?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.133
[linux/fpc-iii.git] / drivers / nvme / host / core.c
blob0d60f2f8f3eecf8f33e045e20a15204e92108ef0
1 /*
2 * NVM Express device driver
3 * Copyright (c) 2011-2014, Intel Corporation.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 */
14
15 #include <linux/blkdev.h>
16 #include <linux/blk-mq.h>
17 #include <linux/delay.h>
18 #include <linux/errno.h>
19 #include <linux/hdreg.h>
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/list_sort.h>
23 #include <linux/slab.h>
24 #include <linux/types.h>
25 #include <linux/pr.h>
26 #include <linux/ptrace.h>
27 #include <linux/nvme_ioctl.h>
28 #include <linux/t10-pi.h>
29 #include <linux/pm_qos.h>
30 #include <asm/unaligned.h>
31
32 #define CREATE_TRACE_POINTS
33 #include "trace.h"
34
35 #include "nvme.h"
36 #include "fabrics.h"
37
38 #define NVME_MINORS (1U << MINORBITS)
39
40 unsigned int admin_timeout = 60;
41 module_param(admin_timeout, uint, 0644);
42 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
43 EXPORT_SYMBOL_GPL(admin_timeout);
44
45 unsigned int nvme_io_timeout = 30;
46 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
47 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
48 EXPORT_SYMBOL_GPL(nvme_io_timeout);
49
50 static unsigned char shutdown_timeout = 5;
51 module_param(shutdown_timeout, byte, 0644);
52 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
53
54 static u8 nvme_max_retries = 5;
55 module_param_named(max_retries, nvme_max_retries, byte, 0644);
56 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
57
58 static unsigned long default_ps_max_latency_us = 100000;
59 module_param(default_ps_max_latency_us, ulong, 0644);
60 MODULE_PARM_DESC(default_ps_max_latency_us,
61 "max power saving latency for new devices; use PM QOS to change per device");
62
63 static bool force_apst;
64 module_param(force_apst, bool, 0644);
65 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
66
67 static bool streams;
68 module_param(streams, bool, 0644);
69 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
70
71 /*
72 * nvme_wq - hosts nvme related works that are not reset or delete
73 * nvme_reset_wq - hosts nvme reset works
74 * nvme_delete_wq - hosts nvme delete works
75 *
76 * nvme_wq will host works such are scan, aen handling, fw activation,
77 * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
78 * runs reset works which also flush works hosted on nvme_wq for
79 * serialization purposes. nvme_delete_wq host controller deletion
80 * works which flush reset works for serialization.
81 */
82 struct workqueue_struct *nvme_wq;
83 EXPORT_SYMBOL_GPL(nvme_wq);
84
85 struct workqueue_struct *nvme_reset_wq;
86 EXPORT_SYMBOL_GPL(nvme_reset_wq);
87
88 struct workqueue_struct *nvme_delete_wq;
89 EXPORT_SYMBOL_GPL(nvme_delete_wq);
90
91 static DEFINE_IDA(nvme_subsystems_ida);
92 static LIST_HEAD(nvme_subsystems);
93 static DEFINE_MUTEX(nvme_subsystems_lock);
94
95 static DEFINE_IDA(nvme_instance_ida);
96 static dev_t nvme_chr_devt;
97 static struct class *nvme_class;
98 static struct class *nvme_subsys_class;
99
100 static void nvme_ns_remove(struct nvme_ns *ns);
101 static int nvme_revalidate_disk(struct gendisk *disk);
102 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
103 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
104 unsigned nsid);
105
106 static void nvme_set_queue_dying(struct nvme_ns *ns)
107 {
108 /*
109 * Revalidating a dead namespace sets capacity to 0. This will end
110 * buffered writers dirtying pages that can't be synced.
111 */
112 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
113 return;
114 blk_set_queue_dying(ns->queue);
115 /* Forcibly unquiesce queues to avoid blocking dispatch */
116 blk_mq_unquiesce_queue(ns->queue);
117 /*
118 * Revalidate after unblocking dispatchers that may be holding bd_butex
119 */
120 revalidate_disk(ns->disk);
121 }
122
123 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
124 {
125 /*
126 * Only new queue scan work when admin and IO queues are both alive
127 */
128 if (ctrl->state == NVME_CTRL_LIVE)
129 queue_work(nvme_wq, &ctrl->scan_work);
130 }
131
132 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
133 {
134 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
135 return -EBUSY;
136 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
137 return -EBUSY;
138 return 0;
139 }
140 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
141
142 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
143 {
144 int ret;
145
146 ret = nvme_reset_ctrl(ctrl);
147 if (!ret) {
148 flush_work(&ctrl->reset_work);
149 if (ctrl->state != NVME_CTRL_LIVE &&
150 ctrl->state != NVME_CTRL_ADMIN_ONLY)
151 ret = -ENETRESET;
152 }
153
154 return ret;
155 }
156 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
157
158 static void nvme_delete_ctrl_work(struct work_struct *work)
159 {
160 struct nvme_ctrl *ctrl =
161 container_of(work, struct nvme_ctrl, delete_work);
162
163 dev_info(ctrl->device,
164 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
165
166 flush_work(&ctrl->reset_work);
167 nvme_stop_ctrl(ctrl);
168 nvme_remove_namespaces(ctrl);
169 ctrl->ops->delete_ctrl(ctrl);
170 nvme_uninit_ctrl(ctrl);
171 nvme_put_ctrl(ctrl);
172 }
173
174 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
175 {
176 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
177 return -EBUSY;
178 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
179 return -EBUSY;
180 return 0;
181 }
182 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
183
184 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
185 {
186 int ret = 0;
187
188 /*
189 * Keep a reference until the work is flushed since ->delete_ctrl
190 * can free the controller.
191 */
192 nvme_get_ctrl(ctrl);
193 ret = nvme_delete_ctrl(ctrl);
194 if (!ret)
195 flush_work(&ctrl->delete_work);
196 nvme_put_ctrl(ctrl);
197 return ret;
198 }
199 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
200
201 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
202 {
203 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
204 }
205
206 static blk_status_t nvme_error_status(struct request *req)
207 {
208 switch (nvme_req(req)->status & 0x7ff) {
209 case NVME_SC_SUCCESS:
210 return BLK_STS_OK;
211 case NVME_SC_CAP_EXCEEDED:
212 return BLK_STS_NOSPC;
213 case NVME_SC_LBA_RANGE:
214 return BLK_STS_TARGET;
215 case NVME_SC_BAD_ATTRIBUTES:
216 case NVME_SC_ONCS_NOT_SUPPORTED:
217 case NVME_SC_INVALID_OPCODE:
218 case NVME_SC_INVALID_FIELD:
219 case NVME_SC_INVALID_NS:
220 return BLK_STS_NOTSUPP;
221 case NVME_SC_WRITE_FAULT:
222 case NVME_SC_READ_ERROR:
223 case NVME_SC_UNWRITTEN_BLOCK:
224 case NVME_SC_ACCESS_DENIED:
225 case NVME_SC_READ_ONLY:
226 case NVME_SC_COMPARE_FAILED:
227 return BLK_STS_MEDIUM;
228 case NVME_SC_GUARD_CHECK:
229 case NVME_SC_APPTAG_CHECK:
230 case NVME_SC_REFTAG_CHECK:
231 case NVME_SC_INVALID_PI:
232 return BLK_STS_PROTECTION;
233 case NVME_SC_RESERVATION_CONFLICT:
234 return BLK_STS_NEXUS;
235 default:
236 return BLK_STS_IOERR;
237 }
238 }
239
240 static inline bool nvme_req_needs_retry(struct request *req)
241 {
242 if (blk_noretry_request(req))
243 return false;
244 if (nvme_req(req)->status & NVME_SC_DNR)
245 return false;
246 if (nvme_req(req)->retries >= nvme_max_retries)
247 return false;
248 return true;
249 }
250
251 void nvme_complete_rq(struct request *req)
252 {
253 blk_status_t status = nvme_error_status(req);
254
255 trace_nvme_complete_rq(req);
256
257 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
258 if ((req->cmd_flags & REQ_NVME_MPATH) &&
259 blk_path_error(status)) {
260 nvme_failover_req(req);
261 return;
262 }
263
264 if (!blk_queue_dying(req->q)) {
265 nvme_req(req)->retries++;
266 blk_mq_requeue_request(req, true);
267 return;
268 }
269 }
270 blk_mq_end_request(req, status);
271 }
272 EXPORT_SYMBOL_GPL(nvme_complete_rq);
273
274 void nvme_cancel_request(struct request *req, void *data, bool reserved)
275 {
276 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
277 "Cancelling I/O %d", req->tag);
278
279 nvme_req(req)->status = NVME_SC_ABORT_REQ;
280 blk_mq_complete_request(req);
281
282 }
283 EXPORT_SYMBOL_GPL(nvme_cancel_request);
284
285 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
286 enum nvme_ctrl_state new_state)
287 {
288 enum nvme_ctrl_state old_state;
289 unsigned long flags;
290 bool changed = false;
291
292 spin_lock_irqsave(&ctrl->lock, flags);
293
294 old_state = ctrl->state;
295 switch (new_state) {
296 case NVME_CTRL_ADMIN_ONLY:
297 switch (old_state) {
298 case NVME_CTRL_CONNECTING:
299 changed = true;
300 /* FALLTHRU */
301 default:
302 break;
303 }
304 break;
305 case NVME_CTRL_LIVE:
306 switch (old_state) {
307 case NVME_CTRL_NEW:
308 case NVME_CTRL_RESETTING:
309 case NVME_CTRL_CONNECTING:
310 changed = true;
311 /* FALLTHRU */
312 default:
313 break;
314 }
315 break;
316 case NVME_CTRL_RESETTING:
317 switch (old_state) {
318 case NVME_CTRL_NEW:
319 case NVME_CTRL_LIVE:
320 case NVME_CTRL_ADMIN_ONLY:
321 changed = true;
322 /* FALLTHRU */
323 default:
324 break;
325 }
326 break;
327 case NVME_CTRL_CONNECTING:
328 switch (old_state) {
329 case NVME_CTRL_NEW:
330 case NVME_CTRL_RESETTING:
331 changed = true;
332 /* FALLTHRU */
333 default:
334 break;
335 }
336 break;
337 case NVME_CTRL_DELETING:
338 switch (old_state) {
339 case NVME_CTRL_LIVE:
340 case NVME_CTRL_ADMIN_ONLY:
341 case NVME_CTRL_RESETTING:
342 case NVME_CTRL_CONNECTING:
343 changed = true;
344 /* FALLTHRU */
345 default:
346 break;
347 }
348 break;
349 case NVME_CTRL_DEAD:
350 switch (old_state) {
351 case NVME_CTRL_DELETING:
352 changed = true;
353 /* FALLTHRU */
354 default:
355 break;
356 }
357 break;
358 default:
359 break;
360 }
361
362 if (changed)
363 ctrl->state = new_state;
364
365 spin_unlock_irqrestore(&ctrl->lock, flags);
366 if (changed && ctrl->state == NVME_CTRL_LIVE)
367 nvme_kick_requeue_lists(ctrl);
368 return changed;
369 }
370 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
371
372 static void nvme_free_ns_head(struct kref *ref)
373 {
374 struct nvme_ns_head *head =
375 container_of(ref, struct nvme_ns_head, ref);
376
377 nvme_mpath_remove_disk(head);
378 ida_simple_remove(&head->subsys->ns_ida, head->instance);
379 list_del_init(&head->entry);
380 cleanup_srcu_struct_quiesced(&head->srcu);
381 nvme_put_subsystem(head->subsys);
382 kfree(head);
383 }
384
385 static void nvme_put_ns_head(struct nvme_ns_head *head)
386 {
387 kref_put(&head->ref, nvme_free_ns_head);
388 }
389
390 static void nvme_free_ns(struct kref *kref)
391 {
392 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
393
394 if (ns->ndev)
395 nvme_nvm_unregister(ns);
396
397 put_disk(ns->disk);
398 nvme_put_ns_head(ns->head);
399 nvme_put_ctrl(ns->ctrl);
400 kfree(ns);
401 }
402
403 static void nvme_put_ns(struct nvme_ns *ns)
404 {
405 kref_put(&ns->kref, nvme_free_ns);
406 }
407
408 static inline void nvme_clear_nvme_request(struct request *req)
409 {
410 if (!(req->rq_flags & RQF_DONTPREP)) {
411 nvme_req(req)->retries = 0;
412 nvme_req(req)->flags = 0;
413 req->rq_flags |= RQF_DONTPREP;
414 }
415 }
416
417 struct request *nvme_alloc_request(struct request_queue *q,
418 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
419 {
420 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
421 struct request *req;
422
423 if (qid == NVME_QID_ANY) {
424 req = blk_mq_alloc_request(q, op, flags);
425 } else {
426 req = blk_mq_alloc_request_hctx(q, op, flags,
427 qid ? qid - 1 : 0);
428 }
429 if (IS_ERR(req))
430 return req;
431
432 req->cmd_flags |= REQ_FAILFAST_DRIVER;
433 nvme_clear_nvme_request(req);
434 nvme_req(req)->cmd = cmd;
435
436 return req;
437 }
438 EXPORT_SYMBOL_GPL(nvme_alloc_request);
439
440 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
441 {
442 struct nvme_command c;
443
444 memset(&c, 0, sizeof(c));
445
446 c.directive.opcode = nvme_admin_directive_send;
447 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
448 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
449 c.directive.dtype = NVME_DIR_IDENTIFY;
450 c.directive.tdtype = NVME_DIR_STREAMS;
451 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
452
453 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
454 }
455
456 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
457 {
458 return nvme_toggle_streams(ctrl, false);
459 }
460
461 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
462 {
463 return nvme_toggle_streams(ctrl, true);
464 }
465
466 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
467 struct streams_directive_params *s, u32 nsid)
468 {
469 struct nvme_command c;
470
471 memset(&c, 0, sizeof(c));
472 memset(s, 0, sizeof(*s));
473
474 c.directive.opcode = nvme_admin_directive_recv;
475 c.directive.nsid = cpu_to_le32(nsid);
476 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
477 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
478 c.directive.dtype = NVME_DIR_STREAMS;
479
480 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
481 }
482
483 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
484 {
485 struct streams_directive_params s;
486 int ret;
487
488 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
489 return 0;
490 if (!streams)
491 return 0;
492
493 ret = nvme_enable_streams(ctrl);
494 if (ret)
495 return ret;
496
497 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
498 if (ret)
499 return ret;
500
501 ctrl->nssa = le16_to_cpu(s.nssa);
502 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
503 dev_info(ctrl->device, "too few streams (%u) available\n",
504 ctrl->nssa);
505 nvme_disable_streams(ctrl);
506 return 0;
507 }
508
509 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
510 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
511 return 0;
512 }
513
514 /*
515 * Check if 'req' has a write hint associated with it. If it does, assign
516 * a valid namespace stream to the write.
517 */
518 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
519 struct request *req, u16 *control,
520 u32 *dsmgmt)
521 {
522 enum rw_hint streamid = req->write_hint;
523
524 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
525 streamid = 0;
526 else {
527 streamid--;
528 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
529 return;
530
531 *control |= NVME_RW_DTYPE_STREAMS;
532 *dsmgmt |= streamid << 16;
533 }
534
535 if (streamid < ARRAY_SIZE(req->q->write_hints))
536 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
537 }
538
539 static inline void nvme_setup_flush(struct nvme_ns *ns,
540 struct nvme_command *cmnd)
541 {
542 memset(cmnd, 0, sizeof(*cmnd));
543 cmnd->common.opcode = nvme_cmd_flush;
544 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
545 }
546
547 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
548 struct nvme_command *cmnd)
549 {
550 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
551 struct nvme_dsm_range *range;
552 struct bio *bio;
553
554 /*
555 * Some devices do not consider the DSM 'Number of Ranges' field when
556 * determining how much data to DMA. Always allocate memory for maximum
557 * number of segments to prevent device reading beyond end of buffer.
558 */
559 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
560
561 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
562 if (!range) {
563 /*
564 * If we fail allocation our range, fallback to the controller
565 * discard page. If that's also busy, it's safe to return
566 * busy, as we know we can make progress once that's freed.
567 */
568 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
569 return BLK_STS_RESOURCE;
570
571 range = page_address(ns->ctrl->discard_page);
572 }
573
574 __rq_for_each_bio(bio, req) {
575 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
576 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
577
578 if (n < segments) {
579 range[n].cattr = cpu_to_le32(0);
580 range[n].nlb = cpu_to_le32(nlb);
581 range[n].slba = cpu_to_le64(slba);
582 }
583 n++;
584 }
585
586 if (WARN_ON_ONCE(n != segments)) {
587 if (virt_to_page(range) == ns->ctrl->discard_page)
588 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
589 else
590 kfree(range);
591 return BLK_STS_IOERR;
592 }
593
594 memset(cmnd, 0, sizeof(*cmnd));
595 cmnd->dsm.opcode = nvme_cmd_dsm;
596 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
597 cmnd->dsm.nr = cpu_to_le32(segments - 1);
598 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
599
600 req->special_vec.bv_page = virt_to_page(range);
601 req->special_vec.bv_offset = offset_in_page(range);
602 req->special_vec.bv_len = alloc_size;
603 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
604
605 return BLK_STS_OK;
606 }
607
608 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
609 struct request *req, struct nvme_command *cmnd)
610 {
611 struct nvme_ctrl *ctrl = ns->ctrl;
612 u16 control = 0;
613 u32 dsmgmt = 0;
614
615 if (req->cmd_flags & REQ_FUA)
616 control |= NVME_RW_FUA;
617 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
618 control |= NVME_RW_LR;
619
620 if (req->cmd_flags & REQ_RAHEAD)
621 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
622
623 memset(cmnd, 0, sizeof(*cmnd));
624 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
625 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
626 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
627 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
628
629 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
630 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
631
632 if (ns->ms) {
633 /*
634 * If formated with metadata, the block layer always provides a
635 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
636 * we enable the PRACT bit for protection information or set the
637 * namespace capacity to zero to prevent any I/O.
638 */
639 if (!blk_integrity_rq(req)) {
640 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
641 return BLK_STS_NOTSUPP;
642 control |= NVME_RW_PRINFO_PRACT;
643 } else if (req_op(req) == REQ_OP_WRITE) {
644 t10_pi_prepare(req, ns->pi_type);
645 }
646
647 switch (ns->pi_type) {
648 case NVME_NS_DPS_PI_TYPE3:
649 control |= NVME_RW_PRINFO_PRCHK_GUARD;
650 break;
651 case NVME_NS_DPS_PI_TYPE1:
652 case NVME_NS_DPS_PI_TYPE2:
653 control |= NVME_RW_PRINFO_PRCHK_GUARD |
654 NVME_RW_PRINFO_PRCHK_REF;
655 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
656 break;
657 }
658 }
659
660 cmnd->rw.control = cpu_to_le16(control);
661 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
662 return 0;
663 }
664
665 void nvme_cleanup_cmd(struct request *req)
666 {
667 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
668 nvme_req(req)->status == 0) {
669 struct nvme_ns *ns = req->rq_disk->private_data;
670
671 t10_pi_complete(req, ns->pi_type,
672 blk_rq_bytes(req) >> ns->lba_shift);
673 }
674 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
675 struct nvme_ns *ns = req->rq_disk->private_data;
676 struct page *page = req->special_vec.bv_page;
677
678 if (page == ns->ctrl->discard_page)
679 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
680 else
681 kfree(page_address(page) + req->special_vec.bv_offset);
682 }
683 }
684 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
685
686 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
687 struct nvme_command *cmd)
688 {
689 blk_status_t ret = BLK_STS_OK;
690
691 nvme_clear_nvme_request(req);
692
693 switch (req_op(req)) {
694 case REQ_OP_DRV_IN:
695 case REQ_OP_DRV_OUT:
696 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
697 break;
698 case REQ_OP_FLUSH:
699 nvme_setup_flush(ns, cmd);
700 break;
701 case REQ_OP_WRITE_ZEROES:
702 /* currently only aliased to deallocate for a few ctrls: */
703 case REQ_OP_DISCARD:
704 ret = nvme_setup_discard(ns, req, cmd);
705 break;
706 case REQ_OP_READ:
707 case REQ_OP_WRITE:
708 ret = nvme_setup_rw(ns, req, cmd);
709 break;
710 default:
711 WARN_ON_ONCE(1);
712 return BLK_STS_IOERR;
713 }
714
715 cmd->common.command_id = req->tag;
716 trace_nvme_setup_cmd(req, cmd);
717 return ret;
718 }
719 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
720
721 /*
722 * Returns 0 on success. If the result is negative, it's a Linux error code;
723 * if the result is positive, it's an NVM Express status code
724 */
725 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
726 union nvme_result *result, void *buffer, unsigned bufflen,
727 unsigned timeout, int qid, int at_head,
728 blk_mq_req_flags_t flags)
729 {
730 struct request *req;
731 int ret;
732
733 req = nvme_alloc_request(q, cmd, flags, qid);
734 if (IS_ERR(req))
735 return PTR_ERR(req);
736
737 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
738
739 if (buffer && bufflen) {
740 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
741 if (ret)
742 goto out;
743 }
744
745 blk_execute_rq(req->q, NULL, req, at_head);
746 if (result)
747 *result = nvme_req(req)->result;
748 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
749 ret = -EINTR;
750 else
751 ret = nvme_req(req)->status;
752 out:
753 blk_mq_free_request(req);
754 return ret;
755 }
756 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
757
758 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
759 void *buffer, unsigned bufflen)
760 {
761 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
762 NVME_QID_ANY, 0, 0);
763 }
764 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
765
766 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
767 unsigned len, u32 seed, bool write)
768 {
769 struct bio_integrity_payload *bip;
770 int ret = -ENOMEM;
771 void *buf;
772
773 buf = kmalloc(len, GFP_KERNEL);
774 if (!buf)
775 goto out;
776
777 ret = -EFAULT;
778 if (write && copy_from_user(buf, ubuf, len))
779 goto out_free_meta;
780
781 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
782 if (IS_ERR(bip)) {
783 ret = PTR_ERR(bip);
784 goto out_free_meta;
785 }
786
787 bip->bip_iter.bi_size = len;
788 bip->bip_iter.bi_sector = seed;
789 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
790 offset_in_page(buf));
791 if (ret == len)
792 return buf;
793 ret = -ENOMEM;
794 out_free_meta:
795 kfree(buf);
796 out:
797 return ERR_PTR(ret);
798 }
799
800 static int nvme_submit_user_cmd(struct request_queue *q,
801 struct nvme_command *cmd, void __user *ubuffer,
802 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
803 u32 meta_seed, u32 *result, unsigned timeout)
804 {
805 bool write = nvme_is_write(cmd);
806 struct nvme_ns *ns = q->queuedata;
807 struct gendisk *disk = ns ? ns->disk : NULL;
808 struct request *req;
809 struct bio *bio = NULL;
810 void *meta = NULL;
811 int ret;
812
813 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
814 if (IS_ERR(req))
815 return PTR_ERR(req);
816
817 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
818 nvme_req(req)->flags |= NVME_REQ_USERCMD;
819
820 if (ubuffer && bufflen) {
821 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
822 GFP_KERNEL);
823 if (ret)
824 goto out;
825 bio = req->bio;
826 bio->bi_disk = disk;
827 if (disk && meta_buffer && meta_len) {
828 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
829 meta_seed, write);
830 if (IS_ERR(meta)) {
831 ret = PTR_ERR(meta);
832 goto out_unmap;
833 }
834 req->cmd_flags |= REQ_INTEGRITY;
835 }
836 }
837
838 blk_execute_rq(req->q, disk, req, 0);
839 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
840 ret = -EINTR;
841 else
842 ret = nvme_req(req)->status;
843 if (result)
844 *result = le32_to_cpu(nvme_req(req)->result.u32);
845 if (meta && !ret && !write) {
846 if (copy_to_user(meta_buffer, meta, meta_len))
847 ret = -EFAULT;
848 }
849 kfree(meta);
850 out_unmap:
851 if (bio)
852 blk_rq_unmap_user(bio);
853 out:
854 blk_mq_free_request(req);
855 return ret;
856 }
857
858 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
859 {
860 struct nvme_ctrl *ctrl = rq->end_io_data;
861 unsigned long flags;
862 bool startka = false;
863
864 blk_mq_free_request(rq);
865
866 if (status) {
867 dev_err(ctrl->device,
868 "failed nvme_keep_alive_end_io error=%d\n",
869 status);
870 return;
871 }
872
873 spin_lock_irqsave(&ctrl->lock, flags);
874 if (ctrl->state == NVME_CTRL_LIVE ||
875 ctrl->state == NVME_CTRL_CONNECTING)
876 startka = true;
877 spin_unlock_irqrestore(&ctrl->lock, flags);
878 if (startka)
879 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
880 }
881
882 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
883 {
884 struct request *rq;
885
886 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
887 NVME_QID_ANY);
888 if (IS_ERR(rq))
889 return PTR_ERR(rq);
890
891 rq->timeout = ctrl->kato * HZ;
892 rq->end_io_data = ctrl;
893
894 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
895
896 return 0;
897 }
898
899 static void nvme_keep_alive_work(struct work_struct *work)
900 {
901 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
902 struct nvme_ctrl, ka_work);
903
904 if (nvme_keep_alive(ctrl)) {
905 /* allocation failure, reset the controller */
906 dev_err(ctrl->device, "keep-alive failed\n");
907 nvme_reset_ctrl(ctrl);
908 return;
909 }
910 }
911
912 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
913 {
914 if (unlikely(ctrl->kato == 0))
915 return;
916
917 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
918 }
919
920 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
921 {
922 if (unlikely(ctrl->kato == 0))
923 return;
924
925 cancel_delayed_work_sync(&ctrl->ka_work);
926 }
927 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
928
929 /*
930 * In NVMe 1.0 the CNS field was just a binary controller or namespace
931 * flag, thus sending any new CNS opcodes has a big chance of not working.
932 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
933 * (but not for any later version).
934 */
935 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
936 {
937 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
938 return ctrl->vs < NVME_VS(1, 2, 0);
939 return ctrl->vs < NVME_VS(1, 1, 0);
940 }
941
942 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
943 {
944 struct nvme_command c = { };
945 int error;
946
947 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
948 c.identify.opcode = nvme_admin_identify;
949 c.identify.cns = NVME_ID_CNS_CTRL;
950
951 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
952 if (!*id)
953 return -ENOMEM;
954
955 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
956 sizeof(struct nvme_id_ctrl));
957 if (error)
958 kfree(*id);
959 return error;
960 }
961
962 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
963 struct nvme_ns_ids *ids)
964 {
965 struct nvme_command c = { };
966 int status;
967 void *data;
968 int pos;
969 int len;
970
971 c.identify.opcode = nvme_admin_identify;
972 c.identify.nsid = cpu_to_le32(nsid);
973 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
974
975 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
976 if (!data)
977 return -ENOMEM;
978
979 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
980 NVME_IDENTIFY_DATA_SIZE);
981 if (status)
982 goto free_data;
983
984 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
985 struct nvme_ns_id_desc *cur = data + pos;
986
987 if (cur->nidl == 0)
988 break;
989
990 switch (cur->nidt) {
991 case NVME_NIDT_EUI64:
992 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
993 dev_warn(ctrl->device,
994 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
995 cur->nidl);
996 goto free_data;
997 }
998 len = NVME_NIDT_EUI64_LEN;
999 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1000 break;
1001 case NVME_NIDT_NGUID:
1002 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1003 dev_warn(ctrl->device,
1004 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1005 cur->nidl);
1006 goto free_data;
1007 }
1008 len = NVME_NIDT_NGUID_LEN;
1009 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1010 break;
1011 case NVME_NIDT_UUID:
1012 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1013 dev_warn(ctrl->device,
1014 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1015 cur->nidl);
1016 goto free_data;
1017 }
1018 len = NVME_NIDT_UUID_LEN;
1019 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1020 break;
1021 default:
1022 /* Skip unnkown types */
1023 len = cur->nidl;
1024 break;
1025 }
1026
1027 len += sizeof(*cur);
1028 }
1029 free_data:
1030 kfree(data);
1031 return status;
1032 }
1033
1034 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1035 {
1036 struct nvme_command c = { };
1037
1038 c.identify.opcode = nvme_admin_identify;
1039 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1040 c.identify.nsid = cpu_to_le32(nsid);
1041 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1042 NVME_IDENTIFY_DATA_SIZE);
1043 }
1044
1045 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
1046 unsigned nsid)
1047 {
1048 struct nvme_id_ns *id;
1049 struct nvme_command c = { };
1050 int error;
1051
1052 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1053 c.identify.opcode = nvme_admin_identify;
1054 c.identify.nsid = cpu_to_le32(nsid);
1055 c.identify.cns = NVME_ID_CNS_NS;
1056
1057 id = kmalloc(sizeof(*id), GFP_KERNEL);
1058 if (!id)
1059 return NULL;
1060
1061 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1062 if (error) {
1063 dev_warn(ctrl->device, "Identify namespace failed\n");
1064 kfree(id);
1065 return NULL;
1066 }
1067
1068 return id;
1069 }
1070
1071 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
1072 void *buffer, size_t buflen, u32 *result)
1073 {
1074 union nvme_result res = { 0 };
1075 struct nvme_command c;
1076 int ret;
1077
1078 memset(&c, 0, sizeof(c));
1079 c.features.opcode = nvme_admin_set_features;
1080 c.features.fid = cpu_to_le32(fid);
1081 c.features.dword11 = cpu_to_le32(dword11);
1082
1083 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1084 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1085 if (ret >= 0 && result)
1086 *result = le32_to_cpu(res.u32);
1087 return ret;
1088 }
1089
1090 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1091 {
1092 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1093 u32 result;
1094 int status, nr_io_queues;
1095
1096 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1097 &result);
1098 if (status < 0)
1099 return status;
1100
1101 /*
1102 * Degraded controllers might return an error when setting the queue
1103 * count. We still want to be able to bring them online and offer
1104 * access to the admin queue, as that might be only way to fix them up.
1105 */
1106 if (status > 0) {
1107 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1108 *count = 0;
1109 } else {
1110 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1111 *count = min(*count, nr_io_queues);
1112 }
1113
1114 return 0;
1115 }
1116 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1117
1118 #define NVME_AEN_SUPPORTED \
1119 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1120
1121 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1122 {
1123 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1124 int status;
1125
1126 if (!supported_aens)
1127 return;
1128
1129 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1130 NULL, 0, &result);
1131 if (status)
1132 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1133 supported_aens);
1134 }
1135
1136 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1137 {
1138 struct nvme_user_io io;
1139 struct nvme_command c;
1140 unsigned length, meta_len;
1141 void __user *metadata;
1142
1143 if (copy_from_user(&io, uio, sizeof(io)))
1144 return -EFAULT;
1145 if (io.flags)
1146 return -EINVAL;
1147
1148 switch (io.opcode) {
1149 case nvme_cmd_write:
1150 case nvme_cmd_read:
1151 case nvme_cmd_compare:
1152 break;
1153 default:
1154 return -EINVAL;
1155 }
1156
1157 length = (io.nblocks + 1) << ns->lba_shift;
1158 meta_len = (io.nblocks + 1) * ns->ms;
1159 metadata = (void __user *)(uintptr_t)io.metadata;
1160
1161 if (ns->ext) {
1162 length += meta_len;
1163 meta_len = 0;
1164 } else if (meta_len) {
1165 if ((io.metadata & 3) || !io.metadata)
1166 return -EINVAL;
1167 }
1168
1169 memset(&c, 0, sizeof(c));
1170 c.rw.opcode = io.opcode;
1171 c.rw.flags = io.flags;
1172 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1173 c.rw.slba = cpu_to_le64(io.slba);
1174 c.rw.length = cpu_to_le16(io.nblocks);
1175 c.rw.control = cpu_to_le16(io.control);
1176 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1177 c.rw.reftag = cpu_to_le32(io.reftag);
1178 c.rw.apptag = cpu_to_le16(io.apptag);
1179 c.rw.appmask = cpu_to_le16(io.appmask);
1180
1181 return nvme_submit_user_cmd(ns->queue, &c,
1182 (void __user *)(uintptr_t)io.addr, length,
1183 metadata, meta_len, io.slba, NULL, 0);
1184 }
1185
1186 static u32 nvme_known_admin_effects(u8 opcode)
1187 {
1188 switch (opcode) {
1189 case nvme_admin_format_nvm:
1190 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1191 NVME_CMD_EFFECTS_CSE_MASK;
1192 case nvme_admin_sanitize_nvm:
1193 return NVME_CMD_EFFECTS_CSE_MASK;
1194 default:
1195 break;
1196 }
1197 return 0;
1198 }
1199
1200 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1201 u8 opcode)
1202 {
1203 u32 effects = 0;
1204
1205 if (ns) {
1206 if (ctrl->effects)
1207 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1208 if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1209 dev_warn(ctrl->device,
1210 "IO command:%02x has unhandled effects:%08x\n",
1211 opcode, effects);
1212 return 0;
1213 }
1214
1215 if (ctrl->effects)
1216 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1217 else
1218 effects = nvme_known_admin_effects(opcode);
1219
1220 /*
1221 * For simplicity, IO to all namespaces is quiesced even if the command
1222 * effects say only one namespace is affected.
1223 */
1224 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1225 mutex_lock(&ctrl->scan_lock);
1226 mutex_lock(&ctrl->subsys->lock);
1227 nvme_mpath_start_freeze(ctrl->subsys);
1228 nvme_mpath_wait_freeze(ctrl->subsys);
1229 nvme_start_freeze(ctrl);
1230 nvme_wait_freeze(ctrl);
1231 }
1232 return effects;
1233 }
1234
1235 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1236 {
1237 struct nvme_ns *ns;
1238
1239 down_read(&ctrl->namespaces_rwsem);
1240 list_for_each_entry(ns, &ctrl->namespaces, list)
1241 if (ns->disk && nvme_revalidate_disk(ns->disk))
1242 nvme_set_queue_dying(ns);
1243 up_read(&ctrl->namespaces_rwsem);
1244
1245 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1246 }
1247
1248 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1249 {
1250 /*
1251 * Revalidate LBA changes prior to unfreezing. This is necessary to
1252 * prevent memory corruption if a logical block size was changed by
1253 * this command.
1254 */
1255 if (effects & NVME_CMD_EFFECTS_LBCC)
1256 nvme_update_formats(ctrl);
1257 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1258 nvme_unfreeze(ctrl);
1259 nvme_mpath_unfreeze(ctrl->subsys);
1260 mutex_unlock(&ctrl->subsys->lock);
1261 mutex_unlock(&ctrl->scan_lock);
1262 }
1263 if (effects & NVME_CMD_EFFECTS_CCC)
1264 nvme_init_identify(ctrl);
1265 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1266 nvme_queue_scan(ctrl);
1267 }
1268
1269 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1270 struct nvme_passthru_cmd __user *ucmd)
1271 {
1272 struct nvme_passthru_cmd cmd;
1273 struct nvme_command c;
1274 unsigned timeout = 0;
1275 u32 effects;
1276 int status;
1277
1278 if (!capable(CAP_SYS_ADMIN))
1279 return -EACCES;
1280 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1281 return -EFAULT;
1282 if (cmd.flags)
1283 return -EINVAL;
1284
1285 memset(&c, 0, sizeof(c));
1286 c.common.opcode = cmd.opcode;
1287 c.common.flags = cmd.flags;
1288 c.common.nsid = cpu_to_le32(cmd.nsid);
1289 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1290 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1291 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1292 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1293 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1294 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1295 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1296 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1297
1298 if (cmd.timeout_ms)
1299 timeout = msecs_to_jiffies(cmd.timeout_ms);
1300
1301 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1302 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1303 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1304 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1305 0, &cmd.result, timeout);
1306 nvme_passthru_end(ctrl, effects);
1307
1308 if (status >= 0) {
1309 if (put_user(cmd.result, &ucmd->result))
1310 return -EFAULT;
1311 }
1312
1313 return status;
1314 }
1315
1316 /*
1317 * Issue ioctl requests on the first available path. Note that unlike normal
1318 * block layer requests we will not retry failed request on another controller.
1319 */
1320 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1321 struct nvme_ns_head **head, int *srcu_idx)
1322 {
1323 #ifdef CONFIG_NVME_MULTIPATH
1324 if (disk->fops == &nvme_ns_head_ops) {
1325 struct nvme_ns *ns;
1326
1327 *head = disk->private_data;
1328 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1329 ns = nvme_find_path(*head);
1330 if (!ns)
1331 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1332 return ns;
1333 }
1334 #endif
1335 *head = NULL;
1336 *srcu_idx = -1;
1337 return disk->private_data;
1338 }
1339
1340 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1341 {
1342 if (head)
1343 srcu_read_unlock(&head->srcu, idx);
1344 }
1345
1346 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1347 unsigned int cmd, unsigned long arg)
1348 {
1349 struct nvme_ns_head *head = NULL;
1350 void __user *argp = (void __user *)arg;
1351 struct nvme_ns *ns;
1352 int srcu_idx, ret;
1353
1354 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1355 if (unlikely(!ns))
1356 return -EWOULDBLOCK;
1357
1358 /*
1359 * Handle ioctls that apply to the controller instead of the namespace
1360 * seperately and drop the ns SRCU reference early. This avoids a
1361 * deadlock when deleting namespaces using the passthrough interface.
1362 */
1363 if (cmd == NVME_IOCTL_ADMIN_CMD || is_sed_ioctl(cmd)) {
1364 struct nvme_ctrl *ctrl = ns->ctrl;
1365
1366 nvme_get_ctrl(ns->ctrl);
1367 nvme_put_ns_from_disk(head, srcu_idx);
1368
1369 if (cmd == NVME_IOCTL_ADMIN_CMD)
1370 ret = nvme_user_cmd(ctrl, NULL, argp);
1371 else
1372 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1373
1374 nvme_put_ctrl(ctrl);
1375 return ret;
1376 }
1377
1378 switch (cmd) {
1379 case NVME_IOCTL_ID:
1380 force_successful_syscall_return();
1381 ret = ns->head->ns_id;
1382 break;
1383 case NVME_IOCTL_IO_CMD:
1384 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1385 break;
1386 case NVME_IOCTL_SUBMIT_IO:
1387 ret = nvme_submit_io(ns, argp);
1388 break;
1389 default:
1390 if (ns->ndev)
1391 ret = nvme_nvm_ioctl(ns, cmd, arg);
1392 else
1393 ret = -ENOTTY;
1394 }
1395
1396 nvme_put_ns_from_disk(head, srcu_idx);
1397 return ret;
1398 }
1399
1400 static int nvme_open(struct block_device *bdev, fmode_t mode)
1401 {
1402 struct nvme_ns *ns = bdev->bd_disk->private_data;
1403
1404 #ifdef CONFIG_NVME_MULTIPATH
1405 /* should never be called due to GENHD_FL_HIDDEN */
1406 if (WARN_ON_ONCE(ns->head->disk))
1407 goto fail;
1408 #endif
1409 if (!kref_get_unless_zero(&ns->kref))
1410 goto fail;
1411 if (!try_module_get(ns->ctrl->ops->module))
1412 goto fail_put_ns;
1413
1414 return 0;
1415
1416 fail_put_ns:
1417 nvme_put_ns(ns);
1418 fail:
1419 return -ENXIO;
1420 }
1421
1422 static void nvme_release(struct gendisk *disk, fmode_t mode)
1423 {
1424 struct nvme_ns *ns = disk->private_data;
1425
1426 module_put(ns->ctrl->ops->module);
1427 nvme_put_ns(ns);
1428 }
1429
1430 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1431 {
1432 /* some standard values */
1433 geo->heads = 1 << 6;
1434 geo->sectors = 1 << 5;
1435 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1436 return 0;
1437 }
1438
1439 #ifdef CONFIG_BLK_DEV_INTEGRITY
1440 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1441 {
1442 struct blk_integrity integrity;
1443
1444 memset(&integrity, 0, sizeof(integrity));
1445 switch (pi_type) {
1446 case NVME_NS_DPS_PI_TYPE3:
1447 integrity.profile = &t10_pi_type3_crc;
1448 integrity.tag_size = sizeof(u16) + sizeof(u32);
1449 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1450 break;
1451 case NVME_NS_DPS_PI_TYPE1:
1452 case NVME_NS_DPS_PI_TYPE2:
1453 integrity.profile = &t10_pi_type1_crc;
1454 integrity.tag_size = sizeof(u16);
1455 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1456 break;
1457 default:
1458 integrity.profile = NULL;
1459 break;
1460 }
1461 integrity.tuple_size = ms;
1462 blk_integrity_register(disk, &integrity);
1463 blk_queue_max_integrity_segments(disk->queue, 1);
1464 }
1465 #else
1466 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1467 {
1468 }
1469 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1470
1471 static void nvme_set_chunk_size(struct nvme_ns *ns)
1472 {
1473 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1474 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1475 }
1476
1477 static void nvme_config_discard(struct nvme_ns *ns)
1478 {
1479 struct nvme_ctrl *ctrl = ns->ctrl;
1480 struct request_queue *queue = ns->queue;
1481 u32 size = queue_logical_block_size(queue);
1482
1483 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1484 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1485 return;
1486 }
1487
1488 if (ctrl->nr_streams && ns->sws && ns->sgs)
1489 size *= ns->sws * ns->sgs;
1490
1491 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1492 NVME_DSM_MAX_RANGES);
1493
1494 queue->limits.discard_alignment = 0;
1495 queue->limits.discard_granularity = size;
1496
1497 /* If discard is already enabled, don't reset queue limits */
1498 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1499 return;
1500
1501 blk_queue_max_discard_sectors(queue, UINT_MAX);
1502 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1503
1504 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1505 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1506 }
1507
1508 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1509 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1510 {
1511 memset(ids, 0, sizeof(*ids));
1512
1513 if (ctrl->vs >= NVME_VS(1, 1, 0))
1514 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1515 if (ctrl->vs >= NVME_VS(1, 2, 0))
1516 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1517 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1518 /* Don't treat error as fatal we potentially
1519 * already have a NGUID or EUI-64
1520 */
1521 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1522 dev_warn(ctrl->device,
1523 "%s: Identify Descriptors failed\n", __func__);
1524 }
1525 }
1526
1527 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1528 {
1529 return !uuid_is_null(&ids->uuid) ||
1530 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1531 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1532 }
1533
1534 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1535 {
1536 return uuid_equal(&a->uuid, &b->uuid) &&
1537 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1538 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1539 }
1540
1541 static void nvme_update_disk_info(struct gendisk *disk,
1542 struct nvme_ns *ns, struct nvme_id_ns *id)
1543 {
1544 sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1545 unsigned short bs = 1 << ns->lba_shift;
1546
1547 if (ns->lba_shift > PAGE_SHIFT) {
1548 /* unsupported block size, set capacity to 0 later */
1549 bs = (1 << 9);
1550 }
1551 blk_mq_freeze_queue(disk->queue);
1552 blk_integrity_unregister(disk);
1553
1554 blk_queue_logical_block_size(disk->queue, bs);
1555 blk_queue_physical_block_size(disk->queue, bs);
1556 blk_queue_io_min(disk->queue, bs);
1557
1558 if (ns->ms && !ns->ext &&
1559 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1560 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1561 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1562 ns->lba_shift > PAGE_SHIFT)
1563 capacity = 0;
1564
1565 set_capacity(disk, capacity);
1566 nvme_config_discard(ns);
1567
1568 if (id->nsattr & (1 << 0))
1569 set_disk_ro(disk, true);
1570 else
1571 set_disk_ro(disk, false);
1572
1573 blk_mq_unfreeze_queue(disk->queue);
1574 }
1575
1576 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1577 {
1578 struct nvme_ns *ns = disk->private_data;
1579
1580 /*
1581 * If identify namespace failed, use default 512 byte block size so
1582 * block layer can use before failing read/write for 0 capacity.
1583 */
1584 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1585 if (ns->lba_shift == 0)
1586 ns->lba_shift = 9;
1587 ns->noiob = le16_to_cpu(id->noiob);
1588 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1589 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1590 /* the PI implementation requires metadata equal t10 pi tuple size */
1591 if (ns->ms == sizeof(struct t10_pi_tuple))
1592 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1593 else
1594 ns->pi_type = 0;
1595
1596 if (ns->noiob)
1597 nvme_set_chunk_size(ns);
1598 nvme_update_disk_info(disk, ns, id);
1599 if (ns->ndev)
1600 nvme_nvm_update_nvm_info(ns);
1601 #ifdef CONFIG_NVME_MULTIPATH
1602 if (ns->head->disk) {
1603 nvme_update_disk_info(ns->head->disk, ns, id);
1604 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1605 revalidate_disk(ns->head->disk);
1606 }
1607 #endif
1608 }
1609
1610 static int nvme_revalidate_disk(struct gendisk *disk)
1611 {
1612 struct nvme_ns *ns = disk->private_data;
1613 struct nvme_ctrl *ctrl = ns->ctrl;
1614 struct nvme_id_ns *id;
1615 struct nvme_ns_ids ids;
1616 int ret = 0;
1617
1618 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1619 set_capacity(disk, 0);
1620 return -ENODEV;
1621 }
1622
1623 id = nvme_identify_ns(ctrl, ns->head->ns_id);
1624 if (!id)
1625 return -ENODEV;
1626
1627 if (id->ncap == 0) {
1628 ret = -ENODEV;
1629 goto out;
1630 }
1631
1632 __nvme_revalidate_disk(disk, id);
1633 nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1634 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1635 dev_err(ctrl->device,
1636 "identifiers changed for nsid %d\n", ns->head->ns_id);
1637 ret = -ENODEV;
1638 }
1639
1640 out:
1641 kfree(id);
1642 return ret;
1643 }
1644
1645 static char nvme_pr_type(enum pr_type type)
1646 {
1647 switch (type) {
1648 case PR_WRITE_EXCLUSIVE:
1649 return 1;
1650 case PR_EXCLUSIVE_ACCESS:
1651 return 2;
1652 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1653 return 3;
1654 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1655 return 4;
1656 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1657 return 5;
1658 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1659 return 6;
1660 default:
1661 return 0;
1662 }
1663 };
1664
1665 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1666 u64 key, u64 sa_key, u8 op)
1667 {
1668 struct nvme_ns_head *head = NULL;
1669 struct nvme_ns *ns;
1670 struct nvme_command c;
1671 int srcu_idx, ret;
1672 u8 data[16] = { 0, };
1673
1674 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1675 if (unlikely(!ns))
1676 return -EWOULDBLOCK;
1677
1678 put_unaligned_le64(key, &data[0]);
1679 put_unaligned_le64(sa_key, &data[8]);
1680
1681 memset(&c, 0, sizeof(c));
1682 c.common.opcode = op;
1683 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1684 c.common.cdw10[0] = cpu_to_le32(cdw10);
1685
1686 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1687 nvme_put_ns_from_disk(head, srcu_idx);
1688 return ret;
1689 }
1690
1691 static int nvme_pr_register(struct block_device *bdev, u64 old,
1692 u64 new, unsigned flags)
1693 {
1694 u32 cdw10;
1695
1696 if (flags & ~PR_FL_IGNORE_KEY)
1697 return -EOPNOTSUPP;
1698
1699 cdw10 = old ? 2 : 0;
1700 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1701 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1702 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1703 }
1704
1705 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1706 enum pr_type type, unsigned flags)
1707 {
1708 u32 cdw10;
1709
1710 if (flags & ~PR_FL_IGNORE_KEY)
1711 return -EOPNOTSUPP;
1712
1713 cdw10 = nvme_pr_type(type) << 8;
1714 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1715 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1716 }
1717
1718 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1719 enum pr_type type, bool abort)
1720 {
1721 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1722 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1723 }
1724
1725 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1726 {
1727 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1728 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1729 }
1730
1731 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1732 {
1733 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1734 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1735 }
1736
1737 static const struct pr_ops nvme_pr_ops = {
1738 .pr_register = nvme_pr_register,
1739 .pr_reserve = nvme_pr_reserve,
1740 .pr_release = nvme_pr_release,
1741 .pr_preempt = nvme_pr_preempt,
1742 .pr_clear = nvme_pr_clear,
1743 };
1744
1745 #ifdef CONFIG_BLK_SED_OPAL
1746 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1747 bool send)
1748 {
1749 struct nvme_ctrl *ctrl = data;
1750 struct nvme_command cmd;
1751
1752 memset(&cmd, 0, sizeof(cmd));
1753 if (send)
1754 cmd.common.opcode = nvme_admin_security_send;
1755 else
1756 cmd.common.opcode = nvme_admin_security_recv;
1757 cmd.common.nsid = 0;
1758 cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1759 cmd.common.cdw10[1] = cpu_to_le32(len);
1760
1761 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1762 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1763 }
1764 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1765 #endif /* CONFIG_BLK_SED_OPAL */
1766
1767 static const struct block_device_operations nvme_fops = {
1768 .owner = THIS_MODULE,
1769 .ioctl = nvme_ioctl,
1770 .compat_ioctl = nvme_ioctl,
1771 .open = nvme_open,
1772 .release = nvme_release,
1773 .getgeo = nvme_getgeo,
1774 .revalidate_disk= nvme_revalidate_disk,
1775 .pr_ops = &nvme_pr_ops,
1776 };
1777
1778 #ifdef CONFIG_NVME_MULTIPATH
1779 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1780 {
1781 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1782
1783 if (!kref_get_unless_zero(&head->ref))
1784 return -ENXIO;
1785 return 0;
1786 }
1787
1788 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1789 {
1790 nvme_put_ns_head(disk->private_data);
1791 }
1792
1793 const struct block_device_operations nvme_ns_head_ops = {
1794 .owner = THIS_MODULE,
1795 .open = nvme_ns_head_open,
1796 .release = nvme_ns_head_release,
1797 .ioctl = nvme_ioctl,
1798 .compat_ioctl = nvme_ioctl,
1799 .getgeo = nvme_getgeo,
1800 .pr_ops = &nvme_pr_ops,
1801 };
1802 #endif /* CONFIG_NVME_MULTIPATH */
1803
1804 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1805 {
1806 unsigned long timeout =
1807 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1808 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1809 int ret;
1810
1811 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1812 if (csts == ~0)
1813 return -ENODEV;
1814 if ((csts & NVME_CSTS_RDY) == bit)
1815 break;
1816
1817 msleep(100);
1818 if (fatal_signal_pending(current))
1819 return -EINTR;
1820 if (time_after(jiffies, timeout)) {
1821 dev_err(ctrl->device,
1822 "Device not ready; aborting %s\n", enabled ?
1823 "initialisation" : "reset");
1824 return -ENODEV;
1825 }
1826 }
1827
1828 return ret;
1829 }
1830
1831 /*
1832 * If the device has been passed off to us in an enabled state, just clear
1833 * the enabled bit. The spec says we should set the 'shutdown notification
1834 * bits', but doing so may cause the device to complete commands to the
1835 * admin queue ... and we don't know what memory that might be pointing at!
1836 */
1837 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1838 {
1839 int ret;
1840
1841 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1842 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1843
1844 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1845 if (ret)
1846 return ret;
1847
1848 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1849 msleep(NVME_QUIRK_DELAY_AMOUNT);
1850
1851 return nvme_wait_ready(ctrl, cap, false);
1852 }
1853 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1854
1855 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1856 {
1857 /*
1858 * Default to a 4K page size, with the intention to update this
1859 * path in the future to accomodate architectures with differing
1860 * kernel and IO page sizes.
1861 */
1862 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1863 int ret;
1864
1865 if (page_shift < dev_page_min) {
1866 dev_err(ctrl->device,
1867 "Minimum device page size %u too large for host (%u)\n",
1868 1 << dev_page_min, 1 << page_shift);
1869 return -ENODEV;
1870 }
1871
1872 ctrl->page_size = 1 << page_shift;
1873
1874 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1875 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1876 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1877 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1878 ctrl->ctrl_config |= NVME_CC_ENABLE;
1879
1880 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1881 if (ret)
1882 return ret;
1883 return nvme_wait_ready(ctrl, cap, true);
1884 }
1885 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1886
1887 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1888 {
1889 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1890 u32 csts;
1891 int ret;
1892
1893 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1894 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1895
1896 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1897 if (ret)
1898 return ret;
1899
1900 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1901 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1902 break;
1903
1904 msleep(100);
1905 if (fatal_signal_pending(current))
1906 return -EINTR;
1907 if (time_after(jiffies, timeout)) {
1908 dev_err(ctrl->device,
1909 "Device shutdown incomplete; abort shutdown\n");
1910 return -ENODEV;
1911 }
1912 }
1913
1914 return ret;
1915 }
1916 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1917
1918 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1919 struct request_queue *q)
1920 {
1921 bool vwc = false;
1922
1923 if (ctrl->max_hw_sectors) {
1924 u32 max_segments =
1925 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1926
1927 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1928 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1929 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1930 }
1931 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1932 is_power_of_2(ctrl->max_hw_sectors))
1933 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1934 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1935 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1936 vwc = true;
1937 blk_queue_write_cache(q, vwc, vwc);
1938 }
1939
1940 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1941 {
1942 __le64 ts;
1943 int ret;
1944
1945 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1946 return 0;
1947
1948 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1949 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1950 NULL);
1951 if (ret)
1952 dev_warn_once(ctrl->device,
1953 "could not set timestamp (%d)\n", ret);
1954 return ret;
1955 }
1956
1957 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1958 {
1959 /*
1960 * APST (Autonomous Power State Transition) lets us program a
1961 * table of power state transitions that the controller will
1962 * perform automatically. We configure it with a simple
1963 * heuristic: we are willing to spend at most 2% of the time
1964 * transitioning between power states. Therefore, when running
1965 * in any given state, we will enter the next lower-power
1966 * non-operational state after waiting 50 * (enlat + exlat)
1967 * microseconds, as long as that state's exit latency is under
1968 * the requested maximum latency.
1969 *
1970 * We will not autonomously enter any non-operational state for
1971 * which the total latency exceeds ps_max_latency_us. Users
1972 * can set ps_max_latency_us to zero to turn off APST.
1973 */
1974
1975 unsigned apste;
1976 struct nvme_feat_auto_pst *table;
1977 u64 max_lat_us = 0;
1978 int max_ps = -1;
1979 int ret;
1980
1981 /*
1982 * If APST isn't supported or if we haven't been initialized yet,
1983 * then don't do anything.
1984 */
1985 if (!ctrl->apsta)
1986 return 0;
1987
1988 if (ctrl->npss > 31) {
1989 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1990 return 0;
1991 }
1992
1993 table = kzalloc(sizeof(*table), GFP_KERNEL);
1994 if (!table)
1995 return 0;
1996
1997 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1998 /* Turn off APST. */
1999 apste = 0;
2000 dev_dbg(ctrl->device, "APST disabled\n");
2001 } else {
2002 __le64 target = cpu_to_le64(0);
2003 int state;
2004
2005 /*
2006 * Walk through all states from lowest- to highest-power.
2007 * According to the spec, lower-numbered states use more
2008 * power. NPSS, despite the name, is the index of the
2009 * lowest-power state, not the number of states.
2010 */
2011 for (state = (int)ctrl->npss; state >= 0; state--) {
2012 u64 total_latency_us, exit_latency_us, transition_ms;
2013
2014 if (target)
2015 table->entries[state] = target;
2016
2017 /*
2018 * Don't allow transitions to the deepest state
2019 * if it's quirked off.
2020 */
2021 if (state == ctrl->npss &&
2022 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2023 continue;
2024
2025 /*
2026 * Is this state a useful non-operational state for
2027 * higher-power states to autonomously transition to?
2028 */
2029 if (!(ctrl->psd[state].flags &
2030 NVME_PS_FLAGS_NON_OP_STATE))
2031 continue;
2032
2033 exit_latency_us =
2034 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2035 if (exit_latency_us > ctrl->ps_max_latency_us)
2036 continue;
2037
2038 total_latency_us =
2039 exit_latency_us +
2040 le32_to_cpu(ctrl->psd[state].entry_lat);
2041
2042 /*
2043 * This state is good. Use it as the APST idle
2044 * target for higher power states.
2045 */
2046 transition_ms = total_latency_us + 19;
2047 do_div(transition_ms, 20);
2048 if (transition_ms > (1 << 24) - 1)
2049 transition_ms = (1 << 24) - 1;
2050
2051 target = cpu_to_le64((state << 3) |
2052 (transition_ms << 8));
2053
2054 if (max_ps == -1)
2055 max_ps = state;
2056
2057 if (total_latency_us > max_lat_us)
2058 max_lat_us = total_latency_us;
2059 }
2060
2061 apste = 1;
2062
2063 if (max_ps == -1) {
2064 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2065 } else {
2066 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2067 max_ps, max_lat_us, (int)sizeof(*table), table);
2068 }
2069 }
2070
2071 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2072 table, sizeof(*table), NULL);
2073 if (ret)
2074 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2075
2076 kfree(table);
2077 return ret;
2078 }
2079
2080 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2081 {
2082 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2083 u64 latency;
2084
2085 switch (val) {
2086 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2087 case PM_QOS_LATENCY_ANY:
2088 latency = U64_MAX;
2089 break;
2090
2091 default:
2092 latency = val;
2093 }
2094
2095 if (ctrl->ps_max_latency_us != latency) {
2096 ctrl->ps_max_latency_us = latency;
2097 nvme_configure_apst(ctrl);
2098 }
2099 }
2100
2101 struct nvme_core_quirk_entry {
2102 /*
2103 * NVMe model and firmware strings are padded with spaces. For
2104 * simplicity, strings in the quirk table are padded with NULLs
2105 * instead.
2106 */
2107 u16 vid;
2108 const char *mn;
2109 const char *fr;
2110 unsigned long quirks;
2111 };
2112
2113 static const struct nvme_core_quirk_entry core_quirks[] = {
2114 {
2115 /*
2116 * This Toshiba device seems to die using any APST states. See:
2117 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2118 */
2119 .vid = 0x1179,
2120 .mn = "THNSF5256GPUK TOSHIBA",
2121 .quirks = NVME_QUIRK_NO_APST,
2122 }
2123 };
2124
2125 /* match is null-terminated but idstr is space-padded. */
2126 static bool string_matches(const char *idstr, const char *match, size_t len)
2127 {
2128 size_t matchlen;
2129
2130 if (!match)
2131 return true;
2132
2133 matchlen = strlen(match);
2134 WARN_ON_ONCE(matchlen > len);
2135
2136 if (memcmp(idstr, match, matchlen))
2137 return false;
2138
2139 for (; matchlen < len; matchlen++)
2140 if (idstr[matchlen] != ' ')
2141 return false;
2142
2143 return true;
2144 }
2145
2146 static bool quirk_matches(const struct nvme_id_ctrl *id,
2147 const struct nvme_core_quirk_entry *q)
2148 {
2149 return q->vid == le16_to_cpu(id->vid) &&
2150 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2151 string_matches(id->fr, q->fr, sizeof(id->fr));
2152 }
2153
2154 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2155 struct nvme_id_ctrl *id)
2156 {
2157 size_t nqnlen;
2158 int off;
2159
2160 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2161 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2162 strncpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2163 return;
2164 }
2165
2166 if (ctrl->vs >= NVME_VS(1, 2, 1))
2167 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2168
2169 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2170 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2171 "nqn.2014.08.org.nvmexpress:%04x%04x",
2172 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2173 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2174 off += sizeof(id->sn);
2175 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2176 off += sizeof(id->mn);
2177 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2178 }
2179
2180 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2181 {
2182 ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2183 kfree(subsys);
2184 }
2185
2186 static void nvme_release_subsystem(struct device *dev)
2187 {
2188 __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2189 }
2190
2191 static void nvme_destroy_subsystem(struct kref *ref)
2192 {
2193 struct nvme_subsystem *subsys =
2194 container_of(ref, struct nvme_subsystem, ref);
2195
2196 mutex_lock(&nvme_subsystems_lock);
2197 list_del(&subsys->entry);
2198 mutex_unlock(&nvme_subsystems_lock);
2199
2200 ida_destroy(&subsys->ns_ida);
2201 device_del(&subsys->dev);
2202 put_device(&subsys->dev);
2203 }
2204
2205 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2206 {
2207 kref_put(&subsys->ref, nvme_destroy_subsystem);
2208 }
2209
2210 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2211 {
2212 struct nvme_subsystem *subsys;
2213
2214 lockdep_assert_held(&nvme_subsystems_lock);
2215
2216 /*
2217 * Fail matches for discovery subsystems. This results
2218 * in each discovery controller bound to a unique subsystem.
2219 * This avoids issues with validating controller values
2220 * that can only be true when there is a single unique subsystem.
2221 * There may be multiple and completely independent entities
2222 * that provide discovery controllers.
2223 */
2224 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2225 return NULL;
2226
2227 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2228 if (strcmp(subsys->subnqn, subsysnqn))
2229 continue;
2230 if (!kref_get_unless_zero(&subsys->ref))
2231 continue;
2232 return subsys;
2233 }
2234
2235 return NULL;
2236 }
2237
2238 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2239 struct device_attribute subsys_attr_##_name = \
2240 __ATTR(_name, _mode, _show, NULL)
2241
2242 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2243 struct device_attribute *attr,
2244 char *buf)
2245 {
2246 struct nvme_subsystem *subsys =
2247 container_of(dev, struct nvme_subsystem, dev);
2248
2249 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2250 }
2251 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2252
2253 #define nvme_subsys_show_str_function(field) \
2254 static ssize_t subsys_##field##_show(struct device *dev, \
2255 struct device_attribute *attr, char *buf) \
2256 { \
2257 struct nvme_subsystem *subsys = \
2258 container_of(dev, struct nvme_subsystem, dev); \
2259 return sprintf(buf, "%.*s\n", \
2260 (int)sizeof(subsys->field), subsys->field); \
2261 } \
2262 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2263
2264 nvme_subsys_show_str_function(model);
2265 nvme_subsys_show_str_function(serial);
2266 nvme_subsys_show_str_function(firmware_rev);
2267
2268 static struct attribute *nvme_subsys_attrs[] = {
2269 &subsys_attr_model.attr,
2270 &subsys_attr_serial.attr,
2271 &subsys_attr_firmware_rev.attr,
2272 &subsys_attr_subsysnqn.attr,
2273 NULL,
2274 };
2275
2276 static struct attribute_group nvme_subsys_attrs_group = {
2277 .attrs = nvme_subsys_attrs,
2278 };
2279
2280 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2281 &nvme_subsys_attrs_group,
2282 NULL,
2283 };
2284
2285 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2286 {
2287 int count = 0;
2288 struct nvme_ctrl *ctrl;
2289
2290 mutex_lock(&subsys->lock);
2291 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2292 if (ctrl->state != NVME_CTRL_DELETING &&
2293 ctrl->state != NVME_CTRL_DEAD)
2294 count++;
2295 }
2296 mutex_unlock(&subsys->lock);
2297
2298 return count;
2299 }
2300
2301 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2302 {
2303 struct nvme_subsystem *subsys, *found;
2304 int ret;
2305
2306 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2307 if (!subsys)
2308 return -ENOMEM;
2309 ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2310 if (ret < 0) {
2311 kfree(subsys);
2312 return ret;
2313 }
2314 subsys->instance = ret;
2315 mutex_init(&subsys->lock);
2316 kref_init(&subsys->ref);
2317 INIT_LIST_HEAD(&subsys->ctrls);
2318 INIT_LIST_HEAD(&subsys->nsheads);
2319 nvme_init_subnqn(subsys, ctrl, id);
2320 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2321 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2322 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2323 subsys->vendor_id = le16_to_cpu(id->vid);
2324 subsys->cmic = id->cmic;
2325
2326 subsys->dev.class = nvme_subsys_class;
2327 subsys->dev.release = nvme_release_subsystem;
2328 subsys->dev.groups = nvme_subsys_attrs_groups;
2329 dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2330 device_initialize(&subsys->dev);
2331
2332 mutex_lock(&nvme_subsystems_lock);
2333 found = __nvme_find_get_subsystem(subsys->subnqn);
2334 if (found) {
2335 /*
2336 * Verify that the subsystem actually supports multiple
2337 * controllers, else bail out.
2338 */
2339 if (!(ctrl->opts && ctrl->opts->discovery_nqn) &&
2340 nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2341 dev_err(ctrl->device,
2342 "ignoring ctrl due to duplicate subnqn (%s).\n",
2343 found->subnqn);
2344 nvme_put_subsystem(found);
2345 ret = -EINVAL;
2346 goto out_unlock;
2347 }
2348
2349 __nvme_release_subsystem(subsys);
2350 subsys = found;
2351 } else {
2352 ret = device_add(&subsys->dev);
2353 if (ret) {
2354 dev_err(ctrl->device,
2355 "failed to register subsystem device.\n");
2356 goto out_unlock;
2357 }
2358 ida_init(&subsys->ns_ida);
2359 list_add_tail(&subsys->entry, &nvme_subsystems);
2360 }
2361
2362 ctrl->subsys = subsys;
2363 mutex_unlock(&nvme_subsystems_lock);
2364
2365 if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2366 dev_name(ctrl->device))) {
2367 dev_err(ctrl->device,
2368 "failed to create sysfs link from subsystem.\n");
2369 /* the transport driver will eventually put the subsystem */
2370 return -EINVAL;
2371 }
2372
2373 mutex_lock(&subsys->lock);
2374 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2375 mutex_unlock(&subsys->lock);
2376
2377 return 0;
2378
2379 out_unlock:
2380 mutex_unlock(&nvme_subsystems_lock);
2381 put_device(&subsys->dev);
2382 return ret;
2383 }
2384
2385 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2386 void *log, size_t size, u64 offset)
2387 {
2388 struct nvme_command c = { };
2389 unsigned long dwlen = size / 4 - 1;
2390
2391 c.get_log_page.opcode = nvme_admin_get_log_page;
2392 c.get_log_page.nsid = cpu_to_le32(nsid);
2393 c.get_log_page.lid = log_page;
2394 c.get_log_page.lsp = lsp;
2395 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2396 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2397 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2398 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2399
2400 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2401 }
2402
2403 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2404 {
2405 int ret;
2406
2407 if (!ctrl->effects)
2408 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2409
2410 if (!ctrl->effects)
2411 return 0;
2412
2413 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2414 ctrl->effects, sizeof(*ctrl->effects), 0);
2415 if (ret) {
2416 kfree(ctrl->effects);
2417 ctrl->effects = NULL;
2418 }
2419 return ret;
2420 }
2421
2422 /*
2423 * Initialize the cached copies of the Identify data and various controller
2424 * register in our nvme_ctrl structure. This should be called as soon as
2425 * the admin queue is fully up and running.
2426 */
2427 int nvme_init_identify(struct nvme_ctrl *ctrl)
2428 {
2429 struct nvme_id_ctrl *id;
2430 u64 cap;
2431 int ret, page_shift;
2432 u32 max_hw_sectors;
2433 bool prev_apst_enabled;
2434
2435 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2436 if (ret) {
2437 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2438 return ret;
2439 }
2440
2441 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2442 if (ret) {
2443 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2444 return ret;
2445 }
2446 page_shift = NVME_CAP_MPSMIN(cap) + 12;
2447
2448 if (ctrl->vs >= NVME_VS(1, 1, 0))
2449 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2450
2451 ret = nvme_identify_ctrl(ctrl, &id);
2452 if (ret) {
2453 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2454 return -EIO;
2455 }
2456
2457 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2458 ret = nvme_get_effects_log(ctrl);
2459 if (ret < 0)
2460 goto out_free;
2461 }
2462
2463 if (!ctrl->identified) {
2464 int i;
2465
2466 ret = nvme_init_subsystem(ctrl, id);
2467 if (ret)
2468 goto out_free;
2469
2470 /*
2471 * Check for quirks. Quirk can depend on firmware version,
2472 * so, in principle, the set of quirks present can change
2473 * across a reset. As a possible future enhancement, we
2474 * could re-scan for quirks every time we reinitialize
2475 * the device, but we'd have to make sure that the driver
2476 * behaves intelligently if the quirks change.
2477 */
2478 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2479 if (quirk_matches(id, &core_quirks[i]))
2480 ctrl->quirks |= core_quirks[i].quirks;
2481 }
2482 }
2483
2484 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2485 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2486 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2487 }
2488
2489 ctrl->oacs = le16_to_cpu(id->oacs);
2490 ctrl->oncs = le16_to_cpup(&id->oncs);
2491 ctrl->oaes = le32_to_cpu(id->oaes);
2492 atomic_set(&ctrl->abort_limit, id->acl + 1);
2493 ctrl->vwc = id->vwc;
2494 ctrl->cntlid = le16_to_cpup(&id->cntlid);
2495 if (id->mdts)
2496 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2497 else
2498 max_hw_sectors = UINT_MAX;
2499 ctrl->max_hw_sectors =
2500 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2501
2502 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2503 ctrl->sgls = le32_to_cpu(id->sgls);
2504 ctrl->kas = le16_to_cpu(id->kas);
2505 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2506
2507 if (id->rtd3e) {
2508 /* us -> s */
2509 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2510
2511 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2512 shutdown_timeout, 60);
2513
2514 if (ctrl->shutdown_timeout != shutdown_timeout)
2515 dev_info(ctrl->device,
2516 "Shutdown timeout set to %u seconds\n",
2517 ctrl->shutdown_timeout);
2518 } else
2519 ctrl->shutdown_timeout = shutdown_timeout;
2520
2521 ctrl->npss = id->npss;
2522 ctrl->apsta = id->apsta;
2523 prev_apst_enabled = ctrl->apst_enabled;
2524 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2525 if (force_apst && id->apsta) {
2526 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2527 ctrl->apst_enabled = true;
2528 } else {
2529 ctrl->apst_enabled = false;
2530 }
2531 } else {
2532 ctrl->apst_enabled = id->apsta;
2533 }
2534 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2535
2536 if (ctrl->ops->flags & NVME_F_FABRICS) {
2537 ctrl->icdoff = le16_to_cpu(id->icdoff);
2538 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2539 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2540 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2541
2542 /*
2543 * In fabrics we need to verify the cntlid matches the
2544 * admin connect
2545 */
2546 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2547 ret = -EINVAL;
2548 goto out_free;
2549 }
2550
2551 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2552 dev_err(ctrl->device,
2553 "keep-alive support is mandatory for fabrics\n");
2554 ret = -EINVAL;
2555 goto out_free;
2556 }
2557 } else {
2558 ctrl->cntlid = le16_to_cpu(id->cntlid);
2559 ctrl->hmpre = le32_to_cpu(id->hmpre);
2560 ctrl->hmmin = le32_to_cpu(id->hmmin);
2561 ctrl->hmminds = le32_to_cpu(id->hmminds);
2562 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2563 }
2564
2565 ret = nvme_mpath_init(ctrl, id);
2566 kfree(id);
2567
2568 if (ret < 0)
2569 return ret;
2570
2571 if (ctrl->apst_enabled && !prev_apst_enabled)
2572 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2573 else if (!ctrl->apst_enabled && prev_apst_enabled)
2574 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2575
2576 ret = nvme_configure_apst(ctrl);
2577 if (ret < 0)
2578 return ret;
2579
2580 ret = nvme_configure_timestamp(ctrl);
2581 if (ret < 0)
2582 return ret;
2583
2584 ret = nvme_configure_directives(ctrl);
2585 if (ret < 0)
2586 return ret;
2587
2588 ctrl->identified = true;
2589
2590 return 0;
2591
2592 out_free:
2593 kfree(id);
2594 return ret;
2595 }
2596 EXPORT_SYMBOL_GPL(nvme_init_identify);
2597
2598 static int nvme_dev_open(struct inode *inode, struct file *file)
2599 {
2600 struct nvme_ctrl *ctrl =
2601 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2602
2603 switch (ctrl->state) {
2604 case NVME_CTRL_LIVE:
2605 case NVME_CTRL_ADMIN_ONLY:
2606 break;
2607 default:
2608 return -EWOULDBLOCK;
2609 }
2610
2611 file->private_data = ctrl;
2612 return 0;
2613 }
2614
2615 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2616 {
2617 struct nvme_ns *ns;
2618 int ret;
2619
2620 down_read(&ctrl->namespaces_rwsem);
2621 if (list_empty(&ctrl->namespaces)) {
2622 ret = -ENOTTY;
2623 goto out_unlock;
2624 }
2625
2626 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2627 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2628 dev_warn(ctrl->device,
2629 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2630 ret = -EINVAL;
2631 goto out_unlock;
2632 }
2633
2634 dev_warn(ctrl->device,
2635 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2636 kref_get(&ns->kref);
2637 up_read(&ctrl->namespaces_rwsem);
2638
2639 ret = nvme_user_cmd(ctrl, ns, argp);
2640 nvme_put_ns(ns);
2641 return ret;
2642
2643 out_unlock:
2644 up_read(&ctrl->namespaces_rwsem);
2645 return ret;
2646 }
2647
2648 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2649 unsigned long arg)
2650 {
2651 struct nvme_ctrl *ctrl = file->private_data;
2652 void __user *argp = (void __user *)arg;
2653
2654 switch (cmd) {
2655 case NVME_IOCTL_ADMIN_CMD:
2656 return nvme_user_cmd(ctrl, NULL, argp);
2657 case NVME_IOCTL_IO_CMD:
2658 return nvme_dev_user_cmd(ctrl, argp);
2659 case NVME_IOCTL_RESET:
2660 dev_warn(ctrl->device, "resetting controller\n");
2661 return nvme_reset_ctrl_sync(ctrl);
2662 case NVME_IOCTL_SUBSYS_RESET:
2663 return nvme_reset_subsystem(ctrl);
2664 case NVME_IOCTL_RESCAN:
2665 nvme_queue_scan(ctrl);
2666 return 0;
2667 default:
2668 return -ENOTTY;
2669 }
2670 }
2671
2672 static const struct file_operations nvme_dev_fops = {
2673 .owner = THIS_MODULE,
2674 .open = nvme_dev_open,
2675 .unlocked_ioctl = nvme_dev_ioctl,
2676 .compat_ioctl = nvme_dev_ioctl,
2677 };
2678
2679 static ssize_t nvme_sysfs_reset(struct device *dev,
2680 struct device_attribute *attr, const char *buf,
2681 size_t count)
2682 {
2683 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2684 int ret;
2685
2686 ret = nvme_reset_ctrl_sync(ctrl);
2687 if (ret < 0)
2688 return ret;
2689 return count;
2690 }
2691 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2692
2693 static ssize_t nvme_sysfs_rescan(struct device *dev,
2694 struct device_attribute *attr, const char *buf,
2695 size_t count)
2696 {
2697 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2698
2699 nvme_queue_scan(ctrl);
2700 return count;
2701 }
2702 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2703
2704 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2705 {
2706 struct gendisk *disk = dev_to_disk(dev);
2707
2708 if (disk->fops == &nvme_fops)
2709 return nvme_get_ns_from_dev(dev)->head;
2710 else
2711 return disk->private_data;
2712 }
2713
2714 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2715 char *buf)
2716 {
2717 struct nvme_ns_head *head = dev_to_ns_head(dev);
2718 struct nvme_ns_ids *ids = &head->ids;
2719 struct nvme_subsystem *subsys = head->subsys;
2720 int serial_len = sizeof(subsys->serial);
2721 int model_len = sizeof(subsys->model);
2722
2723 if (!uuid_is_null(&ids->uuid))
2724 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2725
2726 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2727 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2728
2729 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2730 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2731
2732 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2733 subsys->serial[serial_len - 1] == '\0'))
2734 serial_len--;
2735 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2736 subsys->model[model_len - 1] == '\0'))
2737 model_len--;
2738
2739 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2740 serial_len, subsys->serial, model_len, subsys->model,
2741 head->ns_id);
2742 }
2743 static DEVICE_ATTR_RO(wwid);
2744
2745 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2746 char *buf)
2747 {
2748 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2749 }
2750 static DEVICE_ATTR_RO(nguid);
2751
2752 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2753 char *buf)
2754 {
2755 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2756
2757 /* For backward compatibility expose the NGUID to userspace if
2758 * we have no UUID set
2759 */
2760 if (uuid_is_null(&ids->uuid)) {
2761 printk_ratelimited(KERN_WARNING
2762 "No UUID available providing old NGUID\n");
2763 return sprintf(buf, "%pU\n", ids->nguid);
2764 }
2765 return sprintf(buf, "%pU\n", &ids->uuid);
2766 }
2767 static DEVICE_ATTR_RO(uuid);
2768
2769 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2770 char *buf)
2771 {
2772 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2773 }
2774 static DEVICE_ATTR_RO(eui);
2775
2776 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2777 char *buf)
2778 {
2779 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2780 }
2781 static DEVICE_ATTR_RO(nsid);
2782
2783 static struct attribute *nvme_ns_id_attrs[] = {
2784 &dev_attr_wwid.attr,
2785 &dev_attr_uuid.attr,
2786 &dev_attr_nguid.attr,
2787 &dev_attr_eui.attr,
2788 &dev_attr_nsid.attr,
2789 #ifdef CONFIG_NVME_MULTIPATH
2790 &dev_attr_ana_grpid.attr,
2791 &dev_attr_ana_state.attr,
2792 #endif
2793 NULL,
2794 };
2795
2796 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2797 struct attribute *a, int n)
2798 {
2799 struct device *dev = container_of(kobj, struct device, kobj);
2800 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2801
2802 if (a == &dev_attr_uuid.attr) {
2803 if (uuid_is_null(&ids->uuid) &&
2804 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2805 return 0;
2806 }
2807 if (a == &dev_attr_nguid.attr) {
2808 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2809 return 0;
2810 }
2811 if (a == &dev_attr_eui.attr) {
2812 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2813 return 0;
2814 }
2815 #ifdef CONFIG_NVME_MULTIPATH
2816 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2817 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2818 return 0;
2819 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2820 return 0;
2821 }
2822 #endif
2823 return a->mode;
2824 }
2825
2826 const struct attribute_group nvme_ns_id_attr_group = {
2827 .attrs = nvme_ns_id_attrs,
2828 .is_visible = nvme_ns_id_attrs_are_visible,
2829 };
2830
2831 #define nvme_show_str_function(field) \
2832 static ssize_t field##_show(struct device *dev, \
2833 struct device_attribute *attr, char *buf) \
2834 { \
2835 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2836 return sprintf(buf, "%.*s\n", \
2837 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
2838 } \
2839 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2840
2841 nvme_show_str_function(model);
2842 nvme_show_str_function(serial);
2843 nvme_show_str_function(firmware_rev);
2844
2845 #define nvme_show_int_function(field) \
2846 static ssize_t field##_show(struct device *dev, \
2847 struct device_attribute *attr, char *buf) \
2848 { \
2849 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2850 return sprintf(buf, "%d\n", ctrl->field); \
2851 } \
2852 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2853
2854 nvme_show_int_function(cntlid);
2855
2856 static ssize_t nvme_sysfs_delete(struct device *dev,
2857 struct device_attribute *attr, const char *buf,
2858 size_t count)
2859 {
2860 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2861
2862 if (device_remove_file_self(dev, attr))
2863 nvme_delete_ctrl_sync(ctrl);
2864 return count;
2865 }
2866 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2867
2868 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2869 struct device_attribute *attr,
2870 char *buf)
2871 {
2872 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2873
2874 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2875 }
2876 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2877
2878 static ssize_t nvme_sysfs_show_state(struct device *dev,
2879 struct device_attribute *attr,
2880 char *buf)
2881 {
2882 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2883 static const char *const state_name[] = {
2884 [NVME_CTRL_NEW] = "new",
2885 [NVME_CTRL_LIVE] = "live",
2886 [NVME_CTRL_ADMIN_ONLY] = "only-admin",
2887 [NVME_CTRL_RESETTING] = "resetting",
2888 [NVME_CTRL_CONNECTING] = "connecting",
2889 [NVME_CTRL_DELETING] = "deleting",
2890 [NVME_CTRL_DEAD] = "dead",
2891 };
2892
2893 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2894 state_name[ctrl->state])
2895 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2896
2897 return sprintf(buf, "unknown state\n");
2898 }
2899
2900 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2901
2902 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2903 struct device_attribute *attr,
2904 char *buf)
2905 {
2906 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2907
2908 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2909 }
2910 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2911
2912 static ssize_t nvme_sysfs_show_address(struct device *dev,
2913 struct device_attribute *attr,
2914 char *buf)
2915 {
2916 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2917
2918 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2919 }
2920 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2921
2922 static struct attribute *nvme_dev_attrs[] = {
2923 &dev_attr_reset_controller.attr,
2924 &dev_attr_rescan_controller.attr,
2925 &dev_attr_model.attr,
2926 &dev_attr_serial.attr,
2927 &dev_attr_firmware_rev.attr,
2928 &dev_attr_cntlid.attr,
2929 &dev_attr_delete_controller.attr,
2930 &dev_attr_transport.attr,
2931 &dev_attr_subsysnqn.attr,
2932 &dev_attr_address.attr,
2933 &dev_attr_state.attr,
2934 NULL
2935 };
2936
2937 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2938 struct attribute *a, int n)
2939 {
2940 struct device *dev = container_of(kobj, struct device, kobj);
2941 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2942
2943 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2944 return 0;
2945 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2946 return 0;
2947
2948 return a->mode;
2949 }
2950
2951 static struct attribute_group nvme_dev_attrs_group = {
2952 .attrs = nvme_dev_attrs,
2953 .is_visible = nvme_dev_attrs_are_visible,
2954 };
2955
2956 static const struct attribute_group *nvme_dev_attr_groups[] = {
2957 &nvme_dev_attrs_group,
2958 NULL,
2959 };
2960
2961 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2962 unsigned nsid)
2963 {
2964 struct nvme_ns_head *h;
2965
2966 lockdep_assert_held(&subsys->lock);
2967
2968 list_for_each_entry(h, &subsys->nsheads, entry) {
2969 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2970 return h;
2971 }
2972
2973 return NULL;
2974 }
2975
2976 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2977 struct nvme_ns_head *new)
2978 {
2979 struct nvme_ns_head *h;
2980
2981 lockdep_assert_held(&subsys->lock);
2982
2983 list_for_each_entry(h, &subsys->nsheads, entry) {
2984 if (nvme_ns_ids_valid(&new->ids) &&
2985 !list_empty(&h->list) &&
2986 nvme_ns_ids_equal(&new->ids, &h->ids))
2987 return -EINVAL;
2988 }
2989
2990 return 0;
2991 }
2992
2993 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2994 unsigned nsid, struct nvme_id_ns *id)
2995 {
2996 struct nvme_ns_head *head;
2997 int ret = -ENOMEM;
2998
2999 head = kzalloc(sizeof(*head), GFP_KERNEL);
3000 if (!head)
3001 goto out;
3002 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3003 if (ret < 0)
3004 goto out_free_head;
3005 head->instance = ret;
3006 INIT_LIST_HEAD(&head->list);
3007 ret = init_srcu_struct(&head->srcu);
3008 if (ret)
3009 goto out_ida_remove;
3010 head->subsys = ctrl->subsys;
3011 head->ns_id = nsid;
3012 kref_init(&head->ref);
3013
3014 nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3015
3016 ret = __nvme_check_ids(ctrl->subsys, head);
3017 if (ret) {
3018 dev_err(ctrl->device,
3019 "duplicate IDs for nsid %d\n", nsid);
3020 goto out_cleanup_srcu;
3021 }
3022
3023 ret = nvme_mpath_alloc_disk(ctrl, head);
3024 if (ret)
3025 goto out_cleanup_srcu;
3026
3027 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3028
3029 kref_get(&ctrl->subsys->ref);
3030
3031 return head;
3032 out_cleanup_srcu:
3033 cleanup_srcu_struct(&head->srcu);
3034 out_ida_remove:
3035 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3036 out_free_head:
3037 kfree(head);
3038 out:
3039 return ERR_PTR(ret);
3040 }
3041
3042 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3043 struct nvme_id_ns *id)
3044 {
3045 struct nvme_ctrl *ctrl = ns->ctrl;
3046 bool is_shared = id->nmic & (1 << 0);
3047 struct nvme_ns_head *head = NULL;
3048 int ret = 0;
3049
3050 mutex_lock(&ctrl->subsys->lock);
3051 if (is_shared)
3052 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3053 if (!head) {
3054 head = nvme_alloc_ns_head(ctrl, nsid, id);
3055 if (IS_ERR(head)) {
3056 ret = PTR_ERR(head);
3057 goto out_unlock;
3058 }
3059 } else {
3060 struct nvme_ns_ids ids;
3061
3062 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3063 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3064 dev_err(ctrl->device,
3065 "IDs don't match for shared namespace %d\n",
3066 nsid);
3067 ret = -EINVAL;
3068 goto out_unlock;
3069 }
3070 }
3071
3072 list_add_tail(&ns->siblings, &head->list);
3073 ns->head = head;
3074
3075 out_unlock:
3076 mutex_unlock(&ctrl->subsys->lock);
3077 return ret;
3078 }
3079
3080 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3081 {
3082 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3083 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3084
3085 return nsa->head->ns_id - nsb->head->ns_id;
3086 }
3087
3088 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3089 {
3090 struct nvme_ns *ns, *ret = NULL;
3091
3092 down_read(&ctrl->namespaces_rwsem);
3093 list_for_each_entry(ns, &ctrl->namespaces, list) {
3094 if (ns->head->ns_id == nsid) {
3095 if (!kref_get_unless_zero(&ns->kref))
3096 continue;
3097 ret = ns;
3098 break;
3099 }
3100 if (ns->head->ns_id > nsid)
3101 break;
3102 }
3103 up_read(&ctrl->namespaces_rwsem);
3104 return ret;
3105 }
3106
3107 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3108 {
3109 struct streams_directive_params s;
3110 int ret;
3111
3112 if (!ctrl->nr_streams)
3113 return 0;
3114
3115 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3116 if (ret)
3117 return ret;
3118
3119 ns->sws = le32_to_cpu(s.sws);
3120 ns->sgs = le16_to_cpu(s.sgs);
3121
3122 if (ns->sws) {
3123 unsigned int bs = 1 << ns->lba_shift;
3124
3125 blk_queue_io_min(ns->queue, bs * ns->sws);
3126 if (ns->sgs)
3127 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3128 }
3129
3130 return 0;
3131 }
3132
3133 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3134 {
3135 struct nvme_ns *ns;
3136 struct gendisk *disk;
3137 struct nvme_id_ns *id;
3138 char disk_name[DISK_NAME_LEN];
3139 int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
3140
3141 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3142 if (!ns)
3143 return;
3144
3145 ns->queue = blk_mq_init_queue(ctrl->tagset);
3146 if (IS_ERR(ns->queue))
3147 goto out_free_ns;
3148 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3149 ns->queue->queuedata = ns;
3150 ns->ctrl = ctrl;
3151
3152 kref_init(&ns->kref);
3153 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3154
3155 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3156 nvme_set_queue_limits(ctrl, ns->queue);
3157
3158 id = nvme_identify_ns(ctrl, nsid);
3159 if (!id)
3160 goto out_free_queue;
3161
3162 if (id->ncap == 0)
3163 goto out_free_id;
3164
3165 if (nvme_init_ns_head(ns, nsid, id))
3166 goto out_free_id;
3167 nvme_setup_streams_ns(ctrl, ns);
3168 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3169
3170 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3171 if (nvme_nvm_register(ns, disk_name, node)) {
3172 dev_warn(ctrl->device, "LightNVM init failure\n");
3173 goto out_unlink_ns;
3174 }
3175 }
3176
3177 disk = alloc_disk_node(0, node);
3178 if (!disk)
3179 goto out_unlink_ns;
3180
3181 disk->fops = &nvme_fops;
3182 disk->private_data = ns;
3183 disk->queue = ns->queue;
3184 disk->flags = flags;
3185 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3186 ns->disk = disk;
3187
3188 __nvme_revalidate_disk(disk, id);
3189
3190 down_write(&ctrl->namespaces_rwsem);
3191 list_add_tail(&ns->list, &ctrl->namespaces);
3192 up_write(&ctrl->namespaces_rwsem);
3193
3194 nvme_get_ctrl(ctrl);
3195
3196 device_add_disk(ctrl->device, ns->disk);
3197 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
3198 &nvme_ns_id_attr_group))
3199 pr_warn("%s: failed to create sysfs group for identification\n",
3200 ns->disk->disk_name);
3201 if (ns->ndev && nvme_nvm_register_sysfs(ns))
3202 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
3203 ns->disk->disk_name);
3204
3205 nvme_mpath_add_disk(ns, id);
3206 nvme_fault_inject_init(ns);
3207 kfree(id);
3208
3209 return;
3210 out_unlink_ns:
3211 mutex_lock(&ctrl->subsys->lock);
3212 list_del_rcu(&ns->siblings);
3213 mutex_unlock(&ctrl->subsys->lock);
3214 out_free_id:
3215 kfree(id);
3216 out_free_queue:
3217 blk_cleanup_queue(ns->queue);
3218 out_free_ns:
3219 kfree(ns);
3220 }
3221
3222 static void nvme_ns_remove(struct nvme_ns *ns)
3223 {
3224 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3225 return;
3226
3227 nvme_fault_inject_fini(ns);
3228
3229 mutex_lock(&ns->ctrl->subsys->lock);
3230 list_del_rcu(&ns->siblings);
3231 mutex_unlock(&ns->ctrl->subsys->lock);
3232 synchronize_rcu(); /* guarantee not available in head->list */
3233 nvme_mpath_clear_current_path(ns);
3234 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3235
3236 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3237 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
3238 &nvme_ns_id_attr_group);
3239 if (ns->ndev)
3240 nvme_nvm_unregister_sysfs(ns);
3241 del_gendisk(ns->disk);
3242 blk_cleanup_queue(ns->queue);
3243 if (blk_get_integrity(ns->disk))
3244 blk_integrity_unregister(ns->disk);
3245 }
3246
3247 down_write(&ns->ctrl->namespaces_rwsem);
3248 list_del_init(&ns->list);
3249 up_write(&ns->ctrl->namespaces_rwsem);
3250
3251 nvme_mpath_check_last_path(ns);
3252 nvme_put_ns(ns);
3253 }
3254
3255 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3256 {
3257 struct nvme_ns *ns;
3258
3259 ns = nvme_find_get_ns(ctrl, nsid);
3260 if (ns) {
3261 if (ns->disk && revalidate_disk(ns->disk))
3262 nvme_ns_remove(ns);
3263 nvme_put_ns(ns);
3264 } else
3265 nvme_alloc_ns(ctrl, nsid);
3266 }
3267
3268 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3269 unsigned nsid)
3270 {
3271 struct nvme_ns *ns, *next;
3272 LIST_HEAD(rm_list);
3273
3274 down_write(&ctrl->namespaces_rwsem);
3275 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3276 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3277 list_move_tail(&ns->list, &rm_list);
3278 }
3279 up_write(&ctrl->namespaces_rwsem);
3280
3281 list_for_each_entry_safe(ns, next, &rm_list, list)
3282 nvme_ns_remove(ns);
3283
3284 }
3285
3286 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3287 {
3288 struct nvme_ns *ns;
3289 __le32 *ns_list;
3290 unsigned i, j, nsid, prev = 0;
3291 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3292 int ret = 0;
3293
3294 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3295 if (!ns_list)
3296 return -ENOMEM;
3297
3298 for (i = 0; i < num_lists; i++) {
3299 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3300 if (ret)
3301 goto free;
3302
3303 for (j = 0; j < min(nn, 1024U); j++) {
3304 nsid = le32_to_cpu(ns_list[j]);
3305 if (!nsid)
3306 goto out;
3307
3308 nvme_validate_ns(ctrl, nsid);
3309
3310 while (++prev < nsid) {
3311 ns = nvme_find_get_ns(ctrl, prev);
3312 if (ns) {
3313 nvme_ns_remove(ns);
3314 nvme_put_ns(ns);
3315 }
3316 }
3317 }
3318 nn -= j;
3319 }
3320 out:
3321 nvme_remove_invalid_namespaces(ctrl, prev);
3322 free:
3323 kfree(ns_list);
3324 return ret;
3325 }
3326
3327 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3328 {
3329 unsigned i;
3330
3331 for (i = 1; i <= nn; i++)
3332 nvme_validate_ns(ctrl, i);
3333
3334 nvme_remove_invalid_namespaces(ctrl, nn);
3335 }
3336
3337 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3338 {
3339 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3340 __le32 *log;
3341 int error;
3342
3343 log = kzalloc(log_size, GFP_KERNEL);
3344 if (!log)
3345 return;
3346
3347 /*
3348 * We need to read the log to clear the AEN, but we don't want to rely
3349 * on it for the changed namespace information as userspace could have
3350 * raced with us in reading the log page, which could cause us to miss
3351 * updates.
3352 */
3353 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3354 log_size, 0);
3355 if (error)
3356 dev_warn(ctrl->device,
3357 "reading changed ns log failed: %d\n", error);
3358
3359 kfree(log);
3360 }
3361
3362 static void nvme_scan_work(struct work_struct *work)
3363 {
3364 struct nvme_ctrl *ctrl =
3365 container_of(work, struct nvme_ctrl, scan_work);
3366 struct nvme_id_ctrl *id;
3367 unsigned nn;
3368
3369 if (ctrl->state != NVME_CTRL_LIVE)
3370 return;
3371
3372 WARN_ON_ONCE(!ctrl->tagset);
3373
3374 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3375 dev_info(ctrl->device, "rescanning namespaces.\n");
3376 nvme_clear_changed_ns_log(ctrl);
3377 }
3378
3379 if (nvme_identify_ctrl(ctrl, &id))
3380 return;
3381
3382 mutex_lock(&ctrl->scan_lock);
3383 nn = le32_to_cpu(id->nn);
3384 if (!nvme_ctrl_limited_cns(ctrl)) {
3385 if (!nvme_scan_ns_list(ctrl, nn))
3386 goto out_free_id;
3387 }
3388 nvme_scan_ns_sequential(ctrl, nn);
3389 out_free_id:
3390 mutex_unlock(&ctrl->scan_lock);
3391 kfree(id);
3392 down_write(&ctrl->namespaces_rwsem);
3393 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3394 up_write(&ctrl->namespaces_rwsem);
3395 }
3396
3397 /*
3398 * This function iterates the namespace list unlocked to allow recovery from
3399 * controller failure. It is up to the caller to ensure the namespace list is
3400 * not modified by scan work while this function is executing.
3401 */
3402 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3403 {
3404 struct nvme_ns *ns, *next;
3405 LIST_HEAD(ns_list);
3406
3407 /* prevent racing with ns scanning */
3408 flush_work(&ctrl->scan_work);
3409
3410 /*
3411 * The dead states indicates the controller was not gracefully
3412 * disconnected. In that case, we won't be able to flush any data while
3413 * removing the namespaces' disks; fail all the queues now to avoid
3414 * potentially having to clean up the failed sync later.
3415 */
3416 if (ctrl->state == NVME_CTRL_DEAD)
3417 nvme_kill_queues(ctrl);
3418
3419 down_write(&ctrl->namespaces_rwsem);
3420 list_splice_init(&ctrl->namespaces, &ns_list);
3421 up_write(&ctrl->namespaces_rwsem);
3422
3423 list_for_each_entry_safe(ns, next, &ns_list, list)
3424 nvme_ns_remove(ns);
3425 }
3426 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3427
3428 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3429 {
3430 char *envp[2] = { NULL, NULL };
3431 u32 aen_result = ctrl->aen_result;
3432
3433 ctrl->aen_result = 0;
3434 if (!aen_result)
3435 return;
3436
3437 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3438 if (!envp[0])
3439 return;
3440 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3441 kfree(envp[0]);
3442 }
3443
3444 static void nvme_async_event_work(struct work_struct *work)
3445 {
3446 struct nvme_ctrl *ctrl =
3447 container_of(work, struct nvme_ctrl, async_event_work);
3448
3449 nvme_aen_uevent(ctrl);
3450 ctrl->ops->submit_async_event(ctrl);
3451 }
3452
3453 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3454 {
3455
3456 u32 csts;
3457
3458 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3459 return false;
3460
3461 if (csts == ~0)
3462 return false;
3463
3464 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3465 }
3466
3467 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3468 {
3469 struct nvme_fw_slot_info_log *log;
3470
3471 log = kmalloc(sizeof(*log), GFP_KERNEL);
3472 if (!log)
3473 return;
3474
3475 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, log,
3476 sizeof(*log), 0))
3477 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3478 kfree(log);
3479 }
3480
3481 static void nvme_fw_act_work(struct work_struct *work)
3482 {
3483 struct nvme_ctrl *ctrl = container_of(work,
3484 struct nvme_ctrl, fw_act_work);
3485 unsigned long fw_act_timeout;
3486
3487 if (ctrl->mtfa)
3488 fw_act_timeout = jiffies +
3489 msecs_to_jiffies(ctrl->mtfa * 100);
3490 else
3491 fw_act_timeout = jiffies +
3492 msecs_to_jiffies(admin_timeout * 1000);
3493
3494 nvme_stop_queues(ctrl);
3495 while (nvme_ctrl_pp_status(ctrl)) {
3496 if (time_after(jiffies, fw_act_timeout)) {
3497 dev_warn(ctrl->device,
3498 "Fw activation timeout, reset controller\n");
3499 nvme_reset_ctrl(ctrl);
3500 break;
3501 }
3502 msleep(100);
3503 }
3504
3505 if (ctrl->state != NVME_CTRL_LIVE)
3506 return;
3507
3508 nvme_start_queues(ctrl);
3509 /* read FW slot information to clear the AER */
3510 nvme_get_fw_slot_info(ctrl);
3511 }
3512
3513 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3514 {
3515 switch ((result & 0xff00) >> 8) {
3516 case NVME_AER_NOTICE_NS_CHANGED:
3517 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3518 nvme_queue_scan(ctrl);
3519 break;
3520 case NVME_AER_NOTICE_FW_ACT_STARTING:
3521 queue_work(nvme_wq, &ctrl->fw_act_work);
3522 break;
3523 #ifdef CONFIG_NVME_MULTIPATH
3524 case NVME_AER_NOTICE_ANA:
3525 if (!ctrl->ana_log_buf)
3526 break;
3527 queue_work(nvme_wq, &ctrl->ana_work);
3528 break;
3529 #endif
3530 default:
3531 dev_warn(ctrl->device, "async event result %08x\n", result);
3532 }
3533 }
3534
3535 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3536 volatile union nvme_result *res)
3537 {
3538 u32 result = le32_to_cpu(res->u32);
3539
3540 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3541 return;
3542
3543 switch (result & 0x7) {
3544 case NVME_AER_NOTICE:
3545 nvme_handle_aen_notice(ctrl, result);
3546 break;
3547 case NVME_AER_ERROR:
3548 case NVME_AER_SMART:
3549 case NVME_AER_CSS:
3550 case NVME_AER_VS:
3551 ctrl->aen_result = result;
3552 break;
3553 default:
3554 break;
3555 }
3556 queue_work(nvme_wq, &ctrl->async_event_work);
3557 }
3558 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3559
3560 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3561 {
3562 nvme_mpath_stop(ctrl);
3563 nvme_stop_keep_alive(ctrl);
3564 flush_work(&ctrl->async_event_work);
3565 cancel_work_sync(&ctrl->fw_act_work);
3566 if (ctrl->ops->stop_ctrl)
3567 ctrl->ops->stop_ctrl(ctrl);
3568 }
3569 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3570
3571 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3572 {
3573 if (ctrl->kato)
3574 nvme_start_keep_alive(ctrl);
3575
3576 if (ctrl->queue_count > 1) {
3577 nvme_queue_scan(ctrl);
3578 nvme_enable_aen(ctrl);
3579 queue_work(nvme_wq, &ctrl->async_event_work);
3580 nvme_start_queues(ctrl);
3581 }
3582 }
3583 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3584
3585 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3586 {
3587 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3588 cdev_device_del(&ctrl->cdev, ctrl->device);
3589 }
3590 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3591
3592 static void nvme_free_ctrl(struct device *dev)
3593 {
3594 struct nvme_ctrl *ctrl =
3595 container_of(dev, struct nvme_ctrl, ctrl_device);
3596 struct nvme_subsystem *subsys = ctrl->subsys;
3597
3598 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3599 kfree(ctrl->effects);
3600 nvme_mpath_uninit(ctrl);
3601 __free_page(ctrl->discard_page);
3602
3603 if (subsys) {
3604 mutex_lock(&subsys->lock);
3605 list_del(&ctrl->subsys_entry);
3606 mutex_unlock(&subsys->lock);
3607 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3608 }
3609
3610 ctrl->ops->free_ctrl(ctrl);
3611
3612 if (subsys)
3613 nvme_put_subsystem(subsys);
3614 }
3615
3616 /*
3617 * Initialize a NVMe controller structures. This needs to be called during
3618 * earliest initialization so that we have the initialized structured around
3619 * during probing.
3620 */
3621 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3622 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3623 {
3624 int ret;
3625
3626 ctrl->state = NVME_CTRL_NEW;
3627 spin_lock_init(&ctrl->lock);
3628 mutex_init(&ctrl->scan_lock);
3629 INIT_LIST_HEAD(&ctrl->namespaces);
3630 init_rwsem(&ctrl->namespaces_rwsem);
3631 ctrl->dev = dev;
3632 ctrl->ops = ops;
3633 ctrl->quirks = quirks;
3634 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3635 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3636 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3637 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3638
3639 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3640 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3641 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3642
3643 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3644 PAGE_SIZE);
3645 ctrl->discard_page = alloc_page(GFP_KERNEL);
3646 if (!ctrl->discard_page) {
3647 ret = -ENOMEM;
3648 goto out;
3649 }
3650
3651 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3652 if (ret < 0)
3653 goto out;
3654 ctrl->instance = ret;
3655
3656 device_initialize(&ctrl->ctrl_device);
3657 ctrl->device = &ctrl->ctrl_device;
3658 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3659 ctrl->device->class = nvme_class;
3660 ctrl->device->parent = ctrl->dev;
3661 ctrl->device->groups = nvme_dev_attr_groups;
3662 ctrl->device->release = nvme_free_ctrl;
3663 dev_set_drvdata(ctrl->device, ctrl);
3664 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3665 if (ret)
3666 goto out_release_instance;
3667
3668 cdev_init(&ctrl->cdev, &nvme_dev_fops);
3669 ctrl->cdev.owner = ops->module;
3670 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3671 if (ret)
3672 goto out_free_name;
3673
3674 /*
3675 * Initialize latency tolerance controls. The sysfs files won't
3676 * be visible to userspace unless the device actually supports APST.
3677 */
3678 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3679 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3680 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3681
3682 return 0;
3683 out_free_name:
3684 kfree_const(ctrl->device->kobj.name);
3685 out_release_instance:
3686 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3687 out:
3688 if (ctrl->discard_page)
3689 __free_page(ctrl->discard_page);
3690 return ret;
3691 }
3692 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3693
3694 /**
3695 * nvme_kill_queues(): Ends all namespace queues
3696 * @ctrl: the dead controller that needs to end
3697 *
3698 * Call this function when the driver determines it is unable to get the
3699 * controller in a state capable of servicing IO.
3700 */
3701 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3702 {
3703 struct nvme_ns *ns;
3704
3705 down_read(&ctrl->namespaces_rwsem);
3706
3707 /* Forcibly unquiesce queues to avoid blocking dispatch */
3708 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3709 blk_mq_unquiesce_queue(ctrl->admin_q);
3710
3711 list_for_each_entry(ns, &ctrl->namespaces, list)
3712 nvme_set_queue_dying(ns);
3713
3714 up_read(&ctrl->namespaces_rwsem);
3715 }
3716 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3717
3718 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3719 {
3720 struct nvme_ns *ns;
3721
3722 down_read(&ctrl->namespaces_rwsem);
3723 list_for_each_entry(ns, &ctrl->namespaces, list)
3724 blk_mq_unfreeze_queue(ns->queue);
3725 up_read(&ctrl->namespaces_rwsem);
3726 }
3727 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3728
3729 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3730 {
3731 struct nvme_ns *ns;
3732
3733 down_read(&ctrl->namespaces_rwsem);
3734 list_for_each_entry(ns, &ctrl->namespaces, list) {
3735 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3736 if (timeout <= 0)
3737 break;
3738 }
3739 up_read(&ctrl->namespaces_rwsem);
3740 }
3741 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3742
3743 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3744 {
3745 struct nvme_ns *ns;
3746
3747 down_read(&ctrl->namespaces_rwsem);
3748 list_for_each_entry(ns, &ctrl->namespaces, list)
3749 blk_mq_freeze_queue_wait(ns->queue);
3750 up_read(&ctrl->namespaces_rwsem);
3751 }
3752 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3753
3754 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3755 {
3756 struct nvme_ns *ns;
3757
3758 down_read(&ctrl->namespaces_rwsem);
3759 list_for_each_entry(ns, &ctrl->namespaces, list)
3760 blk_freeze_queue_start(ns->queue);
3761 up_read(&ctrl->namespaces_rwsem);
3762 }
3763 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3764
3765 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3766 {
3767 struct nvme_ns *ns;
3768
3769 down_read(&ctrl->namespaces_rwsem);
3770 list_for_each_entry(ns, &ctrl->namespaces, list)
3771 blk_mq_quiesce_queue(ns->queue);
3772 up_read(&ctrl->namespaces_rwsem);
3773 }
3774 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3775
3776 void nvme_start_queues(struct nvme_ctrl *ctrl)
3777 {
3778 struct nvme_ns *ns;
3779
3780 down_read(&ctrl->namespaces_rwsem);
3781 list_for_each_entry(ns, &ctrl->namespaces, list)
3782 blk_mq_unquiesce_queue(ns->queue);
3783 up_read(&ctrl->namespaces_rwsem);
3784 }
3785 EXPORT_SYMBOL_GPL(nvme_start_queues);
3786
3787 int __init nvme_core_init(void)
3788 {
3789 int result = -ENOMEM;
3790
3791 nvme_wq = alloc_workqueue("nvme-wq",
3792 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3793 if (!nvme_wq)
3794 goto out;
3795
3796 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3797 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3798 if (!nvme_reset_wq)
3799 goto destroy_wq;
3800
3801 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3802 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3803 if (!nvme_delete_wq)
3804 goto destroy_reset_wq;
3805
3806 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3807 if (result < 0)
3808 goto destroy_delete_wq;
3809
3810 nvme_class = class_create(THIS_MODULE, "nvme");
3811 if (IS_ERR(nvme_class)) {
3812 result = PTR_ERR(nvme_class);
3813 goto unregister_chrdev;
3814 }
3815
3816 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3817 if (IS_ERR(nvme_subsys_class)) {
3818 result = PTR_ERR(nvme_subsys_class);
3819 goto destroy_class;
3820 }
3821 return 0;
3822
3823 destroy_class:
3824 class_destroy(nvme_class);
3825 unregister_chrdev:
3826 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3827 destroy_delete_wq:
3828 destroy_workqueue(nvme_delete_wq);
3829 destroy_reset_wq:
3830 destroy_workqueue(nvme_reset_wq);
3831 destroy_wq:
3832 destroy_workqueue(nvme_wq);
3833 out:
3834 return result;
3835 }
3836
3837 void nvme_core_exit(void)
3838 {
3839 ida_destroy(&nvme_subsystems_ida);
3840 class_destroy(nvme_subsys_class);
3841 class_destroy(nvme_class);
3842 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3843 destroy_workqueue(nvme_delete_wq);
3844 destroy_workqueue(nvme_reset_wq);
3845 destroy_workqueue(nvme_wq);
3846 }
3847
3848 MODULE_LICENSE("GPL");
3849 MODULE_VERSION("1.0");
3850 module_init(nvme_core_init);
3851 module_exit(nvme_core_exit);
Linux with added FPC-III support