dm writecache: add cond_resched to loop in persistent_memory_claim()
[linux/fpc-iii.git] / drivers / nvme / host / core.c
blob7b4cbe2c695417e592d973ab1da2b506cb545089
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * NVM Express device driver
4 * Copyright (c) 2011-2014, Intel Corporation.
5 */
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/compat.h>
10 #include <linux/delay.h>
11 #include <linux/errno.h>
12 #include <linux/hdreg.h>
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/list_sort.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
19 #include <linux/pr.h>
20 #include <linux/ptrace.h>
21 #include <linux/nvme_ioctl.h>
22 #include <linux/t10-pi.h>
23 #include <linux/pm_qos.h>
24 #include <asm/unaligned.h>
26 #include "nvme.h"
27 #include "fabrics.h"
29 #define CREATE_TRACE_POINTS
30 #include "trace.h"
32 #define NVME_MINORS (1U << MINORBITS)
34 unsigned int admin_timeout = 60;
35 module_param(admin_timeout, uint, 0644);
36 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
37 EXPORT_SYMBOL_GPL(admin_timeout);
39 unsigned int nvme_io_timeout = 30;
40 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
41 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
42 EXPORT_SYMBOL_GPL(nvme_io_timeout);
44 static unsigned char shutdown_timeout = 5;
45 module_param(shutdown_timeout, byte, 0644);
46 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
48 static u8 nvme_max_retries = 5;
49 module_param_named(max_retries, nvme_max_retries, byte, 0644);
50 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
52 static unsigned long default_ps_max_latency_us = 100000;
53 module_param(default_ps_max_latency_us, ulong, 0644);
54 MODULE_PARM_DESC(default_ps_max_latency_us,
55 "max power saving latency for new devices; use PM QOS to change per device");
57 static bool force_apst;
58 module_param(force_apst, bool, 0644);
59 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
61 static bool streams;
62 module_param(streams, bool, 0644);
63 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
66 * nvme_wq - hosts nvme related works that are not reset or delete
67 * nvme_reset_wq - hosts nvme reset works
68 * nvme_delete_wq - hosts nvme delete works
70 * nvme_wq will host works such as scan, aen handling, fw activation,
71 * keep-alive, periodic reconnects etc. nvme_reset_wq
72 * runs reset works which also flush works hosted on nvme_wq for
73 * serialization purposes. nvme_delete_wq host controller deletion
74 * works which flush reset works for serialization.
76 struct workqueue_struct *nvme_wq;
77 EXPORT_SYMBOL_GPL(nvme_wq);
79 struct workqueue_struct *nvme_reset_wq;
80 EXPORT_SYMBOL_GPL(nvme_reset_wq);
82 struct workqueue_struct *nvme_delete_wq;
83 EXPORT_SYMBOL_GPL(nvme_delete_wq);
85 static LIST_HEAD(nvme_subsystems);
86 static DEFINE_MUTEX(nvme_subsystems_lock);
88 static DEFINE_IDA(nvme_instance_ida);
89 static dev_t nvme_chr_devt;
90 static struct class *nvme_class;
91 static struct class *nvme_subsys_class;
93 static int nvme_revalidate_disk(struct gendisk *disk);
94 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
95 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
96 unsigned nsid);
98 static void nvme_set_queue_dying(struct nvme_ns *ns)
101 * Revalidating a dead namespace sets capacity to 0. This will end
102 * buffered writers dirtying pages that can't be synced.
104 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
105 return;
106 blk_set_queue_dying(ns->queue);
107 /* Forcibly unquiesce queues to avoid blocking dispatch */
108 blk_mq_unquiesce_queue(ns->queue);
110 * Revalidate after unblocking dispatchers that may be holding bd_butex
112 revalidate_disk(ns->disk);
115 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
118 * Only new queue scan work when admin and IO queues are both alive
120 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
121 queue_work(nvme_wq, &ctrl->scan_work);
125 * Use this function to proceed with scheduling reset_work for a controller
126 * that had previously been set to the resetting state. This is intended for
127 * code paths that can't be interrupted by other reset attempts. A hot removal
128 * may prevent this from succeeding.
130 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
132 if (ctrl->state != NVME_CTRL_RESETTING)
133 return -EBUSY;
134 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
135 return -EBUSY;
136 return 0;
138 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
140 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
142 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
143 return -EBUSY;
144 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
145 return -EBUSY;
146 return 0;
148 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
150 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
152 int ret;
154 ret = nvme_reset_ctrl(ctrl);
155 if (!ret) {
156 flush_work(&ctrl->reset_work);
157 if (ctrl->state != NVME_CTRL_LIVE)
158 ret = -ENETRESET;
161 return ret;
163 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
165 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
167 dev_info(ctrl->device,
168 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
170 flush_work(&ctrl->reset_work);
171 nvme_stop_ctrl(ctrl);
172 nvme_remove_namespaces(ctrl);
173 ctrl->ops->delete_ctrl(ctrl);
174 nvme_uninit_ctrl(ctrl);
177 static void nvme_delete_ctrl_work(struct work_struct *work)
179 struct nvme_ctrl *ctrl =
180 container_of(work, struct nvme_ctrl, delete_work);
182 nvme_do_delete_ctrl(ctrl);
185 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
187 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
188 return -EBUSY;
189 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
190 return -EBUSY;
191 return 0;
193 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
195 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
198 * Keep a reference until nvme_do_delete_ctrl() complete,
199 * since ->delete_ctrl can free the controller.
201 nvme_get_ctrl(ctrl);
202 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
203 nvme_do_delete_ctrl(ctrl);
204 nvme_put_ctrl(ctrl);
207 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
209 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
212 static blk_status_t nvme_error_status(u16 status)
214 switch (status & 0x7ff) {
215 case NVME_SC_SUCCESS:
216 return BLK_STS_OK;
217 case NVME_SC_CAP_EXCEEDED:
218 return BLK_STS_NOSPC;
219 case NVME_SC_LBA_RANGE:
220 case NVME_SC_CMD_INTERRUPTED:
221 case NVME_SC_NS_NOT_READY:
222 return BLK_STS_TARGET;
223 case NVME_SC_BAD_ATTRIBUTES:
224 case NVME_SC_ONCS_NOT_SUPPORTED:
225 case NVME_SC_INVALID_OPCODE:
226 case NVME_SC_INVALID_FIELD:
227 case NVME_SC_INVALID_NS:
228 return BLK_STS_NOTSUPP;
229 case NVME_SC_WRITE_FAULT:
230 case NVME_SC_READ_ERROR:
231 case NVME_SC_UNWRITTEN_BLOCK:
232 case NVME_SC_ACCESS_DENIED:
233 case NVME_SC_READ_ONLY:
234 case NVME_SC_COMPARE_FAILED:
235 return BLK_STS_MEDIUM;
236 case NVME_SC_GUARD_CHECK:
237 case NVME_SC_APPTAG_CHECK:
238 case NVME_SC_REFTAG_CHECK:
239 case NVME_SC_INVALID_PI:
240 return BLK_STS_PROTECTION;
241 case NVME_SC_RESERVATION_CONFLICT:
242 return BLK_STS_NEXUS;
243 case NVME_SC_HOST_PATH_ERROR:
244 return BLK_STS_TRANSPORT;
245 default:
246 return BLK_STS_IOERR;
250 static inline bool nvme_req_needs_retry(struct request *req)
252 if (blk_noretry_request(req))
253 return false;
254 if (nvme_req(req)->status & NVME_SC_DNR)
255 return false;
256 if (nvme_req(req)->retries >= nvme_max_retries)
257 return false;
258 return true;
261 static void nvme_retry_req(struct request *req)
263 struct nvme_ns *ns = req->q->queuedata;
264 unsigned long delay = 0;
265 u16 crd;
267 /* The mask and shift result must be <= 3 */
268 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
269 if (ns && crd)
270 delay = ns->ctrl->crdt[crd - 1] * 100;
272 nvme_req(req)->retries++;
273 blk_mq_requeue_request(req, false);
274 blk_mq_delay_kick_requeue_list(req->q, delay);
277 void nvme_complete_rq(struct request *req)
279 blk_status_t status = nvme_error_status(nvme_req(req)->status);
281 trace_nvme_complete_rq(req);
283 nvme_cleanup_cmd(req);
285 if (nvme_req(req)->ctrl->kas)
286 nvme_req(req)->ctrl->comp_seen = true;
288 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
289 if ((req->cmd_flags & REQ_NVME_MPATH) && nvme_failover_req(req))
290 return;
292 if (!blk_queue_dying(req->q)) {
293 nvme_retry_req(req);
294 return;
298 nvme_trace_bio_complete(req, status);
299 blk_mq_end_request(req, status);
301 EXPORT_SYMBOL_GPL(nvme_complete_rq);
303 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
305 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
306 "Cancelling I/O %d", req->tag);
308 /* don't abort one completed request */
309 if (blk_mq_request_completed(req))
310 return true;
312 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
313 blk_mq_complete_request(req);
314 return true;
316 EXPORT_SYMBOL_GPL(nvme_cancel_request);
318 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
319 enum nvme_ctrl_state new_state)
321 enum nvme_ctrl_state old_state;
322 unsigned long flags;
323 bool changed = false;
325 spin_lock_irqsave(&ctrl->lock, flags);
327 old_state = ctrl->state;
328 switch (new_state) {
329 case NVME_CTRL_LIVE:
330 switch (old_state) {
331 case NVME_CTRL_NEW:
332 case NVME_CTRL_RESETTING:
333 case NVME_CTRL_CONNECTING:
334 changed = true;
335 /* FALLTHRU */
336 default:
337 break;
339 break;
340 case NVME_CTRL_RESETTING:
341 switch (old_state) {
342 case NVME_CTRL_NEW:
343 case NVME_CTRL_LIVE:
344 changed = true;
345 /* FALLTHRU */
346 default:
347 break;
349 break;
350 case NVME_CTRL_CONNECTING:
351 switch (old_state) {
352 case NVME_CTRL_NEW:
353 case NVME_CTRL_RESETTING:
354 changed = true;
355 /* FALLTHRU */
356 default:
357 break;
359 break;
360 case NVME_CTRL_DELETING:
361 switch (old_state) {
362 case NVME_CTRL_LIVE:
363 case NVME_CTRL_RESETTING:
364 case NVME_CTRL_CONNECTING:
365 changed = true;
366 /* FALLTHRU */
367 default:
368 break;
370 break;
371 case NVME_CTRL_DEAD:
372 switch (old_state) {
373 case NVME_CTRL_DELETING:
374 changed = true;
375 /* FALLTHRU */
376 default:
377 break;
379 break;
380 default:
381 break;
384 if (changed) {
385 ctrl->state = new_state;
386 wake_up_all(&ctrl->state_wq);
389 spin_unlock_irqrestore(&ctrl->lock, flags);
390 if (changed && ctrl->state == NVME_CTRL_LIVE)
391 nvme_kick_requeue_lists(ctrl);
392 return changed;
394 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
397 * Returns true for sink states that can't ever transition back to live.
399 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
401 switch (ctrl->state) {
402 case NVME_CTRL_NEW:
403 case NVME_CTRL_LIVE:
404 case NVME_CTRL_RESETTING:
405 case NVME_CTRL_CONNECTING:
406 return false;
407 case NVME_CTRL_DELETING:
408 case NVME_CTRL_DEAD:
409 return true;
410 default:
411 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
412 return true;
417 * Waits for the controller state to be resetting, or returns false if it is
418 * not possible to ever transition to that state.
420 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
422 wait_event(ctrl->state_wq,
423 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
424 nvme_state_terminal(ctrl));
425 return ctrl->state == NVME_CTRL_RESETTING;
427 EXPORT_SYMBOL_GPL(nvme_wait_reset);
429 static void nvme_free_ns_head(struct kref *ref)
431 struct nvme_ns_head *head =
432 container_of(ref, struct nvme_ns_head, ref);
434 nvme_mpath_remove_disk(head);
435 ida_simple_remove(&head->subsys->ns_ida, head->instance);
436 list_del_init(&head->entry);
437 cleanup_srcu_struct(&head->srcu);
438 nvme_put_subsystem(head->subsys);
439 kfree(head);
442 static void nvme_put_ns_head(struct nvme_ns_head *head)
444 kref_put(&head->ref, nvme_free_ns_head);
447 static void nvme_free_ns(struct kref *kref)
449 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
451 if (ns->ndev)
452 nvme_nvm_unregister(ns);
454 put_disk(ns->disk);
455 nvme_put_ns_head(ns->head);
456 nvme_put_ctrl(ns->ctrl);
457 kfree(ns);
460 static void nvme_put_ns(struct nvme_ns *ns)
462 kref_put(&ns->kref, nvme_free_ns);
465 static inline void nvme_clear_nvme_request(struct request *req)
467 if (!(req->rq_flags & RQF_DONTPREP)) {
468 nvme_req(req)->retries = 0;
469 nvme_req(req)->flags = 0;
470 req->rq_flags |= RQF_DONTPREP;
474 struct request *nvme_alloc_request(struct request_queue *q,
475 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
477 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
478 struct request *req;
480 if (qid == NVME_QID_ANY) {
481 req = blk_mq_alloc_request(q, op, flags);
482 } else {
483 req = blk_mq_alloc_request_hctx(q, op, flags,
484 qid ? qid - 1 : 0);
486 if (IS_ERR(req))
487 return req;
489 req->cmd_flags |= REQ_FAILFAST_DRIVER;
490 nvme_clear_nvme_request(req);
491 nvme_req(req)->cmd = cmd;
493 return req;
495 EXPORT_SYMBOL_GPL(nvme_alloc_request);
497 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
499 struct nvme_command c;
501 memset(&c, 0, sizeof(c));
503 c.directive.opcode = nvme_admin_directive_send;
504 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
505 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
506 c.directive.dtype = NVME_DIR_IDENTIFY;
507 c.directive.tdtype = NVME_DIR_STREAMS;
508 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
510 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
513 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
515 return nvme_toggle_streams(ctrl, false);
518 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
520 return nvme_toggle_streams(ctrl, true);
523 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
524 struct streams_directive_params *s, u32 nsid)
526 struct nvme_command c;
528 memset(&c, 0, sizeof(c));
529 memset(s, 0, sizeof(*s));
531 c.directive.opcode = nvme_admin_directive_recv;
532 c.directive.nsid = cpu_to_le32(nsid);
533 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
534 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
535 c.directive.dtype = NVME_DIR_STREAMS;
537 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
540 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
542 struct streams_directive_params s;
543 int ret;
545 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
546 return 0;
547 if (!streams)
548 return 0;
550 ret = nvme_enable_streams(ctrl);
551 if (ret)
552 return ret;
554 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
555 if (ret)
556 return ret;
558 ctrl->nssa = le16_to_cpu(s.nssa);
559 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
560 dev_info(ctrl->device, "too few streams (%u) available\n",
561 ctrl->nssa);
562 nvme_disable_streams(ctrl);
563 return 0;
566 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
567 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
568 return 0;
572 * Check if 'req' has a write hint associated with it. If it does, assign
573 * a valid namespace stream to the write.
575 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
576 struct request *req, u16 *control,
577 u32 *dsmgmt)
579 enum rw_hint streamid = req->write_hint;
581 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
582 streamid = 0;
583 else {
584 streamid--;
585 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
586 return;
588 *control |= NVME_RW_DTYPE_STREAMS;
589 *dsmgmt |= streamid << 16;
592 if (streamid < ARRAY_SIZE(req->q->write_hints))
593 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
596 static inline void nvme_setup_flush(struct nvme_ns *ns,
597 struct nvme_command *cmnd)
599 cmnd->common.opcode = nvme_cmd_flush;
600 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
603 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
604 struct nvme_command *cmnd)
606 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
607 struct nvme_dsm_range *range;
608 struct bio *bio;
611 * Some devices do not consider the DSM 'Number of Ranges' field when
612 * determining how much data to DMA. Always allocate memory for maximum
613 * number of segments to prevent device reading beyond end of buffer.
615 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
617 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
618 if (!range) {
620 * If we fail allocation our range, fallback to the controller
621 * discard page. If that's also busy, it's safe to return
622 * busy, as we know we can make progress once that's freed.
624 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
625 return BLK_STS_RESOURCE;
627 range = page_address(ns->ctrl->discard_page);
630 __rq_for_each_bio(bio, req) {
631 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
632 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
634 if (n < segments) {
635 range[n].cattr = cpu_to_le32(0);
636 range[n].nlb = cpu_to_le32(nlb);
637 range[n].slba = cpu_to_le64(slba);
639 n++;
642 if (WARN_ON_ONCE(n != segments)) {
643 if (virt_to_page(range) == ns->ctrl->discard_page)
644 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
645 else
646 kfree(range);
647 return BLK_STS_IOERR;
650 cmnd->dsm.opcode = nvme_cmd_dsm;
651 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
652 cmnd->dsm.nr = cpu_to_le32(segments - 1);
653 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
655 req->special_vec.bv_page = virt_to_page(range);
656 req->special_vec.bv_offset = offset_in_page(range);
657 req->special_vec.bv_len = alloc_size;
658 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
660 return BLK_STS_OK;
663 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
664 struct request *req, struct nvme_command *cmnd)
666 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
667 return nvme_setup_discard(ns, req, cmnd);
669 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
670 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
671 cmnd->write_zeroes.slba =
672 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
673 cmnd->write_zeroes.length =
674 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
675 cmnd->write_zeroes.control = 0;
676 return BLK_STS_OK;
679 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
680 struct request *req, struct nvme_command *cmnd)
682 struct nvme_ctrl *ctrl = ns->ctrl;
683 u16 control = 0;
684 u32 dsmgmt = 0;
686 if (req->cmd_flags & REQ_FUA)
687 control |= NVME_RW_FUA;
688 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
689 control |= NVME_RW_LR;
691 if (req->cmd_flags & REQ_RAHEAD)
692 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
694 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
695 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
696 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
697 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
699 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
700 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
702 if (ns->ms) {
704 * If formated with metadata, the block layer always provides a
705 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
706 * we enable the PRACT bit for protection information or set the
707 * namespace capacity to zero to prevent any I/O.
709 if (!blk_integrity_rq(req)) {
710 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
711 return BLK_STS_NOTSUPP;
712 control |= NVME_RW_PRINFO_PRACT;
715 switch (ns->pi_type) {
716 case NVME_NS_DPS_PI_TYPE3:
717 control |= NVME_RW_PRINFO_PRCHK_GUARD;
718 break;
719 case NVME_NS_DPS_PI_TYPE1:
720 case NVME_NS_DPS_PI_TYPE2:
721 control |= NVME_RW_PRINFO_PRCHK_GUARD |
722 NVME_RW_PRINFO_PRCHK_REF;
723 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
724 break;
728 cmnd->rw.control = cpu_to_le16(control);
729 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
730 return 0;
733 void nvme_cleanup_cmd(struct request *req)
735 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
736 struct nvme_ns *ns = req->rq_disk->private_data;
737 struct page *page = req->special_vec.bv_page;
739 if (page == ns->ctrl->discard_page)
740 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
741 else
742 kfree(page_address(page) + req->special_vec.bv_offset);
745 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
747 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
748 struct nvme_command *cmd)
750 blk_status_t ret = BLK_STS_OK;
752 nvme_clear_nvme_request(req);
754 memset(cmd, 0, sizeof(*cmd));
755 switch (req_op(req)) {
756 case REQ_OP_DRV_IN:
757 case REQ_OP_DRV_OUT:
758 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
759 break;
760 case REQ_OP_FLUSH:
761 nvme_setup_flush(ns, cmd);
762 break;
763 case REQ_OP_WRITE_ZEROES:
764 ret = nvme_setup_write_zeroes(ns, req, cmd);
765 break;
766 case REQ_OP_DISCARD:
767 ret = nvme_setup_discard(ns, req, cmd);
768 break;
769 case REQ_OP_READ:
770 case REQ_OP_WRITE:
771 ret = nvme_setup_rw(ns, req, cmd);
772 break;
773 default:
774 WARN_ON_ONCE(1);
775 return BLK_STS_IOERR;
778 cmd->common.command_id = req->tag;
779 trace_nvme_setup_cmd(req, cmd);
780 return ret;
782 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
784 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
786 struct completion *waiting = rq->end_io_data;
788 rq->end_io_data = NULL;
789 complete(waiting);
792 static void nvme_execute_rq_polled(struct request_queue *q,
793 struct gendisk *bd_disk, struct request *rq, int at_head)
795 DECLARE_COMPLETION_ONSTACK(wait);
797 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
799 rq->cmd_flags |= REQ_HIPRI;
800 rq->end_io_data = &wait;
801 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
803 while (!completion_done(&wait)) {
804 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
805 cond_resched();
810 * Returns 0 on success. If the result is negative, it's a Linux error code;
811 * if the result is positive, it's an NVM Express status code
813 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
814 union nvme_result *result, void *buffer, unsigned bufflen,
815 unsigned timeout, int qid, int at_head,
816 blk_mq_req_flags_t flags, bool poll)
818 struct request *req;
819 int ret;
821 req = nvme_alloc_request(q, cmd, flags, qid);
822 if (IS_ERR(req))
823 return PTR_ERR(req);
825 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
827 if (buffer && bufflen) {
828 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
829 if (ret)
830 goto out;
833 if (poll)
834 nvme_execute_rq_polled(req->q, NULL, req, at_head);
835 else
836 blk_execute_rq(req->q, NULL, req, at_head);
837 if (result)
838 *result = nvme_req(req)->result;
839 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
840 ret = -EINTR;
841 else
842 ret = nvme_req(req)->status;
843 out:
844 blk_mq_free_request(req);
845 return ret;
847 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
849 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
850 void *buffer, unsigned bufflen)
852 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
853 NVME_QID_ANY, 0, 0, false);
855 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
857 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
858 unsigned len, u32 seed, bool write)
860 struct bio_integrity_payload *bip;
861 int ret = -ENOMEM;
862 void *buf;
864 buf = kmalloc(len, GFP_KERNEL);
865 if (!buf)
866 goto out;
868 ret = -EFAULT;
869 if (write && copy_from_user(buf, ubuf, len))
870 goto out_free_meta;
872 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
873 if (IS_ERR(bip)) {
874 ret = PTR_ERR(bip);
875 goto out_free_meta;
878 bip->bip_iter.bi_size = len;
879 bip->bip_iter.bi_sector = seed;
880 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
881 offset_in_page(buf));
882 if (ret == len)
883 return buf;
884 ret = -ENOMEM;
885 out_free_meta:
886 kfree(buf);
887 out:
888 return ERR_PTR(ret);
891 static int nvme_submit_user_cmd(struct request_queue *q,
892 struct nvme_command *cmd, void __user *ubuffer,
893 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
894 u32 meta_seed, u64 *result, unsigned timeout)
896 bool write = nvme_is_write(cmd);
897 struct nvme_ns *ns = q->queuedata;
898 struct gendisk *disk = ns ? ns->disk : NULL;
899 struct request *req;
900 struct bio *bio = NULL;
901 void *meta = NULL;
902 int ret;
904 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
905 if (IS_ERR(req))
906 return PTR_ERR(req);
908 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
909 nvme_req(req)->flags |= NVME_REQ_USERCMD;
911 if (ubuffer && bufflen) {
912 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
913 GFP_KERNEL);
914 if (ret)
915 goto out;
916 bio = req->bio;
917 bio->bi_disk = disk;
918 if (disk && meta_buffer && meta_len) {
919 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
920 meta_seed, write);
921 if (IS_ERR(meta)) {
922 ret = PTR_ERR(meta);
923 goto out_unmap;
925 req->cmd_flags |= REQ_INTEGRITY;
929 blk_execute_rq(req->q, disk, req, 0);
930 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
931 ret = -EINTR;
932 else
933 ret = nvme_req(req)->status;
934 if (result)
935 *result = le64_to_cpu(nvme_req(req)->result.u64);
936 if (meta && !ret && !write) {
937 if (copy_to_user(meta_buffer, meta, meta_len))
938 ret = -EFAULT;
940 kfree(meta);
941 out_unmap:
942 if (bio)
943 blk_rq_unmap_user(bio);
944 out:
945 blk_mq_free_request(req);
946 return ret;
949 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
951 struct nvme_ctrl *ctrl = rq->end_io_data;
952 unsigned long flags;
953 bool startka = false;
955 blk_mq_free_request(rq);
957 if (status) {
958 dev_err(ctrl->device,
959 "failed nvme_keep_alive_end_io error=%d\n",
960 status);
961 return;
964 ctrl->comp_seen = false;
965 spin_lock_irqsave(&ctrl->lock, flags);
966 if (ctrl->state == NVME_CTRL_LIVE ||
967 ctrl->state == NVME_CTRL_CONNECTING)
968 startka = true;
969 spin_unlock_irqrestore(&ctrl->lock, flags);
970 if (startka)
971 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
974 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
976 struct request *rq;
978 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
979 NVME_QID_ANY);
980 if (IS_ERR(rq))
981 return PTR_ERR(rq);
983 rq->timeout = ctrl->kato * HZ;
984 rq->end_io_data = ctrl;
986 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
988 return 0;
991 static void nvme_keep_alive_work(struct work_struct *work)
993 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
994 struct nvme_ctrl, ka_work);
995 bool comp_seen = ctrl->comp_seen;
997 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
998 dev_dbg(ctrl->device,
999 "reschedule traffic based keep-alive timer\n");
1000 ctrl->comp_seen = false;
1001 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1002 return;
1005 if (nvme_keep_alive(ctrl)) {
1006 /* allocation failure, reset the controller */
1007 dev_err(ctrl->device, "keep-alive failed\n");
1008 nvme_reset_ctrl(ctrl);
1009 return;
1013 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1015 if (unlikely(ctrl->kato == 0))
1016 return;
1018 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1021 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1023 if (unlikely(ctrl->kato == 0))
1024 return;
1026 cancel_delayed_work_sync(&ctrl->ka_work);
1028 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1031 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1032 * flag, thus sending any new CNS opcodes has a big chance of not working.
1033 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1034 * (but not for any later version).
1036 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1038 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1039 return ctrl->vs < NVME_VS(1, 2, 0);
1040 return ctrl->vs < NVME_VS(1, 1, 0);
1043 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1045 struct nvme_command c = { };
1046 int error;
1048 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1049 c.identify.opcode = nvme_admin_identify;
1050 c.identify.cns = NVME_ID_CNS_CTRL;
1052 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1053 if (!*id)
1054 return -ENOMEM;
1056 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1057 sizeof(struct nvme_id_ctrl));
1058 if (error)
1059 kfree(*id);
1060 return error;
1063 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1064 struct nvme_ns_id_desc *cur)
1066 const char *warn_str = "ctrl returned bogus length:";
1067 void *data = cur;
1069 switch (cur->nidt) {
1070 case NVME_NIDT_EUI64:
1071 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1072 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1073 warn_str, cur->nidl);
1074 return -1;
1076 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1077 return NVME_NIDT_EUI64_LEN;
1078 case NVME_NIDT_NGUID:
1079 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1080 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1081 warn_str, cur->nidl);
1082 return -1;
1084 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1085 return NVME_NIDT_NGUID_LEN;
1086 case NVME_NIDT_UUID:
1087 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1088 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1089 warn_str, cur->nidl);
1090 return -1;
1092 uuid_copy(&ids->uuid, data + sizeof(*cur));
1093 return NVME_NIDT_UUID_LEN;
1094 default:
1095 /* Skip unknown types */
1096 return cur->nidl;
1100 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1101 struct nvme_ns_ids *ids)
1103 struct nvme_command c = { };
1104 int status;
1105 void *data;
1106 int pos;
1107 int len;
1109 c.identify.opcode = nvme_admin_identify;
1110 c.identify.nsid = cpu_to_le32(nsid);
1111 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1113 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1114 if (!data)
1115 return -ENOMEM;
1117 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1118 NVME_IDENTIFY_DATA_SIZE);
1119 if (status) {
1120 dev_warn(ctrl->device,
1121 "Identify Descriptors failed (%d)\n", status);
1123 * Don't treat an error as fatal, as we potentially already
1124 * have a NGUID or EUI-64.
1126 if (status > 0 && !(status & NVME_SC_DNR))
1127 status = 0;
1128 goto free_data;
1131 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1132 struct nvme_ns_id_desc *cur = data + pos;
1134 if (cur->nidl == 0)
1135 break;
1137 len = nvme_process_ns_desc(ctrl, ids, cur);
1138 if (len < 0)
1139 goto free_data;
1141 len += sizeof(*cur);
1143 free_data:
1144 kfree(data);
1145 return status;
1148 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1150 struct nvme_command c = { };
1152 c.identify.opcode = nvme_admin_identify;
1153 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1154 c.identify.nsid = cpu_to_le32(nsid);
1155 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1156 NVME_IDENTIFY_DATA_SIZE);
1159 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1160 unsigned nsid, struct nvme_id_ns **id)
1162 struct nvme_command c = { };
1163 int error;
1165 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1166 c.identify.opcode = nvme_admin_identify;
1167 c.identify.nsid = cpu_to_le32(nsid);
1168 c.identify.cns = NVME_ID_CNS_NS;
1170 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1171 if (!*id)
1172 return -ENOMEM;
1174 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1175 if (error) {
1176 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1177 kfree(*id);
1180 return error;
1183 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1184 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1186 union nvme_result res = { 0 };
1187 struct nvme_command c;
1188 int ret;
1190 memset(&c, 0, sizeof(c));
1191 c.features.opcode = op;
1192 c.features.fid = cpu_to_le32(fid);
1193 c.features.dword11 = cpu_to_le32(dword11);
1195 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1196 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1197 if (ret >= 0 && result)
1198 *result = le32_to_cpu(res.u32);
1199 return ret;
1202 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1203 unsigned int dword11, void *buffer, size_t buflen,
1204 u32 *result)
1206 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1207 buflen, result);
1209 EXPORT_SYMBOL_GPL(nvme_set_features);
1211 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1212 unsigned int dword11, void *buffer, size_t buflen,
1213 u32 *result)
1215 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1216 buflen, result);
1218 EXPORT_SYMBOL_GPL(nvme_get_features);
1220 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1222 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1223 u32 result;
1224 int status, nr_io_queues;
1226 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1227 &result);
1228 if (status < 0)
1229 return status;
1232 * Degraded controllers might return an error when setting the queue
1233 * count. We still want to be able to bring them online and offer
1234 * access to the admin queue, as that might be only way to fix them up.
1236 if (status > 0) {
1237 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1238 *count = 0;
1239 } else {
1240 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1241 *count = min(*count, nr_io_queues);
1244 return 0;
1246 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1248 #define NVME_AEN_SUPPORTED \
1249 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1250 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1252 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1254 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1255 int status;
1257 if (!supported_aens)
1258 return;
1260 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1261 NULL, 0, &result);
1262 if (status)
1263 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1264 supported_aens);
1266 queue_work(nvme_wq, &ctrl->async_event_work);
1270 * Convert integer values from ioctl structures to user pointers, silently
1271 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1272 * kernels.
1274 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1276 if (in_compat_syscall())
1277 ptrval = (compat_uptr_t)ptrval;
1278 return (void __user *)ptrval;
1281 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1283 struct nvme_user_io io;
1284 struct nvme_command c;
1285 unsigned length, meta_len;
1286 void __user *metadata;
1288 if (copy_from_user(&io, uio, sizeof(io)))
1289 return -EFAULT;
1290 if (io.flags)
1291 return -EINVAL;
1293 switch (io.opcode) {
1294 case nvme_cmd_write:
1295 case nvme_cmd_read:
1296 case nvme_cmd_compare:
1297 break;
1298 default:
1299 return -EINVAL;
1302 length = (io.nblocks + 1) << ns->lba_shift;
1303 meta_len = (io.nblocks + 1) * ns->ms;
1304 metadata = nvme_to_user_ptr(io.metadata);
1306 if (ns->ext) {
1307 length += meta_len;
1308 meta_len = 0;
1309 } else if (meta_len) {
1310 if ((io.metadata & 3) || !io.metadata)
1311 return -EINVAL;
1314 memset(&c, 0, sizeof(c));
1315 c.rw.opcode = io.opcode;
1316 c.rw.flags = io.flags;
1317 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1318 c.rw.slba = cpu_to_le64(io.slba);
1319 c.rw.length = cpu_to_le16(io.nblocks);
1320 c.rw.control = cpu_to_le16(io.control);
1321 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1322 c.rw.reftag = cpu_to_le32(io.reftag);
1323 c.rw.apptag = cpu_to_le16(io.apptag);
1324 c.rw.appmask = cpu_to_le16(io.appmask);
1326 return nvme_submit_user_cmd(ns->queue, &c,
1327 nvme_to_user_ptr(io.addr), length,
1328 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1331 static u32 nvme_known_admin_effects(u8 opcode)
1333 switch (opcode) {
1334 case nvme_admin_format_nvm:
1335 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1336 NVME_CMD_EFFECTS_CSE_MASK;
1337 case nvme_admin_sanitize_nvm:
1338 return NVME_CMD_EFFECTS_CSE_MASK;
1339 default:
1340 break;
1342 return 0;
1345 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1346 u8 opcode)
1348 u32 effects = 0;
1350 if (ns) {
1351 if (ctrl->effects)
1352 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1353 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1354 dev_warn(ctrl->device,
1355 "IO command:%02x has unhandled effects:%08x\n",
1356 opcode, effects);
1357 return 0;
1360 if (ctrl->effects)
1361 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1362 effects |= nvme_known_admin_effects(opcode);
1365 * For simplicity, IO to all namespaces is quiesced even if the command
1366 * effects say only one namespace is affected.
1368 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1369 mutex_lock(&ctrl->scan_lock);
1370 mutex_lock(&ctrl->subsys->lock);
1371 nvme_mpath_start_freeze(ctrl->subsys);
1372 nvme_mpath_wait_freeze(ctrl->subsys);
1373 nvme_start_freeze(ctrl);
1374 nvme_wait_freeze(ctrl);
1376 return effects;
1379 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1381 struct nvme_ns *ns;
1383 down_read(&ctrl->namespaces_rwsem);
1384 list_for_each_entry(ns, &ctrl->namespaces, list)
1385 if (ns->disk && nvme_revalidate_disk(ns->disk))
1386 nvme_set_queue_dying(ns);
1387 up_read(&ctrl->namespaces_rwsem);
1390 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1393 * Revalidate LBA changes prior to unfreezing. This is necessary to
1394 * prevent memory corruption if a logical block size was changed by
1395 * this command.
1397 if (effects & NVME_CMD_EFFECTS_LBCC)
1398 nvme_update_formats(ctrl);
1399 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1400 nvme_unfreeze(ctrl);
1401 nvme_mpath_unfreeze(ctrl->subsys);
1402 mutex_unlock(&ctrl->subsys->lock);
1403 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1404 mutex_unlock(&ctrl->scan_lock);
1406 if (effects & NVME_CMD_EFFECTS_CCC)
1407 nvme_init_identify(ctrl);
1408 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1409 nvme_queue_scan(ctrl);
1412 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1413 struct nvme_passthru_cmd __user *ucmd)
1415 struct nvme_passthru_cmd cmd;
1416 struct nvme_command c;
1417 unsigned timeout = 0;
1418 u32 effects;
1419 u64 result;
1420 int status;
1422 if (!capable(CAP_SYS_ADMIN))
1423 return -EACCES;
1424 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1425 return -EFAULT;
1426 if (cmd.flags)
1427 return -EINVAL;
1429 memset(&c, 0, sizeof(c));
1430 c.common.opcode = cmd.opcode;
1431 c.common.flags = cmd.flags;
1432 c.common.nsid = cpu_to_le32(cmd.nsid);
1433 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1434 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1435 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1436 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1437 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1438 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1439 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1440 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1442 if (cmd.timeout_ms)
1443 timeout = msecs_to_jiffies(cmd.timeout_ms);
1445 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1446 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1447 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1448 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1449 0, &result, timeout);
1450 nvme_passthru_end(ctrl, effects);
1452 if (status >= 0) {
1453 if (put_user(result, &ucmd->result))
1454 return -EFAULT;
1457 return status;
1460 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1461 struct nvme_passthru_cmd64 __user *ucmd)
1463 struct nvme_passthru_cmd64 cmd;
1464 struct nvme_command c;
1465 unsigned timeout = 0;
1466 u32 effects;
1467 int status;
1469 if (!capable(CAP_SYS_ADMIN))
1470 return -EACCES;
1471 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1472 return -EFAULT;
1473 if (cmd.flags)
1474 return -EINVAL;
1476 memset(&c, 0, sizeof(c));
1477 c.common.opcode = cmd.opcode;
1478 c.common.flags = cmd.flags;
1479 c.common.nsid = cpu_to_le32(cmd.nsid);
1480 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1481 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1482 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1483 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1484 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1485 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1486 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1487 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1489 if (cmd.timeout_ms)
1490 timeout = msecs_to_jiffies(cmd.timeout_ms);
1492 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1493 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1494 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1495 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1496 0, &cmd.result, timeout);
1497 nvme_passthru_end(ctrl, effects);
1499 if (status >= 0) {
1500 if (put_user(cmd.result, &ucmd->result))
1501 return -EFAULT;
1504 return status;
1508 * Issue ioctl requests on the first available path. Note that unlike normal
1509 * block layer requests we will not retry failed request on another controller.
1511 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1512 struct nvme_ns_head **head, int *srcu_idx)
1514 #ifdef CONFIG_NVME_MULTIPATH
1515 if (disk->fops == &nvme_ns_head_ops) {
1516 struct nvme_ns *ns;
1518 *head = disk->private_data;
1519 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1520 ns = nvme_find_path(*head);
1521 if (!ns)
1522 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1523 return ns;
1525 #endif
1526 *head = NULL;
1527 *srcu_idx = -1;
1528 return disk->private_data;
1531 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1533 if (head)
1534 srcu_read_unlock(&head->srcu, idx);
1537 static bool is_ctrl_ioctl(unsigned int cmd)
1539 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1540 return true;
1541 if (is_sed_ioctl(cmd))
1542 return true;
1543 return false;
1546 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1547 void __user *argp,
1548 struct nvme_ns_head *head,
1549 int srcu_idx)
1551 struct nvme_ctrl *ctrl = ns->ctrl;
1552 int ret;
1554 nvme_get_ctrl(ns->ctrl);
1555 nvme_put_ns_from_disk(head, srcu_idx);
1557 switch (cmd) {
1558 case NVME_IOCTL_ADMIN_CMD:
1559 ret = nvme_user_cmd(ctrl, NULL, argp);
1560 break;
1561 case NVME_IOCTL_ADMIN64_CMD:
1562 ret = nvme_user_cmd64(ctrl, NULL, argp);
1563 break;
1564 default:
1565 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1566 break;
1568 nvme_put_ctrl(ctrl);
1569 return ret;
1572 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1573 unsigned int cmd, unsigned long arg)
1575 struct nvme_ns_head *head = NULL;
1576 void __user *argp = (void __user *)arg;
1577 struct nvme_ns *ns;
1578 int srcu_idx, ret;
1580 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1581 if (unlikely(!ns))
1582 return -EWOULDBLOCK;
1585 * Handle ioctls that apply to the controller instead of the namespace
1586 * seperately and drop the ns SRCU reference early. This avoids a
1587 * deadlock when deleting namespaces using the passthrough interface.
1589 if (is_ctrl_ioctl(cmd))
1590 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1592 switch (cmd) {
1593 case NVME_IOCTL_ID:
1594 force_successful_syscall_return();
1595 ret = ns->head->ns_id;
1596 break;
1597 case NVME_IOCTL_IO_CMD:
1598 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1599 break;
1600 case NVME_IOCTL_SUBMIT_IO:
1601 ret = nvme_submit_io(ns, argp);
1602 break;
1603 case NVME_IOCTL_IO64_CMD:
1604 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1605 break;
1606 default:
1607 if (ns->ndev)
1608 ret = nvme_nvm_ioctl(ns, cmd, arg);
1609 else
1610 ret = -ENOTTY;
1613 nvme_put_ns_from_disk(head, srcu_idx);
1614 return ret;
1617 #ifdef CONFIG_COMPAT
1618 struct nvme_user_io32 {
1619 __u8 opcode;
1620 __u8 flags;
1621 __u16 control;
1622 __u16 nblocks;
1623 __u16 rsvd;
1624 __u64 metadata;
1625 __u64 addr;
1626 __u64 slba;
1627 __u32 dsmgmt;
1628 __u32 reftag;
1629 __u16 apptag;
1630 __u16 appmask;
1631 } __attribute__((__packed__));
1633 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1635 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1636 unsigned int cmd, unsigned long arg)
1639 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1640 * between 32 bit programs and 64 bit kernel.
1641 * The cause is that the results of sizeof(struct nvme_user_io),
1642 * which is used to define NVME_IOCTL_SUBMIT_IO,
1643 * are not same between 32 bit compiler and 64 bit compiler.
1644 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1645 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1646 * Other IOCTL numbers are same between 32 bit and 64 bit.
1647 * So there is nothing to do regarding to other IOCTL numbers.
1649 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1650 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1652 return nvme_ioctl(bdev, mode, cmd, arg);
1654 #else
1655 #define nvme_compat_ioctl NULL
1656 #endif /* CONFIG_COMPAT */
1658 static int nvme_open(struct block_device *bdev, fmode_t mode)
1660 struct nvme_ns *ns = bdev->bd_disk->private_data;
1662 #ifdef CONFIG_NVME_MULTIPATH
1663 /* should never be called due to GENHD_FL_HIDDEN */
1664 if (WARN_ON_ONCE(ns->head->disk))
1665 goto fail;
1666 #endif
1667 if (!kref_get_unless_zero(&ns->kref))
1668 goto fail;
1669 if (!try_module_get(ns->ctrl->ops->module))
1670 goto fail_put_ns;
1672 return 0;
1674 fail_put_ns:
1675 nvme_put_ns(ns);
1676 fail:
1677 return -ENXIO;
1680 static void nvme_release(struct gendisk *disk, fmode_t mode)
1682 struct nvme_ns *ns = disk->private_data;
1684 module_put(ns->ctrl->ops->module);
1685 nvme_put_ns(ns);
1688 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1690 /* some standard values */
1691 geo->heads = 1 << 6;
1692 geo->sectors = 1 << 5;
1693 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1694 return 0;
1697 #ifdef CONFIG_BLK_DEV_INTEGRITY
1698 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1700 struct blk_integrity integrity;
1702 memset(&integrity, 0, sizeof(integrity));
1703 switch (pi_type) {
1704 case NVME_NS_DPS_PI_TYPE3:
1705 integrity.profile = &t10_pi_type3_crc;
1706 integrity.tag_size = sizeof(u16) + sizeof(u32);
1707 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1708 break;
1709 case NVME_NS_DPS_PI_TYPE1:
1710 case NVME_NS_DPS_PI_TYPE2:
1711 integrity.profile = &t10_pi_type1_crc;
1712 integrity.tag_size = sizeof(u16);
1713 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1714 break;
1715 default:
1716 integrity.profile = NULL;
1717 break;
1719 integrity.tuple_size = ms;
1720 blk_integrity_register(disk, &integrity);
1721 blk_queue_max_integrity_segments(disk->queue, 1);
1723 #else
1724 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1727 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1729 static void nvme_set_chunk_size(struct nvme_ns *ns)
1731 u32 chunk_size = nvme_lba_to_sect(ns, ns->noiob);
1732 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1735 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1737 struct nvme_ctrl *ctrl = ns->ctrl;
1738 struct request_queue *queue = disk->queue;
1739 u32 size = queue_logical_block_size(queue);
1741 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1742 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1743 return;
1746 if (ctrl->nr_streams && ns->sws && ns->sgs)
1747 size *= ns->sws * ns->sgs;
1749 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1750 NVME_DSM_MAX_RANGES);
1752 queue->limits.discard_alignment = 0;
1753 queue->limits.discard_granularity = size;
1755 /* If discard is already enabled, don't reset queue limits */
1756 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1757 return;
1759 blk_queue_max_discard_sectors(queue, UINT_MAX);
1760 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1762 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1763 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1766 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1768 u64 max_blocks;
1770 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1771 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1772 return;
1774 * Even though NVMe spec explicitly states that MDTS is not
1775 * applicable to the write-zeroes:- "The restriction does not apply to
1776 * commands that do not transfer data between the host and the
1777 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1778 * In order to be more cautious use controller's max_hw_sectors value
1779 * to configure the maximum sectors for the write-zeroes which is
1780 * configured based on the controller's MDTS field in the
1781 * nvme_init_identify() if available.
1783 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1784 max_blocks = (u64)USHRT_MAX + 1;
1785 else
1786 max_blocks = ns->ctrl->max_hw_sectors + 1;
1788 blk_queue_max_write_zeroes_sectors(disk->queue,
1789 nvme_lba_to_sect(ns, max_blocks));
1792 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1793 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1795 memset(ids, 0, sizeof(*ids));
1797 if (ctrl->vs >= NVME_VS(1, 1, 0))
1798 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1799 if (ctrl->vs >= NVME_VS(1, 2, 0))
1800 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1801 if (ctrl->vs >= NVME_VS(1, 3, 0))
1802 return nvme_identify_ns_descs(ctrl, nsid, ids);
1803 return 0;
1806 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1808 return !uuid_is_null(&ids->uuid) ||
1809 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1810 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1813 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1815 return uuid_equal(&a->uuid, &b->uuid) &&
1816 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1817 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1820 static void nvme_update_disk_info(struct gendisk *disk,
1821 struct nvme_ns *ns, struct nvme_id_ns *id)
1823 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1824 unsigned short bs = 1 << ns->lba_shift;
1825 u32 atomic_bs, phys_bs, io_opt;
1827 if (ns->lba_shift > PAGE_SHIFT) {
1828 /* unsupported block size, set capacity to 0 later */
1829 bs = (1 << 9);
1831 blk_mq_freeze_queue(disk->queue);
1832 blk_integrity_unregister(disk);
1834 if (id->nabo == 0) {
1836 * Bit 1 indicates whether NAWUPF is defined for this namespace
1837 * and whether it should be used instead of AWUPF. If NAWUPF ==
1838 * 0 then AWUPF must be used instead.
1840 if (id->nsfeat & (1 << 1) && id->nawupf)
1841 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1842 else
1843 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1844 } else {
1845 atomic_bs = bs;
1847 phys_bs = bs;
1848 io_opt = bs;
1849 if (id->nsfeat & (1 << 4)) {
1850 /* NPWG = Namespace Preferred Write Granularity */
1851 phys_bs *= 1 + le16_to_cpu(id->npwg);
1852 /* NOWS = Namespace Optimal Write Size */
1853 io_opt *= 1 + le16_to_cpu(id->nows);
1856 blk_queue_logical_block_size(disk->queue, bs);
1858 * Linux filesystems assume writing a single physical block is
1859 * an atomic operation. Hence limit the physical block size to the
1860 * value of the Atomic Write Unit Power Fail parameter.
1862 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1863 blk_queue_io_min(disk->queue, phys_bs);
1864 blk_queue_io_opt(disk->queue, io_opt);
1866 if (ns->ms && !ns->ext &&
1867 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1868 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1869 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1870 ns->lba_shift > PAGE_SHIFT)
1871 capacity = 0;
1873 set_capacity_revalidate_and_notify(disk, capacity, false);
1875 nvme_config_discard(disk, ns);
1876 nvme_config_write_zeroes(disk, ns);
1878 if (id->nsattr & (1 << 0))
1879 set_disk_ro(disk, true);
1880 else
1881 set_disk_ro(disk, false);
1883 blk_mq_unfreeze_queue(disk->queue);
1886 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1888 struct nvme_ns *ns = disk->private_data;
1891 * If identify namespace failed, use default 512 byte block size so
1892 * block layer can use before failing read/write for 0 capacity.
1894 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1895 if (ns->lba_shift == 0)
1896 ns->lba_shift = 9;
1897 ns->noiob = le16_to_cpu(id->noiob);
1898 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1899 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1900 /* the PI implementation requires metadata equal t10 pi tuple size */
1901 if (ns->ms == sizeof(struct t10_pi_tuple))
1902 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1903 else
1904 ns->pi_type = 0;
1906 if (ns->noiob)
1907 nvme_set_chunk_size(ns);
1908 nvme_update_disk_info(disk, ns, id);
1909 #ifdef CONFIG_NVME_MULTIPATH
1910 if (ns->head->disk) {
1911 nvme_update_disk_info(ns->head->disk, ns, id);
1912 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1913 if (bdi_cap_stable_pages_required(ns->queue->backing_dev_info)) {
1914 struct backing_dev_info *info =
1915 ns->head->disk->queue->backing_dev_info;
1917 info->capabilities |= BDI_CAP_STABLE_WRITES;
1920 revalidate_disk(ns->head->disk);
1922 #endif
1925 static int nvme_revalidate_disk(struct gendisk *disk)
1927 struct nvme_ns *ns = disk->private_data;
1928 struct nvme_ctrl *ctrl = ns->ctrl;
1929 struct nvme_id_ns *id;
1930 struct nvme_ns_ids ids;
1931 int ret = 0;
1933 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1934 set_capacity(disk, 0);
1935 return -ENODEV;
1938 ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1939 if (ret)
1940 goto out;
1942 if (id->ncap == 0) {
1943 ret = -ENODEV;
1944 goto free_id;
1947 __nvme_revalidate_disk(disk, id);
1948 ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1949 if (ret)
1950 goto free_id;
1952 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1953 dev_err(ctrl->device,
1954 "identifiers changed for nsid %d\n", ns->head->ns_id);
1955 ret = -ENODEV;
1958 free_id:
1959 kfree(id);
1960 out:
1962 * Only fail the function if we got a fatal error back from the
1963 * device, otherwise ignore the error and just move on.
1965 if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1966 ret = 0;
1967 else if (ret > 0)
1968 ret = blk_status_to_errno(nvme_error_status(ret));
1969 return ret;
1972 static char nvme_pr_type(enum pr_type type)
1974 switch (type) {
1975 case PR_WRITE_EXCLUSIVE:
1976 return 1;
1977 case PR_EXCLUSIVE_ACCESS:
1978 return 2;
1979 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1980 return 3;
1981 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1982 return 4;
1983 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1984 return 5;
1985 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1986 return 6;
1987 default:
1988 return 0;
1992 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1993 u64 key, u64 sa_key, u8 op)
1995 struct nvme_ns_head *head = NULL;
1996 struct nvme_ns *ns;
1997 struct nvme_command c;
1998 int srcu_idx, ret;
1999 u8 data[16] = { 0, };
2001 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2002 if (unlikely(!ns))
2003 return -EWOULDBLOCK;
2005 put_unaligned_le64(key, &data[0]);
2006 put_unaligned_le64(sa_key, &data[8]);
2008 memset(&c, 0, sizeof(c));
2009 c.common.opcode = op;
2010 c.common.nsid = cpu_to_le32(ns->head->ns_id);
2011 c.common.cdw10 = cpu_to_le32(cdw10);
2013 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2014 nvme_put_ns_from_disk(head, srcu_idx);
2015 return ret;
2018 static int nvme_pr_register(struct block_device *bdev, u64 old,
2019 u64 new, unsigned flags)
2021 u32 cdw10;
2023 if (flags & ~PR_FL_IGNORE_KEY)
2024 return -EOPNOTSUPP;
2026 cdw10 = old ? 2 : 0;
2027 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2028 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2029 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2032 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2033 enum pr_type type, unsigned flags)
2035 u32 cdw10;
2037 if (flags & ~PR_FL_IGNORE_KEY)
2038 return -EOPNOTSUPP;
2040 cdw10 = nvme_pr_type(type) << 8;
2041 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2042 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2045 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2046 enum pr_type type, bool abort)
2048 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2049 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2052 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2054 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2055 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2058 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2060 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2061 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2064 static const struct pr_ops nvme_pr_ops = {
2065 .pr_register = nvme_pr_register,
2066 .pr_reserve = nvme_pr_reserve,
2067 .pr_release = nvme_pr_release,
2068 .pr_preempt = nvme_pr_preempt,
2069 .pr_clear = nvme_pr_clear,
2072 #ifdef CONFIG_BLK_SED_OPAL
2073 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2074 bool send)
2076 struct nvme_ctrl *ctrl = data;
2077 struct nvme_command cmd;
2079 memset(&cmd, 0, sizeof(cmd));
2080 if (send)
2081 cmd.common.opcode = nvme_admin_security_send;
2082 else
2083 cmd.common.opcode = nvme_admin_security_recv;
2084 cmd.common.nsid = 0;
2085 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2086 cmd.common.cdw11 = cpu_to_le32(len);
2088 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2089 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2091 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2092 #endif /* CONFIG_BLK_SED_OPAL */
2094 static const struct block_device_operations nvme_fops = {
2095 .owner = THIS_MODULE,
2096 .ioctl = nvme_ioctl,
2097 .compat_ioctl = nvme_compat_ioctl,
2098 .open = nvme_open,
2099 .release = nvme_release,
2100 .getgeo = nvme_getgeo,
2101 .revalidate_disk= nvme_revalidate_disk,
2102 .pr_ops = &nvme_pr_ops,
2105 #ifdef CONFIG_NVME_MULTIPATH
2106 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2108 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2110 if (!kref_get_unless_zero(&head->ref))
2111 return -ENXIO;
2112 return 0;
2115 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2117 nvme_put_ns_head(disk->private_data);
2120 const struct block_device_operations nvme_ns_head_ops = {
2121 .owner = THIS_MODULE,
2122 .open = nvme_ns_head_open,
2123 .release = nvme_ns_head_release,
2124 .ioctl = nvme_ioctl,
2125 .compat_ioctl = nvme_compat_ioctl,
2126 .getgeo = nvme_getgeo,
2127 .pr_ops = &nvme_pr_ops,
2129 #endif /* CONFIG_NVME_MULTIPATH */
2131 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2133 unsigned long timeout =
2134 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2135 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2136 int ret;
2138 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2139 if (csts == ~0)
2140 return -ENODEV;
2141 if ((csts & NVME_CSTS_RDY) == bit)
2142 break;
2144 usleep_range(1000, 2000);
2145 if (fatal_signal_pending(current))
2146 return -EINTR;
2147 if (time_after(jiffies, timeout)) {
2148 dev_err(ctrl->device,
2149 "Device not ready; aborting %s, CSTS=0x%x\n",
2150 enabled ? "initialisation" : "reset", csts);
2151 return -ENODEV;
2155 return ret;
2159 * If the device has been passed off to us in an enabled state, just clear
2160 * the enabled bit. The spec says we should set the 'shutdown notification
2161 * bits', but doing so may cause the device to complete commands to the
2162 * admin queue ... and we don't know what memory that might be pointing at!
2164 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2166 int ret;
2168 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2169 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2171 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2172 if (ret)
2173 return ret;
2175 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2176 msleep(NVME_QUIRK_DELAY_AMOUNT);
2178 return nvme_wait_ready(ctrl, ctrl->cap, false);
2180 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2182 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2185 * Default to a 4K page size, with the intention to update this
2186 * path in the future to accomodate architectures with differing
2187 * kernel and IO page sizes.
2189 unsigned dev_page_min, page_shift = 12;
2190 int ret;
2192 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2193 if (ret) {
2194 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2195 return ret;
2197 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2199 if (page_shift < dev_page_min) {
2200 dev_err(ctrl->device,
2201 "Minimum device page size %u too large for host (%u)\n",
2202 1 << dev_page_min, 1 << page_shift);
2203 return -ENODEV;
2206 ctrl->page_size = 1 << page_shift;
2208 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2209 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2210 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2211 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2212 ctrl->ctrl_config |= NVME_CC_ENABLE;
2214 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2215 if (ret)
2216 return ret;
2217 return nvme_wait_ready(ctrl, ctrl->cap, true);
2219 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2221 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2223 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2224 u32 csts;
2225 int ret;
2227 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2228 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2230 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2231 if (ret)
2232 return ret;
2234 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2235 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2236 break;
2238 msleep(100);
2239 if (fatal_signal_pending(current))
2240 return -EINTR;
2241 if (time_after(jiffies, timeout)) {
2242 dev_err(ctrl->device,
2243 "Device shutdown incomplete; abort shutdown\n");
2244 return -ENODEV;
2248 return ret;
2250 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2252 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2253 struct request_queue *q)
2255 bool vwc = false;
2257 if (ctrl->max_hw_sectors) {
2258 u32 max_segments =
2259 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2261 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2262 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2263 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2265 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2266 is_power_of_2(ctrl->max_hw_sectors))
2267 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2268 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2269 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2270 vwc = true;
2271 blk_queue_write_cache(q, vwc, vwc);
2274 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2276 __le64 ts;
2277 int ret;
2279 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2280 return 0;
2282 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2283 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2284 NULL);
2285 if (ret)
2286 dev_warn_once(ctrl->device,
2287 "could not set timestamp (%d)\n", ret);
2288 return ret;
2291 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2293 struct nvme_feat_host_behavior *host;
2294 int ret;
2296 /* Don't bother enabling the feature if retry delay is not reported */
2297 if (!ctrl->crdt[0])
2298 return 0;
2300 host = kzalloc(sizeof(*host), GFP_KERNEL);
2301 if (!host)
2302 return 0;
2304 host->acre = NVME_ENABLE_ACRE;
2305 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2306 host, sizeof(*host), NULL);
2307 kfree(host);
2308 return ret;
2311 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2314 * APST (Autonomous Power State Transition) lets us program a
2315 * table of power state transitions that the controller will
2316 * perform automatically. We configure it with a simple
2317 * heuristic: we are willing to spend at most 2% of the time
2318 * transitioning between power states. Therefore, when running
2319 * in any given state, we will enter the next lower-power
2320 * non-operational state after waiting 50 * (enlat + exlat)
2321 * microseconds, as long as that state's exit latency is under
2322 * the requested maximum latency.
2324 * We will not autonomously enter any non-operational state for
2325 * which the total latency exceeds ps_max_latency_us. Users
2326 * can set ps_max_latency_us to zero to turn off APST.
2329 unsigned apste;
2330 struct nvme_feat_auto_pst *table;
2331 u64 max_lat_us = 0;
2332 int max_ps = -1;
2333 int ret;
2336 * If APST isn't supported or if we haven't been initialized yet,
2337 * then don't do anything.
2339 if (!ctrl->apsta)
2340 return 0;
2342 if (ctrl->npss > 31) {
2343 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2344 return 0;
2347 table = kzalloc(sizeof(*table), GFP_KERNEL);
2348 if (!table)
2349 return 0;
2351 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2352 /* Turn off APST. */
2353 apste = 0;
2354 dev_dbg(ctrl->device, "APST disabled\n");
2355 } else {
2356 __le64 target = cpu_to_le64(0);
2357 int state;
2360 * Walk through all states from lowest- to highest-power.
2361 * According to the spec, lower-numbered states use more
2362 * power. NPSS, despite the name, is the index of the
2363 * lowest-power state, not the number of states.
2365 for (state = (int)ctrl->npss; state >= 0; state--) {
2366 u64 total_latency_us, exit_latency_us, transition_ms;
2368 if (target)
2369 table->entries[state] = target;
2372 * Don't allow transitions to the deepest state
2373 * if it's quirked off.
2375 if (state == ctrl->npss &&
2376 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2377 continue;
2380 * Is this state a useful non-operational state for
2381 * higher-power states to autonomously transition to?
2383 if (!(ctrl->psd[state].flags &
2384 NVME_PS_FLAGS_NON_OP_STATE))
2385 continue;
2387 exit_latency_us =
2388 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2389 if (exit_latency_us > ctrl->ps_max_latency_us)
2390 continue;
2392 total_latency_us =
2393 exit_latency_us +
2394 le32_to_cpu(ctrl->psd[state].entry_lat);
2397 * This state is good. Use it as the APST idle
2398 * target for higher power states.
2400 transition_ms = total_latency_us + 19;
2401 do_div(transition_ms, 20);
2402 if (transition_ms > (1 << 24) - 1)
2403 transition_ms = (1 << 24) - 1;
2405 target = cpu_to_le64((state << 3) |
2406 (transition_ms << 8));
2408 if (max_ps == -1)
2409 max_ps = state;
2411 if (total_latency_us > max_lat_us)
2412 max_lat_us = total_latency_us;
2415 apste = 1;
2417 if (max_ps == -1) {
2418 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2419 } else {
2420 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2421 max_ps, max_lat_us, (int)sizeof(*table), table);
2425 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2426 table, sizeof(*table), NULL);
2427 if (ret)
2428 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2430 kfree(table);
2431 return ret;
2434 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2436 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2437 u64 latency;
2439 switch (val) {
2440 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2441 case PM_QOS_LATENCY_ANY:
2442 latency = U64_MAX;
2443 break;
2445 default:
2446 latency = val;
2449 if (ctrl->ps_max_latency_us != latency) {
2450 ctrl->ps_max_latency_us = latency;
2451 nvme_configure_apst(ctrl);
2455 struct nvme_core_quirk_entry {
2457 * NVMe model and firmware strings are padded with spaces. For
2458 * simplicity, strings in the quirk table are padded with NULLs
2459 * instead.
2461 u16 vid;
2462 const char *mn;
2463 const char *fr;
2464 unsigned long quirks;
2467 static const struct nvme_core_quirk_entry core_quirks[] = {
2470 * This Toshiba device seems to die using any APST states. See:
2471 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2473 .vid = 0x1179,
2474 .mn = "THNSF5256GPUK TOSHIBA",
2475 .quirks = NVME_QUIRK_NO_APST,
2479 * This LiteON CL1-3D*-Q11 firmware version has a race
2480 * condition associated with actions related to suspend to idle
2481 * LiteON has resolved the problem in future firmware
2483 .vid = 0x14a4,
2484 .fr = "22301111",
2485 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2489 /* match is null-terminated but idstr is space-padded. */
2490 static bool string_matches(const char *idstr, const char *match, size_t len)
2492 size_t matchlen;
2494 if (!match)
2495 return true;
2497 matchlen = strlen(match);
2498 WARN_ON_ONCE(matchlen > len);
2500 if (memcmp(idstr, match, matchlen))
2501 return false;
2503 for (; matchlen < len; matchlen++)
2504 if (idstr[matchlen] != ' ')
2505 return false;
2507 return true;
2510 static bool quirk_matches(const struct nvme_id_ctrl *id,
2511 const struct nvme_core_quirk_entry *q)
2513 return q->vid == le16_to_cpu(id->vid) &&
2514 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2515 string_matches(id->fr, q->fr, sizeof(id->fr));
2518 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2519 struct nvme_id_ctrl *id)
2521 size_t nqnlen;
2522 int off;
2524 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2525 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2526 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2527 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2528 return;
2531 if (ctrl->vs >= NVME_VS(1, 2, 1))
2532 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2535 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2536 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2537 "nqn.2014.08.org.nvmexpress:%04x%04x",
2538 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2539 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2540 off += sizeof(id->sn);
2541 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2542 off += sizeof(id->mn);
2543 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2546 static void nvme_release_subsystem(struct device *dev)
2548 struct nvme_subsystem *subsys =
2549 container_of(dev, struct nvme_subsystem, dev);
2551 if (subsys->instance >= 0)
2552 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2553 kfree(subsys);
2556 static void nvme_destroy_subsystem(struct kref *ref)
2558 struct nvme_subsystem *subsys =
2559 container_of(ref, struct nvme_subsystem, ref);
2561 mutex_lock(&nvme_subsystems_lock);
2562 list_del(&subsys->entry);
2563 mutex_unlock(&nvme_subsystems_lock);
2565 ida_destroy(&subsys->ns_ida);
2566 device_del(&subsys->dev);
2567 put_device(&subsys->dev);
2570 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2572 kref_put(&subsys->ref, nvme_destroy_subsystem);
2575 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2577 struct nvme_subsystem *subsys;
2579 lockdep_assert_held(&nvme_subsystems_lock);
2582 * Fail matches for discovery subsystems. This results
2583 * in each discovery controller bound to a unique subsystem.
2584 * This avoids issues with validating controller values
2585 * that can only be true when there is a single unique subsystem.
2586 * There may be multiple and completely independent entities
2587 * that provide discovery controllers.
2589 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2590 return NULL;
2592 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2593 if (strcmp(subsys->subnqn, subsysnqn))
2594 continue;
2595 if (!kref_get_unless_zero(&subsys->ref))
2596 continue;
2597 return subsys;
2600 return NULL;
2603 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2604 struct device_attribute subsys_attr_##_name = \
2605 __ATTR(_name, _mode, _show, NULL)
2607 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2608 struct device_attribute *attr,
2609 char *buf)
2611 struct nvme_subsystem *subsys =
2612 container_of(dev, struct nvme_subsystem, dev);
2614 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2616 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2618 #define nvme_subsys_show_str_function(field) \
2619 static ssize_t subsys_##field##_show(struct device *dev, \
2620 struct device_attribute *attr, char *buf) \
2622 struct nvme_subsystem *subsys = \
2623 container_of(dev, struct nvme_subsystem, dev); \
2624 return sprintf(buf, "%.*s\n", \
2625 (int)sizeof(subsys->field), subsys->field); \
2627 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2629 nvme_subsys_show_str_function(model);
2630 nvme_subsys_show_str_function(serial);
2631 nvme_subsys_show_str_function(firmware_rev);
2633 static struct attribute *nvme_subsys_attrs[] = {
2634 &subsys_attr_model.attr,
2635 &subsys_attr_serial.attr,
2636 &subsys_attr_firmware_rev.attr,
2637 &subsys_attr_subsysnqn.attr,
2638 #ifdef CONFIG_NVME_MULTIPATH
2639 &subsys_attr_iopolicy.attr,
2640 #endif
2641 NULL,
2644 static struct attribute_group nvme_subsys_attrs_group = {
2645 .attrs = nvme_subsys_attrs,
2648 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2649 &nvme_subsys_attrs_group,
2650 NULL,
2653 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2654 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2656 struct nvme_ctrl *tmp;
2658 lockdep_assert_held(&nvme_subsystems_lock);
2660 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2661 if (nvme_state_terminal(tmp))
2662 continue;
2664 if (tmp->cntlid == ctrl->cntlid) {
2665 dev_err(ctrl->device,
2666 "Duplicate cntlid %u with %s, rejecting\n",
2667 ctrl->cntlid, dev_name(tmp->device));
2668 return false;
2671 if ((id->cmic & (1 << 1)) ||
2672 (ctrl->opts && ctrl->opts->discovery_nqn))
2673 continue;
2675 dev_err(ctrl->device,
2676 "Subsystem does not support multiple controllers\n");
2677 return false;
2680 return true;
2683 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2685 struct nvme_subsystem *subsys, *found;
2686 int ret;
2688 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2689 if (!subsys)
2690 return -ENOMEM;
2692 subsys->instance = -1;
2693 mutex_init(&subsys->lock);
2694 kref_init(&subsys->ref);
2695 INIT_LIST_HEAD(&subsys->ctrls);
2696 INIT_LIST_HEAD(&subsys->nsheads);
2697 nvme_init_subnqn(subsys, ctrl, id);
2698 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2699 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2700 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2701 subsys->vendor_id = le16_to_cpu(id->vid);
2702 subsys->cmic = id->cmic;
2703 subsys->awupf = le16_to_cpu(id->awupf);
2704 #ifdef CONFIG_NVME_MULTIPATH
2705 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2706 #endif
2708 subsys->dev.class = nvme_subsys_class;
2709 subsys->dev.release = nvme_release_subsystem;
2710 subsys->dev.groups = nvme_subsys_attrs_groups;
2711 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2712 device_initialize(&subsys->dev);
2714 mutex_lock(&nvme_subsystems_lock);
2715 found = __nvme_find_get_subsystem(subsys->subnqn);
2716 if (found) {
2717 put_device(&subsys->dev);
2718 subsys = found;
2720 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2721 ret = -EINVAL;
2722 goto out_put_subsystem;
2724 } else {
2725 ret = device_add(&subsys->dev);
2726 if (ret) {
2727 dev_err(ctrl->device,
2728 "failed to register subsystem device.\n");
2729 put_device(&subsys->dev);
2730 goto out_unlock;
2732 ida_init(&subsys->ns_ida);
2733 list_add_tail(&subsys->entry, &nvme_subsystems);
2736 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2737 dev_name(ctrl->device));
2738 if (ret) {
2739 dev_err(ctrl->device,
2740 "failed to create sysfs link from subsystem.\n");
2741 goto out_put_subsystem;
2744 if (!found)
2745 subsys->instance = ctrl->instance;
2746 ctrl->subsys = subsys;
2747 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2748 mutex_unlock(&nvme_subsystems_lock);
2749 return 0;
2751 out_put_subsystem:
2752 nvme_put_subsystem(subsys);
2753 out_unlock:
2754 mutex_unlock(&nvme_subsystems_lock);
2755 return ret;
2758 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2759 void *log, size_t size, u64 offset)
2761 struct nvme_command c = { };
2762 unsigned long dwlen = size / 4 - 1;
2764 c.get_log_page.opcode = nvme_admin_get_log_page;
2765 c.get_log_page.nsid = cpu_to_le32(nsid);
2766 c.get_log_page.lid = log_page;
2767 c.get_log_page.lsp = lsp;
2768 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2769 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2770 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2771 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2773 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2776 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2778 int ret;
2780 if (!ctrl->effects)
2781 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2783 if (!ctrl->effects)
2784 return 0;
2786 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2787 ctrl->effects, sizeof(*ctrl->effects), 0);
2788 if (ret) {
2789 kfree(ctrl->effects);
2790 ctrl->effects = NULL;
2792 return ret;
2796 * Initialize the cached copies of the Identify data and various controller
2797 * register in our nvme_ctrl structure. This should be called as soon as
2798 * the admin queue is fully up and running.
2800 int nvme_init_identify(struct nvme_ctrl *ctrl)
2802 struct nvme_id_ctrl *id;
2803 int ret, page_shift;
2804 u32 max_hw_sectors;
2805 bool prev_apst_enabled;
2807 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2808 if (ret) {
2809 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2810 return ret;
2812 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2813 ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2815 if (ctrl->vs >= NVME_VS(1, 1, 0))
2816 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2818 ret = nvme_identify_ctrl(ctrl, &id);
2819 if (ret) {
2820 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2821 return -EIO;
2824 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2825 ret = nvme_get_effects_log(ctrl);
2826 if (ret < 0)
2827 goto out_free;
2830 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2831 ctrl->cntlid = le16_to_cpu(id->cntlid);
2833 if (!ctrl->identified) {
2834 int i;
2836 ret = nvme_init_subsystem(ctrl, id);
2837 if (ret)
2838 goto out_free;
2841 * Check for quirks. Quirk can depend on firmware version,
2842 * so, in principle, the set of quirks present can change
2843 * across a reset. As a possible future enhancement, we
2844 * could re-scan for quirks every time we reinitialize
2845 * the device, but we'd have to make sure that the driver
2846 * behaves intelligently if the quirks change.
2848 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2849 if (quirk_matches(id, &core_quirks[i]))
2850 ctrl->quirks |= core_quirks[i].quirks;
2854 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2855 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2856 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2859 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2860 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2861 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2863 ctrl->oacs = le16_to_cpu(id->oacs);
2864 ctrl->oncs = le16_to_cpu(id->oncs);
2865 ctrl->mtfa = le16_to_cpu(id->mtfa);
2866 ctrl->oaes = le32_to_cpu(id->oaes);
2867 ctrl->wctemp = le16_to_cpu(id->wctemp);
2868 ctrl->cctemp = le16_to_cpu(id->cctemp);
2870 atomic_set(&ctrl->abort_limit, id->acl + 1);
2871 ctrl->vwc = id->vwc;
2872 if (id->mdts)
2873 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2874 else
2875 max_hw_sectors = UINT_MAX;
2876 ctrl->max_hw_sectors =
2877 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2879 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2880 ctrl->sgls = le32_to_cpu(id->sgls);
2881 ctrl->kas = le16_to_cpu(id->kas);
2882 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2883 ctrl->ctratt = le32_to_cpu(id->ctratt);
2885 if (id->rtd3e) {
2886 /* us -> s */
2887 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2889 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2890 shutdown_timeout, 60);
2892 if (ctrl->shutdown_timeout != shutdown_timeout)
2893 dev_info(ctrl->device,
2894 "Shutdown timeout set to %u seconds\n",
2895 ctrl->shutdown_timeout);
2896 } else
2897 ctrl->shutdown_timeout = shutdown_timeout;
2899 ctrl->npss = id->npss;
2900 ctrl->apsta = id->apsta;
2901 prev_apst_enabled = ctrl->apst_enabled;
2902 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2903 if (force_apst && id->apsta) {
2904 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2905 ctrl->apst_enabled = true;
2906 } else {
2907 ctrl->apst_enabled = false;
2909 } else {
2910 ctrl->apst_enabled = id->apsta;
2912 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2914 if (ctrl->ops->flags & NVME_F_FABRICS) {
2915 ctrl->icdoff = le16_to_cpu(id->icdoff);
2916 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2917 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2918 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2921 * In fabrics we need to verify the cntlid matches the
2922 * admin connect
2924 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2925 dev_err(ctrl->device,
2926 "Mismatching cntlid: Connect %u vs Identify "
2927 "%u, rejecting\n",
2928 ctrl->cntlid, le16_to_cpu(id->cntlid));
2929 ret = -EINVAL;
2930 goto out_free;
2933 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2934 dev_err(ctrl->device,
2935 "keep-alive support is mandatory for fabrics\n");
2936 ret = -EINVAL;
2937 goto out_free;
2939 } else {
2940 ctrl->hmpre = le32_to_cpu(id->hmpre);
2941 ctrl->hmmin = le32_to_cpu(id->hmmin);
2942 ctrl->hmminds = le32_to_cpu(id->hmminds);
2943 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2946 ret = nvme_mpath_init(ctrl, id);
2947 kfree(id);
2949 if (ret < 0)
2950 return ret;
2952 if (ctrl->apst_enabled && !prev_apst_enabled)
2953 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2954 else if (!ctrl->apst_enabled && prev_apst_enabled)
2955 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2957 ret = nvme_configure_apst(ctrl);
2958 if (ret < 0)
2959 return ret;
2961 ret = nvme_configure_timestamp(ctrl);
2962 if (ret < 0)
2963 return ret;
2965 ret = nvme_configure_directives(ctrl);
2966 if (ret < 0)
2967 return ret;
2969 ret = nvme_configure_acre(ctrl);
2970 if (ret < 0)
2971 return ret;
2973 if (!ctrl->identified)
2974 nvme_hwmon_init(ctrl);
2976 ctrl->identified = true;
2978 return 0;
2980 out_free:
2981 kfree(id);
2982 return ret;
2984 EXPORT_SYMBOL_GPL(nvme_init_identify);
2986 static int nvme_dev_open(struct inode *inode, struct file *file)
2988 struct nvme_ctrl *ctrl =
2989 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2991 switch (ctrl->state) {
2992 case NVME_CTRL_LIVE:
2993 break;
2994 default:
2995 return -EWOULDBLOCK;
2998 file->private_data = ctrl;
2999 return 0;
3002 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3004 struct nvme_ns *ns;
3005 int ret;
3007 down_read(&ctrl->namespaces_rwsem);
3008 if (list_empty(&ctrl->namespaces)) {
3009 ret = -ENOTTY;
3010 goto out_unlock;
3013 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3014 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3015 dev_warn(ctrl->device,
3016 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3017 ret = -EINVAL;
3018 goto out_unlock;
3021 dev_warn(ctrl->device,
3022 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3023 kref_get(&ns->kref);
3024 up_read(&ctrl->namespaces_rwsem);
3026 ret = nvme_user_cmd(ctrl, ns, argp);
3027 nvme_put_ns(ns);
3028 return ret;
3030 out_unlock:
3031 up_read(&ctrl->namespaces_rwsem);
3032 return ret;
3035 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3036 unsigned long arg)
3038 struct nvme_ctrl *ctrl = file->private_data;
3039 void __user *argp = (void __user *)arg;
3041 switch (cmd) {
3042 case NVME_IOCTL_ADMIN_CMD:
3043 return nvme_user_cmd(ctrl, NULL, argp);
3044 case NVME_IOCTL_ADMIN64_CMD:
3045 return nvme_user_cmd64(ctrl, NULL, argp);
3046 case NVME_IOCTL_IO_CMD:
3047 return nvme_dev_user_cmd(ctrl, argp);
3048 case NVME_IOCTL_RESET:
3049 dev_warn(ctrl->device, "resetting controller\n");
3050 return nvme_reset_ctrl_sync(ctrl);
3051 case NVME_IOCTL_SUBSYS_RESET:
3052 return nvme_reset_subsystem(ctrl);
3053 case NVME_IOCTL_RESCAN:
3054 nvme_queue_scan(ctrl);
3055 return 0;
3056 default:
3057 return -ENOTTY;
3061 static const struct file_operations nvme_dev_fops = {
3062 .owner = THIS_MODULE,
3063 .open = nvme_dev_open,
3064 .unlocked_ioctl = nvme_dev_ioctl,
3065 .compat_ioctl = compat_ptr_ioctl,
3068 static ssize_t nvme_sysfs_reset(struct device *dev,
3069 struct device_attribute *attr, const char *buf,
3070 size_t count)
3072 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3073 int ret;
3075 ret = nvme_reset_ctrl_sync(ctrl);
3076 if (ret < 0)
3077 return ret;
3078 return count;
3080 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3082 static ssize_t nvme_sysfs_rescan(struct device *dev,
3083 struct device_attribute *attr, const char *buf,
3084 size_t count)
3086 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3088 nvme_queue_scan(ctrl);
3089 return count;
3091 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3093 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3095 struct gendisk *disk = dev_to_disk(dev);
3097 if (disk->fops == &nvme_fops)
3098 return nvme_get_ns_from_dev(dev)->head;
3099 else
3100 return disk->private_data;
3103 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3104 char *buf)
3106 struct nvme_ns_head *head = dev_to_ns_head(dev);
3107 struct nvme_ns_ids *ids = &head->ids;
3108 struct nvme_subsystem *subsys = head->subsys;
3109 int serial_len = sizeof(subsys->serial);
3110 int model_len = sizeof(subsys->model);
3112 if (!uuid_is_null(&ids->uuid))
3113 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3115 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3116 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3118 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3119 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3121 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3122 subsys->serial[serial_len - 1] == '\0'))
3123 serial_len--;
3124 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3125 subsys->model[model_len - 1] == '\0'))
3126 model_len--;
3128 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3129 serial_len, subsys->serial, model_len, subsys->model,
3130 head->ns_id);
3132 static DEVICE_ATTR_RO(wwid);
3134 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3135 char *buf)
3137 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3139 static DEVICE_ATTR_RO(nguid);
3141 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3142 char *buf)
3144 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3146 /* For backward compatibility expose the NGUID to userspace if
3147 * we have no UUID set
3149 if (uuid_is_null(&ids->uuid)) {
3150 printk_ratelimited(KERN_WARNING
3151 "No UUID available providing old NGUID\n");
3152 return sprintf(buf, "%pU\n", ids->nguid);
3154 return sprintf(buf, "%pU\n", &ids->uuid);
3156 static DEVICE_ATTR_RO(uuid);
3158 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3159 char *buf)
3161 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3163 static DEVICE_ATTR_RO(eui);
3165 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3166 char *buf)
3168 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3170 static DEVICE_ATTR_RO(nsid);
3172 static struct attribute *nvme_ns_id_attrs[] = {
3173 &dev_attr_wwid.attr,
3174 &dev_attr_uuid.attr,
3175 &dev_attr_nguid.attr,
3176 &dev_attr_eui.attr,
3177 &dev_attr_nsid.attr,
3178 #ifdef CONFIG_NVME_MULTIPATH
3179 &dev_attr_ana_grpid.attr,
3180 &dev_attr_ana_state.attr,
3181 #endif
3182 NULL,
3185 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3186 struct attribute *a, int n)
3188 struct device *dev = container_of(kobj, struct device, kobj);
3189 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3191 if (a == &dev_attr_uuid.attr) {
3192 if (uuid_is_null(&ids->uuid) &&
3193 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3194 return 0;
3196 if (a == &dev_attr_nguid.attr) {
3197 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3198 return 0;
3200 if (a == &dev_attr_eui.attr) {
3201 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3202 return 0;
3204 #ifdef CONFIG_NVME_MULTIPATH
3205 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3206 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3207 return 0;
3208 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3209 return 0;
3211 #endif
3212 return a->mode;
3215 static const struct attribute_group nvme_ns_id_attr_group = {
3216 .attrs = nvme_ns_id_attrs,
3217 .is_visible = nvme_ns_id_attrs_are_visible,
3220 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3221 &nvme_ns_id_attr_group,
3222 #ifdef CONFIG_NVM
3223 &nvme_nvm_attr_group,
3224 #endif
3225 NULL,
3228 #define nvme_show_str_function(field) \
3229 static ssize_t field##_show(struct device *dev, \
3230 struct device_attribute *attr, char *buf) \
3232 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3233 return sprintf(buf, "%.*s\n", \
3234 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3236 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3238 nvme_show_str_function(model);
3239 nvme_show_str_function(serial);
3240 nvme_show_str_function(firmware_rev);
3242 #define nvme_show_int_function(field) \
3243 static ssize_t field##_show(struct device *dev, \
3244 struct device_attribute *attr, char *buf) \
3246 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3247 return sprintf(buf, "%d\n", ctrl->field); \
3249 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3251 nvme_show_int_function(cntlid);
3252 nvme_show_int_function(numa_node);
3253 nvme_show_int_function(queue_count);
3254 nvme_show_int_function(sqsize);
3256 static ssize_t nvme_sysfs_delete(struct device *dev,
3257 struct device_attribute *attr, const char *buf,
3258 size_t count)
3260 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3262 /* Can't delete non-created controllers */
3263 if (!ctrl->created)
3264 return -EBUSY;
3266 if (device_remove_file_self(dev, attr))
3267 nvme_delete_ctrl_sync(ctrl);
3268 return count;
3270 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3272 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3273 struct device_attribute *attr,
3274 char *buf)
3276 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3278 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3280 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3282 static ssize_t nvme_sysfs_show_state(struct device *dev,
3283 struct device_attribute *attr,
3284 char *buf)
3286 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3287 static const char *const state_name[] = {
3288 [NVME_CTRL_NEW] = "new",
3289 [NVME_CTRL_LIVE] = "live",
3290 [NVME_CTRL_RESETTING] = "resetting",
3291 [NVME_CTRL_CONNECTING] = "connecting",
3292 [NVME_CTRL_DELETING] = "deleting",
3293 [NVME_CTRL_DEAD] = "dead",
3296 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3297 state_name[ctrl->state])
3298 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3300 return sprintf(buf, "unknown state\n");
3303 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3305 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3306 struct device_attribute *attr,
3307 char *buf)
3309 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3311 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3313 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3315 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3316 struct device_attribute *attr,
3317 char *buf)
3319 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3321 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3323 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3325 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3326 struct device_attribute *attr,
3327 char *buf)
3329 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3331 return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3333 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3335 static ssize_t nvme_sysfs_show_address(struct device *dev,
3336 struct device_attribute *attr,
3337 char *buf)
3339 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3341 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3343 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3345 static struct attribute *nvme_dev_attrs[] = {
3346 &dev_attr_reset_controller.attr,
3347 &dev_attr_rescan_controller.attr,
3348 &dev_attr_model.attr,
3349 &dev_attr_serial.attr,
3350 &dev_attr_firmware_rev.attr,
3351 &dev_attr_cntlid.attr,
3352 &dev_attr_delete_controller.attr,
3353 &dev_attr_transport.attr,
3354 &dev_attr_subsysnqn.attr,
3355 &dev_attr_address.attr,
3356 &dev_attr_state.attr,
3357 &dev_attr_numa_node.attr,
3358 &dev_attr_queue_count.attr,
3359 &dev_attr_sqsize.attr,
3360 &dev_attr_hostnqn.attr,
3361 &dev_attr_hostid.attr,
3362 NULL
3365 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3366 struct attribute *a, int n)
3368 struct device *dev = container_of(kobj, struct device, kobj);
3369 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3371 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3372 return 0;
3373 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3374 return 0;
3375 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3376 return 0;
3377 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3378 return 0;
3380 return a->mode;
3383 static struct attribute_group nvme_dev_attrs_group = {
3384 .attrs = nvme_dev_attrs,
3385 .is_visible = nvme_dev_attrs_are_visible,
3388 static const struct attribute_group *nvme_dev_attr_groups[] = {
3389 &nvme_dev_attrs_group,
3390 NULL,
3393 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3394 unsigned nsid)
3396 struct nvme_ns_head *h;
3398 lockdep_assert_held(&subsys->lock);
3400 list_for_each_entry(h, &subsys->nsheads, entry) {
3401 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3402 return h;
3405 return NULL;
3408 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3409 struct nvme_ns_head *new)
3411 struct nvme_ns_head *h;
3413 lockdep_assert_held(&subsys->lock);
3415 list_for_each_entry(h, &subsys->nsheads, entry) {
3416 if (nvme_ns_ids_valid(&new->ids) &&
3417 !list_empty(&h->list) &&
3418 nvme_ns_ids_equal(&new->ids, &h->ids))
3419 return -EINVAL;
3422 return 0;
3425 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3426 unsigned nsid, struct nvme_id_ns *id,
3427 struct nvme_ns_ids *ids)
3429 struct nvme_ns_head *head;
3430 size_t size = sizeof(*head);
3431 int ret = -ENOMEM;
3433 #ifdef CONFIG_NVME_MULTIPATH
3434 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3435 #endif
3437 head = kzalloc(size, GFP_KERNEL);
3438 if (!head)
3439 goto out;
3440 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3441 if (ret < 0)
3442 goto out_free_head;
3443 head->instance = ret;
3444 INIT_LIST_HEAD(&head->list);
3445 ret = init_srcu_struct(&head->srcu);
3446 if (ret)
3447 goto out_ida_remove;
3448 head->subsys = ctrl->subsys;
3449 head->ns_id = nsid;
3450 head->ids = *ids;
3451 kref_init(&head->ref);
3453 ret = __nvme_check_ids(ctrl->subsys, head);
3454 if (ret) {
3455 dev_err(ctrl->device,
3456 "duplicate IDs for nsid %d\n", nsid);
3457 goto out_cleanup_srcu;
3460 ret = nvme_mpath_alloc_disk(ctrl, head);
3461 if (ret)
3462 goto out_cleanup_srcu;
3464 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3466 kref_get(&ctrl->subsys->ref);
3468 return head;
3469 out_cleanup_srcu:
3470 cleanup_srcu_struct(&head->srcu);
3471 out_ida_remove:
3472 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3473 out_free_head:
3474 kfree(head);
3475 out:
3476 if (ret > 0)
3477 ret = blk_status_to_errno(nvme_error_status(ret));
3478 return ERR_PTR(ret);
3481 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3482 struct nvme_id_ns *id)
3484 struct nvme_ctrl *ctrl = ns->ctrl;
3485 bool is_shared = id->nmic & (1 << 0);
3486 struct nvme_ns_head *head = NULL;
3487 struct nvme_ns_ids ids;
3488 int ret = 0;
3490 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3491 if (ret)
3492 goto out;
3494 mutex_lock(&ctrl->subsys->lock);
3495 if (is_shared)
3496 head = nvme_find_ns_head(ctrl->subsys, nsid);
3497 if (!head) {
3498 head = nvme_alloc_ns_head(ctrl, nsid, id, &ids);
3499 if (IS_ERR(head)) {
3500 ret = PTR_ERR(head);
3501 goto out_unlock;
3503 } else {
3504 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3505 dev_err(ctrl->device,
3506 "IDs don't match for shared namespace %d\n",
3507 nsid);
3508 ret = -EINVAL;
3509 goto out_unlock;
3513 list_add_tail(&ns->siblings, &head->list);
3514 ns->head = head;
3516 out_unlock:
3517 mutex_unlock(&ctrl->subsys->lock);
3518 out:
3519 if (ret > 0)
3520 ret = blk_status_to_errno(nvme_error_status(ret));
3521 return ret;
3524 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3526 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3527 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3529 return nsa->head->ns_id - nsb->head->ns_id;
3532 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3534 struct nvme_ns *ns, *ret = NULL;
3536 down_read(&ctrl->namespaces_rwsem);
3537 list_for_each_entry(ns, &ctrl->namespaces, list) {
3538 if (ns->head->ns_id == nsid) {
3539 if (!kref_get_unless_zero(&ns->kref))
3540 continue;
3541 ret = ns;
3542 break;
3544 if (ns->head->ns_id > nsid)
3545 break;
3547 up_read(&ctrl->namespaces_rwsem);
3548 return ret;
3551 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3553 struct streams_directive_params s;
3554 int ret;
3556 if (!ctrl->nr_streams)
3557 return 0;
3559 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3560 if (ret)
3561 return ret;
3563 ns->sws = le32_to_cpu(s.sws);
3564 ns->sgs = le16_to_cpu(s.sgs);
3566 if (ns->sws) {
3567 unsigned int bs = 1 << ns->lba_shift;
3569 blk_queue_io_min(ns->queue, bs * ns->sws);
3570 if (ns->sgs)
3571 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3574 return 0;
3577 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3579 struct nvme_ns *ns;
3580 struct gendisk *disk;
3581 struct nvme_id_ns *id;
3582 char disk_name[DISK_NAME_LEN];
3583 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3585 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3586 if (!ns)
3587 return;
3589 ns->queue = blk_mq_init_queue(ctrl->tagset);
3590 if (IS_ERR(ns->queue))
3591 goto out_free_ns;
3593 if (ctrl->opts && ctrl->opts->data_digest)
3594 ns->queue->backing_dev_info->capabilities
3595 |= BDI_CAP_STABLE_WRITES;
3597 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3598 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3599 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3601 ns->queue->queuedata = ns;
3602 ns->ctrl = ctrl;
3604 kref_init(&ns->kref);
3605 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3607 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3608 nvme_set_queue_limits(ctrl, ns->queue);
3610 ret = nvme_identify_ns(ctrl, nsid, &id);
3611 if (ret)
3612 goto out_free_queue;
3614 if (id->ncap == 0) /* no namespace (legacy quirk) */
3615 goto out_free_id;
3617 ret = nvme_init_ns_head(ns, nsid, id);
3618 if (ret)
3619 goto out_free_id;
3620 nvme_setup_streams_ns(ctrl, ns);
3621 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3623 disk = alloc_disk_node(0, node);
3624 if (!disk)
3625 goto out_unlink_ns;
3627 disk->fops = &nvme_fops;
3628 disk->private_data = ns;
3629 disk->queue = ns->queue;
3630 disk->flags = flags;
3631 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3632 ns->disk = disk;
3634 __nvme_revalidate_disk(disk, id);
3636 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3637 ret = nvme_nvm_register(ns, disk_name, node);
3638 if (ret) {
3639 dev_warn(ctrl->device, "LightNVM init failure\n");
3640 goto out_put_disk;
3644 down_write(&ctrl->namespaces_rwsem);
3645 list_add_tail(&ns->list, &ctrl->namespaces);
3646 up_write(&ctrl->namespaces_rwsem);
3648 nvme_get_ctrl(ctrl);
3650 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3652 nvme_mpath_add_disk(ns, id);
3653 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3654 kfree(id);
3656 return;
3657 out_put_disk:
3658 /* prevent double queue cleanup */
3659 ns->disk->queue = NULL;
3660 put_disk(ns->disk);
3661 out_unlink_ns:
3662 mutex_lock(&ctrl->subsys->lock);
3663 list_del_rcu(&ns->siblings);
3664 mutex_unlock(&ctrl->subsys->lock);
3665 nvme_put_ns_head(ns->head);
3666 out_free_id:
3667 kfree(id);
3668 out_free_queue:
3669 blk_cleanup_queue(ns->queue);
3670 out_free_ns:
3671 kfree(ns);
3674 static void nvme_ns_remove(struct nvme_ns *ns)
3676 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3677 return;
3679 nvme_fault_inject_fini(&ns->fault_inject);
3681 mutex_lock(&ns->ctrl->subsys->lock);
3682 list_del_rcu(&ns->siblings);
3683 mutex_unlock(&ns->ctrl->subsys->lock);
3684 synchronize_rcu(); /* guarantee not available in head->list */
3685 nvme_mpath_clear_current_path(ns);
3686 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3688 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3689 del_gendisk(ns->disk);
3690 blk_cleanup_queue(ns->queue);
3691 if (blk_get_integrity(ns->disk))
3692 blk_integrity_unregister(ns->disk);
3695 down_write(&ns->ctrl->namespaces_rwsem);
3696 list_del_init(&ns->list);
3697 up_write(&ns->ctrl->namespaces_rwsem);
3699 nvme_mpath_check_last_path(ns);
3700 nvme_put_ns(ns);
3703 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3705 struct nvme_ns *ns;
3707 ns = nvme_find_get_ns(ctrl, nsid);
3708 if (ns) {
3709 if (ns->disk && revalidate_disk(ns->disk))
3710 nvme_ns_remove(ns);
3711 nvme_put_ns(ns);
3712 } else
3713 nvme_alloc_ns(ctrl, nsid);
3716 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3717 unsigned nsid)
3719 struct nvme_ns *ns, *next;
3720 LIST_HEAD(rm_list);
3722 down_write(&ctrl->namespaces_rwsem);
3723 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3724 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3725 list_move_tail(&ns->list, &rm_list);
3727 up_write(&ctrl->namespaces_rwsem);
3729 list_for_each_entry_safe(ns, next, &rm_list, list)
3730 nvme_ns_remove(ns);
3734 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3736 struct nvme_ns *ns;
3737 __le32 *ns_list;
3738 unsigned i, j, nsid, prev = 0;
3739 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3740 int ret = 0;
3742 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3743 if (!ns_list)
3744 return -ENOMEM;
3746 for (i = 0; i < num_lists; i++) {
3747 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3748 if (ret)
3749 goto free;
3751 for (j = 0; j < min(nn, 1024U); j++) {
3752 nsid = le32_to_cpu(ns_list[j]);
3753 if (!nsid)
3754 goto out;
3756 nvme_validate_ns(ctrl, nsid);
3758 while (++prev < nsid) {
3759 ns = nvme_find_get_ns(ctrl, prev);
3760 if (ns) {
3761 nvme_ns_remove(ns);
3762 nvme_put_ns(ns);
3766 nn -= j;
3768 out:
3769 nvme_remove_invalid_namespaces(ctrl, prev);
3770 free:
3771 kfree(ns_list);
3772 return ret;
3775 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3777 unsigned i;
3779 for (i = 1; i <= nn; i++)
3780 nvme_validate_ns(ctrl, i);
3782 nvme_remove_invalid_namespaces(ctrl, nn);
3785 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3787 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3788 __le32 *log;
3789 int error;
3791 log = kzalloc(log_size, GFP_KERNEL);
3792 if (!log)
3793 return;
3796 * We need to read the log to clear the AEN, but we don't want to rely
3797 * on it for the changed namespace information as userspace could have
3798 * raced with us in reading the log page, which could cause us to miss
3799 * updates.
3801 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3802 log_size, 0);
3803 if (error)
3804 dev_warn(ctrl->device,
3805 "reading changed ns log failed: %d\n", error);
3807 kfree(log);
3810 static void nvme_scan_work(struct work_struct *work)
3812 struct nvme_ctrl *ctrl =
3813 container_of(work, struct nvme_ctrl, scan_work);
3814 struct nvme_id_ctrl *id;
3815 unsigned nn;
3817 /* No tagset on a live ctrl means IO queues could not created */
3818 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3819 return;
3821 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3822 dev_info(ctrl->device, "rescanning namespaces.\n");
3823 nvme_clear_changed_ns_log(ctrl);
3826 if (nvme_identify_ctrl(ctrl, &id))
3827 return;
3829 mutex_lock(&ctrl->scan_lock);
3830 nn = le32_to_cpu(id->nn);
3831 if (!nvme_ctrl_limited_cns(ctrl)) {
3832 if (!nvme_scan_ns_list(ctrl, nn))
3833 goto out_free_id;
3835 nvme_scan_ns_sequential(ctrl, nn);
3836 out_free_id:
3837 mutex_unlock(&ctrl->scan_lock);
3838 kfree(id);
3839 down_write(&ctrl->namespaces_rwsem);
3840 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3841 up_write(&ctrl->namespaces_rwsem);
3845 * This function iterates the namespace list unlocked to allow recovery from
3846 * controller failure. It is up to the caller to ensure the namespace list is
3847 * not modified by scan work while this function is executing.
3849 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3851 struct nvme_ns *ns, *next;
3852 LIST_HEAD(ns_list);
3855 * make sure to requeue I/O to all namespaces as these
3856 * might result from the scan itself and must complete
3857 * for the scan_work to make progress
3859 nvme_mpath_clear_ctrl_paths(ctrl);
3861 /* prevent racing with ns scanning */
3862 flush_work(&ctrl->scan_work);
3865 * The dead states indicates the controller was not gracefully
3866 * disconnected. In that case, we won't be able to flush any data while
3867 * removing the namespaces' disks; fail all the queues now to avoid
3868 * potentially having to clean up the failed sync later.
3870 if (ctrl->state == NVME_CTRL_DEAD)
3871 nvme_kill_queues(ctrl);
3873 down_write(&ctrl->namespaces_rwsem);
3874 list_splice_init(&ctrl->namespaces, &ns_list);
3875 up_write(&ctrl->namespaces_rwsem);
3877 list_for_each_entry_safe(ns, next, &ns_list, list)
3878 nvme_ns_remove(ns);
3880 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3882 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3884 struct nvme_ctrl *ctrl =
3885 container_of(dev, struct nvme_ctrl, ctrl_device);
3886 struct nvmf_ctrl_options *opts = ctrl->opts;
3887 int ret;
3889 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3890 if (ret)
3891 return ret;
3893 if (opts) {
3894 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3895 if (ret)
3896 return ret;
3898 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3899 opts->trsvcid ?: "none");
3900 if (ret)
3901 return ret;
3903 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3904 opts->host_traddr ?: "none");
3906 return ret;
3909 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3911 char *envp[2] = { NULL, NULL };
3912 u32 aen_result = ctrl->aen_result;
3914 ctrl->aen_result = 0;
3915 if (!aen_result)
3916 return;
3918 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3919 if (!envp[0])
3920 return;
3921 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3922 kfree(envp[0]);
3925 static void nvme_async_event_work(struct work_struct *work)
3927 struct nvme_ctrl *ctrl =
3928 container_of(work, struct nvme_ctrl, async_event_work);
3930 nvme_aen_uevent(ctrl);
3931 ctrl->ops->submit_async_event(ctrl);
3934 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3937 u32 csts;
3939 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3940 return false;
3942 if (csts == ~0)
3943 return false;
3945 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3948 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3950 struct nvme_fw_slot_info_log *log;
3952 log = kmalloc(sizeof(*log), GFP_KERNEL);
3953 if (!log)
3954 return;
3956 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, log,
3957 sizeof(*log), 0))
3958 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3959 kfree(log);
3962 static void nvme_fw_act_work(struct work_struct *work)
3964 struct nvme_ctrl *ctrl = container_of(work,
3965 struct nvme_ctrl, fw_act_work);
3966 unsigned long fw_act_timeout;
3968 if (ctrl->mtfa)
3969 fw_act_timeout = jiffies +
3970 msecs_to_jiffies(ctrl->mtfa * 100);
3971 else
3972 fw_act_timeout = jiffies +
3973 msecs_to_jiffies(admin_timeout * 1000);
3975 nvme_stop_queues(ctrl);
3976 while (nvme_ctrl_pp_status(ctrl)) {
3977 if (time_after(jiffies, fw_act_timeout)) {
3978 dev_warn(ctrl->device,
3979 "Fw activation timeout, reset controller\n");
3980 nvme_try_sched_reset(ctrl);
3981 return;
3983 msleep(100);
3986 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
3987 return;
3989 nvme_start_queues(ctrl);
3990 /* read FW slot information to clear the AER */
3991 nvme_get_fw_slot_info(ctrl);
3994 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3996 u32 aer_notice_type = (result & 0xff00) >> 8;
3998 trace_nvme_async_event(ctrl, aer_notice_type);
4000 switch (aer_notice_type) {
4001 case NVME_AER_NOTICE_NS_CHANGED:
4002 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4003 nvme_queue_scan(ctrl);
4004 break;
4005 case NVME_AER_NOTICE_FW_ACT_STARTING:
4007 * We are (ab)using the RESETTING state to prevent subsequent
4008 * recovery actions from interfering with the controller's
4009 * firmware activation.
4011 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4012 queue_work(nvme_wq, &ctrl->fw_act_work);
4013 break;
4014 #ifdef CONFIG_NVME_MULTIPATH
4015 case NVME_AER_NOTICE_ANA:
4016 if (!ctrl->ana_log_buf)
4017 break;
4018 queue_work(nvme_wq, &ctrl->ana_work);
4019 break;
4020 #endif
4021 case NVME_AER_NOTICE_DISC_CHANGED:
4022 ctrl->aen_result = result;
4023 break;
4024 default:
4025 dev_warn(ctrl->device, "async event result %08x\n", result);
4029 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4030 volatile union nvme_result *res)
4032 u32 result = le32_to_cpu(res->u32);
4033 u32 aer_type = result & 0x07;
4035 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4036 return;
4038 switch (aer_type) {
4039 case NVME_AER_NOTICE:
4040 nvme_handle_aen_notice(ctrl, result);
4041 break;
4042 case NVME_AER_ERROR:
4043 case NVME_AER_SMART:
4044 case NVME_AER_CSS:
4045 case NVME_AER_VS:
4046 trace_nvme_async_event(ctrl, aer_type);
4047 ctrl->aen_result = result;
4048 break;
4049 default:
4050 break;
4052 queue_work(nvme_wq, &ctrl->async_event_work);
4054 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4056 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4058 nvme_mpath_stop(ctrl);
4059 nvme_stop_keep_alive(ctrl);
4060 flush_work(&ctrl->async_event_work);
4061 cancel_work_sync(&ctrl->fw_act_work);
4063 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4065 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4067 if (ctrl->kato)
4068 nvme_start_keep_alive(ctrl);
4070 nvme_enable_aen(ctrl);
4072 if (ctrl->queue_count > 1) {
4073 nvme_queue_scan(ctrl);
4074 nvme_start_queues(ctrl);
4076 ctrl->created = true;
4078 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4080 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4082 nvme_fault_inject_fini(&ctrl->fault_inject);
4083 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4084 cdev_device_del(&ctrl->cdev, ctrl->device);
4085 nvme_put_ctrl(ctrl);
4087 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4089 static void nvme_free_ctrl(struct device *dev)
4091 struct nvme_ctrl *ctrl =
4092 container_of(dev, struct nvme_ctrl, ctrl_device);
4093 struct nvme_subsystem *subsys = ctrl->subsys;
4095 if (subsys && ctrl->instance != subsys->instance)
4096 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4098 kfree(ctrl->effects);
4099 nvme_mpath_uninit(ctrl);
4100 __free_page(ctrl->discard_page);
4102 if (subsys) {
4103 mutex_lock(&nvme_subsystems_lock);
4104 list_del(&ctrl->subsys_entry);
4105 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4106 mutex_unlock(&nvme_subsystems_lock);
4109 ctrl->ops->free_ctrl(ctrl);
4111 if (subsys)
4112 nvme_put_subsystem(subsys);
4116 * Initialize a NVMe controller structures. This needs to be called during
4117 * earliest initialization so that we have the initialized structured around
4118 * during probing.
4120 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4121 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4123 int ret;
4125 ctrl->state = NVME_CTRL_NEW;
4126 spin_lock_init(&ctrl->lock);
4127 mutex_init(&ctrl->scan_lock);
4128 INIT_LIST_HEAD(&ctrl->namespaces);
4129 init_rwsem(&ctrl->namespaces_rwsem);
4130 ctrl->dev = dev;
4131 ctrl->ops = ops;
4132 ctrl->quirks = quirks;
4133 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4134 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4135 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4136 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4137 init_waitqueue_head(&ctrl->state_wq);
4139 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4140 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4141 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4143 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4144 PAGE_SIZE);
4145 ctrl->discard_page = alloc_page(GFP_KERNEL);
4146 if (!ctrl->discard_page) {
4147 ret = -ENOMEM;
4148 goto out;
4151 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4152 if (ret < 0)
4153 goto out;
4154 ctrl->instance = ret;
4156 device_initialize(&ctrl->ctrl_device);
4157 ctrl->device = &ctrl->ctrl_device;
4158 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4159 ctrl->device->class = nvme_class;
4160 ctrl->device->parent = ctrl->dev;
4161 ctrl->device->groups = nvme_dev_attr_groups;
4162 ctrl->device->release = nvme_free_ctrl;
4163 dev_set_drvdata(ctrl->device, ctrl);
4164 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4165 if (ret)
4166 goto out_release_instance;
4168 nvme_get_ctrl(ctrl);
4169 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4170 ctrl->cdev.owner = ops->module;
4171 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4172 if (ret)
4173 goto out_free_name;
4176 * Initialize latency tolerance controls. The sysfs files won't
4177 * be visible to userspace unless the device actually supports APST.
4179 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4180 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4181 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4183 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4185 return 0;
4186 out_free_name:
4187 nvme_put_ctrl(ctrl);
4188 kfree_const(ctrl->device->kobj.name);
4189 out_release_instance:
4190 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4191 out:
4192 if (ctrl->discard_page)
4193 __free_page(ctrl->discard_page);
4194 return ret;
4196 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4199 * nvme_kill_queues(): Ends all namespace queues
4200 * @ctrl: the dead controller that needs to end
4202 * Call this function when the driver determines it is unable to get the
4203 * controller in a state capable of servicing IO.
4205 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4207 struct nvme_ns *ns;
4209 down_read(&ctrl->namespaces_rwsem);
4211 /* Forcibly unquiesce queues to avoid blocking dispatch */
4212 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4213 blk_mq_unquiesce_queue(ctrl->admin_q);
4215 list_for_each_entry(ns, &ctrl->namespaces, list)
4216 nvme_set_queue_dying(ns);
4218 up_read(&ctrl->namespaces_rwsem);
4220 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4222 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4224 struct nvme_ns *ns;
4226 down_read(&ctrl->namespaces_rwsem);
4227 list_for_each_entry(ns, &ctrl->namespaces, list)
4228 blk_mq_unfreeze_queue(ns->queue);
4229 up_read(&ctrl->namespaces_rwsem);
4231 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4233 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4235 struct nvme_ns *ns;
4237 down_read(&ctrl->namespaces_rwsem);
4238 list_for_each_entry(ns, &ctrl->namespaces, list) {
4239 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4240 if (timeout <= 0)
4241 break;
4243 up_read(&ctrl->namespaces_rwsem);
4245 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4247 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4249 struct nvme_ns *ns;
4251 down_read(&ctrl->namespaces_rwsem);
4252 list_for_each_entry(ns, &ctrl->namespaces, list)
4253 blk_mq_freeze_queue_wait(ns->queue);
4254 up_read(&ctrl->namespaces_rwsem);
4256 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4258 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4260 struct nvme_ns *ns;
4262 down_read(&ctrl->namespaces_rwsem);
4263 list_for_each_entry(ns, &ctrl->namespaces, list)
4264 blk_freeze_queue_start(ns->queue);
4265 up_read(&ctrl->namespaces_rwsem);
4267 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4269 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4271 struct nvme_ns *ns;
4273 down_read(&ctrl->namespaces_rwsem);
4274 list_for_each_entry(ns, &ctrl->namespaces, list)
4275 blk_mq_quiesce_queue(ns->queue);
4276 up_read(&ctrl->namespaces_rwsem);
4278 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4280 void nvme_start_queues(struct nvme_ctrl *ctrl)
4282 struct nvme_ns *ns;
4284 down_read(&ctrl->namespaces_rwsem);
4285 list_for_each_entry(ns, &ctrl->namespaces, list)
4286 blk_mq_unquiesce_queue(ns->queue);
4287 up_read(&ctrl->namespaces_rwsem);
4289 EXPORT_SYMBOL_GPL(nvme_start_queues);
4292 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4294 struct nvme_ns *ns;
4296 down_read(&ctrl->namespaces_rwsem);
4297 list_for_each_entry(ns, &ctrl->namespaces, list)
4298 blk_sync_queue(ns->queue);
4299 up_read(&ctrl->namespaces_rwsem);
4301 if (ctrl->admin_q)
4302 blk_sync_queue(ctrl->admin_q);
4304 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4307 * Check we didn't inadvertently grow the command structure sizes:
4309 static inline void _nvme_check_size(void)
4311 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4312 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4313 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4314 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4315 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4316 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4317 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4318 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4319 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4320 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4321 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4322 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4323 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4324 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4325 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4326 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4327 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4331 static int __init nvme_core_init(void)
4333 int result = -ENOMEM;
4335 _nvme_check_size();
4337 nvme_wq = alloc_workqueue("nvme-wq",
4338 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4339 if (!nvme_wq)
4340 goto out;
4342 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4343 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4344 if (!nvme_reset_wq)
4345 goto destroy_wq;
4347 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4348 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4349 if (!nvme_delete_wq)
4350 goto destroy_reset_wq;
4352 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4353 if (result < 0)
4354 goto destroy_delete_wq;
4356 nvme_class = class_create(THIS_MODULE, "nvme");
4357 if (IS_ERR(nvme_class)) {
4358 result = PTR_ERR(nvme_class);
4359 goto unregister_chrdev;
4361 nvme_class->dev_uevent = nvme_class_uevent;
4363 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4364 if (IS_ERR(nvme_subsys_class)) {
4365 result = PTR_ERR(nvme_subsys_class);
4366 goto destroy_class;
4368 return 0;
4370 destroy_class:
4371 class_destroy(nvme_class);
4372 unregister_chrdev:
4373 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4374 destroy_delete_wq:
4375 destroy_workqueue(nvme_delete_wq);
4376 destroy_reset_wq:
4377 destroy_workqueue(nvme_reset_wq);
4378 destroy_wq:
4379 destroy_workqueue(nvme_wq);
4380 out:
4381 return result;
4384 static void __exit nvme_core_exit(void)
4386 class_destroy(nvme_subsys_class);
4387 class_destroy(nvme_class);
4388 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4389 destroy_workqueue(nvme_delete_wq);
4390 destroy_workqueue(nvme_reset_wq);
4391 destroy_workqueue(nvme_wq);
4392 ida_destroy(&nvme_instance_ida);
4395 MODULE_LICENSE("GPL");
4396 MODULE_VERSION("1.0");
4397 module_init(nvme_core_init);
4398 module_exit(nvme_core_exit);