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/slab.h>
17 #include <linux/types.h>
19 #include <linux/ptrace.h>
20 #include <linux/nvme_ioctl.h>
21 #include <linux/pm_qos.h>
22 #include <asm/unaligned.h>
27 #define CREATE_TRACE_POINTS
30 #define NVME_MINORS (1U << MINORBITS)
32 unsigned int admin_timeout = 60;
33 module_param(admin_timeout, uint, 0644);
34 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
35 EXPORT_SYMBOL_GPL(admin_timeout);
37 unsigned int nvme_io_timeout = 30;
38 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
39 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
40 EXPORT_SYMBOL_GPL(nvme_io_timeout);
42 static unsigned char shutdown_timeout = 5;
43 module_param(shutdown_timeout, byte, 0644);
44 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
46 static u8 nvme_max_retries = 5;
47 module_param_named(max_retries, nvme_max_retries, byte, 0644);
48 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
50 static unsigned long default_ps_max_latency_us = 100000;
51 module_param(default_ps_max_latency_us, ulong, 0644);
52 MODULE_PARM_DESC(default_ps_max_latency_us,
53 "max power saving latency for new devices; use PM QOS to change per device");
55 static bool force_apst;
56 module_param(force_apst, bool, 0644);
57 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
60 module_param(streams, bool, 0644);
61 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
64 * nvme_wq - hosts nvme related works that are not reset or delete
65 * nvme_reset_wq - hosts nvme reset works
66 * nvme_delete_wq - hosts nvme delete works
68 * nvme_wq will host works such as scan, aen handling, fw activation,
69 * keep-alive, periodic reconnects etc. nvme_reset_wq
70 * runs reset works which also flush works hosted on nvme_wq for
71 * serialization purposes. nvme_delete_wq host controller deletion
72 * works which flush reset works for serialization.
74 struct workqueue_struct *nvme_wq;
75 EXPORT_SYMBOL_GPL(nvme_wq);
77 struct workqueue_struct *nvme_reset_wq;
78 EXPORT_SYMBOL_GPL(nvme_reset_wq);
80 struct workqueue_struct *nvme_delete_wq;
81 EXPORT_SYMBOL_GPL(nvme_delete_wq);
83 static LIST_HEAD(nvme_subsystems);
84 static DEFINE_MUTEX(nvme_subsystems_lock);
86 static DEFINE_IDA(nvme_instance_ida);
87 static dev_t nvme_chr_devt;
88 static struct class *nvme_class;
89 static struct class *nvme_subsys_class;
91 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
92 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
95 static void nvme_update_bdev_size(struct gendisk *disk)
97 struct block_device *bdev = bdget_disk(disk, 0);
100 bd_set_nr_sectors(bdev, get_capacity(disk));
106 * Prepare a queue for teardown.
108 * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
109 * the capacity to 0 after that to avoid blocking dispatchers that may be
110 * holding bd_butex. This will end buffered writers dirtying pages that can't
113 static void nvme_set_queue_dying(struct nvme_ns *ns)
115 if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
118 blk_set_queue_dying(ns->queue);
119 blk_mq_unquiesce_queue(ns->queue);
121 set_capacity(ns->disk, 0);
122 nvme_update_bdev_size(ns->disk);
125 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
128 * Only new queue scan work when admin and IO queues are both alive
130 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
131 queue_work(nvme_wq, &ctrl->scan_work);
135 * Use this function to proceed with scheduling reset_work for a controller
136 * that had previously been set to the resetting state. This is intended for
137 * code paths that can't be interrupted by other reset attempts. A hot removal
138 * may prevent this from succeeding.
140 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
142 if (ctrl->state != NVME_CTRL_RESETTING)
144 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
148 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
150 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
152 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
154 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
158 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
160 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
164 ret = nvme_reset_ctrl(ctrl);
166 flush_work(&ctrl->reset_work);
167 if (ctrl->state != NVME_CTRL_LIVE)
173 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
175 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
177 dev_info(ctrl->device,
178 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
180 flush_work(&ctrl->reset_work);
181 nvme_stop_ctrl(ctrl);
182 nvme_remove_namespaces(ctrl);
183 ctrl->ops->delete_ctrl(ctrl);
184 nvme_uninit_ctrl(ctrl);
187 static void nvme_delete_ctrl_work(struct work_struct *work)
189 struct nvme_ctrl *ctrl =
190 container_of(work, struct nvme_ctrl, delete_work);
192 nvme_do_delete_ctrl(ctrl);
195 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
197 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
199 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
203 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
205 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
208 * Keep a reference until nvme_do_delete_ctrl() complete,
209 * since ->delete_ctrl can free the controller.
212 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
213 nvme_do_delete_ctrl(ctrl);
217 static blk_status_t nvme_error_status(u16 status)
219 switch (status & 0x7ff) {
220 case NVME_SC_SUCCESS:
222 case NVME_SC_CAP_EXCEEDED:
223 return BLK_STS_NOSPC;
224 case NVME_SC_LBA_RANGE:
225 case NVME_SC_CMD_INTERRUPTED:
226 case NVME_SC_NS_NOT_READY:
227 return BLK_STS_TARGET;
228 case NVME_SC_BAD_ATTRIBUTES:
229 case NVME_SC_ONCS_NOT_SUPPORTED:
230 case NVME_SC_INVALID_OPCODE:
231 case NVME_SC_INVALID_FIELD:
232 case NVME_SC_INVALID_NS:
233 return BLK_STS_NOTSUPP;
234 case NVME_SC_WRITE_FAULT:
235 case NVME_SC_READ_ERROR:
236 case NVME_SC_UNWRITTEN_BLOCK:
237 case NVME_SC_ACCESS_DENIED:
238 case NVME_SC_READ_ONLY:
239 case NVME_SC_COMPARE_FAILED:
240 return BLK_STS_MEDIUM;
241 case NVME_SC_GUARD_CHECK:
242 case NVME_SC_APPTAG_CHECK:
243 case NVME_SC_REFTAG_CHECK:
244 case NVME_SC_INVALID_PI:
245 return BLK_STS_PROTECTION;
246 case NVME_SC_RESERVATION_CONFLICT:
247 return BLK_STS_NEXUS;
248 case NVME_SC_HOST_PATH_ERROR:
249 return BLK_STS_TRANSPORT;
250 case NVME_SC_ZONE_TOO_MANY_ACTIVE:
251 return BLK_STS_ZONE_ACTIVE_RESOURCE;
252 case NVME_SC_ZONE_TOO_MANY_OPEN:
253 return BLK_STS_ZONE_OPEN_RESOURCE;
255 return BLK_STS_IOERR;
259 static void nvme_retry_req(struct request *req)
261 struct nvme_ns *ns = req->q->queuedata;
262 unsigned long delay = 0;
265 /* The mask and shift result must be <= 3 */
266 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
268 delay = ns->ctrl->crdt[crd - 1] * 100;
270 nvme_req(req)->retries++;
271 blk_mq_requeue_request(req, false);
272 blk_mq_delay_kick_requeue_list(req->q, delay);
275 enum nvme_disposition {
281 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
283 if (likely(nvme_req(req)->status == 0))
286 if (blk_noretry_request(req) ||
287 (nvme_req(req)->status & NVME_SC_DNR) ||
288 nvme_req(req)->retries >= nvme_max_retries)
291 if (req->cmd_flags & REQ_NVME_MPATH) {
292 if (nvme_is_path_error(nvme_req(req)->status) ||
293 blk_queue_dying(req->q))
296 if (blk_queue_dying(req->q))
303 static inline void nvme_end_req(struct request *req)
305 blk_status_t status = nvme_error_status(nvme_req(req)->status);
307 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
308 req_op(req) == REQ_OP_ZONE_APPEND)
309 req->__sector = nvme_lba_to_sect(req->q->queuedata,
310 le64_to_cpu(nvme_req(req)->result.u64));
312 nvme_trace_bio_complete(req, status);
313 blk_mq_end_request(req, status);
316 void nvme_complete_rq(struct request *req)
318 trace_nvme_complete_rq(req);
319 nvme_cleanup_cmd(req);
321 if (nvme_req(req)->ctrl->kas)
322 nvme_req(req)->ctrl->comp_seen = true;
324 switch (nvme_decide_disposition(req)) {
332 nvme_failover_req(req);
336 EXPORT_SYMBOL_GPL(nvme_complete_rq);
338 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
340 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
341 "Cancelling I/O %d", req->tag);
343 /* don't abort one completed request */
344 if (blk_mq_request_completed(req))
347 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
348 nvme_req(req)->flags |= NVME_REQ_CANCELLED;
349 blk_mq_complete_request(req);
352 EXPORT_SYMBOL_GPL(nvme_cancel_request);
354 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
357 blk_mq_tagset_busy_iter(ctrl->tagset,
358 nvme_cancel_request, ctrl);
359 blk_mq_tagset_wait_completed_request(ctrl->tagset);
362 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
364 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
366 if (ctrl->admin_tagset) {
367 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
368 nvme_cancel_request, ctrl);
369 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
372 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
374 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
375 enum nvme_ctrl_state new_state)
377 enum nvme_ctrl_state old_state;
379 bool changed = false;
381 spin_lock_irqsave(&ctrl->lock, flags);
383 old_state = ctrl->state;
388 case NVME_CTRL_RESETTING:
389 case NVME_CTRL_CONNECTING:
396 case NVME_CTRL_RESETTING:
406 case NVME_CTRL_CONNECTING:
409 case NVME_CTRL_RESETTING:
416 case NVME_CTRL_DELETING:
419 case NVME_CTRL_RESETTING:
420 case NVME_CTRL_CONNECTING:
427 case NVME_CTRL_DELETING_NOIO:
429 case NVME_CTRL_DELETING:
439 case NVME_CTRL_DELETING:
451 ctrl->state = new_state;
452 wake_up_all(&ctrl->state_wq);
455 spin_unlock_irqrestore(&ctrl->lock, flags);
456 if (changed && ctrl->state == NVME_CTRL_LIVE)
457 nvme_kick_requeue_lists(ctrl);
460 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
463 * Returns true for sink states that can't ever transition back to live.
465 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
467 switch (ctrl->state) {
470 case NVME_CTRL_RESETTING:
471 case NVME_CTRL_CONNECTING:
473 case NVME_CTRL_DELETING:
474 case NVME_CTRL_DELETING_NOIO:
478 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
484 * Waits for the controller state to be resetting, or returns false if it is
485 * not possible to ever transition to that state.
487 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
489 wait_event(ctrl->state_wq,
490 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
491 nvme_state_terminal(ctrl));
492 return ctrl->state == NVME_CTRL_RESETTING;
494 EXPORT_SYMBOL_GPL(nvme_wait_reset);
496 static void nvme_free_ns_head(struct kref *ref)
498 struct nvme_ns_head *head =
499 container_of(ref, struct nvme_ns_head, ref);
501 nvme_mpath_remove_disk(head);
502 ida_simple_remove(&head->subsys->ns_ida, head->instance);
503 cleanup_srcu_struct(&head->srcu);
504 nvme_put_subsystem(head->subsys);
508 static void nvme_put_ns_head(struct nvme_ns_head *head)
510 kref_put(&head->ref, nvme_free_ns_head);
513 static void nvme_free_ns(struct kref *kref)
515 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
518 nvme_nvm_unregister(ns);
521 nvme_put_ns_head(ns->head);
522 nvme_put_ctrl(ns->ctrl);
526 void nvme_put_ns(struct nvme_ns *ns)
528 kref_put(&ns->kref, nvme_free_ns);
530 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
532 static inline void nvme_clear_nvme_request(struct request *req)
534 nvme_req(req)->retries = 0;
535 nvme_req(req)->flags = 0;
536 req->rq_flags |= RQF_DONTPREP;
539 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
541 return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
544 static inline void nvme_init_request(struct request *req,
545 struct nvme_command *cmd)
547 if (req->q->queuedata)
548 req->timeout = NVME_IO_TIMEOUT;
549 else /* no queuedata implies admin queue */
550 req->timeout = ADMIN_TIMEOUT;
552 req->cmd_flags |= REQ_FAILFAST_DRIVER;
553 nvme_clear_nvme_request(req);
554 nvme_req(req)->cmd = cmd;
557 struct request *nvme_alloc_request(struct request_queue *q,
558 struct nvme_command *cmd, blk_mq_req_flags_t flags)
562 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
564 nvme_init_request(req, cmd);
567 EXPORT_SYMBOL_GPL(nvme_alloc_request);
569 struct request *nvme_alloc_request_qid(struct request_queue *q,
570 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
574 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
577 nvme_init_request(req, cmd);
580 EXPORT_SYMBOL_GPL(nvme_alloc_request_qid);
582 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
584 struct nvme_command c;
586 memset(&c, 0, sizeof(c));
588 c.directive.opcode = nvme_admin_directive_send;
589 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
590 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
591 c.directive.dtype = NVME_DIR_IDENTIFY;
592 c.directive.tdtype = NVME_DIR_STREAMS;
593 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
595 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
598 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
600 return nvme_toggle_streams(ctrl, false);
603 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
605 return nvme_toggle_streams(ctrl, true);
608 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
609 struct streams_directive_params *s, u32 nsid)
611 struct nvme_command c;
613 memset(&c, 0, sizeof(c));
614 memset(s, 0, sizeof(*s));
616 c.directive.opcode = nvme_admin_directive_recv;
617 c.directive.nsid = cpu_to_le32(nsid);
618 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
619 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
620 c.directive.dtype = NVME_DIR_STREAMS;
622 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
625 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
627 struct streams_directive_params s;
630 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
635 ret = nvme_enable_streams(ctrl);
639 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
641 goto out_disable_stream;
643 ctrl->nssa = le16_to_cpu(s.nssa);
644 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
645 dev_info(ctrl->device, "too few streams (%u) available\n",
647 goto out_disable_stream;
650 ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
651 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
655 nvme_disable_streams(ctrl);
660 * Check if 'req' has a write hint associated with it. If it does, assign
661 * a valid namespace stream to the write.
663 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
664 struct request *req, u16 *control,
667 enum rw_hint streamid = req->write_hint;
669 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
673 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
676 *control |= NVME_RW_DTYPE_STREAMS;
677 *dsmgmt |= streamid << 16;
680 if (streamid < ARRAY_SIZE(req->q->write_hints))
681 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
684 static inline void nvme_setup_passthrough(struct request *req,
685 struct nvme_command *cmd)
687 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
688 /* passthru commands should let the driver set the SGL flags */
689 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
692 static inline void nvme_setup_flush(struct nvme_ns *ns,
693 struct nvme_command *cmnd)
695 cmnd->common.opcode = nvme_cmd_flush;
696 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
699 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
700 struct nvme_command *cmnd)
702 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
703 struct nvme_dsm_range *range;
707 * Some devices do not consider the DSM 'Number of Ranges' field when
708 * determining how much data to DMA. Always allocate memory for maximum
709 * number of segments to prevent device reading beyond end of buffer.
711 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
713 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
716 * If we fail allocation our range, fallback to the controller
717 * discard page. If that's also busy, it's safe to return
718 * busy, as we know we can make progress once that's freed.
720 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
721 return BLK_STS_RESOURCE;
723 range = page_address(ns->ctrl->discard_page);
726 __rq_for_each_bio(bio, req) {
727 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
728 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
731 range[n].cattr = cpu_to_le32(0);
732 range[n].nlb = cpu_to_le32(nlb);
733 range[n].slba = cpu_to_le64(slba);
738 if (WARN_ON_ONCE(n != segments)) {
739 if (virt_to_page(range) == ns->ctrl->discard_page)
740 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
743 return BLK_STS_IOERR;
746 cmnd->dsm.opcode = nvme_cmd_dsm;
747 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
748 cmnd->dsm.nr = cpu_to_le32(segments - 1);
749 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
751 req->special_vec.bv_page = virt_to_page(range);
752 req->special_vec.bv_offset = offset_in_page(range);
753 req->special_vec.bv_len = alloc_size;
754 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
759 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
760 struct request *req, struct nvme_command *cmnd)
762 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
763 return nvme_setup_discard(ns, req, cmnd);
765 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
766 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
767 cmnd->write_zeroes.slba =
768 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
769 cmnd->write_zeroes.length =
770 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
771 if (nvme_ns_has_pi(ns))
772 cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
774 cmnd->write_zeroes.control = 0;
778 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
779 struct request *req, struct nvme_command *cmnd,
782 struct nvme_ctrl *ctrl = ns->ctrl;
786 if (req->cmd_flags & REQ_FUA)
787 control |= NVME_RW_FUA;
788 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
789 control |= NVME_RW_LR;
791 if (req->cmd_flags & REQ_RAHEAD)
792 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
794 cmnd->rw.opcode = op;
795 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
796 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
797 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
799 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
800 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
804 * If formated with metadata, the block layer always provides a
805 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
806 * we enable the PRACT bit for protection information or set the
807 * namespace capacity to zero to prevent any I/O.
809 if (!blk_integrity_rq(req)) {
810 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
811 return BLK_STS_NOTSUPP;
812 control |= NVME_RW_PRINFO_PRACT;
815 switch (ns->pi_type) {
816 case NVME_NS_DPS_PI_TYPE3:
817 control |= NVME_RW_PRINFO_PRCHK_GUARD;
819 case NVME_NS_DPS_PI_TYPE1:
820 case NVME_NS_DPS_PI_TYPE2:
821 control |= NVME_RW_PRINFO_PRCHK_GUARD |
822 NVME_RW_PRINFO_PRCHK_REF;
823 if (op == nvme_cmd_zone_append)
824 control |= NVME_RW_APPEND_PIREMAP;
825 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
830 cmnd->rw.control = cpu_to_le16(control);
831 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
835 void nvme_cleanup_cmd(struct request *req)
837 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
838 struct nvme_ns *ns = req->rq_disk->private_data;
839 struct page *page = req->special_vec.bv_page;
841 if (page == ns->ctrl->discard_page)
842 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
844 kfree(page_address(page) + req->special_vec.bv_offset);
847 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
849 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
850 struct nvme_command *cmd)
852 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
853 blk_status_t ret = BLK_STS_OK;
855 if (!(req->rq_flags & RQF_DONTPREP))
856 nvme_clear_nvme_request(req);
858 memset(cmd, 0, sizeof(*cmd));
859 switch (req_op(req)) {
862 nvme_setup_passthrough(req, cmd);
865 nvme_setup_flush(ns, cmd);
867 case REQ_OP_ZONE_RESET_ALL:
868 case REQ_OP_ZONE_RESET:
869 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
871 case REQ_OP_ZONE_OPEN:
872 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
874 case REQ_OP_ZONE_CLOSE:
875 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
877 case REQ_OP_ZONE_FINISH:
878 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
880 case REQ_OP_WRITE_ZEROES:
881 ret = nvme_setup_write_zeroes(ns, req, cmd);
884 ret = nvme_setup_discard(ns, req, cmd);
887 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
890 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
892 case REQ_OP_ZONE_APPEND:
893 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
897 return BLK_STS_IOERR;
900 if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
901 nvme_req(req)->genctr++;
902 cmd->common.command_id = nvme_cid(req);
903 trace_nvme_setup_cmd(req, cmd);
906 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
908 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
910 struct completion *waiting = rq->end_io_data;
912 rq->end_io_data = NULL;
916 static void nvme_execute_rq_polled(struct request_queue *q,
917 struct gendisk *bd_disk, struct request *rq, int at_head)
919 DECLARE_COMPLETION_ONSTACK(wait);
921 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
923 rq->cmd_flags |= REQ_HIPRI;
924 rq->end_io_data = &wait;
925 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
927 while (!completion_done(&wait)) {
928 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
934 * Returns 0 on success. If the result is negative, it's a Linux error code;
935 * if the result is positive, it's an NVM Express status code
937 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
938 union nvme_result *result, void *buffer, unsigned bufflen,
939 unsigned timeout, int qid, int at_head,
940 blk_mq_req_flags_t flags, bool poll)
945 if (qid == NVME_QID_ANY)
946 req = nvme_alloc_request(q, cmd, flags);
948 req = nvme_alloc_request_qid(q, cmd, flags, qid);
953 req->timeout = timeout;
955 if (buffer && bufflen) {
956 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
962 nvme_execute_rq_polled(req->q, NULL, req, at_head);
964 blk_execute_rq(req->q, NULL, req, at_head);
966 *result = nvme_req(req)->result;
967 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
970 ret = nvme_req(req)->status;
972 blk_mq_free_request(req);
975 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
977 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
978 void *buffer, unsigned bufflen)
980 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
981 NVME_QID_ANY, 0, 0, false);
983 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
985 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
986 unsigned len, u32 seed, bool write)
988 struct bio_integrity_payload *bip;
992 buf = kmalloc(len, GFP_KERNEL);
997 if (write && copy_from_user(buf, ubuf, len))
1000 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
1006 bip->bip_iter.bi_size = len;
1007 bip->bip_iter.bi_sector = seed;
1008 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
1009 offset_in_page(buf));
1016 return ERR_PTR(ret);
1019 static u32 nvme_known_admin_effects(u8 opcode)
1022 case nvme_admin_format_nvm:
1023 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1024 NVME_CMD_EFFECTS_CSE_MASK;
1025 case nvme_admin_sanitize_nvm:
1026 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1033 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1038 if (ns->head->effects)
1039 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1040 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1041 dev_warn(ctrl->device,
1042 "IO command:%02x has unhandled effects:%08x\n",
1048 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1049 effects |= nvme_known_admin_effects(opcode);
1053 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1055 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1058 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1061 * For simplicity, IO to all namespaces is quiesced even if the command
1062 * effects say only one namespace is affected.
1064 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1065 mutex_lock(&ctrl->scan_lock);
1066 mutex_lock(&ctrl->subsys->lock);
1067 nvme_mpath_start_freeze(ctrl->subsys);
1068 nvme_mpath_wait_freeze(ctrl->subsys);
1069 nvme_start_freeze(ctrl);
1070 nvme_wait_freeze(ctrl);
1075 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1077 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1078 nvme_unfreeze(ctrl);
1079 nvme_mpath_unfreeze(ctrl->subsys);
1080 mutex_unlock(&ctrl->subsys->lock);
1081 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1082 mutex_unlock(&ctrl->scan_lock);
1084 if (effects & NVME_CMD_EFFECTS_CCC)
1085 nvme_init_identify(ctrl);
1086 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1087 nvme_queue_scan(ctrl);
1088 flush_work(&ctrl->scan_work);
1092 void nvme_execute_passthru_rq(struct request *rq)
1094 struct nvme_command *cmd = nvme_req(rq)->cmd;
1095 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1096 struct nvme_ns *ns = rq->q->queuedata;
1097 struct gendisk *disk = ns ? ns->disk : NULL;
1100 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1101 blk_execute_rq(rq->q, disk, rq, 0);
1102 nvme_passthru_end(ctrl, effects);
1104 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1106 static int nvme_submit_user_cmd(struct request_queue *q,
1107 struct nvme_command *cmd, void __user *ubuffer,
1108 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1109 u32 meta_seed, u64 *result, unsigned timeout)
1111 bool write = nvme_is_write(cmd);
1112 struct nvme_ns *ns = q->queuedata;
1113 struct gendisk *disk = ns ? ns->disk : NULL;
1114 struct request *req;
1115 struct bio *bio = NULL;
1119 req = nvme_alloc_request(q, cmd, 0);
1121 return PTR_ERR(req);
1124 req->timeout = timeout;
1125 nvme_req(req)->flags |= NVME_REQ_USERCMD;
1127 if (ubuffer && bufflen) {
1128 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1133 bio->bi_disk = disk;
1134 if (disk && meta_buffer && meta_len) {
1135 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1138 ret = PTR_ERR(meta);
1141 req->cmd_flags |= REQ_INTEGRITY;
1145 nvme_execute_passthru_rq(req);
1146 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1149 ret = nvme_req(req)->status;
1151 *result = le64_to_cpu(nvme_req(req)->result.u64);
1152 if (meta && !ret && !write) {
1153 if (copy_to_user(meta_buffer, meta, meta_len))
1159 blk_rq_unmap_user(bio);
1161 blk_mq_free_request(req);
1165 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1167 struct nvme_ctrl *ctrl = rq->end_io_data;
1168 unsigned long flags;
1169 bool startka = false;
1171 blk_mq_free_request(rq);
1174 dev_err(ctrl->device,
1175 "failed nvme_keep_alive_end_io error=%d\n",
1180 ctrl->comp_seen = false;
1181 spin_lock_irqsave(&ctrl->lock, flags);
1182 if (ctrl->state == NVME_CTRL_LIVE ||
1183 ctrl->state == NVME_CTRL_CONNECTING)
1185 spin_unlock_irqrestore(&ctrl->lock, flags);
1187 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1190 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1194 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1195 BLK_MQ_REQ_RESERVED);
1199 rq->timeout = ctrl->kato * HZ;
1200 rq->end_io_data = ctrl;
1202 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
1207 static void nvme_keep_alive_work(struct work_struct *work)
1209 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1210 struct nvme_ctrl, ka_work);
1211 bool comp_seen = ctrl->comp_seen;
1213 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1214 dev_dbg(ctrl->device,
1215 "reschedule traffic based keep-alive timer\n");
1216 ctrl->comp_seen = false;
1217 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1221 if (nvme_keep_alive(ctrl)) {
1222 /* allocation failure, reset the controller */
1223 dev_err(ctrl->device, "keep-alive failed\n");
1224 nvme_reset_ctrl(ctrl);
1229 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1231 if (unlikely(ctrl->kato == 0))
1234 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1237 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1239 if (unlikely(ctrl->kato == 0))
1242 cancel_delayed_work_sync(&ctrl->ka_work);
1244 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1247 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1248 * flag, thus sending any new CNS opcodes has a big chance of not working.
1249 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1250 * (but not for any later version).
1252 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1254 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1255 return ctrl->vs < NVME_VS(1, 2, 0);
1256 return ctrl->vs < NVME_VS(1, 1, 0);
1259 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1261 struct nvme_command c = { };
1264 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1265 c.identify.opcode = nvme_admin_identify;
1266 c.identify.cns = NVME_ID_CNS_CTRL;
1268 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1272 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1273 sizeof(struct nvme_id_ctrl));
1279 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1281 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1284 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1285 struct nvme_ns_id_desc *cur, bool *csi_seen)
1287 const char *warn_str = "ctrl returned bogus length:";
1290 switch (cur->nidt) {
1291 case NVME_NIDT_EUI64:
1292 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1293 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1294 warn_str, cur->nidl);
1297 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1298 return NVME_NIDT_EUI64_LEN;
1299 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1300 return NVME_NIDT_EUI64_LEN;
1301 case NVME_NIDT_NGUID:
1302 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1303 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1304 warn_str, cur->nidl);
1307 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1308 return NVME_NIDT_NGUID_LEN;
1309 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1310 return NVME_NIDT_NGUID_LEN;
1311 case NVME_NIDT_UUID:
1312 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1313 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1314 warn_str, cur->nidl);
1317 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1318 return NVME_NIDT_UUID_LEN;
1319 uuid_copy(&ids->uuid, data + sizeof(*cur));
1320 return NVME_NIDT_UUID_LEN;
1322 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1323 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1324 warn_str, cur->nidl);
1327 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1329 return NVME_NIDT_CSI_LEN;
1331 /* Skip unknown types */
1336 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1337 struct nvme_ns_ids *ids)
1339 struct nvme_command c = { };
1340 bool csi_seen = false;
1341 int status, pos, len;
1344 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1346 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1349 c.identify.opcode = nvme_admin_identify;
1350 c.identify.nsid = cpu_to_le32(nsid);
1351 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1353 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1357 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1358 NVME_IDENTIFY_DATA_SIZE);
1360 dev_warn(ctrl->device,
1361 "Identify Descriptors failed (%d)\n", status);
1365 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1366 struct nvme_ns_id_desc *cur = data + pos;
1371 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1375 len += sizeof(*cur);
1378 if (nvme_multi_css(ctrl) && !csi_seen) {
1379 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1389 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1390 struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1392 struct nvme_command c = { };
1395 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1396 c.identify.opcode = nvme_admin_identify;
1397 c.identify.nsid = cpu_to_le32(nsid);
1398 c.identify.cns = NVME_ID_CNS_NS;
1400 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1404 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1406 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1410 error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1411 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1415 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1416 dev_info(ctrl->device,
1417 "Ignoring bogus Namespace Identifiers\n");
1419 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1420 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1421 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1422 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1423 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1424 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1434 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1435 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1437 union nvme_result res = { 0 };
1438 struct nvme_command c;
1441 memset(&c, 0, sizeof(c));
1442 c.features.opcode = op;
1443 c.features.fid = cpu_to_le32(fid);
1444 c.features.dword11 = cpu_to_le32(dword11);
1446 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1447 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1448 if (ret >= 0 && result)
1449 *result = le32_to_cpu(res.u32);
1453 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1454 unsigned int dword11, void *buffer, size_t buflen,
1457 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1460 EXPORT_SYMBOL_GPL(nvme_set_features);
1462 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1463 unsigned int dword11, void *buffer, size_t buflen,
1466 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1469 EXPORT_SYMBOL_GPL(nvme_get_features);
1471 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1473 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1475 int status, nr_io_queues;
1477 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1483 * Degraded controllers might return an error when setting the queue
1484 * count. We still want to be able to bring them online and offer
1485 * access to the admin queue, as that might be only way to fix them up.
1488 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1491 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1492 *count = min(*count, nr_io_queues);
1497 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1499 #define NVME_AEN_SUPPORTED \
1500 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1501 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1503 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1505 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1508 if (!supported_aens)
1511 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1514 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1517 queue_work(nvme_wq, &ctrl->async_event_work);
1521 * Convert integer values from ioctl structures to user pointers, silently
1522 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1525 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1527 if (in_compat_syscall())
1528 ptrval = (compat_uptr_t)ptrval;
1529 return (void __user *)ptrval;
1532 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1534 struct nvme_user_io io;
1535 struct nvme_command c;
1536 unsigned length, meta_len;
1537 void __user *metadata;
1539 if (copy_from_user(&io, uio, sizeof(io)))
1544 switch (io.opcode) {
1545 case nvme_cmd_write:
1547 case nvme_cmd_compare:
1553 length = (io.nblocks + 1) << ns->lba_shift;
1555 if ((io.control & NVME_RW_PRINFO_PRACT) &&
1556 ns->ms == sizeof(struct t10_pi_tuple)) {
1558 * Protection information is stripped/inserted by the
1561 if (nvme_to_user_ptr(io.metadata))
1566 meta_len = (io.nblocks + 1) * ns->ms;
1567 metadata = nvme_to_user_ptr(io.metadata);
1570 if (ns->features & NVME_NS_EXT_LBAS) {
1573 } else if (meta_len) {
1574 if ((io.metadata & 3) || !io.metadata)
1578 memset(&c, 0, sizeof(c));
1579 c.rw.opcode = io.opcode;
1580 c.rw.flags = io.flags;
1581 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1582 c.rw.slba = cpu_to_le64(io.slba);
1583 c.rw.length = cpu_to_le16(io.nblocks);
1584 c.rw.control = cpu_to_le16(io.control);
1585 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1586 c.rw.reftag = cpu_to_le32(io.reftag);
1587 c.rw.apptag = cpu_to_le16(io.apptag);
1588 c.rw.appmask = cpu_to_le16(io.appmask);
1590 return nvme_submit_user_cmd(ns->queue, &c,
1591 nvme_to_user_ptr(io.addr), length,
1592 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1595 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1596 struct nvme_passthru_cmd __user *ucmd)
1598 struct nvme_passthru_cmd cmd;
1599 struct nvme_command c;
1600 unsigned timeout = 0;
1604 if (!capable(CAP_SYS_ADMIN))
1606 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1611 memset(&c, 0, sizeof(c));
1612 c.common.opcode = cmd.opcode;
1613 c.common.flags = cmd.flags;
1614 c.common.nsid = cpu_to_le32(cmd.nsid);
1615 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1616 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1617 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1618 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1619 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1620 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1621 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1622 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1625 timeout = msecs_to_jiffies(cmd.timeout_ms);
1627 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1628 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1629 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1630 0, &result, timeout);
1633 if (put_user(result, &ucmd->result))
1640 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1641 struct nvme_passthru_cmd64 __user *ucmd)
1643 struct nvme_passthru_cmd64 cmd;
1644 struct nvme_command c;
1645 unsigned timeout = 0;
1648 if (!capable(CAP_SYS_ADMIN))
1650 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1655 memset(&c, 0, sizeof(c));
1656 c.common.opcode = cmd.opcode;
1657 c.common.flags = cmd.flags;
1658 c.common.nsid = cpu_to_le32(cmd.nsid);
1659 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1660 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1661 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1662 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1663 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1664 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1665 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1666 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1669 timeout = msecs_to_jiffies(cmd.timeout_ms);
1671 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1672 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1673 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1674 0, &cmd.result, timeout);
1677 if (put_user(cmd.result, &ucmd->result))
1685 * Issue ioctl requests on the first available path. Note that unlike normal
1686 * block layer requests we will not retry failed request on another controller.
1688 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1689 struct nvme_ns_head **head, int *srcu_idx)
1691 #ifdef CONFIG_NVME_MULTIPATH
1692 if (disk->fops == &nvme_ns_head_ops) {
1695 *head = disk->private_data;
1696 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1697 ns = nvme_find_path(*head);
1699 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1705 return disk->private_data;
1708 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1711 srcu_read_unlock(&head->srcu, idx);
1714 static bool is_ctrl_ioctl(unsigned int cmd)
1716 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1718 if (is_sed_ioctl(cmd))
1723 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1725 struct nvme_ns_head *head,
1728 struct nvme_ctrl *ctrl = ns->ctrl;
1731 nvme_get_ctrl(ns->ctrl);
1732 nvme_put_ns_from_disk(head, srcu_idx);
1735 case NVME_IOCTL_ADMIN_CMD:
1736 ret = nvme_user_cmd(ctrl, NULL, argp);
1738 case NVME_IOCTL_ADMIN64_CMD:
1739 ret = nvme_user_cmd64(ctrl, NULL, argp);
1742 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1745 nvme_put_ctrl(ctrl);
1749 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1750 unsigned int cmd, unsigned long arg)
1752 struct nvme_ns_head *head = NULL;
1753 void __user *argp = (void __user *)arg;
1757 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1759 return -EWOULDBLOCK;
1762 * Handle ioctls that apply to the controller instead of the namespace
1763 * seperately and drop the ns SRCU reference early. This avoids a
1764 * deadlock when deleting namespaces using the passthrough interface.
1766 if (is_ctrl_ioctl(cmd))
1767 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1771 force_successful_syscall_return();
1772 ret = ns->head->ns_id;
1774 case NVME_IOCTL_IO_CMD:
1775 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1777 case NVME_IOCTL_SUBMIT_IO:
1778 ret = nvme_submit_io(ns, argp);
1780 case NVME_IOCTL_IO64_CMD:
1781 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1785 ret = nvme_nvm_ioctl(ns, cmd, arg);
1790 nvme_put_ns_from_disk(head, srcu_idx);
1794 #ifdef CONFIG_COMPAT
1795 struct nvme_user_io32 {
1808 } __attribute__((__packed__));
1810 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1812 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1813 unsigned int cmd, unsigned long arg)
1816 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1817 * between 32 bit programs and 64 bit kernel.
1818 * The cause is that the results of sizeof(struct nvme_user_io),
1819 * which is used to define NVME_IOCTL_SUBMIT_IO,
1820 * are not same between 32 bit compiler and 64 bit compiler.
1821 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1822 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1823 * Other IOCTL numbers are same between 32 bit and 64 bit.
1824 * So there is nothing to do regarding to other IOCTL numbers.
1826 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1827 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1829 return nvme_ioctl(bdev, mode, cmd, arg);
1832 #define nvme_compat_ioctl NULL
1833 #endif /* CONFIG_COMPAT */
1835 static int nvme_open(struct block_device *bdev, fmode_t mode)
1837 struct nvme_ns *ns = bdev->bd_disk->private_data;
1839 #ifdef CONFIG_NVME_MULTIPATH
1840 /* should never be called due to GENHD_FL_HIDDEN */
1841 if (WARN_ON_ONCE(ns->head->disk))
1844 if (!kref_get_unless_zero(&ns->kref))
1846 if (!try_module_get(ns->ctrl->ops->module))
1857 static void nvme_release(struct gendisk *disk, fmode_t mode)
1859 struct nvme_ns *ns = disk->private_data;
1861 module_put(ns->ctrl->ops->module);
1865 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1867 /* some standard values */
1868 geo->heads = 1 << 6;
1869 geo->sectors = 1 << 5;
1870 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1874 #ifdef CONFIG_BLK_DEV_INTEGRITY
1875 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1876 u32 max_integrity_segments)
1878 struct blk_integrity integrity;
1880 memset(&integrity, 0, sizeof(integrity));
1882 case NVME_NS_DPS_PI_TYPE3:
1883 integrity.profile = &t10_pi_type3_crc;
1884 integrity.tag_size = sizeof(u16) + sizeof(u32);
1885 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1887 case NVME_NS_DPS_PI_TYPE1:
1888 case NVME_NS_DPS_PI_TYPE2:
1889 integrity.profile = &t10_pi_type1_crc;
1890 integrity.tag_size = sizeof(u16);
1891 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1894 integrity.profile = NULL;
1897 integrity.tuple_size = ms;
1898 blk_integrity_register(disk, &integrity);
1899 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1902 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1903 u32 max_integrity_segments)
1906 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1908 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1910 struct nvme_ctrl *ctrl = ns->ctrl;
1911 struct request_queue *queue = disk->queue;
1912 u32 size = queue_logical_block_size(queue);
1914 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1915 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1919 if (ctrl->nr_streams && ns->sws && ns->sgs)
1920 size *= ns->sws * ns->sgs;
1922 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1923 NVME_DSM_MAX_RANGES);
1925 queue->limits.discard_alignment = 0;
1926 queue->limits.discard_granularity = size;
1928 /* If discard is already enabled, don't reset queue limits */
1929 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1932 blk_queue_max_discard_sectors(queue, UINT_MAX);
1933 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1935 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1936 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1940 * Even though NVMe spec explicitly states that MDTS is not applicable to the
1941 * write-zeroes, we are cautious and limit the size to the controllers
1942 * max_hw_sectors value, which is based on the MDTS field and possibly other
1945 static void nvme_config_write_zeroes(struct request_queue *q,
1946 struct nvme_ctrl *ctrl)
1948 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
1949 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1950 blk_queue_max_write_zeroes_sectors(q, ctrl->max_hw_sectors);
1953 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1955 return !uuid_is_null(&ids->uuid) ||
1956 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1957 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1960 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1962 return uuid_equal(&a->uuid, &b->uuid) &&
1963 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1964 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1968 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1969 u32 *phys_bs, u32 *io_opt)
1971 struct streams_directive_params s;
1974 if (!ctrl->nr_streams)
1977 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1981 ns->sws = le32_to_cpu(s.sws);
1982 ns->sgs = le16_to_cpu(s.sgs);
1985 *phys_bs = ns->sws * (1 << ns->lba_shift);
1987 *io_opt = *phys_bs * ns->sgs;
1993 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1995 struct nvme_ctrl *ctrl = ns->ctrl;
1998 * The PI implementation requires the metadata size to be equal to the
1999 * t10 pi tuple size.
2001 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
2002 if (ns->ms == sizeof(struct t10_pi_tuple))
2003 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
2007 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2008 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
2010 if (ctrl->ops->flags & NVME_F_FABRICS) {
2012 * The NVMe over Fabrics specification only supports metadata as
2013 * part of the extended data LBA. We rely on HCA/HBA support to
2014 * remap the separate metadata buffer from the block layer.
2016 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
2018 if (ctrl->max_integrity_segments)
2020 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2023 * For PCIe controllers, we can't easily remap the separate
2024 * metadata buffer from the block layer and thus require a
2025 * separate metadata buffer for block layer metadata/PI support.
2026 * We allow extended LBAs for the passthrough interface, though.
2028 if (id->flbas & NVME_NS_FLBAS_META_EXT)
2029 ns->features |= NVME_NS_EXT_LBAS;
2031 ns->features |= NVME_NS_METADATA_SUPPORTED;
2037 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2038 struct request_queue *q)
2040 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2042 if (ctrl->max_hw_sectors) {
2044 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
2046 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2047 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2048 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2050 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
2051 blk_queue_dma_alignment(q, 3);
2052 blk_queue_write_cache(q, vwc, vwc);
2055 static void nvme_update_disk_info(struct gendisk *disk,
2056 struct nvme_ns *ns, struct nvme_id_ns *id)
2058 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
2059 unsigned short bs = 1 << ns->lba_shift;
2060 u32 atomic_bs, phys_bs, io_opt = 0;
2063 * The block layer can't support LBA sizes larger than the page size
2064 * yet, so catch this early and don't allow block I/O.
2066 if (ns->lba_shift > PAGE_SHIFT) {
2071 blk_integrity_unregister(disk);
2073 atomic_bs = phys_bs = bs;
2074 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2075 if (id->nabo == 0) {
2077 * Bit 1 indicates whether NAWUPF is defined for this namespace
2078 * and whether it should be used instead of AWUPF. If NAWUPF ==
2079 * 0 then AWUPF must be used instead.
2081 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2082 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2084 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2087 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2088 /* NPWG = Namespace Preferred Write Granularity */
2089 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2090 /* NOWS = Namespace Optimal Write Size */
2091 io_opt = bs * (1 + le16_to_cpu(id->nows));
2094 blk_queue_logical_block_size(disk->queue, bs);
2096 * Linux filesystems assume writing a single physical block is
2097 * an atomic operation. Hence limit the physical block size to the
2098 * value of the Atomic Write Unit Power Fail parameter.
2100 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2101 blk_queue_io_min(disk->queue, phys_bs);
2102 blk_queue_io_opt(disk->queue, io_opt);
2105 * Register a metadata profile for PI, or the plain non-integrity NVMe
2106 * metadata masquerading as Type 0 if supported, otherwise reject block
2107 * I/O to namespaces with metadata except when the namespace supports
2108 * PI, as it can strip/insert in that case.
2111 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2112 (ns->features & NVME_NS_METADATA_SUPPORTED))
2113 nvme_init_integrity(disk, ns->ms, ns->pi_type,
2114 ns->ctrl->max_integrity_segments);
2115 else if (!nvme_ns_has_pi(ns))
2119 set_capacity_revalidate_and_notify(disk, capacity, false);
2121 nvme_config_discard(disk, ns);
2122 nvme_config_write_zeroes(disk->queue, ns->ctrl);
2124 if (id->nsattr & NVME_NS_ATTR_RO)
2125 set_disk_ro(disk, true);
2128 static inline bool nvme_first_scan(struct gendisk *disk)
2130 /* nvme_alloc_ns() scans the disk prior to adding it */
2131 return !(disk->flags & GENHD_FL_UP);
2134 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2136 struct nvme_ctrl *ctrl = ns->ctrl;
2139 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2140 is_power_of_2(ctrl->max_hw_sectors))
2141 iob = ctrl->max_hw_sectors;
2143 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2148 if (!is_power_of_2(iob)) {
2149 if (nvme_first_scan(ns->disk))
2150 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2151 ns->disk->disk_name, iob);
2155 if (blk_queue_is_zoned(ns->disk->queue)) {
2156 if (nvme_first_scan(ns->disk))
2157 pr_warn("%s: ignoring zoned namespace IO boundary\n",
2158 ns->disk->disk_name);
2162 blk_queue_chunk_sectors(ns->queue, iob);
2165 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
2167 unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2170 blk_mq_freeze_queue(ns->disk->queue);
2171 ns->lba_shift = id->lbaf[lbaf].ds;
2172 nvme_set_queue_limits(ns->ctrl, ns->queue);
2174 if (ns->head->ids.csi == NVME_CSI_ZNS) {
2175 ret = nvme_update_zone_info(ns, lbaf);
2180 ret = nvme_configure_metadata(ns, id);
2183 nvme_set_chunk_sectors(ns, id);
2184 nvme_update_disk_info(ns->disk, ns, id);
2185 blk_mq_unfreeze_queue(ns->disk->queue);
2187 if (blk_queue_is_zoned(ns->queue)) {
2188 ret = nvme_revalidate_zones(ns);
2189 if (ret && !nvme_first_scan(ns->disk))
2193 #ifdef CONFIG_NVME_MULTIPATH
2194 if (ns->head->disk) {
2195 blk_mq_freeze_queue(ns->head->disk->queue);
2196 nvme_update_disk_info(ns->head->disk, ns, id);
2197 blk_stack_limits(&ns->head->disk->queue->limits,
2198 &ns->queue->limits, 0);
2199 blk_queue_update_readahead(ns->head->disk->queue);
2200 nvme_update_bdev_size(ns->head->disk);
2201 blk_mq_unfreeze_queue(ns->head->disk->queue);
2207 blk_mq_unfreeze_queue(ns->disk->queue);
2211 static char nvme_pr_type(enum pr_type type)
2214 case PR_WRITE_EXCLUSIVE:
2216 case PR_EXCLUSIVE_ACCESS:
2218 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2220 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2222 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2224 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2231 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2232 u64 key, u64 sa_key, u8 op)
2234 struct nvme_ns_head *head = NULL;
2236 struct nvme_command c;
2238 u8 data[16] = { 0, };
2240 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2242 return -EWOULDBLOCK;
2244 put_unaligned_le64(key, &data[0]);
2245 put_unaligned_le64(sa_key, &data[8]);
2247 memset(&c, 0, sizeof(c));
2248 c.common.opcode = op;
2249 c.common.nsid = cpu_to_le32(ns->head->ns_id);
2250 c.common.cdw10 = cpu_to_le32(cdw10);
2252 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2253 nvme_put_ns_from_disk(head, srcu_idx);
2257 static int nvme_pr_register(struct block_device *bdev, u64 old,
2258 u64 new, unsigned flags)
2262 if (flags & ~PR_FL_IGNORE_KEY)
2265 cdw10 = old ? 2 : 0;
2266 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2267 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2268 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2271 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2272 enum pr_type type, unsigned flags)
2276 if (flags & ~PR_FL_IGNORE_KEY)
2279 cdw10 = nvme_pr_type(type) << 8;
2280 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2281 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2284 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2285 enum pr_type type, bool abort)
2287 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2289 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2292 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2294 u32 cdw10 = 1 | (key ? 0 : 1 << 3);
2296 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2299 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2301 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 0 : 1 << 3);
2303 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2306 static const struct pr_ops nvme_pr_ops = {
2307 .pr_register = nvme_pr_register,
2308 .pr_reserve = nvme_pr_reserve,
2309 .pr_release = nvme_pr_release,
2310 .pr_preempt = nvme_pr_preempt,
2311 .pr_clear = nvme_pr_clear,
2314 #ifdef CONFIG_BLK_SED_OPAL
2315 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2318 struct nvme_ctrl *ctrl = data;
2319 struct nvme_command cmd;
2321 memset(&cmd, 0, sizeof(cmd));
2323 cmd.common.opcode = nvme_admin_security_send;
2325 cmd.common.opcode = nvme_admin_security_recv;
2326 cmd.common.nsid = 0;
2327 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2328 cmd.common.cdw11 = cpu_to_le32(len);
2330 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2331 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2333 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2334 #endif /* CONFIG_BLK_SED_OPAL */
2336 static const struct block_device_operations nvme_fops = {
2337 .owner = THIS_MODULE,
2338 .ioctl = nvme_ioctl,
2339 .compat_ioctl = nvme_compat_ioctl,
2341 .release = nvme_release,
2342 .getgeo = nvme_getgeo,
2343 .report_zones = nvme_report_zones,
2344 .pr_ops = &nvme_pr_ops,
2347 #ifdef CONFIG_NVME_MULTIPATH
2348 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2350 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2352 if (!kref_get_unless_zero(&head->ref))
2357 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2359 nvme_put_ns_head(disk->private_data);
2362 const struct block_device_operations nvme_ns_head_ops = {
2363 .owner = THIS_MODULE,
2364 .submit_bio = nvme_ns_head_submit_bio,
2365 .open = nvme_ns_head_open,
2366 .release = nvme_ns_head_release,
2367 .ioctl = nvme_ioctl,
2368 .compat_ioctl = nvme_compat_ioctl,
2369 .getgeo = nvme_getgeo,
2370 .report_zones = nvme_report_zones,
2371 .pr_ops = &nvme_pr_ops,
2373 #endif /* CONFIG_NVME_MULTIPATH */
2375 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2377 unsigned long timeout =
2378 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2379 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2382 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2385 if ((csts & NVME_CSTS_RDY) == bit)
2388 usleep_range(1000, 2000);
2389 if (fatal_signal_pending(current))
2391 if (time_after(jiffies, timeout)) {
2392 dev_err(ctrl->device,
2393 "Device not ready; aborting %s, CSTS=0x%x\n",
2394 enabled ? "initialisation" : "reset", csts);
2403 * If the device has been passed off to us in an enabled state, just clear
2404 * the enabled bit. The spec says we should set the 'shutdown notification
2405 * bits', but doing so may cause the device to complete commands to the
2406 * admin queue ... and we don't know what memory that might be pointing at!
2408 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2412 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2413 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2415 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2419 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2420 msleep(NVME_QUIRK_DELAY_AMOUNT);
2422 return nvme_wait_ready(ctrl, ctrl->cap, false);
2424 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2426 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2428 unsigned dev_page_min;
2431 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2433 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2436 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2438 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2439 dev_err(ctrl->device,
2440 "Minimum device page size %u too large for host (%u)\n",
2441 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2445 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2446 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2448 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2449 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2450 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2451 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2452 ctrl->ctrl_config |= NVME_CC_ENABLE;
2454 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2457 return nvme_wait_ready(ctrl, ctrl->cap, true);
2459 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2461 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2463 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2467 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2468 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2470 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2474 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2475 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2479 if (fatal_signal_pending(current))
2481 if (time_after(jiffies, timeout)) {
2482 dev_err(ctrl->device,
2483 "Device shutdown incomplete; abort shutdown\n");
2490 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2492 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2497 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2500 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2501 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2504 dev_warn_once(ctrl->device,
2505 "could not set timestamp (%d)\n", ret);
2509 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2511 struct nvme_feat_host_behavior *host;
2514 /* Don't bother enabling the feature if retry delay is not reported */
2518 host = kzalloc(sizeof(*host), GFP_KERNEL);
2522 host->acre = NVME_ENABLE_ACRE;
2523 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2524 host, sizeof(*host), NULL);
2529 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2532 * APST (Autonomous Power State Transition) lets us program a
2533 * table of power state transitions that the controller will
2534 * perform automatically. We configure it with a simple
2535 * heuristic: we are willing to spend at most 2% of the time
2536 * transitioning between power states. Therefore, when running
2537 * in any given state, we will enter the next lower-power
2538 * non-operational state after waiting 50 * (enlat + exlat)
2539 * microseconds, as long as that state's exit latency is under
2540 * the requested maximum latency.
2542 * We will not autonomously enter any non-operational state for
2543 * which the total latency exceeds ps_max_latency_us. Users
2544 * can set ps_max_latency_us to zero to turn off APST.
2548 struct nvme_feat_auto_pst *table;
2554 * If APST isn't supported or if we haven't been initialized yet,
2555 * then don't do anything.
2560 if (ctrl->npss > 31) {
2561 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2565 table = kzalloc(sizeof(*table), GFP_KERNEL);
2569 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2570 /* Turn off APST. */
2572 dev_dbg(ctrl->device, "APST disabled\n");
2574 __le64 target = cpu_to_le64(0);
2578 * Walk through all states from lowest- to highest-power.
2579 * According to the spec, lower-numbered states use more
2580 * power. NPSS, despite the name, is the index of the
2581 * lowest-power state, not the number of states.
2583 for (state = (int)ctrl->npss; state >= 0; state--) {
2584 u64 total_latency_us, exit_latency_us, transition_ms;
2587 table->entries[state] = target;
2590 * Don't allow transitions to the deepest state
2591 * if it's quirked off.
2593 if (state == ctrl->npss &&
2594 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2598 * Is this state a useful non-operational state for
2599 * higher-power states to autonomously transition to?
2601 if (!(ctrl->psd[state].flags &
2602 NVME_PS_FLAGS_NON_OP_STATE))
2606 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2607 if (exit_latency_us > ctrl->ps_max_latency_us)
2612 le32_to_cpu(ctrl->psd[state].entry_lat);
2615 * This state is good. Use it as the APST idle
2616 * target for higher power states.
2618 transition_ms = total_latency_us + 19;
2619 do_div(transition_ms, 20);
2620 if (transition_ms > (1 << 24) - 1)
2621 transition_ms = (1 << 24) - 1;
2623 target = cpu_to_le64((state << 3) |
2624 (transition_ms << 8));
2629 if (total_latency_us > max_lat_us)
2630 max_lat_us = total_latency_us;
2636 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2638 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2639 max_ps, max_lat_us, (int)sizeof(*table), table);
2643 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2644 table, sizeof(*table), NULL);
2646 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2652 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2654 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2658 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2659 case PM_QOS_LATENCY_ANY:
2667 if (ctrl->ps_max_latency_us != latency) {
2668 ctrl->ps_max_latency_us = latency;
2669 if (ctrl->state == NVME_CTRL_LIVE)
2670 nvme_configure_apst(ctrl);
2674 struct nvme_core_quirk_entry {
2676 * NVMe model and firmware strings are padded with spaces. For
2677 * simplicity, strings in the quirk table are padded with NULLs
2683 unsigned long quirks;
2686 static const struct nvme_core_quirk_entry core_quirks[] = {
2689 * This Toshiba device seems to die using any APST states. See:
2690 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2693 .mn = "THNSF5256GPUK TOSHIBA",
2694 .quirks = NVME_QUIRK_NO_APST,
2698 * This LiteON CL1-3D*-Q11 firmware version has a race
2699 * condition associated with actions related to suspend to idle
2700 * LiteON has resolved the problem in future firmware
2704 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2708 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2709 * aborts I/O during any load, but more easily reproducible
2710 * with discards (fstrim).
2712 * The device is left in a state where it is also not possible
2713 * to use "nvme set-feature" to disable APST, but booting with
2714 * nvme_core.default_ps_max_latency=0 works.
2717 .mn = "KCD6XVUL6T40",
2718 .quirks = NVME_QUIRK_NO_APST,
2722 * The external Samsung X5 SSD fails initialization without a
2723 * delay before checking if it is ready and has a whole set of
2724 * other problems. To make this even more interesting, it
2725 * shares the PCI ID with internal Samsung 970 Evo Plus that
2726 * does not need or want these quirks.
2729 .mn = "Samsung Portable SSD X5",
2730 .quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2731 NVME_QUIRK_NO_DEEPEST_PS |
2732 NVME_QUIRK_IGNORE_DEV_SUBNQN,
2736 /* match is null-terminated but idstr is space-padded. */
2737 static bool string_matches(const char *idstr, const char *match, size_t len)
2744 matchlen = strlen(match);
2745 WARN_ON_ONCE(matchlen > len);
2747 if (memcmp(idstr, match, matchlen))
2750 for (; matchlen < len; matchlen++)
2751 if (idstr[matchlen] != ' ')
2757 static bool quirk_matches(const struct nvme_id_ctrl *id,
2758 const struct nvme_core_quirk_entry *q)
2760 return q->vid == le16_to_cpu(id->vid) &&
2761 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2762 string_matches(id->fr, q->fr, sizeof(id->fr));
2765 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2766 struct nvme_id_ctrl *id)
2771 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2772 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2773 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2774 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2778 if (ctrl->vs >= NVME_VS(1, 2, 1))
2779 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2782 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2783 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2784 "nqn.2014.08.org.nvmexpress:%04x%04x",
2785 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2786 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2787 off += sizeof(id->sn);
2788 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2789 off += sizeof(id->mn);
2790 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2793 static void nvme_release_subsystem(struct device *dev)
2795 struct nvme_subsystem *subsys =
2796 container_of(dev, struct nvme_subsystem, dev);
2798 if (subsys->instance >= 0)
2799 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2803 static void nvme_destroy_subsystem(struct kref *ref)
2805 struct nvme_subsystem *subsys =
2806 container_of(ref, struct nvme_subsystem, ref);
2808 mutex_lock(&nvme_subsystems_lock);
2809 list_del(&subsys->entry);
2810 mutex_unlock(&nvme_subsystems_lock);
2812 ida_destroy(&subsys->ns_ida);
2813 device_del(&subsys->dev);
2814 put_device(&subsys->dev);
2817 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2819 kref_put(&subsys->ref, nvme_destroy_subsystem);
2822 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2824 struct nvme_subsystem *subsys;
2826 lockdep_assert_held(&nvme_subsystems_lock);
2829 * Fail matches for discovery subsystems. This results
2830 * in each discovery controller bound to a unique subsystem.
2831 * This avoids issues with validating controller values
2832 * that can only be true when there is a single unique subsystem.
2833 * There may be multiple and completely independent entities
2834 * that provide discovery controllers.
2836 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2839 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2840 if (strcmp(subsys->subnqn, subsysnqn))
2842 if (!kref_get_unless_zero(&subsys->ref))
2850 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2851 struct device_attribute subsys_attr_##_name = \
2852 __ATTR(_name, _mode, _show, NULL)
2854 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2855 struct device_attribute *attr,
2858 struct nvme_subsystem *subsys =
2859 container_of(dev, struct nvme_subsystem, dev);
2861 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2863 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2865 #define nvme_subsys_show_str_function(field) \
2866 static ssize_t subsys_##field##_show(struct device *dev, \
2867 struct device_attribute *attr, char *buf) \
2869 struct nvme_subsystem *subsys = \
2870 container_of(dev, struct nvme_subsystem, dev); \
2871 return sysfs_emit(buf, "%.*s\n", \
2872 (int)sizeof(subsys->field), subsys->field); \
2874 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2876 nvme_subsys_show_str_function(model);
2877 nvme_subsys_show_str_function(serial);
2878 nvme_subsys_show_str_function(firmware_rev);
2880 static struct attribute *nvme_subsys_attrs[] = {
2881 &subsys_attr_model.attr,
2882 &subsys_attr_serial.attr,
2883 &subsys_attr_firmware_rev.attr,
2884 &subsys_attr_subsysnqn.attr,
2885 #ifdef CONFIG_NVME_MULTIPATH
2886 &subsys_attr_iopolicy.attr,
2891 static struct attribute_group nvme_subsys_attrs_group = {
2892 .attrs = nvme_subsys_attrs,
2895 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2896 &nvme_subsys_attrs_group,
2900 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2902 return ctrl->opts && ctrl->opts->discovery_nqn;
2905 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2906 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2908 struct nvme_ctrl *tmp;
2910 lockdep_assert_held(&nvme_subsystems_lock);
2912 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2913 if (nvme_state_terminal(tmp))
2916 if (tmp->cntlid == ctrl->cntlid) {
2917 dev_err(ctrl->device,
2918 "Duplicate cntlid %u with %s, rejecting\n",
2919 ctrl->cntlid, dev_name(tmp->device));
2923 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2924 nvme_discovery_ctrl(ctrl))
2927 dev_err(ctrl->device,
2928 "Subsystem does not support multiple controllers\n");
2935 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2937 struct nvme_subsystem *subsys, *found;
2940 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2944 subsys->instance = -1;
2945 mutex_init(&subsys->lock);
2946 kref_init(&subsys->ref);
2947 INIT_LIST_HEAD(&subsys->ctrls);
2948 INIT_LIST_HEAD(&subsys->nsheads);
2949 nvme_init_subnqn(subsys, ctrl, id);
2950 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2951 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2952 subsys->vendor_id = le16_to_cpu(id->vid);
2953 subsys->cmic = id->cmic;
2954 subsys->awupf = le16_to_cpu(id->awupf);
2955 #ifdef CONFIG_NVME_MULTIPATH
2956 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2959 subsys->dev.class = nvme_subsys_class;
2960 subsys->dev.release = nvme_release_subsystem;
2961 subsys->dev.groups = nvme_subsys_attrs_groups;
2962 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2963 device_initialize(&subsys->dev);
2965 mutex_lock(&nvme_subsystems_lock);
2966 found = __nvme_find_get_subsystem(subsys->subnqn);
2968 put_device(&subsys->dev);
2971 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2973 goto out_put_subsystem;
2976 ret = device_add(&subsys->dev);
2978 dev_err(ctrl->device,
2979 "failed to register subsystem device.\n");
2980 put_device(&subsys->dev);
2983 ida_init(&subsys->ns_ida);
2984 list_add_tail(&subsys->entry, &nvme_subsystems);
2987 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2988 dev_name(ctrl->device));
2990 dev_err(ctrl->device,
2991 "failed to create sysfs link from subsystem.\n");
2992 goto out_put_subsystem;
2996 subsys->instance = ctrl->instance;
2997 ctrl->subsys = subsys;
2998 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2999 mutex_unlock(&nvme_subsystems_lock);
3003 nvme_put_subsystem(subsys);
3005 mutex_unlock(&nvme_subsystems_lock);
3009 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
3010 void *log, size_t size, u64 offset)
3012 struct nvme_command c = { };
3013 u32 dwlen = nvme_bytes_to_numd(size);
3015 c.get_log_page.opcode = nvme_admin_get_log_page;
3016 c.get_log_page.nsid = cpu_to_le32(nsid);
3017 c.get_log_page.lid = log_page;
3018 c.get_log_page.lsp = lsp;
3019 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3020 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3021 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3022 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3023 c.get_log_page.csi = csi;
3025 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3028 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3029 struct nvme_effects_log **log)
3031 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3037 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3041 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3042 cel, sizeof(*cel), 0);
3048 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3055 * Initialize the cached copies of the Identify data and various controller
3056 * register in our nvme_ctrl structure. This should be called as soon as
3057 * the admin queue is fully up and running.
3059 int nvme_init_identify(struct nvme_ctrl *ctrl)
3061 struct nvme_id_ctrl *id;
3062 int ret, page_shift;
3064 bool prev_apst_enabled;
3066 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3068 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3071 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
3072 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3074 if (ctrl->vs >= NVME_VS(1, 1, 0))
3075 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3077 ret = nvme_identify_ctrl(ctrl, &id);
3079 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3083 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3084 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3089 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3090 ctrl->cntlid = le16_to_cpu(id->cntlid);
3092 if (!ctrl->identified) {
3095 ret = nvme_init_subsystem(ctrl, id);
3100 * Check for quirks. Quirk can depend on firmware version,
3101 * so, in principle, the set of quirks present can change
3102 * across a reset. As a possible future enhancement, we
3103 * could re-scan for quirks every time we reinitialize
3104 * the device, but we'd have to make sure that the driver
3105 * behaves intelligently if the quirks change.
3107 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3108 if (quirk_matches(id, &core_quirks[i]))
3109 ctrl->quirks |= core_quirks[i].quirks;
3112 memcpy(ctrl->subsys->firmware_rev, id->fr,
3113 sizeof(ctrl->subsys->firmware_rev));
3115 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3116 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3117 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3120 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3121 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3122 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3124 ctrl->oacs = le16_to_cpu(id->oacs);
3125 ctrl->oncs = le16_to_cpu(id->oncs);
3126 ctrl->mtfa = le16_to_cpu(id->mtfa);
3127 ctrl->oaes = le32_to_cpu(id->oaes);
3128 ctrl->wctemp = le16_to_cpu(id->wctemp);
3129 ctrl->cctemp = le16_to_cpu(id->cctemp);
3131 atomic_set(&ctrl->abort_limit, id->acl + 1);
3132 ctrl->vwc = id->vwc;
3134 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
3136 max_hw_sectors = UINT_MAX;
3137 ctrl->max_hw_sectors =
3138 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3140 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3141 ctrl->sgls = le32_to_cpu(id->sgls);
3142 ctrl->kas = le16_to_cpu(id->kas);
3143 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3144 ctrl->ctratt = le32_to_cpu(id->ctratt);
3148 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3150 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3151 shutdown_timeout, 60);
3153 if (ctrl->shutdown_timeout != shutdown_timeout)
3154 dev_info(ctrl->device,
3155 "Shutdown timeout set to %u seconds\n",
3156 ctrl->shutdown_timeout);
3158 ctrl->shutdown_timeout = shutdown_timeout;
3160 ctrl->npss = id->npss;
3161 ctrl->apsta = id->apsta;
3162 prev_apst_enabled = ctrl->apst_enabled;
3163 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3164 if (force_apst && id->apsta) {
3165 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3166 ctrl->apst_enabled = true;
3168 ctrl->apst_enabled = false;
3171 ctrl->apst_enabled = id->apsta;
3173 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3175 if (ctrl->ops->flags & NVME_F_FABRICS) {
3176 ctrl->icdoff = le16_to_cpu(id->icdoff);
3177 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3178 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3179 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3182 * In fabrics we need to verify the cntlid matches the
3185 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3186 dev_err(ctrl->device,
3187 "Mismatching cntlid: Connect %u vs Identify "
3189 ctrl->cntlid, le16_to_cpu(id->cntlid));
3194 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3195 dev_err(ctrl->device,
3196 "keep-alive support is mandatory for fabrics\n");
3201 ctrl->hmpre = le32_to_cpu(id->hmpre);
3202 ctrl->hmmin = le32_to_cpu(id->hmmin);
3203 ctrl->hmminds = le32_to_cpu(id->hmminds);
3204 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3207 ret = nvme_mpath_init_identify(ctrl, id);
3213 if (ctrl->apst_enabled && !prev_apst_enabled)
3214 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3215 else if (!ctrl->apst_enabled && prev_apst_enabled)
3216 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3218 ret = nvme_configure_apst(ctrl);
3222 ret = nvme_configure_timestamp(ctrl);
3226 ret = nvme_configure_directives(ctrl);
3230 ret = nvme_configure_acre(ctrl);
3234 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3236 * Do not return errors unless we are in a controller reset,
3237 * the controller works perfectly fine without hwmon.
3239 ret = nvme_hwmon_init(ctrl);
3244 ctrl->identified = true;
3252 EXPORT_SYMBOL_GPL(nvme_init_identify);
3254 static int nvme_dev_open(struct inode *inode, struct file *file)
3256 struct nvme_ctrl *ctrl =
3257 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3259 switch (ctrl->state) {
3260 case NVME_CTRL_LIVE:
3263 return -EWOULDBLOCK;
3266 nvme_get_ctrl(ctrl);
3267 if (!try_module_get(ctrl->ops->module)) {
3268 nvme_put_ctrl(ctrl);
3272 file->private_data = ctrl;
3276 static int nvme_dev_release(struct inode *inode, struct file *file)
3278 struct nvme_ctrl *ctrl =
3279 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3281 module_put(ctrl->ops->module);
3282 nvme_put_ctrl(ctrl);
3286 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3291 down_read(&ctrl->namespaces_rwsem);
3292 if (list_empty(&ctrl->namespaces)) {
3297 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3298 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3299 dev_warn(ctrl->device,
3300 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3305 dev_warn(ctrl->device,
3306 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3307 kref_get(&ns->kref);
3308 up_read(&ctrl->namespaces_rwsem);
3310 ret = nvme_user_cmd(ctrl, ns, argp);
3315 up_read(&ctrl->namespaces_rwsem);
3319 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3322 struct nvme_ctrl *ctrl = file->private_data;
3323 void __user *argp = (void __user *)arg;
3326 case NVME_IOCTL_ADMIN_CMD:
3327 return nvme_user_cmd(ctrl, NULL, argp);
3328 case NVME_IOCTL_ADMIN64_CMD:
3329 return nvme_user_cmd64(ctrl, NULL, argp);
3330 case NVME_IOCTL_IO_CMD:
3331 return nvme_dev_user_cmd(ctrl, argp);
3332 case NVME_IOCTL_RESET:
3333 dev_warn(ctrl->device, "resetting controller\n");
3334 return nvme_reset_ctrl_sync(ctrl);
3335 case NVME_IOCTL_SUBSYS_RESET:
3336 return nvme_reset_subsystem(ctrl);
3337 case NVME_IOCTL_RESCAN:
3338 nvme_queue_scan(ctrl);
3345 static const struct file_operations nvme_dev_fops = {
3346 .owner = THIS_MODULE,
3347 .open = nvme_dev_open,
3348 .release = nvme_dev_release,
3349 .unlocked_ioctl = nvme_dev_ioctl,
3350 .compat_ioctl = compat_ptr_ioctl,
3353 static ssize_t nvme_sysfs_reset(struct device *dev,
3354 struct device_attribute *attr, const char *buf,
3357 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3360 ret = nvme_reset_ctrl_sync(ctrl);
3365 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3367 static ssize_t nvme_sysfs_rescan(struct device *dev,
3368 struct device_attribute *attr, const char *buf,
3371 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3373 nvme_queue_scan(ctrl);
3376 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3378 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3380 struct gendisk *disk = dev_to_disk(dev);
3382 if (disk->fops == &nvme_fops)
3383 return nvme_get_ns_from_dev(dev)->head;
3385 return disk->private_data;
3388 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3391 struct nvme_ns_head *head = dev_to_ns_head(dev);
3392 struct nvme_ns_ids *ids = &head->ids;
3393 struct nvme_subsystem *subsys = head->subsys;
3394 int serial_len = sizeof(subsys->serial);
3395 int model_len = sizeof(subsys->model);
3397 if (!uuid_is_null(&ids->uuid))
3398 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3400 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3401 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3403 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3404 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3406 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3407 subsys->serial[serial_len - 1] == '\0'))
3409 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3410 subsys->model[model_len - 1] == '\0'))
3413 return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3414 serial_len, subsys->serial, model_len, subsys->model,
3417 static DEVICE_ATTR_RO(wwid);
3419 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3422 return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3424 static DEVICE_ATTR_RO(nguid);
3426 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3429 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3431 /* For backward compatibility expose the NGUID to userspace if
3432 * we have no UUID set
3434 if (uuid_is_null(&ids->uuid)) {
3435 dev_warn_ratelimited(dev,
3436 "No UUID available providing old NGUID\n");
3437 return sysfs_emit(buf, "%pU\n", ids->nguid);
3439 return sysfs_emit(buf, "%pU\n", &ids->uuid);
3441 static DEVICE_ATTR_RO(uuid);
3443 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3446 return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3448 static DEVICE_ATTR_RO(eui);
3450 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3453 return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3455 static DEVICE_ATTR_RO(nsid);
3457 static struct attribute *nvme_ns_id_attrs[] = {
3458 &dev_attr_wwid.attr,
3459 &dev_attr_uuid.attr,
3460 &dev_attr_nguid.attr,
3462 &dev_attr_nsid.attr,
3463 #ifdef CONFIG_NVME_MULTIPATH
3464 &dev_attr_ana_grpid.attr,
3465 &dev_attr_ana_state.attr,
3470 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3471 struct attribute *a, int n)
3473 struct device *dev = container_of(kobj, struct device, kobj);
3474 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3476 if (a == &dev_attr_uuid.attr) {
3477 if (uuid_is_null(&ids->uuid) &&
3478 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3481 if (a == &dev_attr_nguid.attr) {
3482 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3485 if (a == &dev_attr_eui.attr) {
3486 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3489 #ifdef CONFIG_NVME_MULTIPATH
3490 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3491 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3493 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3500 static const struct attribute_group nvme_ns_id_attr_group = {
3501 .attrs = nvme_ns_id_attrs,
3502 .is_visible = nvme_ns_id_attrs_are_visible,
3505 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3506 &nvme_ns_id_attr_group,
3508 &nvme_nvm_attr_group,
3513 #define nvme_show_str_function(field) \
3514 static ssize_t field##_show(struct device *dev, \
3515 struct device_attribute *attr, char *buf) \
3517 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3518 return sysfs_emit(buf, "%.*s\n", \
3519 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3521 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3523 nvme_show_str_function(model);
3524 nvme_show_str_function(serial);
3525 nvme_show_str_function(firmware_rev);
3527 #define nvme_show_int_function(field) \
3528 static ssize_t field##_show(struct device *dev, \
3529 struct device_attribute *attr, char *buf) \
3531 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3532 return sysfs_emit(buf, "%d\n", ctrl->field); \
3534 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3536 nvme_show_int_function(cntlid);
3537 nvme_show_int_function(numa_node);
3538 nvme_show_int_function(queue_count);
3539 nvme_show_int_function(sqsize);
3541 static ssize_t nvme_sysfs_delete(struct device *dev,
3542 struct device_attribute *attr, const char *buf,
3545 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3547 if (device_remove_file_self(dev, attr))
3548 nvme_delete_ctrl_sync(ctrl);
3551 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3553 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3554 struct device_attribute *attr,
3557 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3559 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3561 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3563 static ssize_t nvme_sysfs_show_state(struct device *dev,
3564 struct device_attribute *attr,
3567 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3568 static const char *const state_name[] = {
3569 [NVME_CTRL_NEW] = "new",
3570 [NVME_CTRL_LIVE] = "live",
3571 [NVME_CTRL_RESETTING] = "resetting",
3572 [NVME_CTRL_CONNECTING] = "connecting",
3573 [NVME_CTRL_DELETING] = "deleting",
3574 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3575 [NVME_CTRL_DEAD] = "dead",
3578 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3579 state_name[ctrl->state])
3580 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3582 return sysfs_emit(buf, "unknown state\n");
3585 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3587 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3588 struct device_attribute *attr,
3591 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3593 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3595 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3597 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3598 struct device_attribute *attr,
3601 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3603 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3605 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3607 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3608 struct device_attribute *attr,
3611 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3613 return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3615 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3617 static ssize_t nvme_sysfs_show_address(struct device *dev,
3618 struct device_attribute *attr,
3621 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3623 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3625 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3627 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3628 struct device_attribute *attr, char *buf)
3630 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3631 struct nvmf_ctrl_options *opts = ctrl->opts;
3633 if (ctrl->opts->max_reconnects == -1)
3634 return sysfs_emit(buf, "off\n");
3635 return sysfs_emit(buf, "%d\n",
3636 opts->max_reconnects * opts->reconnect_delay);
3639 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3640 struct device_attribute *attr, const char *buf, size_t count)
3642 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3643 struct nvmf_ctrl_options *opts = ctrl->opts;
3644 int ctrl_loss_tmo, err;
3646 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3650 else if (ctrl_loss_tmo < 0)
3651 opts->max_reconnects = -1;
3653 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3654 opts->reconnect_delay);
3657 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3658 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3660 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3661 struct device_attribute *attr, char *buf)
3663 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3665 if (ctrl->opts->reconnect_delay == -1)
3666 return sysfs_emit(buf, "off\n");
3667 return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3670 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3671 struct device_attribute *attr, const char *buf, size_t count)
3673 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3677 err = kstrtou32(buf, 10, &v);
3681 ctrl->opts->reconnect_delay = v;
3684 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3685 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3687 static struct attribute *nvme_dev_attrs[] = {
3688 &dev_attr_reset_controller.attr,
3689 &dev_attr_rescan_controller.attr,
3690 &dev_attr_model.attr,
3691 &dev_attr_serial.attr,
3692 &dev_attr_firmware_rev.attr,
3693 &dev_attr_cntlid.attr,
3694 &dev_attr_delete_controller.attr,
3695 &dev_attr_transport.attr,
3696 &dev_attr_subsysnqn.attr,
3697 &dev_attr_address.attr,
3698 &dev_attr_state.attr,
3699 &dev_attr_numa_node.attr,
3700 &dev_attr_queue_count.attr,
3701 &dev_attr_sqsize.attr,
3702 &dev_attr_hostnqn.attr,
3703 &dev_attr_hostid.attr,
3704 &dev_attr_ctrl_loss_tmo.attr,
3705 &dev_attr_reconnect_delay.attr,
3709 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3710 struct attribute *a, int n)
3712 struct device *dev = container_of(kobj, struct device, kobj);
3713 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3715 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3717 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3719 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3721 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3723 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3725 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3731 static struct attribute_group nvme_dev_attrs_group = {
3732 .attrs = nvme_dev_attrs,
3733 .is_visible = nvme_dev_attrs_are_visible,
3736 static const struct attribute_group *nvme_dev_attr_groups[] = {
3737 &nvme_dev_attrs_group,
3741 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3744 struct nvme_ns_head *h;
3746 lockdep_assert_held(&subsys->lock);
3748 list_for_each_entry(h, &subsys->nsheads, entry) {
3749 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3756 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3757 struct nvme_ns_ids *ids)
3759 struct nvme_ns_head *h;
3761 lockdep_assert_held(&subsys->lock);
3763 list_for_each_entry(h, &subsys->nsheads, entry) {
3764 if (nvme_ns_ids_valid(ids) && nvme_ns_ids_equal(ids, &h->ids))
3771 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3772 unsigned nsid, struct nvme_ns_ids *ids)
3774 struct nvme_ns_head *head;
3775 size_t size = sizeof(*head);
3778 #ifdef CONFIG_NVME_MULTIPATH
3779 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3782 head = kzalloc(size, GFP_KERNEL);
3785 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3788 head->instance = ret;
3789 INIT_LIST_HEAD(&head->list);
3790 ret = init_srcu_struct(&head->srcu);
3792 goto out_ida_remove;
3793 head->subsys = ctrl->subsys;
3796 kref_init(&head->ref);
3798 ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &head->ids);
3800 dev_err(ctrl->device,
3801 "duplicate IDs for nsid %d\n", nsid);
3802 goto out_cleanup_srcu;
3805 if (head->ids.csi) {
3806 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3808 goto out_cleanup_srcu;
3810 head->effects = ctrl->effects;
3812 ret = nvme_mpath_alloc_disk(ctrl, head);
3814 goto out_cleanup_srcu;
3816 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3818 kref_get(&ctrl->subsys->ref);
3822 cleanup_srcu_struct(&head->srcu);
3824 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3829 ret = blk_status_to_errno(nvme_error_status(ret));
3830 return ERR_PTR(ret);
3833 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3834 struct nvme_ns_ids *ids, bool is_shared)
3836 struct nvme_ctrl *ctrl = ns->ctrl;
3837 struct nvme_ns_head *head = NULL;
3840 mutex_lock(&ctrl->subsys->lock);
3841 head = nvme_find_ns_head(ctrl->subsys, nsid);
3843 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3845 ret = PTR_ERR(head);
3848 head->shared = is_shared;
3851 if (!is_shared || !head->shared) {
3852 dev_err(ctrl->device,
3853 "Duplicate unshared namespace %d\n", nsid);
3854 goto out_put_ns_head;
3856 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3857 dev_err(ctrl->device,
3858 "IDs don't match for shared namespace %d\n",
3860 goto out_put_ns_head;
3864 list_add_tail(&ns->siblings, &head->list);
3866 mutex_unlock(&ctrl->subsys->lock);
3870 nvme_put_ns_head(head);
3872 mutex_unlock(&ctrl->subsys->lock);
3876 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3878 struct nvme_ns *ns, *ret = NULL;
3880 down_read(&ctrl->namespaces_rwsem);
3881 list_for_each_entry(ns, &ctrl->namespaces, list) {
3882 if (ns->head->ns_id == nsid) {
3883 if (!kref_get_unless_zero(&ns->kref))
3888 if (ns->head->ns_id > nsid)
3891 up_read(&ctrl->namespaces_rwsem);
3894 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3897 * Add the namespace to the controller list while keeping the list ordered.
3899 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3901 struct nvme_ns *tmp;
3903 list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3904 if (tmp->head->ns_id < ns->head->ns_id) {
3905 list_add(&ns->list, &tmp->list);
3909 list_add(&ns->list, &ns->ctrl->namespaces);
3912 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3913 struct nvme_ns_ids *ids)
3916 struct gendisk *disk;
3917 struct nvme_id_ns *id;
3918 char disk_name[DISK_NAME_LEN];
3919 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3921 if (nvme_identify_ns(ctrl, nsid, ids, &id))
3924 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3928 ns->queue = blk_mq_init_queue(ctrl->tagset);
3929 if (IS_ERR(ns->queue))
3932 if (ctrl->opts && ctrl->opts->data_digest)
3933 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3935 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3936 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3937 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3939 ns->queue->queuedata = ns;
3941 kref_init(&ns->kref);
3943 ret = nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED);
3945 goto out_free_queue;
3946 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3948 disk = alloc_disk_node(0, node);
3952 disk->fops = &nvme_fops;
3953 disk->private_data = ns;
3954 disk->queue = ns->queue;
3955 disk->flags = flags;
3956 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3959 if (nvme_update_ns_info(ns, id))
3962 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3963 ret = nvme_nvm_register(ns, disk_name, node);
3965 dev_warn(ctrl->device, "LightNVM init failure\n");
3970 down_write(&ctrl->namespaces_rwsem);
3971 nvme_ns_add_to_ctrl_list(ns);
3972 up_write(&ctrl->namespaces_rwsem);
3973 nvme_get_ctrl(ctrl);
3975 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3977 nvme_mpath_add_disk(ns, id);
3978 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3983 /* prevent double queue cleanup */
3984 ns->disk->queue = NULL;
3987 mutex_lock(&ctrl->subsys->lock);
3988 list_del_rcu(&ns->siblings);
3989 if (list_empty(&ns->head->list))
3990 list_del_init(&ns->head->entry);
3991 mutex_unlock(&ctrl->subsys->lock);
3992 nvme_put_ns_head(ns->head);
3994 blk_cleanup_queue(ns->queue);
4001 static void nvme_ns_remove(struct nvme_ns *ns)
4003 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4006 set_capacity(ns->disk, 0);
4007 nvme_fault_inject_fini(&ns->fault_inject);
4009 mutex_lock(&ns->ctrl->subsys->lock);
4010 list_del_rcu(&ns->siblings);
4011 if (list_empty(&ns->head->list))
4012 list_del_init(&ns->head->entry);
4013 mutex_unlock(&ns->ctrl->subsys->lock);
4015 synchronize_rcu(); /* guarantee not available in head->list */
4016 nvme_mpath_clear_current_path(ns);
4017 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
4019 if (ns->disk->flags & GENHD_FL_UP) {
4020 del_gendisk(ns->disk);
4021 blk_cleanup_queue(ns->queue);
4022 if (blk_get_integrity(ns->disk))
4023 blk_integrity_unregister(ns->disk);
4026 down_write(&ns->ctrl->namespaces_rwsem);
4027 list_del_init(&ns->list);
4028 up_write(&ns->ctrl->namespaces_rwsem);
4030 nvme_mpath_check_last_path(ns);
4034 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4036 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4044 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4046 struct nvme_id_ns *id;
4047 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4049 if (test_bit(NVME_NS_DEAD, &ns->flags))
4052 ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4056 ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4057 if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4058 dev_err(ns->ctrl->device,
4059 "identifiers changed for nsid %d\n", ns->head->ns_id);
4063 ret = nvme_update_ns_info(ns, id);
4069 * Only remove the namespace if we got a fatal error back from the
4070 * device, otherwise ignore the error and just move on.
4072 * TODO: we should probably schedule a delayed retry here.
4074 if (ret > 0 && (ret & NVME_SC_DNR))
4077 revalidate_disk_size(ns->disk, true);
4080 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4082 struct nvme_ns_ids ids = { };
4085 if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4088 ns = nvme_find_get_ns(ctrl, nsid);
4090 nvme_validate_ns(ns, &ids);
4097 nvme_alloc_ns(ctrl, nsid, &ids);
4100 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4101 dev_warn(ctrl->device,
4102 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4106 if (!nvme_multi_css(ctrl)) {
4107 dev_warn(ctrl->device,
4108 "command set not reported for nsid: %d\n",
4112 nvme_alloc_ns(ctrl, nsid, &ids);
4115 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4121 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4124 struct nvme_ns *ns, *next;
4127 down_write(&ctrl->namespaces_rwsem);
4128 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4129 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4130 list_move_tail(&ns->list, &rm_list);
4132 up_write(&ctrl->namespaces_rwsem);
4134 list_for_each_entry_safe(ns, next, &rm_list, list)
4139 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4141 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4146 if (nvme_ctrl_limited_cns(ctrl))
4149 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4154 struct nvme_command cmd = {
4155 .identify.opcode = nvme_admin_identify,
4156 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4157 .identify.nsid = cpu_to_le32(prev),
4160 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4161 NVME_IDENTIFY_DATA_SIZE);
4165 for (i = 0; i < nr_entries; i++) {
4166 u32 nsid = le32_to_cpu(ns_list[i]);
4168 if (!nsid) /* end of the list? */
4170 nvme_validate_or_alloc_ns(ctrl, nsid);
4171 while (++prev < nsid)
4172 nvme_ns_remove_by_nsid(ctrl, prev);
4176 nvme_remove_invalid_namespaces(ctrl, prev);
4182 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4184 struct nvme_id_ctrl *id;
4187 if (nvme_identify_ctrl(ctrl, &id))
4189 nn = le32_to_cpu(id->nn);
4192 for (i = 1; i <= nn; i++)
4193 nvme_validate_or_alloc_ns(ctrl, i);
4195 nvme_remove_invalid_namespaces(ctrl, nn);
4198 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4200 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4204 log = kzalloc(log_size, GFP_KERNEL);
4209 * We need to read the log to clear the AEN, but we don't want to rely
4210 * on it for the changed namespace information as userspace could have
4211 * raced with us in reading the log page, which could cause us to miss
4214 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4215 NVME_CSI_NVM, log, log_size, 0);
4217 dev_warn(ctrl->device,
4218 "reading changed ns log failed: %d\n", error);
4223 static void nvme_scan_work(struct work_struct *work)
4225 struct nvme_ctrl *ctrl =
4226 container_of(work, struct nvme_ctrl, scan_work);
4228 /* No tagset on a live ctrl means IO queues could not created */
4229 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4232 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4233 dev_info(ctrl->device, "rescanning namespaces.\n");
4234 nvme_clear_changed_ns_log(ctrl);
4237 mutex_lock(&ctrl->scan_lock);
4238 if (nvme_scan_ns_list(ctrl) != 0)
4239 nvme_scan_ns_sequential(ctrl);
4240 mutex_unlock(&ctrl->scan_lock);
4244 * This function iterates the namespace list unlocked to allow recovery from
4245 * controller failure. It is up to the caller to ensure the namespace list is
4246 * not modified by scan work while this function is executing.
4248 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4250 struct nvme_ns *ns, *next;
4254 * make sure to requeue I/O to all namespaces as these
4255 * might result from the scan itself and must complete
4256 * for the scan_work to make progress
4258 nvme_mpath_clear_ctrl_paths(ctrl);
4260 /* prevent racing with ns scanning */
4261 flush_work(&ctrl->scan_work);
4264 * The dead states indicates the controller was not gracefully
4265 * disconnected. In that case, we won't be able to flush any data while
4266 * removing the namespaces' disks; fail all the queues now to avoid
4267 * potentially having to clean up the failed sync later.
4269 if (ctrl->state == NVME_CTRL_DEAD)
4270 nvme_kill_queues(ctrl);
4272 /* this is a no-op when called from the controller reset handler */
4273 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4275 down_write(&ctrl->namespaces_rwsem);
4276 list_splice_init(&ctrl->namespaces, &ns_list);
4277 up_write(&ctrl->namespaces_rwsem);
4279 list_for_each_entry_safe(ns, next, &ns_list, list)
4282 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4284 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4286 struct nvme_ctrl *ctrl =
4287 container_of(dev, struct nvme_ctrl, ctrl_device);
4288 struct nvmf_ctrl_options *opts = ctrl->opts;
4291 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4296 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4300 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4301 opts->trsvcid ?: "none");
4305 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4306 opts->host_traddr ?: "none");
4311 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4313 char *envp[2] = { NULL, NULL };
4314 u32 aen_result = ctrl->aen_result;
4316 ctrl->aen_result = 0;
4320 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4323 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4327 static void nvme_async_event_work(struct work_struct *work)
4329 struct nvme_ctrl *ctrl =
4330 container_of(work, struct nvme_ctrl, async_event_work);
4332 nvme_aen_uevent(ctrl);
4335 * The transport drivers must guarantee AER submission here is safe by
4336 * flushing ctrl async_event_work after changing the controller state
4337 * from LIVE and before freeing the admin queue.
4339 if (ctrl->state == NVME_CTRL_LIVE)
4340 ctrl->ops->submit_async_event(ctrl);
4343 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4348 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4354 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4357 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4359 struct nvme_fw_slot_info_log *log;
4361 log = kmalloc(sizeof(*log), GFP_KERNEL);
4365 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4366 log, sizeof(*log), 0))
4367 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4371 static void nvme_fw_act_work(struct work_struct *work)
4373 struct nvme_ctrl *ctrl = container_of(work,
4374 struct nvme_ctrl, fw_act_work);
4375 unsigned long fw_act_timeout;
4378 fw_act_timeout = jiffies +
4379 msecs_to_jiffies(ctrl->mtfa * 100);
4381 fw_act_timeout = jiffies +
4382 msecs_to_jiffies(admin_timeout * 1000);
4384 nvme_stop_queues(ctrl);
4385 while (nvme_ctrl_pp_status(ctrl)) {
4386 if (time_after(jiffies, fw_act_timeout)) {
4387 dev_warn(ctrl->device,
4388 "Fw activation timeout, reset controller\n");
4389 nvme_try_sched_reset(ctrl);
4395 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4398 nvme_start_queues(ctrl);
4399 /* read FW slot information to clear the AER */
4400 nvme_get_fw_slot_info(ctrl);
4403 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4405 u32 aer_notice_type = (result & 0xff00) >> 8;
4407 trace_nvme_async_event(ctrl, aer_notice_type);
4409 switch (aer_notice_type) {
4410 case NVME_AER_NOTICE_NS_CHANGED:
4411 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4412 nvme_queue_scan(ctrl);
4414 case NVME_AER_NOTICE_FW_ACT_STARTING:
4416 * We are (ab)using the RESETTING state to prevent subsequent
4417 * recovery actions from interfering with the controller's
4418 * firmware activation.
4420 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4421 queue_work(nvme_wq, &ctrl->fw_act_work);
4423 #ifdef CONFIG_NVME_MULTIPATH
4424 case NVME_AER_NOTICE_ANA:
4425 if (!ctrl->ana_log_buf)
4427 queue_work(nvme_wq, &ctrl->ana_work);
4430 case NVME_AER_NOTICE_DISC_CHANGED:
4431 ctrl->aen_result = result;
4434 dev_warn(ctrl->device, "async event result %08x\n", result);
4438 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4439 volatile union nvme_result *res)
4441 u32 result = le32_to_cpu(res->u32);
4442 u32 aer_type = result & 0x07;
4444 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4448 case NVME_AER_NOTICE:
4449 nvme_handle_aen_notice(ctrl, result);
4451 case NVME_AER_ERROR:
4452 case NVME_AER_SMART:
4455 trace_nvme_async_event(ctrl, aer_type);
4456 ctrl->aen_result = result;
4461 queue_work(nvme_wq, &ctrl->async_event_work);
4463 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4465 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4467 nvme_mpath_stop(ctrl);
4468 nvme_stop_keep_alive(ctrl);
4469 flush_work(&ctrl->async_event_work);
4470 cancel_work_sync(&ctrl->fw_act_work);
4471 if (ctrl->ops->stop_ctrl)
4472 ctrl->ops->stop_ctrl(ctrl);
4474 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4476 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4478 nvme_start_keep_alive(ctrl);
4480 nvme_enable_aen(ctrl);
4482 if (ctrl->queue_count > 1) {
4483 nvme_queue_scan(ctrl);
4484 nvme_start_queues(ctrl);
4485 nvme_mpath_update(ctrl);
4488 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4490 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4492 nvme_hwmon_exit(ctrl);
4493 nvme_fault_inject_fini(&ctrl->fault_inject);
4494 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4495 cdev_device_del(&ctrl->cdev, ctrl->device);
4496 nvme_put_ctrl(ctrl);
4498 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4500 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4502 struct nvme_effects_log *cel;
4505 xa_for_each (&ctrl->cels, i, cel) {
4506 xa_erase(&ctrl->cels, i);
4510 xa_destroy(&ctrl->cels);
4513 static void nvme_free_ctrl(struct device *dev)
4515 struct nvme_ctrl *ctrl =
4516 container_of(dev, struct nvme_ctrl, ctrl_device);
4517 struct nvme_subsystem *subsys = ctrl->subsys;
4519 if (!subsys || ctrl->instance != subsys->instance)
4520 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4522 nvme_free_cels(ctrl);
4523 nvme_mpath_uninit(ctrl);
4524 __free_page(ctrl->discard_page);
4527 mutex_lock(&nvme_subsystems_lock);
4528 list_del(&ctrl->subsys_entry);
4529 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4530 mutex_unlock(&nvme_subsystems_lock);
4533 ctrl->ops->free_ctrl(ctrl);
4536 nvme_put_subsystem(subsys);
4540 * Initialize a NVMe controller structures. This needs to be called during
4541 * earliest initialization so that we have the initialized structured around
4544 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4545 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4549 ctrl->state = NVME_CTRL_NEW;
4550 spin_lock_init(&ctrl->lock);
4551 mutex_init(&ctrl->scan_lock);
4552 INIT_LIST_HEAD(&ctrl->namespaces);
4553 xa_init(&ctrl->cels);
4554 init_rwsem(&ctrl->namespaces_rwsem);
4557 ctrl->quirks = quirks;
4558 ctrl->numa_node = NUMA_NO_NODE;
4559 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4560 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4561 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4562 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4563 init_waitqueue_head(&ctrl->state_wq);
4565 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4566 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4567 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4569 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4571 ctrl->discard_page = alloc_page(GFP_KERNEL);
4572 if (!ctrl->discard_page) {
4577 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4580 ctrl->instance = ret;
4582 device_initialize(&ctrl->ctrl_device);
4583 ctrl->device = &ctrl->ctrl_device;
4584 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4585 ctrl->device->class = nvme_class;
4586 ctrl->device->parent = ctrl->dev;
4587 ctrl->device->groups = nvme_dev_attr_groups;
4588 ctrl->device->release = nvme_free_ctrl;
4589 dev_set_drvdata(ctrl->device, ctrl);
4590 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4592 goto out_release_instance;
4594 nvme_get_ctrl(ctrl);
4595 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4596 ctrl->cdev.owner = ops->module;
4597 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4602 * Initialize latency tolerance controls. The sysfs files won't
4603 * be visible to userspace unless the device actually supports APST.
4605 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4606 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4607 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4609 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4610 nvme_mpath_init_ctrl(ctrl);
4614 nvme_put_ctrl(ctrl);
4615 kfree_const(ctrl->device->kobj.name);
4616 out_release_instance:
4617 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4619 if (ctrl->discard_page)
4620 __free_page(ctrl->discard_page);
4623 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4626 * nvme_kill_queues(): Ends all namespace queues
4627 * @ctrl: the dead controller that needs to end
4629 * Call this function when the driver determines it is unable to get the
4630 * controller in a state capable of servicing IO.
4632 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4636 down_read(&ctrl->namespaces_rwsem);
4638 /* Forcibly unquiesce queues to avoid blocking dispatch */
4639 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4640 blk_mq_unquiesce_queue(ctrl->admin_q);
4642 list_for_each_entry(ns, &ctrl->namespaces, list)
4643 nvme_set_queue_dying(ns);
4645 up_read(&ctrl->namespaces_rwsem);
4647 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4649 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4653 down_read(&ctrl->namespaces_rwsem);
4654 list_for_each_entry(ns, &ctrl->namespaces, list)
4655 blk_mq_unfreeze_queue(ns->queue);
4656 up_read(&ctrl->namespaces_rwsem);
4658 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4660 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4664 down_read(&ctrl->namespaces_rwsem);
4665 list_for_each_entry(ns, &ctrl->namespaces, list) {
4666 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4670 up_read(&ctrl->namespaces_rwsem);
4673 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4675 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4679 down_read(&ctrl->namespaces_rwsem);
4680 list_for_each_entry(ns, &ctrl->namespaces, list)
4681 blk_mq_freeze_queue_wait(ns->queue);
4682 up_read(&ctrl->namespaces_rwsem);
4684 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4686 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4690 down_read(&ctrl->namespaces_rwsem);
4691 list_for_each_entry(ns, &ctrl->namespaces, list)
4692 blk_freeze_queue_start(ns->queue);
4693 up_read(&ctrl->namespaces_rwsem);
4695 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4697 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4701 down_read(&ctrl->namespaces_rwsem);
4702 list_for_each_entry(ns, &ctrl->namespaces, list)
4703 blk_mq_quiesce_queue(ns->queue);
4704 up_read(&ctrl->namespaces_rwsem);
4706 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4708 void nvme_start_queues(struct nvme_ctrl *ctrl)
4712 down_read(&ctrl->namespaces_rwsem);
4713 list_for_each_entry(ns, &ctrl->namespaces, list)
4714 blk_mq_unquiesce_queue(ns->queue);
4715 up_read(&ctrl->namespaces_rwsem);
4717 EXPORT_SYMBOL_GPL(nvme_start_queues);
4719 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4723 down_read(&ctrl->namespaces_rwsem);
4724 list_for_each_entry(ns, &ctrl->namespaces, list)
4725 blk_sync_queue(ns->queue);
4726 up_read(&ctrl->namespaces_rwsem);
4728 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4730 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4732 nvme_sync_io_queues(ctrl);
4734 blk_sync_queue(ctrl->admin_q);
4736 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4738 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4740 if (file->f_op != &nvme_dev_fops)
4742 return file->private_data;
4744 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4747 * Check we didn't inadvertently grow the command structure sizes:
4749 static inline void _nvme_check_size(void)
4751 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4752 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4753 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4754 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4755 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4756 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4757 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4758 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4759 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4760 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4761 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4762 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4763 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4764 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4765 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4766 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4767 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4768 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4769 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4773 static int __init nvme_core_init(void)
4775 int result = -ENOMEM;
4779 nvme_wq = alloc_workqueue("nvme-wq",
4780 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4784 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4785 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4789 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4790 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4791 if (!nvme_delete_wq)
4792 goto destroy_reset_wq;
4794 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4796 goto destroy_delete_wq;
4798 nvme_class = class_create(THIS_MODULE, "nvme");
4799 if (IS_ERR(nvme_class)) {
4800 result = PTR_ERR(nvme_class);
4801 goto unregister_chrdev;
4803 nvme_class->dev_uevent = nvme_class_uevent;
4805 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4806 if (IS_ERR(nvme_subsys_class)) {
4807 result = PTR_ERR(nvme_subsys_class);
4813 class_destroy(nvme_class);
4815 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4817 destroy_workqueue(nvme_delete_wq);
4819 destroy_workqueue(nvme_reset_wq);
4821 destroy_workqueue(nvme_wq);
4826 static void __exit nvme_core_exit(void)
4828 class_destroy(nvme_subsys_class);
4829 class_destroy(nvme_class);
4830 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4831 destroy_workqueue(nvme_delete_wq);
4832 destroy_workqueue(nvme_reset_wq);
4833 destroy_workqueue(nvme_wq);
4834 ida_destroy(&nvme_instance_ida);
4837 MODULE_LICENSE("GPL");
4838 MODULE_VERSION("1.0");
4839 module_init(nvme_core_init);
4840 module_exit(nvme_core_exit);