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/t10-pi.h>
23 #include <linux/pm_qos.h>
24 #include <asm/unaligned.h>
29 #define CREATE_TRACE_POINTS
32 #define NVME_MINORS (1U << MINORBITS)
34 unsigned int admin_timeout = 60;
35 module_param(admin_timeout, uint, 0644);
36 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
37 EXPORT_SYMBOL_GPL(admin_timeout);
39 unsigned int nvme_io_timeout = 30;
40 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
41 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
42 EXPORT_SYMBOL_GPL(nvme_io_timeout);
44 static unsigned char shutdown_timeout = 5;
45 module_param(shutdown_timeout, byte, 0644);
46 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
48 static u8 nvme_max_retries = 5;
49 module_param_named(max_retries, nvme_max_retries, byte, 0644);
50 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
52 static unsigned long default_ps_max_latency_us = 100000;
53 module_param(default_ps_max_latency_us, ulong, 0644);
54 MODULE_PARM_DESC(default_ps_max_latency_us,
55 "max power saving latency for new devices; use PM QOS to change per device");
57 static bool force_apst;
58 module_param(force_apst, bool, 0644);
59 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
62 module_param(streams, bool, 0644);
63 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
66 * nvme_wq - hosts nvme related works that are not reset or delete
67 * nvme_reset_wq - hosts nvme reset works
68 * nvme_delete_wq - hosts nvme delete works
70 * nvme_wq will host works such as scan, aen handling, fw activation,
71 * keep-alive, periodic reconnects etc. nvme_reset_wq
72 * runs reset works which also flush works hosted on nvme_wq for
73 * serialization purposes. nvme_delete_wq host controller deletion
74 * works which flush reset works for serialization.
76 struct workqueue_struct *nvme_wq;
77 EXPORT_SYMBOL_GPL(nvme_wq);
79 struct workqueue_struct *nvme_reset_wq;
80 EXPORT_SYMBOL_GPL(nvme_reset_wq);
82 struct workqueue_struct *nvme_delete_wq;
83 EXPORT_SYMBOL_GPL(nvme_delete_wq);
85 static LIST_HEAD(nvme_subsystems);
86 static DEFINE_MUTEX(nvme_subsystems_lock);
88 static DEFINE_IDA(nvme_instance_ida);
89 static dev_t nvme_chr_devt;
90 static struct class *nvme_class;
91 static struct class *nvme_subsys_class;
93 static int nvme_revalidate_disk(struct gendisk *disk);
94 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
95 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
98 static void nvme_set_queue_dying(struct nvme_ns *ns)
101 * Revalidating a dead namespace sets capacity to 0. This will end
102 * buffered writers dirtying pages that can't be synced.
104 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
106 blk_set_queue_dying(ns->queue);
107 /* Forcibly unquiesce queues to avoid blocking dispatch */
108 blk_mq_unquiesce_queue(ns->queue);
110 * Revalidate after unblocking dispatchers that may be holding bd_butex
112 revalidate_disk(ns->disk);
115 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
118 * Only new queue scan work when admin and IO queues are both alive
120 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
121 queue_work(nvme_wq, &ctrl->scan_work);
125 * Use this function to proceed with scheduling reset_work for a controller
126 * that had previously been set to the resetting state. This is intended for
127 * code paths that can't be interrupted by other reset attempts. A hot removal
128 * may prevent this from succeeding.
130 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
132 if (ctrl->state != NVME_CTRL_RESETTING)
134 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
138 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
140 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
142 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
144 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
148 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
150 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
154 ret = nvme_reset_ctrl(ctrl);
156 flush_work(&ctrl->reset_work);
157 if (ctrl->state != NVME_CTRL_LIVE)
163 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
165 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
167 dev_info(ctrl->device,
168 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
170 flush_work(&ctrl->reset_work);
171 nvme_stop_ctrl(ctrl);
172 nvme_remove_namespaces(ctrl);
173 ctrl->ops->delete_ctrl(ctrl);
174 nvme_uninit_ctrl(ctrl);
177 static void nvme_delete_ctrl_work(struct work_struct *work)
179 struct nvme_ctrl *ctrl =
180 container_of(work, struct nvme_ctrl, delete_work);
182 nvme_do_delete_ctrl(ctrl);
185 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
187 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
189 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
193 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
195 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
198 * Keep a reference until nvme_do_delete_ctrl() complete,
199 * since ->delete_ctrl can free the controller.
202 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
203 nvme_do_delete_ctrl(ctrl);
207 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
209 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
212 static blk_status_t nvme_error_status(u16 status)
214 switch (status & 0x7ff) {
215 case NVME_SC_SUCCESS:
217 case NVME_SC_CAP_EXCEEDED:
218 return BLK_STS_NOSPC;
219 case NVME_SC_LBA_RANGE:
220 case NVME_SC_CMD_INTERRUPTED:
221 case NVME_SC_NS_NOT_READY:
222 return BLK_STS_TARGET;
223 case NVME_SC_BAD_ATTRIBUTES:
224 case NVME_SC_ONCS_NOT_SUPPORTED:
225 case NVME_SC_INVALID_OPCODE:
226 case NVME_SC_INVALID_FIELD:
227 case NVME_SC_INVALID_NS:
228 return BLK_STS_NOTSUPP;
229 case NVME_SC_WRITE_FAULT:
230 case NVME_SC_READ_ERROR:
231 case NVME_SC_UNWRITTEN_BLOCK:
232 case NVME_SC_ACCESS_DENIED:
233 case NVME_SC_READ_ONLY:
234 case NVME_SC_COMPARE_FAILED:
235 return BLK_STS_MEDIUM;
236 case NVME_SC_GUARD_CHECK:
237 case NVME_SC_APPTAG_CHECK:
238 case NVME_SC_REFTAG_CHECK:
239 case NVME_SC_INVALID_PI:
240 return BLK_STS_PROTECTION;
241 case NVME_SC_RESERVATION_CONFLICT:
242 return BLK_STS_NEXUS;
243 case NVME_SC_HOST_PATH_ERROR:
244 return BLK_STS_TRANSPORT;
246 return BLK_STS_IOERR;
250 static inline bool nvme_req_needs_retry(struct request *req)
252 if (blk_noretry_request(req))
254 if (nvme_req(req)->status & NVME_SC_DNR)
256 if (nvme_req(req)->retries >= nvme_max_retries)
261 static void nvme_retry_req(struct request *req)
263 struct nvme_ns *ns = req->q->queuedata;
264 unsigned long delay = 0;
267 /* The mask and shift result must be <= 3 */
268 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
270 delay = ns->ctrl->crdt[crd - 1] * 100;
272 nvme_req(req)->retries++;
273 blk_mq_requeue_request(req, false);
274 blk_mq_delay_kick_requeue_list(req->q, delay);
277 void nvme_complete_rq(struct request *req)
279 blk_status_t status = nvme_error_status(nvme_req(req)->status);
281 trace_nvme_complete_rq(req);
283 nvme_cleanup_cmd(req);
285 if (nvme_req(req)->ctrl->kas)
286 nvme_req(req)->ctrl->comp_seen = true;
288 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
289 if ((req->cmd_flags & REQ_NVME_MPATH) && nvme_failover_req(req))
292 if (!blk_queue_dying(req->q)) {
298 nvme_trace_bio_complete(req, status);
299 blk_mq_end_request(req, status);
301 EXPORT_SYMBOL_GPL(nvme_complete_rq);
303 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
305 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
306 "Cancelling I/O %d", req->tag);
308 /* don't abort one completed request */
309 if (blk_mq_request_completed(req))
312 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
313 blk_mq_complete_request(req);
316 EXPORT_SYMBOL_GPL(nvme_cancel_request);
318 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
319 enum nvme_ctrl_state new_state)
321 enum nvme_ctrl_state old_state;
323 bool changed = false;
325 spin_lock_irqsave(&ctrl->lock, flags);
327 old_state = ctrl->state;
332 case NVME_CTRL_RESETTING:
333 case NVME_CTRL_CONNECTING:
340 case NVME_CTRL_RESETTING:
350 case NVME_CTRL_CONNECTING:
353 case NVME_CTRL_RESETTING:
360 case NVME_CTRL_DELETING:
363 case NVME_CTRL_RESETTING:
364 case NVME_CTRL_CONNECTING:
373 case NVME_CTRL_DELETING:
385 ctrl->state = new_state;
386 wake_up_all(&ctrl->state_wq);
389 spin_unlock_irqrestore(&ctrl->lock, flags);
390 if (changed && ctrl->state == NVME_CTRL_LIVE)
391 nvme_kick_requeue_lists(ctrl);
394 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
397 * Returns true for sink states that can't ever transition back to live.
399 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
401 switch (ctrl->state) {
404 case NVME_CTRL_RESETTING:
405 case NVME_CTRL_CONNECTING:
407 case NVME_CTRL_DELETING:
411 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
417 * Waits for the controller state to be resetting, or returns false if it is
418 * not possible to ever transition to that state.
420 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
422 wait_event(ctrl->state_wq,
423 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
424 nvme_state_terminal(ctrl));
425 return ctrl->state == NVME_CTRL_RESETTING;
427 EXPORT_SYMBOL_GPL(nvme_wait_reset);
429 static void nvme_free_ns_head(struct kref *ref)
431 struct nvme_ns_head *head =
432 container_of(ref, struct nvme_ns_head, ref);
434 nvme_mpath_remove_disk(head);
435 ida_simple_remove(&head->subsys->ns_ida, head->instance);
436 cleanup_srcu_struct(&head->srcu);
437 nvme_put_subsystem(head->subsys);
441 static void nvme_put_ns_head(struct nvme_ns_head *head)
443 kref_put(&head->ref, nvme_free_ns_head);
446 static void nvme_free_ns(struct kref *kref)
448 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
451 nvme_nvm_unregister(ns);
454 nvme_put_ns_head(ns->head);
455 nvme_put_ctrl(ns->ctrl);
459 static void nvme_put_ns(struct nvme_ns *ns)
461 kref_put(&ns->kref, nvme_free_ns);
464 static inline void nvme_clear_nvme_request(struct request *req)
466 if (!(req->rq_flags & RQF_DONTPREP)) {
467 nvme_req(req)->retries = 0;
468 nvme_req(req)->flags = 0;
469 req->rq_flags |= RQF_DONTPREP;
473 struct request *nvme_alloc_request(struct request_queue *q,
474 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
476 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
479 if (qid == NVME_QID_ANY) {
480 req = blk_mq_alloc_request(q, op, flags);
482 req = blk_mq_alloc_request_hctx(q, op, flags,
488 req->cmd_flags |= REQ_FAILFAST_DRIVER;
489 nvme_clear_nvme_request(req);
490 nvme_req(req)->cmd = cmd;
494 EXPORT_SYMBOL_GPL(nvme_alloc_request);
496 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
498 struct nvme_command c;
500 memset(&c, 0, sizeof(c));
502 c.directive.opcode = nvme_admin_directive_send;
503 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
504 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
505 c.directive.dtype = NVME_DIR_IDENTIFY;
506 c.directive.tdtype = NVME_DIR_STREAMS;
507 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
509 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
512 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
514 return nvme_toggle_streams(ctrl, false);
517 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
519 return nvme_toggle_streams(ctrl, true);
522 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
523 struct streams_directive_params *s, u32 nsid)
525 struct nvme_command c;
527 memset(&c, 0, sizeof(c));
528 memset(s, 0, sizeof(*s));
530 c.directive.opcode = nvme_admin_directive_recv;
531 c.directive.nsid = cpu_to_le32(nsid);
532 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
533 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
534 c.directive.dtype = NVME_DIR_STREAMS;
536 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
539 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
541 struct streams_directive_params s;
544 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
549 ret = nvme_enable_streams(ctrl);
553 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
557 ctrl->nssa = le16_to_cpu(s.nssa);
558 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
559 dev_info(ctrl->device, "too few streams (%u) available\n",
561 nvme_disable_streams(ctrl);
565 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
566 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
571 * Check if 'req' has a write hint associated with it. If it does, assign
572 * a valid namespace stream to the write.
574 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
575 struct request *req, u16 *control,
578 enum rw_hint streamid = req->write_hint;
580 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
584 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
587 *control |= NVME_RW_DTYPE_STREAMS;
588 *dsmgmt |= streamid << 16;
591 if (streamid < ARRAY_SIZE(req->q->write_hints))
592 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
595 static inline void nvme_setup_flush(struct nvme_ns *ns,
596 struct nvme_command *cmnd)
598 cmnd->common.opcode = nvme_cmd_flush;
599 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
602 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
603 struct nvme_command *cmnd)
605 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
606 struct nvme_dsm_range *range;
610 * Some devices do not consider the DSM 'Number of Ranges' field when
611 * determining how much data to DMA. Always allocate memory for maximum
612 * number of segments to prevent device reading beyond end of buffer.
614 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
616 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
619 * If we fail allocation our range, fallback to the controller
620 * discard page. If that's also busy, it's safe to return
621 * busy, as we know we can make progress once that's freed.
623 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
624 return BLK_STS_RESOURCE;
626 range = page_address(ns->ctrl->discard_page);
629 __rq_for_each_bio(bio, req) {
630 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
631 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
634 range[n].cattr = cpu_to_le32(0);
635 range[n].nlb = cpu_to_le32(nlb);
636 range[n].slba = cpu_to_le64(slba);
641 if (WARN_ON_ONCE(n != segments)) {
642 if (virt_to_page(range) == ns->ctrl->discard_page)
643 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
646 return BLK_STS_IOERR;
649 cmnd->dsm.opcode = nvme_cmd_dsm;
650 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
651 cmnd->dsm.nr = cpu_to_le32(segments - 1);
652 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
654 req->special_vec.bv_page = virt_to_page(range);
655 req->special_vec.bv_offset = offset_in_page(range);
656 req->special_vec.bv_len = alloc_size;
657 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
662 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
663 struct request *req, struct nvme_command *cmnd)
665 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
666 return nvme_setup_discard(ns, req, cmnd);
668 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
669 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
670 cmnd->write_zeroes.slba =
671 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
672 cmnd->write_zeroes.length =
673 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
674 cmnd->write_zeroes.control = 0;
678 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
679 struct request *req, struct nvme_command *cmnd)
681 struct nvme_ctrl *ctrl = ns->ctrl;
685 if (req->cmd_flags & REQ_FUA)
686 control |= NVME_RW_FUA;
687 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
688 control |= NVME_RW_LR;
690 if (req->cmd_flags & REQ_RAHEAD)
691 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
693 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
694 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
695 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
696 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
698 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
699 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
703 * If formated with metadata, the block layer always provides a
704 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
705 * we enable the PRACT bit for protection information or set the
706 * namespace capacity to zero to prevent any I/O.
708 if (!blk_integrity_rq(req)) {
709 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
710 return BLK_STS_NOTSUPP;
711 control |= NVME_RW_PRINFO_PRACT;
714 switch (ns->pi_type) {
715 case NVME_NS_DPS_PI_TYPE3:
716 control |= NVME_RW_PRINFO_PRCHK_GUARD;
718 case NVME_NS_DPS_PI_TYPE1:
719 case NVME_NS_DPS_PI_TYPE2:
720 control |= NVME_RW_PRINFO_PRCHK_GUARD |
721 NVME_RW_PRINFO_PRCHK_REF;
722 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
727 cmnd->rw.control = cpu_to_le16(control);
728 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
732 void nvme_cleanup_cmd(struct request *req)
734 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
735 struct nvme_ns *ns = req->rq_disk->private_data;
736 struct page *page = req->special_vec.bv_page;
738 if (page == ns->ctrl->discard_page)
739 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
741 kfree(page_address(page) + req->special_vec.bv_offset);
744 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
746 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
747 struct nvme_command *cmd)
749 blk_status_t ret = BLK_STS_OK;
751 nvme_clear_nvme_request(req);
753 memset(cmd, 0, sizeof(*cmd));
754 switch (req_op(req)) {
757 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
760 nvme_setup_flush(ns, cmd);
762 case REQ_OP_WRITE_ZEROES:
763 ret = nvme_setup_write_zeroes(ns, req, cmd);
766 ret = nvme_setup_discard(ns, req, cmd);
770 ret = nvme_setup_rw(ns, req, cmd);
774 return BLK_STS_IOERR;
777 cmd->common.command_id = req->tag;
778 trace_nvme_setup_cmd(req, cmd);
781 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
783 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
785 struct completion *waiting = rq->end_io_data;
787 rq->end_io_data = NULL;
791 static void nvme_execute_rq_polled(struct request_queue *q,
792 struct gendisk *bd_disk, struct request *rq, int at_head)
794 DECLARE_COMPLETION_ONSTACK(wait);
796 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
798 rq->cmd_flags |= REQ_HIPRI;
799 rq->end_io_data = &wait;
800 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
802 while (!completion_done(&wait)) {
803 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
809 * Returns 0 on success. If the result is negative, it's a Linux error code;
810 * if the result is positive, it's an NVM Express status code
812 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
813 union nvme_result *result, void *buffer, unsigned bufflen,
814 unsigned timeout, int qid, int at_head,
815 blk_mq_req_flags_t flags, bool poll)
820 req = nvme_alloc_request(q, cmd, flags, qid);
824 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
826 if (buffer && bufflen) {
827 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
833 nvme_execute_rq_polled(req->q, NULL, req, at_head);
835 blk_execute_rq(req->q, NULL, req, at_head);
837 *result = nvme_req(req)->result;
838 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
841 ret = nvme_req(req)->status;
843 blk_mq_free_request(req);
846 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
848 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
849 void *buffer, unsigned bufflen)
851 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
852 NVME_QID_ANY, 0, 0, false);
854 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
856 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
857 unsigned len, u32 seed, bool write)
859 struct bio_integrity_payload *bip;
863 buf = kmalloc(len, GFP_KERNEL);
868 if (write && copy_from_user(buf, ubuf, len))
871 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
877 bip->bip_iter.bi_size = len;
878 bip->bip_iter.bi_sector = seed;
879 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
880 offset_in_page(buf));
890 static int nvme_submit_user_cmd(struct request_queue *q,
891 struct nvme_command *cmd, void __user *ubuffer,
892 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
893 u32 meta_seed, u64 *result, unsigned timeout)
895 bool write = nvme_is_write(cmd);
896 struct nvme_ns *ns = q->queuedata;
897 struct gendisk *disk = ns ? ns->disk : NULL;
899 struct bio *bio = NULL;
903 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
907 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
908 nvme_req(req)->flags |= NVME_REQ_USERCMD;
910 if (ubuffer && bufflen) {
911 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
917 if (disk && meta_buffer && meta_len) {
918 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
924 req->cmd_flags |= REQ_INTEGRITY;
928 blk_execute_rq(req->q, disk, req, 0);
929 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
932 ret = nvme_req(req)->status;
934 *result = le64_to_cpu(nvme_req(req)->result.u64);
935 if (meta && !ret && !write) {
936 if (copy_to_user(meta_buffer, meta, meta_len))
942 blk_rq_unmap_user(bio);
944 blk_mq_free_request(req);
948 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
950 struct nvme_ctrl *ctrl = rq->end_io_data;
952 bool startka = false;
954 blk_mq_free_request(rq);
957 dev_err(ctrl->device,
958 "failed nvme_keep_alive_end_io error=%d\n",
963 ctrl->comp_seen = false;
964 spin_lock_irqsave(&ctrl->lock, flags);
965 if (ctrl->state == NVME_CTRL_LIVE ||
966 ctrl->state == NVME_CTRL_CONNECTING)
968 spin_unlock_irqrestore(&ctrl->lock, flags);
970 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
973 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
977 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
982 rq->timeout = ctrl->kato * HZ;
983 rq->end_io_data = ctrl;
985 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
990 static void nvme_keep_alive_work(struct work_struct *work)
992 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
993 struct nvme_ctrl, ka_work);
994 bool comp_seen = ctrl->comp_seen;
996 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
997 dev_dbg(ctrl->device,
998 "reschedule traffic based keep-alive timer\n");
999 ctrl->comp_seen = false;
1000 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1004 if (nvme_keep_alive(ctrl)) {
1005 /* allocation failure, reset the controller */
1006 dev_err(ctrl->device, "keep-alive failed\n");
1007 nvme_reset_ctrl(ctrl);
1012 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1014 if (unlikely(ctrl->kato == 0))
1017 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1020 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1022 if (unlikely(ctrl->kato == 0))
1025 cancel_delayed_work_sync(&ctrl->ka_work);
1027 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1030 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1031 * flag, thus sending any new CNS opcodes has a big chance of not working.
1032 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1033 * (but not for any later version).
1035 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1037 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1038 return ctrl->vs < NVME_VS(1, 2, 0);
1039 return ctrl->vs < NVME_VS(1, 1, 0);
1042 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1044 struct nvme_command c = { };
1047 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1048 c.identify.opcode = nvme_admin_identify;
1049 c.identify.cns = NVME_ID_CNS_CTRL;
1051 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1055 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1056 sizeof(struct nvme_id_ctrl));
1062 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1063 struct nvme_ns_id_desc *cur)
1065 const char *warn_str = "ctrl returned bogus length:";
1068 switch (cur->nidt) {
1069 case NVME_NIDT_EUI64:
1070 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1071 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1072 warn_str, cur->nidl);
1075 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1076 return NVME_NIDT_EUI64_LEN;
1077 case NVME_NIDT_NGUID:
1078 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1079 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1080 warn_str, cur->nidl);
1083 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1084 return NVME_NIDT_NGUID_LEN;
1085 case NVME_NIDT_UUID:
1086 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1087 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1088 warn_str, cur->nidl);
1091 uuid_copy(&ids->uuid, data + sizeof(*cur));
1092 return NVME_NIDT_UUID_LEN;
1094 /* Skip unknown types */
1099 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1100 struct nvme_ns_ids *ids)
1102 struct nvme_command c = { };
1108 c.identify.opcode = nvme_admin_identify;
1109 c.identify.nsid = cpu_to_le32(nsid);
1110 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1112 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1116 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1117 NVME_IDENTIFY_DATA_SIZE);
1119 dev_warn(ctrl->device,
1120 "Identify Descriptors failed (%d)\n", status);
1122 * Don't treat an error as fatal, as we potentially already
1123 * have a NGUID or EUI-64.
1125 if (status > 0 && !(status & NVME_SC_DNR))
1130 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1131 struct nvme_ns_id_desc *cur = data + pos;
1136 len = nvme_process_ns_desc(ctrl, ids, cur);
1140 len += sizeof(*cur);
1147 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1149 struct nvme_command c = { };
1151 c.identify.opcode = nvme_admin_identify;
1152 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1153 c.identify.nsid = cpu_to_le32(nsid);
1154 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1155 NVME_IDENTIFY_DATA_SIZE);
1158 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1159 unsigned nsid, struct nvme_id_ns **id)
1161 struct nvme_command c = { };
1164 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1165 c.identify.opcode = nvme_admin_identify;
1166 c.identify.nsid = cpu_to_le32(nsid);
1167 c.identify.cns = NVME_ID_CNS_NS;
1169 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1173 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1175 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1182 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1183 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1185 union nvme_result res = { 0 };
1186 struct nvme_command c;
1189 memset(&c, 0, sizeof(c));
1190 c.features.opcode = op;
1191 c.features.fid = cpu_to_le32(fid);
1192 c.features.dword11 = cpu_to_le32(dword11);
1194 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1195 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1196 if (ret >= 0 && result)
1197 *result = le32_to_cpu(res.u32);
1201 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1202 unsigned int dword11, void *buffer, size_t buflen,
1205 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1208 EXPORT_SYMBOL_GPL(nvme_set_features);
1210 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1211 unsigned int dword11, void *buffer, size_t buflen,
1214 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1217 EXPORT_SYMBOL_GPL(nvme_get_features);
1219 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1221 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1223 int status, nr_io_queues;
1225 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1231 * Degraded controllers might return an error when setting the queue
1232 * count. We still want to be able to bring them online and offer
1233 * access to the admin queue, as that might be only way to fix them up.
1236 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1239 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1240 *count = min(*count, nr_io_queues);
1245 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1247 #define NVME_AEN_SUPPORTED \
1248 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1249 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1251 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1253 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1256 if (!supported_aens)
1259 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1262 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1265 queue_work(nvme_wq, &ctrl->async_event_work);
1269 * Convert integer values from ioctl structures to user pointers, silently
1270 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1273 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1275 if (in_compat_syscall())
1276 ptrval = (compat_uptr_t)ptrval;
1277 return (void __user *)ptrval;
1280 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1282 struct nvme_user_io io;
1283 struct nvme_command c;
1284 unsigned length, meta_len;
1285 void __user *metadata;
1287 if (copy_from_user(&io, uio, sizeof(io)))
1292 switch (io.opcode) {
1293 case nvme_cmd_write:
1295 case nvme_cmd_compare:
1301 length = (io.nblocks + 1) << ns->lba_shift;
1302 meta_len = (io.nblocks + 1) * ns->ms;
1303 metadata = nvme_to_user_ptr(io.metadata);
1308 } else if (meta_len) {
1309 if ((io.metadata & 3) || !io.metadata)
1313 memset(&c, 0, sizeof(c));
1314 c.rw.opcode = io.opcode;
1315 c.rw.flags = io.flags;
1316 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1317 c.rw.slba = cpu_to_le64(io.slba);
1318 c.rw.length = cpu_to_le16(io.nblocks);
1319 c.rw.control = cpu_to_le16(io.control);
1320 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1321 c.rw.reftag = cpu_to_le32(io.reftag);
1322 c.rw.apptag = cpu_to_le16(io.apptag);
1323 c.rw.appmask = cpu_to_le16(io.appmask);
1325 return nvme_submit_user_cmd(ns->queue, &c,
1326 nvme_to_user_ptr(io.addr), length,
1327 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1330 static u32 nvme_known_admin_effects(u8 opcode)
1333 case nvme_admin_format_nvm:
1334 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1335 NVME_CMD_EFFECTS_CSE_MASK;
1336 case nvme_admin_sanitize_nvm:
1337 return NVME_CMD_EFFECTS_CSE_MASK;
1344 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1351 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1352 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1353 dev_warn(ctrl->device,
1354 "IO command:%02x has unhandled effects:%08x\n",
1360 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1361 effects |= nvme_known_admin_effects(opcode);
1364 * For simplicity, IO to all namespaces is quiesced even if the command
1365 * effects say only one namespace is affected.
1367 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1368 mutex_lock(&ctrl->scan_lock);
1369 mutex_lock(&ctrl->subsys->lock);
1370 nvme_mpath_start_freeze(ctrl->subsys);
1371 nvme_mpath_wait_freeze(ctrl->subsys);
1372 nvme_start_freeze(ctrl);
1373 nvme_wait_freeze(ctrl);
1378 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1382 down_read(&ctrl->namespaces_rwsem);
1383 list_for_each_entry(ns, &ctrl->namespaces, list)
1384 if (ns->disk && nvme_revalidate_disk(ns->disk))
1385 nvme_set_queue_dying(ns);
1386 up_read(&ctrl->namespaces_rwsem);
1389 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1392 * Revalidate LBA changes prior to unfreezing. This is necessary to
1393 * prevent memory corruption if a logical block size was changed by
1396 if (effects & NVME_CMD_EFFECTS_LBCC)
1397 nvme_update_formats(ctrl);
1398 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1399 nvme_unfreeze(ctrl);
1400 nvme_mpath_unfreeze(ctrl->subsys);
1401 mutex_unlock(&ctrl->subsys->lock);
1402 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1403 mutex_unlock(&ctrl->scan_lock);
1405 if (effects & NVME_CMD_EFFECTS_CCC)
1406 nvme_init_identify(ctrl);
1407 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1408 nvme_queue_scan(ctrl);
1411 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1412 struct nvme_passthru_cmd __user *ucmd)
1414 struct nvme_passthru_cmd cmd;
1415 struct nvme_command c;
1416 unsigned timeout = 0;
1421 if (!capable(CAP_SYS_ADMIN))
1423 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1428 memset(&c, 0, sizeof(c));
1429 c.common.opcode = cmd.opcode;
1430 c.common.flags = cmd.flags;
1431 c.common.nsid = cpu_to_le32(cmd.nsid);
1432 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1433 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1434 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1435 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1436 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1437 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1438 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1439 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1442 timeout = msecs_to_jiffies(cmd.timeout_ms);
1444 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1445 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1446 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1447 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1448 0, &result, timeout);
1449 nvme_passthru_end(ctrl, effects);
1452 if (put_user(result, &ucmd->result))
1459 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1460 struct nvme_passthru_cmd64 __user *ucmd)
1462 struct nvme_passthru_cmd64 cmd;
1463 struct nvme_command c;
1464 unsigned timeout = 0;
1468 if (!capable(CAP_SYS_ADMIN))
1470 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1475 memset(&c, 0, sizeof(c));
1476 c.common.opcode = cmd.opcode;
1477 c.common.flags = cmd.flags;
1478 c.common.nsid = cpu_to_le32(cmd.nsid);
1479 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1480 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1481 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1482 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1483 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1484 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1485 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1486 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1489 timeout = msecs_to_jiffies(cmd.timeout_ms);
1491 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1492 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1493 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1494 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1495 0, &cmd.result, timeout);
1496 nvme_passthru_end(ctrl, effects);
1499 if (put_user(cmd.result, &ucmd->result))
1507 * Issue ioctl requests on the first available path. Note that unlike normal
1508 * block layer requests we will not retry failed request on another controller.
1510 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1511 struct nvme_ns_head **head, int *srcu_idx)
1513 #ifdef CONFIG_NVME_MULTIPATH
1514 if (disk->fops == &nvme_ns_head_ops) {
1517 *head = disk->private_data;
1518 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1519 ns = nvme_find_path(*head);
1521 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1527 return disk->private_data;
1530 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1533 srcu_read_unlock(&head->srcu, idx);
1536 static bool is_ctrl_ioctl(unsigned int cmd)
1538 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1540 if (is_sed_ioctl(cmd))
1545 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1547 struct nvme_ns_head *head,
1550 struct nvme_ctrl *ctrl = ns->ctrl;
1553 nvme_get_ctrl(ns->ctrl);
1554 nvme_put_ns_from_disk(head, srcu_idx);
1557 case NVME_IOCTL_ADMIN_CMD:
1558 ret = nvme_user_cmd(ctrl, NULL, argp);
1560 case NVME_IOCTL_ADMIN64_CMD:
1561 ret = nvme_user_cmd64(ctrl, NULL, argp);
1564 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1567 nvme_put_ctrl(ctrl);
1571 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1572 unsigned int cmd, unsigned long arg)
1574 struct nvme_ns_head *head = NULL;
1575 void __user *argp = (void __user *)arg;
1579 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1581 return -EWOULDBLOCK;
1584 * Handle ioctls that apply to the controller instead of the namespace
1585 * seperately and drop the ns SRCU reference early. This avoids a
1586 * deadlock when deleting namespaces using the passthrough interface.
1588 if (is_ctrl_ioctl(cmd))
1589 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1593 force_successful_syscall_return();
1594 ret = ns->head->ns_id;
1596 case NVME_IOCTL_IO_CMD:
1597 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1599 case NVME_IOCTL_SUBMIT_IO:
1600 ret = nvme_submit_io(ns, argp);
1602 case NVME_IOCTL_IO64_CMD:
1603 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1607 ret = nvme_nvm_ioctl(ns, cmd, arg);
1612 nvme_put_ns_from_disk(head, srcu_idx);
1616 #ifdef CONFIG_COMPAT
1617 struct nvme_user_io32 {
1630 } __attribute__((__packed__));
1632 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1634 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1635 unsigned int cmd, unsigned long arg)
1638 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1639 * between 32 bit programs and 64 bit kernel.
1640 * The cause is that the results of sizeof(struct nvme_user_io),
1641 * which is used to define NVME_IOCTL_SUBMIT_IO,
1642 * are not same between 32 bit compiler and 64 bit compiler.
1643 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1644 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1645 * Other IOCTL numbers are same between 32 bit and 64 bit.
1646 * So there is nothing to do regarding to other IOCTL numbers.
1648 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1649 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1651 return nvme_ioctl(bdev, mode, cmd, arg);
1654 #define nvme_compat_ioctl NULL
1655 #endif /* CONFIG_COMPAT */
1657 static int nvme_open(struct block_device *bdev, fmode_t mode)
1659 struct nvme_ns *ns = bdev->bd_disk->private_data;
1661 #ifdef CONFIG_NVME_MULTIPATH
1662 /* should never be called due to GENHD_FL_HIDDEN */
1663 if (WARN_ON_ONCE(ns->head->disk))
1666 if (!kref_get_unless_zero(&ns->kref))
1668 if (!try_module_get(ns->ctrl->ops->module))
1679 static void nvme_release(struct gendisk *disk, fmode_t mode)
1681 struct nvme_ns *ns = disk->private_data;
1683 module_put(ns->ctrl->ops->module);
1687 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1689 /* some standard values */
1690 geo->heads = 1 << 6;
1691 geo->sectors = 1 << 5;
1692 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1696 #ifdef CONFIG_BLK_DEV_INTEGRITY
1697 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
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, 1);
1723 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1726 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1728 static void nvme_set_chunk_size(struct nvme_ns *ns)
1730 u32 chunk_size = nvme_lba_to_sect(ns, ns->noiob);
1731 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1734 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1736 struct nvme_ctrl *ctrl = ns->ctrl;
1737 struct request_queue *queue = disk->queue;
1738 u32 size = queue_logical_block_size(queue);
1740 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1741 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1745 if (ctrl->nr_streams && ns->sws && ns->sgs)
1746 size *= ns->sws * ns->sgs;
1748 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1749 NVME_DSM_MAX_RANGES);
1751 queue->limits.discard_alignment = 0;
1752 queue->limits.discard_granularity = size;
1754 /* If discard is already enabled, don't reset queue limits */
1755 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1758 blk_queue_max_discard_sectors(queue, UINT_MAX);
1759 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1761 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1762 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1765 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1769 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1770 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1773 * Even though NVMe spec explicitly states that MDTS is not
1774 * applicable to the write-zeroes:- "The restriction does not apply to
1775 * commands that do not transfer data between the host and the
1776 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1777 * In order to be more cautious use controller's max_hw_sectors value
1778 * to configure the maximum sectors for the write-zeroes which is
1779 * configured based on the controller's MDTS field in the
1780 * nvme_init_identify() if available.
1782 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1783 max_blocks = (u64)USHRT_MAX + 1;
1785 max_blocks = ns->ctrl->max_hw_sectors + 1;
1787 blk_queue_max_write_zeroes_sectors(disk->queue,
1788 nvme_lba_to_sect(ns, max_blocks));
1791 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1792 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1794 memset(ids, 0, sizeof(*ids));
1796 if (ctrl->vs >= NVME_VS(1, 1, 0))
1797 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1798 if (ctrl->vs >= NVME_VS(1, 2, 0))
1799 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1800 if (ctrl->vs >= NVME_VS(1, 3, 0))
1801 return nvme_identify_ns_descs(ctrl, nsid, ids);
1805 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1807 return !uuid_is_null(&ids->uuid) ||
1808 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1809 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1812 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1814 return uuid_equal(&a->uuid, &b->uuid) &&
1815 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1816 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1819 static void nvme_update_disk_info(struct gendisk *disk,
1820 struct nvme_ns *ns, struct nvme_id_ns *id)
1822 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1823 unsigned short bs = 1 << ns->lba_shift;
1824 u32 atomic_bs, phys_bs, io_opt;
1826 if (ns->lba_shift > PAGE_SHIFT) {
1827 /* unsupported block size, set capacity to 0 later */
1830 blk_mq_freeze_queue(disk->queue);
1831 blk_integrity_unregister(disk);
1833 if (id->nabo == 0) {
1835 * Bit 1 indicates whether NAWUPF is defined for this namespace
1836 * and whether it should be used instead of AWUPF. If NAWUPF ==
1837 * 0 then AWUPF must be used instead.
1839 if (id->nsfeat & (1 << 1) && id->nawupf)
1840 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1842 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1848 if (id->nsfeat & (1 << 4)) {
1849 /* NPWG = Namespace Preferred Write Granularity */
1850 phys_bs *= 1 + le16_to_cpu(id->npwg);
1851 /* NOWS = Namespace Optimal Write Size */
1852 io_opt *= 1 + le16_to_cpu(id->nows);
1855 blk_queue_logical_block_size(disk->queue, bs);
1857 * Linux filesystems assume writing a single physical block is
1858 * an atomic operation. Hence limit the physical block size to the
1859 * value of the Atomic Write Unit Power Fail parameter.
1861 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1862 blk_queue_io_min(disk->queue, phys_bs);
1863 blk_queue_io_opt(disk->queue, io_opt);
1865 if (ns->ms && !ns->ext &&
1866 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1867 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1868 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1869 ns->lba_shift > PAGE_SHIFT)
1872 set_capacity_revalidate_and_notify(disk, capacity, false);
1874 nvme_config_discard(disk, ns);
1875 nvme_config_write_zeroes(disk, ns);
1877 if (id->nsattr & (1 << 0))
1878 set_disk_ro(disk, true);
1880 set_disk_ro(disk, false);
1882 blk_mq_unfreeze_queue(disk->queue);
1885 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1887 struct nvme_ns *ns = disk->private_data;
1890 * If identify namespace failed, use default 512 byte block size so
1891 * block layer can use before failing read/write for 0 capacity.
1893 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1894 if (ns->lba_shift == 0)
1896 ns->noiob = le16_to_cpu(id->noiob);
1897 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1898 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1899 /* the PI implementation requires metadata equal t10 pi tuple size */
1900 if (ns->ms == sizeof(struct t10_pi_tuple))
1901 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1906 nvme_set_chunk_size(ns);
1907 nvme_update_disk_info(disk, ns, id);
1908 #ifdef CONFIG_NVME_MULTIPATH
1909 if (ns->head->disk) {
1910 nvme_update_disk_info(ns->head->disk, ns, id);
1911 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1912 revalidate_disk(ns->head->disk);
1917 static int nvme_revalidate_disk(struct gendisk *disk)
1919 struct nvme_ns *ns = disk->private_data;
1920 struct nvme_ctrl *ctrl = ns->ctrl;
1921 struct nvme_id_ns *id;
1922 struct nvme_ns_ids ids;
1925 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1926 set_capacity(disk, 0);
1930 ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1934 if (id->ncap == 0) {
1939 ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1943 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1944 dev_err(ctrl->device,
1945 "identifiers changed for nsid %d\n", ns->head->ns_id);
1950 __nvme_revalidate_disk(disk, id);
1955 * Only fail the function if we got a fatal error back from the
1956 * device, otherwise ignore the error and just move on.
1958 if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1961 ret = blk_status_to_errno(nvme_error_status(ret));
1965 static char nvme_pr_type(enum pr_type type)
1968 case PR_WRITE_EXCLUSIVE:
1970 case PR_EXCLUSIVE_ACCESS:
1972 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1974 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1976 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1978 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1985 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1986 u64 key, u64 sa_key, u8 op)
1988 struct nvme_ns_head *head = NULL;
1990 struct nvme_command c;
1992 u8 data[16] = { 0, };
1994 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1996 return -EWOULDBLOCK;
1998 put_unaligned_le64(key, &data[0]);
1999 put_unaligned_le64(sa_key, &data[8]);
2001 memset(&c, 0, sizeof(c));
2002 c.common.opcode = op;
2003 c.common.nsid = cpu_to_le32(ns->head->ns_id);
2004 c.common.cdw10 = cpu_to_le32(cdw10);
2006 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2007 nvme_put_ns_from_disk(head, srcu_idx);
2011 static int nvme_pr_register(struct block_device *bdev, u64 old,
2012 u64 new, unsigned flags)
2016 if (flags & ~PR_FL_IGNORE_KEY)
2019 cdw10 = old ? 2 : 0;
2020 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2021 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2022 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2025 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2026 enum pr_type type, unsigned flags)
2030 if (flags & ~PR_FL_IGNORE_KEY)
2033 cdw10 = nvme_pr_type(type) << 8;
2034 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2035 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2038 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2039 enum pr_type type, bool abort)
2041 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2042 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2045 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2047 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2048 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2051 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2053 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2054 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2057 static const struct pr_ops nvme_pr_ops = {
2058 .pr_register = nvme_pr_register,
2059 .pr_reserve = nvme_pr_reserve,
2060 .pr_release = nvme_pr_release,
2061 .pr_preempt = nvme_pr_preempt,
2062 .pr_clear = nvme_pr_clear,
2065 #ifdef CONFIG_BLK_SED_OPAL
2066 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2069 struct nvme_ctrl *ctrl = data;
2070 struct nvme_command cmd;
2072 memset(&cmd, 0, sizeof(cmd));
2074 cmd.common.opcode = nvme_admin_security_send;
2076 cmd.common.opcode = nvme_admin_security_recv;
2077 cmd.common.nsid = 0;
2078 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2079 cmd.common.cdw11 = cpu_to_le32(len);
2081 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2082 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2084 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2085 #endif /* CONFIG_BLK_SED_OPAL */
2087 static const struct block_device_operations nvme_fops = {
2088 .owner = THIS_MODULE,
2089 .ioctl = nvme_ioctl,
2090 .compat_ioctl = nvme_compat_ioctl,
2092 .release = nvme_release,
2093 .getgeo = nvme_getgeo,
2094 .revalidate_disk= nvme_revalidate_disk,
2095 .pr_ops = &nvme_pr_ops,
2098 #ifdef CONFIG_NVME_MULTIPATH
2099 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2101 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2103 if (!kref_get_unless_zero(&head->ref))
2108 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2110 nvme_put_ns_head(disk->private_data);
2113 const struct block_device_operations nvme_ns_head_ops = {
2114 .owner = THIS_MODULE,
2115 .open = nvme_ns_head_open,
2116 .release = nvme_ns_head_release,
2117 .ioctl = nvme_ioctl,
2118 .compat_ioctl = nvme_compat_ioctl,
2119 .getgeo = nvme_getgeo,
2120 .pr_ops = &nvme_pr_ops,
2122 #endif /* CONFIG_NVME_MULTIPATH */
2124 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2126 unsigned long timeout =
2127 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2128 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2131 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2134 if ((csts & NVME_CSTS_RDY) == bit)
2137 usleep_range(1000, 2000);
2138 if (fatal_signal_pending(current))
2140 if (time_after(jiffies, timeout)) {
2141 dev_err(ctrl->device,
2142 "Device not ready; aborting %s, CSTS=0x%x\n",
2143 enabled ? "initialisation" : "reset", csts);
2152 * If the device has been passed off to us in an enabled state, just clear
2153 * the enabled bit. The spec says we should set the 'shutdown notification
2154 * bits', but doing so may cause the device to complete commands to the
2155 * admin queue ... and we don't know what memory that might be pointing at!
2157 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2161 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2162 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2164 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2168 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2169 msleep(NVME_QUIRK_DELAY_AMOUNT);
2171 return nvme_wait_ready(ctrl, ctrl->cap, false);
2173 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2175 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2178 * Default to a 4K page size, with the intention to update this
2179 * path in the future to accomodate architectures with differing
2180 * kernel and IO page sizes.
2182 unsigned dev_page_min, page_shift = 12;
2185 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2187 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2190 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2192 if (page_shift < dev_page_min) {
2193 dev_err(ctrl->device,
2194 "Minimum device page size %u too large for host (%u)\n",
2195 1 << dev_page_min, 1 << page_shift);
2199 ctrl->page_size = 1 << page_shift;
2201 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2202 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2203 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2204 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2205 ctrl->ctrl_config |= NVME_CC_ENABLE;
2207 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2210 return nvme_wait_ready(ctrl, ctrl->cap, true);
2212 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2214 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2216 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2220 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2221 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2223 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2227 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2228 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2232 if (fatal_signal_pending(current))
2234 if (time_after(jiffies, timeout)) {
2235 dev_err(ctrl->device,
2236 "Device shutdown incomplete; abort shutdown\n");
2243 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2245 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2246 struct request_queue *q)
2250 if (ctrl->max_hw_sectors) {
2252 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2254 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2255 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2256 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2258 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2259 is_power_of_2(ctrl->max_hw_sectors))
2260 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2261 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2262 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2264 blk_queue_write_cache(q, vwc, vwc);
2267 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2272 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2275 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2276 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2279 dev_warn_once(ctrl->device,
2280 "could not set timestamp (%d)\n", ret);
2284 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2286 struct nvme_feat_host_behavior *host;
2289 /* Don't bother enabling the feature if retry delay is not reported */
2293 host = kzalloc(sizeof(*host), GFP_KERNEL);
2297 host->acre = NVME_ENABLE_ACRE;
2298 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2299 host, sizeof(*host), NULL);
2304 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2307 * APST (Autonomous Power State Transition) lets us program a
2308 * table of power state transitions that the controller will
2309 * perform automatically. We configure it with a simple
2310 * heuristic: we are willing to spend at most 2% of the time
2311 * transitioning between power states. Therefore, when running
2312 * in any given state, we will enter the next lower-power
2313 * non-operational state after waiting 50 * (enlat + exlat)
2314 * microseconds, as long as that state's exit latency is under
2315 * the requested maximum latency.
2317 * We will not autonomously enter any non-operational state for
2318 * which the total latency exceeds ps_max_latency_us. Users
2319 * can set ps_max_latency_us to zero to turn off APST.
2323 struct nvme_feat_auto_pst *table;
2329 * If APST isn't supported or if we haven't been initialized yet,
2330 * then don't do anything.
2335 if (ctrl->npss > 31) {
2336 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2340 table = kzalloc(sizeof(*table), GFP_KERNEL);
2344 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2345 /* Turn off APST. */
2347 dev_dbg(ctrl->device, "APST disabled\n");
2349 __le64 target = cpu_to_le64(0);
2353 * Walk through all states from lowest- to highest-power.
2354 * According to the spec, lower-numbered states use more
2355 * power. NPSS, despite the name, is the index of the
2356 * lowest-power state, not the number of states.
2358 for (state = (int)ctrl->npss; state >= 0; state--) {
2359 u64 total_latency_us, exit_latency_us, transition_ms;
2362 table->entries[state] = target;
2365 * Don't allow transitions to the deepest state
2366 * if it's quirked off.
2368 if (state == ctrl->npss &&
2369 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2373 * Is this state a useful non-operational state for
2374 * higher-power states to autonomously transition to?
2376 if (!(ctrl->psd[state].flags &
2377 NVME_PS_FLAGS_NON_OP_STATE))
2381 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2382 if (exit_latency_us > ctrl->ps_max_latency_us)
2387 le32_to_cpu(ctrl->psd[state].entry_lat);
2390 * This state is good. Use it as the APST idle
2391 * target for higher power states.
2393 transition_ms = total_latency_us + 19;
2394 do_div(transition_ms, 20);
2395 if (transition_ms > (1 << 24) - 1)
2396 transition_ms = (1 << 24) - 1;
2398 target = cpu_to_le64((state << 3) |
2399 (transition_ms << 8));
2404 if (total_latency_us > max_lat_us)
2405 max_lat_us = total_latency_us;
2411 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2413 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2414 max_ps, max_lat_us, (int)sizeof(*table), table);
2418 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2419 table, sizeof(*table), NULL);
2421 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2427 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2429 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2433 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2434 case PM_QOS_LATENCY_ANY:
2442 if (ctrl->ps_max_latency_us != latency) {
2443 ctrl->ps_max_latency_us = latency;
2444 nvme_configure_apst(ctrl);
2448 struct nvme_core_quirk_entry {
2450 * NVMe model and firmware strings are padded with spaces. For
2451 * simplicity, strings in the quirk table are padded with NULLs
2457 unsigned long quirks;
2460 static const struct nvme_core_quirk_entry core_quirks[] = {
2463 * This Toshiba device seems to die using any APST states. See:
2464 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2467 .mn = "THNSF5256GPUK TOSHIBA",
2468 .quirks = NVME_QUIRK_NO_APST,
2472 * This LiteON CL1-3D*-Q11 firmware version has a race
2473 * condition associated with actions related to suspend to idle
2474 * LiteON has resolved the problem in future firmware
2478 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2482 /* match is null-terminated but idstr is space-padded. */
2483 static bool string_matches(const char *idstr, const char *match, size_t len)
2490 matchlen = strlen(match);
2491 WARN_ON_ONCE(matchlen > len);
2493 if (memcmp(idstr, match, matchlen))
2496 for (; matchlen < len; matchlen++)
2497 if (idstr[matchlen] != ' ')
2503 static bool quirk_matches(const struct nvme_id_ctrl *id,
2504 const struct nvme_core_quirk_entry *q)
2506 return q->vid == le16_to_cpu(id->vid) &&
2507 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2508 string_matches(id->fr, q->fr, sizeof(id->fr));
2511 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2512 struct nvme_id_ctrl *id)
2517 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2518 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2519 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2520 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2524 if (ctrl->vs >= NVME_VS(1, 2, 1))
2525 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2528 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2529 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2530 "nqn.2014.08.org.nvmexpress:%04x%04x",
2531 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2532 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2533 off += sizeof(id->sn);
2534 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2535 off += sizeof(id->mn);
2536 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2539 static void nvme_release_subsystem(struct device *dev)
2541 struct nvme_subsystem *subsys =
2542 container_of(dev, struct nvme_subsystem, dev);
2544 if (subsys->instance >= 0)
2545 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2549 static void nvme_destroy_subsystem(struct kref *ref)
2551 struct nvme_subsystem *subsys =
2552 container_of(ref, struct nvme_subsystem, ref);
2554 mutex_lock(&nvme_subsystems_lock);
2555 list_del(&subsys->entry);
2556 mutex_unlock(&nvme_subsystems_lock);
2558 ida_destroy(&subsys->ns_ida);
2559 device_del(&subsys->dev);
2560 put_device(&subsys->dev);
2563 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2565 kref_put(&subsys->ref, nvme_destroy_subsystem);
2568 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2570 struct nvme_subsystem *subsys;
2572 lockdep_assert_held(&nvme_subsystems_lock);
2575 * Fail matches for discovery subsystems. This results
2576 * in each discovery controller bound to a unique subsystem.
2577 * This avoids issues with validating controller values
2578 * that can only be true when there is a single unique subsystem.
2579 * There may be multiple and completely independent entities
2580 * that provide discovery controllers.
2582 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2585 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2586 if (strcmp(subsys->subnqn, subsysnqn))
2588 if (!kref_get_unless_zero(&subsys->ref))
2596 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2597 struct device_attribute subsys_attr_##_name = \
2598 __ATTR(_name, _mode, _show, NULL)
2600 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2601 struct device_attribute *attr,
2604 struct nvme_subsystem *subsys =
2605 container_of(dev, struct nvme_subsystem, dev);
2607 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2609 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2611 #define nvme_subsys_show_str_function(field) \
2612 static ssize_t subsys_##field##_show(struct device *dev, \
2613 struct device_attribute *attr, char *buf) \
2615 struct nvme_subsystem *subsys = \
2616 container_of(dev, struct nvme_subsystem, dev); \
2617 return sprintf(buf, "%.*s\n", \
2618 (int)sizeof(subsys->field), subsys->field); \
2620 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2622 nvme_subsys_show_str_function(model);
2623 nvme_subsys_show_str_function(serial);
2624 nvme_subsys_show_str_function(firmware_rev);
2626 static struct attribute *nvme_subsys_attrs[] = {
2627 &subsys_attr_model.attr,
2628 &subsys_attr_serial.attr,
2629 &subsys_attr_firmware_rev.attr,
2630 &subsys_attr_subsysnqn.attr,
2631 #ifdef CONFIG_NVME_MULTIPATH
2632 &subsys_attr_iopolicy.attr,
2637 static struct attribute_group nvme_subsys_attrs_group = {
2638 .attrs = nvme_subsys_attrs,
2641 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2642 &nvme_subsys_attrs_group,
2646 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2647 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2649 struct nvme_ctrl *tmp;
2651 lockdep_assert_held(&nvme_subsystems_lock);
2653 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2654 if (nvme_state_terminal(tmp))
2657 if (tmp->cntlid == ctrl->cntlid) {
2658 dev_err(ctrl->device,
2659 "Duplicate cntlid %u with %s, rejecting\n",
2660 ctrl->cntlid, dev_name(tmp->device));
2664 if ((id->cmic & (1 << 1)) ||
2665 (ctrl->opts && ctrl->opts->discovery_nqn))
2668 dev_err(ctrl->device,
2669 "Subsystem does not support multiple controllers\n");
2676 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2678 struct nvme_subsystem *subsys, *found;
2681 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2685 subsys->instance = -1;
2686 mutex_init(&subsys->lock);
2687 kref_init(&subsys->ref);
2688 INIT_LIST_HEAD(&subsys->ctrls);
2689 INIT_LIST_HEAD(&subsys->nsheads);
2690 nvme_init_subnqn(subsys, ctrl, id);
2691 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2692 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2693 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2694 subsys->vendor_id = le16_to_cpu(id->vid);
2695 subsys->cmic = id->cmic;
2696 subsys->awupf = le16_to_cpu(id->awupf);
2697 #ifdef CONFIG_NVME_MULTIPATH
2698 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2701 subsys->dev.class = nvme_subsys_class;
2702 subsys->dev.release = nvme_release_subsystem;
2703 subsys->dev.groups = nvme_subsys_attrs_groups;
2704 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2705 device_initialize(&subsys->dev);
2707 mutex_lock(&nvme_subsystems_lock);
2708 found = __nvme_find_get_subsystem(subsys->subnqn);
2710 put_device(&subsys->dev);
2713 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2715 goto out_put_subsystem;
2718 ret = device_add(&subsys->dev);
2720 dev_err(ctrl->device,
2721 "failed to register subsystem device.\n");
2722 put_device(&subsys->dev);
2725 ida_init(&subsys->ns_ida);
2726 list_add_tail(&subsys->entry, &nvme_subsystems);
2729 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2730 dev_name(ctrl->device));
2732 dev_err(ctrl->device,
2733 "failed to create sysfs link from subsystem.\n");
2734 goto out_put_subsystem;
2738 subsys->instance = ctrl->instance;
2739 ctrl->subsys = subsys;
2740 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2741 mutex_unlock(&nvme_subsystems_lock);
2745 nvme_put_subsystem(subsys);
2747 mutex_unlock(&nvme_subsystems_lock);
2751 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2752 void *log, size_t size, u64 offset)
2754 struct nvme_command c = { };
2755 u32 dwlen = nvme_bytes_to_numd(size);
2757 c.get_log_page.opcode = nvme_admin_get_log_page;
2758 c.get_log_page.nsid = cpu_to_le32(nsid);
2759 c.get_log_page.lid = log_page;
2760 c.get_log_page.lsp = lsp;
2761 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2762 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2763 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2764 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2766 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2769 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2774 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2779 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2780 ctrl->effects, sizeof(*ctrl->effects), 0);
2782 kfree(ctrl->effects);
2783 ctrl->effects = NULL;
2789 * Initialize the cached copies of the Identify data and various controller
2790 * register in our nvme_ctrl structure. This should be called as soon as
2791 * the admin queue is fully up and running.
2793 int nvme_init_identify(struct nvme_ctrl *ctrl)
2795 struct nvme_id_ctrl *id;
2796 int ret, page_shift;
2798 bool prev_apst_enabled;
2800 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2802 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2805 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2806 ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2808 if (ctrl->vs >= NVME_VS(1, 1, 0))
2809 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2811 ret = nvme_identify_ctrl(ctrl, &id);
2813 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2817 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2818 ret = nvme_get_effects_log(ctrl);
2823 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2824 ctrl->cntlid = le16_to_cpu(id->cntlid);
2826 if (!ctrl->identified) {
2829 ret = nvme_init_subsystem(ctrl, id);
2834 * Check for quirks. Quirk can depend on firmware version,
2835 * so, in principle, the set of quirks present can change
2836 * across a reset. As a possible future enhancement, we
2837 * could re-scan for quirks every time we reinitialize
2838 * the device, but we'd have to make sure that the driver
2839 * behaves intelligently if the quirks change.
2841 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2842 if (quirk_matches(id, &core_quirks[i]))
2843 ctrl->quirks |= core_quirks[i].quirks;
2847 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2848 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2849 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2852 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2853 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2854 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2856 ctrl->oacs = le16_to_cpu(id->oacs);
2857 ctrl->oncs = le16_to_cpu(id->oncs);
2858 ctrl->mtfa = le16_to_cpu(id->mtfa);
2859 ctrl->oaes = le32_to_cpu(id->oaes);
2860 ctrl->wctemp = le16_to_cpu(id->wctemp);
2861 ctrl->cctemp = le16_to_cpu(id->cctemp);
2863 atomic_set(&ctrl->abort_limit, id->acl + 1);
2864 ctrl->vwc = id->vwc;
2866 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2868 max_hw_sectors = UINT_MAX;
2869 ctrl->max_hw_sectors =
2870 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2872 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2873 ctrl->sgls = le32_to_cpu(id->sgls);
2874 ctrl->kas = le16_to_cpu(id->kas);
2875 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2876 ctrl->ctratt = le32_to_cpu(id->ctratt);
2880 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2882 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2883 shutdown_timeout, 60);
2885 if (ctrl->shutdown_timeout != shutdown_timeout)
2886 dev_info(ctrl->device,
2887 "Shutdown timeout set to %u seconds\n",
2888 ctrl->shutdown_timeout);
2890 ctrl->shutdown_timeout = shutdown_timeout;
2892 ctrl->npss = id->npss;
2893 ctrl->apsta = id->apsta;
2894 prev_apst_enabled = ctrl->apst_enabled;
2895 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2896 if (force_apst && id->apsta) {
2897 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2898 ctrl->apst_enabled = true;
2900 ctrl->apst_enabled = false;
2903 ctrl->apst_enabled = id->apsta;
2905 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2907 if (ctrl->ops->flags & NVME_F_FABRICS) {
2908 ctrl->icdoff = le16_to_cpu(id->icdoff);
2909 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2910 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2911 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2914 * In fabrics we need to verify the cntlid matches the
2917 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2918 dev_err(ctrl->device,
2919 "Mismatching cntlid: Connect %u vs Identify "
2921 ctrl->cntlid, le16_to_cpu(id->cntlid));
2926 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2927 dev_err(ctrl->device,
2928 "keep-alive support is mandatory for fabrics\n");
2933 ctrl->hmpre = le32_to_cpu(id->hmpre);
2934 ctrl->hmmin = le32_to_cpu(id->hmmin);
2935 ctrl->hmminds = le32_to_cpu(id->hmminds);
2936 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2939 ret = nvme_mpath_init(ctrl, id);
2945 if (ctrl->apst_enabled && !prev_apst_enabled)
2946 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2947 else if (!ctrl->apst_enabled && prev_apst_enabled)
2948 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2950 ret = nvme_configure_apst(ctrl);
2954 ret = nvme_configure_timestamp(ctrl);
2958 ret = nvme_configure_directives(ctrl);
2962 ret = nvme_configure_acre(ctrl);
2966 if (!ctrl->identified)
2967 nvme_hwmon_init(ctrl);
2969 ctrl->identified = true;
2977 EXPORT_SYMBOL_GPL(nvme_init_identify);
2979 static int nvme_dev_open(struct inode *inode, struct file *file)
2981 struct nvme_ctrl *ctrl =
2982 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2984 switch (ctrl->state) {
2985 case NVME_CTRL_LIVE:
2988 return -EWOULDBLOCK;
2991 file->private_data = ctrl;
2995 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3000 down_read(&ctrl->namespaces_rwsem);
3001 if (list_empty(&ctrl->namespaces)) {
3006 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3007 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3008 dev_warn(ctrl->device,
3009 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3014 dev_warn(ctrl->device,
3015 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3016 kref_get(&ns->kref);
3017 up_read(&ctrl->namespaces_rwsem);
3019 ret = nvme_user_cmd(ctrl, ns, argp);
3024 up_read(&ctrl->namespaces_rwsem);
3028 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3031 struct nvme_ctrl *ctrl = file->private_data;
3032 void __user *argp = (void __user *)arg;
3035 case NVME_IOCTL_ADMIN_CMD:
3036 return nvme_user_cmd(ctrl, NULL, argp);
3037 case NVME_IOCTL_ADMIN64_CMD:
3038 return nvme_user_cmd64(ctrl, NULL, argp);
3039 case NVME_IOCTL_IO_CMD:
3040 return nvme_dev_user_cmd(ctrl, argp);
3041 case NVME_IOCTL_RESET:
3042 dev_warn(ctrl->device, "resetting controller\n");
3043 return nvme_reset_ctrl_sync(ctrl);
3044 case NVME_IOCTL_SUBSYS_RESET:
3045 return nvme_reset_subsystem(ctrl);
3046 case NVME_IOCTL_RESCAN:
3047 nvme_queue_scan(ctrl);
3054 static const struct file_operations nvme_dev_fops = {
3055 .owner = THIS_MODULE,
3056 .open = nvme_dev_open,
3057 .unlocked_ioctl = nvme_dev_ioctl,
3058 .compat_ioctl = compat_ptr_ioctl,
3061 static ssize_t nvme_sysfs_reset(struct device *dev,
3062 struct device_attribute *attr, const char *buf,
3065 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3068 ret = nvme_reset_ctrl_sync(ctrl);
3073 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3075 static ssize_t nvme_sysfs_rescan(struct device *dev,
3076 struct device_attribute *attr, const char *buf,
3079 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3081 nvme_queue_scan(ctrl);
3084 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3086 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3088 struct gendisk *disk = dev_to_disk(dev);
3090 if (disk->fops == &nvme_fops)
3091 return nvme_get_ns_from_dev(dev)->head;
3093 return disk->private_data;
3096 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3099 struct nvme_ns_head *head = dev_to_ns_head(dev);
3100 struct nvme_ns_ids *ids = &head->ids;
3101 struct nvme_subsystem *subsys = head->subsys;
3102 int serial_len = sizeof(subsys->serial);
3103 int model_len = sizeof(subsys->model);
3105 if (!uuid_is_null(&ids->uuid))
3106 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3108 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3109 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3111 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3112 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3114 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3115 subsys->serial[serial_len - 1] == '\0'))
3117 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3118 subsys->model[model_len - 1] == '\0'))
3121 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3122 serial_len, subsys->serial, model_len, subsys->model,
3125 static DEVICE_ATTR_RO(wwid);
3127 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3130 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3132 static DEVICE_ATTR_RO(nguid);
3134 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3137 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3139 /* For backward compatibility expose the NGUID to userspace if
3140 * we have no UUID set
3142 if (uuid_is_null(&ids->uuid)) {
3143 printk_ratelimited(KERN_WARNING
3144 "No UUID available providing old NGUID\n");
3145 return sprintf(buf, "%pU\n", ids->nguid);
3147 return sprintf(buf, "%pU\n", &ids->uuid);
3149 static DEVICE_ATTR_RO(uuid);
3151 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3154 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3156 static DEVICE_ATTR_RO(eui);
3158 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3161 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3163 static DEVICE_ATTR_RO(nsid);
3165 static struct attribute *nvme_ns_id_attrs[] = {
3166 &dev_attr_wwid.attr,
3167 &dev_attr_uuid.attr,
3168 &dev_attr_nguid.attr,
3170 &dev_attr_nsid.attr,
3171 #ifdef CONFIG_NVME_MULTIPATH
3172 &dev_attr_ana_grpid.attr,
3173 &dev_attr_ana_state.attr,
3178 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3179 struct attribute *a, int n)
3181 struct device *dev = container_of(kobj, struct device, kobj);
3182 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3184 if (a == &dev_attr_uuid.attr) {
3185 if (uuid_is_null(&ids->uuid) &&
3186 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3189 if (a == &dev_attr_nguid.attr) {
3190 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3193 if (a == &dev_attr_eui.attr) {
3194 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3197 #ifdef CONFIG_NVME_MULTIPATH
3198 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3199 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3201 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3208 static const struct attribute_group nvme_ns_id_attr_group = {
3209 .attrs = nvme_ns_id_attrs,
3210 .is_visible = nvme_ns_id_attrs_are_visible,
3213 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3214 &nvme_ns_id_attr_group,
3216 &nvme_nvm_attr_group,
3221 #define nvme_show_str_function(field) \
3222 static ssize_t field##_show(struct device *dev, \
3223 struct device_attribute *attr, char *buf) \
3225 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3226 return sprintf(buf, "%.*s\n", \
3227 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3229 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3231 nvme_show_str_function(model);
3232 nvme_show_str_function(serial);
3233 nvme_show_str_function(firmware_rev);
3235 #define nvme_show_int_function(field) \
3236 static ssize_t field##_show(struct device *dev, \
3237 struct device_attribute *attr, char *buf) \
3239 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3240 return sprintf(buf, "%d\n", ctrl->field); \
3242 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3244 nvme_show_int_function(cntlid);
3245 nvme_show_int_function(numa_node);
3246 nvme_show_int_function(queue_count);
3247 nvme_show_int_function(sqsize);
3249 static ssize_t nvme_sysfs_delete(struct device *dev,
3250 struct device_attribute *attr, const char *buf,
3253 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3255 /* Can't delete non-created controllers */
3259 if (device_remove_file_self(dev, attr))
3260 nvme_delete_ctrl_sync(ctrl);
3263 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3265 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3266 struct device_attribute *attr,
3269 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3271 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3273 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3275 static ssize_t nvme_sysfs_show_state(struct device *dev,
3276 struct device_attribute *attr,
3279 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3280 static const char *const state_name[] = {
3281 [NVME_CTRL_NEW] = "new",
3282 [NVME_CTRL_LIVE] = "live",
3283 [NVME_CTRL_RESETTING] = "resetting",
3284 [NVME_CTRL_CONNECTING] = "connecting",
3285 [NVME_CTRL_DELETING] = "deleting",
3286 [NVME_CTRL_DEAD] = "dead",
3289 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3290 state_name[ctrl->state])
3291 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3293 return sprintf(buf, "unknown state\n");
3296 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3298 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3299 struct device_attribute *attr,
3302 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3304 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3306 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3308 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3309 struct device_attribute *attr,
3312 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3314 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3316 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3318 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3319 struct device_attribute *attr,
3322 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3324 return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3326 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3328 static ssize_t nvme_sysfs_show_address(struct device *dev,
3329 struct device_attribute *attr,
3332 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3334 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3336 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3338 static struct attribute *nvme_dev_attrs[] = {
3339 &dev_attr_reset_controller.attr,
3340 &dev_attr_rescan_controller.attr,
3341 &dev_attr_model.attr,
3342 &dev_attr_serial.attr,
3343 &dev_attr_firmware_rev.attr,
3344 &dev_attr_cntlid.attr,
3345 &dev_attr_delete_controller.attr,
3346 &dev_attr_transport.attr,
3347 &dev_attr_subsysnqn.attr,
3348 &dev_attr_address.attr,
3349 &dev_attr_state.attr,
3350 &dev_attr_numa_node.attr,
3351 &dev_attr_queue_count.attr,
3352 &dev_attr_sqsize.attr,
3353 &dev_attr_hostnqn.attr,
3354 &dev_attr_hostid.attr,
3358 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3359 struct attribute *a, int n)
3361 struct device *dev = container_of(kobj, struct device, kobj);
3362 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3364 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3366 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3368 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3370 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3376 static struct attribute_group nvme_dev_attrs_group = {
3377 .attrs = nvme_dev_attrs,
3378 .is_visible = nvme_dev_attrs_are_visible,
3381 static const struct attribute_group *nvme_dev_attr_groups[] = {
3382 &nvme_dev_attrs_group,
3386 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3389 struct nvme_ns_head *h;
3391 lockdep_assert_held(&subsys->lock);
3393 list_for_each_entry(h, &subsys->nsheads, entry) {
3394 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3401 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3402 struct nvme_ns_head *new)
3404 struct nvme_ns_head *h;
3406 lockdep_assert_held(&subsys->lock);
3408 list_for_each_entry(h, &subsys->nsheads, entry) {
3409 if (nvme_ns_ids_valid(&new->ids) &&
3410 nvme_ns_ids_equal(&new->ids, &h->ids))
3417 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3418 unsigned nsid, struct nvme_ns_ids *ids)
3420 struct nvme_ns_head *head;
3421 size_t size = sizeof(*head);
3424 #ifdef CONFIG_NVME_MULTIPATH
3425 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3428 head = kzalloc(size, GFP_KERNEL);
3431 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3434 head->instance = ret;
3435 INIT_LIST_HEAD(&head->list);
3436 ret = init_srcu_struct(&head->srcu);
3438 goto out_ida_remove;
3439 head->subsys = ctrl->subsys;
3442 kref_init(&head->ref);
3444 ret = __nvme_check_ids(ctrl->subsys, head);
3446 dev_err(ctrl->device,
3447 "duplicate IDs for nsid %d\n", nsid);
3448 goto out_cleanup_srcu;
3451 ret = nvme_mpath_alloc_disk(ctrl, head);
3453 goto out_cleanup_srcu;
3455 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3457 kref_get(&ctrl->subsys->ref);
3461 cleanup_srcu_struct(&head->srcu);
3463 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3468 ret = blk_status_to_errno(nvme_error_status(ret));
3469 return ERR_PTR(ret);
3472 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3473 struct nvme_id_ns *id)
3475 struct nvme_ctrl *ctrl = ns->ctrl;
3476 bool is_shared = id->nmic & (1 << 0);
3477 struct nvme_ns_head *head = NULL;
3478 struct nvme_ns_ids ids;
3481 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3485 mutex_lock(&ctrl->subsys->lock);
3486 head = nvme_find_ns_head(ctrl->subsys, nsid);
3488 head = nvme_alloc_ns_head(ctrl, nsid, &ids);
3490 ret = PTR_ERR(head);
3493 head->shared = is_shared;
3495 if (!is_shared || !head->shared) {
3496 dev_err(ctrl->device,
3497 "Duplicate unshared namespace %d\n",
3500 nvme_put_ns_head(head);
3503 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3504 dev_err(ctrl->device,
3505 "IDs don't match for shared namespace %d\n",
3508 nvme_put_ns_head(head);
3513 list_add_tail(&ns->siblings, &head->list);
3517 mutex_unlock(&ctrl->subsys->lock);
3520 ret = blk_status_to_errno(nvme_error_status(ret));
3524 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3526 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3527 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3529 return nsa->head->ns_id - nsb->head->ns_id;
3532 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3534 struct nvme_ns *ns, *ret = NULL;
3536 down_read(&ctrl->namespaces_rwsem);
3537 list_for_each_entry(ns, &ctrl->namespaces, list) {
3538 if (ns->head->ns_id == nsid) {
3539 if (!kref_get_unless_zero(&ns->kref))
3544 if (ns->head->ns_id > nsid)
3547 up_read(&ctrl->namespaces_rwsem);
3551 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3553 struct streams_directive_params s;
3556 if (!ctrl->nr_streams)
3559 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3563 ns->sws = le32_to_cpu(s.sws);
3564 ns->sgs = le16_to_cpu(s.sgs);
3567 unsigned int bs = 1 << ns->lba_shift;
3569 blk_queue_io_min(ns->queue, bs * ns->sws);
3571 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3577 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3580 struct gendisk *disk;
3581 struct nvme_id_ns *id;
3582 char disk_name[DISK_NAME_LEN];
3583 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3585 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3589 ns->queue = blk_mq_init_queue(ctrl->tagset);
3590 if (IS_ERR(ns->queue))
3593 if (ctrl->opts && ctrl->opts->data_digest)
3594 ns->queue->backing_dev_info->capabilities
3595 |= BDI_CAP_STABLE_WRITES;
3597 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3598 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3599 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3601 ns->queue->queuedata = ns;
3604 kref_init(&ns->kref);
3605 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3607 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3608 nvme_set_queue_limits(ctrl, ns->queue);
3610 ret = nvme_identify_ns(ctrl, nsid, &id);
3612 goto out_free_queue;
3614 if (id->ncap == 0) /* no namespace (legacy quirk) */
3617 ret = nvme_init_ns_head(ns, nsid, id);
3620 nvme_setup_streams_ns(ctrl, ns);
3621 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3623 disk = alloc_disk_node(0, node);
3627 disk->fops = &nvme_fops;
3628 disk->private_data = ns;
3629 disk->queue = ns->queue;
3630 disk->flags = flags;
3631 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3634 __nvme_revalidate_disk(disk, id);
3636 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3637 ret = nvme_nvm_register(ns, disk_name, node);
3639 dev_warn(ctrl->device, "LightNVM init failure\n");
3644 down_write(&ctrl->namespaces_rwsem);
3645 list_add_tail(&ns->list, &ctrl->namespaces);
3646 up_write(&ctrl->namespaces_rwsem);
3648 nvme_get_ctrl(ctrl);
3650 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3652 nvme_mpath_add_disk(ns, id);
3653 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3658 /* prevent double queue cleanup */
3659 ns->disk->queue = NULL;
3662 mutex_lock(&ctrl->subsys->lock);
3663 list_del_rcu(&ns->siblings);
3664 if (list_empty(&ns->head->list))
3665 list_del_init(&ns->head->entry);
3666 mutex_unlock(&ctrl->subsys->lock);
3667 nvme_put_ns_head(ns->head);
3671 blk_cleanup_queue(ns->queue);
3676 static void nvme_ns_remove(struct nvme_ns *ns)
3678 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3681 nvme_fault_inject_fini(&ns->fault_inject);
3683 mutex_lock(&ns->ctrl->subsys->lock);
3684 list_del_rcu(&ns->siblings);
3685 if (list_empty(&ns->head->list))
3686 list_del_init(&ns->head->entry);
3687 mutex_unlock(&ns->ctrl->subsys->lock);
3689 synchronize_rcu(); /* guarantee not available in head->list */
3690 nvme_mpath_clear_current_path(ns);
3691 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3693 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3694 del_gendisk(ns->disk);
3695 blk_cleanup_queue(ns->queue);
3696 if (blk_get_integrity(ns->disk))
3697 blk_integrity_unregister(ns->disk);
3700 down_write(&ns->ctrl->namespaces_rwsem);
3701 list_del_init(&ns->list);
3702 up_write(&ns->ctrl->namespaces_rwsem);
3704 nvme_mpath_check_last_path(ns);
3708 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3710 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3718 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3722 ns = nvme_find_get_ns(ctrl, nsid);
3724 if (ns->disk && revalidate_disk(ns->disk))
3728 nvme_alloc_ns(ctrl, nsid);
3731 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3734 struct nvme_ns *ns, *next;
3737 down_write(&ctrl->namespaces_rwsem);
3738 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3739 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3740 list_move_tail(&ns->list, &rm_list);
3742 up_write(&ctrl->namespaces_rwsem);
3744 list_for_each_entry_safe(ns, next, &rm_list, list)
3749 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3751 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3756 if (nvme_ctrl_limited_cns(ctrl))
3759 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3764 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3768 for (i = 0; i < nr_entries; i++) {
3769 u32 nsid = le32_to_cpu(ns_list[i]);
3771 if (!nsid) /* end of the list? */
3773 nvme_validate_ns(ctrl, nsid);
3774 while (++prev < nsid)
3775 nvme_ns_remove_by_nsid(ctrl, prev);
3779 nvme_remove_invalid_namespaces(ctrl, prev);
3785 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3787 struct nvme_id_ctrl *id;
3790 if (nvme_identify_ctrl(ctrl, &id))
3792 nn = le32_to_cpu(id->nn);
3795 for (i = 1; i <= nn; i++)
3796 nvme_validate_ns(ctrl, i);
3798 nvme_remove_invalid_namespaces(ctrl, nn);
3801 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3803 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3807 log = kzalloc(log_size, GFP_KERNEL);
3812 * We need to read the log to clear the AEN, but we don't want to rely
3813 * on it for the changed namespace information as userspace could have
3814 * raced with us in reading the log page, which could cause us to miss
3817 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3820 dev_warn(ctrl->device,
3821 "reading changed ns log failed: %d\n", error);
3826 static void nvme_scan_work(struct work_struct *work)
3828 struct nvme_ctrl *ctrl =
3829 container_of(work, struct nvme_ctrl, scan_work);
3831 /* No tagset on a live ctrl means IO queues could not created */
3832 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3835 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3836 dev_info(ctrl->device, "rescanning namespaces.\n");
3837 nvme_clear_changed_ns_log(ctrl);
3840 mutex_lock(&ctrl->scan_lock);
3841 if (nvme_scan_ns_list(ctrl) != 0)
3842 nvme_scan_ns_sequential(ctrl);
3843 mutex_unlock(&ctrl->scan_lock);
3845 down_write(&ctrl->namespaces_rwsem);
3846 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3847 up_write(&ctrl->namespaces_rwsem);
3851 * This function iterates the namespace list unlocked to allow recovery from
3852 * controller failure. It is up to the caller to ensure the namespace list is
3853 * not modified by scan work while this function is executing.
3855 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3857 struct nvme_ns *ns, *next;
3861 * make sure to requeue I/O to all namespaces as these
3862 * might result from the scan itself and must complete
3863 * for the scan_work to make progress
3865 nvme_mpath_clear_ctrl_paths(ctrl);
3867 /* prevent racing with ns scanning */
3868 flush_work(&ctrl->scan_work);
3871 * The dead states indicates the controller was not gracefully
3872 * disconnected. In that case, we won't be able to flush any data while
3873 * removing the namespaces' disks; fail all the queues now to avoid
3874 * potentially having to clean up the failed sync later.
3876 if (ctrl->state == NVME_CTRL_DEAD)
3877 nvme_kill_queues(ctrl);
3879 down_write(&ctrl->namespaces_rwsem);
3880 list_splice_init(&ctrl->namespaces, &ns_list);
3881 up_write(&ctrl->namespaces_rwsem);
3883 list_for_each_entry_safe(ns, next, &ns_list, list)
3886 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3888 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3890 struct nvme_ctrl *ctrl =
3891 container_of(dev, struct nvme_ctrl, ctrl_device);
3892 struct nvmf_ctrl_options *opts = ctrl->opts;
3895 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3900 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3904 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3905 opts->trsvcid ?: "none");
3909 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3910 opts->host_traddr ?: "none");
3915 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3917 char *envp[2] = { NULL, NULL };
3918 u32 aen_result = ctrl->aen_result;
3920 ctrl->aen_result = 0;
3924 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3927 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3931 static void nvme_async_event_work(struct work_struct *work)
3933 struct nvme_ctrl *ctrl =
3934 container_of(work, struct nvme_ctrl, async_event_work);
3936 nvme_aen_uevent(ctrl);
3937 ctrl->ops->submit_async_event(ctrl);
3940 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3945 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3951 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3954 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3956 struct nvme_fw_slot_info_log *log;
3958 log = kmalloc(sizeof(*log), GFP_KERNEL);
3962 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, log,
3964 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3968 static void nvme_fw_act_work(struct work_struct *work)
3970 struct nvme_ctrl *ctrl = container_of(work,
3971 struct nvme_ctrl, fw_act_work);
3972 unsigned long fw_act_timeout;
3975 fw_act_timeout = jiffies +
3976 msecs_to_jiffies(ctrl->mtfa * 100);
3978 fw_act_timeout = jiffies +
3979 msecs_to_jiffies(admin_timeout * 1000);
3981 nvme_stop_queues(ctrl);
3982 while (nvme_ctrl_pp_status(ctrl)) {
3983 if (time_after(jiffies, fw_act_timeout)) {
3984 dev_warn(ctrl->device,
3985 "Fw activation timeout, reset controller\n");
3986 nvme_try_sched_reset(ctrl);
3992 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
3995 nvme_start_queues(ctrl);
3996 /* read FW slot information to clear the AER */
3997 nvme_get_fw_slot_info(ctrl);
4000 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4002 u32 aer_notice_type = (result & 0xff00) >> 8;
4004 trace_nvme_async_event(ctrl, aer_notice_type);
4006 switch (aer_notice_type) {
4007 case NVME_AER_NOTICE_NS_CHANGED:
4008 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4009 nvme_queue_scan(ctrl);
4011 case NVME_AER_NOTICE_FW_ACT_STARTING:
4013 * We are (ab)using the RESETTING state to prevent subsequent
4014 * recovery actions from interfering with the controller's
4015 * firmware activation.
4017 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4018 queue_work(nvme_wq, &ctrl->fw_act_work);
4020 #ifdef CONFIG_NVME_MULTIPATH
4021 case NVME_AER_NOTICE_ANA:
4022 if (!ctrl->ana_log_buf)
4024 queue_work(nvme_wq, &ctrl->ana_work);
4027 case NVME_AER_NOTICE_DISC_CHANGED:
4028 ctrl->aen_result = result;
4031 dev_warn(ctrl->device, "async event result %08x\n", result);
4035 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4036 volatile union nvme_result *res)
4038 u32 result = le32_to_cpu(res->u32);
4039 u32 aer_type = result & 0x07;
4041 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4045 case NVME_AER_NOTICE:
4046 nvme_handle_aen_notice(ctrl, result);
4048 case NVME_AER_ERROR:
4049 case NVME_AER_SMART:
4052 trace_nvme_async_event(ctrl, aer_type);
4053 ctrl->aen_result = result;
4058 queue_work(nvme_wq, &ctrl->async_event_work);
4060 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4062 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4064 nvme_mpath_stop(ctrl);
4065 nvme_stop_keep_alive(ctrl);
4066 flush_work(&ctrl->async_event_work);
4067 cancel_work_sync(&ctrl->fw_act_work);
4069 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4071 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4074 nvme_start_keep_alive(ctrl);
4076 nvme_enable_aen(ctrl);
4078 if (ctrl->queue_count > 1) {
4079 nvme_queue_scan(ctrl);
4080 nvme_start_queues(ctrl);
4082 ctrl->created = true;
4084 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4086 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4088 nvme_fault_inject_fini(&ctrl->fault_inject);
4089 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4090 cdev_device_del(&ctrl->cdev, ctrl->device);
4091 nvme_put_ctrl(ctrl);
4093 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4095 static void nvme_free_ctrl(struct device *dev)
4097 struct nvme_ctrl *ctrl =
4098 container_of(dev, struct nvme_ctrl, ctrl_device);
4099 struct nvme_subsystem *subsys = ctrl->subsys;
4101 if (subsys && ctrl->instance != subsys->instance)
4102 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4104 kfree(ctrl->effects);
4105 nvme_mpath_uninit(ctrl);
4106 __free_page(ctrl->discard_page);
4109 mutex_lock(&nvme_subsystems_lock);
4110 list_del(&ctrl->subsys_entry);
4111 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4112 mutex_unlock(&nvme_subsystems_lock);
4115 ctrl->ops->free_ctrl(ctrl);
4118 nvme_put_subsystem(subsys);
4122 * Initialize a NVMe controller structures. This needs to be called during
4123 * earliest initialization so that we have the initialized structured around
4126 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4127 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4131 ctrl->state = NVME_CTRL_NEW;
4132 spin_lock_init(&ctrl->lock);
4133 mutex_init(&ctrl->scan_lock);
4134 INIT_LIST_HEAD(&ctrl->namespaces);
4135 init_rwsem(&ctrl->namespaces_rwsem);
4138 ctrl->quirks = quirks;
4139 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4140 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4141 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4142 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4143 init_waitqueue_head(&ctrl->state_wq);
4145 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4146 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4147 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4149 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4151 ctrl->discard_page = alloc_page(GFP_KERNEL);
4152 if (!ctrl->discard_page) {
4157 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4160 ctrl->instance = ret;
4162 device_initialize(&ctrl->ctrl_device);
4163 ctrl->device = &ctrl->ctrl_device;
4164 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4165 ctrl->device->class = nvme_class;
4166 ctrl->device->parent = ctrl->dev;
4167 ctrl->device->groups = nvme_dev_attr_groups;
4168 ctrl->device->release = nvme_free_ctrl;
4169 dev_set_drvdata(ctrl->device, ctrl);
4170 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4172 goto out_release_instance;
4174 nvme_get_ctrl(ctrl);
4175 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4176 ctrl->cdev.owner = ops->module;
4177 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4182 * Initialize latency tolerance controls. The sysfs files won't
4183 * be visible to userspace unless the device actually supports APST.
4185 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4186 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4187 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4189 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4193 nvme_put_ctrl(ctrl);
4194 kfree_const(ctrl->device->kobj.name);
4195 out_release_instance:
4196 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4198 if (ctrl->discard_page)
4199 __free_page(ctrl->discard_page);
4202 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4205 * nvme_kill_queues(): Ends all namespace queues
4206 * @ctrl: the dead controller that needs to end
4208 * Call this function when the driver determines it is unable to get the
4209 * controller in a state capable of servicing IO.
4211 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4215 down_read(&ctrl->namespaces_rwsem);
4217 /* Forcibly unquiesce queues to avoid blocking dispatch */
4218 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4219 blk_mq_unquiesce_queue(ctrl->admin_q);
4221 list_for_each_entry(ns, &ctrl->namespaces, list)
4222 nvme_set_queue_dying(ns);
4224 up_read(&ctrl->namespaces_rwsem);
4226 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4228 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4232 down_read(&ctrl->namespaces_rwsem);
4233 list_for_each_entry(ns, &ctrl->namespaces, list)
4234 blk_mq_unfreeze_queue(ns->queue);
4235 up_read(&ctrl->namespaces_rwsem);
4237 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4239 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4243 down_read(&ctrl->namespaces_rwsem);
4244 list_for_each_entry(ns, &ctrl->namespaces, list) {
4245 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4249 up_read(&ctrl->namespaces_rwsem);
4251 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4253 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4257 down_read(&ctrl->namespaces_rwsem);
4258 list_for_each_entry(ns, &ctrl->namespaces, list)
4259 blk_mq_freeze_queue_wait(ns->queue);
4260 up_read(&ctrl->namespaces_rwsem);
4262 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4264 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4268 down_read(&ctrl->namespaces_rwsem);
4269 list_for_each_entry(ns, &ctrl->namespaces, list)
4270 blk_freeze_queue_start(ns->queue);
4271 up_read(&ctrl->namespaces_rwsem);
4273 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4275 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4279 down_read(&ctrl->namespaces_rwsem);
4280 list_for_each_entry(ns, &ctrl->namespaces, list)
4281 blk_mq_quiesce_queue(ns->queue);
4282 up_read(&ctrl->namespaces_rwsem);
4284 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4286 void nvme_start_queues(struct nvme_ctrl *ctrl)
4290 down_read(&ctrl->namespaces_rwsem);
4291 list_for_each_entry(ns, &ctrl->namespaces, list)
4292 blk_mq_unquiesce_queue(ns->queue);
4293 up_read(&ctrl->namespaces_rwsem);
4295 EXPORT_SYMBOL_GPL(nvme_start_queues);
4298 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4302 down_read(&ctrl->namespaces_rwsem);
4303 list_for_each_entry(ns, &ctrl->namespaces, list)
4304 blk_sync_queue(ns->queue);
4305 up_read(&ctrl->namespaces_rwsem);
4308 blk_sync_queue(ctrl->admin_q);
4310 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4313 * Check we didn't inadvertently grow the command structure sizes:
4315 static inline void _nvme_check_size(void)
4317 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4318 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4319 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4320 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4321 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4322 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4323 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4324 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4325 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4326 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4327 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4328 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4329 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4330 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4331 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4332 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4333 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4337 static int __init nvme_core_init(void)
4339 int result = -ENOMEM;
4343 nvme_wq = alloc_workqueue("nvme-wq",
4344 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4348 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4349 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4353 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4354 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4355 if (!nvme_delete_wq)
4356 goto destroy_reset_wq;
4358 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4360 goto destroy_delete_wq;
4362 nvme_class = class_create(THIS_MODULE, "nvme");
4363 if (IS_ERR(nvme_class)) {
4364 result = PTR_ERR(nvme_class);
4365 goto unregister_chrdev;
4367 nvme_class->dev_uevent = nvme_class_uevent;
4369 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4370 if (IS_ERR(nvme_subsys_class)) {
4371 result = PTR_ERR(nvme_subsys_class);
4377 class_destroy(nvme_class);
4379 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4381 destroy_workqueue(nvme_delete_wq);
4383 destroy_workqueue(nvme_reset_wq);
4385 destroy_workqueue(nvme_wq);
4390 static void __exit nvme_core_exit(void)
4392 class_destroy(nvme_subsys_class);
4393 class_destroy(nvme_class);
4394 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4395 destroy_workqueue(nvme_delete_wq);
4396 destroy_workqueue(nvme_reset_wq);
4397 destroy_workqueue(nvme_wq);
4398 ida_destroy(&nvme_instance_ida);
4401 MODULE_LICENSE("GPL");
4402 MODULE_VERSION("1.0");
4403 module_init(nvme_core_init);
4404 module_exit(nvme_core_exit);