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>
28 #define CREATE_TRACE_POINTS
31 #define NVME_MINORS (1U << MINORBITS)
33 unsigned int admin_timeout = 60;
34 module_param(admin_timeout, uint, 0644);
35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
36 EXPORT_SYMBOL_GPL(admin_timeout);
38 unsigned int nvme_io_timeout = 30;
39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
43 static unsigned char shutdown_timeout = 5;
44 module_param(shutdown_timeout, byte, 0644);
45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
47 static u8 nvme_max_retries = 5;
48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
51 static unsigned long default_ps_max_latency_us = 100000;
52 module_param(default_ps_max_latency_us, ulong, 0644);
53 MODULE_PARM_DESC(default_ps_max_latency_us,
54 "max power saving latency for new devices; use PM QOS to change per device");
56 static bool force_apst;
57 module_param(force_apst, bool, 0644);
58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
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 LIST_HEAD(nvme_subsystems);
85 static DEFINE_MUTEX(nvme_subsystems_lock);
87 static DEFINE_IDA(nvme_instance_ida);
88 static dev_t nvme_chr_devt;
89 static struct class *nvme_class;
90 static struct class *nvme_subsys_class;
92 static int nvme_revalidate_disk(struct gendisk *disk);
93 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
94 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
97 static void nvme_set_queue_dying(struct nvme_ns *ns)
100 * Revalidating a dead namespace sets capacity to 0. This will end
101 * buffered writers dirtying pages that can't be synced.
103 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
105 blk_set_queue_dying(ns->queue);
106 /* Forcibly unquiesce queues to avoid blocking dispatch */
107 blk_mq_unquiesce_queue(ns->queue);
109 * Revalidate after unblocking dispatchers that may be holding bd_butex
111 revalidate_disk(ns->disk);
114 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
117 * Only new queue scan work when admin and IO queues are both alive
119 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
120 queue_work(nvme_wq, &ctrl->scan_work);
123 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
125 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
127 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
131 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
133 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
137 ret = nvme_reset_ctrl(ctrl);
139 flush_work(&ctrl->reset_work);
140 if (ctrl->state != NVME_CTRL_LIVE)
146 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
148 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
150 dev_info(ctrl->device,
151 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
153 flush_work(&ctrl->reset_work);
154 nvme_stop_ctrl(ctrl);
155 nvme_remove_namespaces(ctrl);
156 ctrl->ops->delete_ctrl(ctrl);
157 nvme_uninit_ctrl(ctrl);
161 static void nvme_delete_ctrl_work(struct work_struct *work)
163 struct nvme_ctrl *ctrl =
164 container_of(work, struct nvme_ctrl, delete_work);
166 nvme_do_delete_ctrl(ctrl);
169 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
171 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
173 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
177 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
179 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
184 * Keep a reference until nvme_do_delete_ctrl() complete,
185 * since ->delete_ctrl can free the controller.
188 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
191 nvme_do_delete_ctrl(ctrl);
196 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
198 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
201 static blk_status_t nvme_error_status(u16 status)
203 switch (status & 0x7ff) {
204 case NVME_SC_SUCCESS:
206 case NVME_SC_CAP_EXCEEDED:
207 return BLK_STS_NOSPC;
208 case NVME_SC_LBA_RANGE:
209 return BLK_STS_TARGET;
210 case NVME_SC_BAD_ATTRIBUTES:
211 case NVME_SC_ONCS_NOT_SUPPORTED:
212 case NVME_SC_INVALID_OPCODE:
213 case NVME_SC_INVALID_FIELD:
214 case NVME_SC_INVALID_NS:
215 return BLK_STS_NOTSUPP;
216 case NVME_SC_WRITE_FAULT:
217 case NVME_SC_READ_ERROR:
218 case NVME_SC_UNWRITTEN_BLOCK:
219 case NVME_SC_ACCESS_DENIED:
220 case NVME_SC_READ_ONLY:
221 case NVME_SC_COMPARE_FAILED:
222 return BLK_STS_MEDIUM;
223 case NVME_SC_GUARD_CHECK:
224 case NVME_SC_APPTAG_CHECK:
225 case NVME_SC_REFTAG_CHECK:
226 case NVME_SC_INVALID_PI:
227 return BLK_STS_PROTECTION;
228 case NVME_SC_RESERVATION_CONFLICT:
229 return BLK_STS_NEXUS;
230 case NVME_SC_HOST_PATH_ERROR:
231 return BLK_STS_TRANSPORT;
233 return BLK_STS_IOERR;
237 static inline bool nvme_req_needs_retry(struct request *req)
239 if (blk_noretry_request(req))
241 if (nvme_req(req)->status & NVME_SC_DNR)
243 if (nvme_req(req)->retries >= nvme_max_retries)
248 static void nvme_retry_req(struct request *req)
250 struct nvme_ns *ns = req->q->queuedata;
251 unsigned long delay = 0;
254 /* The mask and shift result must be <= 3 */
255 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
257 delay = ns->ctrl->crdt[crd - 1] * 100;
259 nvme_req(req)->retries++;
260 blk_mq_requeue_request(req, false);
261 blk_mq_delay_kick_requeue_list(req->q, delay);
264 void nvme_complete_rq(struct request *req)
266 blk_status_t status = nvme_error_status(nvme_req(req)->status);
268 trace_nvme_complete_rq(req);
270 if (nvme_req(req)->ctrl->kas)
271 nvme_req(req)->ctrl->comp_seen = true;
273 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
274 if ((req->cmd_flags & REQ_NVME_MPATH) &&
275 blk_path_error(status)) {
276 nvme_failover_req(req);
280 if (!blk_queue_dying(req->q)) {
286 nvme_trace_bio_complete(req, status);
287 blk_mq_end_request(req, status);
289 EXPORT_SYMBOL_GPL(nvme_complete_rq);
291 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
293 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
294 "Cancelling I/O %d", req->tag);
296 /* don't abort one completed request */
297 if (blk_mq_request_completed(req))
300 nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
301 blk_mq_complete_request(req);
304 EXPORT_SYMBOL_GPL(nvme_cancel_request);
306 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
307 enum nvme_ctrl_state new_state)
309 enum nvme_ctrl_state old_state;
311 bool changed = false;
313 spin_lock_irqsave(&ctrl->lock, flags);
315 old_state = ctrl->state;
320 case NVME_CTRL_RESETTING:
321 case NVME_CTRL_CONNECTING:
328 case NVME_CTRL_RESETTING:
338 case NVME_CTRL_CONNECTING:
341 case NVME_CTRL_RESETTING:
348 case NVME_CTRL_DELETING:
351 case NVME_CTRL_RESETTING:
352 case NVME_CTRL_CONNECTING:
361 case NVME_CTRL_DELETING:
373 ctrl->state = new_state;
375 spin_unlock_irqrestore(&ctrl->lock, flags);
376 if (changed && ctrl->state == NVME_CTRL_LIVE)
377 nvme_kick_requeue_lists(ctrl);
380 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
382 static void nvme_free_ns_head(struct kref *ref)
384 struct nvme_ns_head *head =
385 container_of(ref, struct nvme_ns_head, ref);
387 nvme_mpath_remove_disk(head);
388 ida_simple_remove(&head->subsys->ns_ida, head->instance);
389 list_del_init(&head->entry);
390 cleanup_srcu_struct(&head->srcu);
391 nvme_put_subsystem(head->subsys);
395 static void nvme_put_ns_head(struct nvme_ns_head *head)
397 kref_put(&head->ref, nvme_free_ns_head);
400 static void nvme_free_ns(struct kref *kref)
402 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
405 nvme_nvm_unregister(ns);
408 nvme_put_ns_head(ns->head);
409 nvme_put_ctrl(ns->ctrl);
413 static void nvme_put_ns(struct nvme_ns *ns)
415 kref_put(&ns->kref, nvme_free_ns);
418 static inline void nvme_clear_nvme_request(struct request *req)
420 if (!(req->rq_flags & RQF_DONTPREP)) {
421 nvme_req(req)->retries = 0;
422 nvme_req(req)->flags = 0;
423 req->rq_flags |= RQF_DONTPREP;
427 struct request *nvme_alloc_request(struct request_queue *q,
428 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
430 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
433 if (qid == NVME_QID_ANY) {
434 req = blk_mq_alloc_request(q, op, flags);
436 req = blk_mq_alloc_request_hctx(q, op, flags,
442 req->cmd_flags |= REQ_FAILFAST_DRIVER;
443 nvme_clear_nvme_request(req);
444 nvme_req(req)->cmd = cmd;
448 EXPORT_SYMBOL_GPL(nvme_alloc_request);
450 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
452 struct nvme_command c;
454 memset(&c, 0, sizeof(c));
456 c.directive.opcode = nvme_admin_directive_send;
457 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
458 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
459 c.directive.dtype = NVME_DIR_IDENTIFY;
460 c.directive.tdtype = NVME_DIR_STREAMS;
461 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
463 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
466 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
468 return nvme_toggle_streams(ctrl, false);
471 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
473 return nvme_toggle_streams(ctrl, true);
476 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
477 struct streams_directive_params *s, u32 nsid)
479 struct nvme_command c;
481 memset(&c, 0, sizeof(c));
482 memset(s, 0, sizeof(*s));
484 c.directive.opcode = nvme_admin_directive_recv;
485 c.directive.nsid = cpu_to_le32(nsid);
486 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
487 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
488 c.directive.dtype = NVME_DIR_STREAMS;
490 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
493 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
495 struct streams_directive_params s;
498 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
503 ret = nvme_enable_streams(ctrl);
507 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
511 ctrl->nssa = le16_to_cpu(s.nssa);
512 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
513 dev_info(ctrl->device, "too few streams (%u) available\n",
515 nvme_disable_streams(ctrl);
519 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
520 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
525 * Check if 'req' has a write hint associated with it. If it does, assign
526 * a valid namespace stream to the write.
528 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
529 struct request *req, u16 *control,
532 enum rw_hint streamid = req->write_hint;
534 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
538 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
541 *control |= NVME_RW_DTYPE_STREAMS;
542 *dsmgmt |= streamid << 16;
545 if (streamid < ARRAY_SIZE(req->q->write_hints))
546 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
549 static inline void nvme_setup_flush(struct nvme_ns *ns,
550 struct nvme_command *cmnd)
552 cmnd->common.opcode = nvme_cmd_flush;
553 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
556 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
557 struct nvme_command *cmnd)
559 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
560 struct nvme_dsm_range *range;
563 range = kmalloc_array(segments, sizeof(*range),
564 GFP_ATOMIC | __GFP_NOWARN);
567 * If we fail allocation our range, fallback to the controller
568 * discard page. If that's also busy, it's safe to return
569 * busy, as we know we can make progress once that's freed.
571 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
572 return BLK_STS_RESOURCE;
574 range = page_address(ns->ctrl->discard_page);
577 __rq_for_each_bio(bio, req) {
578 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
579 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
582 range[n].cattr = cpu_to_le32(0);
583 range[n].nlb = cpu_to_le32(nlb);
584 range[n].slba = cpu_to_le64(slba);
589 if (WARN_ON_ONCE(n != segments)) {
590 if (virt_to_page(range) == ns->ctrl->discard_page)
591 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
594 return BLK_STS_IOERR;
597 cmnd->dsm.opcode = nvme_cmd_dsm;
598 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
599 cmnd->dsm.nr = cpu_to_le32(segments - 1);
600 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
602 req->special_vec.bv_page = virt_to_page(range);
603 req->special_vec.bv_offset = offset_in_page(range);
604 req->special_vec.bv_len = sizeof(*range) * segments;
605 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
610 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
611 struct request *req, struct nvme_command *cmnd)
613 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
614 return nvme_setup_discard(ns, req, cmnd);
616 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
617 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
618 cmnd->write_zeroes.slba =
619 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
620 cmnd->write_zeroes.length =
621 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
622 cmnd->write_zeroes.control = 0;
626 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
627 struct request *req, struct nvme_command *cmnd)
629 struct nvme_ctrl *ctrl = ns->ctrl;
633 if (req->cmd_flags & REQ_FUA)
634 control |= NVME_RW_FUA;
635 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
636 control |= NVME_RW_LR;
638 if (req->cmd_flags & REQ_RAHEAD)
639 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
641 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
642 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
643 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
644 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
646 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
647 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
651 * If formated with metadata, the block layer always provides a
652 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
653 * we enable the PRACT bit for protection information or set the
654 * namespace capacity to zero to prevent any I/O.
656 if (!blk_integrity_rq(req)) {
657 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
658 return BLK_STS_NOTSUPP;
659 control |= NVME_RW_PRINFO_PRACT;
662 switch (ns->pi_type) {
663 case NVME_NS_DPS_PI_TYPE3:
664 control |= NVME_RW_PRINFO_PRCHK_GUARD;
666 case NVME_NS_DPS_PI_TYPE1:
667 case NVME_NS_DPS_PI_TYPE2:
668 control |= NVME_RW_PRINFO_PRCHK_GUARD |
669 NVME_RW_PRINFO_PRCHK_REF;
670 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
675 cmnd->rw.control = cpu_to_le16(control);
676 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
680 void nvme_cleanup_cmd(struct request *req)
682 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
683 struct nvme_ns *ns = req->rq_disk->private_data;
684 struct page *page = req->special_vec.bv_page;
686 if (page == ns->ctrl->discard_page)
687 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
689 kfree(page_address(page) + req->special_vec.bv_offset);
692 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
694 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
695 struct nvme_command *cmd)
697 blk_status_t ret = BLK_STS_OK;
699 nvme_clear_nvme_request(req);
701 memset(cmd, 0, sizeof(*cmd));
702 switch (req_op(req)) {
705 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
708 nvme_setup_flush(ns, cmd);
710 case REQ_OP_WRITE_ZEROES:
711 ret = nvme_setup_write_zeroes(ns, req, cmd);
714 ret = nvme_setup_discard(ns, req, cmd);
718 ret = nvme_setup_rw(ns, req, cmd);
722 return BLK_STS_IOERR;
725 cmd->common.command_id = req->tag;
726 trace_nvme_setup_cmd(req, cmd);
729 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
731 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
733 struct completion *waiting = rq->end_io_data;
735 rq->end_io_data = NULL;
739 static void nvme_execute_rq_polled(struct request_queue *q,
740 struct gendisk *bd_disk, struct request *rq, int at_head)
742 DECLARE_COMPLETION_ONSTACK(wait);
744 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
746 rq->cmd_flags |= REQ_HIPRI;
747 rq->end_io_data = &wait;
748 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
750 while (!completion_done(&wait)) {
751 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
757 * Returns 0 on success. If the result is negative, it's a Linux error code;
758 * if the result is positive, it's an NVM Express status code
760 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
761 union nvme_result *result, void *buffer, unsigned bufflen,
762 unsigned timeout, int qid, int at_head,
763 blk_mq_req_flags_t flags, bool poll)
768 req = nvme_alloc_request(q, cmd, flags, qid);
772 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
774 if (buffer && bufflen) {
775 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
781 nvme_execute_rq_polled(req->q, NULL, req, at_head);
783 blk_execute_rq(req->q, NULL, req, at_head);
785 *result = nvme_req(req)->result;
786 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
789 ret = nvme_req(req)->status;
791 blk_mq_free_request(req);
794 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
796 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
797 void *buffer, unsigned bufflen)
799 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
800 NVME_QID_ANY, 0, 0, false);
802 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
804 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
805 unsigned len, u32 seed, bool write)
807 struct bio_integrity_payload *bip;
811 buf = kmalloc(len, GFP_KERNEL);
816 if (write && copy_from_user(buf, ubuf, len))
819 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
825 bip->bip_iter.bi_size = len;
826 bip->bip_iter.bi_sector = seed;
827 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
828 offset_in_page(buf));
838 static int nvme_submit_user_cmd(struct request_queue *q,
839 struct nvme_command *cmd, void __user *ubuffer,
840 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
841 u32 meta_seed, u64 *result, unsigned timeout)
843 bool write = nvme_is_write(cmd);
844 struct nvme_ns *ns = q->queuedata;
845 struct gendisk *disk = ns ? ns->disk : NULL;
847 struct bio *bio = NULL;
851 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
855 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
856 nvme_req(req)->flags |= NVME_REQ_USERCMD;
858 if (ubuffer && bufflen) {
859 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
865 if (disk && meta_buffer && meta_len) {
866 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
872 req->cmd_flags |= REQ_INTEGRITY;
876 blk_execute_rq(req->q, disk, req, 0);
877 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
880 ret = nvme_req(req)->status;
882 *result = le64_to_cpu(nvme_req(req)->result.u64);
883 if (meta && !ret && !write) {
884 if (copy_to_user(meta_buffer, meta, meta_len))
890 blk_rq_unmap_user(bio);
892 blk_mq_free_request(req);
896 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
898 struct nvme_ctrl *ctrl = rq->end_io_data;
900 bool startka = false;
902 blk_mq_free_request(rq);
905 dev_err(ctrl->device,
906 "failed nvme_keep_alive_end_io error=%d\n",
911 ctrl->comp_seen = false;
912 spin_lock_irqsave(&ctrl->lock, flags);
913 if (ctrl->state == NVME_CTRL_LIVE ||
914 ctrl->state == NVME_CTRL_CONNECTING)
916 spin_unlock_irqrestore(&ctrl->lock, flags);
918 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
921 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
925 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
930 rq->timeout = ctrl->kato * HZ;
931 rq->end_io_data = ctrl;
933 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
938 static void nvme_keep_alive_work(struct work_struct *work)
940 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
941 struct nvme_ctrl, ka_work);
942 bool comp_seen = ctrl->comp_seen;
944 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
945 dev_dbg(ctrl->device,
946 "reschedule traffic based keep-alive timer\n");
947 ctrl->comp_seen = false;
948 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
952 if (nvme_keep_alive(ctrl)) {
953 /* allocation failure, reset the controller */
954 dev_err(ctrl->device, "keep-alive failed\n");
955 nvme_reset_ctrl(ctrl);
960 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
962 if (unlikely(ctrl->kato == 0))
965 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
968 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
970 if (unlikely(ctrl->kato == 0))
973 cancel_delayed_work_sync(&ctrl->ka_work);
975 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
977 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
979 struct nvme_command c = { };
982 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
983 c.identify.opcode = nvme_admin_identify;
984 c.identify.cns = NVME_ID_CNS_CTRL;
986 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
990 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
991 sizeof(struct nvme_id_ctrl));
997 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
998 struct nvme_ns_ids *ids)
1000 struct nvme_command c = { };
1006 c.identify.opcode = nvme_admin_identify;
1007 c.identify.nsid = cpu_to_le32(nsid);
1008 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1010 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1014 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1015 NVME_IDENTIFY_DATA_SIZE);
1019 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1020 struct nvme_ns_id_desc *cur = data + pos;
1025 switch (cur->nidt) {
1026 case NVME_NIDT_EUI64:
1027 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1028 dev_warn(ctrl->device,
1029 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1033 len = NVME_NIDT_EUI64_LEN;
1034 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1036 case NVME_NIDT_NGUID:
1037 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1038 dev_warn(ctrl->device,
1039 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1043 len = NVME_NIDT_NGUID_LEN;
1044 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1046 case NVME_NIDT_UUID:
1047 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1048 dev_warn(ctrl->device,
1049 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1053 len = NVME_NIDT_UUID_LEN;
1054 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1057 /* Skip unknown types */
1062 len += sizeof(*cur);
1069 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1071 struct nvme_command c = { };
1073 c.identify.opcode = nvme_admin_identify;
1074 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1075 c.identify.nsid = cpu_to_le32(nsid);
1076 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1077 NVME_IDENTIFY_DATA_SIZE);
1080 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1081 unsigned nsid, struct nvme_id_ns **id)
1083 struct nvme_command c = { };
1086 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1087 c.identify.opcode = nvme_admin_identify;
1088 c.identify.nsid = cpu_to_le32(nsid);
1089 c.identify.cns = NVME_ID_CNS_NS;
1091 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1095 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1097 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1104 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1105 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1107 struct nvme_command c;
1108 union nvme_result res;
1111 memset(&c, 0, sizeof(c));
1112 c.features.opcode = op;
1113 c.features.fid = cpu_to_le32(fid);
1114 c.features.dword11 = cpu_to_le32(dword11);
1116 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1117 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1118 if (ret >= 0 && result)
1119 *result = le32_to_cpu(res.u32);
1123 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1124 unsigned int dword11, void *buffer, size_t buflen,
1127 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1130 EXPORT_SYMBOL_GPL(nvme_set_features);
1132 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1133 unsigned int dword11, void *buffer, size_t buflen,
1136 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1139 EXPORT_SYMBOL_GPL(nvme_get_features);
1141 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1143 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1145 int status, nr_io_queues;
1147 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1153 * Degraded controllers might return an error when setting the queue
1154 * count. We still want to be able to bring them online and offer
1155 * access to the admin queue, as that might be only way to fix them up.
1158 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1161 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1162 *count = min(*count, nr_io_queues);
1167 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1169 #define NVME_AEN_SUPPORTED \
1170 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1171 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1173 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1175 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1178 if (!supported_aens)
1181 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1184 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1187 queue_work(nvme_wq, &ctrl->async_event_work);
1190 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1192 struct nvme_user_io io;
1193 struct nvme_command c;
1194 unsigned length, meta_len;
1195 void __user *metadata;
1197 if (copy_from_user(&io, uio, sizeof(io)))
1202 switch (io.opcode) {
1203 case nvme_cmd_write:
1205 case nvme_cmd_compare:
1211 length = (io.nblocks + 1) << ns->lba_shift;
1212 meta_len = (io.nblocks + 1) * ns->ms;
1213 metadata = (void __user *)(uintptr_t)io.metadata;
1218 } else if (meta_len) {
1219 if ((io.metadata & 3) || !io.metadata)
1223 memset(&c, 0, sizeof(c));
1224 c.rw.opcode = io.opcode;
1225 c.rw.flags = io.flags;
1226 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1227 c.rw.slba = cpu_to_le64(io.slba);
1228 c.rw.length = cpu_to_le16(io.nblocks);
1229 c.rw.control = cpu_to_le16(io.control);
1230 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1231 c.rw.reftag = cpu_to_le32(io.reftag);
1232 c.rw.apptag = cpu_to_le16(io.apptag);
1233 c.rw.appmask = cpu_to_le16(io.appmask);
1235 return nvme_submit_user_cmd(ns->queue, &c,
1236 (void __user *)(uintptr_t)io.addr, length,
1237 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1240 static u32 nvme_known_admin_effects(u8 opcode)
1243 case nvme_admin_format_nvm:
1244 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1245 NVME_CMD_EFFECTS_CSE_MASK;
1246 case nvme_admin_sanitize_nvm:
1247 return NVME_CMD_EFFECTS_CSE_MASK;
1254 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1261 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1262 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1263 dev_warn(ctrl->device,
1264 "IO command:%02x has unhandled effects:%08x\n",
1270 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1271 effects |= nvme_known_admin_effects(opcode);
1274 * For simplicity, IO to all namespaces is quiesced even if the command
1275 * effects say only one namespace is affected.
1277 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1278 mutex_lock(&ctrl->scan_lock);
1279 mutex_lock(&ctrl->subsys->lock);
1280 nvme_mpath_start_freeze(ctrl->subsys);
1281 nvme_mpath_wait_freeze(ctrl->subsys);
1282 nvme_start_freeze(ctrl);
1283 nvme_wait_freeze(ctrl);
1288 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1292 down_read(&ctrl->namespaces_rwsem);
1293 list_for_each_entry(ns, &ctrl->namespaces, list)
1294 if (ns->disk && nvme_revalidate_disk(ns->disk))
1295 nvme_set_queue_dying(ns);
1296 up_read(&ctrl->namespaces_rwsem);
1299 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1302 * Revalidate LBA changes prior to unfreezing. This is necessary to
1303 * prevent memory corruption if a logical block size was changed by
1306 if (effects & NVME_CMD_EFFECTS_LBCC)
1307 nvme_update_formats(ctrl);
1308 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1309 nvme_unfreeze(ctrl);
1310 nvme_mpath_unfreeze(ctrl->subsys);
1311 mutex_unlock(&ctrl->subsys->lock);
1312 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1313 mutex_unlock(&ctrl->scan_lock);
1315 if (effects & NVME_CMD_EFFECTS_CCC)
1316 nvme_init_identify(ctrl);
1317 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1318 nvme_queue_scan(ctrl);
1321 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1322 struct nvme_passthru_cmd __user *ucmd)
1324 struct nvme_passthru_cmd cmd;
1325 struct nvme_command c;
1326 unsigned timeout = 0;
1331 if (!capable(CAP_SYS_ADMIN))
1333 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1338 memset(&c, 0, sizeof(c));
1339 c.common.opcode = cmd.opcode;
1340 c.common.flags = cmd.flags;
1341 c.common.nsid = cpu_to_le32(cmd.nsid);
1342 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1343 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1344 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1345 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1346 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1347 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1348 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1349 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1352 timeout = msecs_to_jiffies(cmd.timeout_ms);
1354 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1355 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1356 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1357 (void __user *)(uintptr_t)cmd.metadata,
1358 cmd.metadata_len, 0, &result, timeout);
1359 nvme_passthru_end(ctrl, effects);
1362 if (put_user(result, &ucmd->result))
1369 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1370 struct nvme_passthru_cmd64 __user *ucmd)
1372 struct nvme_passthru_cmd64 cmd;
1373 struct nvme_command c;
1374 unsigned timeout = 0;
1378 if (!capable(CAP_SYS_ADMIN))
1380 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1385 memset(&c, 0, sizeof(c));
1386 c.common.opcode = cmd.opcode;
1387 c.common.flags = cmd.flags;
1388 c.common.nsid = cpu_to_le32(cmd.nsid);
1389 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1390 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1391 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1392 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1393 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1394 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1395 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1396 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1399 timeout = msecs_to_jiffies(cmd.timeout_ms);
1401 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1402 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1403 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1404 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1405 0, &cmd.result, timeout);
1406 nvme_passthru_end(ctrl, effects);
1409 if (put_user(cmd.result, &ucmd->result))
1417 * Issue ioctl requests on the first available path. Note that unlike normal
1418 * block layer requests we will not retry failed request on another controller.
1420 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1421 struct nvme_ns_head **head, int *srcu_idx)
1423 #ifdef CONFIG_NVME_MULTIPATH
1424 if (disk->fops == &nvme_ns_head_ops) {
1427 *head = disk->private_data;
1428 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1429 ns = nvme_find_path(*head);
1431 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1437 return disk->private_data;
1440 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1443 srcu_read_unlock(&head->srcu, idx);
1446 static bool is_ctrl_ioctl(unsigned int cmd)
1448 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1450 if (is_sed_ioctl(cmd))
1455 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1457 struct nvme_ns_head *head,
1460 struct nvme_ctrl *ctrl = ns->ctrl;
1463 nvme_get_ctrl(ns->ctrl);
1464 nvme_put_ns_from_disk(head, srcu_idx);
1467 case NVME_IOCTL_ADMIN_CMD:
1468 ret = nvme_user_cmd(ctrl, NULL, argp);
1470 case NVME_IOCTL_ADMIN64_CMD:
1471 ret = nvme_user_cmd64(ctrl, NULL, argp);
1474 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1477 nvme_put_ctrl(ctrl);
1481 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1482 unsigned int cmd, unsigned long arg)
1484 struct nvme_ns_head *head = NULL;
1485 void __user *argp = (void __user *)arg;
1489 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1491 return -EWOULDBLOCK;
1494 * Handle ioctls that apply to the controller instead of the namespace
1495 * seperately and drop the ns SRCU reference early. This avoids a
1496 * deadlock when deleting namespaces using the passthrough interface.
1498 if (is_ctrl_ioctl(cmd))
1499 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1503 force_successful_syscall_return();
1504 ret = ns->head->ns_id;
1506 case NVME_IOCTL_IO_CMD:
1507 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1509 case NVME_IOCTL_SUBMIT_IO:
1510 ret = nvme_submit_io(ns, argp);
1512 case NVME_IOCTL_IO64_CMD:
1513 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1517 ret = nvme_nvm_ioctl(ns, cmd, arg);
1522 nvme_put_ns_from_disk(head, srcu_idx);
1526 static int nvme_open(struct block_device *bdev, fmode_t mode)
1528 struct nvme_ns *ns = bdev->bd_disk->private_data;
1530 #ifdef CONFIG_NVME_MULTIPATH
1531 /* should never be called due to GENHD_FL_HIDDEN */
1532 if (WARN_ON_ONCE(ns->head->disk))
1535 if (!kref_get_unless_zero(&ns->kref))
1537 if (!try_module_get(ns->ctrl->ops->module))
1548 static void nvme_release(struct gendisk *disk, fmode_t mode)
1550 struct nvme_ns *ns = disk->private_data;
1552 module_put(ns->ctrl->ops->module);
1556 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1558 /* some standard values */
1559 geo->heads = 1 << 6;
1560 geo->sectors = 1 << 5;
1561 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1565 #ifdef CONFIG_BLK_DEV_INTEGRITY
1566 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1568 struct blk_integrity integrity;
1570 memset(&integrity, 0, sizeof(integrity));
1572 case NVME_NS_DPS_PI_TYPE3:
1573 integrity.profile = &t10_pi_type3_crc;
1574 integrity.tag_size = sizeof(u16) + sizeof(u32);
1575 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1577 case NVME_NS_DPS_PI_TYPE1:
1578 case NVME_NS_DPS_PI_TYPE2:
1579 integrity.profile = &t10_pi_type1_crc;
1580 integrity.tag_size = sizeof(u16);
1581 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1584 integrity.profile = NULL;
1587 integrity.tuple_size = ms;
1588 blk_integrity_register(disk, &integrity);
1589 blk_queue_max_integrity_segments(disk->queue, 1);
1592 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1595 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1597 static void nvme_set_chunk_size(struct nvme_ns *ns)
1599 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1600 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1603 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1605 struct nvme_ctrl *ctrl = ns->ctrl;
1606 struct request_queue *queue = disk->queue;
1607 u32 size = queue_logical_block_size(queue);
1609 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1610 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1614 if (ctrl->nr_streams && ns->sws && ns->sgs)
1615 size *= ns->sws * ns->sgs;
1617 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1618 NVME_DSM_MAX_RANGES);
1620 queue->limits.discard_alignment = 0;
1621 queue->limits.discard_granularity = size;
1623 /* If discard is already enabled, don't reset queue limits */
1624 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1627 blk_queue_max_discard_sectors(queue, UINT_MAX);
1628 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1630 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1631 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1634 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1637 unsigned short bs = 1 << ns->lba_shift;
1639 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1640 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1643 * Even though NVMe spec explicitly states that MDTS is not
1644 * applicable to the write-zeroes:- "The restriction does not apply to
1645 * commands that do not transfer data between the host and the
1646 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1647 * In order to be more cautious use controller's max_hw_sectors value
1648 * to configure the maximum sectors for the write-zeroes which is
1649 * configured based on the controller's MDTS field in the
1650 * nvme_init_identify() if available.
1652 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1653 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1655 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1657 blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1660 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1661 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1665 memset(ids, 0, sizeof(*ids));
1667 if (ctrl->vs >= NVME_VS(1, 1, 0))
1668 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1669 if (ctrl->vs >= NVME_VS(1, 2, 0))
1670 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1671 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1672 /* Don't treat error as fatal we potentially
1673 * already have a NGUID or EUI-64
1675 ret = nvme_identify_ns_descs(ctrl, nsid, ids);
1677 dev_warn(ctrl->device,
1678 "Identify Descriptors failed (%d)\n", ret);
1683 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1685 return !uuid_is_null(&ids->uuid) ||
1686 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1687 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1690 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1692 return uuid_equal(&a->uuid, &b->uuid) &&
1693 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1694 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1697 static void nvme_update_disk_info(struct gendisk *disk,
1698 struct nvme_ns *ns, struct nvme_id_ns *id)
1700 sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1701 unsigned short bs = 1 << ns->lba_shift;
1702 u32 atomic_bs, phys_bs, io_opt;
1704 if (ns->lba_shift > PAGE_SHIFT) {
1705 /* unsupported block size, set capacity to 0 later */
1708 blk_mq_freeze_queue(disk->queue);
1709 blk_integrity_unregister(disk);
1711 if (id->nabo == 0) {
1713 * Bit 1 indicates whether NAWUPF is defined for this namespace
1714 * and whether it should be used instead of AWUPF. If NAWUPF ==
1715 * 0 then AWUPF must be used instead.
1717 if (id->nsfeat & (1 << 1) && id->nawupf)
1718 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1720 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1726 if (id->nsfeat & (1 << 4)) {
1727 /* NPWG = Namespace Preferred Write Granularity */
1728 phys_bs *= 1 + le16_to_cpu(id->npwg);
1729 /* NOWS = Namespace Optimal Write Size */
1730 io_opt *= 1 + le16_to_cpu(id->nows);
1733 blk_queue_logical_block_size(disk->queue, bs);
1735 * Linux filesystems assume writing a single physical block is
1736 * an atomic operation. Hence limit the physical block size to the
1737 * value of the Atomic Write Unit Power Fail parameter.
1739 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1740 blk_queue_io_min(disk->queue, phys_bs);
1741 blk_queue_io_opt(disk->queue, io_opt);
1743 if (ns->ms && !ns->ext &&
1744 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1745 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1746 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1747 ns->lba_shift > PAGE_SHIFT)
1750 set_capacity(disk, capacity);
1752 nvme_config_discard(disk, ns);
1753 nvme_config_write_zeroes(disk, ns);
1755 if (id->nsattr & (1 << 0))
1756 set_disk_ro(disk, true);
1758 set_disk_ro(disk, false);
1760 blk_mq_unfreeze_queue(disk->queue);
1763 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1765 struct nvme_ns *ns = disk->private_data;
1768 * If identify namespace failed, use default 512 byte block size so
1769 * block layer can use before failing read/write for 0 capacity.
1771 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1772 if (ns->lba_shift == 0)
1774 ns->noiob = le16_to_cpu(id->noiob);
1775 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1776 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1777 /* the PI implementation requires metadata equal t10 pi tuple size */
1778 if (ns->ms == sizeof(struct t10_pi_tuple))
1779 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1784 nvme_set_chunk_size(ns);
1785 nvme_update_disk_info(disk, ns, id);
1786 #ifdef CONFIG_NVME_MULTIPATH
1787 if (ns->head->disk) {
1788 nvme_update_disk_info(ns->head->disk, ns, id);
1789 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1790 revalidate_disk(ns->head->disk);
1795 static int nvme_revalidate_disk(struct gendisk *disk)
1797 struct nvme_ns *ns = disk->private_data;
1798 struct nvme_ctrl *ctrl = ns->ctrl;
1799 struct nvme_id_ns *id;
1800 struct nvme_ns_ids ids;
1803 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1804 set_capacity(disk, 0);
1808 ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1812 if (id->ncap == 0) {
1817 __nvme_revalidate_disk(disk, id);
1818 ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1822 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1823 dev_err(ctrl->device,
1824 "identifiers changed for nsid %d\n", ns->head->ns_id);
1832 * Only fail the function if we got a fatal error back from the
1833 * device, otherwise ignore the error and just move on.
1835 if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1838 ret = blk_status_to_errno(nvme_error_status(ret));
1842 static char nvme_pr_type(enum pr_type type)
1845 case PR_WRITE_EXCLUSIVE:
1847 case PR_EXCLUSIVE_ACCESS:
1849 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1851 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1853 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1855 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1862 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1863 u64 key, u64 sa_key, u8 op)
1865 struct nvme_ns_head *head = NULL;
1867 struct nvme_command c;
1869 u8 data[16] = { 0, };
1871 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1873 return -EWOULDBLOCK;
1875 put_unaligned_le64(key, &data[0]);
1876 put_unaligned_le64(sa_key, &data[8]);
1878 memset(&c, 0, sizeof(c));
1879 c.common.opcode = op;
1880 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1881 c.common.cdw10 = cpu_to_le32(cdw10);
1883 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1884 nvme_put_ns_from_disk(head, srcu_idx);
1888 static int nvme_pr_register(struct block_device *bdev, u64 old,
1889 u64 new, unsigned flags)
1893 if (flags & ~PR_FL_IGNORE_KEY)
1896 cdw10 = old ? 2 : 0;
1897 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1898 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1899 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1902 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1903 enum pr_type type, unsigned flags)
1907 if (flags & ~PR_FL_IGNORE_KEY)
1910 cdw10 = nvme_pr_type(type) << 8;
1911 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1912 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1915 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1916 enum pr_type type, bool abort)
1918 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1919 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1922 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1924 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1925 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1928 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1930 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1931 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1934 static const struct pr_ops nvme_pr_ops = {
1935 .pr_register = nvme_pr_register,
1936 .pr_reserve = nvme_pr_reserve,
1937 .pr_release = nvme_pr_release,
1938 .pr_preempt = nvme_pr_preempt,
1939 .pr_clear = nvme_pr_clear,
1942 #ifdef CONFIG_BLK_SED_OPAL
1943 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1946 struct nvme_ctrl *ctrl = data;
1947 struct nvme_command cmd;
1949 memset(&cmd, 0, sizeof(cmd));
1951 cmd.common.opcode = nvme_admin_security_send;
1953 cmd.common.opcode = nvme_admin_security_recv;
1954 cmd.common.nsid = 0;
1955 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1956 cmd.common.cdw11 = cpu_to_le32(len);
1958 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1959 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1961 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1962 #endif /* CONFIG_BLK_SED_OPAL */
1964 static const struct block_device_operations nvme_fops = {
1965 .owner = THIS_MODULE,
1966 .ioctl = nvme_ioctl,
1967 .compat_ioctl = nvme_ioctl,
1969 .release = nvme_release,
1970 .getgeo = nvme_getgeo,
1971 .revalidate_disk= nvme_revalidate_disk,
1972 .pr_ops = &nvme_pr_ops,
1975 #ifdef CONFIG_NVME_MULTIPATH
1976 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1978 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1980 if (!kref_get_unless_zero(&head->ref))
1985 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1987 nvme_put_ns_head(disk->private_data);
1990 const struct block_device_operations nvme_ns_head_ops = {
1991 .owner = THIS_MODULE,
1992 .open = nvme_ns_head_open,
1993 .release = nvme_ns_head_release,
1994 .ioctl = nvme_ioctl,
1995 .compat_ioctl = nvme_ioctl,
1996 .getgeo = nvme_getgeo,
1997 .pr_ops = &nvme_pr_ops,
1999 #endif /* CONFIG_NVME_MULTIPATH */
2001 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2003 unsigned long timeout =
2004 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2005 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2008 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2011 if ((csts & NVME_CSTS_RDY) == bit)
2015 if (fatal_signal_pending(current))
2017 if (time_after(jiffies, timeout)) {
2018 dev_err(ctrl->device,
2019 "Device not ready; aborting %s\n", enabled ?
2020 "initialisation" : "reset");
2029 * If the device has been passed off to us in an enabled state, just clear
2030 * the enabled bit. The spec says we should set the 'shutdown notification
2031 * bits', but doing so may cause the device to complete commands to the
2032 * admin queue ... and we don't know what memory that might be pointing at!
2034 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2038 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2039 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2041 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2045 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2046 msleep(NVME_QUIRK_DELAY_AMOUNT);
2048 return nvme_wait_ready(ctrl, ctrl->cap, false);
2050 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2052 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2055 * Default to a 4K page size, with the intention to update this
2056 * path in the future to accomodate architectures with differing
2057 * kernel and IO page sizes.
2059 unsigned dev_page_min, page_shift = 12;
2062 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2064 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2067 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2069 if (page_shift < dev_page_min) {
2070 dev_err(ctrl->device,
2071 "Minimum device page size %u too large for host (%u)\n",
2072 1 << dev_page_min, 1 << page_shift);
2076 ctrl->page_size = 1 << page_shift;
2078 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2079 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2080 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2081 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2082 ctrl->ctrl_config |= NVME_CC_ENABLE;
2084 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2087 return nvme_wait_ready(ctrl, ctrl->cap, true);
2089 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2091 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2093 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2097 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2098 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2100 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2104 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2105 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2109 if (fatal_signal_pending(current))
2111 if (time_after(jiffies, timeout)) {
2112 dev_err(ctrl->device,
2113 "Device shutdown incomplete; abort shutdown\n");
2120 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2122 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2123 struct request_queue *q)
2127 if (ctrl->max_hw_sectors) {
2129 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2131 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2132 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2133 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2135 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2136 is_power_of_2(ctrl->max_hw_sectors))
2137 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2138 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2139 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2141 blk_queue_write_cache(q, vwc, vwc);
2144 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2149 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2152 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2153 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2156 dev_warn_once(ctrl->device,
2157 "could not set timestamp (%d)\n", ret);
2161 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2163 struct nvme_feat_host_behavior *host;
2166 /* Don't bother enabling the feature if retry delay is not reported */
2170 host = kzalloc(sizeof(*host), GFP_KERNEL);
2174 host->acre = NVME_ENABLE_ACRE;
2175 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2176 host, sizeof(*host), NULL);
2181 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2184 * APST (Autonomous Power State Transition) lets us program a
2185 * table of power state transitions that the controller will
2186 * perform automatically. We configure it with a simple
2187 * heuristic: we are willing to spend at most 2% of the time
2188 * transitioning between power states. Therefore, when running
2189 * in any given state, we will enter the next lower-power
2190 * non-operational state after waiting 50 * (enlat + exlat)
2191 * microseconds, as long as that state's exit latency is under
2192 * the requested maximum latency.
2194 * We will not autonomously enter any non-operational state for
2195 * which the total latency exceeds ps_max_latency_us. Users
2196 * can set ps_max_latency_us to zero to turn off APST.
2200 struct nvme_feat_auto_pst *table;
2206 * If APST isn't supported or if we haven't been initialized yet,
2207 * then don't do anything.
2212 if (ctrl->npss > 31) {
2213 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2217 table = kzalloc(sizeof(*table), GFP_KERNEL);
2221 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2222 /* Turn off APST. */
2224 dev_dbg(ctrl->device, "APST disabled\n");
2226 __le64 target = cpu_to_le64(0);
2230 * Walk through all states from lowest- to highest-power.
2231 * According to the spec, lower-numbered states use more
2232 * power. NPSS, despite the name, is the index of the
2233 * lowest-power state, not the number of states.
2235 for (state = (int)ctrl->npss; state >= 0; state--) {
2236 u64 total_latency_us, exit_latency_us, transition_ms;
2239 table->entries[state] = target;
2242 * Don't allow transitions to the deepest state
2243 * if it's quirked off.
2245 if (state == ctrl->npss &&
2246 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2250 * Is this state a useful non-operational state for
2251 * higher-power states to autonomously transition to?
2253 if (!(ctrl->psd[state].flags &
2254 NVME_PS_FLAGS_NON_OP_STATE))
2258 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2259 if (exit_latency_us > ctrl->ps_max_latency_us)
2264 le32_to_cpu(ctrl->psd[state].entry_lat);
2267 * This state is good. Use it as the APST idle
2268 * target for higher power states.
2270 transition_ms = total_latency_us + 19;
2271 do_div(transition_ms, 20);
2272 if (transition_ms > (1 << 24) - 1)
2273 transition_ms = (1 << 24) - 1;
2275 target = cpu_to_le64((state << 3) |
2276 (transition_ms << 8));
2281 if (total_latency_us > max_lat_us)
2282 max_lat_us = total_latency_us;
2288 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2290 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2291 max_ps, max_lat_us, (int)sizeof(*table), table);
2295 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2296 table, sizeof(*table), NULL);
2298 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2304 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2306 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2310 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2311 case PM_QOS_LATENCY_ANY:
2319 if (ctrl->ps_max_latency_us != latency) {
2320 ctrl->ps_max_latency_us = latency;
2321 nvme_configure_apst(ctrl);
2325 struct nvme_core_quirk_entry {
2327 * NVMe model and firmware strings are padded with spaces. For
2328 * simplicity, strings in the quirk table are padded with NULLs
2334 unsigned long quirks;
2337 static const struct nvme_core_quirk_entry core_quirks[] = {
2340 * This Toshiba device seems to die using any APST states. See:
2341 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2344 .mn = "THNSF5256GPUK TOSHIBA",
2345 .quirks = NVME_QUIRK_NO_APST,
2349 * This LiteON CL1-3D*-Q11 firmware version has a race
2350 * condition associated with actions related to suspend to idle
2351 * LiteON has resolved the problem in future firmware
2355 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2359 * This Kingston E8FK11.T firmware version has no interrupt
2360 * after resume with actions related to suspend to idle
2361 * https://bugzilla.kernel.org/show_bug.cgi?id=204887
2365 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2369 /* match is null-terminated but idstr is space-padded. */
2370 static bool string_matches(const char *idstr, const char *match, size_t len)
2377 matchlen = strlen(match);
2378 WARN_ON_ONCE(matchlen > len);
2380 if (memcmp(idstr, match, matchlen))
2383 for (; matchlen < len; matchlen++)
2384 if (idstr[matchlen] != ' ')
2390 static bool quirk_matches(const struct nvme_id_ctrl *id,
2391 const struct nvme_core_quirk_entry *q)
2393 return q->vid == le16_to_cpu(id->vid) &&
2394 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2395 string_matches(id->fr, q->fr, sizeof(id->fr));
2398 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2399 struct nvme_id_ctrl *id)
2404 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2405 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2406 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2407 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2411 if (ctrl->vs >= NVME_VS(1, 2, 1))
2412 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2415 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2416 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2417 "nqn.2014.08.org.nvmexpress:%04x%04x",
2418 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2419 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2420 off += sizeof(id->sn);
2421 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2422 off += sizeof(id->mn);
2423 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2426 static void nvme_release_subsystem(struct device *dev)
2428 struct nvme_subsystem *subsys =
2429 container_of(dev, struct nvme_subsystem, dev);
2431 if (subsys->instance >= 0)
2432 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2436 static void nvme_destroy_subsystem(struct kref *ref)
2438 struct nvme_subsystem *subsys =
2439 container_of(ref, struct nvme_subsystem, ref);
2441 mutex_lock(&nvme_subsystems_lock);
2442 list_del(&subsys->entry);
2443 mutex_unlock(&nvme_subsystems_lock);
2445 ida_destroy(&subsys->ns_ida);
2446 device_del(&subsys->dev);
2447 put_device(&subsys->dev);
2450 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2452 kref_put(&subsys->ref, nvme_destroy_subsystem);
2455 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2457 struct nvme_subsystem *subsys;
2459 lockdep_assert_held(&nvme_subsystems_lock);
2462 * Fail matches for discovery subsystems. This results
2463 * in each discovery controller bound to a unique subsystem.
2464 * This avoids issues with validating controller values
2465 * that can only be true when there is a single unique subsystem.
2466 * There may be multiple and completely independent entities
2467 * that provide discovery controllers.
2469 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2472 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2473 if (strcmp(subsys->subnqn, subsysnqn))
2475 if (!kref_get_unless_zero(&subsys->ref))
2483 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2484 struct device_attribute subsys_attr_##_name = \
2485 __ATTR(_name, _mode, _show, NULL)
2487 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2488 struct device_attribute *attr,
2491 struct nvme_subsystem *subsys =
2492 container_of(dev, struct nvme_subsystem, dev);
2494 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2496 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2498 #define nvme_subsys_show_str_function(field) \
2499 static ssize_t subsys_##field##_show(struct device *dev, \
2500 struct device_attribute *attr, char *buf) \
2502 struct nvme_subsystem *subsys = \
2503 container_of(dev, struct nvme_subsystem, dev); \
2504 return sprintf(buf, "%.*s\n", \
2505 (int)sizeof(subsys->field), subsys->field); \
2507 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2509 nvme_subsys_show_str_function(model);
2510 nvme_subsys_show_str_function(serial);
2511 nvme_subsys_show_str_function(firmware_rev);
2513 static struct attribute *nvme_subsys_attrs[] = {
2514 &subsys_attr_model.attr,
2515 &subsys_attr_serial.attr,
2516 &subsys_attr_firmware_rev.attr,
2517 &subsys_attr_subsysnqn.attr,
2518 #ifdef CONFIG_NVME_MULTIPATH
2519 &subsys_attr_iopolicy.attr,
2524 static struct attribute_group nvme_subsys_attrs_group = {
2525 .attrs = nvme_subsys_attrs,
2528 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2529 &nvme_subsys_attrs_group,
2533 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2534 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2536 struct nvme_ctrl *tmp;
2538 lockdep_assert_held(&nvme_subsystems_lock);
2540 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2541 if (tmp->state == NVME_CTRL_DELETING ||
2542 tmp->state == NVME_CTRL_DEAD)
2545 if (tmp->cntlid == ctrl->cntlid) {
2546 dev_err(ctrl->device,
2547 "Duplicate cntlid %u with %s, rejecting\n",
2548 ctrl->cntlid, dev_name(tmp->device));
2552 if ((id->cmic & (1 << 1)) ||
2553 (ctrl->opts && ctrl->opts->discovery_nqn))
2556 dev_err(ctrl->device,
2557 "Subsystem does not support multiple controllers\n");
2564 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2566 struct nvme_subsystem *subsys, *found;
2569 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2573 subsys->instance = -1;
2574 mutex_init(&subsys->lock);
2575 kref_init(&subsys->ref);
2576 INIT_LIST_HEAD(&subsys->ctrls);
2577 INIT_LIST_HEAD(&subsys->nsheads);
2578 nvme_init_subnqn(subsys, ctrl, id);
2579 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2580 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2581 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2582 subsys->vendor_id = le16_to_cpu(id->vid);
2583 subsys->cmic = id->cmic;
2584 subsys->awupf = le16_to_cpu(id->awupf);
2585 #ifdef CONFIG_NVME_MULTIPATH
2586 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2589 subsys->dev.class = nvme_subsys_class;
2590 subsys->dev.release = nvme_release_subsystem;
2591 subsys->dev.groups = nvme_subsys_attrs_groups;
2592 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2593 device_initialize(&subsys->dev);
2595 mutex_lock(&nvme_subsystems_lock);
2596 found = __nvme_find_get_subsystem(subsys->subnqn);
2598 put_device(&subsys->dev);
2601 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2603 goto out_put_subsystem;
2606 ret = device_add(&subsys->dev);
2608 dev_err(ctrl->device,
2609 "failed to register subsystem device.\n");
2610 put_device(&subsys->dev);
2613 ida_init(&subsys->ns_ida);
2614 list_add_tail(&subsys->entry, &nvme_subsystems);
2617 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2618 dev_name(ctrl->device));
2620 dev_err(ctrl->device,
2621 "failed to create sysfs link from subsystem.\n");
2622 goto out_put_subsystem;
2626 subsys->instance = ctrl->instance;
2627 ctrl->subsys = subsys;
2628 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2629 mutex_unlock(&nvme_subsystems_lock);
2633 nvme_put_subsystem(subsys);
2635 mutex_unlock(&nvme_subsystems_lock);
2639 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2640 void *log, size_t size, u64 offset)
2642 struct nvme_command c = { };
2643 unsigned long dwlen = size / 4 - 1;
2645 c.get_log_page.opcode = nvme_admin_get_log_page;
2646 c.get_log_page.nsid = cpu_to_le32(nsid);
2647 c.get_log_page.lid = log_page;
2648 c.get_log_page.lsp = lsp;
2649 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2650 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2651 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2652 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2654 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2657 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2662 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2667 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2668 ctrl->effects, sizeof(*ctrl->effects), 0);
2670 kfree(ctrl->effects);
2671 ctrl->effects = NULL;
2677 * Initialize the cached copies of the Identify data and various controller
2678 * register in our nvme_ctrl structure. This should be called as soon as
2679 * the admin queue is fully up and running.
2681 int nvme_init_identify(struct nvme_ctrl *ctrl)
2683 struct nvme_id_ctrl *id;
2684 int ret, page_shift;
2686 bool prev_apst_enabled;
2688 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2690 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2693 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2694 ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2696 if (ctrl->vs >= NVME_VS(1, 1, 0))
2697 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2699 ret = nvme_identify_ctrl(ctrl, &id);
2701 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2705 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2706 ret = nvme_get_effects_log(ctrl);
2711 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2712 ctrl->cntlid = le16_to_cpu(id->cntlid);
2714 if (!ctrl->identified) {
2717 ret = nvme_init_subsystem(ctrl, id);
2722 * Check for quirks. Quirk can depend on firmware version,
2723 * so, in principle, the set of quirks present can change
2724 * across a reset. As a possible future enhancement, we
2725 * could re-scan for quirks every time we reinitialize
2726 * the device, but we'd have to make sure that the driver
2727 * behaves intelligently if the quirks change.
2729 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2730 if (quirk_matches(id, &core_quirks[i]))
2731 ctrl->quirks |= core_quirks[i].quirks;
2735 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2736 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2737 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2740 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2741 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2742 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2744 ctrl->oacs = le16_to_cpu(id->oacs);
2745 ctrl->oncs = le16_to_cpu(id->oncs);
2746 ctrl->mtfa = le16_to_cpu(id->mtfa);
2747 ctrl->oaes = le32_to_cpu(id->oaes);
2748 atomic_set(&ctrl->abort_limit, id->acl + 1);
2749 ctrl->vwc = id->vwc;
2751 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2753 max_hw_sectors = UINT_MAX;
2754 ctrl->max_hw_sectors =
2755 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2757 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2758 ctrl->sgls = le32_to_cpu(id->sgls);
2759 ctrl->kas = le16_to_cpu(id->kas);
2760 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2761 ctrl->ctratt = le32_to_cpu(id->ctratt);
2765 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2767 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2768 shutdown_timeout, 60);
2770 if (ctrl->shutdown_timeout != shutdown_timeout)
2771 dev_info(ctrl->device,
2772 "Shutdown timeout set to %u seconds\n",
2773 ctrl->shutdown_timeout);
2775 ctrl->shutdown_timeout = shutdown_timeout;
2777 ctrl->npss = id->npss;
2778 ctrl->apsta = id->apsta;
2779 prev_apst_enabled = ctrl->apst_enabled;
2780 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2781 if (force_apst && id->apsta) {
2782 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2783 ctrl->apst_enabled = true;
2785 ctrl->apst_enabled = false;
2788 ctrl->apst_enabled = id->apsta;
2790 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2792 if (ctrl->ops->flags & NVME_F_FABRICS) {
2793 ctrl->icdoff = le16_to_cpu(id->icdoff);
2794 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2795 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2796 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2799 * In fabrics we need to verify the cntlid matches the
2802 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2807 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2808 dev_err(ctrl->device,
2809 "keep-alive support is mandatory for fabrics\n");
2814 ctrl->hmpre = le32_to_cpu(id->hmpre);
2815 ctrl->hmmin = le32_to_cpu(id->hmmin);
2816 ctrl->hmminds = le32_to_cpu(id->hmminds);
2817 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2820 ret = nvme_mpath_init(ctrl, id);
2826 if (ctrl->apst_enabled && !prev_apst_enabled)
2827 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2828 else if (!ctrl->apst_enabled && prev_apst_enabled)
2829 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2831 ret = nvme_configure_apst(ctrl);
2835 ret = nvme_configure_timestamp(ctrl);
2839 ret = nvme_configure_directives(ctrl);
2843 ret = nvme_configure_acre(ctrl);
2847 ctrl->identified = true;
2855 EXPORT_SYMBOL_GPL(nvme_init_identify);
2857 static int nvme_dev_open(struct inode *inode, struct file *file)
2859 struct nvme_ctrl *ctrl =
2860 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2862 switch (ctrl->state) {
2863 case NVME_CTRL_LIVE:
2866 return -EWOULDBLOCK;
2869 file->private_data = ctrl;
2873 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2878 down_read(&ctrl->namespaces_rwsem);
2879 if (list_empty(&ctrl->namespaces)) {
2884 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2885 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2886 dev_warn(ctrl->device,
2887 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2892 dev_warn(ctrl->device,
2893 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2894 kref_get(&ns->kref);
2895 up_read(&ctrl->namespaces_rwsem);
2897 ret = nvme_user_cmd(ctrl, ns, argp);
2902 up_read(&ctrl->namespaces_rwsem);
2906 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2909 struct nvme_ctrl *ctrl = file->private_data;
2910 void __user *argp = (void __user *)arg;
2913 case NVME_IOCTL_ADMIN_CMD:
2914 return nvme_user_cmd(ctrl, NULL, argp);
2915 case NVME_IOCTL_ADMIN64_CMD:
2916 return nvme_user_cmd64(ctrl, NULL, argp);
2917 case NVME_IOCTL_IO_CMD:
2918 return nvme_dev_user_cmd(ctrl, argp);
2919 case NVME_IOCTL_RESET:
2920 dev_warn(ctrl->device, "resetting controller\n");
2921 return nvme_reset_ctrl_sync(ctrl);
2922 case NVME_IOCTL_SUBSYS_RESET:
2923 return nvme_reset_subsystem(ctrl);
2924 case NVME_IOCTL_RESCAN:
2925 nvme_queue_scan(ctrl);
2932 static const struct file_operations nvme_dev_fops = {
2933 .owner = THIS_MODULE,
2934 .open = nvme_dev_open,
2935 .unlocked_ioctl = nvme_dev_ioctl,
2936 .compat_ioctl = nvme_dev_ioctl,
2939 static ssize_t nvme_sysfs_reset(struct device *dev,
2940 struct device_attribute *attr, const char *buf,
2943 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2946 ret = nvme_reset_ctrl_sync(ctrl);
2951 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2953 static ssize_t nvme_sysfs_rescan(struct device *dev,
2954 struct device_attribute *attr, const char *buf,
2957 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2959 nvme_queue_scan(ctrl);
2962 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2964 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2966 struct gendisk *disk = dev_to_disk(dev);
2968 if (disk->fops == &nvme_fops)
2969 return nvme_get_ns_from_dev(dev)->head;
2971 return disk->private_data;
2974 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2977 struct nvme_ns_head *head = dev_to_ns_head(dev);
2978 struct nvme_ns_ids *ids = &head->ids;
2979 struct nvme_subsystem *subsys = head->subsys;
2980 int serial_len = sizeof(subsys->serial);
2981 int model_len = sizeof(subsys->model);
2983 if (!uuid_is_null(&ids->uuid))
2984 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2986 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2987 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2989 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2990 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2992 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2993 subsys->serial[serial_len - 1] == '\0'))
2995 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2996 subsys->model[model_len - 1] == '\0'))
2999 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3000 serial_len, subsys->serial, model_len, subsys->model,
3003 static DEVICE_ATTR_RO(wwid);
3005 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3008 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3010 static DEVICE_ATTR_RO(nguid);
3012 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3015 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3017 /* For backward compatibility expose the NGUID to userspace if
3018 * we have no UUID set
3020 if (uuid_is_null(&ids->uuid)) {
3021 printk_ratelimited(KERN_WARNING
3022 "No UUID available providing old NGUID\n");
3023 return sprintf(buf, "%pU\n", ids->nguid);
3025 return sprintf(buf, "%pU\n", &ids->uuid);
3027 static DEVICE_ATTR_RO(uuid);
3029 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3032 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3034 static DEVICE_ATTR_RO(eui);
3036 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3039 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3041 static DEVICE_ATTR_RO(nsid);
3043 static struct attribute *nvme_ns_id_attrs[] = {
3044 &dev_attr_wwid.attr,
3045 &dev_attr_uuid.attr,
3046 &dev_attr_nguid.attr,
3048 &dev_attr_nsid.attr,
3049 #ifdef CONFIG_NVME_MULTIPATH
3050 &dev_attr_ana_grpid.attr,
3051 &dev_attr_ana_state.attr,
3056 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3057 struct attribute *a, int n)
3059 struct device *dev = container_of(kobj, struct device, kobj);
3060 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3062 if (a == &dev_attr_uuid.attr) {
3063 if (uuid_is_null(&ids->uuid) &&
3064 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3067 if (a == &dev_attr_nguid.attr) {
3068 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3071 if (a == &dev_attr_eui.attr) {
3072 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3075 #ifdef CONFIG_NVME_MULTIPATH
3076 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3077 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3079 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3086 static const struct attribute_group nvme_ns_id_attr_group = {
3087 .attrs = nvme_ns_id_attrs,
3088 .is_visible = nvme_ns_id_attrs_are_visible,
3091 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3092 &nvme_ns_id_attr_group,
3094 &nvme_nvm_attr_group,
3099 #define nvme_show_str_function(field) \
3100 static ssize_t field##_show(struct device *dev, \
3101 struct device_attribute *attr, char *buf) \
3103 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3104 return sprintf(buf, "%.*s\n", \
3105 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3107 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3109 nvme_show_str_function(model);
3110 nvme_show_str_function(serial);
3111 nvme_show_str_function(firmware_rev);
3113 #define nvme_show_int_function(field) \
3114 static ssize_t field##_show(struct device *dev, \
3115 struct device_attribute *attr, char *buf) \
3117 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3118 return sprintf(buf, "%d\n", ctrl->field); \
3120 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3122 nvme_show_int_function(cntlid);
3123 nvme_show_int_function(numa_node);
3124 nvme_show_int_function(queue_count);
3125 nvme_show_int_function(sqsize);
3127 static ssize_t nvme_sysfs_delete(struct device *dev,
3128 struct device_attribute *attr, const char *buf,
3131 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3133 if (device_remove_file_self(dev, attr))
3134 nvme_delete_ctrl_sync(ctrl);
3137 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3139 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3140 struct device_attribute *attr,
3143 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3145 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3147 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3149 static ssize_t nvme_sysfs_show_state(struct device *dev,
3150 struct device_attribute *attr,
3153 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3154 static const char *const state_name[] = {
3155 [NVME_CTRL_NEW] = "new",
3156 [NVME_CTRL_LIVE] = "live",
3157 [NVME_CTRL_RESETTING] = "resetting",
3158 [NVME_CTRL_CONNECTING] = "connecting",
3159 [NVME_CTRL_DELETING] = "deleting",
3160 [NVME_CTRL_DEAD] = "dead",
3163 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3164 state_name[ctrl->state])
3165 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3167 return sprintf(buf, "unknown state\n");
3170 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3172 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3173 struct device_attribute *attr,
3176 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3178 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3180 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3182 static ssize_t nvme_sysfs_show_address(struct device *dev,
3183 struct device_attribute *attr,
3186 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3188 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3190 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3192 static struct attribute *nvme_dev_attrs[] = {
3193 &dev_attr_reset_controller.attr,
3194 &dev_attr_rescan_controller.attr,
3195 &dev_attr_model.attr,
3196 &dev_attr_serial.attr,
3197 &dev_attr_firmware_rev.attr,
3198 &dev_attr_cntlid.attr,
3199 &dev_attr_delete_controller.attr,
3200 &dev_attr_transport.attr,
3201 &dev_attr_subsysnqn.attr,
3202 &dev_attr_address.attr,
3203 &dev_attr_state.attr,
3204 &dev_attr_numa_node.attr,
3205 &dev_attr_queue_count.attr,
3206 &dev_attr_sqsize.attr,
3210 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3211 struct attribute *a, int n)
3213 struct device *dev = container_of(kobj, struct device, kobj);
3214 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3216 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3218 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3224 static struct attribute_group nvme_dev_attrs_group = {
3225 .attrs = nvme_dev_attrs,
3226 .is_visible = nvme_dev_attrs_are_visible,
3229 static const struct attribute_group *nvme_dev_attr_groups[] = {
3230 &nvme_dev_attrs_group,
3234 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3237 struct nvme_ns_head *h;
3239 lockdep_assert_held(&subsys->lock);
3241 list_for_each_entry(h, &subsys->nsheads, entry) {
3242 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3249 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3250 struct nvme_ns_head *new)
3252 struct nvme_ns_head *h;
3254 lockdep_assert_held(&subsys->lock);
3256 list_for_each_entry(h, &subsys->nsheads, entry) {
3257 if (nvme_ns_ids_valid(&new->ids) &&
3258 !list_empty(&h->list) &&
3259 nvme_ns_ids_equal(&new->ids, &h->ids))
3266 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3267 unsigned nsid, struct nvme_id_ns *id)
3269 struct nvme_ns_head *head;
3270 size_t size = sizeof(*head);
3273 #ifdef CONFIG_NVME_MULTIPATH
3274 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3277 head = kzalloc(size, GFP_KERNEL);
3280 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3283 head->instance = ret;
3284 INIT_LIST_HEAD(&head->list);
3285 ret = init_srcu_struct(&head->srcu);
3287 goto out_ida_remove;
3288 head->subsys = ctrl->subsys;
3290 kref_init(&head->ref);
3292 ret = nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3294 goto out_cleanup_srcu;
3296 ret = __nvme_check_ids(ctrl->subsys, head);
3298 dev_err(ctrl->device,
3299 "duplicate IDs for nsid %d\n", nsid);
3300 goto out_cleanup_srcu;
3303 ret = nvme_mpath_alloc_disk(ctrl, head);
3305 goto out_cleanup_srcu;
3307 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3309 kref_get(&ctrl->subsys->ref);
3313 cleanup_srcu_struct(&head->srcu);
3315 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3320 ret = blk_status_to_errno(nvme_error_status(ret));
3321 return ERR_PTR(ret);
3324 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3325 struct nvme_id_ns *id)
3327 struct nvme_ctrl *ctrl = ns->ctrl;
3328 bool is_shared = id->nmic & (1 << 0);
3329 struct nvme_ns_head *head = NULL;
3332 mutex_lock(&ctrl->subsys->lock);
3334 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3336 head = nvme_alloc_ns_head(ctrl, nsid, id);
3338 ret = PTR_ERR(head);
3342 struct nvme_ns_ids ids;
3344 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3348 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3349 dev_err(ctrl->device,
3350 "IDs don't match for shared namespace %d\n",
3357 list_add_tail(&ns->siblings, &head->list);
3361 mutex_unlock(&ctrl->subsys->lock);
3363 ret = blk_status_to_errno(nvme_error_status(ret));
3367 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3369 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3370 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3372 return nsa->head->ns_id - nsb->head->ns_id;
3375 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3377 struct nvme_ns *ns, *ret = NULL;
3379 down_read(&ctrl->namespaces_rwsem);
3380 list_for_each_entry(ns, &ctrl->namespaces, list) {
3381 if (ns->head->ns_id == nsid) {
3382 if (!kref_get_unless_zero(&ns->kref))
3387 if (ns->head->ns_id > nsid)
3390 up_read(&ctrl->namespaces_rwsem);
3394 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3396 struct streams_directive_params s;
3399 if (!ctrl->nr_streams)
3402 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3406 ns->sws = le32_to_cpu(s.sws);
3407 ns->sgs = le16_to_cpu(s.sgs);
3410 unsigned int bs = 1 << ns->lba_shift;
3412 blk_queue_io_min(ns->queue, bs * ns->sws);
3414 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3420 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3423 struct gendisk *disk;
3424 struct nvme_id_ns *id;
3425 char disk_name[DISK_NAME_LEN];
3426 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3428 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3432 ns->queue = blk_mq_init_queue(ctrl->tagset);
3433 if (IS_ERR(ns->queue)) {
3434 ret = PTR_ERR(ns->queue);
3438 if (ctrl->opts && ctrl->opts->data_digest)
3439 ns->queue->backing_dev_info->capabilities
3440 |= BDI_CAP_STABLE_WRITES;
3442 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3443 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3444 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3446 ns->queue->queuedata = ns;
3449 kref_init(&ns->kref);
3450 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3452 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3453 nvme_set_queue_limits(ctrl, ns->queue);
3455 ret = nvme_identify_ns(ctrl, nsid, &id);
3457 goto out_free_queue;
3459 if (id->ncap == 0) {
3464 ret = nvme_init_ns_head(ns, nsid, id);
3467 nvme_setup_streams_ns(ctrl, ns);
3468 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3470 disk = alloc_disk_node(0, node);
3476 disk->fops = &nvme_fops;
3477 disk->private_data = ns;
3478 disk->queue = ns->queue;
3479 disk->flags = flags;
3480 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3483 __nvme_revalidate_disk(disk, id);
3485 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3486 ret = nvme_nvm_register(ns, disk_name, node);
3488 dev_warn(ctrl->device, "LightNVM init failure\n");
3493 down_write(&ctrl->namespaces_rwsem);
3494 list_add_tail(&ns->list, &ctrl->namespaces);
3495 up_write(&ctrl->namespaces_rwsem);
3497 nvme_get_ctrl(ctrl);
3499 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3501 nvme_mpath_add_disk(ns, id);
3502 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3509 mutex_lock(&ctrl->subsys->lock);
3510 list_del_rcu(&ns->siblings);
3511 mutex_unlock(&ctrl->subsys->lock);
3512 nvme_put_ns_head(ns->head);
3516 blk_cleanup_queue(ns->queue);
3520 ret = blk_status_to_errno(nvme_error_status(ret));
3524 static void nvme_ns_remove(struct nvme_ns *ns)
3526 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3529 nvme_fault_inject_fini(&ns->fault_inject);
3531 mutex_lock(&ns->ctrl->subsys->lock);
3532 list_del_rcu(&ns->siblings);
3533 mutex_unlock(&ns->ctrl->subsys->lock);
3534 synchronize_rcu(); /* guarantee not available in head->list */
3535 nvme_mpath_clear_current_path(ns);
3536 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3538 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3539 del_gendisk(ns->disk);
3540 blk_cleanup_queue(ns->queue);
3541 if (blk_get_integrity(ns->disk))
3542 blk_integrity_unregister(ns->disk);
3545 down_write(&ns->ctrl->namespaces_rwsem);
3546 list_del_init(&ns->list);
3547 up_write(&ns->ctrl->namespaces_rwsem);
3549 nvme_mpath_check_last_path(ns);
3553 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3557 ns = nvme_find_get_ns(ctrl, nsid);
3559 if (ns->disk && revalidate_disk(ns->disk))
3563 nvme_alloc_ns(ctrl, nsid);
3566 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3569 struct nvme_ns *ns, *next;
3572 down_write(&ctrl->namespaces_rwsem);
3573 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3574 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3575 list_move_tail(&ns->list, &rm_list);
3577 up_write(&ctrl->namespaces_rwsem);
3579 list_for_each_entry_safe(ns, next, &rm_list, list)
3584 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3588 unsigned i, j, nsid, prev = 0;
3589 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3592 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3596 for (i = 0; i < num_lists; i++) {
3597 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3601 for (j = 0; j < min(nn, 1024U); j++) {
3602 nsid = le32_to_cpu(ns_list[j]);
3606 nvme_validate_ns(ctrl, nsid);
3608 while (++prev < nsid) {
3609 ns = nvme_find_get_ns(ctrl, prev);
3619 nvme_remove_invalid_namespaces(ctrl, prev);
3625 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3629 for (i = 1; i <= nn; i++)
3630 nvme_validate_ns(ctrl, i);
3632 nvme_remove_invalid_namespaces(ctrl, nn);
3635 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3637 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3641 log = kzalloc(log_size, GFP_KERNEL);
3646 * We need to read the log to clear the AEN, but we don't want to rely
3647 * on it for the changed namespace information as userspace could have
3648 * raced with us in reading the log page, which could cause us to miss
3651 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3654 dev_warn(ctrl->device,
3655 "reading changed ns log failed: %d\n", error);
3660 static void nvme_scan_work(struct work_struct *work)
3662 struct nvme_ctrl *ctrl =
3663 container_of(work, struct nvme_ctrl, scan_work);
3664 struct nvme_id_ctrl *id;
3667 /* No tagset on a live ctrl means IO queues could not created */
3668 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3671 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3672 dev_info(ctrl->device, "rescanning namespaces.\n");
3673 nvme_clear_changed_ns_log(ctrl);
3676 if (nvme_identify_ctrl(ctrl, &id))
3679 mutex_lock(&ctrl->scan_lock);
3680 nn = le32_to_cpu(id->nn);
3681 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3682 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3683 if (!nvme_scan_ns_list(ctrl, nn))
3686 nvme_scan_ns_sequential(ctrl, nn);
3688 mutex_unlock(&ctrl->scan_lock);
3690 down_write(&ctrl->namespaces_rwsem);
3691 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3692 up_write(&ctrl->namespaces_rwsem);
3696 * This function iterates the namespace list unlocked to allow recovery from
3697 * controller failure. It is up to the caller to ensure the namespace list is
3698 * not modified by scan work while this function is executing.
3700 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3702 struct nvme_ns *ns, *next;
3706 * make sure to requeue I/O to all namespaces as these
3707 * might result from the scan itself and must complete
3708 * for the scan_work to make progress
3710 nvme_mpath_clear_ctrl_paths(ctrl);
3712 /* prevent racing with ns scanning */
3713 flush_work(&ctrl->scan_work);
3716 * The dead states indicates the controller was not gracefully
3717 * disconnected. In that case, we won't be able to flush any data while
3718 * removing the namespaces' disks; fail all the queues now to avoid
3719 * potentially having to clean up the failed sync later.
3721 if (ctrl->state == NVME_CTRL_DEAD)
3722 nvme_kill_queues(ctrl);
3724 down_write(&ctrl->namespaces_rwsem);
3725 list_splice_init(&ctrl->namespaces, &ns_list);
3726 up_write(&ctrl->namespaces_rwsem);
3728 list_for_each_entry_safe(ns, next, &ns_list, list)
3731 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3733 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3735 struct nvme_ctrl *ctrl =
3736 container_of(dev, struct nvme_ctrl, ctrl_device);
3737 struct nvmf_ctrl_options *opts = ctrl->opts;
3740 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3745 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3749 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3750 opts->trsvcid ?: "none");
3754 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3755 opts->host_traddr ?: "none");
3760 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3762 char *envp[2] = { NULL, NULL };
3763 u32 aen_result = ctrl->aen_result;
3765 ctrl->aen_result = 0;
3769 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3772 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3776 static void nvme_async_event_work(struct work_struct *work)
3778 struct nvme_ctrl *ctrl =
3779 container_of(work, struct nvme_ctrl, async_event_work);
3781 nvme_aen_uevent(ctrl);
3782 ctrl->ops->submit_async_event(ctrl);
3785 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3790 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3796 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3799 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3801 struct nvme_fw_slot_info_log *log;
3803 log = kmalloc(sizeof(*log), GFP_KERNEL);
3807 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3809 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3813 static void nvme_fw_act_work(struct work_struct *work)
3815 struct nvme_ctrl *ctrl = container_of(work,
3816 struct nvme_ctrl, fw_act_work);
3817 unsigned long fw_act_timeout;
3820 fw_act_timeout = jiffies +
3821 msecs_to_jiffies(ctrl->mtfa * 100);
3823 fw_act_timeout = jiffies +
3824 msecs_to_jiffies(admin_timeout * 1000);
3826 nvme_stop_queues(ctrl);
3827 while (nvme_ctrl_pp_status(ctrl)) {
3828 if (time_after(jiffies, fw_act_timeout)) {
3829 dev_warn(ctrl->device,
3830 "Fw activation timeout, reset controller\n");
3831 nvme_reset_ctrl(ctrl);
3837 if (ctrl->state != NVME_CTRL_LIVE)
3840 nvme_start_queues(ctrl);
3841 /* read FW slot information to clear the AER */
3842 nvme_get_fw_slot_info(ctrl);
3845 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3847 u32 aer_notice_type = (result & 0xff00) >> 8;
3849 trace_nvme_async_event(ctrl, aer_notice_type);
3851 switch (aer_notice_type) {
3852 case NVME_AER_NOTICE_NS_CHANGED:
3853 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3854 nvme_queue_scan(ctrl);
3856 case NVME_AER_NOTICE_FW_ACT_STARTING:
3857 queue_work(nvme_wq, &ctrl->fw_act_work);
3859 #ifdef CONFIG_NVME_MULTIPATH
3860 case NVME_AER_NOTICE_ANA:
3861 if (!ctrl->ana_log_buf)
3863 queue_work(nvme_wq, &ctrl->ana_work);
3866 case NVME_AER_NOTICE_DISC_CHANGED:
3867 ctrl->aen_result = result;
3870 dev_warn(ctrl->device, "async event result %08x\n", result);
3874 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3875 volatile union nvme_result *res)
3877 u32 result = le32_to_cpu(res->u32);
3878 u32 aer_type = result & 0x07;
3880 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3884 case NVME_AER_NOTICE:
3885 nvme_handle_aen_notice(ctrl, result);
3887 case NVME_AER_ERROR:
3888 case NVME_AER_SMART:
3891 trace_nvme_async_event(ctrl, aer_type);
3892 ctrl->aen_result = result;
3897 queue_work(nvme_wq, &ctrl->async_event_work);
3899 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3901 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3903 nvme_mpath_stop(ctrl);
3904 nvme_stop_keep_alive(ctrl);
3905 flush_work(&ctrl->async_event_work);
3906 cancel_work_sync(&ctrl->fw_act_work);
3908 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3910 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3913 nvme_start_keep_alive(ctrl);
3915 nvme_enable_aen(ctrl);
3917 if (ctrl->queue_count > 1) {
3918 nvme_queue_scan(ctrl);
3919 nvme_start_queues(ctrl);
3922 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3924 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3926 nvme_fault_inject_fini(&ctrl->fault_inject);
3927 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3928 cdev_device_del(&ctrl->cdev, ctrl->device);
3930 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3932 static void nvme_free_ctrl(struct device *dev)
3934 struct nvme_ctrl *ctrl =
3935 container_of(dev, struct nvme_ctrl, ctrl_device);
3936 struct nvme_subsystem *subsys = ctrl->subsys;
3938 if (subsys && ctrl->instance != subsys->instance)
3939 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3941 kfree(ctrl->effects);
3942 nvme_mpath_uninit(ctrl);
3943 __free_page(ctrl->discard_page);
3946 mutex_lock(&nvme_subsystems_lock);
3947 list_del(&ctrl->subsys_entry);
3948 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3949 mutex_unlock(&nvme_subsystems_lock);
3952 ctrl->ops->free_ctrl(ctrl);
3955 nvme_put_subsystem(subsys);
3959 * Initialize a NVMe controller structures. This needs to be called during
3960 * earliest initialization so that we have the initialized structured around
3963 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3964 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3968 ctrl->state = NVME_CTRL_NEW;
3969 spin_lock_init(&ctrl->lock);
3970 mutex_init(&ctrl->scan_lock);
3971 INIT_LIST_HEAD(&ctrl->namespaces);
3972 init_rwsem(&ctrl->namespaces_rwsem);
3975 ctrl->quirks = quirks;
3976 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3977 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3978 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3979 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3981 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3982 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3983 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3985 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3987 ctrl->discard_page = alloc_page(GFP_KERNEL);
3988 if (!ctrl->discard_page) {
3993 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3996 ctrl->instance = ret;
3998 device_initialize(&ctrl->ctrl_device);
3999 ctrl->device = &ctrl->ctrl_device;
4000 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4001 ctrl->device->class = nvme_class;
4002 ctrl->device->parent = ctrl->dev;
4003 ctrl->device->groups = nvme_dev_attr_groups;
4004 ctrl->device->release = nvme_free_ctrl;
4005 dev_set_drvdata(ctrl->device, ctrl);
4006 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4008 goto out_release_instance;
4010 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4011 ctrl->cdev.owner = ops->module;
4012 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4017 * Initialize latency tolerance controls. The sysfs files won't
4018 * be visible to userspace unless the device actually supports APST.
4020 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4021 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4022 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4024 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4028 kfree_const(ctrl->device->kobj.name);
4029 out_release_instance:
4030 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4032 if (ctrl->discard_page)
4033 __free_page(ctrl->discard_page);
4036 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4039 * nvme_kill_queues(): Ends all namespace queues
4040 * @ctrl: the dead controller that needs to end
4042 * Call this function when the driver determines it is unable to get the
4043 * controller in a state capable of servicing IO.
4045 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4049 down_read(&ctrl->namespaces_rwsem);
4051 /* Forcibly unquiesce queues to avoid blocking dispatch */
4052 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4053 blk_mq_unquiesce_queue(ctrl->admin_q);
4055 list_for_each_entry(ns, &ctrl->namespaces, list)
4056 nvme_set_queue_dying(ns);
4058 up_read(&ctrl->namespaces_rwsem);
4060 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4062 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4066 down_read(&ctrl->namespaces_rwsem);
4067 list_for_each_entry(ns, &ctrl->namespaces, list)
4068 blk_mq_unfreeze_queue(ns->queue);
4069 up_read(&ctrl->namespaces_rwsem);
4071 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4073 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4077 down_read(&ctrl->namespaces_rwsem);
4078 list_for_each_entry(ns, &ctrl->namespaces, list) {
4079 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4083 up_read(&ctrl->namespaces_rwsem);
4085 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4087 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4091 down_read(&ctrl->namespaces_rwsem);
4092 list_for_each_entry(ns, &ctrl->namespaces, list)
4093 blk_mq_freeze_queue_wait(ns->queue);
4094 up_read(&ctrl->namespaces_rwsem);
4096 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4098 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4102 down_read(&ctrl->namespaces_rwsem);
4103 list_for_each_entry(ns, &ctrl->namespaces, list)
4104 blk_freeze_queue_start(ns->queue);
4105 up_read(&ctrl->namespaces_rwsem);
4107 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4109 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4113 down_read(&ctrl->namespaces_rwsem);
4114 list_for_each_entry(ns, &ctrl->namespaces, list)
4115 blk_mq_quiesce_queue(ns->queue);
4116 up_read(&ctrl->namespaces_rwsem);
4118 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4120 void nvme_start_queues(struct nvme_ctrl *ctrl)
4124 down_read(&ctrl->namespaces_rwsem);
4125 list_for_each_entry(ns, &ctrl->namespaces, list)
4126 blk_mq_unquiesce_queue(ns->queue);
4127 up_read(&ctrl->namespaces_rwsem);
4129 EXPORT_SYMBOL_GPL(nvme_start_queues);
4132 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4136 down_read(&ctrl->namespaces_rwsem);
4137 list_for_each_entry(ns, &ctrl->namespaces, list)
4138 blk_sync_queue(ns->queue);
4139 up_read(&ctrl->namespaces_rwsem);
4142 blk_sync_queue(ctrl->admin_q);
4144 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4147 * Check we didn't inadvertently grow the command structure sizes:
4149 static inline void _nvme_check_size(void)
4151 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4152 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4153 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4154 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4155 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4156 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4157 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4158 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4159 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4160 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4161 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4162 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4163 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4164 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4165 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4166 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4167 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4171 static int __init nvme_core_init(void)
4173 int result = -ENOMEM;
4177 nvme_wq = alloc_workqueue("nvme-wq",
4178 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4182 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4183 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4187 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4188 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4189 if (!nvme_delete_wq)
4190 goto destroy_reset_wq;
4192 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4194 goto destroy_delete_wq;
4196 nvme_class = class_create(THIS_MODULE, "nvme");
4197 if (IS_ERR(nvme_class)) {
4198 result = PTR_ERR(nvme_class);
4199 goto unregister_chrdev;
4201 nvme_class->dev_uevent = nvme_class_uevent;
4203 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4204 if (IS_ERR(nvme_subsys_class)) {
4205 result = PTR_ERR(nvme_subsys_class);
4211 class_destroy(nvme_class);
4213 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4215 destroy_workqueue(nvme_delete_wq);
4217 destroy_workqueue(nvme_reset_wq);
4219 destroy_workqueue(nvme_wq);
4224 static void __exit nvme_core_exit(void)
4226 class_destroy(nvme_subsys_class);
4227 class_destroy(nvme_class);
4228 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4229 destroy_workqueue(nvme_delete_wq);
4230 destroy_workqueue(nvme_reset_wq);
4231 destroy_workqueue(nvme_wq);
4234 MODULE_LICENSE("GPL");
4235 MODULE_VERSION("1.0");
4236 module_init(nvme_core_init);
4237 module_exit(nvme_core_exit);