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 if (queue_max_discard_segments(req->q) == 1) {
727 u64 slba = nvme_sect_to_lba(ns, blk_rq_pos(req));
728 u32 nlb = blk_rq_sectors(req) >> (ns->lba_shift - 9);
730 range[0].cattr = cpu_to_le32(0);
731 range[0].nlb = cpu_to_le32(nlb);
732 range[0].slba = cpu_to_le64(slba);
735 __rq_for_each_bio(bio, req) {
736 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
737 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
740 range[n].cattr = cpu_to_le32(0);
741 range[n].nlb = cpu_to_le32(nlb);
742 range[n].slba = cpu_to_le64(slba);
748 if (WARN_ON_ONCE(n != segments)) {
749 if (virt_to_page(range) == ns->ctrl->discard_page)
750 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
753 return BLK_STS_IOERR;
756 cmnd->dsm.opcode = nvme_cmd_dsm;
757 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
758 cmnd->dsm.nr = cpu_to_le32(segments - 1);
759 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
761 req->special_vec.bv_page = virt_to_page(range);
762 req->special_vec.bv_offset = offset_in_page(range);
763 req->special_vec.bv_len = alloc_size;
764 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
769 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
770 struct request *req, struct nvme_command *cmnd)
772 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
773 return nvme_setup_discard(ns, req, cmnd);
775 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
776 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
777 cmnd->write_zeroes.slba =
778 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
779 cmnd->write_zeroes.length =
780 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
781 if (nvme_ns_has_pi(ns))
782 cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
784 cmnd->write_zeroes.control = 0;
788 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
789 struct request *req, struct nvme_command *cmnd,
792 struct nvme_ctrl *ctrl = ns->ctrl;
796 if (req->cmd_flags & REQ_FUA)
797 control |= NVME_RW_FUA;
798 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
799 control |= NVME_RW_LR;
801 if (req->cmd_flags & REQ_RAHEAD)
802 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
804 cmnd->rw.opcode = op;
805 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
806 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
807 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
809 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
810 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
814 * If formated with metadata, the block layer always provides a
815 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
816 * we enable the PRACT bit for protection information or set the
817 * namespace capacity to zero to prevent any I/O.
819 if (!blk_integrity_rq(req)) {
820 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
821 return BLK_STS_NOTSUPP;
822 control |= NVME_RW_PRINFO_PRACT;
825 switch (ns->pi_type) {
826 case NVME_NS_DPS_PI_TYPE3:
827 control |= NVME_RW_PRINFO_PRCHK_GUARD;
829 case NVME_NS_DPS_PI_TYPE1:
830 case NVME_NS_DPS_PI_TYPE2:
831 control |= NVME_RW_PRINFO_PRCHK_GUARD |
832 NVME_RW_PRINFO_PRCHK_REF;
833 if (op == nvme_cmd_zone_append)
834 control |= NVME_RW_APPEND_PIREMAP;
835 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
840 cmnd->rw.control = cpu_to_le16(control);
841 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
845 void nvme_cleanup_cmd(struct request *req)
847 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
848 struct nvme_ns *ns = req->rq_disk->private_data;
849 struct page *page = req->special_vec.bv_page;
851 if (page == ns->ctrl->discard_page)
852 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
854 kfree(page_address(page) + req->special_vec.bv_offset);
857 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
859 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
860 struct nvme_command *cmd)
862 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
863 blk_status_t ret = BLK_STS_OK;
865 if (!(req->rq_flags & RQF_DONTPREP))
866 nvme_clear_nvme_request(req);
868 memset(cmd, 0, sizeof(*cmd));
869 switch (req_op(req)) {
872 nvme_setup_passthrough(req, cmd);
875 nvme_setup_flush(ns, cmd);
877 case REQ_OP_ZONE_RESET_ALL:
878 case REQ_OP_ZONE_RESET:
879 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
881 case REQ_OP_ZONE_OPEN:
882 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
884 case REQ_OP_ZONE_CLOSE:
885 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
887 case REQ_OP_ZONE_FINISH:
888 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
890 case REQ_OP_WRITE_ZEROES:
891 ret = nvme_setup_write_zeroes(ns, req, cmd);
894 ret = nvme_setup_discard(ns, req, cmd);
897 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
900 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
902 case REQ_OP_ZONE_APPEND:
903 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
907 return BLK_STS_IOERR;
910 if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
911 nvme_req(req)->genctr++;
912 cmd->common.command_id = nvme_cid(req);
913 trace_nvme_setup_cmd(req, cmd);
916 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
918 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
920 struct completion *waiting = rq->end_io_data;
922 rq->end_io_data = NULL;
926 static void nvme_execute_rq_polled(struct request_queue *q,
927 struct gendisk *bd_disk, struct request *rq, int at_head)
929 DECLARE_COMPLETION_ONSTACK(wait);
931 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
933 rq->cmd_flags |= REQ_HIPRI;
934 rq->end_io_data = &wait;
935 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
937 while (!completion_done(&wait)) {
938 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
944 * Returns 0 on success. If the result is negative, it's a Linux error code;
945 * if the result is positive, it's an NVM Express status code
947 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
948 union nvme_result *result, void *buffer, unsigned bufflen,
949 unsigned timeout, int qid, int at_head,
950 blk_mq_req_flags_t flags, bool poll)
955 if (qid == NVME_QID_ANY)
956 req = nvme_alloc_request(q, cmd, flags);
958 req = nvme_alloc_request_qid(q, cmd, flags, qid);
963 req->timeout = timeout;
965 if (buffer && bufflen) {
966 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
972 nvme_execute_rq_polled(req->q, NULL, req, at_head);
974 blk_execute_rq(req->q, NULL, req, at_head);
976 *result = nvme_req(req)->result;
977 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
980 ret = nvme_req(req)->status;
982 blk_mq_free_request(req);
985 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
987 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
988 void *buffer, unsigned bufflen)
990 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
991 NVME_QID_ANY, 0, 0, false);
993 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
995 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
996 unsigned len, u32 seed, bool write)
998 struct bio_integrity_payload *bip;
1002 buf = kmalloc(len, GFP_KERNEL);
1007 if (write && copy_from_user(buf, ubuf, len))
1010 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
1016 bip->bip_iter.bi_size = len;
1017 bip->bip_iter.bi_sector = seed;
1018 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
1019 offset_in_page(buf));
1026 return ERR_PTR(ret);
1029 static u32 nvme_known_admin_effects(u8 opcode)
1032 case nvme_admin_format_nvm:
1033 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1034 NVME_CMD_EFFECTS_CSE_MASK;
1035 case nvme_admin_sanitize_nvm:
1036 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1043 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1048 if (ns->head->effects)
1049 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1050 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1051 dev_warn(ctrl->device,
1052 "IO command:%02x has unhandled effects:%08x\n",
1058 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1059 effects |= nvme_known_admin_effects(opcode);
1063 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1065 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1068 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1071 * For simplicity, IO to all namespaces is quiesced even if the command
1072 * effects say only one namespace is affected.
1074 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1075 mutex_lock(&ctrl->scan_lock);
1076 mutex_lock(&ctrl->subsys->lock);
1077 nvme_mpath_start_freeze(ctrl->subsys);
1078 nvme_mpath_wait_freeze(ctrl->subsys);
1079 nvme_start_freeze(ctrl);
1080 nvme_wait_freeze(ctrl);
1085 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1087 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1088 nvme_unfreeze(ctrl);
1089 nvme_mpath_unfreeze(ctrl->subsys);
1090 mutex_unlock(&ctrl->subsys->lock);
1091 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1092 mutex_unlock(&ctrl->scan_lock);
1094 if (effects & NVME_CMD_EFFECTS_CCC)
1095 nvme_init_identify(ctrl);
1096 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1097 nvme_queue_scan(ctrl);
1098 flush_work(&ctrl->scan_work);
1102 void nvme_execute_passthru_rq(struct request *rq)
1104 struct nvme_command *cmd = nvme_req(rq)->cmd;
1105 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1106 struct nvme_ns *ns = rq->q->queuedata;
1107 struct gendisk *disk = ns ? ns->disk : NULL;
1110 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1111 blk_execute_rq(rq->q, disk, rq, 0);
1112 nvme_passthru_end(ctrl, effects);
1114 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1116 static int nvme_submit_user_cmd(struct request_queue *q,
1117 struct nvme_command *cmd, void __user *ubuffer,
1118 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1119 u32 meta_seed, u64 *result, unsigned timeout)
1121 bool write = nvme_is_write(cmd);
1122 struct nvme_ns *ns = q->queuedata;
1123 struct gendisk *disk = ns ? ns->disk : NULL;
1124 struct request *req;
1125 struct bio *bio = NULL;
1129 req = nvme_alloc_request(q, cmd, 0);
1131 return PTR_ERR(req);
1134 req->timeout = timeout;
1135 nvme_req(req)->flags |= NVME_REQ_USERCMD;
1137 if (ubuffer && bufflen) {
1138 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1143 bio->bi_disk = disk;
1144 if (disk && meta_buffer && meta_len) {
1145 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1148 ret = PTR_ERR(meta);
1151 req->cmd_flags |= REQ_INTEGRITY;
1155 nvme_execute_passthru_rq(req);
1156 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1159 ret = nvme_req(req)->status;
1161 *result = le64_to_cpu(nvme_req(req)->result.u64);
1162 if (meta && !ret && !write) {
1163 if (copy_to_user(meta_buffer, meta, meta_len))
1169 blk_rq_unmap_user(bio);
1171 blk_mq_free_request(req);
1175 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1177 struct nvme_ctrl *ctrl = rq->end_io_data;
1178 unsigned long flags;
1179 bool startka = false;
1181 blk_mq_free_request(rq);
1184 dev_err(ctrl->device,
1185 "failed nvme_keep_alive_end_io error=%d\n",
1190 ctrl->comp_seen = false;
1191 spin_lock_irqsave(&ctrl->lock, flags);
1192 if (ctrl->state == NVME_CTRL_LIVE ||
1193 ctrl->state == NVME_CTRL_CONNECTING)
1195 spin_unlock_irqrestore(&ctrl->lock, flags);
1197 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1200 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1204 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1205 BLK_MQ_REQ_RESERVED);
1209 rq->timeout = ctrl->kato * HZ;
1210 rq->end_io_data = ctrl;
1212 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
1217 static void nvme_keep_alive_work(struct work_struct *work)
1219 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1220 struct nvme_ctrl, ka_work);
1221 bool comp_seen = ctrl->comp_seen;
1223 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1224 dev_dbg(ctrl->device,
1225 "reschedule traffic based keep-alive timer\n");
1226 ctrl->comp_seen = false;
1227 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1231 if (nvme_keep_alive(ctrl)) {
1232 /* allocation failure, reset the controller */
1233 dev_err(ctrl->device, "keep-alive failed\n");
1234 nvme_reset_ctrl(ctrl);
1239 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1241 if (unlikely(ctrl->kato == 0))
1244 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1247 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1249 if (unlikely(ctrl->kato == 0))
1252 cancel_delayed_work_sync(&ctrl->ka_work);
1254 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1257 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1258 * flag, thus sending any new CNS opcodes has a big chance of not working.
1259 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1260 * (but not for any later version).
1262 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1264 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1265 return ctrl->vs < NVME_VS(1, 2, 0);
1266 return ctrl->vs < NVME_VS(1, 1, 0);
1269 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1271 struct nvme_command c = { };
1274 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1275 c.identify.opcode = nvme_admin_identify;
1276 c.identify.cns = NVME_ID_CNS_CTRL;
1278 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1282 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1283 sizeof(struct nvme_id_ctrl));
1289 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1291 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1294 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1295 struct nvme_ns_id_desc *cur, bool *csi_seen)
1297 const char *warn_str = "ctrl returned bogus length:";
1300 switch (cur->nidt) {
1301 case NVME_NIDT_EUI64:
1302 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1303 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1304 warn_str, cur->nidl);
1307 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1308 return NVME_NIDT_EUI64_LEN;
1309 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1310 return NVME_NIDT_EUI64_LEN;
1311 case NVME_NIDT_NGUID:
1312 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1313 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1314 warn_str, cur->nidl);
1317 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1318 return NVME_NIDT_NGUID_LEN;
1319 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1320 return NVME_NIDT_NGUID_LEN;
1321 case NVME_NIDT_UUID:
1322 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1323 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1324 warn_str, cur->nidl);
1327 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1328 return NVME_NIDT_UUID_LEN;
1329 uuid_copy(&ids->uuid, data + sizeof(*cur));
1330 return NVME_NIDT_UUID_LEN;
1332 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1333 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1334 warn_str, cur->nidl);
1337 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1339 return NVME_NIDT_CSI_LEN;
1341 /* Skip unknown types */
1346 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1347 struct nvme_ns_ids *ids)
1349 struct nvme_command c = { };
1350 bool csi_seen = false;
1351 int status, pos, len;
1354 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1356 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1359 c.identify.opcode = nvme_admin_identify;
1360 c.identify.nsid = cpu_to_le32(nsid);
1361 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1363 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1367 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1368 NVME_IDENTIFY_DATA_SIZE);
1370 dev_warn(ctrl->device,
1371 "Identify Descriptors failed (%d)\n", status);
1375 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1376 struct nvme_ns_id_desc *cur = data + pos;
1381 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1385 len += sizeof(*cur);
1388 if (nvme_multi_css(ctrl) && !csi_seen) {
1389 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1399 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1400 struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1402 struct nvme_command c = { };
1405 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1406 c.identify.opcode = nvme_admin_identify;
1407 c.identify.nsid = cpu_to_le32(nsid);
1408 c.identify.cns = NVME_ID_CNS_NS;
1410 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1414 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1416 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1420 error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1421 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1425 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1426 dev_info(ctrl->device,
1427 "Ignoring bogus Namespace Identifiers\n");
1429 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1430 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1431 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1432 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1433 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1434 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1444 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1445 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1447 union nvme_result res = { 0 };
1448 struct nvme_command c;
1451 memset(&c, 0, sizeof(c));
1452 c.features.opcode = op;
1453 c.features.fid = cpu_to_le32(fid);
1454 c.features.dword11 = cpu_to_le32(dword11);
1456 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1457 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1458 if (ret >= 0 && result)
1459 *result = le32_to_cpu(res.u32);
1463 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1464 unsigned int dword11, void *buffer, size_t buflen,
1467 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1470 EXPORT_SYMBOL_GPL(nvme_set_features);
1472 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1473 unsigned int dword11, void *buffer, size_t buflen,
1476 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1479 EXPORT_SYMBOL_GPL(nvme_get_features);
1481 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1483 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1485 int status, nr_io_queues;
1487 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1493 * Degraded controllers might return an error when setting the queue
1494 * count. We still want to be able to bring them online and offer
1495 * access to the admin queue, as that might be only way to fix them up.
1498 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1501 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1502 *count = min(*count, nr_io_queues);
1507 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1509 #define NVME_AEN_SUPPORTED \
1510 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1511 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1513 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1515 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1518 if (!supported_aens)
1521 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1524 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1527 queue_work(nvme_wq, &ctrl->async_event_work);
1531 * Convert integer values from ioctl structures to user pointers, silently
1532 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1535 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1537 if (in_compat_syscall())
1538 ptrval = (compat_uptr_t)ptrval;
1539 return (void __user *)ptrval;
1542 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1544 struct nvme_user_io io;
1545 struct nvme_command c;
1546 unsigned length, meta_len;
1547 void __user *metadata;
1549 if (copy_from_user(&io, uio, sizeof(io)))
1554 switch (io.opcode) {
1555 case nvme_cmd_write:
1557 case nvme_cmd_compare:
1563 length = (io.nblocks + 1) << ns->lba_shift;
1565 if ((io.control & NVME_RW_PRINFO_PRACT) &&
1566 ns->ms == sizeof(struct t10_pi_tuple)) {
1568 * Protection information is stripped/inserted by the
1571 if (nvme_to_user_ptr(io.metadata))
1576 meta_len = (io.nblocks + 1) * ns->ms;
1577 metadata = nvme_to_user_ptr(io.metadata);
1580 if (ns->features & NVME_NS_EXT_LBAS) {
1583 } else if (meta_len) {
1584 if ((io.metadata & 3) || !io.metadata)
1588 memset(&c, 0, sizeof(c));
1589 c.rw.opcode = io.opcode;
1590 c.rw.flags = io.flags;
1591 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1592 c.rw.slba = cpu_to_le64(io.slba);
1593 c.rw.length = cpu_to_le16(io.nblocks);
1594 c.rw.control = cpu_to_le16(io.control);
1595 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1596 c.rw.reftag = cpu_to_le32(io.reftag);
1597 c.rw.apptag = cpu_to_le16(io.apptag);
1598 c.rw.appmask = cpu_to_le16(io.appmask);
1600 return nvme_submit_user_cmd(ns->queue, &c,
1601 nvme_to_user_ptr(io.addr), length,
1602 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1605 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1606 struct nvme_passthru_cmd __user *ucmd)
1608 struct nvme_passthru_cmd cmd;
1609 struct nvme_command c;
1610 unsigned timeout = 0;
1614 if (!capable(CAP_SYS_ADMIN))
1616 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1621 memset(&c, 0, sizeof(c));
1622 c.common.opcode = cmd.opcode;
1623 c.common.flags = cmd.flags;
1624 c.common.nsid = cpu_to_le32(cmd.nsid);
1625 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1626 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1627 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1628 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1629 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1630 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1631 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1632 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1635 timeout = msecs_to_jiffies(cmd.timeout_ms);
1637 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1638 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1639 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1640 0, &result, timeout);
1643 if (put_user(result, &ucmd->result))
1650 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1651 struct nvme_passthru_cmd64 __user *ucmd)
1653 struct nvme_passthru_cmd64 cmd;
1654 struct nvme_command c;
1655 unsigned timeout = 0;
1658 if (!capable(CAP_SYS_ADMIN))
1660 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1665 memset(&c, 0, sizeof(c));
1666 c.common.opcode = cmd.opcode;
1667 c.common.flags = cmd.flags;
1668 c.common.nsid = cpu_to_le32(cmd.nsid);
1669 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1670 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1671 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1672 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1673 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1674 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1675 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1676 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1679 timeout = msecs_to_jiffies(cmd.timeout_ms);
1681 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1682 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1683 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1684 0, &cmd.result, timeout);
1687 if (put_user(cmd.result, &ucmd->result))
1695 * Issue ioctl requests on the first available path. Note that unlike normal
1696 * block layer requests we will not retry failed request on another controller.
1698 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1699 struct nvme_ns_head **head, int *srcu_idx)
1701 #ifdef CONFIG_NVME_MULTIPATH
1702 if (disk->fops == &nvme_ns_head_ops) {
1705 *head = disk->private_data;
1706 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1707 ns = nvme_find_path(*head);
1709 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1715 return disk->private_data;
1718 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1721 srcu_read_unlock(&head->srcu, idx);
1724 static bool is_ctrl_ioctl(unsigned int cmd)
1726 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1728 if (is_sed_ioctl(cmd))
1733 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1735 struct nvme_ns_head *head,
1738 struct nvme_ctrl *ctrl = ns->ctrl;
1741 nvme_get_ctrl(ns->ctrl);
1742 nvme_put_ns_from_disk(head, srcu_idx);
1745 case NVME_IOCTL_ADMIN_CMD:
1746 ret = nvme_user_cmd(ctrl, NULL, argp);
1748 case NVME_IOCTL_ADMIN64_CMD:
1749 ret = nvme_user_cmd64(ctrl, NULL, argp);
1752 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1755 nvme_put_ctrl(ctrl);
1759 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1760 unsigned int cmd, unsigned long arg)
1762 struct nvme_ns_head *head = NULL;
1763 void __user *argp = (void __user *)arg;
1767 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1769 return -EWOULDBLOCK;
1772 * Handle ioctls that apply to the controller instead of the namespace
1773 * seperately and drop the ns SRCU reference early. This avoids a
1774 * deadlock when deleting namespaces using the passthrough interface.
1776 if (is_ctrl_ioctl(cmd))
1777 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1781 force_successful_syscall_return();
1782 ret = ns->head->ns_id;
1784 case NVME_IOCTL_IO_CMD:
1785 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1787 case NVME_IOCTL_SUBMIT_IO:
1788 ret = nvme_submit_io(ns, argp);
1790 case NVME_IOCTL_IO64_CMD:
1791 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1795 ret = nvme_nvm_ioctl(ns, cmd, arg);
1800 nvme_put_ns_from_disk(head, srcu_idx);
1804 #ifdef CONFIG_COMPAT
1805 struct nvme_user_io32 {
1818 } __attribute__((__packed__));
1820 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1822 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1823 unsigned int cmd, unsigned long arg)
1826 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1827 * between 32 bit programs and 64 bit kernel.
1828 * The cause is that the results of sizeof(struct nvme_user_io),
1829 * which is used to define NVME_IOCTL_SUBMIT_IO,
1830 * are not same between 32 bit compiler and 64 bit compiler.
1831 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1832 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1833 * Other IOCTL numbers are same between 32 bit and 64 bit.
1834 * So there is nothing to do regarding to other IOCTL numbers.
1836 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1837 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1839 return nvme_ioctl(bdev, mode, cmd, arg);
1842 #define nvme_compat_ioctl NULL
1843 #endif /* CONFIG_COMPAT */
1845 static int nvme_open(struct block_device *bdev, fmode_t mode)
1847 struct nvme_ns *ns = bdev->bd_disk->private_data;
1849 #ifdef CONFIG_NVME_MULTIPATH
1850 /* should never be called due to GENHD_FL_HIDDEN */
1851 if (WARN_ON_ONCE(ns->head->disk))
1854 if (!kref_get_unless_zero(&ns->kref))
1856 if (!try_module_get(ns->ctrl->ops->module))
1867 static void nvme_release(struct gendisk *disk, fmode_t mode)
1869 struct nvme_ns *ns = disk->private_data;
1871 module_put(ns->ctrl->ops->module);
1875 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1877 /* some standard values */
1878 geo->heads = 1 << 6;
1879 geo->sectors = 1 << 5;
1880 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1884 #ifdef CONFIG_BLK_DEV_INTEGRITY
1885 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1886 u32 max_integrity_segments)
1888 struct blk_integrity integrity;
1890 memset(&integrity, 0, sizeof(integrity));
1892 case NVME_NS_DPS_PI_TYPE3:
1893 integrity.profile = &t10_pi_type3_crc;
1894 integrity.tag_size = sizeof(u16) + sizeof(u32);
1895 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1897 case NVME_NS_DPS_PI_TYPE1:
1898 case NVME_NS_DPS_PI_TYPE2:
1899 integrity.profile = &t10_pi_type1_crc;
1900 integrity.tag_size = sizeof(u16);
1901 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1904 integrity.profile = NULL;
1907 integrity.tuple_size = ms;
1908 blk_integrity_register(disk, &integrity);
1909 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1912 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1913 u32 max_integrity_segments)
1916 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1918 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1920 struct nvme_ctrl *ctrl = ns->ctrl;
1921 struct request_queue *queue = disk->queue;
1922 u32 size = queue_logical_block_size(queue);
1924 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1925 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1929 if (ctrl->nr_streams && ns->sws && ns->sgs)
1930 size *= ns->sws * ns->sgs;
1932 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1933 NVME_DSM_MAX_RANGES);
1935 queue->limits.discard_alignment = 0;
1936 queue->limits.discard_granularity = size;
1938 /* If discard is already enabled, don't reset queue limits */
1939 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1942 blk_queue_max_discard_sectors(queue, UINT_MAX);
1943 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1945 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1946 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1950 * Even though NVMe spec explicitly states that MDTS is not applicable to the
1951 * write-zeroes, we are cautious and limit the size to the controllers
1952 * max_hw_sectors value, which is based on the MDTS field and possibly other
1955 static void nvme_config_write_zeroes(struct request_queue *q,
1956 struct nvme_ctrl *ctrl)
1958 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
1959 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1960 blk_queue_max_write_zeroes_sectors(q, ctrl->max_hw_sectors);
1963 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1965 return !uuid_is_null(&ids->uuid) ||
1966 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1967 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1970 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1972 return uuid_equal(&a->uuid, &b->uuid) &&
1973 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1974 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1978 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1979 u32 *phys_bs, u32 *io_opt)
1981 struct streams_directive_params s;
1984 if (!ctrl->nr_streams)
1987 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1991 ns->sws = le32_to_cpu(s.sws);
1992 ns->sgs = le16_to_cpu(s.sgs);
1995 *phys_bs = ns->sws * (1 << ns->lba_shift);
1997 *io_opt = *phys_bs * ns->sgs;
2003 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
2005 struct nvme_ctrl *ctrl = ns->ctrl;
2008 * The PI implementation requires the metadata size to be equal to the
2009 * t10 pi tuple size.
2011 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
2012 if (ns->ms == sizeof(struct t10_pi_tuple))
2013 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
2017 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2018 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
2020 if (ctrl->ops->flags & NVME_F_FABRICS) {
2022 * The NVMe over Fabrics specification only supports metadata as
2023 * part of the extended data LBA. We rely on HCA/HBA support to
2024 * remap the separate metadata buffer from the block layer.
2026 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
2028 if (ctrl->max_integrity_segments)
2030 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2033 * For PCIe controllers, we can't easily remap the separate
2034 * metadata buffer from the block layer and thus require a
2035 * separate metadata buffer for block layer metadata/PI support.
2036 * We allow extended LBAs for the passthrough interface, though.
2038 if (id->flbas & NVME_NS_FLBAS_META_EXT)
2039 ns->features |= NVME_NS_EXT_LBAS;
2041 ns->features |= NVME_NS_METADATA_SUPPORTED;
2047 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2048 struct request_queue *q)
2050 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2052 if (ctrl->max_hw_sectors) {
2054 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
2056 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2057 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2058 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2060 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
2061 blk_queue_dma_alignment(q, 3);
2062 blk_queue_write_cache(q, vwc, vwc);
2065 static void nvme_update_disk_info(struct gendisk *disk,
2066 struct nvme_ns *ns, struct nvme_id_ns *id)
2068 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
2069 unsigned short bs = 1 << ns->lba_shift;
2070 u32 atomic_bs, phys_bs, io_opt = 0;
2073 * The block layer can't support LBA sizes larger than the page size
2074 * or smaller than a sector size yet, so catch this early and don't
2077 if (ns->lba_shift > PAGE_SHIFT || ns->lba_shift < SECTOR_SHIFT) {
2082 blk_integrity_unregister(disk);
2084 atomic_bs = phys_bs = bs;
2085 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2086 if (id->nabo == 0) {
2088 * Bit 1 indicates whether NAWUPF is defined for this namespace
2089 * and whether it should be used instead of AWUPF. If NAWUPF ==
2090 * 0 then AWUPF must be used instead.
2092 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2093 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2095 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2098 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2099 /* NPWG = Namespace Preferred Write Granularity */
2100 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2101 /* NOWS = Namespace Optimal Write Size */
2102 io_opt = bs * (1 + le16_to_cpu(id->nows));
2105 blk_queue_logical_block_size(disk->queue, bs);
2107 * Linux filesystems assume writing a single physical block is
2108 * an atomic operation. Hence limit the physical block size to the
2109 * value of the Atomic Write Unit Power Fail parameter.
2111 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2112 blk_queue_io_min(disk->queue, phys_bs);
2113 blk_queue_io_opt(disk->queue, io_opt);
2116 * Register a metadata profile for PI, or the plain non-integrity NVMe
2117 * metadata masquerading as Type 0 if supported, otherwise reject block
2118 * I/O to namespaces with metadata except when the namespace supports
2119 * PI, as it can strip/insert in that case.
2122 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2123 (ns->features & NVME_NS_METADATA_SUPPORTED))
2124 nvme_init_integrity(disk, ns->ms, ns->pi_type,
2125 ns->ctrl->max_integrity_segments);
2126 else if (!nvme_ns_has_pi(ns))
2130 set_capacity_revalidate_and_notify(disk, capacity, false);
2132 nvme_config_discard(disk, ns);
2133 nvme_config_write_zeroes(disk->queue, ns->ctrl);
2135 if (id->nsattr & NVME_NS_ATTR_RO)
2136 set_disk_ro(disk, true);
2139 static inline bool nvme_first_scan(struct gendisk *disk)
2141 /* nvme_alloc_ns() scans the disk prior to adding it */
2142 return !(disk->flags & GENHD_FL_UP);
2145 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2147 struct nvme_ctrl *ctrl = ns->ctrl;
2150 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2151 is_power_of_2(ctrl->max_hw_sectors))
2152 iob = ctrl->max_hw_sectors;
2154 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2159 if (!is_power_of_2(iob)) {
2160 if (nvme_first_scan(ns->disk))
2161 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2162 ns->disk->disk_name, iob);
2166 if (blk_queue_is_zoned(ns->disk->queue)) {
2167 if (nvme_first_scan(ns->disk))
2168 pr_warn("%s: ignoring zoned namespace IO boundary\n",
2169 ns->disk->disk_name);
2173 blk_queue_chunk_sectors(ns->queue, iob);
2176 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
2178 unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2181 blk_mq_freeze_queue(ns->disk->queue);
2182 ns->lba_shift = id->lbaf[lbaf].ds;
2183 nvme_set_queue_limits(ns->ctrl, ns->queue);
2185 if (ns->head->ids.csi == NVME_CSI_ZNS) {
2186 ret = nvme_update_zone_info(ns, lbaf);
2191 ret = nvme_configure_metadata(ns, id);
2194 nvme_set_chunk_sectors(ns, id);
2195 nvme_update_disk_info(ns->disk, ns, id);
2196 blk_mq_unfreeze_queue(ns->disk->queue);
2198 if (blk_queue_is_zoned(ns->queue)) {
2199 ret = nvme_revalidate_zones(ns);
2200 if (ret && !nvme_first_scan(ns->disk))
2204 #ifdef CONFIG_NVME_MULTIPATH
2205 if (ns->head->disk) {
2206 blk_mq_freeze_queue(ns->head->disk->queue);
2207 nvme_update_disk_info(ns->head->disk, ns, id);
2208 blk_stack_limits(&ns->head->disk->queue->limits,
2209 &ns->queue->limits, 0);
2210 blk_queue_update_readahead(ns->head->disk->queue);
2211 nvme_update_bdev_size(ns->head->disk);
2212 blk_mq_unfreeze_queue(ns->head->disk->queue);
2218 blk_mq_unfreeze_queue(ns->disk->queue);
2222 static char nvme_pr_type(enum pr_type type)
2225 case PR_WRITE_EXCLUSIVE:
2227 case PR_EXCLUSIVE_ACCESS:
2229 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2231 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2233 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2235 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2242 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2243 u64 key, u64 sa_key, u8 op)
2245 struct nvme_ns_head *head = NULL;
2247 struct nvme_command c;
2249 u8 data[16] = { 0, };
2251 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2253 return -EWOULDBLOCK;
2255 put_unaligned_le64(key, &data[0]);
2256 put_unaligned_le64(sa_key, &data[8]);
2258 memset(&c, 0, sizeof(c));
2259 c.common.opcode = op;
2260 c.common.nsid = cpu_to_le32(ns->head->ns_id);
2261 c.common.cdw10 = cpu_to_le32(cdw10);
2263 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2264 nvme_put_ns_from_disk(head, srcu_idx);
2268 static int nvme_pr_register(struct block_device *bdev, u64 old,
2269 u64 new, unsigned flags)
2273 if (flags & ~PR_FL_IGNORE_KEY)
2276 cdw10 = old ? 2 : 0;
2277 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2278 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2279 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2282 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2283 enum pr_type type, unsigned flags)
2287 if (flags & ~PR_FL_IGNORE_KEY)
2290 cdw10 = nvme_pr_type(type) << 8;
2291 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2292 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2295 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2296 enum pr_type type, bool abort)
2298 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2300 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2303 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2305 u32 cdw10 = 1 | (key ? 0 : 1 << 3);
2307 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2310 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2312 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 0 : 1 << 3);
2314 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2317 static const struct pr_ops nvme_pr_ops = {
2318 .pr_register = nvme_pr_register,
2319 .pr_reserve = nvme_pr_reserve,
2320 .pr_release = nvme_pr_release,
2321 .pr_preempt = nvme_pr_preempt,
2322 .pr_clear = nvme_pr_clear,
2325 #ifdef CONFIG_BLK_SED_OPAL
2326 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2329 struct nvme_ctrl *ctrl = data;
2330 struct nvme_command cmd;
2332 memset(&cmd, 0, sizeof(cmd));
2334 cmd.common.opcode = nvme_admin_security_send;
2336 cmd.common.opcode = nvme_admin_security_recv;
2337 cmd.common.nsid = 0;
2338 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2339 cmd.common.cdw11 = cpu_to_le32(len);
2341 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2342 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2344 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2345 #endif /* CONFIG_BLK_SED_OPAL */
2347 static const struct block_device_operations nvme_fops = {
2348 .owner = THIS_MODULE,
2349 .ioctl = nvme_ioctl,
2350 .compat_ioctl = nvme_compat_ioctl,
2352 .release = nvme_release,
2353 .getgeo = nvme_getgeo,
2354 .report_zones = nvme_report_zones,
2355 .pr_ops = &nvme_pr_ops,
2358 #ifdef CONFIG_NVME_MULTIPATH
2359 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2361 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2363 if (!kref_get_unless_zero(&head->ref))
2368 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2370 nvme_put_ns_head(disk->private_data);
2373 const struct block_device_operations nvme_ns_head_ops = {
2374 .owner = THIS_MODULE,
2375 .submit_bio = nvme_ns_head_submit_bio,
2376 .open = nvme_ns_head_open,
2377 .release = nvme_ns_head_release,
2378 .ioctl = nvme_ioctl,
2379 .compat_ioctl = nvme_compat_ioctl,
2380 .getgeo = nvme_getgeo,
2381 .report_zones = nvme_report_zones,
2382 .pr_ops = &nvme_pr_ops,
2384 #endif /* CONFIG_NVME_MULTIPATH */
2386 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2388 unsigned long timeout =
2389 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2390 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2393 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2396 if ((csts & NVME_CSTS_RDY) == bit)
2399 usleep_range(1000, 2000);
2400 if (fatal_signal_pending(current))
2402 if (time_after(jiffies, timeout)) {
2403 dev_err(ctrl->device,
2404 "Device not ready; aborting %s, CSTS=0x%x\n",
2405 enabled ? "initialisation" : "reset", csts);
2414 * If the device has been passed off to us in an enabled state, just clear
2415 * the enabled bit. The spec says we should set the 'shutdown notification
2416 * bits', but doing so may cause the device to complete commands to the
2417 * admin queue ... and we don't know what memory that might be pointing at!
2419 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2423 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2424 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2426 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2430 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2431 msleep(NVME_QUIRK_DELAY_AMOUNT);
2433 return nvme_wait_ready(ctrl, ctrl->cap, false);
2435 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2437 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2439 unsigned dev_page_min;
2442 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2444 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2447 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2449 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2450 dev_err(ctrl->device,
2451 "Minimum device page size %u too large for host (%u)\n",
2452 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2456 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2457 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2459 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2460 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2461 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2462 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2463 ctrl->ctrl_config |= NVME_CC_ENABLE;
2465 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2468 return nvme_wait_ready(ctrl, ctrl->cap, true);
2470 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2472 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2474 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2478 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2479 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2481 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2485 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2486 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2490 if (fatal_signal_pending(current))
2492 if (time_after(jiffies, timeout)) {
2493 dev_err(ctrl->device,
2494 "Device shutdown incomplete; abort shutdown\n");
2501 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2503 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2508 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2511 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2512 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2515 dev_warn_once(ctrl->device,
2516 "could not set timestamp (%d)\n", ret);
2520 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2522 struct nvme_feat_host_behavior *host;
2525 /* Don't bother enabling the feature if retry delay is not reported */
2529 host = kzalloc(sizeof(*host), GFP_KERNEL);
2533 host->acre = NVME_ENABLE_ACRE;
2534 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2535 host, sizeof(*host), NULL);
2540 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2543 * APST (Autonomous Power State Transition) lets us program a
2544 * table of power state transitions that the controller will
2545 * perform automatically. We configure it with a simple
2546 * heuristic: we are willing to spend at most 2% of the time
2547 * transitioning between power states. Therefore, when running
2548 * in any given state, we will enter the next lower-power
2549 * non-operational state after waiting 50 * (enlat + exlat)
2550 * microseconds, as long as that state's exit latency is under
2551 * the requested maximum latency.
2553 * We will not autonomously enter any non-operational state for
2554 * which the total latency exceeds ps_max_latency_us. Users
2555 * can set ps_max_latency_us to zero to turn off APST.
2559 struct nvme_feat_auto_pst *table;
2565 * If APST isn't supported or if we haven't been initialized yet,
2566 * then don't do anything.
2571 if (ctrl->npss > 31) {
2572 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2576 table = kzalloc(sizeof(*table), GFP_KERNEL);
2580 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2581 /* Turn off APST. */
2583 dev_dbg(ctrl->device, "APST disabled\n");
2585 __le64 target = cpu_to_le64(0);
2589 * Walk through all states from lowest- to highest-power.
2590 * According to the spec, lower-numbered states use more
2591 * power. NPSS, despite the name, is the index of the
2592 * lowest-power state, not the number of states.
2594 for (state = (int)ctrl->npss; state >= 0; state--) {
2595 u64 total_latency_us, exit_latency_us, transition_ms;
2598 table->entries[state] = target;
2601 * Don't allow transitions to the deepest state
2602 * if it's quirked off.
2604 if (state == ctrl->npss &&
2605 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2609 * Is this state a useful non-operational state for
2610 * higher-power states to autonomously transition to?
2612 if (!(ctrl->psd[state].flags &
2613 NVME_PS_FLAGS_NON_OP_STATE))
2617 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2618 if (exit_latency_us > ctrl->ps_max_latency_us)
2623 le32_to_cpu(ctrl->psd[state].entry_lat);
2626 * This state is good. Use it as the APST idle
2627 * target for higher power states.
2629 transition_ms = total_latency_us + 19;
2630 do_div(transition_ms, 20);
2631 if (transition_ms > (1 << 24) - 1)
2632 transition_ms = (1 << 24) - 1;
2634 target = cpu_to_le64((state << 3) |
2635 (transition_ms << 8));
2640 if (total_latency_us > max_lat_us)
2641 max_lat_us = total_latency_us;
2647 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2649 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2650 max_ps, max_lat_us, (int)sizeof(*table), table);
2654 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2655 table, sizeof(*table), NULL);
2657 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2663 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2665 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2669 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2670 case PM_QOS_LATENCY_ANY:
2678 if (ctrl->ps_max_latency_us != latency) {
2679 ctrl->ps_max_latency_us = latency;
2680 if (ctrl->state == NVME_CTRL_LIVE)
2681 nvme_configure_apst(ctrl);
2685 struct nvme_core_quirk_entry {
2687 * NVMe model and firmware strings are padded with spaces. For
2688 * simplicity, strings in the quirk table are padded with NULLs
2694 unsigned long quirks;
2697 static const struct nvme_core_quirk_entry core_quirks[] = {
2700 * This Toshiba device seems to die using any APST states. See:
2701 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2704 .mn = "THNSF5256GPUK TOSHIBA",
2705 .quirks = NVME_QUIRK_NO_APST,
2709 * This LiteON CL1-3D*-Q11 firmware version has a race
2710 * condition associated with actions related to suspend to idle
2711 * LiteON has resolved the problem in future firmware
2715 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2719 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2720 * aborts I/O during any load, but more easily reproducible
2721 * with discards (fstrim).
2723 * The device is left in a state where it is also not possible
2724 * to use "nvme set-feature" to disable APST, but booting with
2725 * nvme_core.default_ps_max_latency=0 works.
2728 .mn = "KCD6XVUL6T40",
2729 .quirks = NVME_QUIRK_NO_APST,
2733 * The external Samsung X5 SSD fails initialization without a
2734 * delay before checking if it is ready and has a whole set of
2735 * other problems. To make this even more interesting, it
2736 * shares the PCI ID with internal Samsung 970 Evo Plus that
2737 * does not need or want these quirks.
2740 .mn = "Samsung Portable SSD X5",
2741 .quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2742 NVME_QUIRK_NO_DEEPEST_PS |
2743 NVME_QUIRK_IGNORE_DEV_SUBNQN,
2747 /* match is null-terminated but idstr is space-padded. */
2748 static bool string_matches(const char *idstr, const char *match, size_t len)
2755 matchlen = strlen(match);
2756 WARN_ON_ONCE(matchlen > len);
2758 if (memcmp(idstr, match, matchlen))
2761 for (; matchlen < len; matchlen++)
2762 if (idstr[matchlen] != ' ')
2768 static bool quirk_matches(const struct nvme_id_ctrl *id,
2769 const struct nvme_core_quirk_entry *q)
2771 return q->vid == le16_to_cpu(id->vid) &&
2772 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2773 string_matches(id->fr, q->fr, sizeof(id->fr));
2776 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2777 struct nvme_id_ctrl *id)
2782 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2783 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2784 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2785 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2789 if (ctrl->vs >= NVME_VS(1, 2, 1))
2790 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2793 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2794 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2795 "nqn.2014.08.org.nvmexpress:%04x%04x",
2796 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2797 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2798 off += sizeof(id->sn);
2799 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2800 off += sizeof(id->mn);
2801 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2804 static void nvme_release_subsystem(struct device *dev)
2806 struct nvme_subsystem *subsys =
2807 container_of(dev, struct nvme_subsystem, dev);
2809 if (subsys->instance >= 0)
2810 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2814 static void nvme_destroy_subsystem(struct kref *ref)
2816 struct nvme_subsystem *subsys =
2817 container_of(ref, struct nvme_subsystem, ref);
2819 mutex_lock(&nvme_subsystems_lock);
2820 list_del(&subsys->entry);
2821 mutex_unlock(&nvme_subsystems_lock);
2823 ida_destroy(&subsys->ns_ida);
2824 device_del(&subsys->dev);
2825 put_device(&subsys->dev);
2828 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2830 kref_put(&subsys->ref, nvme_destroy_subsystem);
2833 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2835 struct nvme_subsystem *subsys;
2837 lockdep_assert_held(&nvme_subsystems_lock);
2840 * Fail matches for discovery subsystems. This results
2841 * in each discovery controller bound to a unique subsystem.
2842 * This avoids issues with validating controller values
2843 * that can only be true when there is a single unique subsystem.
2844 * There may be multiple and completely independent entities
2845 * that provide discovery controllers.
2847 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2850 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2851 if (strcmp(subsys->subnqn, subsysnqn))
2853 if (!kref_get_unless_zero(&subsys->ref))
2861 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2862 struct device_attribute subsys_attr_##_name = \
2863 __ATTR(_name, _mode, _show, NULL)
2865 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2866 struct device_attribute *attr,
2869 struct nvme_subsystem *subsys =
2870 container_of(dev, struct nvme_subsystem, dev);
2872 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2874 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2876 #define nvme_subsys_show_str_function(field) \
2877 static ssize_t subsys_##field##_show(struct device *dev, \
2878 struct device_attribute *attr, char *buf) \
2880 struct nvme_subsystem *subsys = \
2881 container_of(dev, struct nvme_subsystem, dev); \
2882 return sysfs_emit(buf, "%.*s\n", \
2883 (int)sizeof(subsys->field), subsys->field); \
2885 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2887 nvme_subsys_show_str_function(model);
2888 nvme_subsys_show_str_function(serial);
2889 nvme_subsys_show_str_function(firmware_rev);
2891 static struct attribute *nvme_subsys_attrs[] = {
2892 &subsys_attr_model.attr,
2893 &subsys_attr_serial.attr,
2894 &subsys_attr_firmware_rev.attr,
2895 &subsys_attr_subsysnqn.attr,
2896 #ifdef CONFIG_NVME_MULTIPATH
2897 &subsys_attr_iopolicy.attr,
2902 static struct attribute_group nvme_subsys_attrs_group = {
2903 .attrs = nvme_subsys_attrs,
2906 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2907 &nvme_subsys_attrs_group,
2911 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2913 return ctrl->opts && ctrl->opts->discovery_nqn;
2916 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2917 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2919 struct nvme_ctrl *tmp;
2921 lockdep_assert_held(&nvme_subsystems_lock);
2923 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2924 if (nvme_state_terminal(tmp))
2927 if (tmp->cntlid == ctrl->cntlid) {
2928 dev_err(ctrl->device,
2929 "Duplicate cntlid %u with %s, rejecting\n",
2930 ctrl->cntlid, dev_name(tmp->device));
2934 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2935 nvme_discovery_ctrl(ctrl))
2938 dev_err(ctrl->device,
2939 "Subsystem does not support multiple controllers\n");
2946 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2948 struct nvme_subsystem *subsys, *found;
2951 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2955 subsys->instance = -1;
2956 mutex_init(&subsys->lock);
2957 kref_init(&subsys->ref);
2958 INIT_LIST_HEAD(&subsys->ctrls);
2959 INIT_LIST_HEAD(&subsys->nsheads);
2960 nvme_init_subnqn(subsys, ctrl, id);
2961 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2962 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2963 subsys->vendor_id = le16_to_cpu(id->vid);
2964 subsys->cmic = id->cmic;
2965 subsys->awupf = le16_to_cpu(id->awupf);
2966 #ifdef CONFIG_NVME_MULTIPATH
2967 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2970 subsys->dev.class = nvme_subsys_class;
2971 subsys->dev.release = nvme_release_subsystem;
2972 subsys->dev.groups = nvme_subsys_attrs_groups;
2973 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2974 device_initialize(&subsys->dev);
2976 mutex_lock(&nvme_subsystems_lock);
2977 found = __nvme_find_get_subsystem(subsys->subnqn);
2979 put_device(&subsys->dev);
2982 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2984 goto out_put_subsystem;
2987 ret = device_add(&subsys->dev);
2989 dev_err(ctrl->device,
2990 "failed to register subsystem device.\n");
2991 put_device(&subsys->dev);
2994 ida_init(&subsys->ns_ida);
2995 list_add_tail(&subsys->entry, &nvme_subsystems);
2998 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2999 dev_name(ctrl->device));
3001 dev_err(ctrl->device,
3002 "failed to create sysfs link from subsystem.\n");
3003 goto out_put_subsystem;
3007 subsys->instance = ctrl->instance;
3008 ctrl->subsys = subsys;
3009 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
3010 mutex_unlock(&nvme_subsystems_lock);
3014 nvme_put_subsystem(subsys);
3016 mutex_unlock(&nvme_subsystems_lock);
3020 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
3021 void *log, size_t size, u64 offset)
3023 struct nvme_command c = { };
3024 u32 dwlen = nvme_bytes_to_numd(size);
3026 c.get_log_page.opcode = nvme_admin_get_log_page;
3027 c.get_log_page.nsid = cpu_to_le32(nsid);
3028 c.get_log_page.lid = log_page;
3029 c.get_log_page.lsp = lsp;
3030 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3031 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3032 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3033 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3034 c.get_log_page.csi = csi;
3036 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3039 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3040 struct nvme_effects_log **log)
3042 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3048 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3052 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3053 cel, sizeof(*cel), 0);
3059 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3066 * Initialize the cached copies of the Identify data and various controller
3067 * register in our nvme_ctrl structure. This should be called as soon as
3068 * the admin queue is fully up and running.
3070 int nvme_init_identify(struct nvme_ctrl *ctrl)
3072 struct nvme_id_ctrl *id;
3073 int ret, page_shift;
3075 bool prev_apst_enabled;
3077 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3079 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3082 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
3083 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3085 if (ctrl->vs >= NVME_VS(1, 1, 0))
3086 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3088 ret = nvme_identify_ctrl(ctrl, &id);
3090 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3094 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3095 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3100 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3101 ctrl->cntlid = le16_to_cpu(id->cntlid);
3103 if (!ctrl->identified) {
3107 * Check for quirks. Quirk can depend on firmware version,
3108 * so, in principle, the set of quirks present can change
3109 * across a reset. As a possible future enhancement, we
3110 * could re-scan for quirks every time we reinitialize
3111 * the device, but we'd have to make sure that the driver
3112 * behaves intelligently if the quirks change.
3114 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3115 if (quirk_matches(id, &core_quirks[i]))
3116 ctrl->quirks |= core_quirks[i].quirks;
3119 ret = nvme_init_subsystem(ctrl, id);
3123 memcpy(ctrl->subsys->firmware_rev, id->fr,
3124 sizeof(ctrl->subsys->firmware_rev));
3126 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3127 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3128 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3131 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3132 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3133 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3135 ctrl->oacs = le16_to_cpu(id->oacs);
3136 ctrl->oncs = le16_to_cpu(id->oncs);
3137 ctrl->mtfa = le16_to_cpu(id->mtfa);
3138 ctrl->oaes = le32_to_cpu(id->oaes);
3139 ctrl->wctemp = le16_to_cpu(id->wctemp);
3140 ctrl->cctemp = le16_to_cpu(id->cctemp);
3142 atomic_set(&ctrl->abort_limit, id->acl + 1);
3143 ctrl->vwc = id->vwc;
3145 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
3147 max_hw_sectors = UINT_MAX;
3148 ctrl->max_hw_sectors =
3149 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3151 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3152 ctrl->sgls = le32_to_cpu(id->sgls);
3153 ctrl->kas = le16_to_cpu(id->kas);
3154 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3155 ctrl->ctratt = le32_to_cpu(id->ctratt);
3159 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3161 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3162 shutdown_timeout, 60);
3164 if (ctrl->shutdown_timeout != shutdown_timeout)
3165 dev_info(ctrl->device,
3166 "Shutdown timeout set to %u seconds\n",
3167 ctrl->shutdown_timeout);
3169 ctrl->shutdown_timeout = shutdown_timeout;
3171 ctrl->npss = id->npss;
3172 ctrl->apsta = id->apsta;
3173 prev_apst_enabled = ctrl->apst_enabled;
3174 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3175 if (force_apst && id->apsta) {
3176 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3177 ctrl->apst_enabled = true;
3179 ctrl->apst_enabled = false;
3182 ctrl->apst_enabled = id->apsta;
3184 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3186 if (ctrl->ops->flags & NVME_F_FABRICS) {
3187 ctrl->icdoff = le16_to_cpu(id->icdoff);
3188 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3189 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3190 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3193 * In fabrics we need to verify the cntlid matches the
3196 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3197 dev_err(ctrl->device,
3198 "Mismatching cntlid: Connect %u vs Identify "
3200 ctrl->cntlid, le16_to_cpu(id->cntlid));
3205 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3206 dev_err(ctrl->device,
3207 "keep-alive support is mandatory for fabrics\n");
3212 ctrl->hmpre = le32_to_cpu(id->hmpre);
3213 ctrl->hmmin = le32_to_cpu(id->hmmin);
3214 ctrl->hmminds = le32_to_cpu(id->hmminds);
3215 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3218 ret = nvme_mpath_init_identify(ctrl, id);
3224 if (ctrl->apst_enabled && !prev_apst_enabled)
3225 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3226 else if (!ctrl->apst_enabled && prev_apst_enabled)
3227 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3229 ret = nvme_configure_apst(ctrl);
3233 ret = nvme_configure_timestamp(ctrl);
3237 ret = nvme_configure_directives(ctrl);
3241 ret = nvme_configure_acre(ctrl);
3245 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3247 * Do not return errors unless we are in a controller reset,
3248 * the controller works perfectly fine without hwmon.
3250 ret = nvme_hwmon_init(ctrl);
3255 ctrl->identified = true;
3263 EXPORT_SYMBOL_GPL(nvme_init_identify);
3265 static int nvme_dev_open(struct inode *inode, struct file *file)
3267 struct nvme_ctrl *ctrl =
3268 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3270 switch (ctrl->state) {
3271 case NVME_CTRL_LIVE:
3274 return -EWOULDBLOCK;
3277 nvme_get_ctrl(ctrl);
3278 if (!try_module_get(ctrl->ops->module)) {
3279 nvme_put_ctrl(ctrl);
3283 file->private_data = ctrl;
3287 static int nvme_dev_release(struct inode *inode, struct file *file)
3289 struct nvme_ctrl *ctrl =
3290 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3292 module_put(ctrl->ops->module);
3293 nvme_put_ctrl(ctrl);
3297 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3302 down_read(&ctrl->namespaces_rwsem);
3303 if (list_empty(&ctrl->namespaces)) {
3308 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3309 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3310 dev_warn(ctrl->device,
3311 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3316 dev_warn(ctrl->device,
3317 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3318 kref_get(&ns->kref);
3319 up_read(&ctrl->namespaces_rwsem);
3321 ret = nvme_user_cmd(ctrl, ns, argp);
3326 up_read(&ctrl->namespaces_rwsem);
3330 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3333 struct nvme_ctrl *ctrl = file->private_data;
3334 void __user *argp = (void __user *)arg;
3337 case NVME_IOCTL_ADMIN_CMD:
3338 return nvme_user_cmd(ctrl, NULL, argp);
3339 case NVME_IOCTL_ADMIN64_CMD:
3340 return nvme_user_cmd64(ctrl, NULL, argp);
3341 case NVME_IOCTL_IO_CMD:
3342 return nvme_dev_user_cmd(ctrl, argp);
3343 case NVME_IOCTL_RESET:
3344 if (!capable(CAP_SYS_ADMIN))
3346 dev_warn(ctrl->device, "resetting controller\n");
3347 return nvme_reset_ctrl_sync(ctrl);
3348 case NVME_IOCTL_SUBSYS_RESET:
3349 if (!capable(CAP_SYS_ADMIN))
3351 return nvme_reset_subsystem(ctrl);
3352 case NVME_IOCTL_RESCAN:
3353 if (!capable(CAP_SYS_ADMIN))
3355 nvme_queue_scan(ctrl);
3362 static const struct file_operations nvme_dev_fops = {
3363 .owner = THIS_MODULE,
3364 .open = nvme_dev_open,
3365 .release = nvme_dev_release,
3366 .unlocked_ioctl = nvme_dev_ioctl,
3367 .compat_ioctl = compat_ptr_ioctl,
3370 static ssize_t nvme_sysfs_reset(struct device *dev,
3371 struct device_attribute *attr, const char *buf,
3374 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3377 ret = nvme_reset_ctrl_sync(ctrl);
3382 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3384 static ssize_t nvme_sysfs_rescan(struct device *dev,
3385 struct device_attribute *attr, const char *buf,
3388 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3390 nvme_queue_scan(ctrl);
3393 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3395 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3397 struct gendisk *disk = dev_to_disk(dev);
3399 if (disk->fops == &nvme_fops)
3400 return nvme_get_ns_from_dev(dev)->head;
3402 return disk->private_data;
3405 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3408 struct nvme_ns_head *head = dev_to_ns_head(dev);
3409 struct nvme_ns_ids *ids = &head->ids;
3410 struct nvme_subsystem *subsys = head->subsys;
3411 int serial_len = sizeof(subsys->serial);
3412 int model_len = sizeof(subsys->model);
3414 if (!uuid_is_null(&ids->uuid))
3415 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3417 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3418 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3420 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3421 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3423 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3424 subsys->serial[serial_len - 1] == '\0'))
3426 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3427 subsys->model[model_len - 1] == '\0'))
3430 return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3431 serial_len, subsys->serial, model_len, subsys->model,
3434 static DEVICE_ATTR_RO(wwid);
3436 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3439 return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3441 static DEVICE_ATTR_RO(nguid);
3443 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3446 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3448 /* For backward compatibility expose the NGUID to userspace if
3449 * we have no UUID set
3451 if (uuid_is_null(&ids->uuid)) {
3452 dev_warn_ratelimited(dev,
3453 "No UUID available providing old NGUID\n");
3454 return sysfs_emit(buf, "%pU\n", ids->nguid);
3456 return sysfs_emit(buf, "%pU\n", &ids->uuid);
3458 static DEVICE_ATTR_RO(uuid);
3460 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3463 return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3465 static DEVICE_ATTR_RO(eui);
3467 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3470 return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3472 static DEVICE_ATTR_RO(nsid);
3474 static struct attribute *nvme_ns_id_attrs[] = {
3475 &dev_attr_wwid.attr,
3476 &dev_attr_uuid.attr,
3477 &dev_attr_nguid.attr,
3479 &dev_attr_nsid.attr,
3480 #ifdef CONFIG_NVME_MULTIPATH
3481 &dev_attr_ana_grpid.attr,
3482 &dev_attr_ana_state.attr,
3487 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3488 struct attribute *a, int n)
3490 struct device *dev = container_of(kobj, struct device, kobj);
3491 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3493 if (a == &dev_attr_uuid.attr) {
3494 if (uuid_is_null(&ids->uuid) &&
3495 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3498 if (a == &dev_attr_nguid.attr) {
3499 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3502 if (a == &dev_attr_eui.attr) {
3503 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3506 #ifdef CONFIG_NVME_MULTIPATH
3507 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3508 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3510 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3517 static const struct attribute_group nvme_ns_id_attr_group = {
3518 .attrs = nvme_ns_id_attrs,
3519 .is_visible = nvme_ns_id_attrs_are_visible,
3522 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3523 &nvme_ns_id_attr_group,
3525 &nvme_nvm_attr_group,
3530 #define nvme_show_str_function(field) \
3531 static ssize_t field##_show(struct device *dev, \
3532 struct device_attribute *attr, char *buf) \
3534 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3535 return sysfs_emit(buf, "%.*s\n", \
3536 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3538 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3540 nvme_show_str_function(model);
3541 nvme_show_str_function(serial);
3542 nvme_show_str_function(firmware_rev);
3544 #define nvme_show_int_function(field) \
3545 static ssize_t field##_show(struct device *dev, \
3546 struct device_attribute *attr, char *buf) \
3548 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3549 return sysfs_emit(buf, "%d\n", ctrl->field); \
3551 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3553 nvme_show_int_function(cntlid);
3554 nvme_show_int_function(numa_node);
3555 nvme_show_int_function(queue_count);
3556 nvme_show_int_function(sqsize);
3558 static ssize_t nvme_sysfs_delete(struct device *dev,
3559 struct device_attribute *attr, const char *buf,
3562 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3564 if (device_remove_file_self(dev, attr))
3565 nvme_delete_ctrl_sync(ctrl);
3568 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3570 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3571 struct device_attribute *attr,
3574 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3576 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3578 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3580 static ssize_t nvme_sysfs_show_state(struct device *dev,
3581 struct device_attribute *attr,
3584 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3585 static const char *const state_name[] = {
3586 [NVME_CTRL_NEW] = "new",
3587 [NVME_CTRL_LIVE] = "live",
3588 [NVME_CTRL_RESETTING] = "resetting",
3589 [NVME_CTRL_CONNECTING] = "connecting",
3590 [NVME_CTRL_DELETING] = "deleting",
3591 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3592 [NVME_CTRL_DEAD] = "dead",
3595 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3596 state_name[ctrl->state])
3597 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3599 return sysfs_emit(buf, "unknown state\n");
3602 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3604 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3605 struct device_attribute *attr,
3608 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3610 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3612 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3614 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3615 struct device_attribute *attr,
3618 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3620 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3622 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3624 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3625 struct device_attribute *attr,
3628 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3630 return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3632 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3634 static ssize_t nvme_sysfs_show_address(struct device *dev,
3635 struct device_attribute *attr,
3638 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3640 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3642 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3644 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3645 struct device_attribute *attr, char *buf)
3647 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3648 struct nvmf_ctrl_options *opts = ctrl->opts;
3650 if (ctrl->opts->max_reconnects == -1)
3651 return sysfs_emit(buf, "off\n");
3652 return sysfs_emit(buf, "%d\n",
3653 opts->max_reconnects * opts->reconnect_delay);
3656 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3657 struct device_attribute *attr, const char *buf, size_t count)
3659 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3660 struct nvmf_ctrl_options *opts = ctrl->opts;
3661 int ctrl_loss_tmo, err;
3663 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3667 else if (ctrl_loss_tmo < 0)
3668 opts->max_reconnects = -1;
3670 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3671 opts->reconnect_delay);
3674 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3675 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3677 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3678 struct device_attribute *attr, char *buf)
3680 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3682 if (ctrl->opts->reconnect_delay == -1)
3683 return sysfs_emit(buf, "off\n");
3684 return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3687 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3688 struct device_attribute *attr, const char *buf, size_t count)
3690 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3694 err = kstrtou32(buf, 10, &v);
3698 ctrl->opts->reconnect_delay = v;
3701 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3702 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3704 static struct attribute *nvme_dev_attrs[] = {
3705 &dev_attr_reset_controller.attr,
3706 &dev_attr_rescan_controller.attr,
3707 &dev_attr_model.attr,
3708 &dev_attr_serial.attr,
3709 &dev_attr_firmware_rev.attr,
3710 &dev_attr_cntlid.attr,
3711 &dev_attr_delete_controller.attr,
3712 &dev_attr_transport.attr,
3713 &dev_attr_subsysnqn.attr,
3714 &dev_attr_address.attr,
3715 &dev_attr_state.attr,
3716 &dev_attr_numa_node.attr,
3717 &dev_attr_queue_count.attr,
3718 &dev_attr_sqsize.attr,
3719 &dev_attr_hostnqn.attr,
3720 &dev_attr_hostid.attr,
3721 &dev_attr_ctrl_loss_tmo.attr,
3722 &dev_attr_reconnect_delay.attr,
3726 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3727 struct attribute *a, int n)
3729 struct device *dev = container_of(kobj, struct device, kobj);
3730 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3732 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3734 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3736 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3738 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3740 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3742 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3748 static struct attribute_group nvme_dev_attrs_group = {
3749 .attrs = nvme_dev_attrs,
3750 .is_visible = nvme_dev_attrs_are_visible,
3753 static const struct attribute_group *nvme_dev_attr_groups[] = {
3754 &nvme_dev_attrs_group,
3758 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3761 struct nvme_ns_head *h;
3763 lockdep_assert_held(&subsys->lock);
3765 list_for_each_entry(h, &subsys->nsheads, entry) {
3766 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3773 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3774 struct nvme_ns_ids *ids)
3776 struct nvme_ns_head *h;
3778 lockdep_assert_held(&subsys->lock);
3780 list_for_each_entry(h, &subsys->nsheads, entry) {
3781 if (nvme_ns_ids_valid(ids) && nvme_ns_ids_equal(ids, &h->ids))
3788 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3789 unsigned nsid, struct nvme_ns_ids *ids)
3791 struct nvme_ns_head *head;
3792 size_t size = sizeof(*head);
3795 #ifdef CONFIG_NVME_MULTIPATH
3796 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3799 head = kzalloc(size, GFP_KERNEL);
3802 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3805 head->instance = ret;
3806 INIT_LIST_HEAD(&head->list);
3807 ret = init_srcu_struct(&head->srcu);
3809 goto out_ida_remove;
3810 head->subsys = ctrl->subsys;
3813 kref_init(&head->ref);
3815 ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &head->ids);
3817 dev_err(ctrl->device,
3818 "duplicate IDs for nsid %d\n", nsid);
3819 goto out_cleanup_srcu;
3822 if (head->ids.csi) {
3823 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3825 goto out_cleanup_srcu;
3827 head->effects = ctrl->effects;
3829 ret = nvme_mpath_alloc_disk(ctrl, head);
3831 goto out_cleanup_srcu;
3833 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3835 kref_get(&ctrl->subsys->ref);
3839 cleanup_srcu_struct(&head->srcu);
3841 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3846 ret = blk_status_to_errno(nvme_error_status(ret));
3847 return ERR_PTR(ret);
3850 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3851 struct nvme_ns_ids *ids, bool is_shared)
3853 struct nvme_ctrl *ctrl = ns->ctrl;
3854 struct nvme_ns_head *head = NULL;
3857 mutex_lock(&ctrl->subsys->lock);
3858 head = nvme_find_ns_head(ctrl->subsys, nsid);
3860 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3862 ret = PTR_ERR(head);
3865 head->shared = is_shared;
3868 if (!is_shared || !head->shared) {
3869 dev_err(ctrl->device,
3870 "Duplicate unshared namespace %d\n", nsid);
3871 goto out_put_ns_head;
3873 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3874 dev_err(ctrl->device,
3875 "IDs don't match for shared namespace %d\n",
3877 goto out_put_ns_head;
3881 list_add_tail(&ns->siblings, &head->list);
3883 mutex_unlock(&ctrl->subsys->lock);
3887 nvme_put_ns_head(head);
3889 mutex_unlock(&ctrl->subsys->lock);
3893 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3895 struct nvme_ns *ns, *ret = NULL;
3897 down_read(&ctrl->namespaces_rwsem);
3898 list_for_each_entry(ns, &ctrl->namespaces, list) {
3899 if (ns->head->ns_id == nsid) {
3900 if (!kref_get_unless_zero(&ns->kref))
3905 if (ns->head->ns_id > nsid)
3908 up_read(&ctrl->namespaces_rwsem);
3911 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3914 * Add the namespace to the controller list while keeping the list ordered.
3916 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3918 struct nvme_ns *tmp;
3920 list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3921 if (tmp->head->ns_id < ns->head->ns_id) {
3922 list_add(&ns->list, &tmp->list);
3926 list_add(&ns->list, &ns->ctrl->namespaces);
3929 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3930 struct nvme_ns_ids *ids)
3933 struct gendisk *disk;
3934 struct nvme_id_ns *id;
3935 char disk_name[DISK_NAME_LEN];
3936 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3938 if (nvme_identify_ns(ctrl, nsid, ids, &id))
3941 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3945 ns->queue = blk_mq_init_queue(ctrl->tagset);
3946 if (IS_ERR(ns->queue))
3949 if (ctrl->opts && ctrl->opts->data_digest)
3950 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3952 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3953 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3954 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3956 ns->queue->queuedata = ns;
3958 kref_init(&ns->kref);
3960 ret = nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED);
3962 goto out_free_queue;
3963 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3965 disk = alloc_disk_node(0, node);
3969 disk->fops = &nvme_fops;
3970 disk->private_data = ns;
3971 disk->queue = ns->queue;
3972 disk->flags = flags;
3973 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3976 if (nvme_update_ns_info(ns, id))
3979 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3980 ret = nvme_nvm_register(ns, disk_name, node);
3982 dev_warn(ctrl->device, "LightNVM init failure\n");
3987 down_write(&ctrl->namespaces_rwsem);
3988 nvme_ns_add_to_ctrl_list(ns);
3989 up_write(&ctrl->namespaces_rwsem);
3990 nvme_get_ctrl(ctrl);
3992 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3994 nvme_mpath_add_disk(ns, id);
3995 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
4000 /* prevent double queue cleanup */
4001 ns->disk->queue = NULL;
4004 mutex_lock(&ctrl->subsys->lock);
4005 list_del_rcu(&ns->siblings);
4006 if (list_empty(&ns->head->list))
4007 list_del_init(&ns->head->entry);
4008 mutex_unlock(&ctrl->subsys->lock);
4009 nvme_put_ns_head(ns->head);
4011 blk_cleanup_queue(ns->queue);
4018 static void nvme_ns_remove(struct nvme_ns *ns)
4020 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4023 set_capacity(ns->disk, 0);
4024 nvme_fault_inject_fini(&ns->fault_inject);
4026 mutex_lock(&ns->ctrl->subsys->lock);
4027 list_del_rcu(&ns->siblings);
4028 if (list_empty(&ns->head->list))
4029 list_del_init(&ns->head->entry);
4030 mutex_unlock(&ns->ctrl->subsys->lock);
4032 synchronize_rcu(); /* guarantee not available in head->list */
4033 nvme_mpath_clear_current_path(ns);
4034 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
4036 if (ns->disk->flags & GENHD_FL_UP) {
4037 del_gendisk(ns->disk);
4038 blk_cleanup_queue(ns->queue);
4039 if (blk_get_integrity(ns->disk))
4040 blk_integrity_unregister(ns->disk);
4043 down_write(&ns->ctrl->namespaces_rwsem);
4044 list_del_init(&ns->list);
4045 up_write(&ns->ctrl->namespaces_rwsem);
4047 nvme_mpath_check_last_path(ns);
4051 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4053 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4061 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4063 struct nvme_id_ns *id;
4064 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4066 if (test_bit(NVME_NS_DEAD, &ns->flags))
4069 ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4073 ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4074 if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4075 dev_err(ns->ctrl->device,
4076 "identifiers changed for nsid %d\n", ns->head->ns_id);
4080 ret = nvme_update_ns_info(ns, id);
4086 * Only remove the namespace if we got a fatal error back from the
4087 * device, otherwise ignore the error and just move on.
4089 * TODO: we should probably schedule a delayed retry here.
4091 if (ret > 0 && (ret & NVME_SC_DNR))
4094 revalidate_disk_size(ns->disk, true);
4097 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4099 struct nvme_ns_ids ids = { };
4102 if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4105 ns = nvme_find_get_ns(ctrl, nsid);
4107 nvme_validate_ns(ns, &ids);
4114 nvme_alloc_ns(ctrl, nsid, &ids);
4117 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4118 dev_warn(ctrl->device,
4119 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4123 if (!nvme_multi_css(ctrl)) {
4124 dev_warn(ctrl->device,
4125 "command set not reported for nsid: %d\n",
4129 nvme_alloc_ns(ctrl, nsid, &ids);
4132 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4138 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4141 struct nvme_ns *ns, *next;
4144 down_write(&ctrl->namespaces_rwsem);
4145 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4146 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4147 list_move_tail(&ns->list, &rm_list);
4149 up_write(&ctrl->namespaces_rwsem);
4151 list_for_each_entry_safe(ns, next, &rm_list, list)
4156 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4158 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4163 if (nvme_ctrl_limited_cns(ctrl))
4166 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4171 struct nvme_command cmd = {
4172 .identify.opcode = nvme_admin_identify,
4173 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4174 .identify.nsid = cpu_to_le32(prev),
4177 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4178 NVME_IDENTIFY_DATA_SIZE);
4182 for (i = 0; i < nr_entries; i++) {
4183 u32 nsid = le32_to_cpu(ns_list[i]);
4185 if (!nsid) /* end of the list? */
4187 nvme_validate_or_alloc_ns(ctrl, nsid);
4188 while (++prev < nsid)
4189 nvme_ns_remove_by_nsid(ctrl, prev);
4193 nvme_remove_invalid_namespaces(ctrl, prev);
4199 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4201 struct nvme_id_ctrl *id;
4204 if (nvme_identify_ctrl(ctrl, &id))
4206 nn = le32_to_cpu(id->nn);
4209 for (i = 1; i <= nn; i++)
4210 nvme_validate_or_alloc_ns(ctrl, i);
4212 nvme_remove_invalid_namespaces(ctrl, nn);
4215 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4217 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4221 log = kzalloc(log_size, GFP_KERNEL);
4226 * We need to read the log to clear the AEN, but we don't want to rely
4227 * on it for the changed namespace information as userspace could have
4228 * raced with us in reading the log page, which could cause us to miss
4231 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4232 NVME_CSI_NVM, log, log_size, 0);
4234 dev_warn(ctrl->device,
4235 "reading changed ns log failed: %d\n", error);
4240 static void nvme_scan_work(struct work_struct *work)
4242 struct nvme_ctrl *ctrl =
4243 container_of(work, struct nvme_ctrl, scan_work);
4245 /* No tagset on a live ctrl means IO queues could not created */
4246 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4249 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4250 dev_info(ctrl->device, "rescanning namespaces.\n");
4251 nvme_clear_changed_ns_log(ctrl);
4254 mutex_lock(&ctrl->scan_lock);
4255 if (nvme_scan_ns_list(ctrl) != 0)
4256 nvme_scan_ns_sequential(ctrl);
4257 mutex_unlock(&ctrl->scan_lock);
4261 * This function iterates the namespace list unlocked to allow recovery from
4262 * controller failure. It is up to the caller to ensure the namespace list is
4263 * not modified by scan work while this function is executing.
4265 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4267 struct nvme_ns *ns, *next;
4271 * make sure to requeue I/O to all namespaces as these
4272 * might result from the scan itself and must complete
4273 * for the scan_work to make progress
4275 nvme_mpath_clear_ctrl_paths(ctrl);
4277 /* prevent racing with ns scanning */
4278 flush_work(&ctrl->scan_work);
4281 * The dead states indicates the controller was not gracefully
4282 * disconnected. In that case, we won't be able to flush any data while
4283 * removing the namespaces' disks; fail all the queues now to avoid
4284 * potentially having to clean up the failed sync later.
4286 if (ctrl->state == NVME_CTRL_DEAD)
4287 nvme_kill_queues(ctrl);
4289 /* this is a no-op when called from the controller reset handler */
4290 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4292 down_write(&ctrl->namespaces_rwsem);
4293 list_splice_init(&ctrl->namespaces, &ns_list);
4294 up_write(&ctrl->namespaces_rwsem);
4296 list_for_each_entry_safe(ns, next, &ns_list, list)
4299 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4301 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4303 struct nvme_ctrl *ctrl =
4304 container_of(dev, struct nvme_ctrl, ctrl_device);
4305 struct nvmf_ctrl_options *opts = ctrl->opts;
4308 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4313 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4317 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4318 opts->trsvcid ?: "none");
4322 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4323 opts->host_traddr ?: "none");
4328 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4330 char *envp[2] = { NULL, NULL };
4331 u32 aen_result = ctrl->aen_result;
4333 ctrl->aen_result = 0;
4337 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4340 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4344 static void nvme_async_event_work(struct work_struct *work)
4346 struct nvme_ctrl *ctrl =
4347 container_of(work, struct nvme_ctrl, async_event_work);
4349 nvme_aen_uevent(ctrl);
4352 * The transport drivers must guarantee AER submission here is safe by
4353 * flushing ctrl async_event_work after changing the controller state
4354 * from LIVE and before freeing the admin queue.
4356 if (ctrl->state == NVME_CTRL_LIVE)
4357 ctrl->ops->submit_async_event(ctrl);
4360 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4365 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4371 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4374 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4376 struct nvme_fw_slot_info_log *log;
4378 log = kmalloc(sizeof(*log), GFP_KERNEL);
4382 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4383 log, sizeof(*log), 0))
4384 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4388 static void nvme_fw_act_work(struct work_struct *work)
4390 struct nvme_ctrl *ctrl = container_of(work,
4391 struct nvme_ctrl, fw_act_work);
4392 unsigned long fw_act_timeout;
4395 fw_act_timeout = jiffies +
4396 msecs_to_jiffies(ctrl->mtfa * 100);
4398 fw_act_timeout = jiffies +
4399 msecs_to_jiffies(admin_timeout * 1000);
4401 nvme_stop_queues(ctrl);
4402 while (nvme_ctrl_pp_status(ctrl)) {
4403 if (time_after(jiffies, fw_act_timeout)) {
4404 dev_warn(ctrl->device,
4405 "Fw activation timeout, reset controller\n");
4406 nvme_try_sched_reset(ctrl);
4412 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4415 nvme_start_queues(ctrl);
4416 /* read FW slot information to clear the AER */
4417 nvme_get_fw_slot_info(ctrl);
4420 static u32 nvme_aer_type(u32 result)
4422 return result & 0x7;
4425 static u32 nvme_aer_subtype(u32 result)
4427 return (result & 0xff00) >> 8;
4430 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4432 u32 aer_notice_type = nvme_aer_subtype(result);
4434 switch (aer_notice_type) {
4435 case NVME_AER_NOTICE_NS_CHANGED:
4436 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4437 nvme_queue_scan(ctrl);
4439 case NVME_AER_NOTICE_FW_ACT_STARTING:
4441 * We are (ab)using the RESETTING state to prevent subsequent
4442 * recovery actions from interfering with the controller's
4443 * firmware activation.
4445 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4446 queue_work(nvme_wq, &ctrl->fw_act_work);
4448 #ifdef CONFIG_NVME_MULTIPATH
4449 case NVME_AER_NOTICE_ANA:
4450 if (!ctrl->ana_log_buf)
4452 queue_work(nvme_wq, &ctrl->ana_work);
4455 case NVME_AER_NOTICE_DISC_CHANGED:
4456 ctrl->aen_result = result;
4459 dev_warn(ctrl->device, "async event result %08x\n", result);
4463 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4465 dev_warn(ctrl->device, "resetting controller due to AER\n");
4466 nvme_reset_ctrl(ctrl);
4469 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4470 volatile union nvme_result *res)
4472 u32 result = le32_to_cpu(res->u32);
4473 u32 aer_type = nvme_aer_type(result);
4474 u32 aer_subtype = nvme_aer_subtype(result);
4476 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4479 trace_nvme_async_event(ctrl, result);
4481 case NVME_AER_NOTICE:
4482 nvme_handle_aen_notice(ctrl, result);
4484 case NVME_AER_ERROR:
4486 * For a persistent internal error, don't run async_event_work
4487 * to submit a new AER. The controller reset will do it.
4489 if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4490 nvme_handle_aer_persistent_error(ctrl);
4494 case NVME_AER_SMART:
4497 ctrl->aen_result = result;
4502 queue_work(nvme_wq, &ctrl->async_event_work);
4504 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4506 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4508 nvme_mpath_stop(ctrl);
4509 nvme_stop_keep_alive(ctrl);
4510 flush_work(&ctrl->async_event_work);
4511 cancel_work_sync(&ctrl->fw_act_work);
4512 if (ctrl->ops->stop_ctrl)
4513 ctrl->ops->stop_ctrl(ctrl);
4515 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4517 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4519 nvme_start_keep_alive(ctrl);
4521 nvme_enable_aen(ctrl);
4523 if (ctrl->queue_count > 1) {
4524 nvme_queue_scan(ctrl);
4525 nvme_start_queues(ctrl);
4526 nvme_mpath_update(ctrl);
4529 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4531 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4533 nvme_hwmon_exit(ctrl);
4534 nvme_fault_inject_fini(&ctrl->fault_inject);
4535 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4536 cdev_device_del(&ctrl->cdev, ctrl->device);
4537 nvme_put_ctrl(ctrl);
4539 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4541 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4543 struct nvme_effects_log *cel;
4546 xa_for_each (&ctrl->cels, i, cel) {
4547 xa_erase(&ctrl->cels, i);
4551 xa_destroy(&ctrl->cels);
4554 static void nvme_free_ctrl(struct device *dev)
4556 struct nvme_ctrl *ctrl =
4557 container_of(dev, struct nvme_ctrl, ctrl_device);
4558 struct nvme_subsystem *subsys = ctrl->subsys;
4560 if (!subsys || ctrl->instance != subsys->instance)
4561 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4563 nvme_free_cels(ctrl);
4564 nvme_mpath_uninit(ctrl);
4565 __free_page(ctrl->discard_page);
4568 mutex_lock(&nvme_subsystems_lock);
4569 list_del(&ctrl->subsys_entry);
4570 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4571 mutex_unlock(&nvme_subsystems_lock);
4574 ctrl->ops->free_ctrl(ctrl);
4577 nvme_put_subsystem(subsys);
4581 * Initialize a NVMe controller structures. This needs to be called during
4582 * earliest initialization so that we have the initialized structured around
4585 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4586 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4590 ctrl->state = NVME_CTRL_NEW;
4591 spin_lock_init(&ctrl->lock);
4592 mutex_init(&ctrl->scan_lock);
4593 INIT_LIST_HEAD(&ctrl->namespaces);
4594 xa_init(&ctrl->cels);
4595 init_rwsem(&ctrl->namespaces_rwsem);
4598 ctrl->quirks = quirks;
4599 ctrl->numa_node = NUMA_NO_NODE;
4600 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4601 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4602 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4603 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4604 init_waitqueue_head(&ctrl->state_wq);
4606 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4607 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4608 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4610 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4612 ctrl->discard_page = alloc_page(GFP_KERNEL);
4613 if (!ctrl->discard_page) {
4618 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4621 ctrl->instance = ret;
4623 device_initialize(&ctrl->ctrl_device);
4624 ctrl->device = &ctrl->ctrl_device;
4625 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4626 ctrl->device->class = nvme_class;
4627 ctrl->device->parent = ctrl->dev;
4628 ctrl->device->groups = nvme_dev_attr_groups;
4629 ctrl->device->release = nvme_free_ctrl;
4630 dev_set_drvdata(ctrl->device, ctrl);
4631 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4633 goto out_release_instance;
4635 nvme_get_ctrl(ctrl);
4636 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4637 ctrl->cdev.owner = ops->module;
4638 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4643 * Initialize latency tolerance controls. The sysfs files won't
4644 * be visible to userspace unless the device actually supports APST.
4646 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4647 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4648 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4650 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4651 nvme_mpath_init_ctrl(ctrl);
4655 nvme_put_ctrl(ctrl);
4656 kfree_const(ctrl->device->kobj.name);
4657 out_release_instance:
4658 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4660 if (ctrl->discard_page)
4661 __free_page(ctrl->discard_page);
4664 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4667 * nvme_kill_queues(): Ends all namespace queues
4668 * @ctrl: the dead controller that needs to end
4670 * Call this function when the driver determines it is unable to get the
4671 * controller in a state capable of servicing IO.
4673 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4677 down_read(&ctrl->namespaces_rwsem);
4679 /* Forcibly unquiesce queues to avoid blocking dispatch */
4680 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4681 blk_mq_unquiesce_queue(ctrl->admin_q);
4683 list_for_each_entry(ns, &ctrl->namespaces, list)
4684 nvme_set_queue_dying(ns);
4686 up_read(&ctrl->namespaces_rwsem);
4688 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4690 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4694 down_read(&ctrl->namespaces_rwsem);
4695 list_for_each_entry(ns, &ctrl->namespaces, list)
4696 blk_mq_unfreeze_queue(ns->queue);
4697 up_read(&ctrl->namespaces_rwsem);
4699 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4701 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4705 down_read(&ctrl->namespaces_rwsem);
4706 list_for_each_entry(ns, &ctrl->namespaces, list) {
4707 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4711 up_read(&ctrl->namespaces_rwsem);
4714 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4716 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4720 down_read(&ctrl->namespaces_rwsem);
4721 list_for_each_entry(ns, &ctrl->namespaces, list)
4722 blk_mq_freeze_queue_wait(ns->queue);
4723 up_read(&ctrl->namespaces_rwsem);
4725 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4727 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4731 down_read(&ctrl->namespaces_rwsem);
4732 list_for_each_entry(ns, &ctrl->namespaces, list)
4733 blk_freeze_queue_start(ns->queue);
4734 up_read(&ctrl->namespaces_rwsem);
4736 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4738 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4742 down_read(&ctrl->namespaces_rwsem);
4743 list_for_each_entry(ns, &ctrl->namespaces, list)
4744 blk_mq_quiesce_queue(ns->queue);
4745 up_read(&ctrl->namespaces_rwsem);
4747 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4749 void nvme_start_queues(struct nvme_ctrl *ctrl)
4753 down_read(&ctrl->namespaces_rwsem);
4754 list_for_each_entry(ns, &ctrl->namespaces, list)
4755 blk_mq_unquiesce_queue(ns->queue);
4756 up_read(&ctrl->namespaces_rwsem);
4758 EXPORT_SYMBOL_GPL(nvme_start_queues);
4760 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4764 down_read(&ctrl->namespaces_rwsem);
4765 list_for_each_entry(ns, &ctrl->namespaces, list)
4766 blk_sync_queue(ns->queue);
4767 up_read(&ctrl->namespaces_rwsem);
4769 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4771 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4773 nvme_sync_io_queues(ctrl);
4775 blk_sync_queue(ctrl->admin_q);
4777 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4779 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4781 if (file->f_op != &nvme_dev_fops)
4783 return file->private_data;
4785 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4788 * Check we didn't inadvertently grow the command structure sizes:
4790 static inline void _nvme_check_size(void)
4792 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4793 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4794 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4795 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4796 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4797 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4798 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4799 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4800 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4801 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4802 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4803 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4804 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4805 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4806 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4807 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4808 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4809 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4810 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4814 static int __init nvme_core_init(void)
4816 int result = -ENOMEM;
4820 nvme_wq = alloc_workqueue("nvme-wq",
4821 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4825 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4826 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4830 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4831 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4832 if (!nvme_delete_wq)
4833 goto destroy_reset_wq;
4835 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4837 goto destroy_delete_wq;
4839 nvme_class = class_create(THIS_MODULE, "nvme");
4840 if (IS_ERR(nvme_class)) {
4841 result = PTR_ERR(nvme_class);
4842 goto unregister_chrdev;
4844 nvme_class->dev_uevent = nvme_class_uevent;
4846 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4847 if (IS_ERR(nvme_subsys_class)) {
4848 result = PTR_ERR(nvme_subsys_class);
4854 class_destroy(nvme_class);
4856 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4858 destroy_workqueue(nvme_delete_wq);
4860 destroy_workqueue(nvme_reset_wq);
4862 destroy_workqueue(nvme_wq);
4867 static void __exit nvme_core_exit(void)
4869 class_destroy(nvme_subsys_class);
4870 class_destroy(nvme_class);
4871 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4872 destroy_workqueue(nvme_delete_wq);
4873 destroy_workqueue(nvme_reset_wq);
4874 destroy_workqueue(nvme_wq);
4875 ida_destroy(&nvme_instance_ida);
4878 MODULE_LICENSE("GPL");
4879 MODULE_VERSION("1.0");
4880 module_init(nvme_core_init);
4881 module_exit(nvme_core_exit);