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/compat.h>
10 #include <linux/delay.h>
11 #include <linux/errno.h>
12 #include <linux/hdreg.h>
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/list_sort.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
20 #include <linux/ptrace.h>
21 #include <linux/nvme_ioctl.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
28 #define CREATE_TRACE_POINTS
31 #define NVME_MINORS (1U << MINORBITS)
33 unsigned int admin_timeout = 60;
34 module_param(admin_timeout, uint, 0644);
35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
36 EXPORT_SYMBOL_GPL(admin_timeout);
38 unsigned int nvme_io_timeout = 30;
39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
43 static unsigned char shutdown_timeout = 5;
44 module_param(shutdown_timeout, byte, 0644);
45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
47 static u8 nvme_max_retries = 5;
48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
51 static unsigned long default_ps_max_latency_us = 100000;
52 module_param(default_ps_max_latency_us, ulong, 0644);
53 MODULE_PARM_DESC(default_ps_max_latency_us,
54 "max power saving latency for new devices; use PM QOS to change per device");
56 static bool force_apst;
57 module_param(force_apst, bool, 0644);
58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
61 module_param(streams, bool, 0644);
62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
65 * nvme_wq - hosts nvme related works that are not reset or delete
66 * nvme_reset_wq - hosts nvme reset works
67 * nvme_delete_wq - hosts nvme delete works
69 * nvme_wq will host works such as scan, aen handling, fw activation,
70 * keep-alive, periodic reconnects etc. nvme_reset_wq
71 * runs reset works which also flush works hosted on nvme_wq for
72 * serialization purposes. nvme_delete_wq host controller deletion
73 * works which flush reset works for serialization.
75 struct workqueue_struct *nvme_wq;
76 EXPORT_SYMBOL_GPL(nvme_wq);
78 struct workqueue_struct *nvme_reset_wq;
79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
81 struct workqueue_struct *nvme_delete_wq;
82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
84 static LIST_HEAD(nvme_subsystems);
85 static DEFINE_MUTEX(nvme_subsystems_lock);
87 static DEFINE_IDA(nvme_instance_ida);
88 static dev_t nvme_chr_devt;
89 static struct class *nvme_class;
90 static struct class *nvme_subsys_class;
92 static int nvme_revalidate_disk(struct gendisk *disk);
93 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
94 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
97 static void nvme_set_queue_dying(struct nvme_ns *ns)
100 * Revalidating a dead namespace sets capacity to 0. This will end
101 * buffered writers dirtying pages that can't be synced.
103 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
105 blk_set_queue_dying(ns->queue);
106 /* Forcibly unquiesce queues to avoid blocking dispatch */
107 blk_mq_unquiesce_queue(ns->queue);
109 * Revalidate after unblocking dispatchers that may be holding bd_butex
111 revalidate_disk(ns->disk);
114 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
117 * Only new queue scan work when admin and IO queues are both alive
119 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
120 queue_work(nvme_wq, &ctrl->scan_work);
124 * Use this function to proceed with scheduling reset_work for a controller
125 * that had previously been set to the resetting state. This is intended for
126 * code paths that can't be interrupted by other reset attempts. A hot removal
127 * may prevent this from succeeding.
129 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
131 if (ctrl->state != NVME_CTRL_RESETTING)
133 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
137 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
139 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
141 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
143 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
147 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
149 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
153 ret = nvme_reset_ctrl(ctrl);
155 flush_work(&ctrl->reset_work);
156 if (ctrl->state != NVME_CTRL_LIVE)
162 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
164 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
166 dev_info(ctrl->device,
167 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
169 flush_work(&ctrl->reset_work);
170 nvme_stop_ctrl(ctrl);
171 nvme_remove_namespaces(ctrl);
172 ctrl->ops->delete_ctrl(ctrl);
173 nvme_uninit_ctrl(ctrl);
176 static void nvme_delete_ctrl_work(struct work_struct *work)
178 struct nvme_ctrl *ctrl =
179 container_of(work, struct nvme_ctrl, delete_work);
181 nvme_do_delete_ctrl(ctrl);
184 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
186 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
188 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
192 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
194 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
197 * Keep a reference until nvme_do_delete_ctrl() complete,
198 * since ->delete_ctrl can free the controller.
201 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
202 nvme_do_delete_ctrl(ctrl);
206 static blk_status_t nvme_error_status(u16 status)
208 switch (status & 0x7ff) {
209 case NVME_SC_SUCCESS:
211 case NVME_SC_CAP_EXCEEDED:
212 return BLK_STS_NOSPC;
213 case NVME_SC_LBA_RANGE:
214 case NVME_SC_CMD_INTERRUPTED:
215 case NVME_SC_NS_NOT_READY:
216 return BLK_STS_TARGET;
217 case NVME_SC_BAD_ATTRIBUTES:
218 case NVME_SC_ONCS_NOT_SUPPORTED:
219 case NVME_SC_INVALID_OPCODE:
220 case NVME_SC_INVALID_FIELD:
221 case NVME_SC_INVALID_NS:
222 return BLK_STS_NOTSUPP;
223 case NVME_SC_WRITE_FAULT:
224 case NVME_SC_READ_ERROR:
225 case NVME_SC_UNWRITTEN_BLOCK:
226 case NVME_SC_ACCESS_DENIED:
227 case NVME_SC_READ_ONLY:
228 case NVME_SC_COMPARE_FAILED:
229 return BLK_STS_MEDIUM;
230 case NVME_SC_GUARD_CHECK:
231 case NVME_SC_APPTAG_CHECK:
232 case NVME_SC_REFTAG_CHECK:
233 case NVME_SC_INVALID_PI:
234 return BLK_STS_PROTECTION;
235 case NVME_SC_RESERVATION_CONFLICT:
236 return BLK_STS_NEXUS;
237 case NVME_SC_HOST_PATH_ERROR:
238 return BLK_STS_TRANSPORT;
240 return BLK_STS_IOERR;
244 static inline bool nvme_req_needs_retry(struct request *req)
246 if (blk_noretry_request(req))
248 if (nvme_req(req)->status & NVME_SC_DNR)
250 if (nvme_req(req)->retries >= nvme_max_retries)
255 static void nvme_retry_req(struct request *req)
257 struct nvme_ns *ns = req->q->queuedata;
258 unsigned long delay = 0;
261 /* The mask and shift result must be <= 3 */
262 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
264 delay = ns->ctrl->crdt[crd - 1] * 100;
266 nvme_req(req)->retries++;
267 blk_mq_requeue_request(req, false);
268 blk_mq_delay_kick_requeue_list(req->q, delay);
271 void nvme_complete_rq(struct request *req)
273 blk_status_t status = nvme_error_status(nvme_req(req)->status);
275 trace_nvme_complete_rq(req);
277 nvme_cleanup_cmd(req);
279 if (nvme_req(req)->ctrl->kas)
280 nvme_req(req)->ctrl->comp_seen = true;
282 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
283 if ((req->cmd_flags & REQ_NVME_MPATH) && nvme_failover_req(req))
286 if (!blk_queue_dying(req->q)) {
292 nvme_trace_bio_complete(req, status);
293 blk_mq_end_request(req, status);
295 EXPORT_SYMBOL_GPL(nvme_complete_rq);
297 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
299 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
300 "Cancelling I/O %d", req->tag);
302 /* don't abort one completed request */
303 if (blk_mq_request_completed(req))
306 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
307 blk_mq_complete_request(req);
310 EXPORT_SYMBOL_GPL(nvme_cancel_request);
312 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
313 enum nvme_ctrl_state new_state)
315 enum nvme_ctrl_state old_state;
317 bool changed = false;
319 spin_lock_irqsave(&ctrl->lock, flags);
321 old_state = ctrl->state;
326 case NVME_CTRL_RESETTING:
327 case NVME_CTRL_CONNECTING:
334 case NVME_CTRL_RESETTING:
344 case NVME_CTRL_CONNECTING:
347 case NVME_CTRL_RESETTING:
354 case NVME_CTRL_DELETING:
357 case NVME_CTRL_RESETTING:
358 case NVME_CTRL_CONNECTING:
367 case NVME_CTRL_DELETING:
379 ctrl->state = new_state;
380 wake_up_all(&ctrl->state_wq);
383 spin_unlock_irqrestore(&ctrl->lock, flags);
384 if (changed && ctrl->state == NVME_CTRL_LIVE)
385 nvme_kick_requeue_lists(ctrl);
388 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
391 * Returns true for sink states that can't ever transition back to live.
393 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
395 switch (ctrl->state) {
398 case NVME_CTRL_RESETTING:
399 case NVME_CTRL_CONNECTING:
401 case NVME_CTRL_DELETING:
405 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
411 * Waits for the controller state to be resetting, or returns false if it is
412 * not possible to ever transition to that state.
414 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
416 wait_event(ctrl->state_wq,
417 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
418 nvme_state_terminal(ctrl));
419 return ctrl->state == NVME_CTRL_RESETTING;
421 EXPORT_SYMBOL_GPL(nvme_wait_reset);
423 static void nvme_free_ns_head(struct kref *ref)
425 struct nvme_ns_head *head =
426 container_of(ref, struct nvme_ns_head, ref);
428 nvme_mpath_remove_disk(head);
429 ida_simple_remove(&head->subsys->ns_ida, head->instance);
430 cleanup_srcu_struct(&head->srcu);
431 nvme_put_subsystem(head->subsys);
435 static void nvme_put_ns_head(struct nvme_ns_head *head)
437 kref_put(&head->ref, nvme_free_ns_head);
440 static void nvme_free_ns(struct kref *kref)
442 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
445 nvme_nvm_unregister(ns);
448 nvme_put_ns_head(ns->head);
449 nvme_put_ctrl(ns->ctrl);
453 static void nvme_put_ns(struct nvme_ns *ns)
455 kref_put(&ns->kref, nvme_free_ns);
458 static inline void nvme_clear_nvme_request(struct request *req)
460 if (!(req->rq_flags & RQF_DONTPREP)) {
461 nvme_req(req)->retries = 0;
462 nvme_req(req)->flags = 0;
463 req->rq_flags |= RQF_DONTPREP;
467 struct request *nvme_alloc_request(struct request_queue *q,
468 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
470 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
473 if (qid == NVME_QID_ANY) {
474 req = blk_mq_alloc_request(q, op, flags);
476 req = blk_mq_alloc_request_hctx(q, op, flags,
482 req->cmd_flags |= REQ_FAILFAST_DRIVER;
483 nvme_clear_nvme_request(req);
484 nvme_req(req)->cmd = cmd;
488 EXPORT_SYMBOL_GPL(nvme_alloc_request);
490 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
492 struct nvme_command c;
494 memset(&c, 0, sizeof(c));
496 c.directive.opcode = nvme_admin_directive_send;
497 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
498 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
499 c.directive.dtype = NVME_DIR_IDENTIFY;
500 c.directive.tdtype = NVME_DIR_STREAMS;
501 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
503 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
506 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
508 return nvme_toggle_streams(ctrl, false);
511 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
513 return nvme_toggle_streams(ctrl, true);
516 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
517 struct streams_directive_params *s, u32 nsid)
519 struct nvme_command c;
521 memset(&c, 0, sizeof(c));
522 memset(s, 0, sizeof(*s));
524 c.directive.opcode = nvme_admin_directive_recv;
525 c.directive.nsid = cpu_to_le32(nsid);
526 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
527 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
528 c.directive.dtype = NVME_DIR_STREAMS;
530 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
533 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
535 struct streams_directive_params s;
538 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
543 ret = nvme_enable_streams(ctrl);
547 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
549 goto out_disable_stream;
551 ctrl->nssa = le16_to_cpu(s.nssa);
552 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
553 dev_info(ctrl->device, "too few streams (%u) available\n",
555 goto out_disable_stream;
558 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
559 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
563 nvme_disable_streams(ctrl);
568 * Check if 'req' has a write hint associated with it. If it does, assign
569 * a valid namespace stream to the write.
571 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
572 struct request *req, u16 *control,
575 enum rw_hint streamid = req->write_hint;
577 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
581 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
584 *control |= NVME_RW_DTYPE_STREAMS;
585 *dsmgmt |= streamid << 16;
588 if (streamid < ARRAY_SIZE(req->q->write_hints))
589 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
592 static inline void nvme_setup_flush(struct nvme_ns *ns,
593 struct nvme_command *cmnd)
595 cmnd->common.opcode = nvme_cmd_flush;
596 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
599 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
600 struct nvme_command *cmnd)
602 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
603 struct nvme_dsm_range *range;
607 * Some devices do not consider the DSM 'Number of Ranges' field when
608 * determining how much data to DMA. Always allocate memory for maximum
609 * number of segments to prevent device reading beyond end of buffer.
611 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
613 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
616 * If we fail allocation our range, fallback to the controller
617 * discard page. If that's also busy, it's safe to return
618 * busy, as we know we can make progress once that's freed.
620 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
621 return BLK_STS_RESOURCE;
623 range = page_address(ns->ctrl->discard_page);
626 __rq_for_each_bio(bio, req) {
627 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
628 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
631 range[n].cattr = cpu_to_le32(0);
632 range[n].nlb = cpu_to_le32(nlb);
633 range[n].slba = cpu_to_le64(slba);
638 if (WARN_ON_ONCE(n != segments)) {
639 if (virt_to_page(range) == ns->ctrl->discard_page)
640 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
643 return BLK_STS_IOERR;
646 cmnd->dsm.opcode = nvme_cmd_dsm;
647 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
648 cmnd->dsm.nr = cpu_to_le32(segments - 1);
649 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
651 req->special_vec.bv_page = virt_to_page(range);
652 req->special_vec.bv_offset = offset_in_page(range);
653 req->special_vec.bv_len = alloc_size;
654 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
659 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
660 struct request *req, struct nvme_command *cmnd)
662 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
663 return nvme_setup_discard(ns, req, cmnd);
665 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
666 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
667 cmnd->write_zeroes.slba =
668 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
669 cmnd->write_zeroes.length =
670 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
671 cmnd->write_zeroes.control = 0;
675 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
676 struct request *req, struct nvme_command *cmnd)
678 struct nvme_ctrl *ctrl = ns->ctrl;
682 if (req->cmd_flags & REQ_FUA)
683 control |= NVME_RW_FUA;
684 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
685 control |= NVME_RW_LR;
687 if (req->cmd_flags & REQ_RAHEAD)
688 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
690 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
691 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
692 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
693 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
695 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
696 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
700 * If formated with metadata, the block layer always provides a
701 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
702 * we enable the PRACT bit for protection information or set the
703 * namespace capacity to zero to prevent any I/O.
705 if (!blk_integrity_rq(req)) {
706 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
707 return BLK_STS_NOTSUPP;
708 control |= NVME_RW_PRINFO_PRACT;
711 switch (ns->pi_type) {
712 case NVME_NS_DPS_PI_TYPE3:
713 control |= NVME_RW_PRINFO_PRCHK_GUARD;
715 case NVME_NS_DPS_PI_TYPE1:
716 case NVME_NS_DPS_PI_TYPE2:
717 control |= NVME_RW_PRINFO_PRCHK_GUARD |
718 NVME_RW_PRINFO_PRCHK_REF;
719 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
724 cmnd->rw.control = cpu_to_le16(control);
725 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
729 void nvme_cleanup_cmd(struct request *req)
731 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
732 struct nvme_ns *ns = req->rq_disk->private_data;
733 struct page *page = req->special_vec.bv_page;
735 if (page == ns->ctrl->discard_page)
736 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
738 kfree(page_address(page) + req->special_vec.bv_offset);
741 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
743 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
744 struct nvme_command *cmd)
746 blk_status_t ret = BLK_STS_OK;
748 nvme_clear_nvme_request(req);
750 memset(cmd, 0, sizeof(*cmd));
751 switch (req_op(req)) {
754 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
757 nvme_setup_flush(ns, cmd);
759 case REQ_OP_WRITE_ZEROES:
760 ret = nvme_setup_write_zeroes(ns, req, cmd);
763 ret = nvme_setup_discard(ns, req, cmd);
767 ret = nvme_setup_rw(ns, req, cmd);
771 return BLK_STS_IOERR;
774 cmd->common.command_id = req->tag;
775 trace_nvme_setup_cmd(req, cmd);
778 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
780 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
782 struct completion *waiting = rq->end_io_data;
784 rq->end_io_data = NULL;
788 static void nvme_execute_rq_polled(struct request_queue *q,
789 struct gendisk *bd_disk, struct request *rq, int at_head)
791 DECLARE_COMPLETION_ONSTACK(wait);
793 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
795 rq->cmd_flags |= REQ_HIPRI;
796 rq->end_io_data = &wait;
797 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
799 while (!completion_done(&wait)) {
800 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
806 * Returns 0 on success. If the result is negative, it's a Linux error code;
807 * if the result is positive, it's an NVM Express status code
809 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
810 union nvme_result *result, void *buffer, unsigned bufflen,
811 unsigned timeout, int qid, int at_head,
812 blk_mq_req_flags_t flags, bool poll)
817 req = nvme_alloc_request(q, cmd, flags, qid);
821 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
823 if (buffer && bufflen) {
824 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
830 nvme_execute_rq_polled(req->q, NULL, req, at_head);
832 blk_execute_rq(req->q, NULL, req, at_head);
834 *result = nvme_req(req)->result;
835 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
838 ret = nvme_req(req)->status;
840 blk_mq_free_request(req);
843 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
845 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
846 void *buffer, unsigned bufflen)
848 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
849 NVME_QID_ANY, 0, 0, false);
851 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
853 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
854 unsigned len, u32 seed, bool write)
856 struct bio_integrity_payload *bip;
860 buf = kmalloc(len, GFP_KERNEL);
865 if (write && copy_from_user(buf, ubuf, len))
868 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
874 bip->bip_iter.bi_size = len;
875 bip->bip_iter.bi_sector = seed;
876 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
877 offset_in_page(buf));
887 static int nvme_submit_user_cmd(struct request_queue *q,
888 struct nvme_command *cmd, void __user *ubuffer,
889 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
890 u32 meta_seed, u64 *result, unsigned timeout)
892 bool write = nvme_is_write(cmd);
893 struct nvme_ns *ns = q->queuedata;
894 struct gendisk *disk = ns ? ns->disk : NULL;
896 struct bio *bio = NULL;
900 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
904 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
905 nvme_req(req)->flags |= NVME_REQ_USERCMD;
907 if (ubuffer && bufflen) {
908 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
914 if (disk && meta_buffer && meta_len) {
915 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
921 req->cmd_flags |= REQ_INTEGRITY;
925 blk_execute_rq(req->q, disk, req, 0);
926 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
929 ret = nvme_req(req)->status;
931 *result = le64_to_cpu(nvme_req(req)->result.u64);
932 if (meta && !ret && !write) {
933 if (copy_to_user(meta_buffer, meta, meta_len))
939 blk_rq_unmap_user(bio);
941 blk_mq_free_request(req);
945 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
947 struct nvme_ctrl *ctrl = rq->end_io_data;
949 bool startka = false;
951 blk_mq_free_request(rq);
954 dev_err(ctrl->device,
955 "failed nvme_keep_alive_end_io error=%d\n",
960 ctrl->comp_seen = false;
961 spin_lock_irqsave(&ctrl->lock, flags);
962 if (ctrl->state == NVME_CTRL_LIVE ||
963 ctrl->state == NVME_CTRL_CONNECTING)
965 spin_unlock_irqrestore(&ctrl->lock, flags);
967 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
970 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
974 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
979 rq->timeout = ctrl->kato * HZ;
980 rq->end_io_data = ctrl;
982 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
987 static void nvme_keep_alive_work(struct work_struct *work)
989 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
990 struct nvme_ctrl, ka_work);
991 bool comp_seen = ctrl->comp_seen;
993 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
994 dev_dbg(ctrl->device,
995 "reschedule traffic based keep-alive timer\n");
996 ctrl->comp_seen = false;
997 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1001 if (nvme_keep_alive(ctrl)) {
1002 /* allocation failure, reset the controller */
1003 dev_err(ctrl->device, "keep-alive failed\n");
1004 nvme_reset_ctrl(ctrl);
1009 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1011 if (unlikely(ctrl->kato == 0))
1014 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1017 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1019 if (unlikely(ctrl->kato == 0))
1022 cancel_delayed_work_sync(&ctrl->ka_work);
1024 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1027 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1028 * flag, thus sending any new CNS opcodes has a big chance of not working.
1029 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1030 * (but not for any later version).
1032 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1034 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1035 return ctrl->vs < NVME_VS(1, 2, 0);
1036 return ctrl->vs < NVME_VS(1, 1, 0);
1039 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1041 struct nvme_command c = { };
1044 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1045 c.identify.opcode = nvme_admin_identify;
1046 c.identify.cns = NVME_ID_CNS_CTRL;
1048 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1052 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1053 sizeof(struct nvme_id_ctrl));
1059 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1060 struct nvme_ns_id_desc *cur)
1062 const char *warn_str = "ctrl returned bogus length:";
1065 switch (cur->nidt) {
1066 case NVME_NIDT_EUI64:
1067 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1068 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1069 warn_str, cur->nidl);
1072 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1073 return NVME_NIDT_EUI64_LEN;
1074 case NVME_NIDT_NGUID:
1075 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1076 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1077 warn_str, cur->nidl);
1080 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1081 return NVME_NIDT_NGUID_LEN;
1082 case NVME_NIDT_UUID:
1083 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1084 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1085 warn_str, cur->nidl);
1088 uuid_copy(&ids->uuid, data + sizeof(*cur));
1089 return NVME_NIDT_UUID_LEN;
1091 /* Skip unknown types */
1096 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1097 struct nvme_ns_ids *ids)
1099 struct nvme_command c = { };
1105 c.identify.opcode = nvme_admin_identify;
1106 c.identify.nsid = cpu_to_le32(nsid);
1107 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1109 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1113 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1114 NVME_IDENTIFY_DATA_SIZE);
1116 dev_warn(ctrl->device,
1117 "Identify Descriptors failed (%d)\n", status);
1119 * Don't treat an error as fatal, as we potentially already
1120 * have a NGUID or EUI-64.
1122 if (status > 0 && !(status & NVME_SC_DNR))
1127 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1128 struct nvme_ns_id_desc *cur = data + pos;
1133 len = nvme_process_ns_desc(ctrl, ids, cur);
1137 len += sizeof(*cur);
1144 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1146 struct nvme_command c = { };
1148 c.identify.opcode = nvme_admin_identify;
1149 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1150 c.identify.nsid = cpu_to_le32(nsid);
1151 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1152 NVME_IDENTIFY_DATA_SIZE);
1155 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1156 unsigned nsid, struct nvme_id_ns **id)
1158 struct nvme_command c = { };
1161 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1162 c.identify.opcode = nvme_admin_identify;
1163 c.identify.nsid = cpu_to_le32(nsid);
1164 c.identify.cns = NVME_ID_CNS_NS;
1166 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1170 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1172 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1179 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1180 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1182 union nvme_result res = { 0 };
1183 struct nvme_command c;
1186 memset(&c, 0, sizeof(c));
1187 c.features.opcode = op;
1188 c.features.fid = cpu_to_le32(fid);
1189 c.features.dword11 = cpu_to_le32(dword11);
1191 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1192 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1193 if (ret >= 0 && result)
1194 *result = le32_to_cpu(res.u32);
1198 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1199 unsigned int dword11, void *buffer, size_t buflen,
1202 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1205 EXPORT_SYMBOL_GPL(nvme_set_features);
1207 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1208 unsigned int dword11, void *buffer, size_t buflen,
1211 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1214 EXPORT_SYMBOL_GPL(nvme_get_features);
1216 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1218 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1220 int status, nr_io_queues;
1222 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1228 * Degraded controllers might return an error when setting the queue
1229 * count. We still want to be able to bring them online and offer
1230 * access to the admin queue, as that might be only way to fix them up.
1233 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1236 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1237 *count = min(*count, nr_io_queues);
1242 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1244 #define NVME_AEN_SUPPORTED \
1245 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1246 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1248 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1250 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1253 if (!supported_aens)
1256 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1259 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1262 queue_work(nvme_wq, &ctrl->async_event_work);
1266 * Convert integer values from ioctl structures to user pointers, silently
1267 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1270 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1272 if (in_compat_syscall())
1273 ptrval = (compat_uptr_t)ptrval;
1274 return (void __user *)ptrval;
1277 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1279 struct nvme_user_io io;
1280 struct nvme_command c;
1281 unsigned length, meta_len;
1282 void __user *metadata;
1284 if (copy_from_user(&io, uio, sizeof(io)))
1289 switch (io.opcode) {
1290 case nvme_cmd_write:
1292 case nvme_cmd_compare:
1298 length = (io.nblocks + 1) << ns->lba_shift;
1299 meta_len = (io.nblocks + 1) * ns->ms;
1300 metadata = nvme_to_user_ptr(io.metadata);
1302 if (ns->features & NVME_NS_EXT_LBAS) {
1305 } else if (meta_len) {
1306 if ((io.metadata & 3) || !io.metadata)
1310 memset(&c, 0, sizeof(c));
1311 c.rw.opcode = io.opcode;
1312 c.rw.flags = io.flags;
1313 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1314 c.rw.slba = cpu_to_le64(io.slba);
1315 c.rw.length = cpu_to_le16(io.nblocks);
1316 c.rw.control = cpu_to_le16(io.control);
1317 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1318 c.rw.reftag = cpu_to_le32(io.reftag);
1319 c.rw.apptag = cpu_to_le16(io.apptag);
1320 c.rw.appmask = cpu_to_le16(io.appmask);
1322 return nvme_submit_user_cmd(ns->queue, &c,
1323 nvme_to_user_ptr(io.addr), length,
1324 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1327 static u32 nvme_known_admin_effects(u8 opcode)
1330 case nvme_admin_format_nvm:
1331 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1332 NVME_CMD_EFFECTS_CSE_MASK;
1333 case nvme_admin_sanitize_nvm:
1334 return NVME_CMD_EFFECTS_CSE_MASK;
1341 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1348 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1349 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1350 dev_warn(ctrl->device,
1351 "IO command:%02x has unhandled effects:%08x\n",
1357 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1358 effects |= nvme_known_admin_effects(opcode);
1361 * For simplicity, IO to all namespaces is quiesced even if the command
1362 * effects say only one namespace is affected.
1364 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1365 mutex_lock(&ctrl->scan_lock);
1366 mutex_lock(&ctrl->subsys->lock);
1367 nvme_mpath_start_freeze(ctrl->subsys);
1368 nvme_mpath_wait_freeze(ctrl->subsys);
1369 nvme_start_freeze(ctrl);
1370 nvme_wait_freeze(ctrl);
1375 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1379 down_read(&ctrl->namespaces_rwsem);
1380 list_for_each_entry(ns, &ctrl->namespaces, list)
1381 if (ns->disk && nvme_revalidate_disk(ns->disk))
1382 nvme_set_queue_dying(ns);
1383 up_read(&ctrl->namespaces_rwsem);
1386 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1389 * Revalidate LBA changes prior to unfreezing. This is necessary to
1390 * prevent memory corruption if a logical block size was changed by
1393 if (effects & NVME_CMD_EFFECTS_LBCC)
1394 nvme_update_formats(ctrl);
1395 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1396 nvme_unfreeze(ctrl);
1397 nvme_mpath_unfreeze(ctrl->subsys);
1398 mutex_unlock(&ctrl->subsys->lock);
1399 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1400 mutex_unlock(&ctrl->scan_lock);
1402 if (effects & NVME_CMD_EFFECTS_CCC)
1403 nvme_init_identify(ctrl);
1404 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1405 nvme_queue_scan(ctrl);
1406 flush_work(&ctrl->scan_work);
1410 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1411 struct nvme_passthru_cmd __user *ucmd)
1413 struct nvme_passthru_cmd cmd;
1414 struct nvme_command c;
1415 unsigned timeout = 0;
1420 if (!capable(CAP_SYS_ADMIN))
1422 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1427 memset(&c, 0, sizeof(c));
1428 c.common.opcode = cmd.opcode;
1429 c.common.flags = cmd.flags;
1430 c.common.nsid = cpu_to_le32(cmd.nsid);
1431 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1432 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1433 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1434 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1435 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1436 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1437 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1438 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1441 timeout = msecs_to_jiffies(cmd.timeout_ms);
1443 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1444 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1445 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1446 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1447 0, &result, timeout);
1448 nvme_passthru_end(ctrl, effects);
1451 if (put_user(result, &ucmd->result))
1458 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1459 struct nvme_passthru_cmd64 __user *ucmd)
1461 struct nvme_passthru_cmd64 cmd;
1462 struct nvme_command c;
1463 unsigned timeout = 0;
1467 if (!capable(CAP_SYS_ADMIN))
1469 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1474 memset(&c, 0, sizeof(c));
1475 c.common.opcode = cmd.opcode;
1476 c.common.flags = cmd.flags;
1477 c.common.nsid = cpu_to_le32(cmd.nsid);
1478 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1479 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1480 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1481 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1482 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1483 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1484 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1485 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1488 timeout = msecs_to_jiffies(cmd.timeout_ms);
1490 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1491 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1492 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1493 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1494 0, &cmd.result, timeout);
1495 nvme_passthru_end(ctrl, effects);
1498 if (put_user(cmd.result, &ucmd->result))
1506 * Issue ioctl requests on the first available path. Note that unlike normal
1507 * block layer requests we will not retry failed request on another controller.
1509 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1510 struct nvme_ns_head **head, int *srcu_idx)
1512 #ifdef CONFIG_NVME_MULTIPATH
1513 if (disk->fops == &nvme_ns_head_ops) {
1516 *head = disk->private_data;
1517 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1518 ns = nvme_find_path(*head);
1520 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1526 return disk->private_data;
1529 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1532 srcu_read_unlock(&head->srcu, idx);
1535 static bool is_ctrl_ioctl(unsigned int cmd)
1537 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1539 if (is_sed_ioctl(cmd))
1544 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1546 struct nvme_ns_head *head,
1549 struct nvme_ctrl *ctrl = ns->ctrl;
1552 nvme_get_ctrl(ns->ctrl);
1553 nvme_put_ns_from_disk(head, srcu_idx);
1556 case NVME_IOCTL_ADMIN_CMD:
1557 ret = nvme_user_cmd(ctrl, NULL, argp);
1559 case NVME_IOCTL_ADMIN64_CMD:
1560 ret = nvme_user_cmd64(ctrl, NULL, argp);
1563 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1566 nvme_put_ctrl(ctrl);
1570 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1571 unsigned int cmd, unsigned long arg)
1573 struct nvme_ns_head *head = NULL;
1574 void __user *argp = (void __user *)arg;
1578 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1580 return -EWOULDBLOCK;
1583 * Handle ioctls that apply to the controller instead of the namespace
1584 * seperately and drop the ns SRCU reference early. This avoids a
1585 * deadlock when deleting namespaces using the passthrough interface.
1587 if (is_ctrl_ioctl(cmd))
1588 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1592 force_successful_syscall_return();
1593 ret = ns->head->ns_id;
1595 case NVME_IOCTL_IO_CMD:
1596 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1598 case NVME_IOCTL_SUBMIT_IO:
1599 ret = nvme_submit_io(ns, argp);
1601 case NVME_IOCTL_IO64_CMD:
1602 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1606 ret = nvme_nvm_ioctl(ns, cmd, arg);
1611 nvme_put_ns_from_disk(head, srcu_idx);
1615 #ifdef CONFIG_COMPAT
1616 struct nvme_user_io32 {
1629 } __attribute__((__packed__));
1631 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1633 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1634 unsigned int cmd, unsigned long arg)
1637 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1638 * between 32 bit programs and 64 bit kernel.
1639 * The cause is that the results of sizeof(struct nvme_user_io),
1640 * which is used to define NVME_IOCTL_SUBMIT_IO,
1641 * are not same between 32 bit compiler and 64 bit compiler.
1642 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1643 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1644 * Other IOCTL numbers are same between 32 bit and 64 bit.
1645 * So there is nothing to do regarding to other IOCTL numbers.
1647 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1648 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1650 return nvme_ioctl(bdev, mode, cmd, arg);
1653 #define nvme_compat_ioctl NULL
1654 #endif /* CONFIG_COMPAT */
1656 static int nvme_open(struct block_device *bdev, fmode_t mode)
1658 struct nvme_ns *ns = bdev->bd_disk->private_data;
1660 #ifdef CONFIG_NVME_MULTIPATH
1661 /* should never be called due to GENHD_FL_HIDDEN */
1662 if (WARN_ON_ONCE(ns->head->disk))
1665 if (!kref_get_unless_zero(&ns->kref))
1667 if (!try_module_get(ns->ctrl->ops->module))
1678 static void nvme_release(struct gendisk *disk, fmode_t mode)
1680 struct nvme_ns *ns = disk->private_data;
1682 module_put(ns->ctrl->ops->module);
1686 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1688 /* some standard values */
1689 geo->heads = 1 << 6;
1690 geo->sectors = 1 << 5;
1691 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1695 #ifdef CONFIG_BLK_DEV_INTEGRITY
1696 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1697 u32 max_integrity_segments)
1699 struct blk_integrity integrity;
1701 memset(&integrity, 0, sizeof(integrity));
1703 case NVME_NS_DPS_PI_TYPE3:
1704 integrity.profile = &t10_pi_type3_crc;
1705 integrity.tag_size = sizeof(u16) + sizeof(u32);
1706 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1708 case NVME_NS_DPS_PI_TYPE1:
1709 case NVME_NS_DPS_PI_TYPE2:
1710 integrity.profile = &t10_pi_type1_crc;
1711 integrity.tag_size = sizeof(u16);
1712 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1715 integrity.profile = NULL;
1718 integrity.tuple_size = ms;
1719 blk_integrity_register(disk, &integrity);
1720 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1723 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1724 u32 max_integrity_segments)
1727 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1729 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1731 struct nvme_ctrl *ctrl = ns->ctrl;
1732 struct request_queue *queue = disk->queue;
1733 u32 size = queue_logical_block_size(queue);
1735 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1736 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1740 if (ctrl->nr_streams && ns->sws && ns->sgs)
1741 size *= ns->sws * ns->sgs;
1743 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1744 NVME_DSM_MAX_RANGES);
1746 queue->limits.discard_alignment = 0;
1747 queue->limits.discard_granularity = size;
1749 /* If discard is already enabled, don't reset queue limits */
1750 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1753 blk_queue_max_discard_sectors(queue, UINT_MAX);
1754 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1756 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1757 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1760 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1764 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1765 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1768 * Even though NVMe spec explicitly states that MDTS is not
1769 * applicable to the write-zeroes:- "The restriction does not apply to
1770 * commands that do not transfer data between the host and the
1771 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1772 * In order to be more cautious use controller's max_hw_sectors value
1773 * to configure the maximum sectors for the write-zeroes which is
1774 * configured based on the controller's MDTS field in the
1775 * nvme_init_identify() if available.
1777 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1778 max_blocks = (u64)USHRT_MAX + 1;
1780 max_blocks = ns->ctrl->max_hw_sectors + 1;
1782 blk_queue_max_write_zeroes_sectors(disk->queue,
1783 nvme_lba_to_sect(ns, max_blocks));
1786 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1787 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1789 memset(ids, 0, sizeof(*ids));
1791 if (ctrl->vs >= NVME_VS(1, 1, 0))
1792 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1793 if (ctrl->vs >= NVME_VS(1, 2, 0))
1794 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1795 if (ctrl->vs >= NVME_VS(1, 3, 0))
1796 return nvme_identify_ns_descs(ctrl, nsid, ids);
1800 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1802 return !uuid_is_null(&ids->uuid) ||
1803 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1804 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1807 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1809 return uuid_equal(&a->uuid, &b->uuid) &&
1810 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1811 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1814 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1815 u32 *phys_bs, u32 *io_opt)
1817 struct streams_directive_params s;
1820 if (!ctrl->nr_streams)
1823 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1827 ns->sws = le32_to_cpu(s.sws);
1828 ns->sgs = le16_to_cpu(s.sgs);
1831 *phys_bs = ns->sws * (1 << ns->lba_shift);
1833 *io_opt = *phys_bs * ns->sgs;
1839 static void nvme_update_disk_info(struct gendisk *disk,
1840 struct nvme_ns *ns, struct nvme_id_ns *id)
1842 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1843 unsigned short bs = 1 << ns->lba_shift;
1844 u32 atomic_bs, phys_bs, io_opt = 0;
1846 if (ns->lba_shift > PAGE_SHIFT) {
1847 /* unsupported block size, set capacity to 0 later */
1850 blk_mq_freeze_queue(disk->queue);
1851 blk_integrity_unregister(disk);
1853 atomic_bs = phys_bs = bs;
1854 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1855 if (id->nabo == 0) {
1857 * Bit 1 indicates whether NAWUPF is defined for this namespace
1858 * and whether it should be used instead of AWUPF. If NAWUPF ==
1859 * 0 then AWUPF must be used instead.
1861 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1862 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1864 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1867 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1868 /* NPWG = Namespace Preferred Write Granularity */
1869 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1870 /* NOWS = Namespace Optimal Write Size */
1871 io_opt = bs * (1 + le16_to_cpu(id->nows));
1874 blk_queue_logical_block_size(disk->queue, bs);
1876 * Linux filesystems assume writing a single physical block is
1877 * an atomic operation. Hence limit the physical block size to the
1878 * value of the Atomic Write Unit Power Fail parameter.
1880 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1881 blk_queue_io_min(disk->queue, phys_bs);
1882 blk_queue_io_opt(disk->queue, io_opt);
1885 * The block layer can't support LBA sizes larger than the page size
1886 * yet, so catch this early and don't allow block I/O.
1888 if (ns->lba_shift > PAGE_SHIFT)
1892 * Register a metadata profile for PI, or the plain non-integrity NVMe
1893 * metadata masquerading as Type 0 if supported, otherwise reject block
1894 * I/O to namespaces with metadata except when the namespace supports
1895 * PI, as it can strip/insert in that case.
1898 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1899 (ns->features & NVME_NS_METADATA_SUPPORTED))
1900 nvme_init_integrity(disk, ns->ms, ns->pi_type,
1901 ns->ctrl->max_integrity_segments);
1902 else if (!nvme_ns_has_pi(ns))
1906 set_capacity_revalidate_and_notify(disk, capacity, false);
1908 nvme_config_discard(disk, ns);
1909 nvme_config_write_zeroes(disk, ns);
1911 if (id->nsattr & NVME_NS_ATTR_RO)
1912 set_disk_ro(disk, true);
1914 set_disk_ro(disk, false);
1916 blk_mq_unfreeze_queue(disk->queue);
1919 static int __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1921 struct nvme_ns *ns = disk->private_data;
1922 struct nvme_ctrl *ctrl = ns->ctrl;
1926 * If identify namespace failed, use default 512 byte block size so
1927 * block layer can use before failing read/write for 0 capacity.
1929 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1930 if (ns->lba_shift == 0)
1933 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1934 is_power_of_2(ctrl->max_hw_sectors))
1935 iob = ctrl->max_hw_sectors;
1937 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1940 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1941 /* the PI implementation requires metadata equal t10 pi tuple size */
1942 if (ns->ms == sizeof(struct t10_pi_tuple))
1943 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1949 * For PCIe only the separate metadata pointer is supported,
1950 * as the block layer supplies metadata in a separate bio_vec
1951 * chain. For Fabrics, only metadata as part of extended data
1952 * LBA is supported on the wire per the Fabrics specification,
1953 * but the HBA/HCA will do the remapping from the separate
1954 * metadata buffers for us.
1956 if (id->flbas & NVME_NS_FLBAS_META_EXT) {
1957 ns->features |= NVME_NS_EXT_LBAS;
1958 if ((ctrl->ops->flags & NVME_F_FABRICS) &&
1959 (ctrl->ops->flags & NVME_F_METADATA_SUPPORTED) &&
1960 ctrl->max_integrity_segments)
1961 ns->features |= NVME_NS_METADATA_SUPPORTED;
1963 if (WARN_ON_ONCE(ctrl->ops->flags & NVME_F_FABRICS))
1965 if (ctrl->ops->flags & NVME_F_METADATA_SUPPORTED)
1966 ns->features |= NVME_NS_METADATA_SUPPORTED;
1971 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(iob));
1972 nvme_update_disk_info(disk, ns, id);
1973 #ifdef CONFIG_NVME_MULTIPATH
1974 if (ns->head->disk) {
1975 nvme_update_disk_info(ns->head->disk, ns, id);
1976 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1977 revalidate_disk(ns->head->disk);
1983 static int nvme_revalidate_disk(struct gendisk *disk)
1985 struct nvme_ns *ns = disk->private_data;
1986 struct nvme_ctrl *ctrl = ns->ctrl;
1987 struct nvme_id_ns *id;
1988 struct nvme_ns_ids ids;
1991 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1992 set_capacity(disk, 0);
1996 ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
2000 if (id->ncap == 0) {
2005 ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
2009 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
2010 dev_err(ctrl->device,
2011 "identifiers changed for nsid %d\n", ns->head->ns_id);
2016 ret = __nvme_revalidate_disk(disk, id);
2021 * Only fail the function if we got a fatal error back from the
2022 * device, otherwise ignore the error and just move on.
2024 if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
2027 ret = blk_status_to_errno(nvme_error_status(ret));
2031 static char nvme_pr_type(enum pr_type type)
2034 case PR_WRITE_EXCLUSIVE:
2036 case PR_EXCLUSIVE_ACCESS:
2038 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2040 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2042 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2044 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2051 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2052 u64 key, u64 sa_key, u8 op)
2054 struct nvme_ns_head *head = NULL;
2056 struct nvme_command c;
2058 u8 data[16] = { 0, };
2060 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2062 return -EWOULDBLOCK;
2064 put_unaligned_le64(key, &data[0]);
2065 put_unaligned_le64(sa_key, &data[8]);
2067 memset(&c, 0, sizeof(c));
2068 c.common.opcode = op;
2069 c.common.nsid = cpu_to_le32(ns->head->ns_id);
2070 c.common.cdw10 = cpu_to_le32(cdw10);
2072 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2073 nvme_put_ns_from_disk(head, srcu_idx);
2077 static int nvme_pr_register(struct block_device *bdev, u64 old,
2078 u64 new, unsigned flags)
2082 if (flags & ~PR_FL_IGNORE_KEY)
2085 cdw10 = old ? 2 : 0;
2086 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2087 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2088 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2091 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2092 enum pr_type type, unsigned flags)
2096 if (flags & ~PR_FL_IGNORE_KEY)
2099 cdw10 = nvme_pr_type(type) << 8;
2100 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2101 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2104 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2105 enum pr_type type, bool abort)
2107 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2108 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2111 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2113 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2114 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2117 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2119 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2120 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2123 static const struct pr_ops nvme_pr_ops = {
2124 .pr_register = nvme_pr_register,
2125 .pr_reserve = nvme_pr_reserve,
2126 .pr_release = nvme_pr_release,
2127 .pr_preempt = nvme_pr_preempt,
2128 .pr_clear = nvme_pr_clear,
2131 #ifdef CONFIG_BLK_SED_OPAL
2132 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2135 struct nvme_ctrl *ctrl = data;
2136 struct nvme_command cmd;
2138 memset(&cmd, 0, sizeof(cmd));
2140 cmd.common.opcode = nvme_admin_security_send;
2142 cmd.common.opcode = nvme_admin_security_recv;
2143 cmd.common.nsid = 0;
2144 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2145 cmd.common.cdw11 = cpu_to_le32(len);
2147 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2148 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2150 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2151 #endif /* CONFIG_BLK_SED_OPAL */
2153 static const struct block_device_operations nvme_fops = {
2154 .owner = THIS_MODULE,
2155 .ioctl = nvme_ioctl,
2156 .compat_ioctl = nvme_compat_ioctl,
2158 .release = nvme_release,
2159 .getgeo = nvme_getgeo,
2160 .revalidate_disk= nvme_revalidate_disk,
2161 .pr_ops = &nvme_pr_ops,
2164 #ifdef CONFIG_NVME_MULTIPATH
2165 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2167 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2169 if (!kref_get_unless_zero(&head->ref))
2174 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2176 nvme_put_ns_head(disk->private_data);
2179 const struct block_device_operations nvme_ns_head_ops = {
2180 .owner = THIS_MODULE,
2181 .submit_bio = nvme_ns_head_submit_bio,
2182 .open = nvme_ns_head_open,
2183 .release = nvme_ns_head_release,
2184 .ioctl = nvme_ioctl,
2185 .compat_ioctl = nvme_compat_ioctl,
2186 .getgeo = nvme_getgeo,
2187 .pr_ops = &nvme_pr_ops,
2189 #endif /* CONFIG_NVME_MULTIPATH */
2191 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2193 unsigned long timeout =
2194 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2195 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2198 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2201 if ((csts & NVME_CSTS_RDY) == bit)
2204 usleep_range(1000, 2000);
2205 if (fatal_signal_pending(current))
2207 if (time_after(jiffies, timeout)) {
2208 dev_err(ctrl->device,
2209 "Device not ready; aborting %s, CSTS=0x%x\n",
2210 enabled ? "initialisation" : "reset", csts);
2219 * If the device has been passed off to us in an enabled state, just clear
2220 * the enabled bit. The spec says we should set the 'shutdown notification
2221 * bits', but doing so may cause the device to complete commands to the
2222 * admin queue ... and we don't know what memory that might be pointing at!
2224 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2228 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2229 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2231 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2235 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2236 msleep(NVME_QUIRK_DELAY_AMOUNT);
2238 return nvme_wait_ready(ctrl, ctrl->cap, false);
2240 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2242 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2245 * Default to a 4K page size, with the intention to update this
2246 * path in the future to accomodate architectures with differing
2247 * kernel and IO page sizes.
2249 unsigned dev_page_min, page_shift = 12;
2252 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2254 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2257 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2259 if (page_shift < dev_page_min) {
2260 dev_err(ctrl->device,
2261 "Minimum device page size %u too large for host (%u)\n",
2262 1 << dev_page_min, 1 << page_shift);
2266 ctrl->page_size = 1 << page_shift;
2268 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2269 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2270 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2271 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2272 ctrl->ctrl_config |= NVME_CC_ENABLE;
2274 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2277 return nvme_wait_ready(ctrl, ctrl->cap, true);
2279 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2281 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2283 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2287 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2288 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2290 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2294 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2295 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2299 if (fatal_signal_pending(current))
2301 if (time_after(jiffies, timeout)) {
2302 dev_err(ctrl->device,
2303 "Device shutdown incomplete; abort shutdown\n");
2310 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2312 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2313 struct request_queue *q)
2317 if (ctrl->max_hw_sectors) {
2319 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2321 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2322 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2323 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2325 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2326 blk_queue_dma_alignment(q, 7);
2327 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2329 blk_queue_write_cache(q, vwc, vwc);
2332 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2337 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2340 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2341 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2344 dev_warn_once(ctrl->device,
2345 "could not set timestamp (%d)\n", ret);
2349 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2351 struct nvme_feat_host_behavior *host;
2354 /* Don't bother enabling the feature if retry delay is not reported */
2358 host = kzalloc(sizeof(*host), GFP_KERNEL);
2362 host->acre = NVME_ENABLE_ACRE;
2363 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2364 host, sizeof(*host), NULL);
2369 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2372 * APST (Autonomous Power State Transition) lets us program a
2373 * table of power state transitions that the controller will
2374 * perform automatically. We configure it with a simple
2375 * heuristic: we are willing to spend at most 2% of the time
2376 * transitioning between power states. Therefore, when running
2377 * in any given state, we will enter the next lower-power
2378 * non-operational state after waiting 50 * (enlat + exlat)
2379 * microseconds, as long as that state's exit latency is under
2380 * the requested maximum latency.
2382 * We will not autonomously enter any non-operational state for
2383 * which the total latency exceeds ps_max_latency_us. Users
2384 * can set ps_max_latency_us to zero to turn off APST.
2388 struct nvme_feat_auto_pst *table;
2394 * If APST isn't supported or if we haven't been initialized yet,
2395 * then don't do anything.
2400 if (ctrl->npss > 31) {
2401 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2405 table = kzalloc(sizeof(*table), GFP_KERNEL);
2409 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2410 /* Turn off APST. */
2412 dev_dbg(ctrl->device, "APST disabled\n");
2414 __le64 target = cpu_to_le64(0);
2418 * Walk through all states from lowest- to highest-power.
2419 * According to the spec, lower-numbered states use more
2420 * power. NPSS, despite the name, is the index of the
2421 * lowest-power state, not the number of states.
2423 for (state = (int)ctrl->npss; state >= 0; state--) {
2424 u64 total_latency_us, exit_latency_us, transition_ms;
2427 table->entries[state] = target;
2430 * Don't allow transitions to the deepest state
2431 * if it's quirked off.
2433 if (state == ctrl->npss &&
2434 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2438 * Is this state a useful non-operational state for
2439 * higher-power states to autonomously transition to?
2441 if (!(ctrl->psd[state].flags &
2442 NVME_PS_FLAGS_NON_OP_STATE))
2446 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2447 if (exit_latency_us > ctrl->ps_max_latency_us)
2452 le32_to_cpu(ctrl->psd[state].entry_lat);
2455 * This state is good. Use it as the APST idle
2456 * target for higher power states.
2458 transition_ms = total_latency_us + 19;
2459 do_div(transition_ms, 20);
2460 if (transition_ms > (1 << 24) - 1)
2461 transition_ms = (1 << 24) - 1;
2463 target = cpu_to_le64((state << 3) |
2464 (transition_ms << 8));
2469 if (total_latency_us > max_lat_us)
2470 max_lat_us = total_latency_us;
2476 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2478 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2479 max_ps, max_lat_us, (int)sizeof(*table), table);
2483 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2484 table, sizeof(*table), NULL);
2486 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2492 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2494 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2498 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2499 case PM_QOS_LATENCY_ANY:
2507 if (ctrl->ps_max_latency_us != latency) {
2508 ctrl->ps_max_latency_us = latency;
2509 nvme_configure_apst(ctrl);
2513 struct nvme_core_quirk_entry {
2515 * NVMe model and firmware strings are padded with spaces. For
2516 * simplicity, strings in the quirk table are padded with NULLs
2522 unsigned long quirks;
2525 static const struct nvme_core_quirk_entry core_quirks[] = {
2528 * This Toshiba device seems to die using any APST states. See:
2529 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2532 .mn = "THNSF5256GPUK TOSHIBA",
2533 .quirks = NVME_QUIRK_NO_APST,
2537 * This LiteON CL1-3D*-Q11 firmware version has a race
2538 * condition associated with actions related to suspend to idle
2539 * LiteON has resolved the problem in future firmware
2543 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2547 /* match is null-terminated but idstr is space-padded. */
2548 static bool string_matches(const char *idstr, const char *match, size_t len)
2555 matchlen = strlen(match);
2556 WARN_ON_ONCE(matchlen > len);
2558 if (memcmp(idstr, match, matchlen))
2561 for (; matchlen < len; matchlen++)
2562 if (idstr[matchlen] != ' ')
2568 static bool quirk_matches(const struct nvme_id_ctrl *id,
2569 const struct nvme_core_quirk_entry *q)
2571 return q->vid == le16_to_cpu(id->vid) &&
2572 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2573 string_matches(id->fr, q->fr, sizeof(id->fr));
2576 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2577 struct nvme_id_ctrl *id)
2582 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2583 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2584 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2585 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2589 if (ctrl->vs >= NVME_VS(1, 2, 1))
2590 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2593 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2594 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2595 "nqn.2014.08.org.nvmexpress:%04x%04x",
2596 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2597 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2598 off += sizeof(id->sn);
2599 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2600 off += sizeof(id->mn);
2601 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2604 static void nvme_release_subsystem(struct device *dev)
2606 struct nvme_subsystem *subsys =
2607 container_of(dev, struct nvme_subsystem, dev);
2609 if (subsys->instance >= 0)
2610 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2614 static void nvme_destroy_subsystem(struct kref *ref)
2616 struct nvme_subsystem *subsys =
2617 container_of(ref, struct nvme_subsystem, ref);
2619 mutex_lock(&nvme_subsystems_lock);
2620 list_del(&subsys->entry);
2621 mutex_unlock(&nvme_subsystems_lock);
2623 ida_destroy(&subsys->ns_ida);
2624 device_del(&subsys->dev);
2625 put_device(&subsys->dev);
2628 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2630 kref_put(&subsys->ref, nvme_destroy_subsystem);
2633 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2635 struct nvme_subsystem *subsys;
2637 lockdep_assert_held(&nvme_subsystems_lock);
2640 * Fail matches for discovery subsystems. This results
2641 * in each discovery controller bound to a unique subsystem.
2642 * This avoids issues with validating controller values
2643 * that can only be true when there is a single unique subsystem.
2644 * There may be multiple and completely independent entities
2645 * that provide discovery controllers.
2647 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2650 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2651 if (strcmp(subsys->subnqn, subsysnqn))
2653 if (!kref_get_unless_zero(&subsys->ref))
2661 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2662 struct device_attribute subsys_attr_##_name = \
2663 __ATTR(_name, _mode, _show, NULL)
2665 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2666 struct device_attribute *attr,
2669 struct nvme_subsystem *subsys =
2670 container_of(dev, struct nvme_subsystem, dev);
2672 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2674 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2676 #define nvme_subsys_show_str_function(field) \
2677 static ssize_t subsys_##field##_show(struct device *dev, \
2678 struct device_attribute *attr, char *buf) \
2680 struct nvme_subsystem *subsys = \
2681 container_of(dev, struct nvme_subsystem, dev); \
2682 return sprintf(buf, "%.*s\n", \
2683 (int)sizeof(subsys->field), subsys->field); \
2685 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2687 nvme_subsys_show_str_function(model);
2688 nvme_subsys_show_str_function(serial);
2689 nvme_subsys_show_str_function(firmware_rev);
2691 static struct attribute *nvme_subsys_attrs[] = {
2692 &subsys_attr_model.attr,
2693 &subsys_attr_serial.attr,
2694 &subsys_attr_firmware_rev.attr,
2695 &subsys_attr_subsysnqn.attr,
2696 #ifdef CONFIG_NVME_MULTIPATH
2697 &subsys_attr_iopolicy.attr,
2702 static struct attribute_group nvme_subsys_attrs_group = {
2703 .attrs = nvme_subsys_attrs,
2706 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2707 &nvme_subsys_attrs_group,
2711 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2712 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2714 struct nvme_ctrl *tmp;
2716 lockdep_assert_held(&nvme_subsystems_lock);
2718 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2719 if (nvme_state_terminal(tmp))
2722 if (tmp->cntlid == ctrl->cntlid) {
2723 dev_err(ctrl->device,
2724 "Duplicate cntlid %u with %s, rejecting\n",
2725 ctrl->cntlid, dev_name(tmp->device));
2729 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2730 (ctrl->opts && ctrl->opts->discovery_nqn))
2733 dev_err(ctrl->device,
2734 "Subsystem does not support multiple controllers\n");
2741 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2743 struct nvme_subsystem *subsys, *found;
2746 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2750 subsys->instance = -1;
2751 mutex_init(&subsys->lock);
2752 kref_init(&subsys->ref);
2753 INIT_LIST_HEAD(&subsys->ctrls);
2754 INIT_LIST_HEAD(&subsys->nsheads);
2755 nvme_init_subnqn(subsys, ctrl, id);
2756 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2757 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2758 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2759 subsys->vendor_id = le16_to_cpu(id->vid);
2760 subsys->cmic = id->cmic;
2761 subsys->awupf = le16_to_cpu(id->awupf);
2762 #ifdef CONFIG_NVME_MULTIPATH
2763 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2766 subsys->dev.class = nvme_subsys_class;
2767 subsys->dev.release = nvme_release_subsystem;
2768 subsys->dev.groups = nvme_subsys_attrs_groups;
2769 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2770 device_initialize(&subsys->dev);
2772 mutex_lock(&nvme_subsystems_lock);
2773 found = __nvme_find_get_subsystem(subsys->subnqn);
2775 put_device(&subsys->dev);
2778 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2780 goto out_put_subsystem;
2783 ret = device_add(&subsys->dev);
2785 dev_err(ctrl->device,
2786 "failed to register subsystem device.\n");
2787 put_device(&subsys->dev);
2790 ida_init(&subsys->ns_ida);
2791 list_add_tail(&subsys->entry, &nvme_subsystems);
2794 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2795 dev_name(ctrl->device));
2797 dev_err(ctrl->device,
2798 "failed to create sysfs link from subsystem.\n");
2799 goto out_put_subsystem;
2803 subsys->instance = ctrl->instance;
2804 ctrl->subsys = subsys;
2805 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2806 mutex_unlock(&nvme_subsystems_lock);
2810 nvme_put_subsystem(subsys);
2812 mutex_unlock(&nvme_subsystems_lock);
2816 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2817 void *log, size_t size, u64 offset)
2819 struct nvme_command c = { };
2820 u32 dwlen = nvme_bytes_to_numd(size);
2822 c.get_log_page.opcode = nvme_admin_get_log_page;
2823 c.get_log_page.nsid = cpu_to_le32(nsid);
2824 c.get_log_page.lid = log_page;
2825 c.get_log_page.lsp = lsp;
2826 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2827 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2828 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2829 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2831 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2834 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2839 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2844 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2845 ctrl->effects, sizeof(*ctrl->effects), 0);
2847 kfree(ctrl->effects);
2848 ctrl->effects = NULL;
2854 * Initialize the cached copies of the Identify data and various controller
2855 * register in our nvme_ctrl structure. This should be called as soon as
2856 * the admin queue is fully up and running.
2858 int nvme_init_identify(struct nvme_ctrl *ctrl)
2860 struct nvme_id_ctrl *id;
2861 int ret, page_shift;
2863 bool prev_apst_enabled;
2865 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2867 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2870 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2871 ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2873 if (ctrl->vs >= NVME_VS(1, 1, 0))
2874 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2876 ret = nvme_identify_ctrl(ctrl, &id);
2878 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2882 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2883 ret = nvme_get_effects_log(ctrl);
2888 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2889 ctrl->cntlid = le16_to_cpu(id->cntlid);
2891 if (!ctrl->identified) {
2894 ret = nvme_init_subsystem(ctrl, id);
2899 * Check for quirks. Quirk can depend on firmware version,
2900 * so, in principle, the set of quirks present can change
2901 * across a reset. As a possible future enhancement, we
2902 * could re-scan for quirks every time we reinitialize
2903 * the device, but we'd have to make sure that the driver
2904 * behaves intelligently if the quirks change.
2906 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2907 if (quirk_matches(id, &core_quirks[i]))
2908 ctrl->quirks |= core_quirks[i].quirks;
2912 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2913 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2914 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2917 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2918 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2919 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2921 ctrl->oacs = le16_to_cpu(id->oacs);
2922 ctrl->oncs = le16_to_cpu(id->oncs);
2923 ctrl->mtfa = le16_to_cpu(id->mtfa);
2924 ctrl->oaes = le32_to_cpu(id->oaes);
2925 ctrl->wctemp = le16_to_cpu(id->wctemp);
2926 ctrl->cctemp = le16_to_cpu(id->cctemp);
2928 atomic_set(&ctrl->abort_limit, id->acl + 1);
2929 ctrl->vwc = id->vwc;
2931 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2933 max_hw_sectors = UINT_MAX;
2934 ctrl->max_hw_sectors =
2935 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2937 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2938 ctrl->sgls = le32_to_cpu(id->sgls);
2939 ctrl->kas = le16_to_cpu(id->kas);
2940 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2941 ctrl->ctratt = le32_to_cpu(id->ctratt);
2945 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2947 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2948 shutdown_timeout, 60);
2950 if (ctrl->shutdown_timeout != shutdown_timeout)
2951 dev_info(ctrl->device,
2952 "Shutdown timeout set to %u seconds\n",
2953 ctrl->shutdown_timeout);
2955 ctrl->shutdown_timeout = shutdown_timeout;
2957 ctrl->npss = id->npss;
2958 ctrl->apsta = id->apsta;
2959 prev_apst_enabled = ctrl->apst_enabled;
2960 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2961 if (force_apst && id->apsta) {
2962 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2963 ctrl->apst_enabled = true;
2965 ctrl->apst_enabled = false;
2968 ctrl->apst_enabled = id->apsta;
2970 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2972 if (ctrl->ops->flags & NVME_F_FABRICS) {
2973 ctrl->icdoff = le16_to_cpu(id->icdoff);
2974 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2975 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2976 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2979 * In fabrics we need to verify the cntlid matches the
2982 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2983 dev_err(ctrl->device,
2984 "Mismatching cntlid: Connect %u vs Identify "
2986 ctrl->cntlid, le16_to_cpu(id->cntlid));
2991 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2992 dev_err(ctrl->device,
2993 "keep-alive support is mandatory for fabrics\n");
2998 ctrl->hmpre = le32_to_cpu(id->hmpre);
2999 ctrl->hmmin = le32_to_cpu(id->hmmin);
3000 ctrl->hmminds = le32_to_cpu(id->hmminds);
3001 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3004 ret = nvme_mpath_init(ctrl, id);
3010 if (ctrl->apst_enabled && !prev_apst_enabled)
3011 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3012 else if (!ctrl->apst_enabled && prev_apst_enabled)
3013 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3015 ret = nvme_configure_apst(ctrl);
3019 ret = nvme_configure_timestamp(ctrl);
3023 ret = nvme_configure_directives(ctrl);
3027 ret = nvme_configure_acre(ctrl);
3031 if (!ctrl->identified)
3032 nvme_hwmon_init(ctrl);
3034 ctrl->identified = true;
3042 EXPORT_SYMBOL_GPL(nvme_init_identify);
3044 static int nvme_dev_open(struct inode *inode, struct file *file)
3046 struct nvme_ctrl *ctrl =
3047 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3049 switch (ctrl->state) {
3050 case NVME_CTRL_LIVE:
3053 return -EWOULDBLOCK;
3056 file->private_data = ctrl;
3060 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3065 down_read(&ctrl->namespaces_rwsem);
3066 if (list_empty(&ctrl->namespaces)) {
3071 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3072 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3073 dev_warn(ctrl->device,
3074 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3079 dev_warn(ctrl->device,
3080 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3081 kref_get(&ns->kref);
3082 up_read(&ctrl->namespaces_rwsem);
3084 ret = nvme_user_cmd(ctrl, ns, argp);
3089 up_read(&ctrl->namespaces_rwsem);
3093 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3096 struct nvme_ctrl *ctrl = file->private_data;
3097 void __user *argp = (void __user *)arg;
3100 case NVME_IOCTL_ADMIN_CMD:
3101 return nvme_user_cmd(ctrl, NULL, argp);
3102 case NVME_IOCTL_ADMIN64_CMD:
3103 return nvme_user_cmd64(ctrl, NULL, argp);
3104 case NVME_IOCTL_IO_CMD:
3105 return nvme_dev_user_cmd(ctrl, argp);
3106 case NVME_IOCTL_RESET:
3107 dev_warn(ctrl->device, "resetting controller\n");
3108 return nvme_reset_ctrl_sync(ctrl);
3109 case NVME_IOCTL_SUBSYS_RESET:
3110 return nvme_reset_subsystem(ctrl);
3111 case NVME_IOCTL_RESCAN:
3112 nvme_queue_scan(ctrl);
3119 static const struct file_operations nvme_dev_fops = {
3120 .owner = THIS_MODULE,
3121 .open = nvme_dev_open,
3122 .unlocked_ioctl = nvme_dev_ioctl,
3123 .compat_ioctl = compat_ptr_ioctl,
3126 static ssize_t nvme_sysfs_reset(struct device *dev,
3127 struct device_attribute *attr, const char *buf,
3130 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3133 ret = nvme_reset_ctrl_sync(ctrl);
3138 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3140 static ssize_t nvme_sysfs_rescan(struct device *dev,
3141 struct device_attribute *attr, const char *buf,
3144 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3146 nvme_queue_scan(ctrl);
3149 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3151 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3153 struct gendisk *disk = dev_to_disk(dev);
3155 if (disk->fops == &nvme_fops)
3156 return nvme_get_ns_from_dev(dev)->head;
3158 return disk->private_data;
3161 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3164 struct nvme_ns_head *head = dev_to_ns_head(dev);
3165 struct nvme_ns_ids *ids = &head->ids;
3166 struct nvme_subsystem *subsys = head->subsys;
3167 int serial_len = sizeof(subsys->serial);
3168 int model_len = sizeof(subsys->model);
3170 if (!uuid_is_null(&ids->uuid))
3171 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3173 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3174 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3176 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3177 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3179 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3180 subsys->serial[serial_len - 1] == '\0'))
3182 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3183 subsys->model[model_len - 1] == '\0'))
3186 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3187 serial_len, subsys->serial, model_len, subsys->model,
3190 static DEVICE_ATTR_RO(wwid);
3192 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3195 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3197 static DEVICE_ATTR_RO(nguid);
3199 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3202 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3204 /* For backward compatibility expose the NGUID to userspace if
3205 * we have no UUID set
3207 if (uuid_is_null(&ids->uuid)) {
3208 printk_ratelimited(KERN_WARNING
3209 "No UUID available providing old NGUID\n");
3210 return sprintf(buf, "%pU\n", ids->nguid);
3212 return sprintf(buf, "%pU\n", &ids->uuid);
3214 static DEVICE_ATTR_RO(uuid);
3216 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3219 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3221 static DEVICE_ATTR_RO(eui);
3223 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3226 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3228 static DEVICE_ATTR_RO(nsid);
3230 static struct attribute *nvme_ns_id_attrs[] = {
3231 &dev_attr_wwid.attr,
3232 &dev_attr_uuid.attr,
3233 &dev_attr_nguid.attr,
3235 &dev_attr_nsid.attr,
3236 #ifdef CONFIG_NVME_MULTIPATH
3237 &dev_attr_ana_grpid.attr,
3238 &dev_attr_ana_state.attr,
3243 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3244 struct attribute *a, int n)
3246 struct device *dev = container_of(kobj, struct device, kobj);
3247 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3249 if (a == &dev_attr_uuid.attr) {
3250 if (uuid_is_null(&ids->uuid) &&
3251 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3254 if (a == &dev_attr_nguid.attr) {
3255 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3258 if (a == &dev_attr_eui.attr) {
3259 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3262 #ifdef CONFIG_NVME_MULTIPATH
3263 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3264 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3266 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3273 static const struct attribute_group nvme_ns_id_attr_group = {
3274 .attrs = nvme_ns_id_attrs,
3275 .is_visible = nvme_ns_id_attrs_are_visible,
3278 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3279 &nvme_ns_id_attr_group,
3281 &nvme_nvm_attr_group,
3286 #define nvme_show_str_function(field) \
3287 static ssize_t field##_show(struct device *dev, \
3288 struct device_attribute *attr, char *buf) \
3290 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3291 return sprintf(buf, "%.*s\n", \
3292 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3294 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3296 nvme_show_str_function(model);
3297 nvme_show_str_function(serial);
3298 nvme_show_str_function(firmware_rev);
3300 #define nvme_show_int_function(field) \
3301 static ssize_t field##_show(struct device *dev, \
3302 struct device_attribute *attr, char *buf) \
3304 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3305 return sprintf(buf, "%d\n", ctrl->field); \
3307 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3309 nvme_show_int_function(cntlid);
3310 nvme_show_int_function(numa_node);
3311 nvme_show_int_function(queue_count);
3312 nvme_show_int_function(sqsize);
3314 static ssize_t nvme_sysfs_delete(struct device *dev,
3315 struct device_attribute *attr, const char *buf,
3318 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3320 /* Can't delete non-created controllers */
3324 if (device_remove_file_self(dev, attr))
3325 nvme_delete_ctrl_sync(ctrl);
3328 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3330 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3331 struct device_attribute *attr,
3334 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3336 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3338 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3340 static ssize_t nvme_sysfs_show_state(struct device *dev,
3341 struct device_attribute *attr,
3344 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3345 static const char *const state_name[] = {
3346 [NVME_CTRL_NEW] = "new",
3347 [NVME_CTRL_LIVE] = "live",
3348 [NVME_CTRL_RESETTING] = "resetting",
3349 [NVME_CTRL_CONNECTING] = "connecting",
3350 [NVME_CTRL_DELETING] = "deleting",
3351 [NVME_CTRL_DEAD] = "dead",
3354 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3355 state_name[ctrl->state])
3356 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3358 return sprintf(buf, "unknown state\n");
3361 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3363 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3364 struct device_attribute *attr,
3367 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3369 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3371 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3373 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3374 struct device_attribute *attr,
3377 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3379 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3381 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3383 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3384 struct device_attribute *attr,
3387 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3389 return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3391 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3393 static ssize_t nvme_sysfs_show_address(struct device *dev,
3394 struct device_attribute *attr,
3397 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3399 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3401 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3403 static struct attribute *nvme_dev_attrs[] = {
3404 &dev_attr_reset_controller.attr,
3405 &dev_attr_rescan_controller.attr,
3406 &dev_attr_model.attr,
3407 &dev_attr_serial.attr,
3408 &dev_attr_firmware_rev.attr,
3409 &dev_attr_cntlid.attr,
3410 &dev_attr_delete_controller.attr,
3411 &dev_attr_transport.attr,
3412 &dev_attr_subsysnqn.attr,
3413 &dev_attr_address.attr,
3414 &dev_attr_state.attr,
3415 &dev_attr_numa_node.attr,
3416 &dev_attr_queue_count.attr,
3417 &dev_attr_sqsize.attr,
3418 &dev_attr_hostnqn.attr,
3419 &dev_attr_hostid.attr,
3423 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3424 struct attribute *a, int n)
3426 struct device *dev = container_of(kobj, struct device, kobj);
3427 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3429 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3431 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3433 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3435 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3441 static struct attribute_group nvme_dev_attrs_group = {
3442 .attrs = nvme_dev_attrs,
3443 .is_visible = nvme_dev_attrs_are_visible,
3446 static const struct attribute_group *nvme_dev_attr_groups[] = {
3447 &nvme_dev_attrs_group,
3451 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3454 struct nvme_ns_head *h;
3456 lockdep_assert_held(&subsys->lock);
3458 list_for_each_entry(h, &subsys->nsheads, entry) {
3459 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3466 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3467 struct nvme_ns_head *new)
3469 struct nvme_ns_head *h;
3471 lockdep_assert_held(&subsys->lock);
3473 list_for_each_entry(h, &subsys->nsheads, entry) {
3474 if (nvme_ns_ids_valid(&new->ids) &&
3475 nvme_ns_ids_equal(&new->ids, &h->ids))
3482 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3483 unsigned nsid, struct nvme_ns_ids *ids)
3485 struct nvme_ns_head *head;
3486 size_t size = sizeof(*head);
3489 #ifdef CONFIG_NVME_MULTIPATH
3490 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3493 head = kzalloc(size, GFP_KERNEL);
3496 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3499 head->instance = ret;
3500 INIT_LIST_HEAD(&head->list);
3501 ret = init_srcu_struct(&head->srcu);
3503 goto out_ida_remove;
3504 head->subsys = ctrl->subsys;
3507 kref_init(&head->ref);
3509 ret = __nvme_check_ids(ctrl->subsys, head);
3511 dev_err(ctrl->device,
3512 "duplicate IDs for nsid %d\n", nsid);
3513 goto out_cleanup_srcu;
3516 ret = nvme_mpath_alloc_disk(ctrl, head);
3518 goto out_cleanup_srcu;
3520 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3522 kref_get(&ctrl->subsys->ref);
3526 cleanup_srcu_struct(&head->srcu);
3528 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3533 ret = blk_status_to_errno(nvme_error_status(ret));
3534 return ERR_PTR(ret);
3537 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3538 struct nvme_id_ns *id)
3540 struct nvme_ctrl *ctrl = ns->ctrl;
3541 bool is_shared = id->nmic & NVME_NS_NMIC_SHARED;
3542 struct nvme_ns_head *head = NULL;
3543 struct nvme_ns_ids ids;
3546 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3550 return blk_status_to_errno(nvme_error_status(ret));
3553 mutex_lock(&ctrl->subsys->lock);
3554 head = nvme_find_ns_head(ctrl->subsys, nsid);
3556 head = nvme_alloc_ns_head(ctrl, nsid, &ids);
3558 ret = PTR_ERR(head);
3561 head->shared = is_shared;
3564 if (!is_shared || !head->shared) {
3565 dev_err(ctrl->device,
3566 "Duplicate unshared namespace %d\n", nsid);
3567 goto out_put_ns_head;
3569 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3570 dev_err(ctrl->device,
3571 "IDs don't match for shared namespace %d\n",
3573 goto out_put_ns_head;
3577 list_add_tail(&ns->siblings, &head->list);
3579 mutex_unlock(&ctrl->subsys->lock);
3583 nvme_put_ns_head(head);
3585 mutex_unlock(&ctrl->subsys->lock);
3589 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3591 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3592 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3594 return nsa->head->ns_id - nsb->head->ns_id;
3597 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3599 struct nvme_ns *ns, *ret = NULL;
3601 down_read(&ctrl->namespaces_rwsem);
3602 list_for_each_entry(ns, &ctrl->namespaces, list) {
3603 if (ns->head->ns_id == nsid) {
3604 if (!kref_get_unless_zero(&ns->kref))
3609 if (ns->head->ns_id > nsid)
3612 up_read(&ctrl->namespaces_rwsem);
3616 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3619 struct gendisk *disk;
3620 struct nvme_id_ns *id;
3621 char disk_name[DISK_NAME_LEN];
3622 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3624 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3628 ns->queue = blk_mq_init_queue(ctrl->tagset);
3629 if (IS_ERR(ns->queue))
3632 if (ctrl->opts && ctrl->opts->data_digest)
3633 ns->queue->backing_dev_info->capabilities
3634 |= BDI_CAP_STABLE_WRITES;
3636 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3637 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3638 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3640 ns->queue->queuedata = ns;
3643 kref_init(&ns->kref);
3644 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3646 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3647 nvme_set_queue_limits(ctrl, ns->queue);
3649 ret = nvme_identify_ns(ctrl, nsid, &id);
3651 goto out_free_queue;
3653 if (id->ncap == 0) /* no namespace (legacy quirk) */
3656 ret = nvme_init_ns_head(ns, nsid, id);
3659 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3661 disk = alloc_disk_node(0, node);
3665 disk->fops = &nvme_fops;
3666 disk->private_data = ns;
3667 disk->queue = ns->queue;
3668 disk->flags = flags;
3669 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3672 if (__nvme_revalidate_disk(disk, id))
3675 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3676 ret = nvme_nvm_register(ns, disk_name, node);
3678 dev_warn(ctrl->device, "LightNVM init failure\n");
3683 down_write(&ctrl->namespaces_rwsem);
3684 list_add_tail(&ns->list, &ctrl->namespaces);
3685 up_write(&ctrl->namespaces_rwsem);
3687 nvme_get_ctrl(ctrl);
3689 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3691 nvme_mpath_add_disk(ns, id);
3692 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3697 /* prevent double queue cleanup */
3698 ns->disk->queue = NULL;
3701 mutex_lock(&ctrl->subsys->lock);
3702 list_del_rcu(&ns->siblings);
3703 if (list_empty(&ns->head->list))
3704 list_del_init(&ns->head->entry);
3705 mutex_unlock(&ctrl->subsys->lock);
3706 nvme_put_ns_head(ns->head);
3710 blk_cleanup_queue(ns->queue);
3715 static void nvme_ns_remove(struct nvme_ns *ns)
3717 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3720 nvme_fault_inject_fini(&ns->fault_inject);
3722 mutex_lock(&ns->ctrl->subsys->lock);
3723 list_del_rcu(&ns->siblings);
3724 if (list_empty(&ns->head->list))
3725 list_del_init(&ns->head->entry);
3726 mutex_unlock(&ns->ctrl->subsys->lock);
3728 synchronize_rcu(); /* guarantee not available in head->list */
3729 nvme_mpath_clear_current_path(ns);
3730 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3732 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3733 del_gendisk(ns->disk);
3734 blk_cleanup_queue(ns->queue);
3735 if (blk_get_integrity(ns->disk))
3736 blk_integrity_unregister(ns->disk);
3739 down_write(&ns->ctrl->namespaces_rwsem);
3740 list_del_init(&ns->list);
3741 up_write(&ns->ctrl->namespaces_rwsem);
3743 nvme_mpath_check_last_path(ns);
3747 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3749 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3757 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3761 ns = nvme_find_get_ns(ctrl, nsid);
3763 if (ns->disk && revalidate_disk(ns->disk))
3767 nvme_alloc_ns(ctrl, nsid);
3770 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3773 struct nvme_ns *ns, *next;
3776 down_write(&ctrl->namespaces_rwsem);
3777 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3778 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3779 list_move_tail(&ns->list, &rm_list);
3781 up_write(&ctrl->namespaces_rwsem);
3783 list_for_each_entry_safe(ns, next, &rm_list, list)
3788 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3790 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3795 if (nvme_ctrl_limited_cns(ctrl))
3798 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3803 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3807 for (i = 0; i < nr_entries; i++) {
3808 u32 nsid = le32_to_cpu(ns_list[i]);
3810 if (!nsid) /* end of the list? */
3812 nvme_validate_ns(ctrl, nsid);
3813 while (++prev < nsid)
3814 nvme_ns_remove_by_nsid(ctrl, prev);
3818 nvme_remove_invalid_namespaces(ctrl, prev);
3824 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3826 struct nvme_id_ctrl *id;
3829 if (nvme_identify_ctrl(ctrl, &id))
3831 nn = le32_to_cpu(id->nn);
3834 for (i = 1; i <= nn; i++)
3835 nvme_validate_ns(ctrl, i);
3837 nvme_remove_invalid_namespaces(ctrl, nn);
3840 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3842 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3846 log = kzalloc(log_size, GFP_KERNEL);
3851 * We need to read the log to clear the AEN, but we don't want to rely
3852 * on it for the changed namespace information as userspace could have
3853 * raced with us in reading the log page, which could cause us to miss
3856 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3859 dev_warn(ctrl->device,
3860 "reading changed ns log failed: %d\n", error);
3865 static void nvme_scan_work(struct work_struct *work)
3867 struct nvme_ctrl *ctrl =
3868 container_of(work, struct nvme_ctrl, scan_work);
3870 /* No tagset on a live ctrl means IO queues could not created */
3871 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3874 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3875 dev_info(ctrl->device, "rescanning namespaces.\n");
3876 nvme_clear_changed_ns_log(ctrl);
3879 mutex_lock(&ctrl->scan_lock);
3880 if (nvme_scan_ns_list(ctrl) != 0)
3881 nvme_scan_ns_sequential(ctrl);
3882 mutex_unlock(&ctrl->scan_lock);
3884 down_write(&ctrl->namespaces_rwsem);
3885 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3886 up_write(&ctrl->namespaces_rwsem);
3890 * This function iterates the namespace list unlocked to allow recovery from
3891 * controller failure. It is up to the caller to ensure the namespace list is
3892 * not modified by scan work while this function is executing.
3894 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3896 struct nvme_ns *ns, *next;
3900 * make sure to requeue I/O to all namespaces as these
3901 * might result from the scan itself and must complete
3902 * for the scan_work to make progress
3904 nvme_mpath_clear_ctrl_paths(ctrl);
3906 /* prevent racing with ns scanning */
3907 flush_work(&ctrl->scan_work);
3910 * The dead states indicates the controller was not gracefully
3911 * disconnected. In that case, we won't be able to flush any data while
3912 * removing the namespaces' disks; fail all the queues now to avoid
3913 * potentially having to clean up the failed sync later.
3915 if (ctrl->state == NVME_CTRL_DEAD)
3916 nvme_kill_queues(ctrl);
3918 down_write(&ctrl->namespaces_rwsem);
3919 list_splice_init(&ctrl->namespaces, &ns_list);
3920 up_write(&ctrl->namespaces_rwsem);
3922 list_for_each_entry_safe(ns, next, &ns_list, list)
3925 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3927 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3929 struct nvme_ctrl *ctrl =
3930 container_of(dev, struct nvme_ctrl, ctrl_device);
3931 struct nvmf_ctrl_options *opts = ctrl->opts;
3934 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3939 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3943 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3944 opts->trsvcid ?: "none");
3948 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3949 opts->host_traddr ?: "none");
3954 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3956 char *envp[2] = { NULL, NULL };
3957 u32 aen_result = ctrl->aen_result;
3959 ctrl->aen_result = 0;
3963 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3966 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3970 static void nvme_async_event_work(struct work_struct *work)
3972 struct nvme_ctrl *ctrl =
3973 container_of(work, struct nvme_ctrl, async_event_work);
3975 nvme_aen_uevent(ctrl);
3976 ctrl->ops->submit_async_event(ctrl);
3979 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3984 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3990 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3993 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3995 struct nvme_fw_slot_info_log *log;
3997 log = kmalloc(sizeof(*log), GFP_KERNEL);
4001 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, log,
4003 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4007 static void nvme_fw_act_work(struct work_struct *work)
4009 struct nvme_ctrl *ctrl = container_of(work,
4010 struct nvme_ctrl, fw_act_work);
4011 unsigned long fw_act_timeout;
4014 fw_act_timeout = jiffies +
4015 msecs_to_jiffies(ctrl->mtfa * 100);
4017 fw_act_timeout = jiffies +
4018 msecs_to_jiffies(admin_timeout * 1000);
4020 nvme_stop_queues(ctrl);
4021 while (nvme_ctrl_pp_status(ctrl)) {
4022 if (time_after(jiffies, fw_act_timeout)) {
4023 dev_warn(ctrl->device,
4024 "Fw activation timeout, reset controller\n");
4025 nvme_try_sched_reset(ctrl);
4031 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4034 nvme_start_queues(ctrl);
4035 /* read FW slot information to clear the AER */
4036 nvme_get_fw_slot_info(ctrl);
4039 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4041 u32 aer_notice_type = (result & 0xff00) >> 8;
4043 trace_nvme_async_event(ctrl, aer_notice_type);
4045 switch (aer_notice_type) {
4046 case NVME_AER_NOTICE_NS_CHANGED:
4047 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4048 nvme_queue_scan(ctrl);
4050 case NVME_AER_NOTICE_FW_ACT_STARTING:
4052 * We are (ab)using the RESETTING state to prevent subsequent
4053 * recovery actions from interfering with the controller's
4054 * firmware activation.
4056 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4057 queue_work(nvme_wq, &ctrl->fw_act_work);
4059 #ifdef CONFIG_NVME_MULTIPATH
4060 case NVME_AER_NOTICE_ANA:
4061 if (!ctrl->ana_log_buf)
4063 queue_work(nvme_wq, &ctrl->ana_work);
4066 case NVME_AER_NOTICE_DISC_CHANGED:
4067 ctrl->aen_result = result;
4070 dev_warn(ctrl->device, "async event result %08x\n", result);
4074 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4075 volatile union nvme_result *res)
4077 u32 result = le32_to_cpu(res->u32);
4078 u32 aer_type = result & 0x07;
4080 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4084 case NVME_AER_NOTICE:
4085 nvme_handle_aen_notice(ctrl, result);
4087 case NVME_AER_ERROR:
4088 case NVME_AER_SMART:
4091 trace_nvme_async_event(ctrl, aer_type);
4092 ctrl->aen_result = result;
4097 queue_work(nvme_wq, &ctrl->async_event_work);
4099 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4101 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4103 nvme_mpath_stop(ctrl);
4104 nvme_stop_keep_alive(ctrl);
4105 flush_work(&ctrl->async_event_work);
4106 cancel_work_sync(&ctrl->fw_act_work);
4108 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4110 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4113 nvme_start_keep_alive(ctrl);
4115 nvme_enable_aen(ctrl);
4117 if (ctrl->queue_count > 1) {
4118 nvme_queue_scan(ctrl);
4119 nvme_start_queues(ctrl);
4121 ctrl->created = true;
4123 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4125 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4127 nvme_fault_inject_fini(&ctrl->fault_inject);
4128 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4129 cdev_device_del(&ctrl->cdev, ctrl->device);
4130 nvme_put_ctrl(ctrl);
4132 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4134 static void nvme_free_ctrl(struct device *dev)
4136 struct nvme_ctrl *ctrl =
4137 container_of(dev, struct nvme_ctrl, ctrl_device);
4138 struct nvme_subsystem *subsys = ctrl->subsys;
4140 if (subsys && ctrl->instance != subsys->instance)
4141 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4143 kfree(ctrl->effects);
4144 nvme_mpath_uninit(ctrl);
4145 __free_page(ctrl->discard_page);
4148 mutex_lock(&nvme_subsystems_lock);
4149 list_del(&ctrl->subsys_entry);
4150 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4151 mutex_unlock(&nvme_subsystems_lock);
4154 ctrl->ops->free_ctrl(ctrl);
4157 nvme_put_subsystem(subsys);
4161 * Initialize a NVMe controller structures. This needs to be called during
4162 * earliest initialization so that we have the initialized structured around
4165 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4166 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4170 ctrl->state = NVME_CTRL_NEW;
4171 spin_lock_init(&ctrl->lock);
4172 mutex_init(&ctrl->scan_lock);
4173 INIT_LIST_HEAD(&ctrl->namespaces);
4174 init_rwsem(&ctrl->namespaces_rwsem);
4177 ctrl->quirks = quirks;
4178 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4179 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4180 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4181 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4182 init_waitqueue_head(&ctrl->state_wq);
4184 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4185 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4186 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4188 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4190 ctrl->discard_page = alloc_page(GFP_KERNEL);
4191 if (!ctrl->discard_page) {
4196 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4199 ctrl->instance = ret;
4201 device_initialize(&ctrl->ctrl_device);
4202 ctrl->device = &ctrl->ctrl_device;
4203 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4204 ctrl->device->class = nvme_class;
4205 ctrl->device->parent = ctrl->dev;
4206 ctrl->device->groups = nvme_dev_attr_groups;
4207 ctrl->device->release = nvme_free_ctrl;
4208 dev_set_drvdata(ctrl->device, ctrl);
4209 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4211 goto out_release_instance;
4213 nvme_get_ctrl(ctrl);
4214 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4215 ctrl->cdev.owner = ops->module;
4216 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4221 * Initialize latency tolerance controls. The sysfs files won't
4222 * be visible to userspace unless the device actually supports APST.
4224 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4225 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4226 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4228 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4232 nvme_put_ctrl(ctrl);
4233 kfree_const(ctrl->device->kobj.name);
4234 out_release_instance:
4235 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4237 if (ctrl->discard_page)
4238 __free_page(ctrl->discard_page);
4241 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4244 * nvme_kill_queues(): Ends all namespace queues
4245 * @ctrl: the dead controller that needs to end
4247 * Call this function when the driver determines it is unable to get the
4248 * controller in a state capable of servicing IO.
4250 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4254 down_read(&ctrl->namespaces_rwsem);
4256 /* Forcibly unquiesce queues to avoid blocking dispatch */
4257 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4258 blk_mq_unquiesce_queue(ctrl->admin_q);
4260 list_for_each_entry(ns, &ctrl->namespaces, list)
4261 nvme_set_queue_dying(ns);
4263 up_read(&ctrl->namespaces_rwsem);
4265 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4267 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4271 down_read(&ctrl->namespaces_rwsem);
4272 list_for_each_entry(ns, &ctrl->namespaces, list)
4273 blk_mq_unfreeze_queue(ns->queue);
4274 up_read(&ctrl->namespaces_rwsem);
4276 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4278 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4282 down_read(&ctrl->namespaces_rwsem);
4283 list_for_each_entry(ns, &ctrl->namespaces, list) {
4284 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4288 up_read(&ctrl->namespaces_rwsem);
4290 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4292 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4296 down_read(&ctrl->namespaces_rwsem);
4297 list_for_each_entry(ns, &ctrl->namespaces, list)
4298 blk_mq_freeze_queue_wait(ns->queue);
4299 up_read(&ctrl->namespaces_rwsem);
4301 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4303 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4307 down_read(&ctrl->namespaces_rwsem);
4308 list_for_each_entry(ns, &ctrl->namespaces, list)
4309 blk_freeze_queue_start(ns->queue);
4310 up_read(&ctrl->namespaces_rwsem);
4312 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4314 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4318 down_read(&ctrl->namespaces_rwsem);
4319 list_for_each_entry(ns, &ctrl->namespaces, list)
4320 blk_mq_quiesce_queue(ns->queue);
4321 up_read(&ctrl->namespaces_rwsem);
4323 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4325 void nvme_start_queues(struct nvme_ctrl *ctrl)
4329 down_read(&ctrl->namespaces_rwsem);
4330 list_for_each_entry(ns, &ctrl->namespaces, list)
4331 blk_mq_unquiesce_queue(ns->queue);
4332 up_read(&ctrl->namespaces_rwsem);
4334 EXPORT_SYMBOL_GPL(nvme_start_queues);
4337 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4341 down_read(&ctrl->namespaces_rwsem);
4342 list_for_each_entry(ns, &ctrl->namespaces, list)
4343 blk_sync_queue(ns->queue);
4344 up_read(&ctrl->namespaces_rwsem);
4347 blk_sync_queue(ctrl->admin_q);
4349 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4352 * Check we didn't inadvertently grow the command structure sizes:
4354 static inline void _nvme_check_size(void)
4356 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4357 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4358 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4359 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4360 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4361 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4362 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4363 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4364 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4365 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4366 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4367 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4368 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4369 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4370 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4371 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4372 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4376 static int __init nvme_core_init(void)
4378 int result = -ENOMEM;
4382 nvme_wq = alloc_workqueue("nvme-wq",
4383 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4387 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4388 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4392 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4393 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4394 if (!nvme_delete_wq)
4395 goto destroy_reset_wq;
4397 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4399 goto destroy_delete_wq;
4401 nvme_class = class_create(THIS_MODULE, "nvme");
4402 if (IS_ERR(nvme_class)) {
4403 result = PTR_ERR(nvme_class);
4404 goto unregister_chrdev;
4406 nvme_class->dev_uevent = nvme_class_uevent;
4408 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4409 if (IS_ERR(nvme_subsys_class)) {
4410 result = PTR_ERR(nvme_subsys_class);
4416 class_destroy(nvme_class);
4418 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4420 destroy_workqueue(nvme_delete_wq);
4422 destroy_workqueue(nvme_reset_wq);
4424 destroy_workqueue(nvme_wq);
4429 static void __exit nvme_core_exit(void)
4431 class_destroy(nvme_subsys_class);
4432 class_destroy(nvme_class);
4433 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4434 destroy_workqueue(nvme_delete_wq);
4435 destroy_workqueue(nvme_reset_wq);
4436 destroy_workqueue(nvme_wq);
4437 ida_destroy(&nvme_instance_ida);
4440 MODULE_LICENSE("GPL");
4441 MODULE_VERSION("1.0");
4442 module_init(nvme_core_init);
4443 module_exit(nvme_core_exit);