2 * NVM Express device driver
3 * Copyright (c) 2011-2014, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 #include <linux/blkdev.h>
16 #include <linux/blk-mq.h>
17 #include <linux/delay.h>
18 #include <linux/errno.h>
19 #include <linux/hdreg.h>
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/list_sort.h>
23 #include <linux/slab.h>
24 #include <linux/types.h>
26 #include <linux/ptrace.h>
27 #include <linux/nvme_ioctl.h>
28 #include <linux/t10-pi.h>
29 #include <linux/pm_qos.h>
30 #include <asm/unaligned.h>
32 #define CREATE_TRACE_POINTS
38 #define NVME_MINORS (1U << MINORBITS)
40 unsigned int admin_timeout = 60;
41 module_param(admin_timeout, uint, 0644);
42 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
43 EXPORT_SYMBOL_GPL(admin_timeout);
45 unsigned int nvme_io_timeout = 30;
46 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
47 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
48 EXPORT_SYMBOL_GPL(nvme_io_timeout);
50 static unsigned char shutdown_timeout = 5;
51 module_param(shutdown_timeout, byte, 0644);
52 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
54 static u8 nvme_max_retries = 5;
55 module_param_named(max_retries, nvme_max_retries, byte, 0644);
56 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
58 static unsigned long default_ps_max_latency_us = 100000;
59 module_param(default_ps_max_latency_us, ulong, 0644);
60 MODULE_PARM_DESC(default_ps_max_latency_us,
61 "max power saving latency for new devices; use PM QOS to change per device");
63 static bool force_apst;
64 module_param(force_apst, bool, 0644);
65 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
68 module_param(streams, bool, 0644);
69 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
72 * nvme_wq - hosts nvme related works that are not reset or delete
73 * nvme_reset_wq - hosts nvme reset works
74 * nvme_delete_wq - hosts nvme delete works
76 * nvme_wq will host works such are scan, aen handling, fw activation,
77 * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
78 * runs reset works which also flush works hosted on nvme_wq for
79 * serialization purposes. nvme_delete_wq host controller deletion
80 * works which flush reset works for serialization.
82 struct workqueue_struct *nvme_wq;
83 EXPORT_SYMBOL_GPL(nvme_wq);
85 struct workqueue_struct *nvme_reset_wq;
86 EXPORT_SYMBOL_GPL(nvme_reset_wq);
88 struct workqueue_struct *nvme_delete_wq;
89 EXPORT_SYMBOL_GPL(nvme_delete_wq);
91 static DEFINE_IDA(nvme_subsystems_ida);
92 static LIST_HEAD(nvme_subsystems);
93 static DEFINE_MUTEX(nvme_subsystems_lock);
95 static DEFINE_IDA(nvme_instance_ida);
96 static dev_t nvme_chr_devt;
97 static struct class *nvme_class;
98 static struct class *nvme_subsys_class;
100 static void nvme_ns_remove(struct nvme_ns *ns);
101 static int nvme_revalidate_disk(struct gendisk *disk);
103 static __le32 nvme_get_log_dw10(u8 lid, size_t size)
105 return cpu_to_le32((((size / 4) - 1) << 16) | lid);
108 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
110 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
112 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
116 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
118 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
122 ret = nvme_reset_ctrl(ctrl);
124 flush_work(&ctrl->reset_work);
127 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
129 static void nvme_delete_ctrl_work(struct work_struct *work)
131 struct nvme_ctrl *ctrl =
132 container_of(work, struct nvme_ctrl, delete_work);
134 flush_work(&ctrl->reset_work);
135 nvme_stop_ctrl(ctrl);
136 nvme_remove_namespaces(ctrl);
137 ctrl->ops->delete_ctrl(ctrl);
138 nvme_uninit_ctrl(ctrl);
142 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
144 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
146 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
150 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
152 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
157 * Keep a reference until the work is flushed since ->delete_ctrl
158 * can free the controller.
161 ret = nvme_delete_ctrl(ctrl);
163 flush_work(&ctrl->delete_work);
167 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
169 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
171 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
174 static blk_status_t nvme_error_status(struct request *req)
176 switch (nvme_req(req)->status & 0x7ff) {
177 case NVME_SC_SUCCESS:
179 case NVME_SC_CAP_EXCEEDED:
180 return BLK_STS_NOSPC;
181 case NVME_SC_LBA_RANGE:
182 return BLK_STS_TARGET;
183 case NVME_SC_BAD_ATTRIBUTES:
184 case NVME_SC_ONCS_NOT_SUPPORTED:
185 case NVME_SC_INVALID_OPCODE:
186 case NVME_SC_INVALID_FIELD:
187 case NVME_SC_INVALID_NS:
188 return BLK_STS_NOTSUPP;
189 case NVME_SC_WRITE_FAULT:
190 case NVME_SC_READ_ERROR:
191 case NVME_SC_UNWRITTEN_BLOCK:
192 case NVME_SC_ACCESS_DENIED:
193 case NVME_SC_READ_ONLY:
194 case NVME_SC_COMPARE_FAILED:
195 return BLK_STS_MEDIUM;
196 case NVME_SC_GUARD_CHECK:
197 case NVME_SC_APPTAG_CHECK:
198 case NVME_SC_REFTAG_CHECK:
199 case NVME_SC_INVALID_PI:
200 return BLK_STS_PROTECTION;
201 case NVME_SC_RESERVATION_CONFLICT:
202 return BLK_STS_NEXUS;
204 return BLK_STS_IOERR;
208 static inline bool nvme_req_needs_retry(struct request *req)
210 if (blk_noretry_request(req))
212 if (nvme_req(req)->status & NVME_SC_DNR)
214 if (nvme_req(req)->retries >= nvme_max_retries)
219 void nvme_complete_rq(struct request *req)
221 blk_status_t status = nvme_error_status(req);
223 trace_nvme_complete_rq(req);
225 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
226 if (nvme_req_needs_failover(req, status)) {
227 nvme_failover_req(req);
231 if (!blk_queue_dying(req->q)) {
232 nvme_req(req)->retries++;
233 blk_mq_requeue_request(req, true);
237 blk_mq_end_request(req, status);
239 EXPORT_SYMBOL_GPL(nvme_complete_rq);
241 void nvme_cancel_request(struct request *req, void *data, bool reserved)
243 if (!blk_mq_request_started(req))
246 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
247 "Cancelling I/O %d", req->tag);
249 nvme_req(req)->status = NVME_SC_ABORT_REQ;
250 blk_mq_complete_request(req);
253 EXPORT_SYMBOL_GPL(nvme_cancel_request);
255 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
256 enum nvme_ctrl_state new_state)
258 enum nvme_ctrl_state old_state;
260 bool changed = false;
262 spin_lock_irqsave(&ctrl->lock, flags);
264 old_state = ctrl->state;
266 case NVME_CTRL_ADMIN_ONLY:
268 case NVME_CTRL_CONNECTING:
278 case NVME_CTRL_RESETTING:
279 case NVME_CTRL_CONNECTING:
286 case NVME_CTRL_RESETTING:
290 case NVME_CTRL_ADMIN_ONLY:
297 case NVME_CTRL_CONNECTING:
300 case NVME_CTRL_RESETTING:
307 case NVME_CTRL_DELETING:
310 case NVME_CTRL_ADMIN_ONLY:
311 case NVME_CTRL_RESETTING:
312 case NVME_CTRL_CONNECTING:
321 case NVME_CTRL_DELETING:
333 ctrl->state = new_state;
335 spin_unlock_irqrestore(&ctrl->lock, flags);
336 if (changed && ctrl->state == NVME_CTRL_LIVE)
337 nvme_kick_requeue_lists(ctrl);
340 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
342 static void nvme_free_ns_head(struct kref *ref)
344 struct nvme_ns_head *head =
345 container_of(ref, struct nvme_ns_head, ref);
347 nvme_mpath_remove_disk(head);
348 ida_simple_remove(&head->subsys->ns_ida, head->instance);
349 list_del_init(&head->entry);
350 cleanup_srcu_struct(&head->srcu);
354 static void nvme_put_ns_head(struct nvme_ns_head *head)
356 kref_put(&head->ref, nvme_free_ns_head);
359 static void nvme_free_ns(struct kref *kref)
361 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
364 nvme_nvm_unregister(ns);
367 nvme_put_ns_head(ns->head);
368 nvme_put_ctrl(ns->ctrl);
372 static void nvme_put_ns(struct nvme_ns *ns)
374 kref_put(&ns->kref, nvme_free_ns);
377 struct request *nvme_alloc_request(struct request_queue *q,
378 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
380 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
383 if (qid == NVME_QID_ANY) {
384 req = blk_mq_alloc_request(q, op, flags);
386 req = blk_mq_alloc_request_hctx(q, op, flags,
392 req->cmd_flags |= REQ_FAILFAST_DRIVER;
393 nvme_req(req)->cmd = cmd;
397 EXPORT_SYMBOL_GPL(nvme_alloc_request);
399 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
401 struct nvme_command c;
403 memset(&c, 0, sizeof(c));
405 c.directive.opcode = nvme_admin_directive_send;
406 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
407 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
408 c.directive.dtype = NVME_DIR_IDENTIFY;
409 c.directive.tdtype = NVME_DIR_STREAMS;
410 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
412 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
415 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
417 return nvme_toggle_streams(ctrl, false);
420 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
422 return nvme_toggle_streams(ctrl, true);
425 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
426 struct streams_directive_params *s, u32 nsid)
428 struct nvme_command c;
430 memset(&c, 0, sizeof(c));
431 memset(s, 0, sizeof(*s));
433 c.directive.opcode = nvme_admin_directive_recv;
434 c.directive.nsid = cpu_to_le32(nsid);
435 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
436 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
437 c.directive.dtype = NVME_DIR_STREAMS;
439 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
442 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
444 struct streams_directive_params s;
447 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
452 ret = nvme_enable_streams(ctrl);
456 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
460 ctrl->nssa = le16_to_cpu(s.nssa);
461 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
462 dev_info(ctrl->device, "too few streams (%u) available\n",
464 nvme_disable_streams(ctrl);
468 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
469 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
474 * Check if 'req' has a write hint associated with it. If it does, assign
475 * a valid namespace stream to the write.
477 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
478 struct request *req, u16 *control,
481 enum rw_hint streamid = req->write_hint;
483 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
487 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
490 *control |= NVME_RW_DTYPE_STREAMS;
491 *dsmgmt |= streamid << 16;
494 if (streamid < ARRAY_SIZE(req->q->write_hints))
495 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
498 static inline void nvme_setup_flush(struct nvme_ns *ns,
499 struct nvme_command *cmnd)
501 memset(cmnd, 0, sizeof(*cmnd));
502 cmnd->common.opcode = nvme_cmd_flush;
503 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
506 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
507 struct nvme_command *cmnd)
509 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
510 struct nvme_dsm_range *range;
513 range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
515 return BLK_STS_RESOURCE;
517 __rq_for_each_bio(bio, req) {
518 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
519 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
522 range[n].cattr = cpu_to_le32(0);
523 range[n].nlb = cpu_to_le32(nlb);
524 range[n].slba = cpu_to_le64(slba);
529 if (WARN_ON_ONCE(n != segments)) {
531 return BLK_STS_IOERR;
534 memset(cmnd, 0, sizeof(*cmnd));
535 cmnd->dsm.opcode = nvme_cmd_dsm;
536 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
537 cmnd->dsm.nr = cpu_to_le32(segments - 1);
538 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
540 req->special_vec.bv_page = virt_to_page(range);
541 req->special_vec.bv_offset = offset_in_page(range);
542 req->special_vec.bv_len = sizeof(*range) * segments;
543 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
548 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
549 struct request *req, struct nvme_command *cmnd)
551 struct nvme_ctrl *ctrl = ns->ctrl;
555 if (req->cmd_flags & REQ_FUA)
556 control |= NVME_RW_FUA;
557 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
558 control |= NVME_RW_LR;
560 if (req->cmd_flags & REQ_RAHEAD)
561 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
563 memset(cmnd, 0, sizeof(*cmnd));
564 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
565 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
566 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
567 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
569 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
570 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
574 * If formated with metadata, the block layer always provides a
575 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
576 * we enable the PRACT bit for protection information or set the
577 * namespace capacity to zero to prevent any I/O.
579 if (!blk_integrity_rq(req)) {
580 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
581 return BLK_STS_NOTSUPP;
582 control |= NVME_RW_PRINFO_PRACT;
585 switch (ns->pi_type) {
586 case NVME_NS_DPS_PI_TYPE3:
587 control |= NVME_RW_PRINFO_PRCHK_GUARD;
589 case NVME_NS_DPS_PI_TYPE1:
590 case NVME_NS_DPS_PI_TYPE2:
591 control |= NVME_RW_PRINFO_PRCHK_GUARD |
592 NVME_RW_PRINFO_PRCHK_REF;
593 cmnd->rw.reftag = cpu_to_le32(
594 nvme_block_nr(ns, blk_rq_pos(req)));
599 cmnd->rw.control = cpu_to_le16(control);
600 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
604 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
605 struct nvme_command *cmd)
607 blk_status_t ret = BLK_STS_OK;
609 if (!(req->rq_flags & RQF_DONTPREP)) {
610 nvme_req(req)->retries = 0;
611 nvme_req(req)->flags = 0;
612 req->rq_flags |= RQF_DONTPREP;
615 switch (req_op(req)) {
618 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
621 nvme_setup_flush(ns, cmd);
623 case REQ_OP_WRITE_ZEROES:
624 /* currently only aliased to deallocate for a few ctrls: */
626 ret = nvme_setup_discard(ns, req, cmd);
630 ret = nvme_setup_rw(ns, req, cmd);
634 return BLK_STS_IOERR;
637 cmd->common.command_id = req->tag;
639 trace_nvme_setup_nvm_cmd(req->q->id, cmd);
641 trace_nvme_setup_admin_cmd(cmd);
644 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
647 * Returns 0 on success. If the result is negative, it's a Linux error code;
648 * if the result is positive, it's an NVM Express status code
650 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
651 union nvme_result *result, void *buffer, unsigned bufflen,
652 unsigned timeout, int qid, int at_head,
653 blk_mq_req_flags_t flags)
658 req = nvme_alloc_request(q, cmd, flags, qid);
662 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
664 if (buffer && bufflen) {
665 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
670 blk_execute_rq(req->q, NULL, req, at_head);
672 *result = nvme_req(req)->result;
673 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
676 ret = nvme_req(req)->status;
678 blk_mq_free_request(req);
681 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
683 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
684 void *buffer, unsigned bufflen)
686 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
689 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
691 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
692 unsigned len, u32 seed, bool write)
694 struct bio_integrity_payload *bip;
698 buf = kmalloc(len, GFP_KERNEL);
703 if (write && copy_from_user(buf, ubuf, len))
706 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
712 bip->bip_iter.bi_size = len;
713 bip->bip_iter.bi_sector = seed;
714 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
715 offset_in_page(buf));
725 static int nvme_submit_user_cmd(struct request_queue *q,
726 struct nvme_command *cmd, void __user *ubuffer,
727 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
728 u32 meta_seed, u32 *result, unsigned timeout)
730 bool write = nvme_is_write(cmd);
731 struct nvme_ns *ns = q->queuedata;
732 struct gendisk *disk = ns ? ns->disk : NULL;
734 struct bio *bio = NULL;
738 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
742 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
744 if (ubuffer && bufflen) {
745 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
751 if (disk && meta_buffer && meta_len) {
752 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
761 blk_execute_rq(req->q, disk, req, 0);
762 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
765 ret = nvme_req(req)->status;
767 *result = le32_to_cpu(nvme_req(req)->result.u32);
768 if (meta && !ret && !write) {
769 if (copy_to_user(meta_buffer, meta, meta_len))
775 blk_rq_unmap_user(bio);
777 blk_mq_free_request(req);
781 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
783 struct nvme_ctrl *ctrl = rq->end_io_data;
785 blk_mq_free_request(rq);
788 dev_err(ctrl->device,
789 "failed nvme_keep_alive_end_io error=%d\n",
794 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
797 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
801 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
806 rq->timeout = ctrl->kato * HZ;
807 rq->end_io_data = ctrl;
809 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
814 static void nvme_keep_alive_work(struct work_struct *work)
816 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
817 struct nvme_ctrl, ka_work);
819 if (nvme_keep_alive(ctrl)) {
820 /* allocation failure, reset the controller */
821 dev_err(ctrl->device, "keep-alive failed\n");
822 nvme_reset_ctrl(ctrl);
827 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
829 if (unlikely(ctrl->kato == 0))
832 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
833 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
834 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
835 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
837 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
839 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
841 if (unlikely(ctrl->kato == 0))
844 cancel_delayed_work_sync(&ctrl->ka_work);
846 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
848 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
850 struct nvme_command c = { };
853 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
854 c.identify.opcode = nvme_admin_identify;
855 c.identify.cns = NVME_ID_CNS_CTRL;
857 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
861 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
862 sizeof(struct nvme_id_ctrl));
868 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
869 struct nvme_ns_ids *ids)
871 struct nvme_command c = { };
877 c.identify.opcode = nvme_admin_identify;
878 c.identify.nsid = cpu_to_le32(nsid);
879 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
881 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
885 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
886 NVME_IDENTIFY_DATA_SIZE);
890 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
891 struct nvme_ns_id_desc *cur = data + pos;
897 case NVME_NIDT_EUI64:
898 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
899 dev_warn(ctrl->device,
900 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
904 len = NVME_NIDT_EUI64_LEN;
905 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
907 case NVME_NIDT_NGUID:
908 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
909 dev_warn(ctrl->device,
910 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
914 len = NVME_NIDT_NGUID_LEN;
915 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
918 if (cur->nidl != NVME_NIDT_UUID_LEN) {
919 dev_warn(ctrl->device,
920 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
924 len = NVME_NIDT_UUID_LEN;
925 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
928 /* Skip unnkown types */
940 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
942 struct nvme_command c = { };
944 c.identify.opcode = nvme_admin_identify;
945 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
946 c.identify.nsid = cpu_to_le32(nsid);
947 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
950 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
953 struct nvme_id_ns *id;
954 struct nvme_command c = { };
957 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
958 c.identify.opcode = nvme_admin_identify;
959 c.identify.nsid = cpu_to_le32(nsid);
960 c.identify.cns = NVME_ID_CNS_NS;
962 id = kmalloc(sizeof(*id), GFP_KERNEL);
966 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
968 dev_warn(ctrl->device, "Identify namespace failed\n");
976 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
977 void *buffer, size_t buflen, u32 *result)
979 struct nvme_command c;
980 union nvme_result res;
983 memset(&c, 0, sizeof(c));
984 c.features.opcode = nvme_admin_set_features;
985 c.features.fid = cpu_to_le32(fid);
986 c.features.dword11 = cpu_to_le32(dword11);
988 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
989 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
990 if (ret >= 0 && result)
991 *result = le32_to_cpu(res.u32);
995 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
997 u32 q_count = (*count - 1) | ((*count - 1) << 16);
999 int status, nr_io_queues;
1001 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1007 * Degraded controllers might return an error when setting the queue
1008 * count. We still want to be able to bring them online and offer
1009 * access to the admin queue, as that might be only way to fix them up.
1012 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1015 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1016 *count = min(*count, nr_io_queues);
1021 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1023 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1025 struct nvme_user_io io;
1026 struct nvme_command c;
1027 unsigned length, meta_len;
1028 void __user *metadata;
1030 if (copy_from_user(&io, uio, sizeof(io)))
1035 switch (io.opcode) {
1036 case nvme_cmd_write:
1038 case nvme_cmd_compare:
1044 length = (io.nblocks + 1) << ns->lba_shift;
1045 meta_len = (io.nblocks + 1) * ns->ms;
1046 metadata = (void __user *)(uintptr_t)io.metadata;
1051 } else if (meta_len) {
1052 if ((io.metadata & 3) || !io.metadata)
1056 memset(&c, 0, sizeof(c));
1057 c.rw.opcode = io.opcode;
1058 c.rw.flags = io.flags;
1059 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1060 c.rw.slba = cpu_to_le64(io.slba);
1061 c.rw.length = cpu_to_le16(io.nblocks);
1062 c.rw.control = cpu_to_le16(io.control);
1063 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1064 c.rw.reftag = cpu_to_le32(io.reftag);
1065 c.rw.apptag = cpu_to_le16(io.apptag);
1066 c.rw.appmask = cpu_to_le16(io.appmask);
1068 return nvme_submit_user_cmd(ns->queue, &c,
1069 (void __user *)(uintptr_t)io.addr, length,
1070 metadata, meta_len, io.slba, NULL, 0);
1073 static u32 nvme_known_admin_effects(u8 opcode)
1076 case nvme_admin_format_nvm:
1077 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1078 NVME_CMD_EFFECTS_CSE_MASK;
1079 case nvme_admin_sanitize_nvm:
1080 return NVME_CMD_EFFECTS_CSE_MASK;
1087 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1094 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1095 if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1096 dev_warn(ctrl->device,
1097 "IO command:%02x has unhandled effects:%08x\n",
1103 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1105 effects = nvme_known_admin_effects(opcode);
1108 * For simplicity, IO to all namespaces is quiesced even if the command
1109 * effects say only one namespace is affected.
1111 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1112 nvme_start_freeze(ctrl);
1113 nvme_wait_freeze(ctrl);
1118 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1120 struct nvme_ns *ns, *next;
1123 mutex_lock(&ctrl->namespaces_mutex);
1124 list_for_each_entry(ns, &ctrl->namespaces, list) {
1125 if (ns->disk && nvme_revalidate_disk(ns->disk)) {
1126 list_move_tail(&ns->list, &rm_list);
1129 mutex_unlock(&ctrl->namespaces_mutex);
1131 list_for_each_entry_safe(ns, next, &rm_list, list)
1135 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1138 * Revalidate LBA changes prior to unfreezing. This is necessary to
1139 * prevent memory corruption if a logical block size was changed by
1142 if (effects & NVME_CMD_EFFECTS_LBCC)
1143 nvme_update_formats(ctrl);
1144 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK))
1145 nvme_unfreeze(ctrl);
1146 if (effects & NVME_CMD_EFFECTS_CCC)
1147 nvme_init_identify(ctrl);
1148 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1149 nvme_queue_scan(ctrl);
1152 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1153 struct nvme_passthru_cmd __user *ucmd)
1155 struct nvme_passthru_cmd cmd;
1156 struct nvme_command c;
1157 unsigned timeout = 0;
1161 if (!capable(CAP_SYS_ADMIN))
1163 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1168 memset(&c, 0, sizeof(c));
1169 c.common.opcode = cmd.opcode;
1170 c.common.flags = cmd.flags;
1171 c.common.nsid = cpu_to_le32(cmd.nsid);
1172 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1173 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1174 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1175 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1176 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1177 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1178 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1179 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1182 timeout = msecs_to_jiffies(cmd.timeout_ms);
1184 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1185 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1186 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1187 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata,
1188 0, &cmd.result, timeout);
1189 nvme_passthru_end(ctrl, effects);
1192 if (put_user(cmd.result, &ucmd->result))
1200 * Issue ioctl requests on the first available path. Note that unlike normal
1201 * block layer requests we will not retry failed request on another controller.
1203 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1204 struct nvme_ns_head **head, int *srcu_idx)
1206 #ifdef CONFIG_NVME_MULTIPATH
1207 if (disk->fops == &nvme_ns_head_ops) {
1208 *head = disk->private_data;
1209 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1210 return nvme_find_path(*head);
1215 return disk->private_data;
1218 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1221 srcu_read_unlock(&head->srcu, idx);
1224 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1228 force_successful_syscall_return();
1229 return ns->head->ns_id;
1230 case NVME_IOCTL_ADMIN_CMD:
1231 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1232 case NVME_IOCTL_IO_CMD:
1233 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1234 case NVME_IOCTL_SUBMIT_IO:
1235 return nvme_submit_io(ns, (void __user *)arg);
1239 return nvme_nvm_ioctl(ns, cmd, arg);
1241 if (is_sed_ioctl(cmd))
1242 return sed_ioctl(ns->ctrl->opal_dev, cmd,
1243 (void __user *) arg);
1248 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1249 unsigned int cmd, unsigned long arg)
1251 struct nvme_ns_head *head = NULL;
1255 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1259 ret = nvme_ns_ioctl(ns, cmd, arg);
1260 nvme_put_ns_from_disk(head, srcu_idx);
1264 static int nvme_open(struct block_device *bdev, fmode_t mode)
1266 struct nvme_ns *ns = bdev->bd_disk->private_data;
1268 #ifdef CONFIG_NVME_MULTIPATH
1269 /* should never be called due to GENHD_FL_HIDDEN */
1270 if (WARN_ON_ONCE(ns->head->disk))
1273 if (!kref_get_unless_zero(&ns->kref))
1275 if (!try_module_get(ns->ctrl->ops->module))
1286 static void nvme_release(struct gendisk *disk, fmode_t mode)
1288 struct nvme_ns *ns = disk->private_data;
1290 module_put(ns->ctrl->ops->module);
1294 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1296 /* some standard values */
1297 geo->heads = 1 << 6;
1298 geo->sectors = 1 << 5;
1299 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1303 #ifdef CONFIG_BLK_DEV_INTEGRITY
1304 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1306 struct blk_integrity integrity;
1308 memset(&integrity, 0, sizeof(integrity));
1310 case NVME_NS_DPS_PI_TYPE3:
1311 integrity.profile = &t10_pi_type3_crc;
1312 integrity.tag_size = sizeof(u16) + sizeof(u32);
1313 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1315 case NVME_NS_DPS_PI_TYPE1:
1316 case NVME_NS_DPS_PI_TYPE2:
1317 integrity.profile = &t10_pi_type1_crc;
1318 integrity.tag_size = sizeof(u16);
1319 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1322 integrity.profile = NULL;
1325 integrity.tuple_size = ms;
1326 blk_integrity_register(disk, &integrity);
1327 blk_queue_max_integrity_segments(disk->queue, 1);
1330 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1333 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1335 static void nvme_set_chunk_size(struct nvme_ns *ns)
1337 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1338 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1341 static void nvme_config_discard(struct nvme_ctrl *ctrl,
1342 unsigned stream_alignment, struct request_queue *queue)
1344 u32 size = queue_logical_block_size(queue);
1346 if (stream_alignment)
1347 size *= stream_alignment;
1349 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1350 NVME_DSM_MAX_RANGES);
1352 queue->limits.discard_alignment = 0;
1353 queue->limits.discard_granularity = size;
1355 blk_queue_max_discard_sectors(queue, UINT_MAX);
1356 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1357 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, queue);
1359 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1360 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1363 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1364 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1366 memset(ids, 0, sizeof(*ids));
1368 if (ctrl->vs >= NVME_VS(1, 1, 0))
1369 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1370 if (ctrl->vs >= NVME_VS(1, 2, 0))
1371 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1372 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1373 /* Don't treat error as fatal we potentially
1374 * already have a NGUID or EUI-64
1376 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1377 dev_warn(ctrl->device,
1378 "%s: Identify Descriptors failed\n", __func__);
1382 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1384 return !uuid_is_null(&ids->uuid) ||
1385 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1386 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1389 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1391 return uuid_equal(&a->uuid, &b->uuid) &&
1392 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1393 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1396 static void nvme_update_disk_info(struct gendisk *disk,
1397 struct nvme_ns *ns, struct nvme_id_ns *id)
1399 sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1400 unsigned short bs = 1 << ns->lba_shift;
1401 unsigned stream_alignment = 0;
1403 if (ns->ctrl->nr_streams && ns->sws && ns->sgs)
1404 stream_alignment = ns->sws * ns->sgs;
1406 blk_mq_freeze_queue(disk->queue);
1407 blk_integrity_unregister(disk);
1409 blk_queue_logical_block_size(disk->queue, bs);
1410 blk_queue_physical_block_size(disk->queue, bs);
1411 blk_queue_io_min(disk->queue, bs);
1413 if (ns->ms && !ns->ext &&
1414 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1415 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1416 if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1418 set_capacity(disk, capacity);
1420 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1421 nvme_config_discard(ns->ctrl, stream_alignment, disk->queue);
1422 blk_mq_unfreeze_queue(disk->queue);
1425 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1427 struct nvme_ns *ns = disk->private_data;
1430 * If identify namespace failed, use default 512 byte block size so
1431 * block layer can use before failing read/write for 0 capacity.
1433 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1434 if (ns->lba_shift == 0)
1436 ns->noiob = le16_to_cpu(id->noiob);
1437 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1438 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1439 /* the PI implementation requires metadata equal t10 pi tuple size */
1440 if (ns->ms == sizeof(struct t10_pi_tuple))
1441 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1446 nvme_set_chunk_size(ns);
1447 nvme_update_disk_info(disk, ns, id);
1448 #ifdef CONFIG_NVME_MULTIPATH
1450 nvme_update_disk_info(ns->head->disk, ns, id);
1454 static int nvme_revalidate_disk(struct gendisk *disk)
1456 struct nvme_ns *ns = disk->private_data;
1457 struct nvme_ctrl *ctrl = ns->ctrl;
1458 struct nvme_id_ns *id;
1459 struct nvme_ns_ids ids;
1462 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1463 set_capacity(disk, 0);
1467 id = nvme_identify_ns(ctrl, ns->head->ns_id);
1471 if (id->ncap == 0) {
1476 __nvme_revalidate_disk(disk, id);
1477 nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1478 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1479 dev_err(ctrl->device,
1480 "identifiers changed for nsid %d\n", ns->head->ns_id);
1489 static char nvme_pr_type(enum pr_type type)
1492 case PR_WRITE_EXCLUSIVE:
1494 case PR_EXCLUSIVE_ACCESS:
1496 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1498 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1500 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1502 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1509 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1510 u64 key, u64 sa_key, u8 op)
1512 struct nvme_ns_head *head = NULL;
1514 struct nvme_command c;
1516 u8 data[16] = { 0, };
1518 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1520 return -EWOULDBLOCK;
1522 put_unaligned_le64(key, &data[0]);
1523 put_unaligned_le64(sa_key, &data[8]);
1525 memset(&c, 0, sizeof(c));
1526 c.common.opcode = op;
1527 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1528 c.common.cdw10[0] = cpu_to_le32(cdw10);
1530 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1531 nvme_put_ns_from_disk(head, srcu_idx);
1535 static int nvme_pr_register(struct block_device *bdev, u64 old,
1536 u64 new, unsigned flags)
1540 if (flags & ~PR_FL_IGNORE_KEY)
1543 cdw10 = old ? 2 : 0;
1544 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1545 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1546 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1549 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1550 enum pr_type type, unsigned flags)
1554 if (flags & ~PR_FL_IGNORE_KEY)
1557 cdw10 = nvme_pr_type(type) << 8;
1558 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1559 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1562 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1563 enum pr_type type, bool abort)
1565 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1566 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1569 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1571 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1572 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1575 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1577 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1578 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1581 static const struct pr_ops nvme_pr_ops = {
1582 .pr_register = nvme_pr_register,
1583 .pr_reserve = nvme_pr_reserve,
1584 .pr_release = nvme_pr_release,
1585 .pr_preempt = nvme_pr_preempt,
1586 .pr_clear = nvme_pr_clear,
1589 #ifdef CONFIG_BLK_SED_OPAL
1590 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1593 struct nvme_ctrl *ctrl = data;
1594 struct nvme_command cmd;
1596 memset(&cmd, 0, sizeof(cmd));
1598 cmd.common.opcode = nvme_admin_security_send;
1600 cmd.common.opcode = nvme_admin_security_recv;
1601 cmd.common.nsid = 0;
1602 cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1603 cmd.common.cdw10[1] = cpu_to_le32(len);
1605 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1606 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1608 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1609 #endif /* CONFIG_BLK_SED_OPAL */
1611 static const struct block_device_operations nvme_fops = {
1612 .owner = THIS_MODULE,
1613 .ioctl = nvme_ioctl,
1614 .compat_ioctl = nvme_ioctl,
1616 .release = nvme_release,
1617 .getgeo = nvme_getgeo,
1618 .revalidate_disk= nvme_revalidate_disk,
1619 .pr_ops = &nvme_pr_ops,
1622 #ifdef CONFIG_NVME_MULTIPATH
1623 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1625 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1627 if (!kref_get_unless_zero(&head->ref))
1632 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1634 nvme_put_ns_head(disk->private_data);
1637 const struct block_device_operations nvme_ns_head_ops = {
1638 .owner = THIS_MODULE,
1639 .open = nvme_ns_head_open,
1640 .release = nvme_ns_head_release,
1641 .ioctl = nvme_ioctl,
1642 .compat_ioctl = nvme_ioctl,
1643 .getgeo = nvme_getgeo,
1644 .pr_ops = &nvme_pr_ops,
1646 #endif /* CONFIG_NVME_MULTIPATH */
1648 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1650 unsigned long timeout =
1651 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1652 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1655 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1658 if ((csts & NVME_CSTS_RDY) == bit)
1662 if (fatal_signal_pending(current))
1664 if (time_after(jiffies, timeout)) {
1665 dev_err(ctrl->device,
1666 "Device not ready; aborting %s\n", enabled ?
1667 "initialisation" : "reset");
1676 * If the device has been passed off to us in an enabled state, just clear
1677 * the enabled bit. The spec says we should set the 'shutdown notification
1678 * bits', but doing so may cause the device to complete commands to the
1679 * admin queue ... and we don't know what memory that might be pointing at!
1681 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1685 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1686 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1688 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1692 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1693 msleep(NVME_QUIRK_DELAY_AMOUNT);
1695 return nvme_wait_ready(ctrl, cap, false);
1697 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1699 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1702 * Default to a 4K page size, with the intention to update this
1703 * path in the future to accomodate architectures with differing
1704 * kernel and IO page sizes.
1706 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1709 if (page_shift < dev_page_min) {
1710 dev_err(ctrl->device,
1711 "Minimum device page size %u too large for host (%u)\n",
1712 1 << dev_page_min, 1 << page_shift);
1716 ctrl->page_size = 1 << page_shift;
1718 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1719 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1720 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1721 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1722 ctrl->ctrl_config |= NVME_CC_ENABLE;
1724 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1727 return nvme_wait_ready(ctrl, cap, true);
1729 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1731 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1733 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1737 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1738 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1740 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1744 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1745 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1749 if (fatal_signal_pending(current))
1751 if (time_after(jiffies, timeout)) {
1752 dev_err(ctrl->device,
1753 "Device shutdown incomplete; abort shutdown\n");
1760 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1762 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1763 struct request_queue *q)
1767 if (ctrl->max_hw_sectors) {
1769 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1771 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1772 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1774 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1775 is_power_of_2(ctrl->max_hw_sectors))
1776 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1777 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1778 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1780 blk_queue_write_cache(q, vwc, vwc);
1783 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1788 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1791 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1792 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1795 dev_warn_once(ctrl->device,
1796 "could not set timestamp (%d)\n", ret);
1800 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1803 * APST (Autonomous Power State Transition) lets us program a
1804 * table of power state transitions that the controller will
1805 * perform automatically. We configure it with a simple
1806 * heuristic: we are willing to spend at most 2% of the time
1807 * transitioning between power states. Therefore, when running
1808 * in any given state, we will enter the next lower-power
1809 * non-operational state after waiting 50 * (enlat + exlat)
1810 * microseconds, as long as that state's exit latency is under
1811 * the requested maximum latency.
1813 * We will not autonomously enter any non-operational state for
1814 * which the total latency exceeds ps_max_latency_us. Users
1815 * can set ps_max_latency_us to zero to turn off APST.
1819 struct nvme_feat_auto_pst *table;
1825 * If APST isn't supported or if we haven't been initialized yet,
1826 * then don't do anything.
1831 if (ctrl->npss > 31) {
1832 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1836 table = kzalloc(sizeof(*table), GFP_KERNEL);
1840 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1841 /* Turn off APST. */
1843 dev_dbg(ctrl->device, "APST disabled\n");
1845 __le64 target = cpu_to_le64(0);
1849 * Walk through all states from lowest- to highest-power.
1850 * According to the spec, lower-numbered states use more
1851 * power. NPSS, despite the name, is the index of the
1852 * lowest-power state, not the number of states.
1854 for (state = (int)ctrl->npss; state >= 0; state--) {
1855 u64 total_latency_us, exit_latency_us, transition_ms;
1858 table->entries[state] = target;
1861 * Don't allow transitions to the deepest state
1862 * if it's quirked off.
1864 if (state == ctrl->npss &&
1865 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1869 * Is this state a useful non-operational state for
1870 * higher-power states to autonomously transition to?
1872 if (!(ctrl->psd[state].flags &
1873 NVME_PS_FLAGS_NON_OP_STATE))
1877 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1878 if (exit_latency_us > ctrl->ps_max_latency_us)
1883 le32_to_cpu(ctrl->psd[state].entry_lat);
1886 * This state is good. Use it as the APST idle
1887 * target for higher power states.
1889 transition_ms = total_latency_us + 19;
1890 do_div(transition_ms, 20);
1891 if (transition_ms > (1 << 24) - 1)
1892 transition_ms = (1 << 24) - 1;
1894 target = cpu_to_le64((state << 3) |
1895 (transition_ms << 8));
1900 if (total_latency_us > max_lat_us)
1901 max_lat_us = total_latency_us;
1907 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1909 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1910 max_ps, max_lat_us, (int)sizeof(*table), table);
1914 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1915 table, sizeof(*table), NULL);
1917 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1923 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1925 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1929 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1930 case PM_QOS_LATENCY_ANY:
1938 if (ctrl->ps_max_latency_us != latency) {
1939 ctrl->ps_max_latency_us = latency;
1940 nvme_configure_apst(ctrl);
1944 struct nvme_core_quirk_entry {
1946 * NVMe model and firmware strings are padded with spaces. For
1947 * simplicity, strings in the quirk table are padded with NULLs
1953 unsigned long quirks;
1956 static const struct nvme_core_quirk_entry core_quirks[] = {
1959 * This Toshiba device seems to die using any APST states. See:
1960 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1963 .mn = "THNSF5256GPUK TOSHIBA",
1964 .quirks = NVME_QUIRK_NO_APST,
1968 /* match is null-terminated but idstr is space-padded. */
1969 static bool string_matches(const char *idstr, const char *match, size_t len)
1976 matchlen = strlen(match);
1977 WARN_ON_ONCE(matchlen > len);
1979 if (memcmp(idstr, match, matchlen))
1982 for (; matchlen < len; matchlen++)
1983 if (idstr[matchlen] != ' ')
1989 static bool quirk_matches(const struct nvme_id_ctrl *id,
1990 const struct nvme_core_quirk_entry *q)
1992 return q->vid == le16_to_cpu(id->vid) &&
1993 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1994 string_matches(id->fr, q->fr, sizeof(id->fr));
1997 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
1998 struct nvme_id_ctrl *id)
2003 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2004 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2005 strncpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2009 if (ctrl->vs >= NVME_VS(1, 2, 1))
2010 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2012 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2013 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2014 "nqn.2014.08.org.nvmexpress:%4x%4x",
2015 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2016 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2017 off += sizeof(id->sn);
2018 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2019 off += sizeof(id->mn);
2020 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2023 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2025 ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2029 static void nvme_release_subsystem(struct device *dev)
2031 __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2034 static void nvme_destroy_subsystem(struct kref *ref)
2036 struct nvme_subsystem *subsys =
2037 container_of(ref, struct nvme_subsystem, ref);
2039 mutex_lock(&nvme_subsystems_lock);
2040 list_del(&subsys->entry);
2041 mutex_unlock(&nvme_subsystems_lock);
2043 ida_destroy(&subsys->ns_ida);
2044 device_del(&subsys->dev);
2045 put_device(&subsys->dev);
2048 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2050 kref_put(&subsys->ref, nvme_destroy_subsystem);
2053 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2055 struct nvme_subsystem *subsys;
2057 lockdep_assert_held(&nvme_subsystems_lock);
2059 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2060 if (strcmp(subsys->subnqn, subsysnqn))
2062 if (!kref_get_unless_zero(&subsys->ref))
2070 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2071 struct device_attribute subsys_attr_##_name = \
2072 __ATTR(_name, _mode, _show, NULL)
2074 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2075 struct device_attribute *attr,
2078 struct nvme_subsystem *subsys =
2079 container_of(dev, struct nvme_subsystem, dev);
2081 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2083 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2085 #define nvme_subsys_show_str_function(field) \
2086 static ssize_t subsys_##field##_show(struct device *dev, \
2087 struct device_attribute *attr, char *buf) \
2089 struct nvme_subsystem *subsys = \
2090 container_of(dev, struct nvme_subsystem, dev); \
2091 return sprintf(buf, "%.*s\n", \
2092 (int)sizeof(subsys->field), subsys->field); \
2094 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2096 nvme_subsys_show_str_function(model);
2097 nvme_subsys_show_str_function(serial);
2098 nvme_subsys_show_str_function(firmware_rev);
2100 static struct attribute *nvme_subsys_attrs[] = {
2101 &subsys_attr_model.attr,
2102 &subsys_attr_serial.attr,
2103 &subsys_attr_firmware_rev.attr,
2104 &subsys_attr_subsysnqn.attr,
2108 static struct attribute_group nvme_subsys_attrs_group = {
2109 .attrs = nvme_subsys_attrs,
2112 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2113 &nvme_subsys_attrs_group,
2117 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2120 struct nvme_ctrl *ctrl;
2122 mutex_lock(&subsys->lock);
2123 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2124 if (ctrl->state != NVME_CTRL_DELETING &&
2125 ctrl->state != NVME_CTRL_DEAD)
2128 mutex_unlock(&subsys->lock);
2133 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2135 struct nvme_subsystem *subsys, *found;
2138 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2141 ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2146 subsys->instance = ret;
2147 mutex_init(&subsys->lock);
2148 kref_init(&subsys->ref);
2149 INIT_LIST_HEAD(&subsys->ctrls);
2150 INIT_LIST_HEAD(&subsys->nsheads);
2151 nvme_init_subnqn(subsys, ctrl, id);
2152 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2153 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2154 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2155 subsys->vendor_id = le16_to_cpu(id->vid);
2156 subsys->cmic = id->cmic;
2158 subsys->dev.class = nvme_subsys_class;
2159 subsys->dev.release = nvme_release_subsystem;
2160 subsys->dev.groups = nvme_subsys_attrs_groups;
2161 dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2162 device_initialize(&subsys->dev);
2164 mutex_lock(&nvme_subsystems_lock);
2165 found = __nvme_find_get_subsystem(subsys->subnqn);
2168 * Verify that the subsystem actually supports multiple
2169 * controllers, else bail out.
2171 if (nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2172 dev_err(ctrl->device,
2173 "ignoring ctrl due to duplicate subnqn (%s).\n",
2175 nvme_put_subsystem(found);
2180 __nvme_release_subsystem(subsys);
2183 ret = device_add(&subsys->dev);
2185 dev_err(ctrl->device,
2186 "failed to register subsystem device.\n");
2189 ida_init(&subsys->ns_ida);
2190 list_add_tail(&subsys->entry, &nvme_subsystems);
2193 ctrl->subsys = subsys;
2194 mutex_unlock(&nvme_subsystems_lock);
2196 if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2197 dev_name(ctrl->device))) {
2198 dev_err(ctrl->device,
2199 "failed to create sysfs link from subsystem.\n");
2200 /* the transport driver will eventually put the subsystem */
2204 mutex_lock(&subsys->lock);
2205 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2206 mutex_unlock(&subsys->lock);
2211 mutex_unlock(&nvme_subsystems_lock);
2212 put_device(&subsys->dev);
2216 static int nvme_get_log(struct nvme_ctrl *ctrl, u8 log_page, void *log,
2219 struct nvme_command c = { };
2221 c.common.opcode = nvme_admin_get_log_page;
2222 c.common.nsid = cpu_to_le32(NVME_NSID_ALL);
2223 c.common.cdw10[0] = nvme_get_log_dw10(log_page, size);
2225 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2228 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2233 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2238 ret = nvme_get_log(ctrl, NVME_LOG_CMD_EFFECTS, ctrl->effects,
2239 sizeof(*ctrl->effects));
2241 kfree(ctrl->effects);
2242 ctrl->effects = NULL;
2248 * Initialize the cached copies of the Identify data and various controller
2249 * register in our nvme_ctrl structure. This should be called as soon as
2250 * the admin queue is fully up and running.
2252 int nvme_init_identify(struct nvme_ctrl *ctrl)
2254 struct nvme_id_ctrl *id;
2256 int ret, page_shift;
2258 bool prev_apst_enabled;
2260 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2262 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2266 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2268 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2271 page_shift = NVME_CAP_MPSMIN(cap) + 12;
2273 if (ctrl->vs >= NVME_VS(1, 1, 0))
2274 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2276 ret = nvme_identify_ctrl(ctrl, &id);
2278 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2282 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2283 ret = nvme_get_effects_log(ctrl);
2288 if (!ctrl->identified) {
2291 ret = nvme_init_subsystem(ctrl, id);
2296 * Check for quirks. Quirk can depend on firmware version,
2297 * so, in principle, the set of quirks present can change
2298 * across a reset. As a possible future enhancement, we
2299 * could re-scan for quirks every time we reinitialize
2300 * the device, but we'd have to make sure that the driver
2301 * behaves intelligently if the quirks change.
2303 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2304 if (quirk_matches(id, &core_quirks[i]))
2305 ctrl->quirks |= core_quirks[i].quirks;
2309 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2310 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2311 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2314 ctrl->oacs = le16_to_cpu(id->oacs);
2315 ctrl->oncs = le16_to_cpup(&id->oncs);
2316 atomic_set(&ctrl->abort_limit, id->acl + 1);
2317 ctrl->vwc = id->vwc;
2318 ctrl->cntlid = le16_to_cpup(&id->cntlid);
2320 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2322 max_hw_sectors = UINT_MAX;
2323 ctrl->max_hw_sectors =
2324 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2326 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2327 ctrl->sgls = le32_to_cpu(id->sgls);
2328 ctrl->kas = le16_to_cpu(id->kas);
2332 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2334 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2335 shutdown_timeout, 60);
2337 if (ctrl->shutdown_timeout != shutdown_timeout)
2338 dev_info(ctrl->device,
2339 "Shutdown timeout set to %u seconds\n",
2340 ctrl->shutdown_timeout);
2342 ctrl->shutdown_timeout = shutdown_timeout;
2344 ctrl->npss = id->npss;
2345 ctrl->apsta = id->apsta;
2346 prev_apst_enabled = ctrl->apst_enabled;
2347 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2348 if (force_apst && id->apsta) {
2349 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2350 ctrl->apst_enabled = true;
2352 ctrl->apst_enabled = false;
2355 ctrl->apst_enabled = id->apsta;
2357 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2359 if (ctrl->ops->flags & NVME_F_FABRICS) {
2360 ctrl->icdoff = le16_to_cpu(id->icdoff);
2361 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2362 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2363 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2366 * In fabrics we need to verify the cntlid matches the
2369 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2374 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2375 dev_err(ctrl->device,
2376 "keep-alive support is mandatory for fabrics\n");
2381 ctrl->cntlid = le16_to_cpu(id->cntlid);
2382 ctrl->hmpre = le32_to_cpu(id->hmpre);
2383 ctrl->hmmin = le32_to_cpu(id->hmmin);
2384 ctrl->hmminds = le32_to_cpu(id->hmminds);
2385 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2390 if (ctrl->apst_enabled && !prev_apst_enabled)
2391 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2392 else if (!ctrl->apst_enabled && prev_apst_enabled)
2393 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2395 ret = nvme_configure_apst(ctrl);
2399 ret = nvme_configure_timestamp(ctrl);
2403 ret = nvme_configure_directives(ctrl);
2407 ctrl->identified = true;
2415 EXPORT_SYMBOL_GPL(nvme_init_identify);
2417 static int nvme_dev_open(struct inode *inode, struct file *file)
2419 struct nvme_ctrl *ctrl =
2420 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2422 switch (ctrl->state) {
2423 case NVME_CTRL_LIVE:
2424 case NVME_CTRL_ADMIN_ONLY:
2427 return -EWOULDBLOCK;
2430 file->private_data = ctrl;
2434 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2439 mutex_lock(&ctrl->namespaces_mutex);
2440 if (list_empty(&ctrl->namespaces)) {
2445 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2446 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2447 dev_warn(ctrl->device,
2448 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2453 dev_warn(ctrl->device,
2454 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2455 kref_get(&ns->kref);
2456 mutex_unlock(&ctrl->namespaces_mutex);
2458 ret = nvme_user_cmd(ctrl, ns, argp);
2463 mutex_unlock(&ctrl->namespaces_mutex);
2467 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2470 struct nvme_ctrl *ctrl = file->private_data;
2471 void __user *argp = (void __user *)arg;
2474 case NVME_IOCTL_ADMIN_CMD:
2475 return nvme_user_cmd(ctrl, NULL, argp);
2476 case NVME_IOCTL_IO_CMD:
2477 return nvme_dev_user_cmd(ctrl, argp);
2478 case NVME_IOCTL_RESET:
2479 dev_warn(ctrl->device, "resetting controller\n");
2480 return nvme_reset_ctrl_sync(ctrl);
2481 case NVME_IOCTL_SUBSYS_RESET:
2482 return nvme_reset_subsystem(ctrl);
2483 case NVME_IOCTL_RESCAN:
2484 nvme_queue_scan(ctrl);
2491 static const struct file_operations nvme_dev_fops = {
2492 .owner = THIS_MODULE,
2493 .open = nvme_dev_open,
2494 .unlocked_ioctl = nvme_dev_ioctl,
2495 .compat_ioctl = nvme_dev_ioctl,
2498 static ssize_t nvme_sysfs_reset(struct device *dev,
2499 struct device_attribute *attr, const char *buf,
2502 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2505 ret = nvme_reset_ctrl_sync(ctrl);
2510 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2512 static ssize_t nvme_sysfs_rescan(struct device *dev,
2513 struct device_attribute *attr, const char *buf,
2516 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2518 nvme_queue_scan(ctrl);
2521 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2523 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2525 struct gendisk *disk = dev_to_disk(dev);
2527 if (disk->fops == &nvme_fops)
2528 return nvme_get_ns_from_dev(dev)->head;
2530 return disk->private_data;
2533 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2536 struct nvme_ns_head *head = dev_to_ns_head(dev);
2537 struct nvme_ns_ids *ids = &head->ids;
2538 struct nvme_subsystem *subsys = head->subsys;
2539 int serial_len = sizeof(subsys->serial);
2540 int model_len = sizeof(subsys->model);
2542 if (!uuid_is_null(&ids->uuid))
2543 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2545 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2546 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2548 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2549 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2551 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2552 subsys->serial[serial_len - 1] == '\0'))
2554 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2555 subsys->model[model_len - 1] == '\0'))
2558 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2559 serial_len, subsys->serial, model_len, subsys->model,
2562 static DEVICE_ATTR_RO(wwid);
2564 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2567 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2569 static DEVICE_ATTR_RO(nguid);
2571 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2574 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2576 /* For backward compatibility expose the NGUID to userspace if
2577 * we have no UUID set
2579 if (uuid_is_null(&ids->uuid)) {
2580 printk_ratelimited(KERN_WARNING
2581 "No UUID available providing old NGUID\n");
2582 return sprintf(buf, "%pU\n", ids->nguid);
2584 return sprintf(buf, "%pU\n", &ids->uuid);
2586 static DEVICE_ATTR_RO(uuid);
2588 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2591 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2593 static DEVICE_ATTR_RO(eui);
2595 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2598 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2600 static DEVICE_ATTR_RO(nsid);
2602 static struct attribute *nvme_ns_id_attrs[] = {
2603 &dev_attr_wwid.attr,
2604 &dev_attr_uuid.attr,
2605 &dev_attr_nguid.attr,
2607 &dev_attr_nsid.attr,
2611 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2612 struct attribute *a, int n)
2614 struct device *dev = container_of(kobj, struct device, kobj);
2615 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2617 if (a == &dev_attr_uuid.attr) {
2618 if (uuid_is_null(&ids->uuid) &&
2619 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2622 if (a == &dev_attr_nguid.attr) {
2623 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2626 if (a == &dev_attr_eui.attr) {
2627 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2633 const struct attribute_group nvme_ns_id_attr_group = {
2634 .attrs = nvme_ns_id_attrs,
2635 .is_visible = nvme_ns_id_attrs_are_visible,
2638 #define nvme_show_str_function(field) \
2639 static ssize_t field##_show(struct device *dev, \
2640 struct device_attribute *attr, char *buf) \
2642 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2643 return sprintf(buf, "%.*s\n", \
2644 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
2646 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2648 nvme_show_str_function(model);
2649 nvme_show_str_function(serial);
2650 nvme_show_str_function(firmware_rev);
2652 #define nvme_show_int_function(field) \
2653 static ssize_t field##_show(struct device *dev, \
2654 struct device_attribute *attr, char *buf) \
2656 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2657 return sprintf(buf, "%d\n", ctrl->field); \
2659 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2661 nvme_show_int_function(cntlid);
2663 static ssize_t nvme_sysfs_delete(struct device *dev,
2664 struct device_attribute *attr, const char *buf,
2667 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2669 if (device_remove_file_self(dev, attr))
2670 nvme_delete_ctrl_sync(ctrl);
2673 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2675 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2676 struct device_attribute *attr,
2679 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2681 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2683 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2685 static ssize_t nvme_sysfs_show_state(struct device *dev,
2686 struct device_attribute *attr,
2689 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2690 static const char *const state_name[] = {
2691 [NVME_CTRL_NEW] = "new",
2692 [NVME_CTRL_LIVE] = "live",
2693 [NVME_CTRL_ADMIN_ONLY] = "only-admin",
2694 [NVME_CTRL_RESETTING] = "resetting",
2695 [NVME_CTRL_CONNECTING] = "connecting",
2696 [NVME_CTRL_DELETING] = "deleting",
2697 [NVME_CTRL_DEAD] = "dead",
2700 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2701 state_name[ctrl->state])
2702 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2704 return sprintf(buf, "unknown state\n");
2707 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2709 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2710 struct device_attribute *attr,
2713 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2715 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2717 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2719 static ssize_t nvme_sysfs_show_address(struct device *dev,
2720 struct device_attribute *attr,
2723 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2725 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2727 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2729 static struct attribute *nvme_dev_attrs[] = {
2730 &dev_attr_reset_controller.attr,
2731 &dev_attr_rescan_controller.attr,
2732 &dev_attr_model.attr,
2733 &dev_attr_serial.attr,
2734 &dev_attr_firmware_rev.attr,
2735 &dev_attr_cntlid.attr,
2736 &dev_attr_delete_controller.attr,
2737 &dev_attr_transport.attr,
2738 &dev_attr_subsysnqn.attr,
2739 &dev_attr_address.attr,
2740 &dev_attr_state.attr,
2744 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2745 struct attribute *a, int n)
2747 struct device *dev = container_of(kobj, struct device, kobj);
2748 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2750 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2752 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2758 static struct attribute_group nvme_dev_attrs_group = {
2759 .attrs = nvme_dev_attrs,
2760 .is_visible = nvme_dev_attrs_are_visible,
2763 static const struct attribute_group *nvme_dev_attr_groups[] = {
2764 &nvme_dev_attrs_group,
2768 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2771 struct nvme_ns_head *h;
2773 lockdep_assert_held(&subsys->lock);
2775 list_for_each_entry(h, &subsys->nsheads, entry) {
2776 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2783 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2784 struct nvme_ns_head *new)
2786 struct nvme_ns_head *h;
2788 lockdep_assert_held(&subsys->lock);
2790 list_for_each_entry(h, &subsys->nsheads, entry) {
2791 if (nvme_ns_ids_valid(&new->ids) &&
2792 nvme_ns_ids_equal(&new->ids, &h->ids))
2799 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2800 unsigned nsid, struct nvme_id_ns *id)
2802 struct nvme_ns_head *head;
2805 head = kzalloc(sizeof(*head), GFP_KERNEL);
2808 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
2811 head->instance = ret;
2812 INIT_LIST_HEAD(&head->list);
2813 init_srcu_struct(&head->srcu);
2814 head->subsys = ctrl->subsys;
2816 kref_init(&head->ref);
2818 nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
2820 ret = __nvme_check_ids(ctrl->subsys, head);
2822 dev_err(ctrl->device,
2823 "duplicate IDs for nsid %d\n", nsid);
2824 goto out_cleanup_srcu;
2827 ret = nvme_mpath_alloc_disk(ctrl, head);
2829 goto out_cleanup_srcu;
2831 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
2834 cleanup_srcu_struct(&head->srcu);
2835 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
2839 return ERR_PTR(ret);
2842 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
2843 struct nvme_id_ns *id, bool *new)
2845 struct nvme_ctrl *ctrl = ns->ctrl;
2846 bool is_shared = id->nmic & (1 << 0);
2847 struct nvme_ns_head *head = NULL;
2850 mutex_lock(&ctrl->subsys->lock);
2852 head = __nvme_find_ns_head(ctrl->subsys, nsid);
2854 head = nvme_alloc_ns_head(ctrl, nsid, id);
2856 ret = PTR_ERR(head);
2862 struct nvme_ns_ids ids;
2864 nvme_report_ns_ids(ctrl, nsid, id, &ids);
2865 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
2866 dev_err(ctrl->device,
2867 "IDs don't match for shared namespace %d\n",
2876 list_add_tail(&ns->siblings, &head->list);
2880 mutex_unlock(&ctrl->subsys->lock);
2884 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2886 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2887 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2889 return nsa->head->ns_id - nsb->head->ns_id;
2892 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2894 struct nvme_ns *ns, *ret = NULL;
2896 mutex_lock(&ctrl->namespaces_mutex);
2897 list_for_each_entry(ns, &ctrl->namespaces, list) {
2898 if (ns->head->ns_id == nsid) {
2899 if (!kref_get_unless_zero(&ns->kref))
2904 if (ns->head->ns_id > nsid)
2907 mutex_unlock(&ctrl->namespaces_mutex);
2911 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
2913 struct streams_directive_params s;
2916 if (!ctrl->nr_streams)
2919 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
2923 ns->sws = le32_to_cpu(s.sws);
2924 ns->sgs = le16_to_cpu(s.sgs);
2927 unsigned int bs = 1 << ns->lba_shift;
2929 blk_queue_io_min(ns->queue, bs * ns->sws);
2931 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
2937 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2940 struct gendisk *disk;
2941 struct nvme_id_ns *id;
2942 char disk_name[DISK_NAME_LEN];
2943 int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
2946 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2950 ns->queue = blk_mq_init_queue(ctrl->tagset);
2951 if (IS_ERR(ns->queue))
2953 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2954 ns->queue->queuedata = ns;
2957 kref_init(&ns->kref);
2958 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2960 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2961 nvme_set_queue_limits(ctrl, ns->queue);
2963 id = nvme_identify_ns(ctrl, nsid);
2965 goto out_free_queue;
2970 if (nvme_init_ns_head(ns, nsid, id, &new))
2972 nvme_setup_streams_ns(ctrl, ns);
2974 #ifdef CONFIG_NVME_MULTIPATH
2976 * If multipathing is enabled we need to always use the subsystem
2977 * instance number for numbering our devices to avoid conflicts
2978 * between subsystems that have multiple controllers and thus use
2979 * the multipath-aware subsystem node and those that have a single
2980 * controller and use the controller node directly.
2982 if (ns->head->disk) {
2983 sprintf(disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
2984 ctrl->cntlid, ns->head->instance);
2985 flags = GENHD_FL_HIDDEN;
2987 sprintf(disk_name, "nvme%dn%d", ctrl->subsys->instance,
2988 ns->head->instance);
2992 * But without the multipath code enabled, multiple controller per
2993 * subsystems are visible as devices and thus we cannot use the
2994 * subsystem instance.
2996 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
2999 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3000 if (nvme_nvm_register(ns, disk_name, node)) {
3001 dev_warn(ctrl->device, "LightNVM init failure\n");
3006 disk = alloc_disk_node(0, node);
3010 disk->fops = &nvme_fops;
3011 disk->private_data = ns;
3012 disk->queue = ns->queue;
3013 disk->flags = flags;
3014 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3017 __nvme_revalidate_disk(disk, id);
3019 mutex_lock(&ctrl->namespaces_mutex);
3020 list_add_tail(&ns->list, &ctrl->namespaces);
3021 mutex_unlock(&ctrl->namespaces_mutex);
3023 nvme_get_ctrl(ctrl);
3027 device_add_disk(ctrl->device, ns->disk);
3028 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
3029 &nvme_ns_id_attr_group))
3030 pr_warn("%s: failed to create sysfs group for identification\n",
3031 ns->disk->disk_name);
3032 if (ns->ndev && nvme_nvm_register_sysfs(ns))
3033 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
3034 ns->disk->disk_name);
3037 nvme_mpath_add_disk(ns->head);
3038 nvme_mpath_add_disk_links(ns);
3041 mutex_lock(&ctrl->subsys->lock);
3042 list_del_rcu(&ns->siblings);
3043 mutex_unlock(&ctrl->subsys->lock);
3047 blk_cleanup_queue(ns->queue);
3052 static void nvme_ns_remove(struct nvme_ns *ns)
3054 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3057 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3058 nvme_mpath_remove_disk_links(ns);
3059 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
3060 &nvme_ns_id_attr_group);
3062 nvme_nvm_unregister_sysfs(ns);
3063 del_gendisk(ns->disk);
3064 blk_cleanup_queue(ns->queue);
3065 if (blk_get_integrity(ns->disk))
3066 blk_integrity_unregister(ns->disk);
3069 mutex_lock(&ns->ctrl->subsys->lock);
3070 nvme_mpath_clear_current_path(ns);
3071 list_del_rcu(&ns->siblings);
3072 mutex_unlock(&ns->ctrl->subsys->lock);
3074 mutex_lock(&ns->ctrl->namespaces_mutex);
3075 list_del_init(&ns->list);
3076 mutex_unlock(&ns->ctrl->namespaces_mutex);
3078 synchronize_srcu(&ns->head->srcu);
3079 nvme_mpath_check_last_path(ns);
3083 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3087 ns = nvme_find_get_ns(ctrl, nsid);
3089 if (ns->disk && revalidate_disk(ns->disk))
3093 nvme_alloc_ns(ctrl, nsid);
3096 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3099 struct nvme_ns *ns, *next;
3101 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3102 if (ns->head->ns_id > nsid)
3107 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3111 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3114 ns_list = kzalloc(0x1000, GFP_KERNEL);
3118 for (i = 0; i < num_lists; i++) {
3119 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3123 for (j = 0; j < min(nn, 1024U); j++) {
3124 nsid = le32_to_cpu(ns_list[j]);
3128 nvme_validate_ns(ctrl, nsid);
3130 while (++prev < nsid) {
3131 ns = nvme_find_get_ns(ctrl, prev);
3141 nvme_remove_invalid_namespaces(ctrl, prev);
3147 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3151 for (i = 1; i <= nn; i++)
3152 nvme_validate_ns(ctrl, i);
3154 nvme_remove_invalid_namespaces(ctrl, nn);
3157 static void nvme_scan_work(struct work_struct *work)
3159 struct nvme_ctrl *ctrl =
3160 container_of(work, struct nvme_ctrl, scan_work);
3161 struct nvme_id_ctrl *id;
3164 if (ctrl->state != NVME_CTRL_LIVE)
3167 WARN_ON_ONCE(!ctrl->tagset);
3169 if (nvme_identify_ctrl(ctrl, &id))
3172 nn = le32_to_cpu(id->nn);
3173 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3174 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3175 if (!nvme_scan_ns_list(ctrl, nn))
3178 nvme_scan_ns_sequential(ctrl, nn);
3180 mutex_lock(&ctrl->namespaces_mutex);
3181 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3182 mutex_unlock(&ctrl->namespaces_mutex);
3186 void nvme_queue_scan(struct nvme_ctrl *ctrl)
3189 * Only new queue scan work when admin and IO queues are both alive
3191 if (ctrl->state == NVME_CTRL_LIVE)
3192 queue_work(nvme_wq, &ctrl->scan_work);
3194 EXPORT_SYMBOL_GPL(nvme_queue_scan);
3197 * This function iterates the namespace list unlocked to allow recovery from
3198 * controller failure. It is up to the caller to ensure the namespace list is
3199 * not modified by scan work while this function is executing.
3201 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3203 struct nvme_ns *ns, *next;
3206 * The dead states indicates the controller was not gracefully
3207 * disconnected. In that case, we won't be able to flush any data while
3208 * removing the namespaces' disks; fail all the queues now to avoid
3209 * potentially having to clean up the failed sync later.
3211 if (ctrl->state == NVME_CTRL_DEAD)
3212 nvme_kill_queues(ctrl);
3214 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
3217 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3219 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3221 char *envp[2] = { NULL, NULL };
3222 u32 aen_result = ctrl->aen_result;
3224 ctrl->aen_result = 0;
3228 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3231 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3235 static void nvme_async_event_work(struct work_struct *work)
3237 struct nvme_ctrl *ctrl =
3238 container_of(work, struct nvme_ctrl, async_event_work);
3240 nvme_aen_uevent(ctrl);
3241 ctrl->ops->submit_async_event(ctrl);
3244 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3249 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3255 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3258 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3260 struct nvme_fw_slot_info_log *log;
3262 log = kmalloc(sizeof(*log), GFP_KERNEL);
3266 if (nvme_get_log(ctrl, NVME_LOG_FW_SLOT, log, sizeof(*log)))
3267 dev_warn(ctrl->device,
3268 "Get FW SLOT INFO log error\n");
3272 static void nvme_fw_act_work(struct work_struct *work)
3274 struct nvme_ctrl *ctrl = container_of(work,
3275 struct nvme_ctrl, fw_act_work);
3276 unsigned long fw_act_timeout;
3279 fw_act_timeout = jiffies +
3280 msecs_to_jiffies(ctrl->mtfa * 100);
3282 fw_act_timeout = jiffies +
3283 msecs_to_jiffies(admin_timeout * 1000);
3285 nvme_stop_queues(ctrl);
3286 while (nvme_ctrl_pp_status(ctrl)) {
3287 if (time_after(jiffies, fw_act_timeout)) {
3288 dev_warn(ctrl->device,
3289 "Fw activation timeout, reset controller\n");
3290 nvme_reset_ctrl(ctrl);
3296 if (ctrl->state != NVME_CTRL_LIVE)
3299 nvme_start_queues(ctrl);
3300 /* read FW slot information to clear the AER */
3301 nvme_get_fw_slot_info(ctrl);
3304 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3305 union nvme_result *res)
3307 u32 result = le32_to_cpu(res->u32);
3309 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3312 switch (result & 0x7) {
3313 case NVME_AER_ERROR:
3314 case NVME_AER_SMART:
3317 ctrl->aen_result = result;
3323 switch (result & 0xff07) {
3324 case NVME_AER_NOTICE_NS_CHANGED:
3325 dev_info(ctrl->device, "rescanning\n");
3326 nvme_queue_scan(ctrl);
3328 case NVME_AER_NOTICE_FW_ACT_STARTING:
3329 queue_work(nvme_wq, &ctrl->fw_act_work);
3332 dev_warn(ctrl->device, "async event result %08x\n", result);
3334 queue_work(nvme_wq, &ctrl->async_event_work);
3336 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3338 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3340 nvme_stop_keep_alive(ctrl);
3341 flush_work(&ctrl->async_event_work);
3342 flush_work(&ctrl->scan_work);
3343 cancel_work_sync(&ctrl->fw_act_work);
3345 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3347 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3350 nvme_start_keep_alive(ctrl);
3352 if (ctrl->queue_count > 1) {
3353 nvme_queue_scan(ctrl);
3354 queue_work(nvme_wq, &ctrl->async_event_work);
3355 nvme_start_queues(ctrl);
3358 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3360 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3362 cdev_device_del(&ctrl->cdev, ctrl->device);
3364 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3366 static void nvme_free_ctrl(struct device *dev)
3368 struct nvme_ctrl *ctrl =
3369 container_of(dev, struct nvme_ctrl, ctrl_device);
3370 struct nvme_subsystem *subsys = ctrl->subsys;
3372 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3373 kfree(ctrl->effects);
3376 mutex_lock(&subsys->lock);
3377 list_del(&ctrl->subsys_entry);
3378 mutex_unlock(&subsys->lock);
3379 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3382 ctrl->ops->free_ctrl(ctrl);
3385 nvme_put_subsystem(subsys);
3389 * Initialize a NVMe controller structures. This needs to be called during
3390 * earliest initialization so that we have the initialized structured around
3393 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3394 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3398 ctrl->state = NVME_CTRL_NEW;
3399 spin_lock_init(&ctrl->lock);
3400 INIT_LIST_HEAD(&ctrl->namespaces);
3401 mutex_init(&ctrl->namespaces_mutex);
3404 ctrl->quirks = quirks;
3405 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3406 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3407 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3408 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3410 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3413 ctrl->instance = ret;
3415 device_initialize(&ctrl->ctrl_device);
3416 ctrl->device = &ctrl->ctrl_device;
3417 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3418 ctrl->device->class = nvme_class;
3419 ctrl->device->parent = ctrl->dev;
3420 ctrl->device->groups = nvme_dev_attr_groups;
3421 ctrl->device->release = nvme_free_ctrl;
3422 dev_set_drvdata(ctrl->device, ctrl);
3423 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3425 goto out_release_instance;
3427 cdev_init(&ctrl->cdev, &nvme_dev_fops);
3428 ctrl->cdev.owner = ops->module;
3429 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3434 * Initialize latency tolerance controls. The sysfs files won't
3435 * be visible to userspace unless the device actually supports APST.
3437 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3438 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3439 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3443 kfree_const(dev->kobj.name);
3444 out_release_instance:
3445 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3449 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3452 * nvme_kill_queues(): Ends all namespace queues
3453 * @ctrl: the dead controller that needs to end
3455 * Call this function when the driver determines it is unable to get the
3456 * controller in a state capable of servicing IO.
3458 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3462 mutex_lock(&ctrl->namespaces_mutex);
3464 /* Forcibly unquiesce queues to avoid blocking dispatch */
3466 blk_mq_unquiesce_queue(ctrl->admin_q);
3468 list_for_each_entry(ns, &ctrl->namespaces, list) {
3470 * Revalidating a dead namespace sets capacity to 0. This will
3471 * end buffered writers dirtying pages that can't be synced.
3473 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
3475 revalidate_disk(ns->disk);
3476 blk_set_queue_dying(ns->queue);
3478 /* Forcibly unquiesce queues to avoid blocking dispatch */
3479 blk_mq_unquiesce_queue(ns->queue);
3481 mutex_unlock(&ctrl->namespaces_mutex);
3483 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3485 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3489 mutex_lock(&ctrl->namespaces_mutex);
3490 list_for_each_entry(ns, &ctrl->namespaces, list)
3491 blk_mq_unfreeze_queue(ns->queue);
3492 mutex_unlock(&ctrl->namespaces_mutex);
3494 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3496 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3500 mutex_lock(&ctrl->namespaces_mutex);
3501 list_for_each_entry(ns, &ctrl->namespaces, list) {
3502 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3506 mutex_unlock(&ctrl->namespaces_mutex);
3508 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3510 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3514 mutex_lock(&ctrl->namespaces_mutex);
3515 list_for_each_entry(ns, &ctrl->namespaces, list)
3516 blk_mq_freeze_queue_wait(ns->queue);
3517 mutex_unlock(&ctrl->namespaces_mutex);
3519 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3521 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3525 mutex_lock(&ctrl->namespaces_mutex);
3526 list_for_each_entry(ns, &ctrl->namespaces, list)
3527 blk_freeze_queue_start(ns->queue);
3528 mutex_unlock(&ctrl->namespaces_mutex);
3530 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3532 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3536 mutex_lock(&ctrl->namespaces_mutex);
3537 list_for_each_entry(ns, &ctrl->namespaces, list)
3538 blk_mq_quiesce_queue(ns->queue);
3539 mutex_unlock(&ctrl->namespaces_mutex);
3541 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3543 void nvme_start_queues(struct nvme_ctrl *ctrl)
3547 mutex_lock(&ctrl->namespaces_mutex);
3548 list_for_each_entry(ns, &ctrl->namespaces, list)
3549 blk_mq_unquiesce_queue(ns->queue);
3550 mutex_unlock(&ctrl->namespaces_mutex);
3552 EXPORT_SYMBOL_GPL(nvme_start_queues);
3554 int nvme_reinit_tagset(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set)
3556 if (!ctrl->ops->reinit_request)
3559 return blk_mq_tagset_iter(set, set->driver_data,
3560 ctrl->ops->reinit_request);
3562 EXPORT_SYMBOL_GPL(nvme_reinit_tagset);
3564 int __init nvme_core_init(void)
3566 int result = -ENOMEM;
3568 nvme_wq = alloc_workqueue("nvme-wq",
3569 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3573 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3574 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3578 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3579 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3580 if (!nvme_delete_wq)
3581 goto destroy_reset_wq;
3583 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3585 goto destroy_delete_wq;
3587 nvme_class = class_create(THIS_MODULE, "nvme");
3588 if (IS_ERR(nvme_class)) {
3589 result = PTR_ERR(nvme_class);
3590 goto unregister_chrdev;
3593 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3594 if (IS_ERR(nvme_subsys_class)) {
3595 result = PTR_ERR(nvme_subsys_class);
3601 class_destroy(nvme_class);
3603 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3605 destroy_workqueue(nvme_delete_wq);
3607 destroy_workqueue(nvme_reset_wq);
3609 destroy_workqueue(nvme_wq);
3614 void nvme_core_exit(void)
3616 ida_destroy(&nvme_subsystems_ida);
3617 class_destroy(nvme_subsys_class);
3618 class_destroy(nvme_class);
3619 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3620 destroy_workqueue(nvme_delete_wq);
3621 destroy_workqueue(nvme_reset_wq);
3622 destroy_workqueue(nvme_wq);
3625 MODULE_LICENSE("GPL");
3626 MODULE_VERSION("1.0");
3627 module_init(nvme_core_init);
3628 module_exit(nvme_core_exit);