1 // SPDX-License-Identifier: GPL-2.0
3 * NVM Express device driver
4 * Copyright (c) 2011-2014, Intel Corporation.
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/delay.h>
10 #include <linux/errno.h>
11 #include <linux/hdreg.h>
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/list_sort.h>
16 #include <linux/slab.h>
17 #include <linux/types.h>
19 #include <linux/ptrace.h>
20 #include <linux/nvme_ioctl.h>
21 #include <linux/t10-pi.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
28 #define CREATE_TRACE_POINTS
31 #define NVME_MINORS (1U << MINORBITS)
33 unsigned int admin_timeout = 60;
34 module_param(admin_timeout, uint, 0644);
35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
36 EXPORT_SYMBOL_GPL(admin_timeout);
38 unsigned int nvme_io_timeout = 30;
39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
43 static unsigned char shutdown_timeout = 5;
44 module_param(shutdown_timeout, byte, 0644);
45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
47 static u8 nvme_max_retries = 5;
48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
51 static unsigned long default_ps_max_latency_us = 100000;
52 module_param(default_ps_max_latency_us, ulong, 0644);
53 MODULE_PARM_DESC(default_ps_max_latency_us,
54 "max power saving latency for new devices; use PM QOS to change per device");
56 static bool force_apst;
57 module_param(force_apst, bool, 0644);
58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
61 module_param(streams, bool, 0644);
62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
65 * nvme_wq - hosts nvme related works that are not reset or delete
66 * nvme_reset_wq - hosts nvme reset works
67 * nvme_delete_wq - hosts nvme delete works
69 * nvme_wq will host works such are scan, aen handling, fw activation,
70 * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
71 * runs reset works which also flush works hosted on nvme_wq for
72 * serialization purposes. nvme_delete_wq host controller deletion
73 * works which flush reset works for serialization.
75 struct workqueue_struct *nvme_wq;
76 EXPORT_SYMBOL_GPL(nvme_wq);
78 struct workqueue_struct *nvme_reset_wq;
79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
81 struct workqueue_struct *nvme_delete_wq;
82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
84 static LIST_HEAD(nvme_subsystems);
85 static DEFINE_MUTEX(nvme_subsystems_lock);
87 static DEFINE_IDA(nvme_instance_ida);
88 static dev_t nvme_chr_devt;
89 static struct class *nvme_class;
90 static struct class *nvme_subsys_class;
92 static int nvme_revalidate_disk(struct gendisk *disk);
93 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
94 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
97 static void nvme_set_queue_dying(struct nvme_ns *ns)
100 * Revalidating a dead namespace sets capacity to 0. This will end
101 * buffered writers dirtying pages that can't be synced.
103 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
105 blk_set_queue_dying(ns->queue);
106 /* Forcibly unquiesce queues to avoid blocking dispatch */
107 blk_mq_unquiesce_queue(ns->queue);
109 * Revalidate after unblocking dispatchers that may be holding bd_butex
111 revalidate_disk(ns->disk);
114 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
117 * Only new queue scan work when admin and IO queues are both alive
119 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
120 queue_work(nvme_wq, &ctrl->scan_work);
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);
177 static void nvme_delete_ctrl_work(struct work_struct *work)
179 struct nvme_ctrl *ctrl =
180 container_of(work, struct nvme_ctrl, delete_work);
182 nvme_do_delete_ctrl(ctrl);
185 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
187 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
189 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
193 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
195 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
200 * Keep a reference until nvme_do_delete_ctrl() complete,
201 * since ->delete_ctrl can free the controller.
204 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
207 nvme_do_delete_ctrl(ctrl);
212 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
214 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
217 static blk_status_t nvme_error_status(u16 status)
219 switch (status & 0x7ff) {
220 case NVME_SC_SUCCESS:
222 case NVME_SC_CAP_EXCEEDED:
223 return BLK_STS_NOSPC;
224 case NVME_SC_LBA_RANGE:
225 return BLK_STS_TARGET;
226 case NVME_SC_BAD_ATTRIBUTES:
227 case NVME_SC_ONCS_NOT_SUPPORTED:
228 case NVME_SC_INVALID_OPCODE:
229 case NVME_SC_INVALID_FIELD:
230 case NVME_SC_INVALID_NS:
231 return BLK_STS_NOTSUPP;
232 case NVME_SC_WRITE_FAULT:
233 case NVME_SC_READ_ERROR:
234 case NVME_SC_UNWRITTEN_BLOCK:
235 case NVME_SC_ACCESS_DENIED:
236 case NVME_SC_READ_ONLY:
237 case NVME_SC_COMPARE_FAILED:
238 return BLK_STS_MEDIUM;
239 case NVME_SC_GUARD_CHECK:
240 case NVME_SC_APPTAG_CHECK:
241 case NVME_SC_REFTAG_CHECK:
242 case NVME_SC_INVALID_PI:
243 return BLK_STS_PROTECTION;
244 case NVME_SC_RESERVATION_CONFLICT:
245 return BLK_STS_NEXUS;
246 case NVME_SC_HOST_PATH_ERROR:
247 return BLK_STS_TRANSPORT;
249 return BLK_STS_IOERR;
253 static inline bool nvme_req_needs_retry(struct request *req)
255 if (blk_noretry_request(req))
257 if (nvme_req(req)->status & NVME_SC_DNR)
259 if (nvme_req(req)->retries >= nvme_max_retries)
264 static void nvme_retry_req(struct request *req)
266 struct nvme_ns *ns = req->q->queuedata;
267 unsigned long delay = 0;
270 /* The mask and shift result must be <= 3 */
271 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
273 delay = ns->ctrl->crdt[crd - 1] * 100;
275 nvme_req(req)->retries++;
276 blk_mq_requeue_request(req, false);
277 blk_mq_delay_kick_requeue_list(req->q, delay);
280 void nvme_complete_rq(struct request *req)
282 blk_status_t status = nvme_error_status(nvme_req(req)->status);
284 trace_nvme_complete_rq(req);
286 if (nvme_req(req)->ctrl->kas)
287 nvme_req(req)->ctrl->comp_seen = true;
289 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
290 if ((req->cmd_flags & REQ_NVME_MPATH) &&
291 blk_path_error(status)) {
292 nvme_failover_req(req);
296 if (!blk_queue_dying(req->q)) {
302 nvme_trace_bio_complete(req, status);
303 blk_mq_end_request(req, status);
305 EXPORT_SYMBOL_GPL(nvme_complete_rq);
307 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
309 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
310 "Cancelling I/O %d", req->tag);
312 /* don't abort one completed request */
313 if (blk_mq_request_completed(req))
316 nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
317 blk_mq_complete_request(req);
320 EXPORT_SYMBOL_GPL(nvme_cancel_request);
322 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
323 enum nvme_ctrl_state new_state)
325 enum nvme_ctrl_state old_state;
327 bool changed = false;
329 spin_lock_irqsave(&ctrl->lock, flags);
331 old_state = ctrl->state;
336 case NVME_CTRL_RESETTING:
337 case NVME_CTRL_CONNECTING:
344 case NVME_CTRL_RESETTING:
354 case NVME_CTRL_CONNECTING:
357 case NVME_CTRL_RESETTING:
364 case NVME_CTRL_DELETING:
367 case NVME_CTRL_RESETTING:
368 case NVME_CTRL_CONNECTING:
377 case NVME_CTRL_DELETING:
389 ctrl->state = new_state;
390 wake_up_all(&ctrl->state_wq);
393 spin_unlock_irqrestore(&ctrl->lock, flags);
394 if (changed && ctrl->state == NVME_CTRL_LIVE)
395 nvme_kick_requeue_lists(ctrl);
398 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
401 * Returns true for sink states that can't ever transition back to live.
403 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
405 switch (ctrl->state) {
408 case NVME_CTRL_RESETTING:
409 case NVME_CTRL_CONNECTING:
411 case NVME_CTRL_DELETING:
415 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
421 * Waits for the controller state to be resetting, or returns false if it is
422 * not possible to ever transition to that state.
424 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
426 wait_event(ctrl->state_wq,
427 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
428 nvme_state_terminal(ctrl));
429 return ctrl->state == NVME_CTRL_RESETTING;
431 EXPORT_SYMBOL_GPL(nvme_wait_reset);
433 static void nvme_free_ns_head(struct kref *ref)
435 struct nvme_ns_head *head =
436 container_of(ref, struct nvme_ns_head, ref);
438 nvme_mpath_remove_disk(head);
439 ida_simple_remove(&head->subsys->ns_ida, head->instance);
440 list_del_init(&head->entry);
441 cleanup_srcu_struct(&head->srcu);
442 nvme_put_subsystem(head->subsys);
446 static void nvme_put_ns_head(struct nvme_ns_head *head)
448 kref_put(&head->ref, nvme_free_ns_head);
451 static void nvme_free_ns(struct kref *kref)
453 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
456 nvme_nvm_unregister(ns);
459 nvme_put_ns_head(ns->head);
460 nvme_put_ctrl(ns->ctrl);
464 static void nvme_put_ns(struct nvme_ns *ns)
466 kref_put(&ns->kref, nvme_free_ns);
469 static inline void nvme_clear_nvme_request(struct request *req)
471 if (!(req->rq_flags & RQF_DONTPREP)) {
472 nvme_req(req)->retries = 0;
473 nvme_req(req)->flags = 0;
474 req->rq_flags |= RQF_DONTPREP;
478 struct request *nvme_alloc_request(struct request_queue *q,
479 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
481 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
484 if (qid == NVME_QID_ANY) {
485 req = blk_mq_alloc_request(q, op, flags);
487 req = blk_mq_alloc_request_hctx(q, op, flags,
493 req->cmd_flags |= REQ_FAILFAST_DRIVER;
494 nvme_clear_nvme_request(req);
495 nvme_req(req)->cmd = cmd;
499 EXPORT_SYMBOL_GPL(nvme_alloc_request);
501 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
503 struct nvme_command c;
505 memset(&c, 0, sizeof(c));
507 c.directive.opcode = nvme_admin_directive_send;
508 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
509 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
510 c.directive.dtype = NVME_DIR_IDENTIFY;
511 c.directive.tdtype = NVME_DIR_STREAMS;
512 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
514 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
517 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
519 return nvme_toggle_streams(ctrl, false);
522 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
524 return nvme_toggle_streams(ctrl, true);
527 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
528 struct streams_directive_params *s, u32 nsid)
530 struct nvme_command c;
532 memset(&c, 0, sizeof(c));
533 memset(s, 0, sizeof(*s));
535 c.directive.opcode = nvme_admin_directive_recv;
536 c.directive.nsid = cpu_to_le32(nsid);
537 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
538 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
539 c.directive.dtype = NVME_DIR_STREAMS;
541 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
544 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
546 struct streams_directive_params s;
549 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
554 ret = nvme_enable_streams(ctrl);
558 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
562 ctrl->nssa = le16_to_cpu(s.nssa);
563 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
564 dev_info(ctrl->device, "too few streams (%u) available\n",
566 nvme_disable_streams(ctrl);
570 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
571 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
576 * Check if 'req' has a write hint associated with it. If it does, assign
577 * a valid namespace stream to the write.
579 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
580 struct request *req, u16 *control,
583 enum rw_hint streamid = req->write_hint;
585 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
589 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
592 *control |= NVME_RW_DTYPE_STREAMS;
593 *dsmgmt |= streamid << 16;
596 if (streamid < ARRAY_SIZE(req->q->write_hints))
597 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
600 static inline void nvme_setup_flush(struct nvme_ns *ns,
601 struct nvme_command *cmnd)
603 cmnd->common.opcode = nvme_cmd_flush;
604 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
607 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
608 struct nvme_command *cmnd)
610 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
611 struct nvme_dsm_range *range;
614 range = kmalloc_array(segments, sizeof(*range),
615 GFP_ATOMIC | __GFP_NOWARN);
618 * If we fail allocation our range, fallback to the controller
619 * discard page. If that's also busy, it's safe to return
620 * busy, as we know we can make progress once that's freed.
622 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
623 return BLK_STS_RESOURCE;
625 range = page_address(ns->ctrl->discard_page);
628 __rq_for_each_bio(bio, req) {
629 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
630 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
633 range[n].cattr = cpu_to_le32(0);
634 range[n].nlb = cpu_to_le32(nlb);
635 range[n].slba = cpu_to_le64(slba);
640 if (WARN_ON_ONCE(n != segments)) {
641 if (virt_to_page(range) == ns->ctrl->discard_page)
642 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
645 return BLK_STS_IOERR;
648 cmnd->dsm.opcode = nvme_cmd_dsm;
649 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
650 cmnd->dsm.nr = cpu_to_le32(segments - 1);
651 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
653 req->special_vec.bv_page = virt_to_page(range);
654 req->special_vec.bv_offset = offset_in_page(range);
655 req->special_vec.bv_len = sizeof(*range) * segments;
656 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
661 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
662 struct request *req, struct nvme_command *cmnd)
664 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
665 return nvme_setup_discard(ns, req, cmnd);
667 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
668 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
669 cmnd->write_zeroes.slba =
670 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
671 cmnd->write_zeroes.length =
672 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
673 cmnd->write_zeroes.control = 0;
677 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
678 struct request *req, struct nvme_command *cmnd)
680 struct nvme_ctrl *ctrl = ns->ctrl;
684 if (req->cmd_flags & REQ_FUA)
685 control |= NVME_RW_FUA;
686 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
687 control |= NVME_RW_LR;
689 if (req->cmd_flags & REQ_RAHEAD)
690 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
692 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
693 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
694 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
695 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
697 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
698 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
702 * If formated with metadata, the block layer always provides a
703 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
704 * we enable the PRACT bit for protection information or set the
705 * namespace capacity to zero to prevent any I/O.
707 if (!blk_integrity_rq(req)) {
708 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
709 return BLK_STS_NOTSUPP;
710 control |= NVME_RW_PRINFO_PRACT;
713 switch (ns->pi_type) {
714 case NVME_NS_DPS_PI_TYPE3:
715 control |= NVME_RW_PRINFO_PRCHK_GUARD;
717 case NVME_NS_DPS_PI_TYPE1:
718 case NVME_NS_DPS_PI_TYPE2:
719 control |= NVME_RW_PRINFO_PRCHK_GUARD |
720 NVME_RW_PRINFO_PRCHK_REF;
721 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
726 cmnd->rw.control = cpu_to_le16(control);
727 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
731 void nvme_cleanup_cmd(struct request *req)
733 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
734 struct nvme_ns *ns = req->rq_disk->private_data;
735 struct page *page = req->special_vec.bv_page;
737 if (page == ns->ctrl->discard_page)
738 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
740 kfree(page_address(page) + req->special_vec.bv_offset);
743 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
745 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
746 struct nvme_command *cmd)
748 blk_status_t ret = BLK_STS_OK;
750 nvme_clear_nvme_request(req);
752 memset(cmd, 0, sizeof(*cmd));
753 switch (req_op(req)) {
756 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
759 nvme_setup_flush(ns, cmd);
761 case REQ_OP_WRITE_ZEROES:
762 ret = nvme_setup_write_zeroes(ns, req, cmd);
765 ret = nvme_setup_discard(ns, req, cmd);
769 ret = nvme_setup_rw(ns, req, cmd);
773 return BLK_STS_IOERR;
776 cmd->common.command_id = req->tag;
777 trace_nvme_setup_cmd(req, cmd);
780 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
782 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
784 struct completion *waiting = rq->end_io_data;
786 rq->end_io_data = NULL;
790 static void nvme_execute_rq_polled(struct request_queue *q,
791 struct gendisk *bd_disk, struct request *rq, int at_head)
793 DECLARE_COMPLETION_ONSTACK(wait);
795 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
797 rq->cmd_flags |= REQ_HIPRI;
798 rq->end_io_data = &wait;
799 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
801 while (!completion_done(&wait)) {
802 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
808 * Returns 0 on success. If the result is negative, it's a Linux error code;
809 * if the result is positive, it's an NVM Express status code
811 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
812 union nvme_result *result, void *buffer, unsigned bufflen,
813 unsigned timeout, int qid, int at_head,
814 blk_mq_req_flags_t flags, bool poll)
819 req = nvme_alloc_request(q, cmd, flags, qid);
823 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
825 if (buffer && bufflen) {
826 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
832 nvme_execute_rq_polled(req->q, NULL, req, at_head);
834 blk_execute_rq(req->q, NULL, req, at_head);
836 *result = nvme_req(req)->result;
837 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
840 ret = nvme_req(req)->status;
842 blk_mq_free_request(req);
845 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
847 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
848 void *buffer, unsigned bufflen)
850 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
851 NVME_QID_ANY, 0, 0, false);
853 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
855 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
856 unsigned len, u32 seed, bool write)
858 struct bio_integrity_payload *bip;
862 buf = kmalloc(len, GFP_KERNEL);
867 if (write && copy_from_user(buf, ubuf, len))
870 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
876 bip->bip_iter.bi_size = len;
877 bip->bip_iter.bi_sector = seed;
878 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
879 offset_in_page(buf));
889 static int nvme_submit_user_cmd(struct request_queue *q,
890 struct nvme_command *cmd, void __user *ubuffer,
891 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
892 u32 meta_seed, u64 *result, unsigned timeout)
894 bool write = nvme_is_write(cmd);
895 struct nvme_ns *ns = q->queuedata;
896 struct gendisk *disk = ns ? ns->disk : NULL;
898 struct bio *bio = NULL;
902 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
906 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
907 nvme_req(req)->flags |= NVME_REQ_USERCMD;
909 if (ubuffer && bufflen) {
910 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
916 if (disk && meta_buffer && meta_len) {
917 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
923 req->cmd_flags |= REQ_INTEGRITY;
927 blk_execute_rq(req->q, disk, req, 0);
928 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
931 ret = nvme_req(req)->status;
933 *result = le64_to_cpu(nvme_req(req)->result.u64);
934 if (meta && !ret && !write) {
935 if (copy_to_user(meta_buffer, meta, meta_len))
941 blk_rq_unmap_user(bio);
943 blk_mq_free_request(req);
947 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
949 struct nvme_ctrl *ctrl = rq->end_io_data;
951 bool startka = false;
953 blk_mq_free_request(rq);
956 dev_err(ctrl->device,
957 "failed nvme_keep_alive_end_io error=%d\n",
962 ctrl->comp_seen = false;
963 spin_lock_irqsave(&ctrl->lock, flags);
964 if (ctrl->state == NVME_CTRL_LIVE ||
965 ctrl->state == NVME_CTRL_CONNECTING)
967 spin_unlock_irqrestore(&ctrl->lock, flags);
969 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
972 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
976 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
981 rq->timeout = ctrl->kato * HZ;
982 rq->end_io_data = ctrl;
984 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
989 static void nvme_keep_alive_work(struct work_struct *work)
991 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
992 struct nvme_ctrl, ka_work);
993 bool comp_seen = ctrl->comp_seen;
995 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
996 dev_dbg(ctrl->device,
997 "reschedule traffic based keep-alive timer\n");
998 ctrl->comp_seen = false;
999 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
1003 if (nvme_keep_alive(ctrl)) {
1004 /* allocation failure, reset the controller */
1005 dev_err(ctrl->device, "keep-alive failed\n");
1006 nvme_reset_ctrl(ctrl);
1011 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1013 if (unlikely(ctrl->kato == 0))
1016 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
1019 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1021 if (unlikely(ctrl->kato == 0))
1024 cancel_delayed_work_sync(&ctrl->ka_work);
1026 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1028 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1030 struct nvme_command c = { };
1033 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1034 c.identify.opcode = nvme_admin_identify;
1035 c.identify.cns = NVME_ID_CNS_CTRL;
1037 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1041 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1042 sizeof(struct nvme_id_ctrl));
1048 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1049 struct nvme_ns_ids *ids)
1051 struct nvme_command c = { };
1057 c.identify.opcode = nvme_admin_identify;
1058 c.identify.nsid = cpu_to_le32(nsid);
1059 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1061 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1065 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1066 NVME_IDENTIFY_DATA_SIZE);
1070 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1071 struct nvme_ns_id_desc *cur = data + pos;
1076 switch (cur->nidt) {
1077 case NVME_NIDT_EUI64:
1078 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1079 dev_warn(ctrl->device,
1080 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1084 len = NVME_NIDT_EUI64_LEN;
1085 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1087 case NVME_NIDT_NGUID:
1088 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1089 dev_warn(ctrl->device,
1090 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1094 len = NVME_NIDT_NGUID_LEN;
1095 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1097 case NVME_NIDT_UUID:
1098 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1099 dev_warn(ctrl->device,
1100 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1104 len = NVME_NIDT_UUID_LEN;
1105 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1108 /* Skip unknown types */
1113 len += sizeof(*cur);
1120 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1122 struct nvme_command c = { };
1124 c.identify.opcode = nvme_admin_identify;
1125 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1126 c.identify.nsid = cpu_to_le32(nsid);
1127 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1128 NVME_IDENTIFY_DATA_SIZE);
1131 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1132 unsigned nsid, struct nvme_id_ns **id)
1134 struct nvme_command c = { };
1137 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1138 c.identify.opcode = nvme_admin_identify;
1139 c.identify.nsid = cpu_to_le32(nsid);
1140 c.identify.cns = NVME_ID_CNS_NS;
1142 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1146 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1148 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1155 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1156 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1158 struct nvme_command c;
1159 union nvme_result res;
1162 memset(&c, 0, sizeof(c));
1163 c.features.opcode = op;
1164 c.features.fid = cpu_to_le32(fid);
1165 c.features.dword11 = cpu_to_le32(dword11);
1167 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1168 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1169 if (ret >= 0 && result)
1170 *result = le32_to_cpu(res.u32);
1174 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1175 unsigned int dword11, void *buffer, size_t buflen,
1178 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1181 EXPORT_SYMBOL_GPL(nvme_set_features);
1183 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1184 unsigned int dword11, void *buffer, size_t buflen,
1187 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1190 EXPORT_SYMBOL_GPL(nvme_get_features);
1192 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1194 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1196 int status, nr_io_queues;
1198 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1204 * Degraded controllers might return an error when setting the queue
1205 * count. We still want to be able to bring them online and offer
1206 * access to the admin queue, as that might be only way to fix them up.
1209 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1212 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1213 *count = min(*count, nr_io_queues);
1218 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1220 #define NVME_AEN_SUPPORTED \
1221 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1222 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1224 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1226 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1229 if (!supported_aens)
1232 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1235 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1238 queue_work(nvme_wq, &ctrl->async_event_work);
1241 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1243 struct nvme_user_io io;
1244 struct nvme_command c;
1245 unsigned length, meta_len;
1246 void __user *metadata;
1248 if (copy_from_user(&io, uio, sizeof(io)))
1253 switch (io.opcode) {
1254 case nvme_cmd_write:
1256 case nvme_cmd_compare:
1262 length = (io.nblocks + 1) << ns->lba_shift;
1263 meta_len = (io.nblocks + 1) * ns->ms;
1264 metadata = (void __user *)(uintptr_t)io.metadata;
1269 } else if (meta_len) {
1270 if ((io.metadata & 3) || !io.metadata)
1274 memset(&c, 0, sizeof(c));
1275 c.rw.opcode = io.opcode;
1276 c.rw.flags = io.flags;
1277 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1278 c.rw.slba = cpu_to_le64(io.slba);
1279 c.rw.length = cpu_to_le16(io.nblocks);
1280 c.rw.control = cpu_to_le16(io.control);
1281 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1282 c.rw.reftag = cpu_to_le32(io.reftag);
1283 c.rw.apptag = cpu_to_le16(io.apptag);
1284 c.rw.appmask = cpu_to_le16(io.appmask);
1286 return nvme_submit_user_cmd(ns->queue, &c,
1287 (void __user *)(uintptr_t)io.addr, length,
1288 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1291 static u32 nvme_known_admin_effects(u8 opcode)
1294 case nvme_admin_format_nvm:
1295 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1296 NVME_CMD_EFFECTS_CSE_MASK;
1297 case nvme_admin_sanitize_nvm:
1298 return NVME_CMD_EFFECTS_CSE_MASK;
1305 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1312 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1313 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1314 dev_warn(ctrl->device,
1315 "IO command:%02x has unhandled effects:%08x\n",
1321 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1322 effects |= nvme_known_admin_effects(opcode);
1325 * For simplicity, IO to all namespaces is quiesced even if the command
1326 * effects say only one namespace is affected.
1328 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1329 mutex_lock(&ctrl->scan_lock);
1330 mutex_lock(&ctrl->subsys->lock);
1331 nvme_mpath_start_freeze(ctrl->subsys);
1332 nvme_mpath_wait_freeze(ctrl->subsys);
1333 nvme_start_freeze(ctrl);
1334 nvme_wait_freeze(ctrl);
1339 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1343 down_read(&ctrl->namespaces_rwsem);
1344 list_for_each_entry(ns, &ctrl->namespaces, list)
1345 if (ns->disk && nvme_revalidate_disk(ns->disk))
1346 nvme_set_queue_dying(ns);
1347 up_read(&ctrl->namespaces_rwsem);
1350 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1353 * Revalidate LBA changes prior to unfreezing. This is necessary to
1354 * prevent memory corruption if a logical block size was changed by
1357 if (effects & NVME_CMD_EFFECTS_LBCC)
1358 nvme_update_formats(ctrl);
1359 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1360 nvme_unfreeze(ctrl);
1361 nvme_mpath_unfreeze(ctrl->subsys);
1362 mutex_unlock(&ctrl->subsys->lock);
1363 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1364 mutex_unlock(&ctrl->scan_lock);
1366 if (effects & NVME_CMD_EFFECTS_CCC)
1367 nvme_init_identify(ctrl);
1368 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1369 nvme_queue_scan(ctrl);
1372 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1373 struct nvme_passthru_cmd __user *ucmd)
1375 struct nvme_passthru_cmd cmd;
1376 struct nvme_command c;
1377 unsigned timeout = 0;
1382 if (!capable(CAP_SYS_ADMIN))
1384 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1389 memset(&c, 0, sizeof(c));
1390 c.common.opcode = cmd.opcode;
1391 c.common.flags = cmd.flags;
1392 c.common.nsid = cpu_to_le32(cmd.nsid);
1393 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1394 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1395 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1396 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1397 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1398 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1399 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1400 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1403 timeout = msecs_to_jiffies(cmd.timeout_ms);
1405 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1406 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1407 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1408 (void __user *)(uintptr_t)cmd.metadata,
1409 cmd.metadata_len, 0, &result, timeout);
1410 nvme_passthru_end(ctrl, effects);
1413 if (put_user(result, &ucmd->result))
1420 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1421 struct nvme_passthru_cmd64 __user *ucmd)
1423 struct nvme_passthru_cmd64 cmd;
1424 struct nvme_command c;
1425 unsigned timeout = 0;
1429 if (!capable(CAP_SYS_ADMIN))
1431 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1436 memset(&c, 0, sizeof(c));
1437 c.common.opcode = cmd.opcode;
1438 c.common.flags = cmd.flags;
1439 c.common.nsid = cpu_to_le32(cmd.nsid);
1440 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1441 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1442 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1443 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1444 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1445 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1446 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1447 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1450 timeout = msecs_to_jiffies(cmd.timeout_ms);
1452 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1453 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1454 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1455 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1456 0, &cmd.result, timeout);
1457 nvme_passthru_end(ctrl, effects);
1460 if (put_user(cmd.result, &ucmd->result))
1468 * Issue ioctl requests on the first available path. Note that unlike normal
1469 * block layer requests we will not retry failed request on another controller.
1471 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1472 struct nvme_ns_head **head, int *srcu_idx)
1474 #ifdef CONFIG_NVME_MULTIPATH
1475 if (disk->fops == &nvme_ns_head_ops) {
1478 *head = disk->private_data;
1479 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1480 ns = nvme_find_path(*head);
1482 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1488 return disk->private_data;
1491 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1494 srcu_read_unlock(&head->srcu, idx);
1497 static bool is_ctrl_ioctl(unsigned int cmd)
1499 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1501 if (is_sed_ioctl(cmd))
1506 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1508 struct nvme_ns_head *head,
1511 struct nvme_ctrl *ctrl = ns->ctrl;
1514 nvme_get_ctrl(ns->ctrl);
1515 nvme_put_ns_from_disk(head, srcu_idx);
1518 case NVME_IOCTL_ADMIN_CMD:
1519 ret = nvme_user_cmd(ctrl, NULL, argp);
1521 case NVME_IOCTL_ADMIN64_CMD:
1522 ret = nvme_user_cmd64(ctrl, NULL, argp);
1525 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1528 nvme_put_ctrl(ctrl);
1532 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1533 unsigned int cmd, unsigned long arg)
1535 struct nvme_ns_head *head = NULL;
1536 void __user *argp = (void __user *)arg;
1540 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1542 return -EWOULDBLOCK;
1545 * Handle ioctls that apply to the controller instead of the namespace
1546 * seperately and drop the ns SRCU reference early. This avoids a
1547 * deadlock when deleting namespaces using the passthrough interface.
1549 if (is_ctrl_ioctl(cmd))
1550 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1554 force_successful_syscall_return();
1555 ret = ns->head->ns_id;
1557 case NVME_IOCTL_IO_CMD:
1558 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1560 case NVME_IOCTL_SUBMIT_IO:
1561 ret = nvme_submit_io(ns, argp);
1563 case NVME_IOCTL_IO64_CMD:
1564 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1568 ret = nvme_nvm_ioctl(ns, cmd, arg);
1573 nvme_put_ns_from_disk(head, srcu_idx);
1577 static int nvme_open(struct block_device *bdev, fmode_t mode)
1579 struct nvme_ns *ns = bdev->bd_disk->private_data;
1581 #ifdef CONFIG_NVME_MULTIPATH
1582 /* should never be called due to GENHD_FL_HIDDEN */
1583 if (WARN_ON_ONCE(ns->head->disk))
1586 if (!kref_get_unless_zero(&ns->kref))
1588 if (!try_module_get(ns->ctrl->ops->module))
1599 static void nvme_release(struct gendisk *disk, fmode_t mode)
1601 struct nvme_ns *ns = disk->private_data;
1603 module_put(ns->ctrl->ops->module);
1607 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1609 /* some standard values */
1610 geo->heads = 1 << 6;
1611 geo->sectors = 1 << 5;
1612 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1616 #ifdef CONFIG_BLK_DEV_INTEGRITY
1617 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1619 struct blk_integrity integrity;
1621 memset(&integrity, 0, sizeof(integrity));
1623 case NVME_NS_DPS_PI_TYPE3:
1624 integrity.profile = &t10_pi_type3_crc;
1625 integrity.tag_size = sizeof(u16) + sizeof(u32);
1626 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1628 case NVME_NS_DPS_PI_TYPE1:
1629 case NVME_NS_DPS_PI_TYPE2:
1630 integrity.profile = &t10_pi_type1_crc;
1631 integrity.tag_size = sizeof(u16);
1632 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1635 integrity.profile = NULL;
1638 integrity.tuple_size = ms;
1639 blk_integrity_register(disk, &integrity);
1640 blk_queue_max_integrity_segments(disk->queue, 1);
1643 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1646 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1648 static void nvme_set_chunk_size(struct nvme_ns *ns)
1650 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1651 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1654 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1656 struct nvme_ctrl *ctrl = ns->ctrl;
1657 struct request_queue *queue = disk->queue;
1658 u32 size = queue_logical_block_size(queue);
1660 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1661 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1665 if (ctrl->nr_streams && ns->sws && ns->sgs)
1666 size *= ns->sws * ns->sgs;
1668 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1669 NVME_DSM_MAX_RANGES);
1671 queue->limits.discard_alignment = 0;
1672 queue->limits.discard_granularity = size;
1674 /* If discard is already enabled, don't reset queue limits */
1675 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1678 blk_queue_max_discard_sectors(queue, UINT_MAX);
1679 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1681 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1682 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1685 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1688 unsigned short bs = 1 << ns->lba_shift;
1690 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1691 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1694 * Even though NVMe spec explicitly states that MDTS is not
1695 * applicable to the write-zeroes:- "The restriction does not apply to
1696 * commands that do not transfer data between the host and the
1697 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1698 * In order to be more cautious use controller's max_hw_sectors value
1699 * to configure the maximum sectors for the write-zeroes which is
1700 * configured based on the controller's MDTS field in the
1701 * nvme_init_identify() if available.
1703 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1704 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1706 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1708 blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1711 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1712 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1716 memset(ids, 0, sizeof(*ids));
1718 if (ctrl->vs >= NVME_VS(1, 1, 0))
1719 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1720 if (ctrl->vs >= NVME_VS(1, 2, 0))
1721 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1722 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1723 /* Don't treat error as fatal we potentially
1724 * already have a NGUID or EUI-64
1726 ret = nvme_identify_ns_descs(ctrl, nsid, ids);
1728 dev_warn(ctrl->device,
1729 "Identify Descriptors failed (%d)\n", ret);
1734 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1736 return !uuid_is_null(&ids->uuid) ||
1737 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1738 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1741 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1743 return uuid_equal(&a->uuid, &b->uuid) &&
1744 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1745 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1748 static void nvme_update_disk_info(struct gendisk *disk,
1749 struct nvme_ns *ns, struct nvme_id_ns *id)
1751 sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1752 unsigned short bs = 1 << ns->lba_shift;
1753 u32 atomic_bs, phys_bs, io_opt;
1755 if (ns->lba_shift > PAGE_SHIFT) {
1756 /* unsupported block size, set capacity to 0 later */
1759 blk_mq_freeze_queue(disk->queue);
1760 blk_integrity_unregister(disk);
1762 if (id->nabo == 0) {
1764 * Bit 1 indicates whether NAWUPF is defined for this namespace
1765 * and whether it should be used instead of AWUPF. If NAWUPF ==
1766 * 0 then AWUPF must be used instead.
1768 if (id->nsfeat & (1 << 1) && id->nawupf)
1769 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1771 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1777 if (id->nsfeat & (1 << 4)) {
1778 /* NPWG = Namespace Preferred Write Granularity */
1779 phys_bs *= 1 + le16_to_cpu(id->npwg);
1780 /* NOWS = Namespace Optimal Write Size */
1781 io_opt *= 1 + le16_to_cpu(id->nows);
1784 blk_queue_logical_block_size(disk->queue, bs);
1786 * Linux filesystems assume writing a single physical block is
1787 * an atomic operation. Hence limit the physical block size to the
1788 * value of the Atomic Write Unit Power Fail parameter.
1790 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1791 blk_queue_io_min(disk->queue, phys_bs);
1792 blk_queue_io_opt(disk->queue, io_opt);
1794 if (ns->ms && !ns->ext &&
1795 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1796 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1797 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1798 ns->lba_shift > PAGE_SHIFT)
1801 set_capacity(disk, capacity);
1803 nvme_config_discard(disk, ns);
1804 nvme_config_write_zeroes(disk, ns);
1806 if (id->nsattr & (1 << 0))
1807 set_disk_ro(disk, true);
1809 set_disk_ro(disk, false);
1811 blk_mq_unfreeze_queue(disk->queue);
1814 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1816 struct nvme_ns *ns = disk->private_data;
1819 * If identify namespace failed, use default 512 byte block size so
1820 * block layer can use before failing read/write for 0 capacity.
1822 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1823 if (ns->lba_shift == 0)
1825 ns->noiob = le16_to_cpu(id->noiob);
1826 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1827 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1828 /* the PI implementation requires metadata equal t10 pi tuple size */
1829 if (ns->ms == sizeof(struct t10_pi_tuple))
1830 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1835 nvme_set_chunk_size(ns);
1836 nvme_update_disk_info(disk, ns, id);
1837 #ifdef CONFIG_NVME_MULTIPATH
1838 if (ns->head->disk) {
1839 nvme_update_disk_info(ns->head->disk, ns, id);
1840 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1841 revalidate_disk(ns->head->disk);
1846 static int nvme_revalidate_disk(struct gendisk *disk)
1848 struct nvme_ns *ns = disk->private_data;
1849 struct nvme_ctrl *ctrl = ns->ctrl;
1850 struct nvme_id_ns *id;
1851 struct nvme_ns_ids ids;
1854 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1855 set_capacity(disk, 0);
1859 ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1863 if (id->ncap == 0) {
1868 __nvme_revalidate_disk(disk, id);
1869 ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1873 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1874 dev_err(ctrl->device,
1875 "identifiers changed for nsid %d\n", ns->head->ns_id);
1883 * Only fail the function if we got a fatal error back from the
1884 * device, otherwise ignore the error and just move on.
1886 if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1889 ret = blk_status_to_errno(nvme_error_status(ret));
1893 static char nvme_pr_type(enum pr_type type)
1896 case PR_WRITE_EXCLUSIVE:
1898 case PR_EXCLUSIVE_ACCESS:
1900 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1902 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1904 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1906 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1913 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1914 u64 key, u64 sa_key, u8 op)
1916 struct nvme_ns_head *head = NULL;
1918 struct nvme_command c;
1920 u8 data[16] = { 0, };
1922 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1924 return -EWOULDBLOCK;
1926 put_unaligned_le64(key, &data[0]);
1927 put_unaligned_le64(sa_key, &data[8]);
1929 memset(&c, 0, sizeof(c));
1930 c.common.opcode = op;
1931 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1932 c.common.cdw10 = cpu_to_le32(cdw10);
1934 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1935 nvme_put_ns_from_disk(head, srcu_idx);
1939 static int nvme_pr_register(struct block_device *bdev, u64 old,
1940 u64 new, unsigned flags)
1944 if (flags & ~PR_FL_IGNORE_KEY)
1947 cdw10 = old ? 2 : 0;
1948 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1949 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1950 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1953 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1954 enum pr_type type, unsigned flags)
1958 if (flags & ~PR_FL_IGNORE_KEY)
1961 cdw10 = nvme_pr_type(type) << 8;
1962 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1963 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1966 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1967 enum pr_type type, bool abort)
1969 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1970 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1973 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1975 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1976 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1979 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1981 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1982 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1985 static const struct pr_ops nvme_pr_ops = {
1986 .pr_register = nvme_pr_register,
1987 .pr_reserve = nvme_pr_reserve,
1988 .pr_release = nvme_pr_release,
1989 .pr_preempt = nvme_pr_preempt,
1990 .pr_clear = nvme_pr_clear,
1993 #ifdef CONFIG_BLK_SED_OPAL
1994 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1997 struct nvme_ctrl *ctrl = data;
1998 struct nvme_command cmd;
2000 memset(&cmd, 0, sizeof(cmd));
2002 cmd.common.opcode = nvme_admin_security_send;
2004 cmd.common.opcode = nvme_admin_security_recv;
2005 cmd.common.nsid = 0;
2006 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2007 cmd.common.cdw11 = cpu_to_le32(len);
2009 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2010 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2012 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2013 #endif /* CONFIG_BLK_SED_OPAL */
2015 static const struct block_device_operations nvme_fops = {
2016 .owner = THIS_MODULE,
2017 .ioctl = nvme_ioctl,
2018 .compat_ioctl = nvme_ioctl,
2020 .release = nvme_release,
2021 .getgeo = nvme_getgeo,
2022 .revalidate_disk= nvme_revalidate_disk,
2023 .pr_ops = &nvme_pr_ops,
2026 #ifdef CONFIG_NVME_MULTIPATH
2027 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2029 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2031 if (!kref_get_unless_zero(&head->ref))
2036 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2038 nvme_put_ns_head(disk->private_data);
2041 const struct block_device_operations nvme_ns_head_ops = {
2042 .owner = THIS_MODULE,
2043 .open = nvme_ns_head_open,
2044 .release = nvme_ns_head_release,
2045 .ioctl = nvme_ioctl,
2046 .compat_ioctl = nvme_ioctl,
2047 .getgeo = nvme_getgeo,
2048 .pr_ops = &nvme_pr_ops,
2050 #endif /* CONFIG_NVME_MULTIPATH */
2052 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2054 unsigned long timeout =
2055 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2056 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2059 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2062 if ((csts & NVME_CSTS_RDY) == bit)
2066 if (fatal_signal_pending(current))
2068 if (time_after(jiffies, timeout)) {
2069 dev_err(ctrl->device,
2070 "Device not ready; aborting %s\n", enabled ?
2071 "initialisation" : "reset");
2080 * If the device has been passed off to us in an enabled state, just clear
2081 * the enabled bit. The spec says we should set the 'shutdown notification
2082 * bits', but doing so may cause the device to complete commands to the
2083 * admin queue ... and we don't know what memory that might be pointing at!
2085 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2089 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2090 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2092 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2096 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2097 msleep(NVME_QUIRK_DELAY_AMOUNT);
2099 return nvme_wait_ready(ctrl, ctrl->cap, false);
2101 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2103 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2106 * Default to a 4K page size, with the intention to update this
2107 * path in the future to accomodate architectures with differing
2108 * kernel and IO page sizes.
2110 unsigned dev_page_min, page_shift = 12;
2113 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2115 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2118 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2120 if (page_shift < dev_page_min) {
2121 dev_err(ctrl->device,
2122 "Minimum device page size %u too large for host (%u)\n",
2123 1 << dev_page_min, 1 << page_shift);
2127 ctrl->page_size = 1 << page_shift;
2129 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2130 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2131 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2132 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2133 ctrl->ctrl_config |= NVME_CC_ENABLE;
2135 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2138 return nvme_wait_ready(ctrl, ctrl->cap, true);
2140 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2142 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2144 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2148 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2149 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2151 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2155 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2156 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2160 if (fatal_signal_pending(current))
2162 if (time_after(jiffies, timeout)) {
2163 dev_err(ctrl->device,
2164 "Device shutdown incomplete; abort shutdown\n");
2171 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2173 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2174 struct request_queue *q)
2178 if (ctrl->max_hw_sectors) {
2180 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2182 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2183 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2184 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2186 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2187 is_power_of_2(ctrl->max_hw_sectors))
2188 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2189 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2190 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2192 blk_queue_write_cache(q, vwc, vwc);
2195 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2200 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2203 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2204 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2207 dev_warn_once(ctrl->device,
2208 "could not set timestamp (%d)\n", ret);
2212 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2214 struct nvme_feat_host_behavior *host;
2217 /* Don't bother enabling the feature if retry delay is not reported */
2221 host = kzalloc(sizeof(*host), GFP_KERNEL);
2225 host->acre = NVME_ENABLE_ACRE;
2226 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2227 host, sizeof(*host), NULL);
2232 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2235 * APST (Autonomous Power State Transition) lets us program a
2236 * table of power state transitions that the controller will
2237 * perform automatically. We configure it with a simple
2238 * heuristic: we are willing to spend at most 2% of the time
2239 * transitioning between power states. Therefore, when running
2240 * in any given state, we will enter the next lower-power
2241 * non-operational state after waiting 50 * (enlat + exlat)
2242 * microseconds, as long as that state's exit latency is under
2243 * the requested maximum latency.
2245 * We will not autonomously enter any non-operational state for
2246 * which the total latency exceeds ps_max_latency_us. Users
2247 * can set ps_max_latency_us to zero to turn off APST.
2251 struct nvme_feat_auto_pst *table;
2257 * If APST isn't supported or if we haven't been initialized yet,
2258 * then don't do anything.
2263 if (ctrl->npss > 31) {
2264 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2268 table = kzalloc(sizeof(*table), GFP_KERNEL);
2272 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2273 /* Turn off APST. */
2275 dev_dbg(ctrl->device, "APST disabled\n");
2277 __le64 target = cpu_to_le64(0);
2281 * Walk through all states from lowest- to highest-power.
2282 * According to the spec, lower-numbered states use more
2283 * power. NPSS, despite the name, is the index of the
2284 * lowest-power state, not the number of states.
2286 for (state = (int)ctrl->npss; state >= 0; state--) {
2287 u64 total_latency_us, exit_latency_us, transition_ms;
2290 table->entries[state] = target;
2293 * Don't allow transitions to the deepest state
2294 * if it's quirked off.
2296 if (state == ctrl->npss &&
2297 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2301 * Is this state a useful non-operational state for
2302 * higher-power states to autonomously transition to?
2304 if (!(ctrl->psd[state].flags &
2305 NVME_PS_FLAGS_NON_OP_STATE))
2309 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2310 if (exit_latency_us > ctrl->ps_max_latency_us)
2315 le32_to_cpu(ctrl->psd[state].entry_lat);
2318 * This state is good. Use it as the APST idle
2319 * target for higher power states.
2321 transition_ms = total_latency_us + 19;
2322 do_div(transition_ms, 20);
2323 if (transition_ms > (1 << 24) - 1)
2324 transition_ms = (1 << 24) - 1;
2326 target = cpu_to_le64((state << 3) |
2327 (transition_ms << 8));
2332 if (total_latency_us > max_lat_us)
2333 max_lat_us = total_latency_us;
2339 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2341 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2342 max_ps, max_lat_us, (int)sizeof(*table), table);
2346 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2347 table, sizeof(*table), NULL);
2349 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2355 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2357 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2361 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2362 case PM_QOS_LATENCY_ANY:
2370 if (ctrl->ps_max_latency_us != latency) {
2371 ctrl->ps_max_latency_us = latency;
2372 nvme_configure_apst(ctrl);
2376 struct nvme_core_quirk_entry {
2378 * NVMe model and firmware strings are padded with spaces. For
2379 * simplicity, strings in the quirk table are padded with NULLs
2385 unsigned long quirks;
2388 static const struct nvme_core_quirk_entry core_quirks[] = {
2391 * This Toshiba device seems to die using any APST states. See:
2392 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2395 .mn = "THNSF5256GPUK TOSHIBA",
2396 .quirks = NVME_QUIRK_NO_APST,
2400 * This LiteON CL1-3D*-Q11 firmware version has a race
2401 * condition associated with actions related to suspend to idle
2402 * LiteON has resolved the problem in future firmware
2406 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2410 * This Kingston E8FK11.T firmware version has no interrupt
2411 * after resume with actions related to suspend to idle
2412 * https://bugzilla.kernel.org/show_bug.cgi?id=204887
2416 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2420 /* match is null-terminated but idstr is space-padded. */
2421 static bool string_matches(const char *idstr, const char *match, size_t len)
2428 matchlen = strlen(match);
2429 WARN_ON_ONCE(matchlen > len);
2431 if (memcmp(idstr, match, matchlen))
2434 for (; matchlen < len; matchlen++)
2435 if (idstr[matchlen] != ' ')
2441 static bool quirk_matches(const struct nvme_id_ctrl *id,
2442 const struct nvme_core_quirk_entry *q)
2444 return q->vid == le16_to_cpu(id->vid) &&
2445 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2446 string_matches(id->fr, q->fr, sizeof(id->fr));
2449 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2450 struct nvme_id_ctrl *id)
2455 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2456 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2457 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2458 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2462 if (ctrl->vs >= NVME_VS(1, 2, 1))
2463 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2466 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2467 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2468 "nqn.2014.08.org.nvmexpress:%04x%04x",
2469 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2470 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2471 off += sizeof(id->sn);
2472 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2473 off += sizeof(id->mn);
2474 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2477 static void nvme_release_subsystem(struct device *dev)
2479 struct nvme_subsystem *subsys =
2480 container_of(dev, struct nvme_subsystem, dev);
2482 if (subsys->instance >= 0)
2483 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2487 static void nvme_destroy_subsystem(struct kref *ref)
2489 struct nvme_subsystem *subsys =
2490 container_of(ref, struct nvme_subsystem, ref);
2492 mutex_lock(&nvme_subsystems_lock);
2493 list_del(&subsys->entry);
2494 mutex_unlock(&nvme_subsystems_lock);
2496 ida_destroy(&subsys->ns_ida);
2497 device_del(&subsys->dev);
2498 put_device(&subsys->dev);
2501 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2503 kref_put(&subsys->ref, nvme_destroy_subsystem);
2506 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2508 struct nvme_subsystem *subsys;
2510 lockdep_assert_held(&nvme_subsystems_lock);
2513 * Fail matches for discovery subsystems. This results
2514 * in each discovery controller bound to a unique subsystem.
2515 * This avoids issues with validating controller values
2516 * that can only be true when there is a single unique subsystem.
2517 * There may be multiple and completely independent entities
2518 * that provide discovery controllers.
2520 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2523 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2524 if (strcmp(subsys->subnqn, subsysnqn))
2526 if (!kref_get_unless_zero(&subsys->ref))
2534 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2535 struct device_attribute subsys_attr_##_name = \
2536 __ATTR(_name, _mode, _show, NULL)
2538 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2539 struct device_attribute *attr,
2542 struct nvme_subsystem *subsys =
2543 container_of(dev, struct nvme_subsystem, dev);
2545 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2547 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2549 #define nvme_subsys_show_str_function(field) \
2550 static ssize_t subsys_##field##_show(struct device *dev, \
2551 struct device_attribute *attr, char *buf) \
2553 struct nvme_subsystem *subsys = \
2554 container_of(dev, struct nvme_subsystem, dev); \
2555 return sprintf(buf, "%.*s\n", \
2556 (int)sizeof(subsys->field), subsys->field); \
2558 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2560 nvme_subsys_show_str_function(model);
2561 nvme_subsys_show_str_function(serial);
2562 nvme_subsys_show_str_function(firmware_rev);
2564 static struct attribute *nvme_subsys_attrs[] = {
2565 &subsys_attr_model.attr,
2566 &subsys_attr_serial.attr,
2567 &subsys_attr_firmware_rev.attr,
2568 &subsys_attr_subsysnqn.attr,
2569 #ifdef CONFIG_NVME_MULTIPATH
2570 &subsys_attr_iopolicy.attr,
2575 static struct attribute_group nvme_subsys_attrs_group = {
2576 .attrs = nvme_subsys_attrs,
2579 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2580 &nvme_subsys_attrs_group,
2584 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2585 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2587 struct nvme_ctrl *tmp;
2589 lockdep_assert_held(&nvme_subsystems_lock);
2591 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2592 if (tmp->state == NVME_CTRL_DELETING ||
2593 tmp->state == NVME_CTRL_DEAD)
2596 if (tmp->cntlid == ctrl->cntlid) {
2597 dev_err(ctrl->device,
2598 "Duplicate cntlid %u with %s, rejecting\n",
2599 ctrl->cntlid, dev_name(tmp->device));
2603 if ((id->cmic & (1 << 1)) ||
2604 (ctrl->opts && ctrl->opts->discovery_nqn))
2607 dev_err(ctrl->device,
2608 "Subsystem does not support multiple controllers\n");
2615 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2617 struct nvme_subsystem *subsys, *found;
2620 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2624 subsys->instance = -1;
2625 mutex_init(&subsys->lock);
2626 kref_init(&subsys->ref);
2627 INIT_LIST_HEAD(&subsys->ctrls);
2628 INIT_LIST_HEAD(&subsys->nsheads);
2629 nvme_init_subnqn(subsys, ctrl, id);
2630 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2631 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2632 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2633 subsys->vendor_id = le16_to_cpu(id->vid);
2634 subsys->cmic = id->cmic;
2635 subsys->awupf = le16_to_cpu(id->awupf);
2636 #ifdef CONFIG_NVME_MULTIPATH
2637 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2640 subsys->dev.class = nvme_subsys_class;
2641 subsys->dev.release = nvme_release_subsystem;
2642 subsys->dev.groups = nvme_subsys_attrs_groups;
2643 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2644 device_initialize(&subsys->dev);
2646 mutex_lock(&nvme_subsystems_lock);
2647 found = __nvme_find_get_subsystem(subsys->subnqn);
2649 put_device(&subsys->dev);
2652 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2654 goto out_put_subsystem;
2657 ret = device_add(&subsys->dev);
2659 dev_err(ctrl->device,
2660 "failed to register subsystem device.\n");
2661 put_device(&subsys->dev);
2664 ida_init(&subsys->ns_ida);
2665 list_add_tail(&subsys->entry, &nvme_subsystems);
2668 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2669 dev_name(ctrl->device));
2671 dev_err(ctrl->device,
2672 "failed to create sysfs link from subsystem.\n");
2673 goto out_put_subsystem;
2677 subsys->instance = ctrl->instance;
2678 ctrl->subsys = subsys;
2679 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2680 mutex_unlock(&nvme_subsystems_lock);
2684 nvme_put_subsystem(subsys);
2686 mutex_unlock(&nvme_subsystems_lock);
2690 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2691 void *log, size_t size, u64 offset)
2693 struct nvme_command c = { };
2694 unsigned long dwlen = size / 4 - 1;
2696 c.get_log_page.opcode = nvme_admin_get_log_page;
2697 c.get_log_page.nsid = cpu_to_le32(nsid);
2698 c.get_log_page.lid = log_page;
2699 c.get_log_page.lsp = lsp;
2700 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2701 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2702 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2703 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2705 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2708 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2713 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2718 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2719 ctrl->effects, sizeof(*ctrl->effects), 0);
2721 kfree(ctrl->effects);
2722 ctrl->effects = NULL;
2728 * Initialize the cached copies of the Identify data and various controller
2729 * register in our nvme_ctrl structure. This should be called as soon as
2730 * the admin queue is fully up and running.
2732 int nvme_init_identify(struct nvme_ctrl *ctrl)
2734 struct nvme_id_ctrl *id;
2735 int ret, page_shift;
2737 bool prev_apst_enabled;
2739 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2741 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2744 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2745 ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2747 if (ctrl->vs >= NVME_VS(1, 1, 0))
2748 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2750 ret = nvme_identify_ctrl(ctrl, &id);
2752 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2756 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2757 ret = nvme_get_effects_log(ctrl);
2762 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2763 ctrl->cntlid = le16_to_cpu(id->cntlid);
2765 if (!ctrl->identified) {
2768 ret = nvme_init_subsystem(ctrl, id);
2773 * Check for quirks. Quirk can depend on firmware version,
2774 * so, in principle, the set of quirks present can change
2775 * across a reset. As a possible future enhancement, we
2776 * could re-scan for quirks every time we reinitialize
2777 * the device, but we'd have to make sure that the driver
2778 * behaves intelligently if the quirks change.
2780 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2781 if (quirk_matches(id, &core_quirks[i]))
2782 ctrl->quirks |= core_quirks[i].quirks;
2786 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2787 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2788 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2791 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2792 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2793 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2795 ctrl->oacs = le16_to_cpu(id->oacs);
2796 ctrl->oncs = le16_to_cpu(id->oncs);
2797 ctrl->mtfa = le16_to_cpu(id->mtfa);
2798 ctrl->oaes = le32_to_cpu(id->oaes);
2799 atomic_set(&ctrl->abort_limit, id->acl + 1);
2800 ctrl->vwc = id->vwc;
2802 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2804 max_hw_sectors = UINT_MAX;
2805 ctrl->max_hw_sectors =
2806 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2808 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2809 ctrl->sgls = le32_to_cpu(id->sgls);
2810 ctrl->kas = le16_to_cpu(id->kas);
2811 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2812 ctrl->ctratt = le32_to_cpu(id->ctratt);
2816 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2818 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2819 shutdown_timeout, 60);
2821 if (ctrl->shutdown_timeout != shutdown_timeout)
2822 dev_info(ctrl->device,
2823 "Shutdown timeout set to %u seconds\n",
2824 ctrl->shutdown_timeout);
2826 ctrl->shutdown_timeout = shutdown_timeout;
2828 ctrl->npss = id->npss;
2829 ctrl->apsta = id->apsta;
2830 prev_apst_enabled = ctrl->apst_enabled;
2831 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2832 if (force_apst && id->apsta) {
2833 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2834 ctrl->apst_enabled = true;
2836 ctrl->apst_enabled = false;
2839 ctrl->apst_enabled = id->apsta;
2841 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2843 if (ctrl->ops->flags & NVME_F_FABRICS) {
2844 ctrl->icdoff = le16_to_cpu(id->icdoff);
2845 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2846 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2847 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2850 * In fabrics we need to verify the cntlid matches the
2853 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2858 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2859 dev_err(ctrl->device,
2860 "keep-alive support is mandatory for fabrics\n");
2865 ctrl->hmpre = le32_to_cpu(id->hmpre);
2866 ctrl->hmmin = le32_to_cpu(id->hmmin);
2867 ctrl->hmminds = le32_to_cpu(id->hmminds);
2868 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2871 ret = nvme_mpath_init(ctrl, id);
2877 if (ctrl->apst_enabled && !prev_apst_enabled)
2878 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2879 else if (!ctrl->apst_enabled && prev_apst_enabled)
2880 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2882 ret = nvme_configure_apst(ctrl);
2886 ret = nvme_configure_timestamp(ctrl);
2890 ret = nvme_configure_directives(ctrl);
2894 ret = nvme_configure_acre(ctrl);
2898 ctrl->identified = true;
2906 EXPORT_SYMBOL_GPL(nvme_init_identify);
2908 static int nvme_dev_open(struct inode *inode, struct file *file)
2910 struct nvme_ctrl *ctrl =
2911 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2913 switch (ctrl->state) {
2914 case NVME_CTRL_LIVE:
2917 return -EWOULDBLOCK;
2920 file->private_data = ctrl;
2924 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2929 down_read(&ctrl->namespaces_rwsem);
2930 if (list_empty(&ctrl->namespaces)) {
2935 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2936 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2937 dev_warn(ctrl->device,
2938 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2943 dev_warn(ctrl->device,
2944 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2945 kref_get(&ns->kref);
2946 up_read(&ctrl->namespaces_rwsem);
2948 ret = nvme_user_cmd(ctrl, ns, argp);
2953 up_read(&ctrl->namespaces_rwsem);
2957 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2960 struct nvme_ctrl *ctrl = file->private_data;
2961 void __user *argp = (void __user *)arg;
2964 case NVME_IOCTL_ADMIN_CMD:
2965 return nvme_user_cmd(ctrl, NULL, argp);
2966 case NVME_IOCTL_ADMIN64_CMD:
2967 return nvme_user_cmd64(ctrl, NULL, argp);
2968 case NVME_IOCTL_IO_CMD:
2969 return nvme_dev_user_cmd(ctrl, argp);
2970 case NVME_IOCTL_RESET:
2971 dev_warn(ctrl->device, "resetting controller\n");
2972 return nvme_reset_ctrl_sync(ctrl);
2973 case NVME_IOCTL_SUBSYS_RESET:
2974 return nvme_reset_subsystem(ctrl);
2975 case NVME_IOCTL_RESCAN:
2976 nvme_queue_scan(ctrl);
2983 static const struct file_operations nvme_dev_fops = {
2984 .owner = THIS_MODULE,
2985 .open = nvme_dev_open,
2986 .unlocked_ioctl = nvme_dev_ioctl,
2987 .compat_ioctl = nvme_dev_ioctl,
2990 static ssize_t nvme_sysfs_reset(struct device *dev,
2991 struct device_attribute *attr, const char *buf,
2994 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2997 ret = nvme_reset_ctrl_sync(ctrl);
3002 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3004 static ssize_t nvme_sysfs_rescan(struct device *dev,
3005 struct device_attribute *attr, const char *buf,
3008 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3010 nvme_queue_scan(ctrl);
3013 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3015 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3017 struct gendisk *disk = dev_to_disk(dev);
3019 if (disk->fops == &nvme_fops)
3020 return nvme_get_ns_from_dev(dev)->head;
3022 return disk->private_data;
3025 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3028 struct nvme_ns_head *head = dev_to_ns_head(dev);
3029 struct nvme_ns_ids *ids = &head->ids;
3030 struct nvme_subsystem *subsys = head->subsys;
3031 int serial_len = sizeof(subsys->serial);
3032 int model_len = sizeof(subsys->model);
3034 if (!uuid_is_null(&ids->uuid))
3035 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3037 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3038 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3040 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3041 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3043 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3044 subsys->serial[serial_len - 1] == '\0'))
3046 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3047 subsys->model[model_len - 1] == '\0'))
3050 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3051 serial_len, subsys->serial, model_len, subsys->model,
3054 static DEVICE_ATTR_RO(wwid);
3056 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3059 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3061 static DEVICE_ATTR_RO(nguid);
3063 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3066 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3068 /* For backward compatibility expose the NGUID to userspace if
3069 * we have no UUID set
3071 if (uuid_is_null(&ids->uuid)) {
3072 printk_ratelimited(KERN_WARNING
3073 "No UUID available providing old NGUID\n");
3074 return sprintf(buf, "%pU\n", ids->nguid);
3076 return sprintf(buf, "%pU\n", &ids->uuid);
3078 static DEVICE_ATTR_RO(uuid);
3080 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3083 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3085 static DEVICE_ATTR_RO(eui);
3087 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3090 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3092 static DEVICE_ATTR_RO(nsid);
3094 static struct attribute *nvme_ns_id_attrs[] = {
3095 &dev_attr_wwid.attr,
3096 &dev_attr_uuid.attr,
3097 &dev_attr_nguid.attr,
3099 &dev_attr_nsid.attr,
3100 #ifdef CONFIG_NVME_MULTIPATH
3101 &dev_attr_ana_grpid.attr,
3102 &dev_attr_ana_state.attr,
3107 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3108 struct attribute *a, int n)
3110 struct device *dev = container_of(kobj, struct device, kobj);
3111 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3113 if (a == &dev_attr_uuid.attr) {
3114 if (uuid_is_null(&ids->uuid) &&
3115 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3118 if (a == &dev_attr_nguid.attr) {
3119 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3122 if (a == &dev_attr_eui.attr) {
3123 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3126 #ifdef CONFIG_NVME_MULTIPATH
3127 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3128 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3130 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3137 static const struct attribute_group nvme_ns_id_attr_group = {
3138 .attrs = nvme_ns_id_attrs,
3139 .is_visible = nvme_ns_id_attrs_are_visible,
3142 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3143 &nvme_ns_id_attr_group,
3145 &nvme_nvm_attr_group,
3150 #define nvme_show_str_function(field) \
3151 static ssize_t field##_show(struct device *dev, \
3152 struct device_attribute *attr, char *buf) \
3154 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3155 return sprintf(buf, "%.*s\n", \
3156 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3158 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3160 nvme_show_str_function(model);
3161 nvme_show_str_function(serial);
3162 nvme_show_str_function(firmware_rev);
3164 #define nvme_show_int_function(field) \
3165 static ssize_t field##_show(struct device *dev, \
3166 struct device_attribute *attr, char *buf) \
3168 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3169 return sprintf(buf, "%d\n", ctrl->field); \
3171 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3173 nvme_show_int_function(cntlid);
3174 nvme_show_int_function(numa_node);
3175 nvme_show_int_function(queue_count);
3176 nvme_show_int_function(sqsize);
3178 static ssize_t nvme_sysfs_delete(struct device *dev,
3179 struct device_attribute *attr, const char *buf,
3182 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3184 if (device_remove_file_self(dev, attr))
3185 nvme_delete_ctrl_sync(ctrl);
3188 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3190 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3191 struct device_attribute *attr,
3194 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3196 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3198 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3200 static ssize_t nvme_sysfs_show_state(struct device *dev,
3201 struct device_attribute *attr,
3204 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3205 static const char *const state_name[] = {
3206 [NVME_CTRL_NEW] = "new",
3207 [NVME_CTRL_LIVE] = "live",
3208 [NVME_CTRL_RESETTING] = "resetting",
3209 [NVME_CTRL_CONNECTING] = "connecting",
3210 [NVME_CTRL_DELETING] = "deleting",
3211 [NVME_CTRL_DEAD] = "dead",
3214 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3215 state_name[ctrl->state])
3216 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3218 return sprintf(buf, "unknown state\n");
3221 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3223 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3224 struct device_attribute *attr,
3227 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3229 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3231 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3233 static ssize_t nvme_sysfs_show_address(struct device *dev,
3234 struct device_attribute *attr,
3237 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3239 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3241 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3243 static struct attribute *nvme_dev_attrs[] = {
3244 &dev_attr_reset_controller.attr,
3245 &dev_attr_rescan_controller.attr,
3246 &dev_attr_model.attr,
3247 &dev_attr_serial.attr,
3248 &dev_attr_firmware_rev.attr,
3249 &dev_attr_cntlid.attr,
3250 &dev_attr_delete_controller.attr,
3251 &dev_attr_transport.attr,
3252 &dev_attr_subsysnqn.attr,
3253 &dev_attr_address.attr,
3254 &dev_attr_state.attr,
3255 &dev_attr_numa_node.attr,
3256 &dev_attr_queue_count.attr,
3257 &dev_attr_sqsize.attr,
3261 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3262 struct attribute *a, int n)
3264 struct device *dev = container_of(kobj, struct device, kobj);
3265 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3267 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3269 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3275 static struct attribute_group nvme_dev_attrs_group = {
3276 .attrs = nvme_dev_attrs,
3277 .is_visible = nvme_dev_attrs_are_visible,
3280 static const struct attribute_group *nvme_dev_attr_groups[] = {
3281 &nvme_dev_attrs_group,
3285 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3288 struct nvme_ns_head *h;
3290 lockdep_assert_held(&subsys->lock);
3292 list_for_each_entry(h, &subsys->nsheads, entry) {
3293 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3300 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3301 struct nvme_ns_head *new)
3303 struct nvme_ns_head *h;
3305 lockdep_assert_held(&subsys->lock);
3307 list_for_each_entry(h, &subsys->nsheads, entry) {
3308 if (nvme_ns_ids_valid(&new->ids) &&
3309 !list_empty(&h->list) &&
3310 nvme_ns_ids_equal(&new->ids, &h->ids))
3317 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3318 unsigned nsid, struct nvme_id_ns *id)
3320 struct nvme_ns_head *head;
3321 size_t size = sizeof(*head);
3324 #ifdef CONFIG_NVME_MULTIPATH
3325 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3328 head = kzalloc(size, GFP_KERNEL);
3331 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3334 head->instance = ret;
3335 INIT_LIST_HEAD(&head->list);
3336 ret = init_srcu_struct(&head->srcu);
3338 goto out_ida_remove;
3339 head->subsys = ctrl->subsys;
3341 kref_init(&head->ref);
3343 ret = nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3345 goto out_cleanup_srcu;
3347 ret = __nvme_check_ids(ctrl->subsys, head);
3349 dev_err(ctrl->device,
3350 "duplicate IDs for nsid %d\n", nsid);
3351 goto out_cleanup_srcu;
3354 ret = nvme_mpath_alloc_disk(ctrl, head);
3356 goto out_cleanup_srcu;
3358 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3360 kref_get(&ctrl->subsys->ref);
3364 cleanup_srcu_struct(&head->srcu);
3366 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3371 ret = blk_status_to_errno(nvme_error_status(ret));
3372 return ERR_PTR(ret);
3375 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3376 struct nvme_id_ns *id)
3378 struct nvme_ctrl *ctrl = ns->ctrl;
3379 bool is_shared = id->nmic & (1 << 0);
3380 struct nvme_ns_head *head = NULL;
3383 mutex_lock(&ctrl->subsys->lock);
3385 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3387 head = nvme_alloc_ns_head(ctrl, nsid, id);
3389 ret = PTR_ERR(head);
3393 struct nvme_ns_ids ids;
3395 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3399 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3400 dev_err(ctrl->device,
3401 "IDs don't match for shared namespace %d\n",
3408 list_add_tail(&ns->siblings, &head->list);
3412 mutex_unlock(&ctrl->subsys->lock);
3414 ret = blk_status_to_errno(nvme_error_status(ret));
3418 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3420 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3421 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3423 return nsa->head->ns_id - nsb->head->ns_id;
3426 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3428 struct nvme_ns *ns, *ret = NULL;
3430 down_read(&ctrl->namespaces_rwsem);
3431 list_for_each_entry(ns, &ctrl->namespaces, list) {
3432 if (ns->head->ns_id == nsid) {
3433 if (!kref_get_unless_zero(&ns->kref))
3438 if (ns->head->ns_id > nsid)
3441 up_read(&ctrl->namespaces_rwsem);
3445 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3447 struct streams_directive_params s;
3450 if (!ctrl->nr_streams)
3453 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3457 ns->sws = le32_to_cpu(s.sws);
3458 ns->sgs = le16_to_cpu(s.sgs);
3461 unsigned int bs = 1 << ns->lba_shift;
3463 blk_queue_io_min(ns->queue, bs * ns->sws);
3465 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3471 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3474 struct gendisk *disk;
3475 struct nvme_id_ns *id;
3476 char disk_name[DISK_NAME_LEN];
3477 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3479 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3483 ns->queue = blk_mq_init_queue(ctrl->tagset);
3484 if (IS_ERR(ns->queue)) {
3485 ret = PTR_ERR(ns->queue);
3489 if (ctrl->opts && ctrl->opts->data_digest)
3490 ns->queue->backing_dev_info->capabilities
3491 |= BDI_CAP_STABLE_WRITES;
3493 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3494 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3495 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3497 ns->queue->queuedata = ns;
3500 kref_init(&ns->kref);
3501 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3503 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3504 nvme_set_queue_limits(ctrl, ns->queue);
3506 ret = nvme_identify_ns(ctrl, nsid, &id);
3508 goto out_free_queue;
3510 if (id->ncap == 0) {
3515 ret = nvme_init_ns_head(ns, nsid, id);
3518 nvme_setup_streams_ns(ctrl, ns);
3519 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3521 disk = alloc_disk_node(0, node);
3527 disk->fops = &nvme_fops;
3528 disk->private_data = ns;
3529 disk->queue = ns->queue;
3530 disk->flags = flags;
3531 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3534 __nvme_revalidate_disk(disk, id);
3536 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3537 ret = nvme_nvm_register(ns, disk_name, node);
3539 dev_warn(ctrl->device, "LightNVM init failure\n");
3544 down_write(&ctrl->namespaces_rwsem);
3545 list_add_tail(&ns->list, &ctrl->namespaces);
3546 up_write(&ctrl->namespaces_rwsem);
3548 nvme_get_ctrl(ctrl);
3550 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3552 nvme_mpath_add_disk(ns, id);
3553 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3560 mutex_lock(&ctrl->subsys->lock);
3561 list_del_rcu(&ns->siblings);
3562 mutex_unlock(&ctrl->subsys->lock);
3563 nvme_put_ns_head(ns->head);
3567 blk_cleanup_queue(ns->queue);
3571 ret = blk_status_to_errno(nvme_error_status(ret));
3575 static void nvme_ns_remove(struct nvme_ns *ns)
3577 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3580 nvme_fault_inject_fini(&ns->fault_inject);
3582 mutex_lock(&ns->ctrl->subsys->lock);
3583 list_del_rcu(&ns->siblings);
3584 mutex_unlock(&ns->ctrl->subsys->lock);
3585 synchronize_rcu(); /* guarantee not available in head->list */
3586 nvme_mpath_clear_current_path(ns);
3587 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3589 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3590 del_gendisk(ns->disk);
3591 blk_cleanup_queue(ns->queue);
3592 if (blk_get_integrity(ns->disk))
3593 blk_integrity_unregister(ns->disk);
3596 down_write(&ns->ctrl->namespaces_rwsem);
3597 list_del_init(&ns->list);
3598 up_write(&ns->ctrl->namespaces_rwsem);
3600 nvme_mpath_check_last_path(ns);
3604 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3608 ns = nvme_find_get_ns(ctrl, nsid);
3610 if (ns->disk && revalidate_disk(ns->disk))
3614 nvme_alloc_ns(ctrl, nsid);
3617 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3620 struct nvme_ns *ns, *next;
3623 down_write(&ctrl->namespaces_rwsem);
3624 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3625 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3626 list_move_tail(&ns->list, &rm_list);
3628 up_write(&ctrl->namespaces_rwsem);
3630 list_for_each_entry_safe(ns, next, &rm_list, list)
3635 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3639 unsigned i, j, nsid, prev = 0;
3640 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3643 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3647 for (i = 0; i < num_lists; i++) {
3648 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3652 for (j = 0; j < min(nn, 1024U); j++) {
3653 nsid = le32_to_cpu(ns_list[j]);
3657 nvme_validate_ns(ctrl, nsid);
3659 while (++prev < nsid) {
3660 ns = nvme_find_get_ns(ctrl, prev);
3670 nvme_remove_invalid_namespaces(ctrl, prev);
3676 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3680 for (i = 1; i <= nn; i++)
3681 nvme_validate_ns(ctrl, i);
3683 nvme_remove_invalid_namespaces(ctrl, nn);
3686 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3688 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3692 log = kzalloc(log_size, GFP_KERNEL);
3697 * We need to read the log to clear the AEN, but we don't want to rely
3698 * on it for the changed namespace information as userspace could have
3699 * raced with us in reading the log page, which could cause us to miss
3702 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3705 dev_warn(ctrl->device,
3706 "reading changed ns log failed: %d\n", error);
3711 static void nvme_scan_work(struct work_struct *work)
3713 struct nvme_ctrl *ctrl =
3714 container_of(work, struct nvme_ctrl, scan_work);
3715 struct nvme_id_ctrl *id;
3718 /* No tagset on a live ctrl means IO queues could not created */
3719 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3722 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3723 dev_info(ctrl->device, "rescanning namespaces.\n");
3724 nvme_clear_changed_ns_log(ctrl);
3727 if (nvme_identify_ctrl(ctrl, &id))
3730 mutex_lock(&ctrl->scan_lock);
3731 nn = le32_to_cpu(id->nn);
3732 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3733 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3734 if (!nvme_scan_ns_list(ctrl, nn))
3737 nvme_scan_ns_sequential(ctrl, nn);
3739 mutex_unlock(&ctrl->scan_lock);
3741 down_write(&ctrl->namespaces_rwsem);
3742 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3743 up_write(&ctrl->namespaces_rwsem);
3747 * This function iterates the namespace list unlocked to allow recovery from
3748 * controller failure. It is up to the caller to ensure the namespace list is
3749 * not modified by scan work while this function is executing.
3751 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3753 struct nvme_ns *ns, *next;
3757 * make sure to requeue I/O to all namespaces as these
3758 * might result from the scan itself and must complete
3759 * for the scan_work to make progress
3761 nvme_mpath_clear_ctrl_paths(ctrl);
3763 /* prevent racing with ns scanning */
3764 flush_work(&ctrl->scan_work);
3767 * The dead states indicates the controller was not gracefully
3768 * disconnected. In that case, we won't be able to flush any data while
3769 * removing the namespaces' disks; fail all the queues now to avoid
3770 * potentially having to clean up the failed sync later.
3772 if (ctrl->state == NVME_CTRL_DEAD)
3773 nvme_kill_queues(ctrl);
3775 down_write(&ctrl->namespaces_rwsem);
3776 list_splice_init(&ctrl->namespaces, &ns_list);
3777 up_write(&ctrl->namespaces_rwsem);
3779 list_for_each_entry_safe(ns, next, &ns_list, list)
3782 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3784 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3786 struct nvme_ctrl *ctrl =
3787 container_of(dev, struct nvme_ctrl, ctrl_device);
3788 struct nvmf_ctrl_options *opts = ctrl->opts;
3791 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3796 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3800 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3801 opts->trsvcid ?: "none");
3805 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3806 opts->host_traddr ?: "none");
3811 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3813 char *envp[2] = { NULL, NULL };
3814 u32 aen_result = ctrl->aen_result;
3816 ctrl->aen_result = 0;
3820 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3823 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3827 static void nvme_async_event_work(struct work_struct *work)
3829 struct nvme_ctrl *ctrl =
3830 container_of(work, struct nvme_ctrl, async_event_work);
3832 nvme_aen_uevent(ctrl);
3833 ctrl->ops->submit_async_event(ctrl);
3836 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3841 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3847 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3850 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3852 struct nvme_fw_slot_info_log *log;
3854 log = kmalloc(sizeof(*log), GFP_KERNEL);
3858 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3860 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3864 static void nvme_fw_act_work(struct work_struct *work)
3866 struct nvme_ctrl *ctrl = container_of(work,
3867 struct nvme_ctrl, fw_act_work);
3868 unsigned long fw_act_timeout;
3871 fw_act_timeout = jiffies +
3872 msecs_to_jiffies(ctrl->mtfa * 100);
3874 fw_act_timeout = jiffies +
3875 msecs_to_jiffies(admin_timeout * 1000);
3877 nvme_stop_queues(ctrl);
3878 while (nvme_ctrl_pp_status(ctrl)) {
3879 if (time_after(jiffies, fw_act_timeout)) {
3880 dev_warn(ctrl->device,
3881 "Fw activation timeout, reset controller\n");
3882 nvme_try_sched_reset(ctrl);
3888 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
3891 nvme_start_queues(ctrl);
3892 /* read FW slot information to clear the AER */
3893 nvme_get_fw_slot_info(ctrl);
3896 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3898 u32 aer_notice_type = (result & 0xff00) >> 8;
3900 trace_nvme_async_event(ctrl, aer_notice_type);
3902 switch (aer_notice_type) {
3903 case NVME_AER_NOTICE_NS_CHANGED:
3904 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3905 nvme_queue_scan(ctrl);
3907 case NVME_AER_NOTICE_FW_ACT_STARTING:
3909 * We are (ab)using the RESETTING state to prevent subsequent
3910 * recovery actions from interfering with the controller's
3911 * firmware activation.
3913 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
3914 queue_work(nvme_wq, &ctrl->fw_act_work);
3916 #ifdef CONFIG_NVME_MULTIPATH
3917 case NVME_AER_NOTICE_ANA:
3918 if (!ctrl->ana_log_buf)
3920 queue_work(nvme_wq, &ctrl->ana_work);
3923 case NVME_AER_NOTICE_DISC_CHANGED:
3924 ctrl->aen_result = result;
3927 dev_warn(ctrl->device, "async event result %08x\n", result);
3931 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3932 volatile union nvme_result *res)
3934 u32 result = le32_to_cpu(res->u32);
3935 u32 aer_type = result & 0x07;
3937 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3941 case NVME_AER_NOTICE:
3942 nvme_handle_aen_notice(ctrl, result);
3944 case NVME_AER_ERROR:
3945 case NVME_AER_SMART:
3948 trace_nvme_async_event(ctrl, aer_type);
3949 ctrl->aen_result = result;
3954 queue_work(nvme_wq, &ctrl->async_event_work);
3956 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3958 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3960 nvme_mpath_stop(ctrl);
3961 nvme_stop_keep_alive(ctrl);
3962 flush_work(&ctrl->async_event_work);
3963 cancel_work_sync(&ctrl->fw_act_work);
3965 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3967 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3970 nvme_start_keep_alive(ctrl);
3972 nvme_enable_aen(ctrl);
3974 if (ctrl->queue_count > 1) {
3975 nvme_queue_scan(ctrl);
3976 nvme_start_queues(ctrl);
3979 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3981 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3983 nvme_fault_inject_fini(&ctrl->fault_inject);
3984 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3985 cdev_device_del(&ctrl->cdev, ctrl->device);
3987 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3989 static void nvme_free_ctrl(struct device *dev)
3991 struct nvme_ctrl *ctrl =
3992 container_of(dev, struct nvme_ctrl, ctrl_device);
3993 struct nvme_subsystem *subsys = ctrl->subsys;
3995 if (subsys && ctrl->instance != subsys->instance)
3996 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3998 kfree(ctrl->effects);
3999 nvme_mpath_uninit(ctrl);
4000 __free_page(ctrl->discard_page);
4003 mutex_lock(&nvme_subsystems_lock);
4004 list_del(&ctrl->subsys_entry);
4005 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4006 mutex_unlock(&nvme_subsystems_lock);
4009 ctrl->ops->free_ctrl(ctrl);
4012 nvme_put_subsystem(subsys);
4016 * Initialize a NVMe controller structures. This needs to be called during
4017 * earliest initialization so that we have the initialized structured around
4020 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4021 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4025 ctrl->state = NVME_CTRL_NEW;
4026 spin_lock_init(&ctrl->lock);
4027 mutex_init(&ctrl->scan_lock);
4028 INIT_LIST_HEAD(&ctrl->namespaces);
4029 init_rwsem(&ctrl->namespaces_rwsem);
4032 ctrl->quirks = quirks;
4033 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4034 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4035 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4036 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4037 init_waitqueue_head(&ctrl->state_wq);
4039 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4040 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4041 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4043 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4045 ctrl->discard_page = alloc_page(GFP_KERNEL);
4046 if (!ctrl->discard_page) {
4051 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4054 ctrl->instance = ret;
4056 device_initialize(&ctrl->ctrl_device);
4057 ctrl->device = &ctrl->ctrl_device;
4058 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4059 ctrl->device->class = nvme_class;
4060 ctrl->device->parent = ctrl->dev;
4061 ctrl->device->groups = nvme_dev_attr_groups;
4062 ctrl->device->release = nvme_free_ctrl;
4063 dev_set_drvdata(ctrl->device, ctrl);
4064 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4066 goto out_release_instance;
4068 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4069 ctrl->cdev.owner = ops->module;
4070 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4075 * Initialize latency tolerance controls. The sysfs files won't
4076 * be visible to userspace unless the device actually supports APST.
4078 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4079 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4080 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4082 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4086 kfree_const(ctrl->device->kobj.name);
4087 out_release_instance:
4088 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4090 if (ctrl->discard_page)
4091 __free_page(ctrl->discard_page);
4094 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4097 * nvme_kill_queues(): Ends all namespace queues
4098 * @ctrl: the dead controller that needs to end
4100 * Call this function when the driver determines it is unable to get the
4101 * controller in a state capable of servicing IO.
4103 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4107 down_read(&ctrl->namespaces_rwsem);
4109 /* Forcibly unquiesce queues to avoid blocking dispatch */
4110 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4111 blk_mq_unquiesce_queue(ctrl->admin_q);
4113 list_for_each_entry(ns, &ctrl->namespaces, list)
4114 nvme_set_queue_dying(ns);
4116 up_read(&ctrl->namespaces_rwsem);
4118 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4120 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4124 down_read(&ctrl->namespaces_rwsem);
4125 list_for_each_entry(ns, &ctrl->namespaces, list)
4126 blk_mq_unfreeze_queue(ns->queue);
4127 up_read(&ctrl->namespaces_rwsem);
4129 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4131 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4135 down_read(&ctrl->namespaces_rwsem);
4136 list_for_each_entry(ns, &ctrl->namespaces, list) {
4137 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4141 up_read(&ctrl->namespaces_rwsem);
4143 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4145 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4149 down_read(&ctrl->namespaces_rwsem);
4150 list_for_each_entry(ns, &ctrl->namespaces, list)
4151 blk_mq_freeze_queue_wait(ns->queue);
4152 up_read(&ctrl->namespaces_rwsem);
4154 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4156 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4160 down_read(&ctrl->namespaces_rwsem);
4161 list_for_each_entry(ns, &ctrl->namespaces, list)
4162 blk_freeze_queue_start(ns->queue);
4163 up_read(&ctrl->namespaces_rwsem);
4165 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4167 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4171 down_read(&ctrl->namespaces_rwsem);
4172 list_for_each_entry(ns, &ctrl->namespaces, list)
4173 blk_mq_quiesce_queue(ns->queue);
4174 up_read(&ctrl->namespaces_rwsem);
4176 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4178 void nvme_start_queues(struct nvme_ctrl *ctrl)
4182 down_read(&ctrl->namespaces_rwsem);
4183 list_for_each_entry(ns, &ctrl->namespaces, list)
4184 blk_mq_unquiesce_queue(ns->queue);
4185 up_read(&ctrl->namespaces_rwsem);
4187 EXPORT_SYMBOL_GPL(nvme_start_queues);
4190 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4194 down_read(&ctrl->namespaces_rwsem);
4195 list_for_each_entry(ns, &ctrl->namespaces, list)
4196 blk_sync_queue(ns->queue);
4197 up_read(&ctrl->namespaces_rwsem);
4200 blk_sync_queue(ctrl->admin_q);
4202 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4205 * Check we didn't inadvertently grow the command structure sizes:
4207 static inline void _nvme_check_size(void)
4209 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4210 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4211 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4212 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4213 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4214 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4215 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4216 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4217 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4218 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4219 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4220 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4221 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4222 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4223 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4224 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4225 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4229 static int __init nvme_core_init(void)
4231 int result = -ENOMEM;
4235 nvme_wq = alloc_workqueue("nvme-wq",
4236 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4240 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4241 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4245 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4246 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4247 if (!nvme_delete_wq)
4248 goto destroy_reset_wq;
4250 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4252 goto destroy_delete_wq;
4254 nvme_class = class_create(THIS_MODULE, "nvme");
4255 if (IS_ERR(nvme_class)) {
4256 result = PTR_ERR(nvme_class);
4257 goto unregister_chrdev;
4259 nvme_class->dev_uevent = nvme_class_uevent;
4261 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4262 if (IS_ERR(nvme_subsys_class)) {
4263 result = PTR_ERR(nvme_subsys_class);
4269 class_destroy(nvme_class);
4271 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4273 destroy_workqueue(nvme_delete_wq);
4275 destroy_workqueue(nvme_reset_wq);
4277 destroy_workqueue(nvme_wq);
4282 static void __exit nvme_core_exit(void)
4284 class_destroy(nvme_subsys_class);
4285 class_destroy(nvme_class);
4286 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4287 destroy_workqueue(nvme_delete_wq);
4288 destroy_workqueue(nvme_reset_wq);
4289 destroy_workqueue(nvme_wq);
4292 MODULE_LICENSE("GPL");
4293 MODULE_VERSION("1.0");
4294 module_init(nvme_core_init);
4295 module_exit(nvme_core_exit);