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 as scan, aen handling, fw activation,
70 * keep-alive, 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);
124 * Use this function to proceed with scheduling reset_work for a controller
125 * that had previously been set to the resetting state. This is intended for
126 * code paths that can't be interrupted by other reset attempts. A hot removal
127 * may prevent this from succeeding.
129 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
131 if (ctrl->state != NVME_CTRL_RESETTING)
133 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
137 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
139 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
141 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
143 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
147 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
149 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
153 ret = nvme_reset_ctrl(ctrl);
155 flush_work(&ctrl->reset_work);
156 if (ctrl->state != NVME_CTRL_LIVE)
162 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
164 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
166 dev_info(ctrl->device,
167 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
169 flush_work(&ctrl->reset_work);
170 nvme_stop_ctrl(ctrl);
171 nvme_remove_namespaces(ctrl);
172 ctrl->ops->delete_ctrl(ctrl);
173 nvme_uninit_ctrl(ctrl);
176 static void nvme_delete_ctrl_work(struct work_struct *work)
178 struct nvme_ctrl *ctrl =
179 container_of(work, struct nvme_ctrl, delete_work);
181 nvme_do_delete_ctrl(ctrl);
184 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
186 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
188 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
192 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
194 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
197 * Keep a reference until nvme_do_delete_ctrl() complete,
198 * since ->delete_ctrl can free the controller.
201 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
202 nvme_do_delete_ctrl(ctrl);
206 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
208 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
211 static blk_status_t nvme_error_status(u16 status)
213 switch (status & 0x7ff) {
214 case NVME_SC_SUCCESS:
216 case NVME_SC_CAP_EXCEEDED:
217 return BLK_STS_NOSPC;
218 case NVME_SC_LBA_RANGE:
219 case NVME_SC_CMD_INTERRUPTED:
220 case NVME_SC_NS_NOT_READY:
221 return BLK_STS_TARGET;
222 case NVME_SC_BAD_ATTRIBUTES:
223 case NVME_SC_ONCS_NOT_SUPPORTED:
224 case NVME_SC_INVALID_OPCODE:
225 case NVME_SC_INVALID_FIELD:
226 case NVME_SC_INVALID_NS:
227 return BLK_STS_NOTSUPP;
228 case NVME_SC_WRITE_FAULT:
229 case NVME_SC_READ_ERROR:
230 case NVME_SC_UNWRITTEN_BLOCK:
231 case NVME_SC_ACCESS_DENIED:
232 case NVME_SC_READ_ONLY:
233 case NVME_SC_COMPARE_FAILED:
234 return BLK_STS_MEDIUM;
235 case NVME_SC_GUARD_CHECK:
236 case NVME_SC_APPTAG_CHECK:
237 case NVME_SC_REFTAG_CHECK:
238 case NVME_SC_INVALID_PI:
239 return BLK_STS_PROTECTION;
240 case NVME_SC_RESERVATION_CONFLICT:
241 return BLK_STS_NEXUS;
242 case NVME_SC_HOST_PATH_ERROR:
243 return BLK_STS_TRANSPORT;
245 return BLK_STS_IOERR;
249 static inline bool nvme_req_needs_retry(struct request *req)
251 if (blk_noretry_request(req))
253 if (nvme_req(req)->status & NVME_SC_DNR)
255 if (nvme_req(req)->retries >= nvme_max_retries)
260 static void nvme_retry_req(struct request *req)
262 struct nvme_ns *ns = req->q->queuedata;
263 unsigned long delay = 0;
266 /* The mask and shift result must be <= 3 */
267 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
269 delay = ns->ctrl->crdt[crd - 1] * 100;
271 nvme_req(req)->retries++;
272 blk_mq_requeue_request(req, false);
273 blk_mq_delay_kick_requeue_list(req->q, delay);
276 void nvme_complete_rq(struct request *req)
278 blk_status_t status = nvme_error_status(nvme_req(req)->status);
280 trace_nvme_complete_rq(req);
282 nvme_cleanup_cmd(req);
284 if (nvme_req(req)->ctrl->kas)
285 nvme_req(req)->ctrl->comp_seen = true;
287 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
288 if ((req->cmd_flags & REQ_NVME_MPATH) && nvme_failover_req(req))
291 if (!blk_queue_dying(req->q)) {
297 nvme_trace_bio_complete(req, status);
298 blk_mq_end_request(req, status);
300 EXPORT_SYMBOL_GPL(nvme_complete_rq);
302 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
304 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
305 "Cancelling I/O %d", req->tag);
307 /* don't abort one completed request */
308 if (blk_mq_request_completed(req))
311 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
312 blk_mq_complete_request(req);
315 EXPORT_SYMBOL_GPL(nvme_cancel_request);
317 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
318 enum nvme_ctrl_state new_state)
320 enum nvme_ctrl_state old_state;
322 bool changed = false;
324 spin_lock_irqsave(&ctrl->lock, flags);
326 old_state = ctrl->state;
331 case NVME_CTRL_RESETTING:
332 case NVME_CTRL_CONNECTING:
339 case NVME_CTRL_RESETTING:
349 case NVME_CTRL_CONNECTING:
352 case NVME_CTRL_RESETTING:
359 case NVME_CTRL_DELETING:
362 case NVME_CTRL_RESETTING:
363 case NVME_CTRL_CONNECTING:
372 case NVME_CTRL_DELETING:
384 ctrl->state = new_state;
385 wake_up_all(&ctrl->state_wq);
388 spin_unlock_irqrestore(&ctrl->lock, flags);
389 if (changed && ctrl->state == NVME_CTRL_LIVE)
390 nvme_kick_requeue_lists(ctrl);
393 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
396 * Returns true for sink states that can't ever transition back to live.
398 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
400 switch (ctrl->state) {
403 case NVME_CTRL_RESETTING:
404 case NVME_CTRL_CONNECTING:
406 case NVME_CTRL_DELETING:
410 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
416 * Waits for the controller state to be resetting, or returns false if it is
417 * not possible to ever transition to that state.
419 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
421 wait_event(ctrl->state_wq,
422 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
423 nvme_state_terminal(ctrl));
424 return ctrl->state == NVME_CTRL_RESETTING;
426 EXPORT_SYMBOL_GPL(nvme_wait_reset);
428 static void nvme_free_ns_head(struct kref *ref)
430 struct nvme_ns_head *head =
431 container_of(ref, struct nvme_ns_head, ref);
433 nvme_mpath_remove_disk(head);
434 ida_simple_remove(&head->subsys->ns_ida, head->instance);
435 list_del_init(&head->entry);
436 cleanup_srcu_struct(&head->srcu);
437 nvme_put_subsystem(head->subsys);
441 static void nvme_put_ns_head(struct nvme_ns_head *head)
443 kref_put(&head->ref, nvme_free_ns_head);
446 static void nvme_free_ns(struct kref *kref)
448 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
451 nvme_nvm_unregister(ns);
454 nvme_put_ns_head(ns->head);
455 nvme_put_ctrl(ns->ctrl);
459 static void nvme_put_ns(struct nvme_ns *ns)
461 kref_put(&ns->kref, nvme_free_ns);
464 static inline void nvme_clear_nvme_request(struct request *req)
466 if (!(req->rq_flags & RQF_DONTPREP)) {
467 nvme_req(req)->retries = 0;
468 nvme_req(req)->flags = 0;
469 req->rq_flags |= RQF_DONTPREP;
473 struct request *nvme_alloc_request(struct request_queue *q,
474 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
476 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
479 if (qid == NVME_QID_ANY) {
480 req = blk_mq_alloc_request(q, op, flags);
482 req = blk_mq_alloc_request_hctx(q, op, flags,
488 req->cmd_flags |= REQ_FAILFAST_DRIVER;
489 nvme_clear_nvme_request(req);
490 nvme_req(req)->cmd = cmd;
494 EXPORT_SYMBOL_GPL(nvme_alloc_request);
496 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
498 struct nvme_command c;
500 memset(&c, 0, sizeof(c));
502 c.directive.opcode = nvme_admin_directive_send;
503 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
504 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
505 c.directive.dtype = NVME_DIR_IDENTIFY;
506 c.directive.tdtype = NVME_DIR_STREAMS;
507 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
509 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
512 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
514 return nvme_toggle_streams(ctrl, false);
517 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
519 return nvme_toggle_streams(ctrl, true);
522 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
523 struct streams_directive_params *s, u32 nsid)
525 struct nvme_command c;
527 memset(&c, 0, sizeof(c));
528 memset(s, 0, sizeof(*s));
530 c.directive.opcode = nvme_admin_directive_recv;
531 c.directive.nsid = cpu_to_le32(nsid);
532 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
533 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
534 c.directive.dtype = NVME_DIR_STREAMS;
536 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
539 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
541 struct streams_directive_params s;
544 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
549 ret = nvme_enable_streams(ctrl);
553 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
557 ctrl->nssa = le16_to_cpu(s.nssa);
558 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
559 dev_info(ctrl->device, "too few streams (%u) available\n",
561 nvme_disable_streams(ctrl);
565 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
566 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
571 * Check if 'req' has a write hint associated with it. If it does, assign
572 * a valid namespace stream to the write.
574 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
575 struct request *req, u16 *control,
578 enum rw_hint streamid = req->write_hint;
580 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
584 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
587 *control |= NVME_RW_DTYPE_STREAMS;
588 *dsmgmt |= streamid << 16;
591 if (streamid < ARRAY_SIZE(req->q->write_hints))
592 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
595 static inline void nvme_setup_flush(struct nvme_ns *ns,
596 struct nvme_command *cmnd)
598 cmnd->common.opcode = nvme_cmd_flush;
599 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
602 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
603 struct nvme_command *cmnd)
605 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
606 struct nvme_dsm_range *range;
610 * Some devices do not consider the DSM 'Number of Ranges' field when
611 * determining how much data to DMA. Always allocate memory for maximum
612 * number of segments to prevent device reading beyond end of buffer.
614 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
616 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
619 * If we fail allocation our range, fallback to the controller
620 * discard page. If that's also busy, it's safe to return
621 * busy, as we know we can make progress once that's freed.
623 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
624 return BLK_STS_RESOURCE;
626 range = page_address(ns->ctrl->discard_page);
629 __rq_for_each_bio(bio, req) {
630 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
631 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
634 range[n].cattr = cpu_to_le32(0);
635 range[n].nlb = cpu_to_le32(nlb);
636 range[n].slba = cpu_to_le64(slba);
641 if (WARN_ON_ONCE(n != segments)) {
642 if (virt_to_page(range) == ns->ctrl->discard_page)
643 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
646 return BLK_STS_IOERR;
649 cmnd->dsm.opcode = nvme_cmd_dsm;
650 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
651 cmnd->dsm.nr = cpu_to_le32(segments - 1);
652 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
654 req->special_vec.bv_page = virt_to_page(range);
655 req->special_vec.bv_offset = offset_in_page(range);
656 req->special_vec.bv_len = alloc_size;
657 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
662 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
663 struct request *req, struct nvme_command *cmnd)
665 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
666 return nvme_setup_discard(ns, req, cmnd);
668 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
669 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
670 cmnd->write_zeroes.slba =
671 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
672 cmnd->write_zeroes.length =
673 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
674 cmnd->write_zeroes.control = 0;
678 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
679 struct request *req, struct nvme_command *cmnd)
681 struct nvme_ctrl *ctrl = ns->ctrl;
685 if (req->cmd_flags & REQ_FUA)
686 control |= NVME_RW_FUA;
687 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
688 control |= NVME_RW_LR;
690 if (req->cmd_flags & REQ_RAHEAD)
691 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
693 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
694 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
695 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
696 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
698 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
699 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
703 * If formated with metadata, the block layer always provides a
704 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
705 * we enable the PRACT bit for protection information or set the
706 * namespace capacity to zero to prevent any I/O.
708 if (!blk_integrity_rq(req)) {
709 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
710 return BLK_STS_NOTSUPP;
711 control |= NVME_RW_PRINFO_PRACT;
714 switch (ns->pi_type) {
715 case NVME_NS_DPS_PI_TYPE3:
716 control |= NVME_RW_PRINFO_PRCHK_GUARD;
718 case NVME_NS_DPS_PI_TYPE1:
719 case NVME_NS_DPS_PI_TYPE2:
720 control |= NVME_RW_PRINFO_PRCHK_GUARD |
721 NVME_RW_PRINFO_PRCHK_REF;
722 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
727 cmnd->rw.control = cpu_to_le16(control);
728 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
732 void nvme_cleanup_cmd(struct request *req)
734 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
735 struct nvme_ns *ns = req->rq_disk->private_data;
736 struct page *page = req->special_vec.bv_page;
738 if (page == ns->ctrl->discard_page)
739 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
741 kfree(page_address(page) + req->special_vec.bv_offset);
744 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
746 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
747 struct nvme_command *cmd)
749 blk_status_t ret = BLK_STS_OK;
751 nvme_clear_nvme_request(req);
753 memset(cmd, 0, sizeof(*cmd));
754 switch (req_op(req)) {
757 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
760 nvme_setup_flush(ns, cmd);
762 case REQ_OP_WRITE_ZEROES:
763 ret = nvme_setup_write_zeroes(ns, req, cmd);
766 ret = nvme_setup_discard(ns, req, cmd);
770 ret = nvme_setup_rw(ns, req, cmd);
774 return BLK_STS_IOERR;
777 cmd->common.command_id = req->tag;
778 trace_nvme_setup_cmd(req, cmd);
781 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
783 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
785 struct completion *waiting = rq->end_io_data;
787 rq->end_io_data = NULL;
791 static void nvme_execute_rq_polled(struct request_queue *q,
792 struct gendisk *bd_disk, struct request *rq, int at_head)
794 DECLARE_COMPLETION_ONSTACK(wait);
796 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
798 rq->cmd_flags |= REQ_HIPRI;
799 rq->end_io_data = &wait;
800 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
802 while (!completion_done(&wait)) {
803 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
809 * Returns 0 on success. If the result is negative, it's a Linux error code;
810 * if the result is positive, it's an NVM Express status code
812 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
813 union nvme_result *result, void *buffer, unsigned bufflen,
814 unsigned timeout, int qid, int at_head,
815 blk_mq_req_flags_t flags, bool poll)
820 req = nvme_alloc_request(q, cmd, flags, qid);
824 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
826 if (buffer && bufflen) {
827 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
833 nvme_execute_rq_polled(req->q, NULL, req, at_head);
835 blk_execute_rq(req->q, NULL, req, at_head);
837 *result = nvme_req(req)->result;
838 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
841 ret = nvme_req(req)->status;
843 blk_mq_free_request(req);
846 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
848 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
849 void *buffer, unsigned bufflen)
851 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
852 NVME_QID_ANY, 0, 0, false);
854 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
856 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
857 unsigned len, u32 seed, bool write)
859 struct bio_integrity_payload *bip;
863 buf = kmalloc(len, GFP_KERNEL);
868 if (write && copy_from_user(buf, ubuf, len))
871 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
877 bip->bip_iter.bi_size = len;
878 bip->bip_iter.bi_sector = seed;
879 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
880 offset_in_page(buf));
890 static int nvme_submit_user_cmd(struct request_queue *q,
891 struct nvme_command *cmd, void __user *ubuffer,
892 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
893 u32 meta_seed, u64 *result, unsigned timeout)
895 bool write = nvme_is_write(cmd);
896 struct nvme_ns *ns = q->queuedata;
897 struct gendisk *disk = ns ? ns->disk : NULL;
899 struct bio *bio = NULL;
903 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
907 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
908 nvme_req(req)->flags |= NVME_REQ_USERCMD;
910 if (ubuffer && bufflen) {
911 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
917 if (disk && meta_buffer && meta_len) {
918 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
924 req->cmd_flags |= REQ_INTEGRITY;
928 blk_execute_rq(req->q, disk, req, 0);
929 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
932 ret = nvme_req(req)->status;
934 *result = le64_to_cpu(nvme_req(req)->result.u64);
935 if (meta && !ret && !write) {
936 if (copy_to_user(meta_buffer, meta, meta_len))
942 blk_rq_unmap_user(bio);
944 blk_mq_free_request(req);
948 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
950 struct nvme_ctrl *ctrl = rq->end_io_data;
952 bool startka = false;
954 blk_mq_free_request(rq);
957 dev_err(ctrl->device,
958 "failed nvme_keep_alive_end_io error=%d\n",
963 ctrl->comp_seen = false;
964 spin_lock_irqsave(&ctrl->lock, flags);
965 if (ctrl->state == NVME_CTRL_LIVE ||
966 ctrl->state == NVME_CTRL_CONNECTING)
968 spin_unlock_irqrestore(&ctrl->lock, flags);
970 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
973 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
977 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
982 rq->timeout = ctrl->kato * HZ;
983 rq->end_io_data = ctrl;
985 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
990 static void nvme_keep_alive_work(struct work_struct *work)
992 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
993 struct nvme_ctrl, ka_work);
994 bool comp_seen = ctrl->comp_seen;
996 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
997 dev_dbg(ctrl->device,
998 "reschedule traffic based keep-alive timer\n");
999 ctrl->comp_seen = false;
1000 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1004 if (nvme_keep_alive(ctrl)) {
1005 /* allocation failure, reset the controller */
1006 dev_err(ctrl->device, "keep-alive failed\n");
1007 nvme_reset_ctrl(ctrl);
1012 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1014 if (unlikely(ctrl->kato == 0))
1017 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1020 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1022 if (unlikely(ctrl->kato == 0))
1025 cancel_delayed_work_sync(&ctrl->ka_work);
1027 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1029 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1031 struct nvme_command c = { };
1034 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1035 c.identify.opcode = nvme_admin_identify;
1036 c.identify.cns = NVME_ID_CNS_CTRL;
1038 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1042 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1043 sizeof(struct nvme_id_ctrl));
1049 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1050 struct nvme_ns_id_desc *cur)
1052 const char *warn_str = "ctrl returned bogus length:";
1055 switch (cur->nidt) {
1056 case NVME_NIDT_EUI64:
1057 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1058 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1059 warn_str, cur->nidl);
1062 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1063 return NVME_NIDT_EUI64_LEN;
1064 case NVME_NIDT_NGUID:
1065 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1066 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1067 warn_str, cur->nidl);
1070 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1071 return NVME_NIDT_NGUID_LEN;
1072 case NVME_NIDT_UUID:
1073 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1074 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1075 warn_str, cur->nidl);
1078 uuid_copy(&ids->uuid, data + sizeof(*cur));
1079 return NVME_NIDT_UUID_LEN;
1081 /* Skip unknown types */
1086 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1087 struct nvme_ns_ids *ids)
1089 struct nvme_command c = { };
1095 c.identify.opcode = nvme_admin_identify;
1096 c.identify.nsid = cpu_to_le32(nsid);
1097 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1099 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1103 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1104 NVME_IDENTIFY_DATA_SIZE);
1106 dev_warn(ctrl->device,
1107 "Identify Descriptors failed (%d)\n", status);
1109 * Don't treat an error as fatal, as we potentially already
1110 * have a NGUID or EUI-64.
1117 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1118 struct nvme_ns_id_desc *cur = data + pos;
1123 len = nvme_process_ns_desc(ctrl, ids, cur);
1127 len += sizeof(*cur);
1134 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1136 struct nvme_command c = { };
1138 c.identify.opcode = nvme_admin_identify;
1139 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1140 c.identify.nsid = cpu_to_le32(nsid);
1141 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1142 NVME_IDENTIFY_DATA_SIZE);
1145 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1146 unsigned nsid, struct nvme_id_ns **id)
1148 struct nvme_command c = { };
1151 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1152 c.identify.opcode = nvme_admin_identify;
1153 c.identify.nsid = cpu_to_le32(nsid);
1154 c.identify.cns = NVME_ID_CNS_NS;
1156 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1160 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1162 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1169 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1170 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1172 union nvme_result res = { 0 };
1173 struct nvme_command c;
1176 memset(&c, 0, sizeof(c));
1177 c.features.opcode = op;
1178 c.features.fid = cpu_to_le32(fid);
1179 c.features.dword11 = cpu_to_le32(dword11);
1181 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1182 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1183 if (ret >= 0 && result)
1184 *result = le32_to_cpu(res.u32);
1188 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1189 unsigned int dword11, void *buffer, size_t buflen,
1192 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1195 EXPORT_SYMBOL_GPL(nvme_set_features);
1197 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1198 unsigned int dword11, void *buffer, size_t buflen,
1201 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1204 EXPORT_SYMBOL_GPL(nvme_get_features);
1206 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1208 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1210 int status, nr_io_queues;
1212 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1218 * Degraded controllers might return an error when setting the queue
1219 * count. We still want to be able to bring them online and offer
1220 * access to the admin queue, as that might be only way to fix them up.
1223 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1226 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1227 *count = min(*count, nr_io_queues);
1232 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1234 #define NVME_AEN_SUPPORTED \
1235 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1236 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1238 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1240 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1243 if (!supported_aens)
1246 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1249 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1252 queue_work(nvme_wq, &ctrl->async_event_work);
1255 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1257 struct nvme_user_io io;
1258 struct nvme_command c;
1259 unsigned length, meta_len;
1260 void __user *metadata;
1262 if (copy_from_user(&io, uio, sizeof(io)))
1267 switch (io.opcode) {
1268 case nvme_cmd_write:
1270 case nvme_cmd_compare:
1276 length = (io.nblocks + 1) << ns->lba_shift;
1277 meta_len = (io.nblocks + 1) * ns->ms;
1278 metadata = (void __user *)(uintptr_t)io.metadata;
1283 } else if (meta_len) {
1284 if ((io.metadata & 3) || !io.metadata)
1288 memset(&c, 0, sizeof(c));
1289 c.rw.opcode = io.opcode;
1290 c.rw.flags = io.flags;
1291 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1292 c.rw.slba = cpu_to_le64(io.slba);
1293 c.rw.length = cpu_to_le16(io.nblocks);
1294 c.rw.control = cpu_to_le16(io.control);
1295 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1296 c.rw.reftag = cpu_to_le32(io.reftag);
1297 c.rw.apptag = cpu_to_le16(io.apptag);
1298 c.rw.appmask = cpu_to_le16(io.appmask);
1300 return nvme_submit_user_cmd(ns->queue, &c,
1301 (void __user *)(uintptr_t)io.addr, length,
1302 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1305 static u32 nvme_known_admin_effects(u8 opcode)
1308 case nvme_admin_format_nvm:
1309 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1310 NVME_CMD_EFFECTS_CSE_MASK;
1311 case nvme_admin_sanitize_nvm:
1312 return NVME_CMD_EFFECTS_CSE_MASK;
1319 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1326 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1327 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1328 dev_warn(ctrl->device,
1329 "IO command:%02x has unhandled effects:%08x\n",
1335 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1336 effects |= nvme_known_admin_effects(opcode);
1339 * For simplicity, IO to all namespaces is quiesced even if the command
1340 * effects say only one namespace is affected.
1342 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1343 mutex_lock(&ctrl->scan_lock);
1344 mutex_lock(&ctrl->subsys->lock);
1345 nvme_mpath_start_freeze(ctrl->subsys);
1346 nvme_mpath_wait_freeze(ctrl->subsys);
1347 nvme_start_freeze(ctrl);
1348 nvme_wait_freeze(ctrl);
1353 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1357 down_read(&ctrl->namespaces_rwsem);
1358 list_for_each_entry(ns, &ctrl->namespaces, list)
1359 if (ns->disk && nvme_revalidate_disk(ns->disk))
1360 nvme_set_queue_dying(ns);
1361 up_read(&ctrl->namespaces_rwsem);
1364 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1367 * Revalidate LBA changes prior to unfreezing. This is necessary to
1368 * prevent memory corruption if a logical block size was changed by
1371 if (effects & NVME_CMD_EFFECTS_LBCC)
1372 nvme_update_formats(ctrl);
1373 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1374 nvme_unfreeze(ctrl);
1375 nvme_mpath_unfreeze(ctrl->subsys);
1376 mutex_unlock(&ctrl->subsys->lock);
1377 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1378 mutex_unlock(&ctrl->scan_lock);
1380 if (effects & NVME_CMD_EFFECTS_CCC)
1381 nvme_init_identify(ctrl);
1382 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1383 nvme_queue_scan(ctrl);
1386 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1387 struct nvme_passthru_cmd __user *ucmd)
1389 struct nvme_passthru_cmd cmd;
1390 struct nvme_command c;
1391 unsigned timeout = 0;
1396 if (!capable(CAP_SYS_ADMIN))
1398 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1403 memset(&c, 0, sizeof(c));
1404 c.common.opcode = cmd.opcode;
1405 c.common.flags = cmd.flags;
1406 c.common.nsid = cpu_to_le32(cmd.nsid);
1407 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1408 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1409 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1410 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1411 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1412 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1413 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1414 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1417 timeout = msecs_to_jiffies(cmd.timeout_ms);
1419 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1420 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1421 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1422 (void __user *)(uintptr_t)cmd.metadata,
1423 cmd.metadata_len, 0, &result, timeout);
1424 nvme_passthru_end(ctrl, effects);
1427 if (put_user(result, &ucmd->result))
1434 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1435 struct nvme_passthru_cmd64 __user *ucmd)
1437 struct nvme_passthru_cmd64 cmd;
1438 struct nvme_command c;
1439 unsigned timeout = 0;
1443 if (!capable(CAP_SYS_ADMIN))
1445 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1450 memset(&c, 0, sizeof(c));
1451 c.common.opcode = cmd.opcode;
1452 c.common.flags = cmd.flags;
1453 c.common.nsid = cpu_to_le32(cmd.nsid);
1454 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1455 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1456 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1457 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1458 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1459 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1460 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1461 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1464 timeout = msecs_to_jiffies(cmd.timeout_ms);
1466 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1467 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1468 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1469 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1470 0, &cmd.result, timeout);
1471 nvme_passthru_end(ctrl, effects);
1474 if (put_user(cmd.result, &ucmd->result))
1482 * Issue ioctl requests on the first available path. Note that unlike normal
1483 * block layer requests we will not retry failed request on another controller.
1485 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1486 struct nvme_ns_head **head, int *srcu_idx)
1488 #ifdef CONFIG_NVME_MULTIPATH
1489 if (disk->fops == &nvme_ns_head_ops) {
1492 *head = disk->private_data;
1493 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1494 ns = nvme_find_path(*head);
1496 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1502 return disk->private_data;
1505 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1508 srcu_read_unlock(&head->srcu, idx);
1511 static bool is_ctrl_ioctl(unsigned int cmd)
1513 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1515 if (is_sed_ioctl(cmd))
1520 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1522 struct nvme_ns_head *head,
1525 struct nvme_ctrl *ctrl = ns->ctrl;
1528 nvme_get_ctrl(ns->ctrl);
1529 nvme_put_ns_from_disk(head, srcu_idx);
1532 case NVME_IOCTL_ADMIN_CMD:
1533 ret = nvme_user_cmd(ctrl, NULL, argp);
1535 case NVME_IOCTL_ADMIN64_CMD:
1536 ret = nvme_user_cmd64(ctrl, NULL, argp);
1539 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1542 nvme_put_ctrl(ctrl);
1546 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1547 unsigned int cmd, unsigned long arg)
1549 struct nvme_ns_head *head = NULL;
1550 void __user *argp = (void __user *)arg;
1554 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1556 return -EWOULDBLOCK;
1559 * Handle ioctls that apply to the controller instead of the namespace
1560 * seperately and drop the ns SRCU reference early. This avoids a
1561 * deadlock when deleting namespaces using the passthrough interface.
1563 if (is_ctrl_ioctl(cmd))
1564 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1568 force_successful_syscall_return();
1569 ret = ns->head->ns_id;
1571 case NVME_IOCTL_IO_CMD:
1572 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1574 case NVME_IOCTL_SUBMIT_IO:
1575 ret = nvme_submit_io(ns, argp);
1577 case NVME_IOCTL_IO64_CMD:
1578 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1582 ret = nvme_nvm_ioctl(ns, cmd, arg);
1587 nvme_put_ns_from_disk(head, srcu_idx);
1591 #ifdef CONFIG_COMPAT
1592 struct nvme_user_io32 {
1605 } __attribute__((__packed__));
1607 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1609 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1610 unsigned int cmd, unsigned long arg)
1613 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1614 * between 32 bit programs and 64 bit kernel.
1615 * The cause is that the results of sizeof(struct nvme_user_io),
1616 * which is used to define NVME_IOCTL_SUBMIT_IO,
1617 * are not same between 32 bit compiler and 64 bit compiler.
1618 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1619 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1620 * Other IOCTL numbers are same between 32 bit and 64 bit.
1621 * So there is nothing to do regarding to other IOCTL numbers.
1623 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1624 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1626 return nvme_ioctl(bdev, mode, cmd, arg);
1629 #define nvme_compat_ioctl NULL
1630 #endif /* CONFIG_COMPAT */
1632 static int nvme_open(struct block_device *bdev, fmode_t mode)
1634 struct nvme_ns *ns = bdev->bd_disk->private_data;
1636 #ifdef CONFIG_NVME_MULTIPATH
1637 /* should never be called due to GENHD_FL_HIDDEN */
1638 if (WARN_ON_ONCE(ns->head->disk))
1641 if (!kref_get_unless_zero(&ns->kref))
1643 if (!try_module_get(ns->ctrl->ops->module))
1654 static void nvme_release(struct gendisk *disk, fmode_t mode)
1656 struct nvme_ns *ns = disk->private_data;
1658 module_put(ns->ctrl->ops->module);
1662 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1664 /* some standard values */
1665 geo->heads = 1 << 6;
1666 geo->sectors = 1 << 5;
1667 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1671 #ifdef CONFIG_BLK_DEV_INTEGRITY
1672 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1674 struct blk_integrity integrity;
1676 memset(&integrity, 0, sizeof(integrity));
1678 case NVME_NS_DPS_PI_TYPE3:
1679 integrity.profile = &t10_pi_type3_crc;
1680 integrity.tag_size = sizeof(u16) + sizeof(u32);
1681 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1683 case NVME_NS_DPS_PI_TYPE1:
1684 case NVME_NS_DPS_PI_TYPE2:
1685 integrity.profile = &t10_pi_type1_crc;
1686 integrity.tag_size = sizeof(u16);
1687 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1690 integrity.profile = NULL;
1693 integrity.tuple_size = ms;
1694 blk_integrity_register(disk, &integrity);
1695 blk_queue_max_integrity_segments(disk->queue, 1);
1698 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1701 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1703 static void nvme_set_chunk_size(struct nvme_ns *ns)
1705 u32 chunk_size = nvme_lba_to_sect(ns, ns->noiob);
1706 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1709 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1711 struct nvme_ctrl *ctrl = ns->ctrl;
1712 struct request_queue *queue = disk->queue;
1713 u32 size = queue_logical_block_size(queue);
1715 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1716 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1720 if (ctrl->nr_streams && ns->sws && ns->sgs)
1721 size *= ns->sws * ns->sgs;
1723 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1724 NVME_DSM_MAX_RANGES);
1726 queue->limits.discard_alignment = 0;
1727 queue->limits.discard_granularity = size;
1729 /* If discard is already enabled, don't reset queue limits */
1730 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1733 blk_queue_max_discard_sectors(queue, UINT_MAX);
1734 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1736 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1737 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1740 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1744 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1745 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1748 * Even though NVMe spec explicitly states that MDTS is not
1749 * applicable to the write-zeroes:- "The restriction does not apply to
1750 * commands that do not transfer data between the host and the
1751 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1752 * In order to be more cautious use controller's max_hw_sectors value
1753 * to configure the maximum sectors for the write-zeroes which is
1754 * configured based on the controller's MDTS field in the
1755 * nvme_init_identify() if available.
1757 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1758 max_blocks = (u64)USHRT_MAX + 1;
1760 max_blocks = ns->ctrl->max_hw_sectors + 1;
1762 blk_queue_max_write_zeroes_sectors(disk->queue,
1763 nvme_lba_to_sect(ns, max_blocks));
1766 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1767 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1769 memset(ids, 0, sizeof(*ids));
1771 if (ctrl->vs >= NVME_VS(1, 1, 0))
1772 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1773 if (ctrl->vs >= NVME_VS(1, 2, 0))
1774 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1775 if (ctrl->vs >= NVME_VS(1, 3, 0))
1776 return nvme_identify_ns_descs(ctrl, nsid, ids);
1780 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1782 return !uuid_is_null(&ids->uuid) ||
1783 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1784 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1787 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1789 return uuid_equal(&a->uuid, &b->uuid) &&
1790 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1791 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1794 static void nvme_update_disk_info(struct gendisk *disk,
1795 struct nvme_ns *ns, struct nvme_id_ns *id)
1797 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1798 unsigned short bs = 1 << ns->lba_shift;
1799 u32 atomic_bs, phys_bs, io_opt;
1801 if (ns->lba_shift > PAGE_SHIFT) {
1802 /* unsupported block size, set capacity to 0 later */
1805 blk_mq_freeze_queue(disk->queue);
1806 blk_integrity_unregister(disk);
1808 if (id->nabo == 0) {
1810 * Bit 1 indicates whether NAWUPF is defined for this namespace
1811 * and whether it should be used instead of AWUPF. If NAWUPF ==
1812 * 0 then AWUPF must be used instead.
1814 if (id->nsfeat & (1 << 1) && id->nawupf)
1815 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1817 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1823 if (id->nsfeat & (1 << 4)) {
1824 /* NPWG = Namespace Preferred Write Granularity */
1825 phys_bs *= 1 + le16_to_cpu(id->npwg);
1826 /* NOWS = Namespace Optimal Write Size */
1827 io_opt *= 1 + le16_to_cpu(id->nows);
1830 blk_queue_logical_block_size(disk->queue, bs);
1832 * Linux filesystems assume writing a single physical block is
1833 * an atomic operation. Hence limit the physical block size to the
1834 * value of the Atomic Write Unit Power Fail parameter.
1836 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1837 blk_queue_io_min(disk->queue, phys_bs);
1838 blk_queue_io_opt(disk->queue, io_opt);
1840 if (ns->ms && !ns->ext &&
1841 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1842 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1843 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1844 ns->lba_shift > PAGE_SHIFT)
1847 set_capacity_revalidate_and_notify(disk, capacity, false);
1849 nvme_config_discard(disk, ns);
1850 nvme_config_write_zeroes(disk, ns);
1852 if (id->nsattr & (1 << 0))
1853 set_disk_ro(disk, true);
1855 set_disk_ro(disk, false);
1857 blk_mq_unfreeze_queue(disk->queue);
1860 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1862 struct nvme_ns *ns = disk->private_data;
1865 * If identify namespace failed, use default 512 byte block size so
1866 * block layer can use before failing read/write for 0 capacity.
1868 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1869 if (ns->lba_shift == 0)
1871 ns->noiob = le16_to_cpu(id->noiob);
1872 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1873 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1874 /* the PI implementation requires metadata equal t10 pi tuple size */
1875 if (ns->ms == sizeof(struct t10_pi_tuple))
1876 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1881 nvme_set_chunk_size(ns);
1882 nvme_update_disk_info(disk, ns, id);
1883 #ifdef CONFIG_NVME_MULTIPATH
1884 if (ns->head->disk) {
1885 nvme_update_disk_info(ns->head->disk, ns, id);
1886 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1887 revalidate_disk(ns->head->disk);
1892 static int nvme_revalidate_disk(struct gendisk *disk)
1894 struct nvme_ns *ns = disk->private_data;
1895 struct nvme_ctrl *ctrl = ns->ctrl;
1896 struct nvme_id_ns *id;
1897 struct nvme_ns_ids ids;
1900 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1901 set_capacity(disk, 0);
1905 ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1909 if (id->ncap == 0) {
1914 __nvme_revalidate_disk(disk, id);
1915 ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1919 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1920 dev_err(ctrl->device,
1921 "identifiers changed for nsid %d\n", ns->head->ns_id);
1929 * Only fail the function if we got a fatal error back from the
1930 * device, otherwise ignore the error and just move on.
1932 if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1935 ret = blk_status_to_errno(nvme_error_status(ret));
1939 static char nvme_pr_type(enum pr_type type)
1942 case PR_WRITE_EXCLUSIVE:
1944 case PR_EXCLUSIVE_ACCESS:
1946 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1948 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1950 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1952 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1959 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1960 u64 key, u64 sa_key, u8 op)
1962 struct nvme_ns_head *head = NULL;
1964 struct nvme_command c;
1966 u8 data[16] = { 0, };
1968 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1970 return -EWOULDBLOCK;
1972 put_unaligned_le64(key, &data[0]);
1973 put_unaligned_le64(sa_key, &data[8]);
1975 memset(&c, 0, sizeof(c));
1976 c.common.opcode = op;
1977 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1978 c.common.cdw10 = cpu_to_le32(cdw10);
1980 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1981 nvme_put_ns_from_disk(head, srcu_idx);
1985 static int nvme_pr_register(struct block_device *bdev, u64 old,
1986 u64 new, unsigned flags)
1990 if (flags & ~PR_FL_IGNORE_KEY)
1993 cdw10 = old ? 2 : 0;
1994 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1995 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1996 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1999 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2000 enum pr_type type, unsigned flags)
2004 if (flags & ~PR_FL_IGNORE_KEY)
2007 cdw10 = nvme_pr_type(type) << 8;
2008 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2009 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2012 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2013 enum pr_type type, bool abort)
2015 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2016 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2019 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2021 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2022 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2025 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2027 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2028 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2031 static const struct pr_ops nvme_pr_ops = {
2032 .pr_register = nvme_pr_register,
2033 .pr_reserve = nvme_pr_reserve,
2034 .pr_release = nvme_pr_release,
2035 .pr_preempt = nvme_pr_preempt,
2036 .pr_clear = nvme_pr_clear,
2039 #ifdef CONFIG_BLK_SED_OPAL
2040 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2043 struct nvme_ctrl *ctrl = data;
2044 struct nvme_command cmd;
2046 memset(&cmd, 0, sizeof(cmd));
2048 cmd.common.opcode = nvme_admin_security_send;
2050 cmd.common.opcode = nvme_admin_security_recv;
2051 cmd.common.nsid = 0;
2052 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2053 cmd.common.cdw11 = cpu_to_le32(len);
2055 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2056 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2058 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2059 #endif /* CONFIG_BLK_SED_OPAL */
2061 static const struct block_device_operations nvme_fops = {
2062 .owner = THIS_MODULE,
2063 .ioctl = nvme_ioctl,
2064 .compat_ioctl = nvme_compat_ioctl,
2066 .release = nvme_release,
2067 .getgeo = nvme_getgeo,
2068 .revalidate_disk= nvme_revalidate_disk,
2069 .pr_ops = &nvme_pr_ops,
2072 #ifdef CONFIG_NVME_MULTIPATH
2073 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2075 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2077 if (!kref_get_unless_zero(&head->ref))
2082 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2084 nvme_put_ns_head(disk->private_data);
2087 const struct block_device_operations nvme_ns_head_ops = {
2088 .owner = THIS_MODULE,
2089 .open = nvme_ns_head_open,
2090 .release = nvme_ns_head_release,
2091 .ioctl = nvme_ioctl,
2092 .compat_ioctl = nvme_compat_ioctl,
2093 .getgeo = nvme_getgeo,
2094 .pr_ops = &nvme_pr_ops,
2096 #endif /* CONFIG_NVME_MULTIPATH */
2098 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2100 unsigned long timeout =
2101 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2102 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2105 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2108 if ((csts & NVME_CSTS_RDY) == bit)
2111 usleep_range(1000, 2000);
2112 if (fatal_signal_pending(current))
2114 if (time_after(jiffies, timeout)) {
2115 dev_err(ctrl->device,
2116 "Device not ready; aborting %s, CSTS=0x%x\n",
2117 enabled ? "initialisation" : "reset", csts);
2126 * If the device has been passed off to us in an enabled state, just clear
2127 * the enabled bit. The spec says we should set the 'shutdown notification
2128 * bits', but doing so may cause the device to complete commands to the
2129 * admin queue ... and we don't know what memory that might be pointing at!
2131 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2135 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2136 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2138 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2142 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2143 msleep(NVME_QUIRK_DELAY_AMOUNT);
2145 return nvme_wait_ready(ctrl, ctrl->cap, false);
2147 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2149 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2152 * Default to a 4K page size, with the intention to update this
2153 * path in the future to accomodate architectures with differing
2154 * kernel and IO page sizes.
2156 unsigned dev_page_min, page_shift = 12;
2159 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2161 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2164 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2166 if (page_shift < dev_page_min) {
2167 dev_err(ctrl->device,
2168 "Minimum device page size %u too large for host (%u)\n",
2169 1 << dev_page_min, 1 << page_shift);
2173 ctrl->page_size = 1 << page_shift;
2175 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2176 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2177 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2178 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2179 ctrl->ctrl_config |= NVME_CC_ENABLE;
2181 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2184 return nvme_wait_ready(ctrl, ctrl->cap, true);
2186 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2188 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2190 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2194 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2195 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2197 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2201 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2202 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2206 if (fatal_signal_pending(current))
2208 if (time_after(jiffies, timeout)) {
2209 dev_err(ctrl->device,
2210 "Device shutdown incomplete; abort shutdown\n");
2217 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2219 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2220 struct request_queue *q)
2224 if (ctrl->max_hw_sectors) {
2226 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2228 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2229 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2230 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2232 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2233 is_power_of_2(ctrl->max_hw_sectors))
2234 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2235 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2236 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2238 blk_queue_write_cache(q, vwc, vwc);
2241 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2246 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2249 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2250 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2253 dev_warn_once(ctrl->device,
2254 "could not set timestamp (%d)\n", ret);
2258 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2260 struct nvme_feat_host_behavior *host;
2263 /* Don't bother enabling the feature if retry delay is not reported */
2267 host = kzalloc(sizeof(*host), GFP_KERNEL);
2271 host->acre = NVME_ENABLE_ACRE;
2272 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2273 host, sizeof(*host), NULL);
2278 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2281 * APST (Autonomous Power State Transition) lets us program a
2282 * table of power state transitions that the controller will
2283 * perform automatically. We configure it with a simple
2284 * heuristic: we are willing to spend at most 2% of the time
2285 * transitioning between power states. Therefore, when running
2286 * in any given state, we will enter the next lower-power
2287 * non-operational state after waiting 50 * (enlat + exlat)
2288 * microseconds, as long as that state's exit latency is under
2289 * the requested maximum latency.
2291 * We will not autonomously enter any non-operational state for
2292 * which the total latency exceeds ps_max_latency_us. Users
2293 * can set ps_max_latency_us to zero to turn off APST.
2297 struct nvme_feat_auto_pst *table;
2303 * If APST isn't supported or if we haven't been initialized yet,
2304 * then don't do anything.
2309 if (ctrl->npss > 31) {
2310 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2314 table = kzalloc(sizeof(*table), GFP_KERNEL);
2318 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2319 /* Turn off APST. */
2321 dev_dbg(ctrl->device, "APST disabled\n");
2323 __le64 target = cpu_to_le64(0);
2327 * Walk through all states from lowest- to highest-power.
2328 * According to the spec, lower-numbered states use more
2329 * power. NPSS, despite the name, is the index of the
2330 * lowest-power state, not the number of states.
2332 for (state = (int)ctrl->npss; state >= 0; state--) {
2333 u64 total_latency_us, exit_latency_us, transition_ms;
2336 table->entries[state] = target;
2339 * Don't allow transitions to the deepest state
2340 * if it's quirked off.
2342 if (state == ctrl->npss &&
2343 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2347 * Is this state a useful non-operational state for
2348 * higher-power states to autonomously transition to?
2350 if (!(ctrl->psd[state].flags &
2351 NVME_PS_FLAGS_NON_OP_STATE))
2355 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2356 if (exit_latency_us > ctrl->ps_max_latency_us)
2361 le32_to_cpu(ctrl->psd[state].entry_lat);
2364 * This state is good. Use it as the APST idle
2365 * target for higher power states.
2367 transition_ms = total_latency_us + 19;
2368 do_div(transition_ms, 20);
2369 if (transition_ms > (1 << 24) - 1)
2370 transition_ms = (1 << 24) - 1;
2372 target = cpu_to_le64((state << 3) |
2373 (transition_ms << 8));
2378 if (total_latency_us > max_lat_us)
2379 max_lat_us = total_latency_us;
2385 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2387 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2388 max_ps, max_lat_us, (int)sizeof(*table), table);
2392 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2393 table, sizeof(*table), NULL);
2395 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2401 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2403 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2407 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2408 case PM_QOS_LATENCY_ANY:
2416 if (ctrl->ps_max_latency_us != latency) {
2417 ctrl->ps_max_latency_us = latency;
2418 nvme_configure_apst(ctrl);
2422 struct nvme_core_quirk_entry {
2424 * NVMe model and firmware strings are padded with spaces. For
2425 * simplicity, strings in the quirk table are padded with NULLs
2431 unsigned long quirks;
2434 static const struct nvme_core_quirk_entry core_quirks[] = {
2437 * This Toshiba device seems to die using any APST states. See:
2438 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2441 .mn = "THNSF5256GPUK TOSHIBA",
2442 .quirks = NVME_QUIRK_NO_APST,
2446 * This LiteON CL1-3D*-Q11 firmware version has a race
2447 * condition associated with actions related to suspend to idle
2448 * LiteON has resolved the problem in future firmware
2452 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2456 /* match is null-terminated but idstr is space-padded. */
2457 static bool string_matches(const char *idstr, const char *match, size_t len)
2464 matchlen = strlen(match);
2465 WARN_ON_ONCE(matchlen > len);
2467 if (memcmp(idstr, match, matchlen))
2470 for (; matchlen < len; matchlen++)
2471 if (idstr[matchlen] != ' ')
2477 static bool quirk_matches(const struct nvme_id_ctrl *id,
2478 const struct nvme_core_quirk_entry *q)
2480 return q->vid == le16_to_cpu(id->vid) &&
2481 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2482 string_matches(id->fr, q->fr, sizeof(id->fr));
2485 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2486 struct nvme_id_ctrl *id)
2491 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2492 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2493 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2494 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2498 if (ctrl->vs >= NVME_VS(1, 2, 1))
2499 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2502 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2503 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2504 "nqn.2014.08.org.nvmexpress:%04x%04x",
2505 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2506 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2507 off += sizeof(id->sn);
2508 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2509 off += sizeof(id->mn);
2510 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2513 static void nvme_release_subsystem(struct device *dev)
2515 struct nvme_subsystem *subsys =
2516 container_of(dev, struct nvme_subsystem, dev);
2518 if (subsys->instance >= 0)
2519 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2523 static void nvme_destroy_subsystem(struct kref *ref)
2525 struct nvme_subsystem *subsys =
2526 container_of(ref, struct nvme_subsystem, ref);
2528 mutex_lock(&nvme_subsystems_lock);
2529 list_del(&subsys->entry);
2530 mutex_unlock(&nvme_subsystems_lock);
2532 ida_destroy(&subsys->ns_ida);
2533 device_del(&subsys->dev);
2534 put_device(&subsys->dev);
2537 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2539 kref_put(&subsys->ref, nvme_destroy_subsystem);
2542 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2544 struct nvme_subsystem *subsys;
2546 lockdep_assert_held(&nvme_subsystems_lock);
2549 * Fail matches for discovery subsystems. This results
2550 * in each discovery controller bound to a unique subsystem.
2551 * This avoids issues with validating controller values
2552 * that can only be true when there is a single unique subsystem.
2553 * There may be multiple and completely independent entities
2554 * that provide discovery controllers.
2556 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2559 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2560 if (strcmp(subsys->subnqn, subsysnqn))
2562 if (!kref_get_unless_zero(&subsys->ref))
2570 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2571 struct device_attribute subsys_attr_##_name = \
2572 __ATTR(_name, _mode, _show, NULL)
2574 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2575 struct device_attribute *attr,
2578 struct nvme_subsystem *subsys =
2579 container_of(dev, struct nvme_subsystem, dev);
2581 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2583 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2585 #define nvme_subsys_show_str_function(field) \
2586 static ssize_t subsys_##field##_show(struct device *dev, \
2587 struct device_attribute *attr, char *buf) \
2589 struct nvme_subsystem *subsys = \
2590 container_of(dev, struct nvme_subsystem, dev); \
2591 return sprintf(buf, "%.*s\n", \
2592 (int)sizeof(subsys->field), subsys->field); \
2594 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2596 nvme_subsys_show_str_function(model);
2597 nvme_subsys_show_str_function(serial);
2598 nvme_subsys_show_str_function(firmware_rev);
2600 static struct attribute *nvme_subsys_attrs[] = {
2601 &subsys_attr_model.attr,
2602 &subsys_attr_serial.attr,
2603 &subsys_attr_firmware_rev.attr,
2604 &subsys_attr_subsysnqn.attr,
2605 #ifdef CONFIG_NVME_MULTIPATH
2606 &subsys_attr_iopolicy.attr,
2611 static struct attribute_group nvme_subsys_attrs_group = {
2612 .attrs = nvme_subsys_attrs,
2615 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2616 &nvme_subsys_attrs_group,
2620 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2621 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2623 struct nvme_ctrl *tmp;
2625 lockdep_assert_held(&nvme_subsystems_lock);
2627 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2628 if (nvme_state_terminal(tmp))
2631 if (tmp->cntlid == ctrl->cntlid) {
2632 dev_err(ctrl->device,
2633 "Duplicate cntlid %u with %s, rejecting\n",
2634 ctrl->cntlid, dev_name(tmp->device));
2638 if ((id->cmic & (1 << 1)) ||
2639 (ctrl->opts && ctrl->opts->discovery_nqn))
2642 dev_err(ctrl->device,
2643 "Subsystem does not support multiple controllers\n");
2650 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2652 struct nvme_subsystem *subsys, *found;
2655 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2659 subsys->instance = -1;
2660 mutex_init(&subsys->lock);
2661 kref_init(&subsys->ref);
2662 INIT_LIST_HEAD(&subsys->ctrls);
2663 INIT_LIST_HEAD(&subsys->nsheads);
2664 nvme_init_subnqn(subsys, ctrl, id);
2665 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2666 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2667 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2668 subsys->vendor_id = le16_to_cpu(id->vid);
2669 subsys->cmic = id->cmic;
2670 subsys->awupf = le16_to_cpu(id->awupf);
2671 #ifdef CONFIG_NVME_MULTIPATH
2672 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2675 subsys->dev.class = nvme_subsys_class;
2676 subsys->dev.release = nvme_release_subsystem;
2677 subsys->dev.groups = nvme_subsys_attrs_groups;
2678 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2679 device_initialize(&subsys->dev);
2681 mutex_lock(&nvme_subsystems_lock);
2682 found = __nvme_find_get_subsystem(subsys->subnqn);
2684 put_device(&subsys->dev);
2687 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2689 goto out_put_subsystem;
2692 ret = device_add(&subsys->dev);
2694 dev_err(ctrl->device,
2695 "failed to register subsystem device.\n");
2696 put_device(&subsys->dev);
2699 ida_init(&subsys->ns_ida);
2700 list_add_tail(&subsys->entry, &nvme_subsystems);
2703 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2704 dev_name(ctrl->device));
2706 dev_err(ctrl->device,
2707 "failed to create sysfs link from subsystem.\n");
2708 goto out_put_subsystem;
2712 subsys->instance = ctrl->instance;
2713 ctrl->subsys = subsys;
2714 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2715 mutex_unlock(&nvme_subsystems_lock);
2719 nvme_put_subsystem(subsys);
2721 mutex_unlock(&nvme_subsystems_lock);
2725 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2726 void *log, size_t size, u64 offset)
2728 struct nvme_command c = { };
2729 unsigned long dwlen = size / 4 - 1;
2731 c.get_log_page.opcode = nvme_admin_get_log_page;
2732 c.get_log_page.nsid = cpu_to_le32(nsid);
2733 c.get_log_page.lid = log_page;
2734 c.get_log_page.lsp = lsp;
2735 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2736 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2737 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2738 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2740 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2743 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2748 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2753 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2754 ctrl->effects, sizeof(*ctrl->effects), 0);
2756 kfree(ctrl->effects);
2757 ctrl->effects = NULL;
2763 * Initialize the cached copies of the Identify data and various controller
2764 * register in our nvme_ctrl structure. This should be called as soon as
2765 * the admin queue is fully up and running.
2767 int nvme_init_identify(struct nvme_ctrl *ctrl)
2769 struct nvme_id_ctrl *id;
2770 int ret, page_shift;
2772 bool prev_apst_enabled;
2774 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2776 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2779 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2780 ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2782 if (ctrl->vs >= NVME_VS(1, 1, 0))
2783 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2785 ret = nvme_identify_ctrl(ctrl, &id);
2787 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2791 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2792 ret = nvme_get_effects_log(ctrl);
2797 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2798 ctrl->cntlid = le16_to_cpu(id->cntlid);
2800 if (!ctrl->identified) {
2803 ret = nvme_init_subsystem(ctrl, id);
2808 * Check for quirks. Quirk can depend on firmware version,
2809 * so, in principle, the set of quirks present can change
2810 * across a reset. As a possible future enhancement, we
2811 * could re-scan for quirks every time we reinitialize
2812 * the device, but we'd have to make sure that the driver
2813 * behaves intelligently if the quirks change.
2815 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2816 if (quirk_matches(id, &core_quirks[i]))
2817 ctrl->quirks |= core_quirks[i].quirks;
2821 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2822 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2823 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2826 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2827 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2828 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2830 ctrl->oacs = le16_to_cpu(id->oacs);
2831 ctrl->oncs = le16_to_cpu(id->oncs);
2832 ctrl->mtfa = le16_to_cpu(id->mtfa);
2833 ctrl->oaes = le32_to_cpu(id->oaes);
2834 ctrl->wctemp = le16_to_cpu(id->wctemp);
2835 ctrl->cctemp = le16_to_cpu(id->cctemp);
2837 atomic_set(&ctrl->abort_limit, id->acl + 1);
2838 ctrl->vwc = id->vwc;
2840 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2842 max_hw_sectors = UINT_MAX;
2843 ctrl->max_hw_sectors =
2844 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2846 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2847 ctrl->sgls = le32_to_cpu(id->sgls);
2848 ctrl->kas = le16_to_cpu(id->kas);
2849 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2850 ctrl->ctratt = le32_to_cpu(id->ctratt);
2854 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2856 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2857 shutdown_timeout, 60);
2859 if (ctrl->shutdown_timeout != shutdown_timeout)
2860 dev_info(ctrl->device,
2861 "Shutdown timeout set to %u seconds\n",
2862 ctrl->shutdown_timeout);
2864 ctrl->shutdown_timeout = shutdown_timeout;
2866 ctrl->npss = id->npss;
2867 ctrl->apsta = id->apsta;
2868 prev_apst_enabled = ctrl->apst_enabled;
2869 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2870 if (force_apst && id->apsta) {
2871 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2872 ctrl->apst_enabled = true;
2874 ctrl->apst_enabled = false;
2877 ctrl->apst_enabled = id->apsta;
2879 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2881 if (ctrl->ops->flags & NVME_F_FABRICS) {
2882 ctrl->icdoff = le16_to_cpu(id->icdoff);
2883 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2884 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2885 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2888 * In fabrics we need to verify the cntlid matches the
2891 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2892 dev_err(ctrl->device,
2893 "Mismatching cntlid: Connect %u vs Identify "
2895 ctrl->cntlid, le16_to_cpu(id->cntlid));
2900 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2901 dev_err(ctrl->device,
2902 "keep-alive support is mandatory for fabrics\n");
2907 ctrl->hmpre = le32_to_cpu(id->hmpre);
2908 ctrl->hmmin = le32_to_cpu(id->hmmin);
2909 ctrl->hmminds = le32_to_cpu(id->hmminds);
2910 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2913 ret = nvme_mpath_init(ctrl, id);
2919 if (ctrl->apst_enabled && !prev_apst_enabled)
2920 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2921 else if (!ctrl->apst_enabled && prev_apst_enabled)
2922 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2924 ret = nvme_configure_apst(ctrl);
2928 ret = nvme_configure_timestamp(ctrl);
2932 ret = nvme_configure_directives(ctrl);
2936 ret = nvme_configure_acre(ctrl);
2940 if (!ctrl->identified)
2941 nvme_hwmon_init(ctrl);
2943 ctrl->identified = true;
2951 EXPORT_SYMBOL_GPL(nvme_init_identify);
2953 static int nvme_dev_open(struct inode *inode, struct file *file)
2955 struct nvme_ctrl *ctrl =
2956 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2958 switch (ctrl->state) {
2959 case NVME_CTRL_LIVE:
2962 return -EWOULDBLOCK;
2965 file->private_data = ctrl;
2969 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2974 down_read(&ctrl->namespaces_rwsem);
2975 if (list_empty(&ctrl->namespaces)) {
2980 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2981 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2982 dev_warn(ctrl->device,
2983 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2988 dev_warn(ctrl->device,
2989 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2990 kref_get(&ns->kref);
2991 up_read(&ctrl->namespaces_rwsem);
2993 ret = nvme_user_cmd(ctrl, ns, argp);
2998 up_read(&ctrl->namespaces_rwsem);
3002 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3005 struct nvme_ctrl *ctrl = file->private_data;
3006 void __user *argp = (void __user *)arg;
3009 case NVME_IOCTL_ADMIN_CMD:
3010 return nvme_user_cmd(ctrl, NULL, argp);
3011 case NVME_IOCTL_ADMIN64_CMD:
3012 return nvme_user_cmd64(ctrl, NULL, argp);
3013 case NVME_IOCTL_IO_CMD:
3014 return nvme_dev_user_cmd(ctrl, argp);
3015 case NVME_IOCTL_RESET:
3016 dev_warn(ctrl->device, "resetting controller\n");
3017 return nvme_reset_ctrl_sync(ctrl);
3018 case NVME_IOCTL_SUBSYS_RESET:
3019 return nvme_reset_subsystem(ctrl);
3020 case NVME_IOCTL_RESCAN:
3021 nvme_queue_scan(ctrl);
3028 static const struct file_operations nvme_dev_fops = {
3029 .owner = THIS_MODULE,
3030 .open = nvme_dev_open,
3031 .unlocked_ioctl = nvme_dev_ioctl,
3032 .compat_ioctl = compat_ptr_ioctl,
3035 static ssize_t nvme_sysfs_reset(struct device *dev,
3036 struct device_attribute *attr, const char *buf,
3039 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3042 ret = nvme_reset_ctrl_sync(ctrl);
3047 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3049 static ssize_t nvme_sysfs_rescan(struct device *dev,
3050 struct device_attribute *attr, const char *buf,
3053 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3055 nvme_queue_scan(ctrl);
3058 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3060 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3062 struct gendisk *disk = dev_to_disk(dev);
3064 if (disk->fops == &nvme_fops)
3065 return nvme_get_ns_from_dev(dev)->head;
3067 return disk->private_data;
3070 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3073 struct nvme_ns_head *head = dev_to_ns_head(dev);
3074 struct nvme_ns_ids *ids = &head->ids;
3075 struct nvme_subsystem *subsys = head->subsys;
3076 int serial_len = sizeof(subsys->serial);
3077 int model_len = sizeof(subsys->model);
3079 if (!uuid_is_null(&ids->uuid))
3080 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3082 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3083 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3085 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3086 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3088 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3089 subsys->serial[serial_len - 1] == '\0'))
3091 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3092 subsys->model[model_len - 1] == '\0'))
3095 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3096 serial_len, subsys->serial, model_len, subsys->model,
3099 static DEVICE_ATTR_RO(wwid);
3101 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3104 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3106 static DEVICE_ATTR_RO(nguid);
3108 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3111 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3113 /* For backward compatibility expose the NGUID to userspace if
3114 * we have no UUID set
3116 if (uuid_is_null(&ids->uuid)) {
3117 printk_ratelimited(KERN_WARNING
3118 "No UUID available providing old NGUID\n");
3119 return sprintf(buf, "%pU\n", ids->nguid);
3121 return sprintf(buf, "%pU\n", &ids->uuid);
3123 static DEVICE_ATTR_RO(uuid);
3125 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3128 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3130 static DEVICE_ATTR_RO(eui);
3132 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3135 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3137 static DEVICE_ATTR_RO(nsid);
3139 static struct attribute *nvme_ns_id_attrs[] = {
3140 &dev_attr_wwid.attr,
3141 &dev_attr_uuid.attr,
3142 &dev_attr_nguid.attr,
3144 &dev_attr_nsid.attr,
3145 #ifdef CONFIG_NVME_MULTIPATH
3146 &dev_attr_ana_grpid.attr,
3147 &dev_attr_ana_state.attr,
3152 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3153 struct attribute *a, int n)
3155 struct device *dev = container_of(kobj, struct device, kobj);
3156 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3158 if (a == &dev_attr_uuid.attr) {
3159 if (uuid_is_null(&ids->uuid) &&
3160 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3163 if (a == &dev_attr_nguid.attr) {
3164 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3167 if (a == &dev_attr_eui.attr) {
3168 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3171 #ifdef CONFIG_NVME_MULTIPATH
3172 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3173 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3175 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3182 static const struct attribute_group nvme_ns_id_attr_group = {
3183 .attrs = nvme_ns_id_attrs,
3184 .is_visible = nvme_ns_id_attrs_are_visible,
3187 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3188 &nvme_ns_id_attr_group,
3190 &nvme_nvm_attr_group,
3195 #define nvme_show_str_function(field) \
3196 static ssize_t field##_show(struct device *dev, \
3197 struct device_attribute *attr, char *buf) \
3199 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3200 return sprintf(buf, "%.*s\n", \
3201 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3203 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3205 nvme_show_str_function(model);
3206 nvme_show_str_function(serial);
3207 nvme_show_str_function(firmware_rev);
3209 #define nvme_show_int_function(field) \
3210 static ssize_t field##_show(struct device *dev, \
3211 struct device_attribute *attr, char *buf) \
3213 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3214 return sprintf(buf, "%d\n", ctrl->field); \
3216 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3218 nvme_show_int_function(cntlid);
3219 nvme_show_int_function(numa_node);
3220 nvme_show_int_function(queue_count);
3221 nvme_show_int_function(sqsize);
3223 static ssize_t nvme_sysfs_delete(struct device *dev,
3224 struct device_attribute *attr, const char *buf,
3227 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3229 /* Can't delete non-created controllers */
3233 if (device_remove_file_self(dev, attr))
3234 nvme_delete_ctrl_sync(ctrl);
3237 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3239 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3240 struct device_attribute *attr,
3243 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3245 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3247 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3249 static ssize_t nvme_sysfs_show_state(struct device *dev,
3250 struct device_attribute *attr,
3253 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3254 static const char *const state_name[] = {
3255 [NVME_CTRL_NEW] = "new",
3256 [NVME_CTRL_LIVE] = "live",
3257 [NVME_CTRL_RESETTING] = "resetting",
3258 [NVME_CTRL_CONNECTING] = "connecting",
3259 [NVME_CTRL_DELETING] = "deleting",
3260 [NVME_CTRL_DEAD] = "dead",
3263 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3264 state_name[ctrl->state])
3265 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3267 return sprintf(buf, "unknown state\n");
3270 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3272 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3273 struct device_attribute *attr,
3276 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3278 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3280 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3282 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3283 struct device_attribute *attr,
3286 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3288 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3290 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3292 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3293 struct device_attribute *attr,
3296 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3298 return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3300 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3302 static ssize_t nvme_sysfs_show_address(struct device *dev,
3303 struct device_attribute *attr,
3306 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3308 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3310 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3312 static struct attribute *nvme_dev_attrs[] = {
3313 &dev_attr_reset_controller.attr,
3314 &dev_attr_rescan_controller.attr,
3315 &dev_attr_model.attr,
3316 &dev_attr_serial.attr,
3317 &dev_attr_firmware_rev.attr,
3318 &dev_attr_cntlid.attr,
3319 &dev_attr_delete_controller.attr,
3320 &dev_attr_transport.attr,
3321 &dev_attr_subsysnqn.attr,
3322 &dev_attr_address.attr,
3323 &dev_attr_state.attr,
3324 &dev_attr_numa_node.attr,
3325 &dev_attr_queue_count.attr,
3326 &dev_attr_sqsize.attr,
3327 &dev_attr_hostnqn.attr,
3328 &dev_attr_hostid.attr,
3332 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3333 struct attribute *a, int n)
3335 struct device *dev = container_of(kobj, struct device, kobj);
3336 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3338 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3340 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3342 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3344 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3350 static struct attribute_group nvme_dev_attrs_group = {
3351 .attrs = nvme_dev_attrs,
3352 .is_visible = nvme_dev_attrs_are_visible,
3355 static const struct attribute_group *nvme_dev_attr_groups[] = {
3356 &nvme_dev_attrs_group,
3360 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3363 struct nvme_ns_head *h;
3365 lockdep_assert_held(&subsys->lock);
3367 list_for_each_entry(h, &subsys->nsheads, entry) {
3368 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3375 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3376 struct nvme_ns_head *new)
3378 struct nvme_ns_head *h;
3380 lockdep_assert_held(&subsys->lock);
3382 list_for_each_entry(h, &subsys->nsheads, entry) {
3383 if (nvme_ns_ids_valid(&new->ids) &&
3384 !list_empty(&h->list) &&
3385 nvme_ns_ids_equal(&new->ids, &h->ids))
3392 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3393 unsigned nsid, struct nvme_id_ns *id,
3394 struct nvme_ns_ids *ids)
3396 struct nvme_ns_head *head;
3397 size_t size = sizeof(*head);
3400 #ifdef CONFIG_NVME_MULTIPATH
3401 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3404 head = kzalloc(size, GFP_KERNEL);
3407 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3410 head->instance = ret;
3411 INIT_LIST_HEAD(&head->list);
3412 ret = init_srcu_struct(&head->srcu);
3414 goto out_ida_remove;
3415 head->subsys = ctrl->subsys;
3418 kref_init(&head->ref);
3420 ret = __nvme_check_ids(ctrl->subsys, head);
3422 dev_err(ctrl->device,
3423 "duplicate IDs for nsid %d\n", nsid);
3424 goto out_cleanup_srcu;
3427 ret = nvme_mpath_alloc_disk(ctrl, head);
3429 goto out_cleanup_srcu;
3431 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3433 kref_get(&ctrl->subsys->ref);
3437 cleanup_srcu_struct(&head->srcu);
3439 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3444 ret = blk_status_to_errno(nvme_error_status(ret));
3445 return ERR_PTR(ret);
3448 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3449 struct nvme_id_ns *id)
3451 struct nvme_ctrl *ctrl = ns->ctrl;
3452 bool is_shared = id->nmic & (1 << 0);
3453 struct nvme_ns_head *head = NULL;
3454 struct nvme_ns_ids ids;
3457 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3461 mutex_lock(&ctrl->subsys->lock);
3463 head = nvme_find_ns_head(ctrl->subsys, nsid);
3465 head = nvme_alloc_ns_head(ctrl, nsid, id, &ids);
3467 ret = PTR_ERR(head);
3471 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3472 dev_err(ctrl->device,
3473 "IDs don't match for shared namespace %d\n",
3480 list_add_tail(&ns->siblings, &head->list);
3484 mutex_unlock(&ctrl->subsys->lock);
3487 ret = blk_status_to_errno(nvme_error_status(ret));
3491 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3493 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3494 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3496 return nsa->head->ns_id - nsb->head->ns_id;
3499 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3501 struct nvme_ns *ns, *ret = NULL;
3503 down_read(&ctrl->namespaces_rwsem);
3504 list_for_each_entry(ns, &ctrl->namespaces, list) {
3505 if (ns->head->ns_id == nsid) {
3506 if (!kref_get_unless_zero(&ns->kref))
3511 if (ns->head->ns_id > nsid)
3514 up_read(&ctrl->namespaces_rwsem);
3518 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3520 struct streams_directive_params s;
3523 if (!ctrl->nr_streams)
3526 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3530 ns->sws = le32_to_cpu(s.sws);
3531 ns->sgs = le16_to_cpu(s.sgs);
3534 unsigned int bs = 1 << ns->lba_shift;
3536 blk_queue_io_min(ns->queue, bs * ns->sws);
3538 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3544 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3547 struct gendisk *disk;
3548 struct nvme_id_ns *id;
3549 char disk_name[DISK_NAME_LEN];
3550 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3552 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3556 ns->queue = blk_mq_init_queue(ctrl->tagset);
3557 if (IS_ERR(ns->queue))
3560 if (ctrl->opts && ctrl->opts->data_digest)
3561 ns->queue->backing_dev_info->capabilities
3562 |= BDI_CAP_STABLE_WRITES;
3564 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3565 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3566 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3568 ns->queue->queuedata = ns;
3571 kref_init(&ns->kref);
3572 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3574 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3575 nvme_set_queue_limits(ctrl, ns->queue);
3577 ret = nvme_identify_ns(ctrl, nsid, &id);
3579 goto out_free_queue;
3581 if (id->ncap == 0) /* no namespace (legacy quirk) */
3584 ret = nvme_init_ns_head(ns, nsid, id);
3587 nvme_setup_streams_ns(ctrl, ns);
3588 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3590 disk = alloc_disk_node(0, node);
3594 disk->fops = &nvme_fops;
3595 disk->private_data = ns;
3596 disk->queue = ns->queue;
3597 disk->flags = flags;
3598 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3601 __nvme_revalidate_disk(disk, id);
3603 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3604 ret = nvme_nvm_register(ns, disk_name, node);
3606 dev_warn(ctrl->device, "LightNVM init failure\n");
3611 down_write(&ctrl->namespaces_rwsem);
3612 list_add_tail(&ns->list, &ctrl->namespaces);
3613 up_write(&ctrl->namespaces_rwsem);
3615 nvme_get_ctrl(ctrl);
3617 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3619 nvme_mpath_add_disk(ns, id);
3620 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3627 mutex_lock(&ctrl->subsys->lock);
3628 list_del_rcu(&ns->siblings);
3629 mutex_unlock(&ctrl->subsys->lock);
3630 nvme_put_ns_head(ns->head);
3634 blk_cleanup_queue(ns->queue);
3639 static void nvme_ns_remove(struct nvme_ns *ns)
3641 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3644 nvme_fault_inject_fini(&ns->fault_inject);
3646 mutex_lock(&ns->ctrl->subsys->lock);
3647 list_del_rcu(&ns->siblings);
3648 mutex_unlock(&ns->ctrl->subsys->lock);
3649 synchronize_rcu(); /* guarantee not available in head->list */
3650 nvme_mpath_clear_current_path(ns);
3651 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3653 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3654 del_gendisk(ns->disk);
3655 blk_cleanup_queue(ns->queue);
3656 if (blk_get_integrity(ns->disk))
3657 blk_integrity_unregister(ns->disk);
3660 down_write(&ns->ctrl->namespaces_rwsem);
3661 list_del_init(&ns->list);
3662 up_write(&ns->ctrl->namespaces_rwsem);
3664 nvme_mpath_check_last_path(ns);
3668 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3672 ns = nvme_find_get_ns(ctrl, nsid);
3674 if (ns->disk && revalidate_disk(ns->disk))
3678 nvme_alloc_ns(ctrl, nsid);
3681 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3684 struct nvme_ns *ns, *next;
3687 down_write(&ctrl->namespaces_rwsem);
3688 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3689 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3690 list_move_tail(&ns->list, &rm_list);
3692 up_write(&ctrl->namespaces_rwsem);
3694 list_for_each_entry_safe(ns, next, &rm_list, list)
3699 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3703 unsigned i, j, nsid, prev = 0;
3704 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3707 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3711 for (i = 0; i < num_lists; i++) {
3712 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3716 for (j = 0; j < min(nn, 1024U); j++) {
3717 nsid = le32_to_cpu(ns_list[j]);
3721 nvme_validate_ns(ctrl, nsid);
3723 while (++prev < nsid) {
3724 ns = nvme_find_get_ns(ctrl, prev);
3734 nvme_remove_invalid_namespaces(ctrl, prev);
3740 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3744 for (i = 1; i <= nn; i++)
3745 nvme_validate_ns(ctrl, i);
3747 nvme_remove_invalid_namespaces(ctrl, nn);
3750 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3752 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3756 log = kzalloc(log_size, GFP_KERNEL);
3761 * We need to read the log to clear the AEN, but we don't want to rely
3762 * on it for the changed namespace information as userspace could have
3763 * raced with us in reading the log page, which could cause us to miss
3766 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3769 dev_warn(ctrl->device,
3770 "reading changed ns log failed: %d\n", error);
3775 static void nvme_scan_work(struct work_struct *work)
3777 struct nvme_ctrl *ctrl =
3778 container_of(work, struct nvme_ctrl, scan_work);
3779 struct nvme_id_ctrl *id;
3782 /* No tagset on a live ctrl means IO queues could not created */
3783 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3786 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3787 dev_info(ctrl->device, "rescanning namespaces.\n");
3788 nvme_clear_changed_ns_log(ctrl);
3791 if (nvme_identify_ctrl(ctrl, &id))
3794 mutex_lock(&ctrl->scan_lock);
3795 nn = le32_to_cpu(id->nn);
3796 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3797 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3798 if (!nvme_scan_ns_list(ctrl, nn))
3801 nvme_scan_ns_sequential(ctrl, nn);
3803 mutex_unlock(&ctrl->scan_lock);
3805 down_write(&ctrl->namespaces_rwsem);
3806 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3807 up_write(&ctrl->namespaces_rwsem);
3811 * This function iterates the namespace list unlocked to allow recovery from
3812 * controller failure. It is up to the caller to ensure the namespace list is
3813 * not modified by scan work while this function is executing.
3815 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3817 struct nvme_ns *ns, *next;
3821 * make sure to requeue I/O to all namespaces as these
3822 * might result from the scan itself and must complete
3823 * for the scan_work to make progress
3825 nvme_mpath_clear_ctrl_paths(ctrl);
3827 /* prevent racing with ns scanning */
3828 flush_work(&ctrl->scan_work);
3831 * The dead states indicates the controller was not gracefully
3832 * disconnected. In that case, we won't be able to flush any data while
3833 * removing the namespaces' disks; fail all the queues now to avoid
3834 * potentially having to clean up the failed sync later.
3836 if (ctrl->state == NVME_CTRL_DEAD)
3837 nvme_kill_queues(ctrl);
3839 down_write(&ctrl->namespaces_rwsem);
3840 list_splice_init(&ctrl->namespaces, &ns_list);
3841 up_write(&ctrl->namespaces_rwsem);
3843 list_for_each_entry_safe(ns, next, &ns_list, list)
3846 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3848 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3850 struct nvme_ctrl *ctrl =
3851 container_of(dev, struct nvme_ctrl, ctrl_device);
3852 struct nvmf_ctrl_options *opts = ctrl->opts;
3855 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3860 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3864 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3865 opts->trsvcid ?: "none");
3869 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3870 opts->host_traddr ?: "none");
3875 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3877 char *envp[2] = { NULL, NULL };
3878 u32 aen_result = ctrl->aen_result;
3880 ctrl->aen_result = 0;
3884 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3887 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3891 static void nvme_async_event_work(struct work_struct *work)
3893 struct nvme_ctrl *ctrl =
3894 container_of(work, struct nvme_ctrl, async_event_work);
3896 nvme_aen_uevent(ctrl);
3897 ctrl->ops->submit_async_event(ctrl);
3900 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3905 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3911 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3914 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3916 struct nvme_fw_slot_info_log *log;
3918 log = kmalloc(sizeof(*log), GFP_KERNEL);
3922 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, log,
3924 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3928 static void nvme_fw_act_work(struct work_struct *work)
3930 struct nvme_ctrl *ctrl = container_of(work,
3931 struct nvme_ctrl, fw_act_work);
3932 unsigned long fw_act_timeout;
3935 fw_act_timeout = jiffies +
3936 msecs_to_jiffies(ctrl->mtfa * 100);
3938 fw_act_timeout = jiffies +
3939 msecs_to_jiffies(admin_timeout * 1000);
3941 nvme_stop_queues(ctrl);
3942 while (nvme_ctrl_pp_status(ctrl)) {
3943 if (time_after(jiffies, fw_act_timeout)) {
3944 dev_warn(ctrl->device,
3945 "Fw activation timeout, reset controller\n");
3946 nvme_try_sched_reset(ctrl);
3952 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
3955 nvme_start_queues(ctrl);
3956 /* read FW slot information to clear the AER */
3957 nvme_get_fw_slot_info(ctrl);
3960 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3962 u32 aer_notice_type = (result & 0xff00) >> 8;
3964 trace_nvme_async_event(ctrl, aer_notice_type);
3966 switch (aer_notice_type) {
3967 case NVME_AER_NOTICE_NS_CHANGED:
3968 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3969 nvme_queue_scan(ctrl);
3971 case NVME_AER_NOTICE_FW_ACT_STARTING:
3973 * We are (ab)using the RESETTING state to prevent subsequent
3974 * recovery actions from interfering with the controller's
3975 * firmware activation.
3977 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
3978 queue_work(nvme_wq, &ctrl->fw_act_work);
3980 #ifdef CONFIG_NVME_MULTIPATH
3981 case NVME_AER_NOTICE_ANA:
3982 if (!ctrl->ana_log_buf)
3984 queue_work(nvme_wq, &ctrl->ana_work);
3987 case NVME_AER_NOTICE_DISC_CHANGED:
3988 ctrl->aen_result = result;
3991 dev_warn(ctrl->device, "async event result %08x\n", result);
3995 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3996 volatile union nvme_result *res)
3998 u32 result = le32_to_cpu(res->u32);
3999 u32 aer_type = result & 0x07;
4001 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4005 case NVME_AER_NOTICE:
4006 nvme_handle_aen_notice(ctrl, result);
4008 case NVME_AER_ERROR:
4009 case NVME_AER_SMART:
4012 trace_nvme_async_event(ctrl, aer_type);
4013 ctrl->aen_result = result;
4018 queue_work(nvme_wq, &ctrl->async_event_work);
4020 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4022 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4024 nvme_mpath_stop(ctrl);
4025 nvme_stop_keep_alive(ctrl);
4026 flush_work(&ctrl->async_event_work);
4027 cancel_work_sync(&ctrl->fw_act_work);
4029 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4031 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4034 nvme_start_keep_alive(ctrl);
4036 nvme_enable_aen(ctrl);
4038 if (ctrl->queue_count > 1) {
4039 nvme_queue_scan(ctrl);
4040 nvme_start_queues(ctrl);
4042 ctrl->created = true;
4044 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4046 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4048 nvme_fault_inject_fini(&ctrl->fault_inject);
4049 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4050 cdev_device_del(&ctrl->cdev, ctrl->device);
4051 nvme_put_ctrl(ctrl);
4053 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4055 static void nvme_free_ctrl(struct device *dev)
4057 struct nvme_ctrl *ctrl =
4058 container_of(dev, struct nvme_ctrl, ctrl_device);
4059 struct nvme_subsystem *subsys = ctrl->subsys;
4061 if (subsys && ctrl->instance != subsys->instance)
4062 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4064 kfree(ctrl->effects);
4065 nvme_mpath_uninit(ctrl);
4066 __free_page(ctrl->discard_page);
4069 mutex_lock(&nvme_subsystems_lock);
4070 list_del(&ctrl->subsys_entry);
4071 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4072 mutex_unlock(&nvme_subsystems_lock);
4075 ctrl->ops->free_ctrl(ctrl);
4078 nvme_put_subsystem(subsys);
4082 * Initialize a NVMe controller structures. This needs to be called during
4083 * earliest initialization so that we have the initialized structured around
4086 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4087 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4091 ctrl->state = NVME_CTRL_NEW;
4092 spin_lock_init(&ctrl->lock);
4093 mutex_init(&ctrl->scan_lock);
4094 INIT_LIST_HEAD(&ctrl->namespaces);
4095 init_rwsem(&ctrl->namespaces_rwsem);
4098 ctrl->quirks = quirks;
4099 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4100 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4101 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4102 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4103 init_waitqueue_head(&ctrl->state_wq);
4105 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4106 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4107 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4109 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4111 ctrl->discard_page = alloc_page(GFP_KERNEL);
4112 if (!ctrl->discard_page) {
4117 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4120 ctrl->instance = ret;
4122 device_initialize(&ctrl->ctrl_device);
4123 ctrl->device = &ctrl->ctrl_device;
4124 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4125 ctrl->device->class = nvme_class;
4126 ctrl->device->parent = ctrl->dev;
4127 ctrl->device->groups = nvme_dev_attr_groups;
4128 ctrl->device->release = nvme_free_ctrl;
4129 dev_set_drvdata(ctrl->device, ctrl);
4130 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4132 goto out_release_instance;
4134 nvme_get_ctrl(ctrl);
4135 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4136 ctrl->cdev.owner = ops->module;
4137 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4142 * Initialize latency tolerance controls. The sysfs files won't
4143 * be visible to userspace unless the device actually supports APST.
4145 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4146 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4147 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4149 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4153 nvme_put_ctrl(ctrl);
4154 kfree_const(ctrl->device->kobj.name);
4155 out_release_instance:
4156 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4158 if (ctrl->discard_page)
4159 __free_page(ctrl->discard_page);
4162 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4165 * nvme_kill_queues(): Ends all namespace queues
4166 * @ctrl: the dead controller that needs to end
4168 * Call this function when the driver determines it is unable to get the
4169 * controller in a state capable of servicing IO.
4171 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4175 down_read(&ctrl->namespaces_rwsem);
4177 /* Forcibly unquiesce queues to avoid blocking dispatch */
4178 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4179 blk_mq_unquiesce_queue(ctrl->admin_q);
4181 list_for_each_entry(ns, &ctrl->namespaces, list)
4182 nvme_set_queue_dying(ns);
4184 up_read(&ctrl->namespaces_rwsem);
4186 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4188 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4192 down_read(&ctrl->namespaces_rwsem);
4193 list_for_each_entry(ns, &ctrl->namespaces, list)
4194 blk_mq_unfreeze_queue(ns->queue);
4195 up_read(&ctrl->namespaces_rwsem);
4197 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4199 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4203 down_read(&ctrl->namespaces_rwsem);
4204 list_for_each_entry(ns, &ctrl->namespaces, list) {
4205 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4209 up_read(&ctrl->namespaces_rwsem);
4211 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4213 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4217 down_read(&ctrl->namespaces_rwsem);
4218 list_for_each_entry(ns, &ctrl->namespaces, list)
4219 blk_mq_freeze_queue_wait(ns->queue);
4220 up_read(&ctrl->namespaces_rwsem);
4222 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4224 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4228 down_read(&ctrl->namespaces_rwsem);
4229 list_for_each_entry(ns, &ctrl->namespaces, list)
4230 blk_freeze_queue_start(ns->queue);
4231 up_read(&ctrl->namespaces_rwsem);
4233 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4235 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4239 down_read(&ctrl->namespaces_rwsem);
4240 list_for_each_entry(ns, &ctrl->namespaces, list)
4241 blk_mq_quiesce_queue(ns->queue);
4242 up_read(&ctrl->namespaces_rwsem);
4244 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4246 void nvme_start_queues(struct nvme_ctrl *ctrl)
4250 down_read(&ctrl->namespaces_rwsem);
4251 list_for_each_entry(ns, &ctrl->namespaces, list)
4252 blk_mq_unquiesce_queue(ns->queue);
4253 up_read(&ctrl->namespaces_rwsem);
4255 EXPORT_SYMBOL_GPL(nvme_start_queues);
4258 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4262 down_read(&ctrl->namespaces_rwsem);
4263 list_for_each_entry(ns, &ctrl->namespaces, list)
4264 blk_sync_queue(ns->queue);
4265 up_read(&ctrl->namespaces_rwsem);
4268 blk_sync_queue(ctrl->admin_q);
4270 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4273 * Check we didn't inadvertently grow the command structure sizes:
4275 static inline void _nvme_check_size(void)
4277 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4278 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4279 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4280 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4281 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4282 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4283 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4284 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4285 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4286 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4287 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4288 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4289 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4290 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4291 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4292 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4293 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4297 static int __init nvme_core_init(void)
4299 int result = -ENOMEM;
4303 nvme_wq = alloc_workqueue("nvme-wq",
4304 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4308 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4309 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4313 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4314 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4315 if (!nvme_delete_wq)
4316 goto destroy_reset_wq;
4318 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4320 goto destroy_delete_wq;
4322 nvme_class = class_create(THIS_MODULE, "nvme");
4323 if (IS_ERR(nvme_class)) {
4324 result = PTR_ERR(nvme_class);
4325 goto unregister_chrdev;
4327 nvme_class->dev_uevent = nvme_class_uevent;
4329 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4330 if (IS_ERR(nvme_subsys_class)) {
4331 result = PTR_ERR(nvme_subsys_class);
4337 class_destroy(nvme_class);
4339 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4341 destroy_workqueue(nvme_delete_wq);
4343 destroy_workqueue(nvme_reset_wq);
4345 destroy_workqueue(nvme_wq);
4350 static void __exit nvme_core_exit(void)
4352 class_destroy(nvme_subsys_class);
4353 class_destroy(nvme_class);
4354 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4355 destroy_workqueue(nvme_delete_wq);
4356 destroy_workqueue(nvme_reset_wq);
4357 destroy_workqueue(nvme_wq);
4358 ida_destroy(&nvme_instance_ida);
4361 MODULE_LICENSE("GPL");
4362 MODULE_VERSION("1.0");
4363 module_init(nvme_core_init);
4364 module_exit(nvme_core_exit);