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/delay.h>
10 #include <linux/errno.h>
11 #include <linux/hdreg.h>
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/list_sort.h>
16 #include <linux/slab.h>
17 #include <linux/types.h>
19 #include <linux/ptrace.h>
20 #include <linux/nvme_ioctl.h>
21 #include <linux/t10-pi.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
25 #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");
61 module_param(streams, bool, 0644);
62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
65 * nvme_wq - hosts nvme related works that are not reset or delete
66 * nvme_reset_wq - hosts nvme reset works
67 * nvme_delete_wq - hosts nvme delete works
69 * nvme_wq will host works such are scan, aen handling, fw activation,
70 * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
71 * runs reset works which also flush works hosted on nvme_wq for
72 * serialization purposes. nvme_delete_wq host controller deletion
73 * works which flush reset works for serialization.
75 struct workqueue_struct *nvme_wq;
76 EXPORT_SYMBOL_GPL(nvme_wq);
78 struct workqueue_struct *nvme_reset_wq;
79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
81 struct workqueue_struct *nvme_delete_wq;
82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
84 static DEFINE_IDA(nvme_subsystems_ida);
85 static LIST_HEAD(nvme_subsystems);
86 static DEFINE_MUTEX(nvme_subsystems_lock);
88 static DEFINE_IDA(nvme_instance_ida);
89 static dev_t nvme_chr_devt;
90 static struct class *nvme_class;
91 static struct class *nvme_subsys_class;
93 static int nvme_revalidate_disk(struct gendisk *disk);
94 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
95 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
98 static void nvme_set_queue_dying(struct nvme_ns *ns)
101 * Revalidating a dead namespace sets capacity to 0. This will end
102 * buffered writers dirtying pages that can't be synced.
104 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
106 revalidate_disk(ns->disk);
107 blk_set_queue_dying(ns->queue);
108 /* Forcibly unquiesce queues to avoid blocking dispatch */
109 blk_mq_unquiesce_queue(ns->queue);
112 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
115 * Only new queue scan work when admin and IO queues are both alive
117 if (ctrl->state == NVME_CTRL_LIVE)
118 queue_work(nvme_wq, &ctrl->scan_work);
121 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
123 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
125 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
129 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
131 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
135 ret = nvme_reset_ctrl(ctrl);
137 flush_work(&ctrl->reset_work);
138 if (ctrl->state != NVME_CTRL_LIVE &&
139 ctrl->state != NVME_CTRL_ADMIN_ONLY)
145 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
147 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
149 dev_info(ctrl->device,
150 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
152 flush_work(&ctrl->reset_work);
153 nvme_stop_ctrl(ctrl);
154 nvme_remove_namespaces(ctrl);
155 ctrl->ops->delete_ctrl(ctrl);
156 nvme_uninit_ctrl(ctrl);
160 static void nvme_delete_ctrl_work(struct work_struct *work)
162 struct nvme_ctrl *ctrl =
163 container_of(work, struct nvme_ctrl, delete_work);
165 nvme_do_delete_ctrl(ctrl);
168 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
170 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
172 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
176 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
178 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
183 * Keep a reference until nvme_do_delete_ctrl() complete,
184 * since ->delete_ctrl can free the controller.
187 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
190 nvme_do_delete_ctrl(ctrl);
195 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
197 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
200 static blk_status_t nvme_error_status(struct request *req)
202 switch (nvme_req(req)->status & 0x7ff) {
203 case NVME_SC_SUCCESS:
205 case NVME_SC_CAP_EXCEEDED:
206 return BLK_STS_NOSPC;
207 case NVME_SC_LBA_RANGE:
208 return BLK_STS_TARGET;
209 case NVME_SC_BAD_ATTRIBUTES:
210 case NVME_SC_ONCS_NOT_SUPPORTED:
211 case NVME_SC_INVALID_OPCODE:
212 case NVME_SC_INVALID_FIELD:
213 case NVME_SC_INVALID_NS:
214 return BLK_STS_NOTSUPP;
215 case NVME_SC_WRITE_FAULT:
216 case NVME_SC_READ_ERROR:
217 case NVME_SC_UNWRITTEN_BLOCK:
218 case NVME_SC_ACCESS_DENIED:
219 case NVME_SC_READ_ONLY:
220 case NVME_SC_COMPARE_FAILED:
221 return BLK_STS_MEDIUM;
222 case NVME_SC_GUARD_CHECK:
223 case NVME_SC_APPTAG_CHECK:
224 case NVME_SC_REFTAG_CHECK:
225 case NVME_SC_INVALID_PI:
226 return BLK_STS_PROTECTION;
227 case NVME_SC_RESERVATION_CONFLICT:
228 return BLK_STS_NEXUS;
230 return BLK_STS_IOERR;
234 static inline bool nvme_req_needs_retry(struct request *req)
236 if (blk_noretry_request(req))
238 if (nvme_req(req)->status & NVME_SC_DNR)
240 if (nvme_req(req)->retries >= nvme_max_retries)
245 static void nvme_retry_req(struct request *req)
247 struct nvme_ns *ns = req->q->queuedata;
248 unsigned long delay = 0;
251 /* The mask and shift result must be <= 3 */
252 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
254 delay = ns->ctrl->crdt[crd - 1] * 100;
256 nvme_req(req)->retries++;
257 blk_mq_requeue_request(req, false);
258 blk_mq_delay_kick_requeue_list(req->q, delay);
261 void nvme_complete_rq(struct request *req)
263 blk_status_t status = nvme_error_status(req);
265 trace_nvme_complete_rq(req);
267 if (nvme_req(req)->ctrl->kas)
268 nvme_req(req)->ctrl->comp_seen = true;
270 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
271 if ((req->cmd_flags & REQ_NVME_MPATH) &&
272 blk_path_error(status)) {
273 nvme_failover_req(req);
277 if (!blk_queue_dying(req->q)) {
282 blk_mq_end_request(req, status);
284 EXPORT_SYMBOL_GPL(nvme_complete_rq);
286 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
288 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
289 "Cancelling I/O %d", req->tag);
291 /* don't abort one completed request */
292 if (blk_mq_request_completed(req))
295 nvme_req(req)->status = NVME_SC_ABORT_REQ;
296 blk_mq_complete_request(req);
299 EXPORT_SYMBOL_GPL(nvme_cancel_request);
301 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
302 enum nvme_ctrl_state new_state)
304 enum nvme_ctrl_state old_state;
306 bool changed = false;
308 spin_lock_irqsave(&ctrl->lock, flags);
310 old_state = ctrl->state;
312 case NVME_CTRL_ADMIN_ONLY:
314 case NVME_CTRL_CONNECTING:
324 case NVME_CTRL_RESETTING:
325 case NVME_CTRL_CONNECTING:
332 case NVME_CTRL_RESETTING:
336 case NVME_CTRL_ADMIN_ONLY:
343 case NVME_CTRL_CONNECTING:
346 case NVME_CTRL_RESETTING:
353 case NVME_CTRL_DELETING:
356 case NVME_CTRL_ADMIN_ONLY:
357 case NVME_CTRL_RESETTING:
358 case NVME_CTRL_CONNECTING:
367 case NVME_CTRL_DELETING:
379 ctrl->state = new_state;
381 spin_unlock_irqrestore(&ctrl->lock, flags);
382 if (changed && ctrl->state == NVME_CTRL_LIVE)
383 nvme_kick_requeue_lists(ctrl);
386 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
388 static void nvme_free_ns_head(struct kref *ref)
390 struct nvme_ns_head *head =
391 container_of(ref, struct nvme_ns_head, ref);
393 nvme_mpath_remove_disk(head);
394 ida_simple_remove(&head->subsys->ns_ida, head->instance);
395 list_del_init(&head->entry);
396 cleanup_srcu_struct(&head->srcu);
397 nvme_put_subsystem(head->subsys);
401 static void nvme_put_ns_head(struct nvme_ns_head *head)
403 kref_put(&head->ref, nvme_free_ns_head);
406 static void nvme_free_ns(struct kref *kref)
408 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
411 nvme_nvm_unregister(ns);
414 nvme_put_ns_head(ns->head);
415 nvme_put_ctrl(ns->ctrl);
419 static void nvme_put_ns(struct nvme_ns *ns)
421 kref_put(&ns->kref, nvme_free_ns);
424 static inline void nvme_clear_nvme_request(struct request *req)
426 if (!(req->rq_flags & RQF_DONTPREP)) {
427 nvme_req(req)->retries = 0;
428 nvme_req(req)->flags = 0;
429 req->rq_flags |= RQF_DONTPREP;
433 struct request *nvme_alloc_request(struct request_queue *q,
434 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
436 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
439 if (qid == NVME_QID_ANY) {
440 req = blk_mq_alloc_request(q, op, flags);
442 req = blk_mq_alloc_request_hctx(q, op, flags,
448 req->cmd_flags |= REQ_FAILFAST_DRIVER;
449 nvme_clear_nvme_request(req);
450 nvme_req(req)->cmd = cmd;
454 EXPORT_SYMBOL_GPL(nvme_alloc_request);
456 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
458 struct nvme_command c;
460 memset(&c, 0, sizeof(c));
462 c.directive.opcode = nvme_admin_directive_send;
463 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
464 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
465 c.directive.dtype = NVME_DIR_IDENTIFY;
466 c.directive.tdtype = NVME_DIR_STREAMS;
467 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
469 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
472 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
474 return nvme_toggle_streams(ctrl, false);
477 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
479 return nvme_toggle_streams(ctrl, true);
482 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
483 struct streams_directive_params *s, u32 nsid)
485 struct nvme_command c;
487 memset(&c, 0, sizeof(c));
488 memset(s, 0, sizeof(*s));
490 c.directive.opcode = nvme_admin_directive_recv;
491 c.directive.nsid = cpu_to_le32(nsid);
492 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
493 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
494 c.directive.dtype = NVME_DIR_STREAMS;
496 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
499 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
501 struct streams_directive_params s;
504 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
509 ret = nvme_enable_streams(ctrl);
513 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
517 ctrl->nssa = le16_to_cpu(s.nssa);
518 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
519 dev_info(ctrl->device, "too few streams (%u) available\n",
521 nvme_disable_streams(ctrl);
525 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
526 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
531 * Check if 'req' has a write hint associated with it. If it does, assign
532 * a valid namespace stream to the write.
534 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
535 struct request *req, u16 *control,
538 enum rw_hint streamid = req->write_hint;
540 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
544 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
547 *control |= NVME_RW_DTYPE_STREAMS;
548 *dsmgmt |= streamid << 16;
551 if (streamid < ARRAY_SIZE(req->q->write_hints))
552 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
555 static inline void nvme_setup_flush(struct nvme_ns *ns,
556 struct nvme_command *cmnd)
558 cmnd->common.opcode = nvme_cmd_flush;
559 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
562 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
563 struct nvme_command *cmnd)
565 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
566 struct nvme_dsm_range *range;
569 range = kmalloc_array(segments, sizeof(*range),
570 GFP_ATOMIC | __GFP_NOWARN);
573 * If we fail allocation our range, fallback to the controller
574 * discard page. If that's also busy, it's safe to return
575 * busy, as we know we can make progress once that's freed.
577 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
578 return BLK_STS_RESOURCE;
580 range = page_address(ns->ctrl->discard_page);
583 __rq_for_each_bio(bio, req) {
584 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
585 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
588 range[n].cattr = cpu_to_le32(0);
589 range[n].nlb = cpu_to_le32(nlb);
590 range[n].slba = cpu_to_le64(slba);
595 if (WARN_ON_ONCE(n != segments)) {
596 if (virt_to_page(range) == ns->ctrl->discard_page)
597 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
600 return BLK_STS_IOERR;
603 cmnd->dsm.opcode = nvme_cmd_dsm;
604 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
605 cmnd->dsm.nr = cpu_to_le32(segments - 1);
606 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
608 req->special_vec.bv_page = virt_to_page(range);
609 req->special_vec.bv_offset = offset_in_page(range);
610 req->special_vec.bv_len = sizeof(*range) * segments;
611 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
616 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
617 struct request *req, struct nvme_command *cmnd)
619 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
620 return nvme_setup_discard(ns, req, cmnd);
622 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
623 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
624 cmnd->write_zeroes.slba =
625 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
626 cmnd->write_zeroes.length =
627 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
628 cmnd->write_zeroes.control = 0;
632 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
633 struct request *req, struct nvme_command *cmnd)
635 struct nvme_ctrl *ctrl = ns->ctrl;
639 if (req->cmd_flags & REQ_FUA)
640 control |= NVME_RW_FUA;
641 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
642 control |= NVME_RW_LR;
644 if (req->cmd_flags & REQ_RAHEAD)
645 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
647 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
648 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
649 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
650 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
652 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
653 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
657 * If formated with metadata, the block layer always provides a
658 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
659 * we enable the PRACT bit for protection information or set the
660 * namespace capacity to zero to prevent any I/O.
662 if (!blk_integrity_rq(req)) {
663 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
664 return BLK_STS_NOTSUPP;
665 control |= NVME_RW_PRINFO_PRACT;
666 } else if (req_op(req) == REQ_OP_WRITE) {
667 t10_pi_prepare(req, ns->pi_type);
670 switch (ns->pi_type) {
671 case NVME_NS_DPS_PI_TYPE3:
672 control |= NVME_RW_PRINFO_PRCHK_GUARD;
674 case NVME_NS_DPS_PI_TYPE1:
675 case NVME_NS_DPS_PI_TYPE2:
676 control |= NVME_RW_PRINFO_PRCHK_GUARD |
677 NVME_RW_PRINFO_PRCHK_REF;
678 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
683 cmnd->rw.control = cpu_to_le16(control);
684 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
688 void nvme_cleanup_cmd(struct request *req)
690 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
691 nvme_req(req)->status == 0) {
692 struct nvme_ns *ns = req->rq_disk->private_data;
694 t10_pi_complete(req, ns->pi_type,
695 blk_rq_bytes(req) >> ns->lba_shift);
697 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
698 struct nvme_ns *ns = req->rq_disk->private_data;
699 struct page *page = req->special_vec.bv_page;
701 if (page == ns->ctrl->discard_page)
702 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
704 kfree(page_address(page) + req->special_vec.bv_offset);
707 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
709 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
710 struct nvme_command *cmd)
712 blk_status_t ret = BLK_STS_OK;
714 nvme_clear_nvme_request(req);
716 memset(cmd, 0, sizeof(*cmd));
717 switch (req_op(req)) {
720 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
723 nvme_setup_flush(ns, cmd);
725 case REQ_OP_WRITE_ZEROES:
726 ret = nvme_setup_write_zeroes(ns, req, cmd);
729 ret = nvme_setup_discard(ns, req, cmd);
733 ret = nvme_setup_rw(ns, req, cmd);
737 return BLK_STS_IOERR;
740 cmd->common.command_id = req->tag;
741 trace_nvme_setup_cmd(req, cmd);
744 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
746 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
748 struct completion *waiting = rq->end_io_data;
750 rq->end_io_data = NULL;
754 static void nvme_execute_rq_polled(struct request_queue *q,
755 struct gendisk *bd_disk, struct request *rq, int at_head)
757 DECLARE_COMPLETION_ONSTACK(wait);
759 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
761 rq->cmd_flags |= REQ_HIPRI;
762 rq->end_io_data = &wait;
763 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
765 while (!completion_done(&wait)) {
766 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
772 * Returns 0 on success. If the result is negative, it's a Linux error code;
773 * if the result is positive, it's an NVM Express status code
775 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
776 union nvme_result *result, void *buffer, unsigned bufflen,
777 unsigned timeout, int qid, int at_head,
778 blk_mq_req_flags_t flags, bool poll)
783 req = nvme_alloc_request(q, cmd, flags, qid);
787 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
789 if (buffer && bufflen) {
790 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
796 nvme_execute_rq_polled(req->q, NULL, req, at_head);
798 blk_execute_rq(req->q, NULL, req, at_head);
800 *result = nvme_req(req)->result;
801 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
804 ret = nvme_req(req)->status;
806 blk_mq_free_request(req);
809 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
811 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
812 void *buffer, unsigned bufflen)
814 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
815 NVME_QID_ANY, 0, 0, false);
817 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
819 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
820 unsigned len, u32 seed, bool write)
822 struct bio_integrity_payload *bip;
826 buf = kmalloc(len, GFP_KERNEL);
831 if (write && copy_from_user(buf, ubuf, len))
834 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
840 bip->bip_iter.bi_size = len;
841 bip->bip_iter.bi_sector = seed;
842 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
843 offset_in_page(buf));
853 static int nvme_submit_user_cmd(struct request_queue *q,
854 struct nvme_command *cmd, void __user *ubuffer,
855 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
856 u32 meta_seed, u32 *result, unsigned timeout)
858 bool write = nvme_is_write(cmd);
859 struct nvme_ns *ns = q->queuedata;
860 struct gendisk *disk = ns ? ns->disk : NULL;
862 struct bio *bio = NULL;
866 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
870 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
871 nvme_req(req)->flags |= NVME_REQ_USERCMD;
873 if (ubuffer && bufflen) {
874 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
880 if (disk && meta_buffer && meta_len) {
881 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
887 req->cmd_flags |= REQ_INTEGRITY;
891 blk_execute_rq(req->q, disk, req, 0);
892 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
895 ret = nvme_req(req)->status;
897 *result = le32_to_cpu(nvme_req(req)->result.u32);
898 if (meta && !ret && !write) {
899 if (copy_to_user(meta_buffer, meta, meta_len))
905 blk_rq_unmap_user(bio);
907 blk_mq_free_request(req);
911 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
913 struct nvme_ctrl *ctrl = rq->end_io_data;
915 bool startka = false;
917 blk_mq_free_request(rq);
920 dev_err(ctrl->device,
921 "failed nvme_keep_alive_end_io error=%d\n",
926 ctrl->comp_seen = false;
927 spin_lock_irqsave(&ctrl->lock, flags);
928 if (ctrl->state == NVME_CTRL_LIVE ||
929 ctrl->state == NVME_CTRL_CONNECTING)
931 spin_unlock_irqrestore(&ctrl->lock, flags);
933 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
936 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
940 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
945 rq->timeout = ctrl->kato * HZ;
946 rq->end_io_data = ctrl;
948 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
953 static void nvme_keep_alive_work(struct work_struct *work)
955 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
956 struct nvme_ctrl, ka_work);
957 bool comp_seen = ctrl->comp_seen;
959 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
960 dev_dbg(ctrl->device,
961 "reschedule traffic based keep-alive timer\n");
962 ctrl->comp_seen = false;
963 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
967 if (nvme_keep_alive(ctrl)) {
968 /* allocation failure, reset the controller */
969 dev_err(ctrl->device, "keep-alive failed\n");
970 nvme_reset_ctrl(ctrl);
975 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
977 if (unlikely(ctrl->kato == 0))
980 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
983 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
985 if (unlikely(ctrl->kato == 0))
988 cancel_delayed_work_sync(&ctrl->ka_work);
990 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
992 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
994 struct nvme_command c = { };
997 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
998 c.identify.opcode = nvme_admin_identify;
999 c.identify.cns = NVME_ID_CNS_CTRL;
1001 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1005 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1006 sizeof(struct nvme_id_ctrl));
1012 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1013 struct nvme_ns_ids *ids)
1015 struct nvme_command c = { };
1021 c.identify.opcode = nvme_admin_identify;
1022 c.identify.nsid = cpu_to_le32(nsid);
1023 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1025 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1029 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1030 NVME_IDENTIFY_DATA_SIZE);
1034 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1035 struct nvme_ns_id_desc *cur = data + pos;
1040 switch (cur->nidt) {
1041 case NVME_NIDT_EUI64:
1042 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1043 dev_warn(ctrl->device,
1044 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1048 len = NVME_NIDT_EUI64_LEN;
1049 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1051 case NVME_NIDT_NGUID:
1052 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1053 dev_warn(ctrl->device,
1054 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1058 len = NVME_NIDT_NGUID_LEN;
1059 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1061 case NVME_NIDT_UUID:
1062 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1063 dev_warn(ctrl->device,
1064 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1068 len = NVME_NIDT_UUID_LEN;
1069 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1072 /* Skip unknown types */
1077 len += sizeof(*cur);
1084 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1086 struct nvme_command c = { };
1088 c.identify.opcode = nvme_admin_identify;
1089 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1090 c.identify.nsid = cpu_to_le32(nsid);
1091 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1092 NVME_IDENTIFY_DATA_SIZE);
1095 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
1098 struct nvme_id_ns *id;
1099 struct nvme_command c = { };
1102 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1103 c.identify.opcode = nvme_admin_identify;
1104 c.identify.nsid = cpu_to_le32(nsid);
1105 c.identify.cns = NVME_ID_CNS_NS;
1107 id = kmalloc(sizeof(*id), GFP_KERNEL);
1111 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1113 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1121 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1122 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1124 struct nvme_command c;
1125 union nvme_result res;
1128 memset(&c, 0, sizeof(c));
1129 c.features.opcode = op;
1130 c.features.fid = cpu_to_le32(fid);
1131 c.features.dword11 = cpu_to_le32(dword11);
1133 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1134 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1135 if (ret >= 0 && result)
1136 *result = le32_to_cpu(res.u32);
1140 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1141 unsigned int dword11, void *buffer, size_t buflen,
1144 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1147 EXPORT_SYMBOL_GPL(nvme_set_features);
1149 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1150 unsigned int dword11, void *buffer, size_t buflen,
1153 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1156 EXPORT_SYMBOL_GPL(nvme_get_features);
1158 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1160 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1162 int status, nr_io_queues;
1164 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1170 * Degraded controllers might return an error when setting the queue
1171 * count. We still want to be able to bring them online and offer
1172 * access to the admin queue, as that might be only way to fix them up.
1175 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1178 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1179 *count = min(*count, nr_io_queues);
1184 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1186 #define NVME_AEN_SUPPORTED \
1187 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1189 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1191 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1194 if (!supported_aens)
1197 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1200 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1204 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1206 struct nvme_user_io io;
1207 struct nvme_command c;
1208 unsigned length, meta_len;
1209 void __user *metadata;
1211 if (copy_from_user(&io, uio, sizeof(io)))
1216 switch (io.opcode) {
1217 case nvme_cmd_write:
1219 case nvme_cmd_compare:
1225 length = (io.nblocks + 1) << ns->lba_shift;
1226 meta_len = (io.nblocks + 1) * ns->ms;
1227 metadata = (void __user *)(uintptr_t)io.metadata;
1232 } else if (meta_len) {
1233 if ((io.metadata & 3) || !io.metadata)
1237 memset(&c, 0, sizeof(c));
1238 c.rw.opcode = io.opcode;
1239 c.rw.flags = io.flags;
1240 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1241 c.rw.slba = cpu_to_le64(io.slba);
1242 c.rw.length = cpu_to_le16(io.nblocks);
1243 c.rw.control = cpu_to_le16(io.control);
1244 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1245 c.rw.reftag = cpu_to_le32(io.reftag);
1246 c.rw.apptag = cpu_to_le16(io.apptag);
1247 c.rw.appmask = cpu_to_le16(io.appmask);
1249 return nvme_submit_user_cmd(ns->queue, &c,
1250 (void __user *)(uintptr_t)io.addr, length,
1251 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1254 static u32 nvme_known_admin_effects(u8 opcode)
1257 case nvme_admin_format_nvm:
1258 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1259 NVME_CMD_EFFECTS_CSE_MASK;
1260 case nvme_admin_sanitize_nvm:
1261 return NVME_CMD_EFFECTS_CSE_MASK;
1268 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1275 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1276 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1277 dev_warn(ctrl->device,
1278 "IO command:%02x has unhandled effects:%08x\n",
1284 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1285 effects |= nvme_known_admin_effects(opcode);
1288 * For simplicity, IO to all namespaces is quiesced even if the command
1289 * effects say only one namespace is affected.
1291 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1292 mutex_lock(&ctrl->scan_lock);
1293 nvme_start_freeze(ctrl);
1294 nvme_wait_freeze(ctrl);
1299 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1303 down_read(&ctrl->namespaces_rwsem);
1304 list_for_each_entry(ns, &ctrl->namespaces, list)
1305 if (ns->disk && nvme_revalidate_disk(ns->disk))
1306 nvme_set_queue_dying(ns);
1307 up_read(&ctrl->namespaces_rwsem);
1309 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1312 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1315 * Revalidate LBA changes prior to unfreezing. This is necessary to
1316 * prevent memory corruption if a logical block size was changed by
1319 if (effects & NVME_CMD_EFFECTS_LBCC)
1320 nvme_update_formats(ctrl);
1321 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1322 nvme_unfreeze(ctrl);
1323 mutex_unlock(&ctrl->scan_lock);
1325 if (effects & NVME_CMD_EFFECTS_CCC)
1326 nvme_init_identify(ctrl);
1327 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1328 nvme_queue_scan(ctrl);
1331 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1332 struct nvme_passthru_cmd __user *ucmd)
1334 struct nvme_passthru_cmd cmd;
1335 struct nvme_command c;
1336 unsigned timeout = 0;
1340 if (!capable(CAP_SYS_ADMIN))
1342 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1347 memset(&c, 0, sizeof(c));
1348 c.common.opcode = cmd.opcode;
1349 c.common.flags = cmd.flags;
1350 c.common.nsid = cpu_to_le32(cmd.nsid);
1351 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1352 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1353 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1354 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1355 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1356 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1357 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1358 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1361 timeout = msecs_to_jiffies(cmd.timeout_ms);
1363 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1364 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1365 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1366 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1367 0, &cmd.result, timeout);
1368 nvme_passthru_end(ctrl, effects);
1371 if (put_user(cmd.result, &ucmd->result))
1379 * Issue ioctl requests on the first available path. Note that unlike normal
1380 * block layer requests we will not retry failed request on another controller.
1382 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1383 struct nvme_ns_head **head, int *srcu_idx)
1385 #ifdef CONFIG_NVME_MULTIPATH
1386 if (disk->fops == &nvme_ns_head_ops) {
1389 *head = disk->private_data;
1390 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1391 ns = nvme_find_path(*head);
1393 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1399 return disk->private_data;
1402 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1405 srcu_read_unlock(&head->srcu, idx);
1408 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1409 unsigned int cmd, unsigned long arg)
1411 struct nvme_ns_head *head = NULL;
1412 void __user *argp = (void __user *)arg;
1416 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1418 return -EWOULDBLOCK;
1421 * Handle ioctls that apply to the controller instead of the namespace
1422 * seperately and drop the ns SRCU reference early. This avoids a
1423 * deadlock when deleting namespaces using the passthrough interface.
1425 if (cmd == NVME_IOCTL_ADMIN_CMD || is_sed_ioctl(cmd)) {
1426 struct nvme_ctrl *ctrl = ns->ctrl;
1428 nvme_get_ctrl(ns->ctrl);
1429 nvme_put_ns_from_disk(head, srcu_idx);
1431 if (cmd == NVME_IOCTL_ADMIN_CMD)
1432 ret = nvme_user_cmd(ctrl, NULL, argp);
1434 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1436 nvme_put_ctrl(ctrl);
1442 force_successful_syscall_return();
1443 ret = ns->head->ns_id;
1445 case NVME_IOCTL_IO_CMD:
1446 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1448 case NVME_IOCTL_SUBMIT_IO:
1449 ret = nvme_submit_io(ns, argp);
1453 ret = nvme_nvm_ioctl(ns, cmd, arg);
1458 nvme_put_ns_from_disk(head, srcu_idx);
1462 static int nvme_open(struct block_device *bdev, fmode_t mode)
1464 struct nvme_ns *ns = bdev->bd_disk->private_data;
1466 #ifdef CONFIG_NVME_MULTIPATH
1467 /* should never be called due to GENHD_FL_HIDDEN */
1468 if (WARN_ON_ONCE(ns->head->disk))
1471 if (!kref_get_unless_zero(&ns->kref))
1473 if (!try_module_get(ns->ctrl->ops->module))
1484 static void nvme_release(struct gendisk *disk, fmode_t mode)
1486 struct nvme_ns *ns = disk->private_data;
1488 module_put(ns->ctrl->ops->module);
1492 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1494 /* some standard values */
1495 geo->heads = 1 << 6;
1496 geo->sectors = 1 << 5;
1497 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1501 #ifdef CONFIG_BLK_DEV_INTEGRITY
1502 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1504 struct blk_integrity integrity;
1506 memset(&integrity, 0, sizeof(integrity));
1508 case NVME_NS_DPS_PI_TYPE3:
1509 integrity.profile = &t10_pi_type3_crc;
1510 integrity.tag_size = sizeof(u16) + sizeof(u32);
1511 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1513 case NVME_NS_DPS_PI_TYPE1:
1514 case NVME_NS_DPS_PI_TYPE2:
1515 integrity.profile = &t10_pi_type1_crc;
1516 integrity.tag_size = sizeof(u16);
1517 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1520 integrity.profile = NULL;
1523 integrity.tuple_size = ms;
1524 blk_integrity_register(disk, &integrity);
1525 blk_queue_max_integrity_segments(disk->queue, 1);
1528 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1531 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1533 static void nvme_set_chunk_size(struct nvme_ns *ns)
1535 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1536 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1539 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1541 struct nvme_ctrl *ctrl = ns->ctrl;
1542 struct request_queue *queue = disk->queue;
1543 u32 size = queue_logical_block_size(queue);
1545 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1546 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1550 if (ctrl->nr_streams && ns->sws && ns->sgs)
1551 size *= ns->sws * ns->sgs;
1553 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1554 NVME_DSM_MAX_RANGES);
1556 queue->limits.discard_alignment = 0;
1557 queue->limits.discard_granularity = size;
1559 /* If discard is already enabled, don't reset queue limits */
1560 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1563 blk_queue_max_discard_sectors(queue, UINT_MAX);
1564 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1566 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1567 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1570 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1573 unsigned short bs = 1 << ns->lba_shift;
1575 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1576 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1579 * Even though NVMe spec explicitly states that MDTS is not
1580 * applicable to the write-zeroes:- "The restriction does not apply to
1581 * commands that do not transfer data between the host and the
1582 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1583 * In order to be more cautious use controller's max_hw_sectors value
1584 * to configure the maximum sectors for the write-zeroes which is
1585 * configured based on the controller's MDTS field in the
1586 * nvme_init_identify() if available.
1588 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1589 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1591 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1593 blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1596 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1597 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1599 memset(ids, 0, sizeof(*ids));
1601 if (ctrl->vs >= NVME_VS(1, 1, 0))
1602 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1603 if (ctrl->vs >= NVME_VS(1, 2, 0))
1604 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1605 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1606 /* Don't treat error as fatal we potentially
1607 * already have a NGUID or EUI-64
1609 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1610 dev_warn(ctrl->device,
1611 "%s: Identify Descriptors failed\n", __func__);
1615 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1617 return !uuid_is_null(&ids->uuid) ||
1618 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1619 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1622 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1624 return uuid_equal(&a->uuid, &b->uuid) &&
1625 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1626 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1629 static void nvme_update_disk_info(struct gendisk *disk,
1630 struct nvme_ns *ns, struct nvme_id_ns *id)
1632 sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1633 unsigned short bs = 1 << ns->lba_shift;
1634 u32 atomic_bs, phys_bs, io_opt;
1636 if (ns->lba_shift > PAGE_SHIFT) {
1637 /* unsupported block size, set capacity to 0 later */
1640 blk_mq_freeze_queue(disk->queue);
1641 blk_integrity_unregister(disk);
1643 if (id->nabo == 0) {
1645 * Bit 1 indicates whether NAWUPF is defined for this namespace
1646 * and whether it should be used instead of AWUPF. If NAWUPF ==
1647 * 0 then AWUPF must be used instead.
1649 if (id->nsfeat & (1 << 1) && id->nawupf)
1650 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1652 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1658 if (id->nsfeat & (1 << 4)) {
1659 /* NPWG = Namespace Preferred Write Granularity */
1660 phys_bs *= 1 + le16_to_cpu(id->npwg);
1661 /* NOWS = Namespace Optimal Write Size */
1662 io_opt *= 1 + le16_to_cpu(id->nows);
1665 blk_queue_logical_block_size(disk->queue, bs);
1667 * Linux filesystems assume writing a single physical block is
1668 * an atomic operation. Hence limit the physical block size to the
1669 * value of the Atomic Write Unit Power Fail parameter.
1671 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1672 blk_queue_io_min(disk->queue, phys_bs);
1673 blk_queue_io_opt(disk->queue, io_opt);
1675 if (ns->ms && !ns->ext &&
1676 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1677 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1678 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1679 ns->lba_shift > PAGE_SHIFT)
1682 set_capacity(disk, capacity);
1684 nvme_config_discard(disk, ns);
1685 nvme_config_write_zeroes(disk, ns);
1687 if (id->nsattr & (1 << 0))
1688 set_disk_ro(disk, true);
1690 set_disk_ro(disk, false);
1692 blk_mq_unfreeze_queue(disk->queue);
1695 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1697 struct nvme_ns *ns = disk->private_data;
1700 * If identify namespace failed, use default 512 byte block size so
1701 * block layer can use before failing read/write for 0 capacity.
1703 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1704 if (ns->lba_shift == 0)
1706 ns->noiob = le16_to_cpu(id->noiob);
1707 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1708 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1709 /* the PI implementation requires metadata equal t10 pi tuple size */
1710 if (ns->ms == sizeof(struct t10_pi_tuple))
1711 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1716 nvme_set_chunk_size(ns);
1717 nvme_update_disk_info(disk, ns, id);
1718 #ifdef CONFIG_NVME_MULTIPATH
1719 if (ns->head->disk) {
1720 nvme_update_disk_info(ns->head->disk, ns, id);
1721 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1726 static int nvme_revalidate_disk(struct gendisk *disk)
1728 struct nvme_ns *ns = disk->private_data;
1729 struct nvme_ctrl *ctrl = ns->ctrl;
1730 struct nvme_id_ns *id;
1731 struct nvme_ns_ids ids;
1734 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1735 set_capacity(disk, 0);
1739 id = nvme_identify_ns(ctrl, ns->head->ns_id);
1743 if (id->ncap == 0) {
1748 __nvme_revalidate_disk(disk, id);
1749 nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1750 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1751 dev_err(ctrl->device,
1752 "identifiers changed for nsid %d\n", ns->head->ns_id);
1761 static char nvme_pr_type(enum pr_type type)
1764 case PR_WRITE_EXCLUSIVE:
1766 case PR_EXCLUSIVE_ACCESS:
1768 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1770 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1772 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1774 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1781 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1782 u64 key, u64 sa_key, u8 op)
1784 struct nvme_ns_head *head = NULL;
1786 struct nvme_command c;
1788 u8 data[16] = { 0, };
1790 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1792 return -EWOULDBLOCK;
1794 put_unaligned_le64(key, &data[0]);
1795 put_unaligned_le64(sa_key, &data[8]);
1797 memset(&c, 0, sizeof(c));
1798 c.common.opcode = op;
1799 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1800 c.common.cdw10 = cpu_to_le32(cdw10);
1802 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1803 nvme_put_ns_from_disk(head, srcu_idx);
1807 static int nvme_pr_register(struct block_device *bdev, u64 old,
1808 u64 new, unsigned flags)
1812 if (flags & ~PR_FL_IGNORE_KEY)
1815 cdw10 = old ? 2 : 0;
1816 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1817 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1818 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1821 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1822 enum pr_type type, unsigned flags)
1826 if (flags & ~PR_FL_IGNORE_KEY)
1829 cdw10 = nvme_pr_type(type) << 8;
1830 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1831 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1834 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1835 enum pr_type type, bool abort)
1837 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1838 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1841 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1843 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1844 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1847 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1849 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1850 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1853 static const struct pr_ops nvme_pr_ops = {
1854 .pr_register = nvme_pr_register,
1855 .pr_reserve = nvme_pr_reserve,
1856 .pr_release = nvme_pr_release,
1857 .pr_preempt = nvme_pr_preempt,
1858 .pr_clear = nvme_pr_clear,
1861 #ifdef CONFIG_BLK_SED_OPAL
1862 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1865 struct nvme_ctrl *ctrl = data;
1866 struct nvme_command cmd;
1868 memset(&cmd, 0, sizeof(cmd));
1870 cmd.common.opcode = nvme_admin_security_send;
1872 cmd.common.opcode = nvme_admin_security_recv;
1873 cmd.common.nsid = 0;
1874 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1875 cmd.common.cdw11 = cpu_to_le32(len);
1877 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1878 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1880 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1881 #endif /* CONFIG_BLK_SED_OPAL */
1883 static const struct block_device_operations nvme_fops = {
1884 .owner = THIS_MODULE,
1885 .ioctl = nvme_ioctl,
1886 .compat_ioctl = nvme_ioctl,
1888 .release = nvme_release,
1889 .getgeo = nvme_getgeo,
1890 .revalidate_disk= nvme_revalidate_disk,
1891 .pr_ops = &nvme_pr_ops,
1894 #ifdef CONFIG_NVME_MULTIPATH
1895 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1897 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1899 if (!kref_get_unless_zero(&head->ref))
1904 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1906 nvme_put_ns_head(disk->private_data);
1909 const struct block_device_operations nvme_ns_head_ops = {
1910 .owner = THIS_MODULE,
1911 .open = nvme_ns_head_open,
1912 .release = nvme_ns_head_release,
1913 .ioctl = nvme_ioctl,
1914 .compat_ioctl = nvme_ioctl,
1915 .getgeo = nvme_getgeo,
1916 .pr_ops = &nvme_pr_ops,
1918 #endif /* CONFIG_NVME_MULTIPATH */
1920 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1922 unsigned long timeout =
1923 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1924 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1927 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1930 if ((csts & NVME_CSTS_RDY) == bit)
1934 if (fatal_signal_pending(current))
1936 if (time_after(jiffies, timeout)) {
1937 dev_err(ctrl->device,
1938 "Device not ready; aborting %s\n", enabled ?
1939 "initialisation" : "reset");
1948 * If the device has been passed off to us in an enabled state, just clear
1949 * the enabled bit. The spec says we should set the 'shutdown notification
1950 * bits', but doing so may cause the device to complete commands to the
1951 * admin queue ... and we don't know what memory that might be pointing at!
1953 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1957 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1958 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1960 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1964 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1965 msleep(NVME_QUIRK_DELAY_AMOUNT);
1967 return nvme_wait_ready(ctrl, cap, false);
1969 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1971 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1974 * Default to a 4K page size, with the intention to update this
1975 * path in the future to accomodate architectures with differing
1976 * kernel and IO page sizes.
1978 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1981 if (page_shift < dev_page_min) {
1982 dev_err(ctrl->device,
1983 "Minimum device page size %u too large for host (%u)\n",
1984 1 << dev_page_min, 1 << page_shift);
1988 ctrl->page_size = 1 << page_shift;
1990 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1991 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1992 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1993 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1994 ctrl->ctrl_config |= NVME_CC_ENABLE;
1996 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1999 return nvme_wait_ready(ctrl, cap, true);
2001 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2003 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2005 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2009 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2010 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2012 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2016 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2017 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2021 if (fatal_signal_pending(current))
2023 if (time_after(jiffies, timeout)) {
2024 dev_err(ctrl->device,
2025 "Device shutdown incomplete; abort shutdown\n");
2032 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2034 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2035 struct request_queue *q)
2039 if (ctrl->max_hw_sectors) {
2041 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2043 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2044 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2045 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2047 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2048 is_power_of_2(ctrl->max_hw_sectors))
2049 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2050 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2051 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2053 blk_queue_write_cache(q, vwc, vwc);
2056 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2061 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2064 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2065 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2068 dev_warn_once(ctrl->device,
2069 "could not set timestamp (%d)\n", ret);
2073 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2075 struct nvme_feat_host_behavior *host;
2078 /* Don't bother enabling the feature if retry delay is not reported */
2082 host = kzalloc(sizeof(*host), GFP_KERNEL);
2086 host->acre = NVME_ENABLE_ACRE;
2087 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2088 host, sizeof(*host), NULL);
2093 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2096 * APST (Autonomous Power State Transition) lets us program a
2097 * table of power state transitions that the controller will
2098 * perform automatically. We configure it with a simple
2099 * heuristic: we are willing to spend at most 2% of the time
2100 * transitioning between power states. Therefore, when running
2101 * in any given state, we will enter the next lower-power
2102 * non-operational state after waiting 50 * (enlat + exlat)
2103 * microseconds, as long as that state's exit latency is under
2104 * the requested maximum latency.
2106 * We will not autonomously enter any non-operational state for
2107 * which the total latency exceeds ps_max_latency_us. Users
2108 * can set ps_max_latency_us to zero to turn off APST.
2112 struct nvme_feat_auto_pst *table;
2118 * If APST isn't supported or if we haven't been initialized yet,
2119 * then don't do anything.
2124 if (ctrl->npss > 31) {
2125 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2129 table = kzalloc(sizeof(*table), GFP_KERNEL);
2133 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2134 /* Turn off APST. */
2136 dev_dbg(ctrl->device, "APST disabled\n");
2138 __le64 target = cpu_to_le64(0);
2142 * Walk through all states from lowest- to highest-power.
2143 * According to the spec, lower-numbered states use more
2144 * power. NPSS, despite the name, is the index of the
2145 * lowest-power state, not the number of states.
2147 for (state = (int)ctrl->npss; state >= 0; state--) {
2148 u64 total_latency_us, exit_latency_us, transition_ms;
2151 table->entries[state] = target;
2154 * Don't allow transitions to the deepest state
2155 * if it's quirked off.
2157 if (state == ctrl->npss &&
2158 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2162 * Is this state a useful non-operational state for
2163 * higher-power states to autonomously transition to?
2165 if (!(ctrl->psd[state].flags &
2166 NVME_PS_FLAGS_NON_OP_STATE))
2170 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2171 if (exit_latency_us > ctrl->ps_max_latency_us)
2176 le32_to_cpu(ctrl->psd[state].entry_lat);
2179 * This state is good. Use it as the APST idle
2180 * target for higher power states.
2182 transition_ms = total_latency_us + 19;
2183 do_div(transition_ms, 20);
2184 if (transition_ms > (1 << 24) - 1)
2185 transition_ms = (1 << 24) - 1;
2187 target = cpu_to_le64((state << 3) |
2188 (transition_ms << 8));
2193 if (total_latency_us > max_lat_us)
2194 max_lat_us = total_latency_us;
2200 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2202 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2203 max_ps, max_lat_us, (int)sizeof(*table), table);
2207 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2208 table, sizeof(*table), NULL);
2210 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2216 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2218 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2222 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2223 case PM_QOS_LATENCY_ANY:
2231 if (ctrl->ps_max_latency_us != latency) {
2232 ctrl->ps_max_latency_us = latency;
2233 nvme_configure_apst(ctrl);
2237 struct nvme_core_quirk_entry {
2239 * NVMe model and firmware strings are padded with spaces. For
2240 * simplicity, strings in the quirk table are padded with NULLs
2246 unsigned long quirks;
2249 static const struct nvme_core_quirk_entry core_quirks[] = {
2252 * This Toshiba device seems to die using any APST states. See:
2253 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2256 .mn = "THNSF5256GPUK TOSHIBA",
2257 .quirks = NVME_QUIRK_NO_APST,
2261 /* match is null-terminated but idstr is space-padded. */
2262 static bool string_matches(const char *idstr, const char *match, size_t len)
2269 matchlen = strlen(match);
2270 WARN_ON_ONCE(matchlen > len);
2272 if (memcmp(idstr, match, matchlen))
2275 for (; matchlen < len; matchlen++)
2276 if (idstr[matchlen] != ' ')
2282 static bool quirk_matches(const struct nvme_id_ctrl *id,
2283 const struct nvme_core_quirk_entry *q)
2285 return q->vid == le16_to_cpu(id->vid) &&
2286 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2287 string_matches(id->fr, q->fr, sizeof(id->fr));
2290 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2291 struct nvme_id_ctrl *id)
2296 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2297 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2298 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2299 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2303 if (ctrl->vs >= NVME_VS(1, 2, 1))
2304 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2307 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2308 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2309 "nqn.2014.08.org.nvmexpress:%04x%04x",
2310 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2311 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2312 off += sizeof(id->sn);
2313 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2314 off += sizeof(id->mn);
2315 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2318 static void nvme_release_subsystem(struct device *dev)
2320 struct nvme_subsystem *subsys =
2321 container_of(dev, struct nvme_subsystem, dev);
2323 ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2327 static void nvme_destroy_subsystem(struct kref *ref)
2329 struct nvme_subsystem *subsys =
2330 container_of(ref, struct nvme_subsystem, ref);
2332 mutex_lock(&nvme_subsystems_lock);
2333 list_del(&subsys->entry);
2334 mutex_unlock(&nvme_subsystems_lock);
2336 ida_destroy(&subsys->ns_ida);
2337 device_del(&subsys->dev);
2338 put_device(&subsys->dev);
2341 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2343 kref_put(&subsys->ref, nvme_destroy_subsystem);
2346 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2348 struct nvme_subsystem *subsys;
2350 lockdep_assert_held(&nvme_subsystems_lock);
2352 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2353 if (strcmp(subsys->subnqn, subsysnqn))
2355 if (!kref_get_unless_zero(&subsys->ref))
2363 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2364 struct device_attribute subsys_attr_##_name = \
2365 __ATTR(_name, _mode, _show, NULL)
2367 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2368 struct device_attribute *attr,
2371 struct nvme_subsystem *subsys =
2372 container_of(dev, struct nvme_subsystem, dev);
2374 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2376 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2378 #define nvme_subsys_show_str_function(field) \
2379 static ssize_t subsys_##field##_show(struct device *dev, \
2380 struct device_attribute *attr, char *buf) \
2382 struct nvme_subsystem *subsys = \
2383 container_of(dev, struct nvme_subsystem, dev); \
2384 return sprintf(buf, "%.*s\n", \
2385 (int)sizeof(subsys->field), subsys->field); \
2387 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2389 nvme_subsys_show_str_function(model);
2390 nvme_subsys_show_str_function(serial);
2391 nvme_subsys_show_str_function(firmware_rev);
2393 static struct attribute *nvme_subsys_attrs[] = {
2394 &subsys_attr_model.attr,
2395 &subsys_attr_serial.attr,
2396 &subsys_attr_firmware_rev.attr,
2397 &subsys_attr_subsysnqn.attr,
2398 #ifdef CONFIG_NVME_MULTIPATH
2399 &subsys_attr_iopolicy.attr,
2404 static struct attribute_group nvme_subsys_attrs_group = {
2405 .attrs = nvme_subsys_attrs,
2408 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2409 &nvme_subsys_attrs_group,
2413 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2414 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2416 struct nvme_ctrl *tmp;
2418 lockdep_assert_held(&nvme_subsystems_lock);
2420 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2421 if (tmp->state == NVME_CTRL_DELETING ||
2422 tmp->state == NVME_CTRL_DEAD)
2425 if (tmp->cntlid == ctrl->cntlid) {
2426 dev_err(ctrl->device,
2427 "Duplicate cntlid %u with %s, rejecting\n",
2428 ctrl->cntlid, dev_name(tmp->device));
2432 if ((id->cmic & (1 << 1)) ||
2433 (ctrl->opts && ctrl->opts->discovery_nqn))
2436 dev_err(ctrl->device,
2437 "Subsystem does not support multiple controllers\n");
2444 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2446 struct nvme_subsystem *subsys, *found;
2449 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2452 ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2457 subsys->instance = ret;
2458 mutex_init(&subsys->lock);
2459 kref_init(&subsys->ref);
2460 INIT_LIST_HEAD(&subsys->ctrls);
2461 INIT_LIST_HEAD(&subsys->nsheads);
2462 nvme_init_subnqn(subsys, ctrl, id);
2463 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2464 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2465 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2466 subsys->vendor_id = le16_to_cpu(id->vid);
2467 subsys->cmic = id->cmic;
2468 subsys->awupf = le16_to_cpu(id->awupf);
2469 #ifdef CONFIG_NVME_MULTIPATH
2470 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2473 subsys->dev.class = nvme_subsys_class;
2474 subsys->dev.release = nvme_release_subsystem;
2475 subsys->dev.groups = nvme_subsys_attrs_groups;
2476 dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2477 device_initialize(&subsys->dev);
2479 mutex_lock(&nvme_subsystems_lock);
2480 found = __nvme_find_get_subsystem(subsys->subnqn);
2482 put_device(&subsys->dev);
2485 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2487 goto out_put_subsystem;
2490 ret = device_add(&subsys->dev);
2492 dev_err(ctrl->device,
2493 "failed to register subsystem device.\n");
2496 ida_init(&subsys->ns_ida);
2497 list_add_tail(&subsys->entry, &nvme_subsystems);
2500 if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2501 dev_name(ctrl->device))) {
2502 dev_err(ctrl->device,
2503 "failed to create sysfs link from subsystem.\n");
2504 goto out_put_subsystem;
2507 ctrl->subsys = subsys;
2508 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2509 mutex_unlock(&nvme_subsystems_lock);
2513 nvme_put_subsystem(subsys);
2515 mutex_unlock(&nvme_subsystems_lock);
2516 put_device(&subsys->dev);
2520 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2521 void *log, size_t size, u64 offset)
2523 struct nvme_command c = { };
2524 unsigned long dwlen = size / 4 - 1;
2526 c.get_log_page.opcode = nvme_admin_get_log_page;
2527 c.get_log_page.nsid = cpu_to_le32(nsid);
2528 c.get_log_page.lid = log_page;
2529 c.get_log_page.lsp = lsp;
2530 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2531 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2532 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2533 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2535 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2538 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2543 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2548 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2549 ctrl->effects, sizeof(*ctrl->effects), 0);
2551 kfree(ctrl->effects);
2552 ctrl->effects = NULL;
2558 * Initialize the cached copies of the Identify data and various controller
2559 * register in our nvme_ctrl structure. This should be called as soon as
2560 * the admin queue is fully up and running.
2562 int nvme_init_identify(struct nvme_ctrl *ctrl)
2564 struct nvme_id_ctrl *id;
2565 int ret, page_shift;
2567 bool prev_apst_enabled;
2569 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2571 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2575 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2577 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2580 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2582 if (ctrl->vs >= NVME_VS(1, 1, 0))
2583 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2585 ret = nvme_identify_ctrl(ctrl, &id);
2587 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2591 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2592 ret = nvme_get_effects_log(ctrl);
2597 if (!ctrl->identified) {
2600 ret = nvme_init_subsystem(ctrl, id);
2605 * Check for quirks. Quirk can depend on firmware version,
2606 * so, in principle, the set of quirks present can change
2607 * across a reset. As a possible future enhancement, we
2608 * could re-scan for quirks every time we reinitialize
2609 * the device, but we'd have to make sure that the driver
2610 * behaves intelligently if the quirks change.
2612 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2613 if (quirk_matches(id, &core_quirks[i]))
2614 ctrl->quirks |= core_quirks[i].quirks;
2618 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2619 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2620 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2623 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2624 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2625 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2627 ctrl->oacs = le16_to_cpu(id->oacs);
2628 ctrl->oncs = le16_to_cpu(id->oncs);
2629 ctrl->mtfa = le16_to_cpu(id->mtfa);
2630 ctrl->oaes = le32_to_cpu(id->oaes);
2631 atomic_set(&ctrl->abort_limit, id->acl + 1);
2632 ctrl->vwc = id->vwc;
2634 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2636 max_hw_sectors = UINT_MAX;
2637 ctrl->max_hw_sectors =
2638 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2640 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2641 ctrl->sgls = le32_to_cpu(id->sgls);
2642 ctrl->kas = le16_to_cpu(id->kas);
2643 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2644 ctrl->ctratt = le32_to_cpu(id->ctratt);
2648 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2650 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2651 shutdown_timeout, 60);
2653 if (ctrl->shutdown_timeout != shutdown_timeout)
2654 dev_info(ctrl->device,
2655 "Shutdown timeout set to %u seconds\n",
2656 ctrl->shutdown_timeout);
2658 ctrl->shutdown_timeout = shutdown_timeout;
2660 ctrl->npss = id->npss;
2661 ctrl->apsta = id->apsta;
2662 prev_apst_enabled = ctrl->apst_enabled;
2663 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2664 if (force_apst && id->apsta) {
2665 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2666 ctrl->apst_enabled = true;
2668 ctrl->apst_enabled = false;
2671 ctrl->apst_enabled = id->apsta;
2673 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2675 if (ctrl->ops->flags & NVME_F_FABRICS) {
2676 ctrl->icdoff = le16_to_cpu(id->icdoff);
2677 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2678 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2679 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2682 * In fabrics we need to verify the cntlid matches the
2685 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2690 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2691 dev_err(ctrl->device,
2692 "keep-alive support is mandatory for fabrics\n");
2697 ctrl->cntlid = le16_to_cpu(id->cntlid);
2698 ctrl->hmpre = le32_to_cpu(id->hmpre);
2699 ctrl->hmmin = le32_to_cpu(id->hmmin);
2700 ctrl->hmminds = le32_to_cpu(id->hmminds);
2701 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2704 ret = nvme_mpath_init(ctrl, id);
2710 if (ctrl->apst_enabled && !prev_apst_enabled)
2711 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2712 else if (!ctrl->apst_enabled && prev_apst_enabled)
2713 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2715 ret = nvme_configure_apst(ctrl);
2719 ret = nvme_configure_timestamp(ctrl);
2723 ret = nvme_configure_directives(ctrl);
2727 ret = nvme_configure_acre(ctrl);
2731 ctrl->identified = true;
2739 EXPORT_SYMBOL_GPL(nvme_init_identify);
2741 static int nvme_dev_open(struct inode *inode, struct file *file)
2743 struct nvme_ctrl *ctrl =
2744 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2746 switch (ctrl->state) {
2747 case NVME_CTRL_LIVE:
2748 case NVME_CTRL_ADMIN_ONLY:
2751 return -EWOULDBLOCK;
2754 file->private_data = ctrl;
2758 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2763 down_read(&ctrl->namespaces_rwsem);
2764 if (list_empty(&ctrl->namespaces)) {
2769 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2770 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2771 dev_warn(ctrl->device,
2772 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2777 dev_warn(ctrl->device,
2778 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2779 kref_get(&ns->kref);
2780 up_read(&ctrl->namespaces_rwsem);
2782 ret = nvme_user_cmd(ctrl, ns, argp);
2787 up_read(&ctrl->namespaces_rwsem);
2791 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2794 struct nvme_ctrl *ctrl = file->private_data;
2795 void __user *argp = (void __user *)arg;
2798 case NVME_IOCTL_ADMIN_CMD:
2799 return nvme_user_cmd(ctrl, NULL, argp);
2800 case NVME_IOCTL_IO_CMD:
2801 return nvme_dev_user_cmd(ctrl, argp);
2802 case NVME_IOCTL_RESET:
2803 dev_warn(ctrl->device, "resetting controller\n");
2804 return nvme_reset_ctrl_sync(ctrl);
2805 case NVME_IOCTL_SUBSYS_RESET:
2806 return nvme_reset_subsystem(ctrl);
2807 case NVME_IOCTL_RESCAN:
2808 nvme_queue_scan(ctrl);
2815 static const struct file_operations nvme_dev_fops = {
2816 .owner = THIS_MODULE,
2817 .open = nvme_dev_open,
2818 .unlocked_ioctl = nvme_dev_ioctl,
2819 .compat_ioctl = nvme_dev_ioctl,
2822 static ssize_t nvme_sysfs_reset(struct device *dev,
2823 struct device_attribute *attr, const char *buf,
2826 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2829 ret = nvme_reset_ctrl_sync(ctrl);
2834 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2836 static ssize_t nvme_sysfs_rescan(struct device *dev,
2837 struct device_attribute *attr, const char *buf,
2840 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2842 nvme_queue_scan(ctrl);
2845 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2847 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2849 struct gendisk *disk = dev_to_disk(dev);
2851 if (disk->fops == &nvme_fops)
2852 return nvme_get_ns_from_dev(dev)->head;
2854 return disk->private_data;
2857 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2860 struct nvme_ns_head *head = dev_to_ns_head(dev);
2861 struct nvme_ns_ids *ids = &head->ids;
2862 struct nvme_subsystem *subsys = head->subsys;
2863 int serial_len = sizeof(subsys->serial);
2864 int model_len = sizeof(subsys->model);
2866 if (!uuid_is_null(&ids->uuid))
2867 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2869 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2870 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2872 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2873 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2875 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2876 subsys->serial[serial_len - 1] == '\0'))
2878 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2879 subsys->model[model_len - 1] == '\0'))
2882 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2883 serial_len, subsys->serial, model_len, subsys->model,
2886 static DEVICE_ATTR_RO(wwid);
2888 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2891 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2893 static DEVICE_ATTR_RO(nguid);
2895 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2898 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2900 /* For backward compatibility expose the NGUID to userspace if
2901 * we have no UUID set
2903 if (uuid_is_null(&ids->uuid)) {
2904 printk_ratelimited(KERN_WARNING
2905 "No UUID available providing old NGUID\n");
2906 return sprintf(buf, "%pU\n", ids->nguid);
2908 return sprintf(buf, "%pU\n", &ids->uuid);
2910 static DEVICE_ATTR_RO(uuid);
2912 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2915 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2917 static DEVICE_ATTR_RO(eui);
2919 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2922 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2924 static DEVICE_ATTR_RO(nsid);
2926 static struct attribute *nvme_ns_id_attrs[] = {
2927 &dev_attr_wwid.attr,
2928 &dev_attr_uuid.attr,
2929 &dev_attr_nguid.attr,
2931 &dev_attr_nsid.attr,
2932 #ifdef CONFIG_NVME_MULTIPATH
2933 &dev_attr_ana_grpid.attr,
2934 &dev_attr_ana_state.attr,
2939 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2940 struct attribute *a, int n)
2942 struct device *dev = container_of(kobj, struct device, kobj);
2943 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2945 if (a == &dev_attr_uuid.attr) {
2946 if (uuid_is_null(&ids->uuid) &&
2947 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2950 if (a == &dev_attr_nguid.attr) {
2951 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2954 if (a == &dev_attr_eui.attr) {
2955 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2958 #ifdef CONFIG_NVME_MULTIPATH
2959 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2960 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2962 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2969 static const struct attribute_group nvme_ns_id_attr_group = {
2970 .attrs = nvme_ns_id_attrs,
2971 .is_visible = nvme_ns_id_attrs_are_visible,
2974 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2975 &nvme_ns_id_attr_group,
2977 &nvme_nvm_attr_group,
2982 #define nvme_show_str_function(field) \
2983 static ssize_t field##_show(struct device *dev, \
2984 struct device_attribute *attr, char *buf) \
2986 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2987 return sprintf(buf, "%.*s\n", \
2988 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
2990 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2992 nvme_show_str_function(model);
2993 nvme_show_str_function(serial);
2994 nvme_show_str_function(firmware_rev);
2996 #define nvme_show_int_function(field) \
2997 static ssize_t field##_show(struct device *dev, \
2998 struct device_attribute *attr, char *buf) \
3000 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3001 return sprintf(buf, "%d\n", ctrl->field); \
3003 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3005 nvme_show_int_function(cntlid);
3006 nvme_show_int_function(numa_node);
3008 static ssize_t nvme_sysfs_delete(struct device *dev,
3009 struct device_attribute *attr, const char *buf,
3012 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3014 if (device_remove_file_self(dev, attr))
3015 nvme_delete_ctrl_sync(ctrl);
3018 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3020 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3021 struct device_attribute *attr,
3024 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3026 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3028 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3030 static ssize_t nvme_sysfs_show_state(struct device *dev,
3031 struct device_attribute *attr,
3034 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3035 static const char *const state_name[] = {
3036 [NVME_CTRL_NEW] = "new",
3037 [NVME_CTRL_LIVE] = "live",
3038 [NVME_CTRL_ADMIN_ONLY] = "only-admin",
3039 [NVME_CTRL_RESETTING] = "resetting",
3040 [NVME_CTRL_CONNECTING] = "connecting",
3041 [NVME_CTRL_DELETING] = "deleting",
3042 [NVME_CTRL_DEAD] = "dead",
3045 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3046 state_name[ctrl->state])
3047 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3049 return sprintf(buf, "unknown state\n");
3052 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3054 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3055 struct device_attribute *attr,
3058 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3060 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3062 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3064 static ssize_t nvme_sysfs_show_address(struct device *dev,
3065 struct device_attribute *attr,
3068 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3070 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3072 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3074 static struct attribute *nvme_dev_attrs[] = {
3075 &dev_attr_reset_controller.attr,
3076 &dev_attr_rescan_controller.attr,
3077 &dev_attr_model.attr,
3078 &dev_attr_serial.attr,
3079 &dev_attr_firmware_rev.attr,
3080 &dev_attr_cntlid.attr,
3081 &dev_attr_delete_controller.attr,
3082 &dev_attr_transport.attr,
3083 &dev_attr_subsysnqn.attr,
3084 &dev_attr_address.attr,
3085 &dev_attr_state.attr,
3086 &dev_attr_numa_node.attr,
3090 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3091 struct attribute *a, int n)
3093 struct device *dev = container_of(kobj, struct device, kobj);
3094 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3096 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3098 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3104 static struct attribute_group nvme_dev_attrs_group = {
3105 .attrs = nvme_dev_attrs,
3106 .is_visible = nvme_dev_attrs_are_visible,
3109 static const struct attribute_group *nvme_dev_attr_groups[] = {
3110 &nvme_dev_attrs_group,
3114 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3117 struct nvme_ns_head *h;
3119 lockdep_assert_held(&subsys->lock);
3121 list_for_each_entry(h, &subsys->nsheads, entry) {
3122 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3129 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3130 struct nvme_ns_head *new)
3132 struct nvme_ns_head *h;
3134 lockdep_assert_held(&subsys->lock);
3136 list_for_each_entry(h, &subsys->nsheads, entry) {
3137 if (nvme_ns_ids_valid(&new->ids) &&
3138 !list_empty(&h->list) &&
3139 nvme_ns_ids_equal(&new->ids, &h->ids))
3146 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3147 unsigned nsid, struct nvme_id_ns *id)
3149 struct nvme_ns_head *head;
3150 size_t size = sizeof(*head);
3153 #ifdef CONFIG_NVME_MULTIPATH
3154 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3157 head = kzalloc(size, GFP_KERNEL);
3160 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3163 head->instance = ret;
3164 INIT_LIST_HEAD(&head->list);
3165 ret = init_srcu_struct(&head->srcu);
3167 goto out_ida_remove;
3168 head->subsys = ctrl->subsys;
3170 kref_init(&head->ref);
3172 nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3174 ret = __nvme_check_ids(ctrl->subsys, head);
3176 dev_err(ctrl->device,
3177 "duplicate IDs for nsid %d\n", nsid);
3178 goto out_cleanup_srcu;
3181 ret = nvme_mpath_alloc_disk(ctrl, head);
3183 goto out_cleanup_srcu;
3185 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3187 kref_get(&ctrl->subsys->ref);
3191 cleanup_srcu_struct(&head->srcu);
3193 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3197 return ERR_PTR(ret);
3200 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3201 struct nvme_id_ns *id)
3203 struct nvme_ctrl *ctrl = ns->ctrl;
3204 bool is_shared = id->nmic & (1 << 0);
3205 struct nvme_ns_head *head = NULL;
3208 mutex_lock(&ctrl->subsys->lock);
3210 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3212 head = nvme_alloc_ns_head(ctrl, nsid, id);
3214 ret = PTR_ERR(head);
3218 struct nvme_ns_ids ids;
3220 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3221 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3222 dev_err(ctrl->device,
3223 "IDs don't match for shared namespace %d\n",
3230 list_add_tail(&ns->siblings, &head->list);
3234 mutex_unlock(&ctrl->subsys->lock);
3238 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3240 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3241 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3243 return nsa->head->ns_id - nsb->head->ns_id;
3246 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3248 struct nvme_ns *ns, *ret = NULL;
3250 down_read(&ctrl->namespaces_rwsem);
3251 list_for_each_entry(ns, &ctrl->namespaces, list) {
3252 if (ns->head->ns_id == nsid) {
3253 if (!kref_get_unless_zero(&ns->kref))
3258 if (ns->head->ns_id > nsid)
3261 up_read(&ctrl->namespaces_rwsem);
3265 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3267 struct streams_directive_params s;
3270 if (!ctrl->nr_streams)
3273 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3277 ns->sws = le32_to_cpu(s.sws);
3278 ns->sgs = le16_to_cpu(s.sgs);
3281 unsigned int bs = 1 << ns->lba_shift;
3283 blk_queue_io_min(ns->queue, bs * ns->sws);
3285 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3291 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3294 struct gendisk *disk;
3295 struct nvme_id_ns *id;
3296 char disk_name[DISK_NAME_LEN];
3297 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3299 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3303 ns->queue = blk_mq_init_queue(ctrl->tagset);
3304 if (IS_ERR(ns->queue)) {
3305 ret = PTR_ERR(ns->queue);
3309 if (ctrl->opts && ctrl->opts->data_digest)
3310 ns->queue->backing_dev_info->capabilities
3311 |= BDI_CAP_STABLE_WRITES;
3313 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3314 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3315 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3317 ns->queue->queuedata = ns;
3320 kref_init(&ns->kref);
3321 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3323 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3324 nvme_set_queue_limits(ctrl, ns->queue);
3326 id = nvme_identify_ns(ctrl, nsid);
3329 goto out_free_queue;
3332 if (id->ncap == 0) {
3337 ret = nvme_init_ns_head(ns, nsid, id);
3340 nvme_setup_streams_ns(ctrl, ns);
3341 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3343 disk = alloc_disk_node(0, node);
3349 disk->fops = &nvme_fops;
3350 disk->private_data = ns;
3351 disk->queue = ns->queue;
3352 disk->flags = flags;
3353 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3356 __nvme_revalidate_disk(disk, id);
3358 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3359 ret = nvme_nvm_register(ns, disk_name, node);
3361 dev_warn(ctrl->device, "LightNVM init failure\n");
3366 down_write(&ctrl->namespaces_rwsem);
3367 list_add_tail(&ns->list, &ctrl->namespaces);
3368 up_write(&ctrl->namespaces_rwsem);
3370 nvme_get_ctrl(ctrl);
3372 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3374 nvme_mpath_add_disk(ns, id);
3375 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3382 mutex_lock(&ctrl->subsys->lock);
3383 list_del_rcu(&ns->siblings);
3384 mutex_unlock(&ctrl->subsys->lock);
3385 nvme_put_ns_head(ns->head);
3389 blk_cleanup_queue(ns->queue);
3395 static void nvme_ns_remove(struct nvme_ns *ns)
3397 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3400 nvme_fault_inject_fini(&ns->fault_inject);
3402 mutex_lock(&ns->ctrl->subsys->lock);
3403 list_del_rcu(&ns->siblings);
3404 mutex_unlock(&ns->ctrl->subsys->lock);
3405 synchronize_rcu(); /* guarantee not available in head->list */
3406 nvme_mpath_clear_current_path(ns);
3407 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3409 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3410 del_gendisk(ns->disk);
3411 blk_cleanup_queue(ns->queue);
3412 if (blk_get_integrity(ns->disk))
3413 blk_integrity_unregister(ns->disk);
3416 down_write(&ns->ctrl->namespaces_rwsem);
3417 list_del_init(&ns->list);
3418 up_write(&ns->ctrl->namespaces_rwsem);
3420 nvme_mpath_check_last_path(ns);
3424 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3428 ns = nvme_find_get_ns(ctrl, nsid);
3430 if (ns->disk && revalidate_disk(ns->disk))
3434 nvme_alloc_ns(ctrl, nsid);
3437 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3440 struct nvme_ns *ns, *next;
3443 down_write(&ctrl->namespaces_rwsem);
3444 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3445 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3446 list_move_tail(&ns->list, &rm_list);
3448 up_write(&ctrl->namespaces_rwsem);
3450 list_for_each_entry_safe(ns, next, &rm_list, list)
3455 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3459 unsigned i, j, nsid, prev = 0;
3460 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3463 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3467 for (i = 0; i < num_lists; i++) {
3468 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3472 for (j = 0; j < min(nn, 1024U); j++) {
3473 nsid = le32_to_cpu(ns_list[j]);
3477 nvme_validate_ns(ctrl, nsid);
3479 while (++prev < nsid) {
3480 ns = nvme_find_get_ns(ctrl, prev);
3490 nvme_remove_invalid_namespaces(ctrl, prev);
3496 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3500 for (i = 1; i <= nn; i++)
3501 nvme_validate_ns(ctrl, i);
3503 nvme_remove_invalid_namespaces(ctrl, nn);
3506 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3508 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3512 log = kzalloc(log_size, GFP_KERNEL);
3517 * We need to read the log to clear the AEN, but we don't want to rely
3518 * on it for the changed namespace information as userspace could have
3519 * raced with us in reading the log page, which could cause us to miss
3522 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3525 dev_warn(ctrl->device,
3526 "reading changed ns log failed: %d\n", error);
3531 static void nvme_scan_work(struct work_struct *work)
3533 struct nvme_ctrl *ctrl =
3534 container_of(work, struct nvme_ctrl, scan_work);
3535 struct nvme_id_ctrl *id;
3538 if (ctrl->state != NVME_CTRL_LIVE)
3541 WARN_ON_ONCE(!ctrl->tagset);
3543 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3544 dev_info(ctrl->device, "rescanning namespaces.\n");
3545 nvme_clear_changed_ns_log(ctrl);
3548 if (nvme_identify_ctrl(ctrl, &id))
3551 mutex_lock(&ctrl->scan_lock);
3552 nn = le32_to_cpu(id->nn);
3553 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3554 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3555 if (!nvme_scan_ns_list(ctrl, nn))
3558 nvme_scan_ns_sequential(ctrl, nn);
3560 mutex_unlock(&ctrl->scan_lock);
3562 down_write(&ctrl->namespaces_rwsem);
3563 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3564 up_write(&ctrl->namespaces_rwsem);
3568 * This function iterates the namespace list unlocked to allow recovery from
3569 * controller failure. It is up to the caller to ensure the namespace list is
3570 * not modified by scan work while this function is executing.
3572 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3574 struct nvme_ns *ns, *next;
3577 /* prevent racing with ns scanning */
3578 flush_work(&ctrl->scan_work);
3581 * The dead states indicates the controller was not gracefully
3582 * disconnected. In that case, we won't be able to flush any data while
3583 * removing the namespaces' disks; fail all the queues now to avoid
3584 * potentially having to clean up the failed sync later.
3586 if (ctrl->state == NVME_CTRL_DEAD)
3587 nvme_kill_queues(ctrl);
3589 down_write(&ctrl->namespaces_rwsem);
3590 list_splice_init(&ctrl->namespaces, &ns_list);
3591 up_write(&ctrl->namespaces_rwsem);
3593 list_for_each_entry_safe(ns, next, &ns_list, list)
3596 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3598 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3600 char *envp[2] = { NULL, NULL };
3601 u32 aen_result = ctrl->aen_result;
3603 ctrl->aen_result = 0;
3607 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3610 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3614 static void nvme_async_event_work(struct work_struct *work)
3616 struct nvme_ctrl *ctrl =
3617 container_of(work, struct nvme_ctrl, async_event_work);
3619 nvme_aen_uevent(ctrl);
3620 ctrl->ops->submit_async_event(ctrl);
3623 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3628 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3634 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3637 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3639 struct nvme_fw_slot_info_log *log;
3641 log = kmalloc(sizeof(*log), GFP_KERNEL);
3645 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3647 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3651 static void nvme_fw_act_work(struct work_struct *work)
3653 struct nvme_ctrl *ctrl = container_of(work,
3654 struct nvme_ctrl, fw_act_work);
3655 unsigned long fw_act_timeout;
3658 fw_act_timeout = jiffies +
3659 msecs_to_jiffies(ctrl->mtfa * 100);
3661 fw_act_timeout = jiffies +
3662 msecs_to_jiffies(admin_timeout * 1000);
3664 nvme_stop_queues(ctrl);
3665 while (nvme_ctrl_pp_status(ctrl)) {
3666 if (time_after(jiffies, fw_act_timeout)) {
3667 dev_warn(ctrl->device,
3668 "Fw activation timeout, reset controller\n");
3669 nvme_reset_ctrl(ctrl);
3675 if (ctrl->state != NVME_CTRL_LIVE)
3678 nvme_start_queues(ctrl);
3679 /* read FW slot information to clear the AER */
3680 nvme_get_fw_slot_info(ctrl);
3683 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3685 u32 aer_notice_type = (result & 0xff00) >> 8;
3687 trace_nvme_async_event(ctrl, aer_notice_type);
3689 switch (aer_notice_type) {
3690 case NVME_AER_NOTICE_NS_CHANGED:
3691 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3692 nvme_queue_scan(ctrl);
3694 case NVME_AER_NOTICE_FW_ACT_STARTING:
3695 queue_work(nvme_wq, &ctrl->fw_act_work);
3697 #ifdef CONFIG_NVME_MULTIPATH
3698 case NVME_AER_NOTICE_ANA:
3699 if (!ctrl->ana_log_buf)
3701 queue_work(nvme_wq, &ctrl->ana_work);
3705 dev_warn(ctrl->device, "async event result %08x\n", result);
3709 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3710 volatile union nvme_result *res)
3712 u32 result = le32_to_cpu(res->u32);
3713 u32 aer_type = result & 0x07;
3715 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3719 case NVME_AER_NOTICE:
3720 nvme_handle_aen_notice(ctrl, result);
3722 case NVME_AER_ERROR:
3723 case NVME_AER_SMART:
3726 trace_nvme_async_event(ctrl, aer_type);
3727 ctrl->aen_result = result;
3732 queue_work(nvme_wq, &ctrl->async_event_work);
3734 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3736 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3738 nvme_mpath_stop(ctrl);
3739 nvme_stop_keep_alive(ctrl);
3740 flush_work(&ctrl->async_event_work);
3741 cancel_work_sync(&ctrl->fw_act_work);
3743 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3745 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3748 nvme_start_keep_alive(ctrl);
3750 if (ctrl->queue_count > 1) {
3751 nvme_queue_scan(ctrl);
3752 nvme_enable_aen(ctrl);
3753 queue_work(nvme_wq, &ctrl->async_event_work);
3754 nvme_start_queues(ctrl);
3757 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3759 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3761 nvme_fault_inject_fini(&ctrl->fault_inject);
3762 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3763 cdev_device_del(&ctrl->cdev, ctrl->device);
3765 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3767 static void nvme_free_ctrl(struct device *dev)
3769 struct nvme_ctrl *ctrl =
3770 container_of(dev, struct nvme_ctrl, ctrl_device);
3771 struct nvme_subsystem *subsys = ctrl->subsys;
3773 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3774 kfree(ctrl->effects);
3775 nvme_mpath_uninit(ctrl);
3776 __free_page(ctrl->discard_page);
3779 mutex_lock(&nvme_subsystems_lock);
3780 list_del(&ctrl->subsys_entry);
3781 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3782 mutex_unlock(&nvme_subsystems_lock);
3785 ctrl->ops->free_ctrl(ctrl);
3788 nvme_put_subsystem(subsys);
3792 * Initialize a NVMe controller structures. This needs to be called during
3793 * earliest initialization so that we have the initialized structured around
3796 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3797 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3801 ctrl->state = NVME_CTRL_NEW;
3802 spin_lock_init(&ctrl->lock);
3803 mutex_init(&ctrl->scan_lock);
3804 INIT_LIST_HEAD(&ctrl->namespaces);
3805 init_rwsem(&ctrl->namespaces_rwsem);
3808 ctrl->quirks = quirks;
3809 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3810 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3811 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3812 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3814 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3815 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3816 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3818 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3820 ctrl->discard_page = alloc_page(GFP_KERNEL);
3821 if (!ctrl->discard_page) {
3826 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3829 ctrl->instance = ret;
3831 device_initialize(&ctrl->ctrl_device);
3832 ctrl->device = &ctrl->ctrl_device;
3833 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3834 ctrl->device->class = nvme_class;
3835 ctrl->device->parent = ctrl->dev;
3836 ctrl->device->groups = nvme_dev_attr_groups;
3837 ctrl->device->release = nvme_free_ctrl;
3838 dev_set_drvdata(ctrl->device, ctrl);
3839 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3841 goto out_release_instance;
3843 cdev_init(&ctrl->cdev, &nvme_dev_fops);
3844 ctrl->cdev.owner = ops->module;
3845 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3850 * Initialize latency tolerance controls. The sysfs files won't
3851 * be visible to userspace unless the device actually supports APST.
3853 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3854 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3855 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3857 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
3861 kfree_const(ctrl->device->kobj.name);
3862 out_release_instance:
3863 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3865 if (ctrl->discard_page)
3866 __free_page(ctrl->discard_page);
3869 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3872 * nvme_kill_queues(): Ends all namespace queues
3873 * @ctrl: the dead controller that needs to end
3875 * Call this function when the driver determines it is unable to get the
3876 * controller in a state capable of servicing IO.
3878 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3882 down_read(&ctrl->namespaces_rwsem);
3884 /* Forcibly unquiesce queues to avoid blocking dispatch */
3885 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3886 blk_mq_unquiesce_queue(ctrl->admin_q);
3888 list_for_each_entry(ns, &ctrl->namespaces, list)
3889 nvme_set_queue_dying(ns);
3891 up_read(&ctrl->namespaces_rwsem);
3893 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3895 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3899 down_read(&ctrl->namespaces_rwsem);
3900 list_for_each_entry(ns, &ctrl->namespaces, list)
3901 blk_mq_unfreeze_queue(ns->queue);
3902 up_read(&ctrl->namespaces_rwsem);
3904 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3906 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3910 down_read(&ctrl->namespaces_rwsem);
3911 list_for_each_entry(ns, &ctrl->namespaces, list) {
3912 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3916 up_read(&ctrl->namespaces_rwsem);
3918 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3920 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3924 down_read(&ctrl->namespaces_rwsem);
3925 list_for_each_entry(ns, &ctrl->namespaces, list)
3926 blk_mq_freeze_queue_wait(ns->queue);
3927 up_read(&ctrl->namespaces_rwsem);
3929 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3931 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3935 down_read(&ctrl->namespaces_rwsem);
3936 list_for_each_entry(ns, &ctrl->namespaces, list)
3937 blk_freeze_queue_start(ns->queue);
3938 up_read(&ctrl->namespaces_rwsem);
3940 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3942 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3946 down_read(&ctrl->namespaces_rwsem);
3947 list_for_each_entry(ns, &ctrl->namespaces, list)
3948 blk_mq_quiesce_queue(ns->queue);
3949 up_read(&ctrl->namespaces_rwsem);
3951 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3953 void nvme_start_queues(struct nvme_ctrl *ctrl)
3957 down_read(&ctrl->namespaces_rwsem);
3958 list_for_each_entry(ns, &ctrl->namespaces, list)
3959 blk_mq_unquiesce_queue(ns->queue);
3960 up_read(&ctrl->namespaces_rwsem);
3962 EXPORT_SYMBOL_GPL(nvme_start_queues);
3965 void nvme_sync_queues(struct nvme_ctrl *ctrl)
3969 down_read(&ctrl->namespaces_rwsem);
3970 list_for_each_entry(ns, &ctrl->namespaces, list)
3971 blk_sync_queue(ns->queue);
3972 up_read(&ctrl->namespaces_rwsem);
3974 EXPORT_SYMBOL_GPL(nvme_sync_queues);
3977 * Check we didn't inadvertently grow the command structure sizes:
3979 static inline void _nvme_check_size(void)
3981 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
3982 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
3983 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
3984 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
3985 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
3986 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
3987 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
3988 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
3989 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
3990 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
3991 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
3992 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
3993 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
3994 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
3995 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
3996 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
3997 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4001 static int __init nvme_core_init(void)
4003 int result = -ENOMEM;
4007 nvme_wq = alloc_workqueue("nvme-wq",
4008 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4012 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4013 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4017 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4018 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4019 if (!nvme_delete_wq)
4020 goto destroy_reset_wq;
4022 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4024 goto destroy_delete_wq;
4026 nvme_class = class_create(THIS_MODULE, "nvme");
4027 if (IS_ERR(nvme_class)) {
4028 result = PTR_ERR(nvme_class);
4029 goto unregister_chrdev;
4032 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4033 if (IS_ERR(nvme_subsys_class)) {
4034 result = PTR_ERR(nvme_subsys_class);
4040 class_destroy(nvme_class);
4042 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4044 destroy_workqueue(nvme_delete_wq);
4046 destroy_workqueue(nvme_reset_wq);
4048 destroy_workqueue(nvme_wq);
4053 static void __exit nvme_core_exit(void)
4055 ida_destroy(&nvme_subsystems_ida);
4056 class_destroy(nvme_subsys_class);
4057 class_destroy(nvme_class);
4058 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4059 destroy_workqueue(nvme_delete_wq);
4060 destroy_workqueue(nvme_reset_wq);
4061 destroy_workqueue(nvme_wq);
4064 MODULE_LICENSE("GPL");
4065 MODULE_VERSION("1.0");
4066 module_init(nvme_core_init);
4067 module_exit(nvme_core_exit);