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)
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, page_shift = 12;
1981 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
1983 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1986 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
1988 if (page_shift < dev_page_min) {
1989 dev_err(ctrl->device,
1990 "Minimum device page size %u too large for host (%u)\n",
1991 1 << dev_page_min, 1 << page_shift);
1995 ctrl->page_size = 1 << page_shift;
1997 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1998 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1999 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2000 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2001 ctrl->ctrl_config |= NVME_CC_ENABLE;
2003 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2006 return nvme_wait_ready(ctrl, ctrl->cap, true);
2008 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2010 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2012 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2016 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2017 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2019 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2023 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2024 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2028 if (fatal_signal_pending(current))
2030 if (time_after(jiffies, timeout)) {
2031 dev_err(ctrl->device,
2032 "Device shutdown incomplete; abort shutdown\n");
2039 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2041 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2042 struct request_queue *q)
2046 if (ctrl->max_hw_sectors) {
2048 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2050 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2051 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2052 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2054 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2055 is_power_of_2(ctrl->max_hw_sectors))
2056 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2057 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2058 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2060 blk_queue_write_cache(q, vwc, vwc);
2063 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2068 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2071 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2072 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2075 dev_warn_once(ctrl->device,
2076 "could not set timestamp (%d)\n", ret);
2080 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2082 struct nvme_feat_host_behavior *host;
2085 /* Don't bother enabling the feature if retry delay is not reported */
2089 host = kzalloc(sizeof(*host), GFP_KERNEL);
2093 host->acre = NVME_ENABLE_ACRE;
2094 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2095 host, sizeof(*host), NULL);
2100 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2103 * APST (Autonomous Power State Transition) lets us program a
2104 * table of power state transitions that the controller will
2105 * perform automatically. We configure it with a simple
2106 * heuristic: we are willing to spend at most 2% of the time
2107 * transitioning between power states. Therefore, when running
2108 * in any given state, we will enter the next lower-power
2109 * non-operational state after waiting 50 * (enlat + exlat)
2110 * microseconds, as long as that state's exit latency is under
2111 * the requested maximum latency.
2113 * We will not autonomously enter any non-operational state for
2114 * which the total latency exceeds ps_max_latency_us. Users
2115 * can set ps_max_latency_us to zero to turn off APST.
2119 struct nvme_feat_auto_pst *table;
2125 * If APST isn't supported or if we haven't been initialized yet,
2126 * then don't do anything.
2131 if (ctrl->npss > 31) {
2132 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2136 table = kzalloc(sizeof(*table), GFP_KERNEL);
2140 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2141 /* Turn off APST. */
2143 dev_dbg(ctrl->device, "APST disabled\n");
2145 __le64 target = cpu_to_le64(0);
2149 * Walk through all states from lowest- to highest-power.
2150 * According to the spec, lower-numbered states use more
2151 * power. NPSS, despite the name, is the index of the
2152 * lowest-power state, not the number of states.
2154 for (state = (int)ctrl->npss; state >= 0; state--) {
2155 u64 total_latency_us, exit_latency_us, transition_ms;
2158 table->entries[state] = target;
2161 * Don't allow transitions to the deepest state
2162 * if it's quirked off.
2164 if (state == ctrl->npss &&
2165 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2169 * Is this state a useful non-operational state for
2170 * higher-power states to autonomously transition to?
2172 if (!(ctrl->psd[state].flags &
2173 NVME_PS_FLAGS_NON_OP_STATE))
2177 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2178 if (exit_latency_us > ctrl->ps_max_latency_us)
2183 le32_to_cpu(ctrl->psd[state].entry_lat);
2186 * This state is good. Use it as the APST idle
2187 * target for higher power states.
2189 transition_ms = total_latency_us + 19;
2190 do_div(transition_ms, 20);
2191 if (transition_ms > (1 << 24) - 1)
2192 transition_ms = (1 << 24) - 1;
2194 target = cpu_to_le64((state << 3) |
2195 (transition_ms << 8));
2200 if (total_latency_us > max_lat_us)
2201 max_lat_us = total_latency_us;
2207 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2209 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2210 max_ps, max_lat_us, (int)sizeof(*table), table);
2214 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2215 table, sizeof(*table), NULL);
2217 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2223 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2225 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2229 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2230 case PM_QOS_LATENCY_ANY:
2238 if (ctrl->ps_max_latency_us != latency) {
2239 ctrl->ps_max_latency_us = latency;
2240 nvme_configure_apst(ctrl);
2244 struct nvme_core_quirk_entry {
2246 * NVMe model and firmware strings are padded with spaces. For
2247 * simplicity, strings in the quirk table are padded with NULLs
2253 unsigned long quirks;
2256 static const struct nvme_core_quirk_entry core_quirks[] = {
2259 * This Toshiba device seems to die using any APST states. See:
2260 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2263 .mn = "THNSF5256GPUK TOSHIBA",
2264 .quirks = NVME_QUIRK_NO_APST,
2268 /* match is null-terminated but idstr is space-padded. */
2269 static bool string_matches(const char *idstr, const char *match, size_t len)
2276 matchlen = strlen(match);
2277 WARN_ON_ONCE(matchlen > len);
2279 if (memcmp(idstr, match, matchlen))
2282 for (; matchlen < len; matchlen++)
2283 if (idstr[matchlen] != ' ')
2289 static bool quirk_matches(const struct nvme_id_ctrl *id,
2290 const struct nvme_core_quirk_entry *q)
2292 return q->vid == le16_to_cpu(id->vid) &&
2293 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2294 string_matches(id->fr, q->fr, sizeof(id->fr));
2297 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2298 struct nvme_id_ctrl *id)
2303 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2304 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2305 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2306 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2310 if (ctrl->vs >= NVME_VS(1, 2, 1))
2311 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2314 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2315 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2316 "nqn.2014.08.org.nvmexpress:%04x%04x",
2317 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2318 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2319 off += sizeof(id->sn);
2320 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2321 off += sizeof(id->mn);
2322 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2325 static void nvme_release_subsystem(struct device *dev)
2327 struct nvme_subsystem *subsys =
2328 container_of(dev, struct nvme_subsystem, dev);
2330 ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2334 static void nvme_destroy_subsystem(struct kref *ref)
2336 struct nvme_subsystem *subsys =
2337 container_of(ref, struct nvme_subsystem, ref);
2339 mutex_lock(&nvme_subsystems_lock);
2340 list_del(&subsys->entry);
2341 mutex_unlock(&nvme_subsystems_lock);
2343 ida_destroy(&subsys->ns_ida);
2344 device_del(&subsys->dev);
2345 put_device(&subsys->dev);
2348 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2350 kref_put(&subsys->ref, nvme_destroy_subsystem);
2353 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2355 struct nvme_subsystem *subsys;
2357 lockdep_assert_held(&nvme_subsystems_lock);
2359 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2360 if (strcmp(subsys->subnqn, subsysnqn))
2362 if (!kref_get_unless_zero(&subsys->ref))
2370 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2371 struct device_attribute subsys_attr_##_name = \
2372 __ATTR(_name, _mode, _show, NULL)
2374 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2375 struct device_attribute *attr,
2378 struct nvme_subsystem *subsys =
2379 container_of(dev, struct nvme_subsystem, dev);
2381 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2383 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2385 #define nvme_subsys_show_str_function(field) \
2386 static ssize_t subsys_##field##_show(struct device *dev, \
2387 struct device_attribute *attr, char *buf) \
2389 struct nvme_subsystem *subsys = \
2390 container_of(dev, struct nvme_subsystem, dev); \
2391 return sprintf(buf, "%.*s\n", \
2392 (int)sizeof(subsys->field), subsys->field); \
2394 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2396 nvme_subsys_show_str_function(model);
2397 nvme_subsys_show_str_function(serial);
2398 nvme_subsys_show_str_function(firmware_rev);
2400 static struct attribute *nvme_subsys_attrs[] = {
2401 &subsys_attr_model.attr,
2402 &subsys_attr_serial.attr,
2403 &subsys_attr_firmware_rev.attr,
2404 &subsys_attr_subsysnqn.attr,
2405 #ifdef CONFIG_NVME_MULTIPATH
2406 &subsys_attr_iopolicy.attr,
2411 static struct attribute_group nvme_subsys_attrs_group = {
2412 .attrs = nvme_subsys_attrs,
2415 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2416 &nvme_subsys_attrs_group,
2420 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2421 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2423 struct nvme_ctrl *tmp;
2425 lockdep_assert_held(&nvme_subsystems_lock);
2427 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2428 if (tmp->state == NVME_CTRL_DELETING ||
2429 tmp->state == NVME_CTRL_DEAD)
2432 if (tmp->cntlid == ctrl->cntlid) {
2433 dev_err(ctrl->device,
2434 "Duplicate cntlid %u with %s, rejecting\n",
2435 ctrl->cntlid, dev_name(tmp->device));
2439 if ((id->cmic & (1 << 1)) ||
2440 (ctrl->opts && ctrl->opts->discovery_nqn))
2443 dev_err(ctrl->device,
2444 "Subsystem does not support multiple controllers\n");
2451 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2453 struct nvme_subsystem *subsys, *found;
2456 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2459 ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2464 subsys->instance = ret;
2465 mutex_init(&subsys->lock);
2466 kref_init(&subsys->ref);
2467 INIT_LIST_HEAD(&subsys->ctrls);
2468 INIT_LIST_HEAD(&subsys->nsheads);
2469 nvme_init_subnqn(subsys, ctrl, id);
2470 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2471 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2472 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2473 subsys->vendor_id = le16_to_cpu(id->vid);
2474 subsys->cmic = id->cmic;
2475 subsys->awupf = le16_to_cpu(id->awupf);
2476 #ifdef CONFIG_NVME_MULTIPATH
2477 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2480 subsys->dev.class = nvme_subsys_class;
2481 subsys->dev.release = nvme_release_subsystem;
2482 subsys->dev.groups = nvme_subsys_attrs_groups;
2483 dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2484 device_initialize(&subsys->dev);
2486 mutex_lock(&nvme_subsystems_lock);
2487 found = __nvme_find_get_subsystem(subsys->subnqn);
2489 put_device(&subsys->dev);
2492 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2494 goto out_put_subsystem;
2497 ret = device_add(&subsys->dev);
2499 dev_err(ctrl->device,
2500 "failed to register subsystem device.\n");
2503 ida_init(&subsys->ns_ida);
2504 list_add_tail(&subsys->entry, &nvme_subsystems);
2507 if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2508 dev_name(ctrl->device))) {
2509 dev_err(ctrl->device,
2510 "failed to create sysfs link from subsystem.\n");
2511 goto out_put_subsystem;
2514 ctrl->subsys = subsys;
2515 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2516 mutex_unlock(&nvme_subsystems_lock);
2520 nvme_put_subsystem(subsys);
2522 mutex_unlock(&nvme_subsystems_lock);
2523 put_device(&subsys->dev);
2527 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2528 void *log, size_t size, u64 offset)
2530 struct nvme_command c = { };
2531 unsigned long dwlen = size / 4 - 1;
2533 c.get_log_page.opcode = nvme_admin_get_log_page;
2534 c.get_log_page.nsid = cpu_to_le32(nsid);
2535 c.get_log_page.lid = log_page;
2536 c.get_log_page.lsp = lsp;
2537 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2538 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2539 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2540 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2542 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2545 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2550 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2555 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2556 ctrl->effects, sizeof(*ctrl->effects), 0);
2558 kfree(ctrl->effects);
2559 ctrl->effects = NULL;
2565 * Initialize the cached copies of the Identify data and various controller
2566 * register in our nvme_ctrl structure. This should be called as soon as
2567 * the admin queue is fully up and running.
2569 int nvme_init_identify(struct nvme_ctrl *ctrl)
2571 struct nvme_id_ctrl *id;
2572 int ret, page_shift;
2574 bool prev_apst_enabled;
2576 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2578 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2581 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2582 ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2584 if (ctrl->vs >= NVME_VS(1, 1, 0))
2585 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2587 ret = nvme_identify_ctrl(ctrl, &id);
2589 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2593 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2594 ret = nvme_get_effects_log(ctrl);
2599 if (!ctrl->identified) {
2602 ret = nvme_init_subsystem(ctrl, id);
2607 * Check for quirks. Quirk can depend on firmware version,
2608 * so, in principle, the set of quirks present can change
2609 * across a reset. As a possible future enhancement, we
2610 * could re-scan for quirks every time we reinitialize
2611 * the device, but we'd have to make sure that the driver
2612 * behaves intelligently if the quirks change.
2614 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2615 if (quirk_matches(id, &core_quirks[i]))
2616 ctrl->quirks |= core_quirks[i].quirks;
2620 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2621 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2622 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2625 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2626 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2627 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2629 ctrl->oacs = le16_to_cpu(id->oacs);
2630 ctrl->oncs = le16_to_cpu(id->oncs);
2631 ctrl->mtfa = le16_to_cpu(id->mtfa);
2632 ctrl->oaes = le32_to_cpu(id->oaes);
2633 atomic_set(&ctrl->abort_limit, id->acl + 1);
2634 ctrl->vwc = id->vwc;
2636 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2638 max_hw_sectors = UINT_MAX;
2639 ctrl->max_hw_sectors =
2640 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2642 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2643 ctrl->sgls = le32_to_cpu(id->sgls);
2644 ctrl->kas = le16_to_cpu(id->kas);
2645 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2646 ctrl->ctratt = le32_to_cpu(id->ctratt);
2650 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2652 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2653 shutdown_timeout, 60);
2655 if (ctrl->shutdown_timeout != shutdown_timeout)
2656 dev_info(ctrl->device,
2657 "Shutdown timeout set to %u seconds\n",
2658 ctrl->shutdown_timeout);
2660 ctrl->shutdown_timeout = shutdown_timeout;
2662 ctrl->npss = id->npss;
2663 ctrl->apsta = id->apsta;
2664 prev_apst_enabled = ctrl->apst_enabled;
2665 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2666 if (force_apst && id->apsta) {
2667 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2668 ctrl->apst_enabled = true;
2670 ctrl->apst_enabled = false;
2673 ctrl->apst_enabled = id->apsta;
2675 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2677 if (ctrl->ops->flags & NVME_F_FABRICS) {
2678 ctrl->icdoff = le16_to_cpu(id->icdoff);
2679 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2680 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2681 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2684 * In fabrics we need to verify the cntlid matches the
2687 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2692 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2693 dev_err(ctrl->device,
2694 "keep-alive support is mandatory for fabrics\n");
2699 ctrl->cntlid = le16_to_cpu(id->cntlid);
2700 ctrl->hmpre = le32_to_cpu(id->hmpre);
2701 ctrl->hmmin = le32_to_cpu(id->hmmin);
2702 ctrl->hmminds = le32_to_cpu(id->hmminds);
2703 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2706 ret = nvme_mpath_init(ctrl, id);
2712 if (ctrl->apst_enabled && !prev_apst_enabled)
2713 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2714 else if (!ctrl->apst_enabled && prev_apst_enabled)
2715 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2717 ret = nvme_configure_apst(ctrl);
2721 ret = nvme_configure_timestamp(ctrl);
2725 ret = nvme_configure_directives(ctrl);
2729 ret = nvme_configure_acre(ctrl);
2733 ctrl->identified = true;
2741 EXPORT_SYMBOL_GPL(nvme_init_identify);
2743 static int nvme_dev_open(struct inode *inode, struct file *file)
2745 struct nvme_ctrl *ctrl =
2746 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2748 switch (ctrl->state) {
2749 case NVME_CTRL_LIVE:
2750 case NVME_CTRL_ADMIN_ONLY:
2753 return -EWOULDBLOCK;
2756 file->private_data = ctrl;
2760 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2765 down_read(&ctrl->namespaces_rwsem);
2766 if (list_empty(&ctrl->namespaces)) {
2771 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2772 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2773 dev_warn(ctrl->device,
2774 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2779 dev_warn(ctrl->device,
2780 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2781 kref_get(&ns->kref);
2782 up_read(&ctrl->namespaces_rwsem);
2784 ret = nvme_user_cmd(ctrl, ns, argp);
2789 up_read(&ctrl->namespaces_rwsem);
2793 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2796 struct nvme_ctrl *ctrl = file->private_data;
2797 void __user *argp = (void __user *)arg;
2800 case NVME_IOCTL_ADMIN_CMD:
2801 return nvme_user_cmd(ctrl, NULL, argp);
2802 case NVME_IOCTL_IO_CMD:
2803 return nvme_dev_user_cmd(ctrl, argp);
2804 case NVME_IOCTL_RESET:
2805 dev_warn(ctrl->device, "resetting controller\n");
2806 return nvme_reset_ctrl_sync(ctrl);
2807 case NVME_IOCTL_SUBSYS_RESET:
2808 return nvme_reset_subsystem(ctrl);
2809 case NVME_IOCTL_RESCAN:
2810 nvme_queue_scan(ctrl);
2817 static const struct file_operations nvme_dev_fops = {
2818 .owner = THIS_MODULE,
2819 .open = nvme_dev_open,
2820 .unlocked_ioctl = nvme_dev_ioctl,
2821 .compat_ioctl = nvme_dev_ioctl,
2824 static ssize_t nvme_sysfs_reset(struct device *dev,
2825 struct device_attribute *attr, const char *buf,
2828 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2831 ret = nvme_reset_ctrl_sync(ctrl);
2836 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2838 static ssize_t nvme_sysfs_rescan(struct device *dev,
2839 struct device_attribute *attr, const char *buf,
2842 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2844 nvme_queue_scan(ctrl);
2847 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2849 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2851 struct gendisk *disk = dev_to_disk(dev);
2853 if (disk->fops == &nvme_fops)
2854 return nvme_get_ns_from_dev(dev)->head;
2856 return disk->private_data;
2859 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2862 struct nvme_ns_head *head = dev_to_ns_head(dev);
2863 struct nvme_ns_ids *ids = &head->ids;
2864 struct nvme_subsystem *subsys = head->subsys;
2865 int serial_len = sizeof(subsys->serial);
2866 int model_len = sizeof(subsys->model);
2868 if (!uuid_is_null(&ids->uuid))
2869 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2871 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2872 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2874 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2875 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2877 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2878 subsys->serial[serial_len - 1] == '\0'))
2880 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2881 subsys->model[model_len - 1] == '\0'))
2884 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2885 serial_len, subsys->serial, model_len, subsys->model,
2888 static DEVICE_ATTR_RO(wwid);
2890 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2893 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2895 static DEVICE_ATTR_RO(nguid);
2897 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2900 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2902 /* For backward compatibility expose the NGUID to userspace if
2903 * we have no UUID set
2905 if (uuid_is_null(&ids->uuid)) {
2906 printk_ratelimited(KERN_WARNING
2907 "No UUID available providing old NGUID\n");
2908 return sprintf(buf, "%pU\n", ids->nguid);
2910 return sprintf(buf, "%pU\n", &ids->uuid);
2912 static DEVICE_ATTR_RO(uuid);
2914 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2917 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2919 static DEVICE_ATTR_RO(eui);
2921 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2924 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2926 static DEVICE_ATTR_RO(nsid);
2928 static struct attribute *nvme_ns_id_attrs[] = {
2929 &dev_attr_wwid.attr,
2930 &dev_attr_uuid.attr,
2931 &dev_attr_nguid.attr,
2933 &dev_attr_nsid.attr,
2934 #ifdef CONFIG_NVME_MULTIPATH
2935 &dev_attr_ana_grpid.attr,
2936 &dev_attr_ana_state.attr,
2941 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2942 struct attribute *a, int n)
2944 struct device *dev = container_of(kobj, struct device, kobj);
2945 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2947 if (a == &dev_attr_uuid.attr) {
2948 if (uuid_is_null(&ids->uuid) &&
2949 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2952 if (a == &dev_attr_nguid.attr) {
2953 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2956 if (a == &dev_attr_eui.attr) {
2957 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2960 #ifdef CONFIG_NVME_MULTIPATH
2961 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2962 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2964 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2971 static const struct attribute_group nvme_ns_id_attr_group = {
2972 .attrs = nvme_ns_id_attrs,
2973 .is_visible = nvme_ns_id_attrs_are_visible,
2976 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2977 &nvme_ns_id_attr_group,
2979 &nvme_nvm_attr_group,
2984 #define nvme_show_str_function(field) \
2985 static ssize_t field##_show(struct device *dev, \
2986 struct device_attribute *attr, char *buf) \
2988 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2989 return sprintf(buf, "%.*s\n", \
2990 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
2992 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2994 nvme_show_str_function(model);
2995 nvme_show_str_function(serial);
2996 nvme_show_str_function(firmware_rev);
2998 #define nvme_show_int_function(field) \
2999 static ssize_t field##_show(struct device *dev, \
3000 struct device_attribute *attr, char *buf) \
3002 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3003 return sprintf(buf, "%d\n", ctrl->field); \
3005 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3007 nvme_show_int_function(cntlid);
3008 nvme_show_int_function(numa_node);
3010 static ssize_t nvme_sysfs_delete(struct device *dev,
3011 struct device_attribute *attr, const char *buf,
3014 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3016 if (device_remove_file_self(dev, attr))
3017 nvme_delete_ctrl_sync(ctrl);
3020 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3022 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3023 struct device_attribute *attr,
3026 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3028 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3030 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3032 static ssize_t nvme_sysfs_show_state(struct device *dev,
3033 struct device_attribute *attr,
3036 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3037 static const char *const state_name[] = {
3038 [NVME_CTRL_NEW] = "new",
3039 [NVME_CTRL_LIVE] = "live",
3040 [NVME_CTRL_ADMIN_ONLY] = "only-admin",
3041 [NVME_CTRL_RESETTING] = "resetting",
3042 [NVME_CTRL_CONNECTING] = "connecting",
3043 [NVME_CTRL_DELETING] = "deleting",
3044 [NVME_CTRL_DEAD] = "dead",
3047 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3048 state_name[ctrl->state])
3049 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3051 return sprintf(buf, "unknown state\n");
3054 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3056 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3057 struct device_attribute *attr,
3060 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3062 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3064 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3066 static ssize_t nvme_sysfs_show_address(struct device *dev,
3067 struct device_attribute *attr,
3070 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3072 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3074 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3076 static struct attribute *nvme_dev_attrs[] = {
3077 &dev_attr_reset_controller.attr,
3078 &dev_attr_rescan_controller.attr,
3079 &dev_attr_model.attr,
3080 &dev_attr_serial.attr,
3081 &dev_attr_firmware_rev.attr,
3082 &dev_attr_cntlid.attr,
3083 &dev_attr_delete_controller.attr,
3084 &dev_attr_transport.attr,
3085 &dev_attr_subsysnqn.attr,
3086 &dev_attr_address.attr,
3087 &dev_attr_state.attr,
3088 &dev_attr_numa_node.attr,
3092 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3093 struct attribute *a, int n)
3095 struct device *dev = container_of(kobj, struct device, kobj);
3096 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3098 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3100 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3106 static struct attribute_group nvme_dev_attrs_group = {
3107 .attrs = nvme_dev_attrs,
3108 .is_visible = nvme_dev_attrs_are_visible,
3111 static const struct attribute_group *nvme_dev_attr_groups[] = {
3112 &nvme_dev_attrs_group,
3116 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3119 struct nvme_ns_head *h;
3121 lockdep_assert_held(&subsys->lock);
3123 list_for_each_entry(h, &subsys->nsheads, entry) {
3124 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3131 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3132 struct nvme_ns_head *new)
3134 struct nvme_ns_head *h;
3136 lockdep_assert_held(&subsys->lock);
3138 list_for_each_entry(h, &subsys->nsheads, entry) {
3139 if (nvme_ns_ids_valid(&new->ids) &&
3140 !list_empty(&h->list) &&
3141 nvme_ns_ids_equal(&new->ids, &h->ids))
3148 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3149 unsigned nsid, struct nvme_id_ns *id)
3151 struct nvme_ns_head *head;
3152 size_t size = sizeof(*head);
3155 #ifdef CONFIG_NVME_MULTIPATH
3156 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3159 head = kzalloc(size, GFP_KERNEL);
3162 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3165 head->instance = ret;
3166 INIT_LIST_HEAD(&head->list);
3167 ret = init_srcu_struct(&head->srcu);
3169 goto out_ida_remove;
3170 head->subsys = ctrl->subsys;
3172 kref_init(&head->ref);
3174 nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3176 ret = __nvme_check_ids(ctrl->subsys, head);
3178 dev_err(ctrl->device,
3179 "duplicate IDs for nsid %d\n", nsid);
3180 goto out_cleanup_srcu;
3183 ret = nvme_mpath_alloc_disk(ctrl, head);
3185 goto out_cleanup_srcu;
3187 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3189 kref_get(&ctrl->subsys->ref);
3193 cleanup_srcu_struct(&head->srcu);
3195 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3199 return ERR_PTR(ret);
3202 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3203 struct nvme_id_ns *id)
3205 struct nvme_ctrl *ctrl = ns->ctrl;
3206 bool is_shared = id->nmic & (1 << 0);
3207 struct nvme_ns_head *head = NULL;
3210 mutex_lock(&ctrl->subsys->lock);
3212 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3214 head = nvme_alloc_ns_head(ctrl, nsid, id);
3216 ret = PTR_ERR(head);
3220 struct nvme_ns_ids ids;
3222 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3223 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3224 dev_err(ctrl->device,
3225 "IDs don't match for shared namespace %d\n",
3232 list_add_tail(&ns->siblings, &head->list);
3236 mutex_unlock(&ctrl->subsys->lock);
3240 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3242 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3243 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3245 return nsa->head->ns_id - nsb->head->ns_id;
3248 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3250 struct nvme_ns *ns, *ret = NULL;
3252 down_read(&ctrl->namespaces_rwsem);
3253 list_for_each_entry(ns, &ctrl->namespaces, list) {
3254 if (ns->head->ns_id == nsid) {
3255 if (!kref_get_unless_zero(&ns->kref))
3260 if (ns->head->ns_id > nsid)
3263 up_read(&ctrl->namespaces_rwsem);
3267 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3269 struct streams_directive_params s;
3272 if (!ctrl->nr_streams)
3275 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3279 ns->sws = le32_to_cpu(s.sws);
3280 ns->sgs = le16_to_cpu(s.sgs);
3283 unsigned int bs = 1 << ns->lba_shift;
3285 blk_queue_io_min(ns->queue, bs * ns->sws);
3287 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3293 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3296 struct gendisk *disk;
3297 struct nvme_id_ns *id;
3298 char disk_name[DISK_NAME_LEN];
3299 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3301 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3305 ns->queue = blk_mq_init_queue(ctrl->tagset);
3306 if (IS_ERR(ns->queue)) {
3307 ret = PTR_ERR(ns->queue);
3311 if (ctrl->opts && ctrl->opts->data_digest)
3312 ns->queue->backing_dev_info->capabilities
3313 |= BDI_CAP_STABLE_WRITES;
3315 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3316 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3317 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3319 ns->queue->queuedata = ns;
3322 kref_init(&ns->kref);
3323 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3325 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3326 nvme_set_queue_limits(ctrl, ns->queue);
3328 id = nvme_identify_ns(ctrl, nsid);
3331 goto out_free_queue;
3334 if (id->ncap == 0) {
3339 ret = nvme_init_ns_head(ns, nsid, id);
3342 nvme_setup_streams_ns(ctrl, ns);
3343 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3345 disk = alloc_disk_node(0, node);
3351 disk->fops = &nvme_fops;
3352 disk->private_data = ns;
3353 disk->queue = ns->queue;
3354 disk->flags = flags;
3355 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3358 __nvme_revalidate_disk(disk, id);
3360 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3361 ret = nvme_nvm_register(ns, disk_name, node);
3363 dev_warn(ctrl->device, "LightNVM init failure\n");
3368 down_write(&ctrl->namespaces_rwsem);
3369 list_add_tail(&ns->list, &ctrl->namespaces);
3370 up_write(&ctrl->namespaces_rwsem);
3372 nvme_get_ctrl(ctrl);
3374 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3376 nvme_mpath_add_disk(ns, id);
3377 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3384 mutex_lock(&ctrl->subsys->lock);
3385 list_del_rcu(&ns->siblings);
3386 mutex_unlock(&ctrl->subsys->lock);
3387 nvme_put_ns_head(ns->head);
3391 blk_cleanup_queue(ns->queue);
3397 static void nvme_ns_remove(struct nvme_ns *ns)
3399 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3402 nvme_fault_inject_fini(&ns->fault_inject);
3404 mutex_lock(&ns->ctrl->subsys->lock);
3405 list_del_rcu(&ns->siblings);
3406 mutex_unlock(&ns->ctrl->subsys->lock);
3407 synchronize_rcu(); /* guarantee not available in head->list */
3408 nvme_mpath_clear_current_path(ns);
3409 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3411 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3412 del_gendisk(ns->disk);
3413 blk_cleanup_queue(ns->queue);
3414 if (blk_get_integrity(ns->disk))
3415 blk_integrity_unregister(ns->disk);
3418 down_write(&ns->ctrl->namespaces_rwsem);
3419 list_del_init(&ns->list);
3420 up_write(&ns->ctrl->namespaces_rwsem);
3422 nvme_mpath_check_last_path(ns);
3426 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3430 ns = nvme_find_get_ns(ctrl, nsid);
3432 if (ns->disk && revalidate_disk(ns->disk))
3436 nvme_alloc_ns(ctrl, nsid);
3439 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3442 struct nvme_ns *ns, *next;
3445 down_write(&ctrl->namespaces_rwsem);
3446 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3447 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3448 list_move_tail(&ns->list, &rm_list);
3450 up_write(&ctrl->namespaces_rwsem);
3452 list_for_each_entry_safe(ns, next, &rm_list, list)
3457 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3461 unsigned i, j, nsid, prev = 0;
3462 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3465 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3469 for (i = 0; i < num_lists; i++) {
3470 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3474 for (j = 0; j < min(nn, 1024U); j++) {
3475 nsid = le32_to_cpu(ns_list[j]);
3479 nvme_validate_ns(ctrl, nsid);
3481 while (++prev < nsid) {
3482 ns = nvme_find_get_ns(ctrl, prev);
3492 nvme_remove_invalid_namespaces(ctrl, prev);
3498 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3502 for (i = 1; i <= nn; i++)
3503 nvme_validate_ns(ctrl, i);
3505 nvme_remove_invalid_namespaces(ctrl, nn);
3508 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3510 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3514 log = kzalloc(log_size, GFP_KERNEL);
3519 * We need to read the log to clear the AEN, but we don't want to rely
3520 * on it for the changed namespace information as userspace could have
3521 * raced with us in reading the log page, which could cause us to miss
3524 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3527 dev_warn(ctrl->device,
3528 "reading changed ns log failed: %d\n", error);
3533 static void nvme_scan_work(struct work_struct *work)
3535 struct nvme_ctrl *ctrl =
3536 container_of(work, struct nvme_ctrl, scan_work);
3537 struct nvme_id_ctrl *id;
3540 if (ctrl->state != NVME_CTRL_LIVE)
3543 WARN_ON_ONCE(!ctrl->tagset);
3545 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3546 dev_info(ctrl->device, "rescanning namespaces.\n");
3547 nvme_clear_changed_ns_log(ctrl);
3550 if (nvme_identify_ctrl(ctrl, &id))
3553 mutex_lock(&ctrl->scan_lock);
3554 nn = le32_to_cpu(id->nn);
3555 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3556 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3557 if (!nvme_scan_ns_list(ctrl, nn))
3560 nvme_scan_ns_sequential(ctrl, nn);
3562 mutex_unlock(&ctrl->scan_lock);
3564 down_write(&ctrl->namespaces_rwsem);
3565 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3566 up_write(&ctrl->namespaces_rwsem);
3570 * This function iterates the namespace list unlocked to allow recovery from
3571 * controller failure. It is up to the caller to ensure the namespace list is
3572 * not modified by scan work while this function is executing.
3574 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3576 struct nvme_ns *ns, *next;
3579 /* prevent racing with ns scanning */
3580 flush_work(&ctrl->scan_work);
3583 * The dead states indicates the controller was not gracefully
3584 * disconnected. In that case, we won't be able to flush any data while
3585 * removing the namespaces' disks; fail all the queues now to avoid
3586 * potentially having to clean up the failed sync later.
3588 if (ctrl->state == NVME_CTRL_DEAD)
3589 nvme_kill_queues(ctrl);
3591 down_write(&ctrl->namespaces_rwsem);
3592 list_splice_init(&ctrl->namespaces, &ns_list);
3593 up_write(&ctrl->namespaces_rwsem);
3595 list_for_each_entry_safe(ns, next, &ns_list, list)
3598 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3600 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3602 char *envp[2] = { NULL, NULL };
3603 u32 aen_result = ctrl->aen_result;
3605 ctrl->aen_result = 0;
3609 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3612 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3616 static void nvme_async_event_work(struct work_struct *work)
3618 struct nvme_ctrl *ctrl =
3619 container_of(work, struct nvme_ctrl, async_event_work);
3621 nvme_aen_uevent(ctrl);
3622 ctrl->ops->submit_async_event(ctrl);
3625 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3630 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3636 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3639 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3641 struct nvme_fw_slot_info_log *log;
3643 log = kmalloc(sizeof(*log), GFP_KERNEL);
3647 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3649 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3653 static void nvme_fw_act_work(struct work_struct *work)
3655 struct nvme_ctrl *ctrl = container_of(work,
3656 struct nvme_ctrl, fw_act_work);
3657 unsigned long fw_act_timeout;
3660 fw_act_timeout = jiffies +
3661 msecs_to_jiffies(ctrl->mtfa * 100);
3663 fw_act_timeout = jiffies +
3664 msecs_to_jiffies(admin_timeout * 1000);
3666 nvme_stop_queues(ctrl);
3667 while (nvme_ctrl_pp_status(ctrl)) {
3668 if (time_after(jiffies, fw_act_timeout)) {
3669 dev_warn(ctrl->device,
3670 "Fw activation timeout, reset controller\n");
3671 nvme_reset_ctrl(ctrl);
3677 if (ctrl->state != NVME_CTRL_LIVE)
3680 nvme_start_queues(ctrl);
3681 /* read FW slot information to clear the AER */
3682 nvme_get_fw_slot_info(ctrl);
3685 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3687 u32 aer_notice_type = (result & 0xff00) >> 8;
3689 trace_nvme_async_event(ctrl, aer_notice_type);
3691 switch (aer_notice_type) {
3692 case NVME_AER_NOTICE_NS_CHANGED:
3693 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3694 nvme_queue_scan(ctrl);
3696 case NVME_AER_NOTICE_FW_ACT_STARTING:
3697 queue_work(nvme_wq, &ctrl->fw_act_work);
3699 #ifdef CONFIG_NVME_MULTIPATH
3700 case NVME_AER_NOTICE_ANA:
3701 if (!ctrl->ana_log_buf)
3703 queue_work(nvme_wq, &ctrl->ana_work);
3707 dev_warn(ctrl->device, "async event result %08x\n", result);
3711 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3712 volatile union nvme_result *res)
3714 u32 result = le32_to_cpu(res->u32);
3715 u32 aer_type = result & 0x07;
3717 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3721 case NVME_AER_NOTICE:
3722 nvme_handle_aen_notice(ctrl, result);
3724 case NVME_AER_ERROR:
3725 case NVME_AER_SMART:
3728 trace_nvme_async_event(ctrl, aer_type);
3729 ctrl->aen_result = result;
3734 queue_work(nvme_wq, &ctrl->async_event_work);
3736 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3738 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3740 nvme_mpath_stop(ctrl);
3741 nvme_stop_keep_alive(ctrl);
3742 flush_work(&ctrl->async_event_work);
3743 cancel_work_sync(&ctrl->fw_act_work);
3745 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3747 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3750 nvme_start_keep_alive(ctrl);
3752 if (ctrl->queue_count > 1) {
3753 nvme_queue_scan(ctrl);
3754 nvme_enable_aen(ctrl);
3755 queue_work(nvme_wq, &ctrl->async_event_work);
3756 nvme_start_queues(ctrl);
3759 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3761 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3763 nvme_fault_inject_fini(&ctrl->fault_inject);
3764 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3765 cdev_device_del(&ctrl->cdev, ctrl->device);
3767 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3769 static void nvme_free_ctrl(struct device *dev)
3771 struct nvme_ctrl *ctrl =
3772 container_of(dev, struct nvme_ctrl, ctrl_device);
3773 struct nvme_subsystem *subsys = ctrl->subsys;
3775 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3776 kfree(ctrl->effects);
3777 nvme_mpath_uninit(ctrl);
3778 __free_page(ctrl->discard_page);
3781 mutex_lock(&nvme_subsystems_lock);
3782 list_del(&ctrl->subsys_entry);
3783 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3784 mutex_unlock(&nvme_subsystems_lock);
3787 ctrl->ops->free_ctrl(ctrl);
3790 nvme_put_subsystem(subsys);
3794 * Initialize a NVMe controller structures. This needs to be called during
3795 * earliest initialization so that we have the initialized structured around
3798 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3799 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3803 ctrl->state = NVME_CTRL_NEW;
3804 spin_lock_init(&ctrl->lock);
3805 mutex_init(&ctrl->scan_lock);
3806 INIT_LIST_HEAD(&ctrl->namespaces);
3807 init_rwsem(&ctrl->namespaces_rwsem);
3810 ctrl->quirks = quirks;
3811 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3812 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3813 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3814 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3816 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3817 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3818 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3820 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3822 ctrl->discard_page = alloc_page(GFP_KERNEL);
3823 if (!ctrl->discard_page) {
3828 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3831 ctrl->instance = ret;
3833 device_initialize(&ctrl->ctrl_device);
3834 ctrl->device = &ctrl->ctrl_device;
3835 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3836 ctrl->device->class = nvme_class;
3837 ctrl->device->parent = ctrl->dev;
3838 ctrl->device->groups = nvme_dev_attr_groups;
3839 ctrl->device->release = nvme_free_ctrl;
3840 dev_set_drvdata(ctrl->device, ctrl);
3841 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3843 goto out_release_instance;
3845 cdev_init(&ctrl->cdev, &nvme_dev_fops);
3846 ctrl->cdev.owner = ops->module;
3847 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3852 * Initialize latency tolerance controls. The sysfs files won't
3853 * be visible to userspace unless the device actually supports APST.
3855 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3856 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3857 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3859 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
3863 kfree_const(ctrl->device->kobj.name);
3864 out_release_instance:
3865 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3867 if (ctrl->discard_page)
3868 __free_page(ctrl->discard_page);
3871 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3874 * nvme_kill_queues(): Ends all namespace queues
3875 * @ctrl: the dead controller that needs to end
3877 * Call this function when the driver determines it is unable to get the
3878 * controller in a state capable of servicing IO.
3880 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3884 down_read(&ctrl->namespaces_rwsem);
3886 /* Forcibly unquiesce queues to avoid blocking dispatch */
3887 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3888 blk_mq_unquiesce_queue(ctrl->admin_q);
3890 list_for_each_entry(ns, &ctrl->namespaces, list)
3891 nvme_set_queue_dying(ns);
3893 up_read(&ctrl->namespaces_rwsem);
3895 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3897 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3901 down_read(&ctrl->namespaces_rwsem);
3902 list_for_each_entry(ns, &ctrl->namespaces, list)
3903 blk_mq_unfreeze_queue(ns->queue);
3904 up_read(&ctrl->namespaces_rwsem);
3906 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3908 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3912 down_read(&ctrl->namespaces_rwsem);
3913 list_for_each_entry(ns, &ctrl->namespaces, list) {
3914 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3918 up_read(&ctrl->namespaces_rwsem);
3920 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3922 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3926 down_read(&ctrl->namespaces_rwsem);
3927 list_for_each_entry(ns, &ctrl->namespaces, list)
3928 blk_mq_freeze_queue_wait(ns->queue);
3929 up_read(&ctrl->namespaces_rwsem);
3931 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3933 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3937 down_read(&ctrl->namespaces_rwsem);
3938 list_for_each_entry(ns, &ctrl->namespaces, list)
3939 blk_freeze_queue_start(ns->queue);
3940 up_read(&ctrl->namespaces_rwsem);
3942 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3944 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3948 down_read(&ctrl->namespaces_rwsem);
3949 list_for_each_entry(ns, &ctrl->namespaces, list)
3950 blk_mq_quiesce_queue(ns->queue);
3951 up_read(&ctrl->namespaces_rwsem);
3953 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3955 void nvme_start_queues(struct nvme_ctrl *ctrl)
3959 down_read(&ctrl->namespaces_rwsem);
3960 list_for_each_entry(ns, &ctrl->namespaces, list)
3961 blk_mq_unquiesce_queue(ns->queue);
3962 up_read(&ctrl->namespaces_rwsem);
3964 EXPORT_SYMBOL_GPL(nvme_start_queues);
3967 void nvme_sync_queues(struct nvme_ctrl *ctrl)
3971 down_read(&ctrl->namespaces_rwsem);
3972 list_for_each_entry(ns, &ctrl->namespaces, list)
3973 blk_sync_queue(ns->queue);
3974 up_read(&ctrl->namespaces_rwsem);
3976 EXPORT_SYMBOL_GPL(nvme_sync_queues);
3979 * Check we didn't inadvertently grow the command structure sizes:
3981 static inline void _nvme_check_size(void)
3983 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
3984 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
3985 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
3986 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
3987 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
3988 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
3989 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
3990 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
3991 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
3992 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
3993 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
3994 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
3995 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
3996 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
3997 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
3998 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
3999 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4003 static int __init nvme_core_init(void)
4005 int result = -ENOMEM;
4009 nvme_wq = alloc_workqueue("nvme-wq",
4010 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4014 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4015 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4019 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4020 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4021 if (!nvme_delete_wq)
4022 goto destroy_reset_wq;
4024 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4026 goto destroy_delete_wq;
4028 nvme_class = class_create(THIS_MODULE, "nvme");
4029 if (IS_ERR(nvme_class)) {
4030 result = PTR_ERR(nvme_class);
4031 goto unregister_chrdev;
4034 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4035 if (IS_ERR(nvme_subsys_class)) {
4036 result = PTR_ERR(nvme_subsys_class);
4042 class_destroy(nvme_class);
4044 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4046 destroy_workqueue(nvme_delete_wq);
4048 destroy_workqueue(nvme_reset_wq);
4050 destroy_workqueue(nvme_wq);
4055 static void __exit nvme_core_exit(void)
4057 ida_destroy(&nvme_subsystems_ida);
4058 class_destroy(nvme_subsys_class);
4059 class_destroy(nvme_class);
4060 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4061 destroy_workqueue(nvme_delete_wq);
4062 destroy_workqueue(nvme_reset_wq);
4063 destroy_workqueue(nvme_wq);
4066 MODULE_LICENSE("GPL");
4067 MODULE_VERSION("1.0");
4068 module_init(nvme_core_init);
4069 module_exit(nvme_core_exit);