?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
bpf: Prevent memory disambiguation attack
[linux/fpc-iii.git] / drivers / nvme / host / core.c
blobf81773570dfd57423df47cbf8e1e408551030f4a
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
103 static __le32 nvme_get_log_dw10(u8 lid, size_t size)
104 {
105 return cpu_to_le32((((size / 4) - 1) << 16) | lid);
106 }
107
108 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
109 {
110 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
111 return -EBUSY;
112 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
113 return -EBUSY;
114 return 0;
115 }
116 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
117
118 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
119 {
120 int ret;
121
122 ret = nvme_reset_ctrl(ctrl);
123 if (!ret) {
124 flush_work(&ctrl->reset_work);
125 if (ctrl->state != NVME_CTRL_LIVE &&
126 ctrl->state != NVME_CTRL_ADMIN_ONLY)
127 ret = -ENETRESET;
128 }
129
130 return ret;
131 }
132 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
133
134 static void nvme_delete_ctrl_work(struct work_struct *work)
135 {
136 struct nvme_ctrl *ctrl =
137 container_of(work, struct nvme_ctrl, delete_work);
138
139 flush_work(&ctrl->reset_work);
140 nvme_stop_ctrl(ctrl);
141 nvme_remove_namespaces(ctrl);
142 ctrl->ops->delete_ctrl(ctrl);
143 nvme_uninit_ctrl(ctrl);
144 nvme_put_ctrl(ctrl);
145 }
146
147 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
148 {
149 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
150 return -EBUSY;
151 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
152 return -EBUSY;
153 return 0;
154 }
155 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
156
157 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
158 {
159 int ret = 0;
160
161 /*
162 * Keep a reference until the work is flushed since ->delete_ctrl
163 * can free the controller.
164 */
165 nvme_get_ctrl(ctrl);
166 ret = nvme_delete_ctrl(ctrl);
167 if (!ret)
168 flush_work(&ctrl->delete_work);
169 nvme_put_ctrl(ctrl);
170 return ret;
171 }
172 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
173
174 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
175 {
176 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
177 }
178
179 static blk_status_t nvme_error_status(struct request *req)
180 {
181 switch (nvme_req(req)->status & 0x7ff) {
182 case NVME_SC_SUCCESS:
183 return BLK_STS_OK;
184 case NVME_SC_CAP_EXCEEDED:
185 return BLK_STS_NOSPC;
186 case NVME_SC_LBA_RANGE:
187 return BLK_STS_TARGET;
188 case NVME_SC_BAD_ATTRIBUTES:
189 case NVME_SC_ONCS_NOT_SUPPORTED:
190 case NVME_SC_INVALID_OPCODE:
191 case NVME_SC_INVALID_FIELD:
192 case NVME_SC_INVALID_NS:
193 return BLK_STS_NOTSUPP;
194 case NVME_SC_WRITE_FAULT:
195 case NVME_SC_READ_ERROR:
196 case NVME_SC_UNWRITTEN_BLOCK:
197 case NVME_SC_ACCESS_DENIED:
198 case NVME_SC_READ_ONLY:
199 case NVME_SC_COMPARE_FAILED:
200 return BLK_STS_MEDIUM;
201 case NVME_SC_GUARD_CHECK:
202 case NVME_SC_APPTAG_CHECK:
203 case NVME_SC_REFTAG_CHECK:
204 case NVME_SC_INVALID_PI:
205 return BLK_STS_PROTECTION;
206 case NVME_SC_RESERVATION_CONFLICT:
207 return BLK_STS_NEXUS;
208 default:
209 return BLK_STS_IOERR;
210 }
211 }
212
213 static inline bool nvme_req_needs_retry(struct request *req)
214 {
215 if (blk_noretry_request(req))
216 return false;
217 if (nvme_req(req)->status & NVME_SC_DNR)
218 return false;
219 if (nvme_req(req)->retries >= nvme_max_retries)
220 return false;
221 return true;
222 }
223
224 void nvme_complete_rq(struct request *req)
225 {
226 blk_status_t status = nvme_error_status(req);
227
228 trace_nvme_complete_rq(req);
229
230 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
231 if (nvme_req_needs_failover(req, status)) {
232 nvme_failover_req(req);
233 return;
234 }
235
236 if (!blk_queue_dying(req->q)) {
237 nvme_req(req)->retries++;
238 blk_mq_requeue_request(req, true);
239 return;
240 }
241 }
242 blk_mq_end_request(req, status);
243 }
244 EXPORT_SYMBOL_GPL(nvme_complete_rq);
245
246 void nvme_cancel_request(struct request *req, void *data, bool reserved)
247 {
248 if (!blk_mq_request_started(req))
249 return;
250
251 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
252 "Cancelling I/O %d", req->tag);
253
254 nvme_req(req)->status = NVME_SC_ABORT_REQ;
255 blk_mq_complete_request(req);
256
257 }
258 EXPORT_SYMBOL_GPL(nvme_cancel_request);
259
260 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
261 enum nvme_ctrl_state new_state)
262 {
263 enum nvme_ctrl_state old_state;
264 unsigned long flags;
265 bool changed = false;
266
267 spin_lock_irqsave(&ctrl->lock, flags);
268
269 old_state = ctrl->state;
270 switch (new_state) {
271 case NVME_CTRL_ADMIN_ONLY:
272 switch (old_state) {
273 case NVME_CTRL_CONNECTING:
274 changed = true;
275 /* FALLTHRU */
276 default:
277 break;
278 }
279 break;
280 case NVME_CTRL_LIVE:
281 switch (old_state) {
282 case NVME_CTRL_NEW:
283 case NVME_CTRL_RESETTING:
284 case NVME_CTRL_CONNECTING:
285 changed = true;
286 /* FALLTHRU */
287 default:
288 break;
289 }
290 break;
291 case NVME_CTRL_RESETTING:
292 switch (old_state) {
293 case NVME_CTRL_NEW:
294 case NVME_CTRL_LIVE:
295 case NVME_CTRL_ADMIN_ONLY:
296 changed = true;
297 /* FALLTHRU */
298 default:
299 break;
300 }
301 break;
302 case NVME_CTRL_CONNECTING:
303 switch (old_state) {
304 case NVME_CTRL_NEW:
305 case NVME_CTRL_RESETTING:
306 changed = true;
307 /* FALLTHRU */
308 default:
309 break;
310 }
311 break;
312 case NVME_CTRL_DELETING:
313 switch (old_state) {
314 case NVME_CTRL_LIVE:
315 case NVME_CTRL_ADMIN_ONLY:
316 case NVME_CTRL_RESETTING:
317 case NVME_CTRL_CONNECTING:
318 changed = true;
319 /* FALLTHRU */
320 default:
321 break;
322 }
323 break;
324 case NVME_CTRL_DEAD:
325 switch (old_state) {
326 case NVME_CTRL_DELETING:
327 changed = true;
328 /* FALLTHRU */
329 default:
330 break;
331 }
332 break;
333 default:
334 break;
335 }
336
337 if (changed)
338 ctrl->state = new_state;
339
340 spin_unlock_irqrestore(&ctrl->lock, flags);
341 if (changed && ctrl->state == NVME_CTRL_LIVE)
342 nvme_kick_requeue_lists(ctrl);
343 return changed;
344 }
345 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
346
347 static void nvme_free_ns_head(struct kref *ref)
348 {
349 struct nvme_ns_head *head =
350 container_of(ref, struct nvme_ns_head, ref);
351
352 nvme_mpath_remove_disk(head);
353 ida_simple_remove(&head->subsys->ns_ida, head->instance);
354 list_del_init(&head->entry);
355 cleanup_srcu_struct(&head->srcu);
356 kfree(head);
357 }
358
359 static void nvme_put_ns_head(struct nvme_ns_head *head)
360 {
361 kref_put(&head->ref, nvme_free_ns_head);
362 }
363
364 static void nvme_free_ns(struct kref *kref)
365 {
366 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
367
368 if (ns->ndev)
369 nvme_nvm_unregister(ns);
370
371 put_disk(ns->disk);
372 nvme_put_ns_head(ns->head);
373 nvme_put_ctrl(ns->ctrl);
374 kfree(ns);
375 }
376
377 static void nvme_put_ns(struct nvme_ns *ns)
378 {
379 kref_put(&ns->kref, nvme_free_ns);
380 }
381
382 struct request *nvme_alloc_request(struct request_queue *q,
383 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
384 {
385 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
386 struct request *req;
387
388 if (qid == NVME_QID_ANY) {
389 req = blk_mq_alloc_request(q, op, flags);
390 } else {
391 req = blk_mq_alloc_request_hctx(q, op, flags,
392 qid ? qid - 1 : 0);
393 }
394 if (IS_ERR(req))
395 return req;
396
397 req->cmd_flags |= REQ_FAILFAST_DRIVER;
398 nvme_req(req)->cmd = cmd;
399
400 return req;
401 }
402 EXPORT_SYMBOL_GPL(nvme_alloc_request);
403
404 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
405 {
406 struct nvme_command c;
407
408 memset(&c, 0, sizeof(c));
409
410 c.directive.opcode = nvme_admin_directive_send;
411 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
412 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
413 c.directive.dtype = NVME_DIR_IDENTIFY;
414 c.directive.tdtype = NVME_DIR_STREAMS;
415 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
416
417 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
418 }
419
420 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
421 {
422 return nvme_toggle_streams(ctrl, false);
423 }
424
425 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
426 {
427 return nvme_toggle_streams(ctrl, true);
428 }
429
430 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
431 struct streams_directive_params *s, u32 nsid)
432 {
433 struct nvme_command c;
434
435 memset(&c, 0, sizeof(c));
436 memset(s, 0, sizeof(*s));
437
438 c.directive.opcode = nvme_admin_directive_recv;
439 c.directive.nsid = cpu_to_le32(nsid);
440 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
441 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
442 c.directive.dtype = NVME_DIR_STREAMS;
443
444 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
445 }
446
447 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
448 {
449 struct streams_directive_params s;
450 int ret;
451
452 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
453 return 0;
454 if (!streams)
455 return 0;
456
457 ret = nvme_enable_streams(ctrl);
458 if (ret)
459 return ret;
460
461 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
462 if (ret)
463 return ret;
464
465 ctrl->nssa = le16_to_cpu(s.nssa);
466 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
467 dev_info(ctrl->device, "too few streams (%u) available\n",
468 ctrl->nssa);
469 nvme_disable_streams(ctrl);
470 return 0;
471 }
472
473 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
474 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
475 return 0;
476 }
477
478 /*
479 * Check if 'req' has a write hint associated with it. If it does, assign
480 * a valid namespace stream to the write.
481 */
482 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
483 struct request *req, u16 *control,
484 u32 *dsmgmt)
485 {
486 enum rw_hint streamid = req->write_hint;
487
488 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
489 streamid = 0;
490 else {
491 streamid--;
492 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
493 return;
494
495 *control |= NVME_RW_DTYPE_STREAMS;
496 *dsmgmt |= streamid << 16;
497 }
498
499 if (streamid < ARRAY_SIZE(req->q->write_hints))
500 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
501 }
502
503 static inline void nvme_setup_flush(struct nvme_ns *ns,
504 struct nvme_command *cmnd)
505 {
506 memset(cmnd, 0, sizeof(*cmnd));
507 cmnd->common.opcode = nvme_cmd_flush;
508 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
509 }
510
511 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
512 struct nvme_command *cmnd)
513 {
514 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
515 struct nvme_dsm_range *range;
516 struct bio *bio;
517
518 range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
519 if (!range)
520 return BLK_STS_RESOURCE;
521
522 __rq_for_each_bio(bio, req) {
523 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
524 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
525
526 if (n < segments) {
527 range[n].cattr = cpu_to_le32(0);
528 range[n].nlb = cpu_to_le32(nlb);
529 range[n].slba = cpu_to_le64(slba);
530 }
531 n++;
532 }
533
534 if (WARN_ON_ONCE(n != segments)) {
535 kfree(range);
536 return BLK_STS_IOERR;
537 }
538
539 memset(cmnd, 0, sizeof(*cmnd));
540 cmnd->dsm.opcode = nvme_cmd_dsm;
541 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
542 cmnd->dsm.nr = cpu_to_le32(segments - 1);
543 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
544
545 req->special_vec.bv_page = virt_to_page(range);
546 req->special_vec.bv_offset = offset_in_page(range);
547 req->special_vec.bv_len = sizeof(*range) * segments;
548 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
549
550 return BLK_STS_OK;
551 }
552
553 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
554 struct request *req, struct nvme_command *cmnd)
555 {
556 struct nvme_ctrl *ctrl = ns->ctrl;
557 u16 control = 0;
558 u32 dsmgmt = 0;
559
560 if (req->cmd_flags & REQ_FUA)
561 control |= NVME_RW_FUA;
562 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
563 control |= NVME_RW_LR;
564
565 if (req->cmd_flags & REQ_RAHEAD)
566 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
567
568 memset(cmnd, 0, sizeof(*cmnd));
569 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
570 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
571 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
572 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
573
574 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
575 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
576
577 if (ns->ms) {
578 /*
579 * If formated with metadata, the block layer always provides a
580 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
581 * we enable the PRACT bit for protection information or set the
582 * namespace capacity to zero to prevent any I/O.
583 */
584 if (!blk_integrity_rq(req)) {
585 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
586 return BLK_STS_NOTSUPP;
587 control |= NVME_RW_PRINFO_PRACT;
588 }
589
590 switch (ns->pi_type) {
591 case NVME_NS_DPS_PI_TYPE3:
592 control |= NVME_RW_PRINFO_PRCHK_GUARD;
593 break;
594 case NVME_NS_DPS_PI_TYPE1:
595 case NVME_NS_DPS_PI_TYPE2:
596 control |= NVME_RW_PRINFO_PRCHK_GUARD |
597 NVME_RW_PRINFO_PRCHK_REF;
598 cmnd->rw.reftag = cpu_to_le32(
599 nvme_block_nr(ns, blk_rq_pos(req)));
600 break;
601 }
602 }
603
604 cmnd->rw.control = cpu_to_le16(control);
605 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
606 return 0;
607 }
608
609 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
610 struct nvme_command *cmd)
611 {
612 blk_status_t ret = BLK_STS_OK;
613
614 if (!(req->rq_flags & RQF_DONTPREP)) {
615 nvme_req(req)->retries = 0;
616 nvme_req(req)->flags = 0;
617 req->rq_flags |= RQF_DONTPREP;
618 }
619
620 switch (req_op(req)) {
621 case REQ_OP_DRV_IN:
622 case REQ_OP_DRV_OUT:
623 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
624 break;
625 case REQ_OP_FLUSH:
626 nvme_setup_flush(ns, cmd);
627 break;
628 case REQ_OP_WRITE_ZEROES:
629 /* currently only aliased to deallocate for a few ctrls: */
630 case REQ_OP_DISCARD:
631 ret = nvme_setup_discard(ns, req, cmd);
632 break;
633 case REQ_OP_READ:
634 case REQ_OP_WRITE:
635 ret = nvme_setup_rw(ns, req, cmd);
636 break;
637 default:
638 WARN_ON_ONCE(1);
639 return BLK_STS_IOERR;
640 }
641
642 cmd->common.command_id = req->tag;
643 if (ns)
644 trace_nvme_setup_nvm_cmd(req->q->id, cmd);
645 else
646 trace_nvme_setup_admin_cmd(cmd);
647 return ret;
648 }
649 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
650
651 /*
652 * Returns 0 on success. If the result is negative, it's a Linux error code;
653 * if the result is positive, it's an NVM Express status code
654 */
655 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
656 union nvme_result *result, void *buffer, unsigned bufflen,
657 unsigned timeout, int qid, int at_head,
658 blk_mq_req_flags_t flags)
659 {
660 struct request *req;
661 int ret;
662
663 req = nvme_alloc_request(q, cmd, flags, qid);
664 if (IS_ERR(req))
665 return PTR_ERR(req);
666
667 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
668
669 if (buffer && bufflen) {
670 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
671 if (ret)
672 goto out;
673 }
674
675 blk_execute_rq(req->q, NULL, req, at_head);
676 if (result)
677 *result = nvme_req(req)->result;
678 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
679 ret = -EINTR;
680 else
681 ret = nvme_req(req)->status;
682 out:
683 blk_mq_free_request(req);
684 return ret;
685 }
686 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
687
688 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
689 void *buffer, unsigned bufflen)
690 {
691 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
692 NVME_QID_ANY, 0, 0);
693 }
694 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
695
696 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
697 unsigned len, u32 seed, bool write)
698 {
699 struct bio_integrity_payload *bip;
700 int ret = -ENOMEM;
701 void *buf;
702
703 buf = kmalloc(len, GFP_KERNEL);
704 if (!buf)
705 goto out;
706
707 ret = -EFAULT;
708 if (write && copy_from_user(buf, ubuf, len))
709 goto out_free_meta;
710
711 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
712 if (IS_ERR(bip)) {
713 ret = PTR_ERR(bip);
714 goto out_free_meta;
715 }
716
717 bip->bip_iter.bi_size = len;
718 bip->bip_iter.bi_sector = seed;
719 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
720 offset_in_page(buf));
721 if (ret == len)
722 return buf;
723 ret = -ENOMEM;
724 out_free_meta:
725 kfree(buf);
726 out:
727 return ERR_PTR(ret);
728 }
729
730 static int nvme_submit_user_cmd(struct request_queue *q,
731 struct nvme_command *cmd, void __user *ubuffer,
732 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
733 u32 meta_seed, u32 *result, unsigned timeout)
734 {
735 bool write = nvme_is_write(cmd);
736 struct nvme_ns *ns = q->queuedata;
737 struct gendisk *disk = ns ? ns->disk : NULL;
738 struct request *req;
739 struct bio *bio = NULL;
740 void *meta = NULL;
741 int ret;
742
743 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
744 if (IS_ERR(req))
745 return PTR_ERR(req);
746
747 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
748
749 if (ubuffer && bufflen) {
750 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
751 GFP_KERNEL);
752 if (ret)
753 goto out;
754 bio = req->bio;
755 bio->bi_disk = disk;
756 if (disk && meta_buffer && meta_len) {
757 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
758 meta_seed, write);
759 if (IS_ERR(meta)) {
760 ret = PTR_ERR(meta);
761 goto out_unmap;
762 }
763 }
764 }
765
766 blk_execute_rq(req->q, disk, req, 0);
767 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
768 ret = -EINTR;
769 else
770 ret = nvme_req(req)->status;
771 if (result)
772 *result = le32_to_cpu(nvme_req(req)->result.u32);
773 if (meta && !ret && !write) {
774 if (copy_to_user(meta_buffer, meta, meta_len))
775 ret = -EFAULT;
776 }
777 kfree(meta);
778 out_unmap:
779 if (bio)
780 blk_rq_unmap_user(bio);
781 out:
782 blk_mq_free_request(req);
783 return ret;
784 }
785
786 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
787 {
788 struct nvme_ctrl *ctrl = rq->end_io_data;
789
790 blk_mq_free_request(rq);
791
792 if (status) {
793 dev_err(ctrl->device,
794 "failed nvme_keep_alive_end_io error=%d\n",
795 status);
796 return;
797 }
798
799 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
800 }
801
802 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
803 {
804 struct request *rq;
805
806 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
807 NVME_QID_ANY);
808 if (IS_ERR(rq))
809 return PTR_ERR(rq);
810
811 rq->timeout = ctrl->kato * HZ;
812 rq->end_io_data = ctrl;
813
814 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
815
816 return 0;
817 }
818
819 static void nvme_keep_alive_work(struct work_struct *work)
820 {
821 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
822 struct nvme_ctrl, ka_work);
823
824 if (nvme_keep_alive(ctrl)) {
825 /* allocation failure, reset the controller */
826 dev_err(ctrl->device, "keep-alive failed\n");
827 nvme_reset_ctrl(ctrl);
828 return;
829 }
830 }
831
832 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
833 {
834 if (unlikely(ctrl->kato == 0))
835 return;
836
837 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
838 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
839 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
840 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
841 }
842 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
843
844 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
845 {
846 if (unlikely(ctrl->kato == 0))
847 return;
848
849 cancel_delayed_work_sync(&ctrl->ka_work);
850 }
851 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
852
853 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
854 {
855 struct nvme_command c = { };
856 int error;
857
858 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
859 c.identify.opcode = nvme_admin_identify;
860 c.identify.cns = NVME_ID_CNS_CTRL;
861
862 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
863 if (!*id)
864 return -ENOMEM;
865
866 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
867 sizeof(struct nvme_id_ctrl));
868 if (error)
869 kfree(*id);
870 return error;
871 }
872
873 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
874 struct nvme_ns_ids *ids)
875 {
876 struct nvme_command c = { };
877 int status;
878 void *data;
879 int pos;
880 int len;
881
882 c.identify.opcode = nvme_admin_identify;
883 c.identify.nsid = cpu_to_le32(nsid);
884 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
885
886 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
887 if (!data)
888 return -ENOMEM;
889
890 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
891 NVME_IDENTIFY_DATA_SIZE);
892 if (status)
893 goto free_data;
894
895 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
896 struct nvme_ns_id_desc *cur = data + pos;
897
898 if (cur->nidl == 0)
899 break;
900
901 switch (cur->nidt) {
902 case NVME_NIDT_EUI64:
903 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
904 dev_warn(ctrl->device,
905 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
906 cur->nidl);
907 goto free_data;
908 }
909 len = NVME_NIDT_EUI64_LEN;
910 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
911 break;
912 case NVME_NIDT_NGUID:
913 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
914 dev_warn(ctrl->device,
915 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
916 cur->nidl);
917 goto free_data;
918 }
919 len = NVME_NIDT_NGUID_LEN;
920 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
921 break;
922 case NVME_NIDT_UUID:
923 if (cur->nidl != NVME_NIDT_UUID_LEN) {
924 dev_warn(ctrl->device,
925 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
926 cur->nidl);
927 goto free_data;
928 }
929 len = NVME_NIDT_UUID_LEN;
930 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
931 break;
932 default:
933 /* Skip unnkown types */
934 len = cur->nidl;
935 break;
936 }
937
938 len += sizeof(*cur);
939 }
940 free_data:
941 kfree(data);
942 return status;
943 }
944
945 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
946 {
947 struct nvme_command c = { };
948
949 c.identify.opcode = nvme_admin_identify;
950 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
951 c.identify.nsid = cpu_to_le32(nsid);
952 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
953 }
954
955 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
956 unsigned nsid)
957 {
958 struct nvme_id_ns *id;
959 struct nvme_command c = { };
960 int error;
961
962 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
963 c.identify.opcode = nvme_admin_identify;
964 c.identify.nsid = cpu_to_le32(nsid);
965 c.identify.cns = NVME_ID_CNS_NS;
966
967 id = kmalloc(sizeof(*id), GFP_KERNEL);
968 if (!id)
969 return NULL;
970
971 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
972 if (error) {
973 dev_warn(ctrl->device, "Identify namespace failed\n");
974 kfree(id);
975 return NULL;
976 }
977
978 return id;
979 }
980
981 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
982 void *buffer, size_t buflen, u32 *result)
983 {
984 struct nvme_command c;
985 union nvme_result res;
986 int ret;
987
988 memset(&c, 0, sizeof(c));
989 c.features.opcode = nvme_admin_set_features;
990 c.features.fid = cpu_to_le32(fid);
991 c.features.dword11 = cpu_to_le32(dword11);
992
993 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
994 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
995 if (ret >= 0 && result)
996 *result = le32_to_cpu(res.u32);
997 return ret;
998 }
999
1000 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1001 {
1002 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1003 u32 result;
1004 int status, nr_io_queues;
1005
1006 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1007 &result);
1008 if (status < 0)
1009 return status;
1010
1011 /*
1012 * Degraded controllers might return an error when setting the queue
1013 * count. We still want to be able to bring them online and offer
1014 * access to the admin queue, as that might be only way to fix them up.
1015 */
1016 if (status > 0) {
1017 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1018 *count = 0;
1019 } else {
1020 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1021 *count = min(*count, nr_io_queues);
1022 }
1023
1024 return 0;
1025 }
1026 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1027
1028 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1029 {
1030 struct nvme_user_io io;
1031 struct nvme_command c;
1032 unsigned length, meta_len;
1033 void __user *metadata;
1034
1035 if (copy_from_user(&io, uio, sizeof(io)))
1036 return -EFAULT;
1037 if (io.flags)
1038 return -EINVAL;
1039
1040 switch (io.opcode) {
1041 case nvme_cmd_write:
1042 case nvme_cmd_read:
1043 case nvme_cmd_compare:
1044 break;
1045 default:
1046 return -EINVAL;
1047 }
1048
1049 length = (io.nblocks + 1) << ns->lba_shift;
1050 meta_len = (io.nblocks + 1) * ns->ms;
1051 metadata = (void __user *)(uintptr_t)io.metadata;
1052
1053 if (ns->ext) {
1054 length += meta_len;
1055 meta_len = 0;
1056 } else if (meta_len) {
1057 if ((io.metadata & 3) || !io.metadata)
1058 return -EINVAL;
1059 }
1060
1061 memset(&c, 0, sizeof(c));
1062 c.rw.opcode = io.opcode;
1063 c.rw.flags = io.flags;
1064 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1065 c.rw.slba = cpu_to_le64(io.slba);
1066 c.rw.length = cpu_to_le16(io.nblocks);
1067 c.rw.control = cpu_to_le16(io.control);
1068 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1069 c.rw.reftag = cpu_to_le32(io.reftag);
1070 c.rw.apptag = cpu_to_le16(io.apptag);
1071 c.rw.appmask = cpu_to_le16(io.appmask);
1072
1073 return nvme_submit_user_cmd(ns->queue, &c,
1074 (void __user *)(uintptr_t)io.addr, length,
1075 metadata, meta_len, io.slba, NULL, 0);
1076 }
1077
1078 static u32 nvme_known_admin_effects(u8 opcode)
1079 {
1080 switch (opcode) {
1081 case nvme_admin_format_nvm:
1082 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1083 NVME_CMD_EFFECTS_CSE_MASK;
1084 case nvme_admin_sanitize_nvm:
1085 return NVME_CMD_EFFECTS_CSE_MASK;
1086 default:
1087 break;
1088 }
1089 return 0;
1090 }
1091
1092 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1093 u8 opcode)
1094 {
1095 u32 effects = 0;
1096
1097 if (ns) {
1098 if (ctrl->effects)
1099 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1100 if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1101 dev_warn(ctrl->device,
1102 "IO command:%02x has unhandled effects:%08x\n",
1103 opcode, effects);
1104 return 0;
1105 }
1106
1107 if (ctrl->effects)
1108 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1109 else
1110 effects = nvme_known_admin_effects(opcode);
1111
1112 /*
1113 * For simplicity, IO to all namespaces is quiesced even if the command
1114 * effects say only one namespace is affected.
1115 */
1116 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1117 nvme_start_freeze(ctrl);
1118 nvme_wait_freeze(ctrl);
1119 }
1120 return effects;
1121 }
1122
1123 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1124 {
1125 struct nvme_ns *ns, *next;
1126 LIST_HEAD(rm_list);
1127
1128 mutex_lock(&ctrl->namespaces_mutex);
1129 list_for_each_entry(ns, &ctrl->namespaces, list) {
1130 if (ns->disk && nvme_revalidate_disk(ns->disk)) {
1131 list_move_tail(&ns->list, &rm_list);
1132 }
1133 }
1134 mutex_unlock(&ctrl->namespaces_mutex);
1135
1136 list_for_each_entry_safe(ns, next, &rm_list, list)
1137 nvme_ns_remove(ns);
1138 }
1139
1140 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1141 {
1142 /*
1143 * Revalidate LBA changes prior to unfreezing. This is necessary to
1144 * prevent memory corruption if a logical block size was changed by
1145 * this command.
1146 */
1147 if (effects & NVME_CMD_EFFECTS_LBCC)
1148 nvme_update_formats(ctrl);
1149 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK))
1150 nvme_unfreeze(ctrl);
1151 if (effects & NVME_CMD_EFFECTS_CCC)
1152 nvme_init_identify(ctrl);
1153 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1154 nvme_queue_scan(ctrl);
1155 }
1156
1157 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1158 struct nvme_passthru_cmd __user *ucmd)
1159 {
1160 struct nvme_passthru_cmd cmd;
1161 struct nvme_command c;
1162 unsigned timeout = 0;
1163 u32 effects;
1164 int status;
1165
1166 if (!capable(CAP_SYS_ADMIN))
1167 return -EACCES;
1168 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1169 return -EFAULT;
1170 if (cmd.flags)
1171 return -EINVAL;
1172
1173 memset(&c, 0, sizeof(c));
1174 c.common.opcode = cmd.opcode;
1175 c.common.flags = cmd.flags;
1176 c.common.nsid = cpu_to_le32(cmd.nsid);
1177 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1178 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1179 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1180 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1181 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1182 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1183 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1184 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1185
1186 if (cmd.timeout_ms)
1187 timeout = msecs_to_jiffies(cmd.timeout_ms);
1188
1189 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1190 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1191 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1192 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata,
1193 0, &cmd.result, timeout);
1194 nvme_passthru_end(ctrl, effects);
1195
1196 if (status >= 0) {
1197 if (put_user(cmd.result, &ucmd->result))
1198 return -EFAULT;
1199 }
1200
1201 return status;
1202 }
1203
1204 /*
1205 * Issue ioctl requests on the first available path. Note that unlike normal
1206 * block layer requests we will not retry failed request on another controller.
1207 */
1208 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1209 struct nvme_ns_head **head, int *srcu_idx)
1210 {
1211 #ifdef CONFIG_NVME_MULTIPATH
1212 if (disk->fops == &nvme_ns_head_ops) {
1213 *head = disk->private_data;
1214 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1215 return nvme_find_path(*head);
1216 }
1217 #endif
1218 *head = NULL;
1219 *srcu_idx = -1;
1220 return disk->private_data;
1221 }
1222
1223 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1224 {
1225 if (head)
1226 srcu_read_unlock(&head->srcu, idx);
1227 }
1228
1229 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1230 {
1231 switch (cmd) {
1232 case NVME_IOCTL_ID:
1233 force_successful_syscall_return();
1234 return ns->head->ns_id;
1235 case NVME_IOCTL_ADMIN_CMD:
1236 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1237 case NVME_IOCTL_IO_CMD:
1238 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1239 case NVME_IOCTL_SUBMIT_IO:
1240 return nvme_submit_io(ns, (void __user *)arg);
1241 default:
1242 #ifdef CONFIG_NVM
1243 if (ns->ndev)
1244 return nvme_nvm_ioctl(ns, cmd, arg);
1245 #endif
1246 if (is_sed_ioctl(cmd))
1247 return sed_ioctl(ns->ctrl->opal_dev, cmd,
1248 (void __user *) arg);
1249 return -ENOTTY;
1250 }
1251 }
1252
1253 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1254 unsigned int cmd, unsigned long arg)
1255 {
1256 struct nvme_ns_head *head = NULL;
1257 struct nvme_ns *ns;
1258 int srcu_idx, ret;
1259
1260 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1261 if (unlikely(!ns))
1262 ret = -EWOULDBLOCK;
1263 else
1264 ret = nvme_ns_ioctl(ns, cmd, arg);
1265 nvme_put_ns_from_disk(head, srcu_idx);
1266 return ret;
1267 }
1268
1269 static int nvme_open(struct block_device *bdev, fmode_t mode)
1270 {
1271 struct nvme_ns *ns = bdev->bd_disk->private_data;
1272
1273 #ifdef CONFIG_NVME_MULTIPATH
1274 /* should never be called due to GENHD_FL_HIDDEN */
1275 if (WARN_ON_ONCE(ns->head->disk))
1276 goto fail;
1277 #endif
1278 if (!kref_get_unless_zero(&ns->kref))
1279 goto fail;
1280 if (!try_module_get(ns->ctrl->ops->module))
1281 goto fail_put_ns;
1282
1283 return 0;
1284
1285 fail_put_ns:
1286 nvme_put_ns(ns);
1287 fail:
1288 return -ENXIO;
1289 }
1290
1291 static void nvme_release(struct gendisk *disk, fmode_t mode)
1292 {
1293 struct nvme_ns *ns = disk->private_data;
1294
1295 module_put(ns->ctrl->ops->module);
1296 nvme_put_ns(ns);
1297 }
1298
1299 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1300 {
1301 /* some standard values */
1302 geo->heads = 1 << 6;
1303 geo->sectors = 1 << 5;
1304 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1305 return 0;
1306 }
1307
1308 #ifdef CONFIG_BLK_DEV_INTEGRITY
1309 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1310 {
1311 struct blk_integrity integrity;
1312
1313 memset(&integrity, 0, sizeof(integrity));
1314 switch (pi_type) {
1315 case NVME_NS_DPS_PI_TYPE3:
1316 integrity.profile = &t10_pi_type3_crc;
1317 integrity.tag_size = sizeof(u16) + sizeof(u32);
1318 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1319 break;
1320 case NVME_NS_DPS_PI_TYPE1:
1321 case NVME_NS_DPS_PI_TYPE2:
1322 integrity.profile = &t10_pi_type1_crc;
1323 integrity.tag_size = sizeof(u16);
1324 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1325 break;
1326 default:
1327 integrity.profile = NULL;
1328 break;
1329 }
1330 integrity.tuple_size = ms;
1331 blk_integrity_register(disk, &integrity);
1332 blk_queue_max_integrity_segments(disk->queue, 1);
1333 }
1334 #else
1335 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1336 {
1337 }
1338 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1339
1340 static void nvme_set_chunk_size(struct nvme_ns *ns)
1341 {
1342 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1343 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1344 }
1345
1346 static void nvme_config_discard(struct nvme_ctrl *ctrl,
1347 unsigned stream_alignment, struct request_queue *queue)
1348 {
1349 u32 size = queue_logical_block_size(queue);
1350
1351 if (stream_alignment)
1352 size *= stream_alignment;
1353
1354 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1355 NVME_DSM_MAX_RANGES);
1356
1357 queue->limits.discard_alignment = 0;
1358 queue->limits.discard_granularity = size;
1359
1360 blk_queue_max_discard_sectors(queue, UINT_MAX);
1361 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1362 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, queue);
1363
1364 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1365 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1366 }
1367
1368 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1369 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1370 {
1371 memset(ids, 0, sizeof(*ids));
1372
1373 if (ctrl->vs >= NVME_VS(1, 1, 0))
1374 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1375 if (ctrl->vs >= NVME_VS(1, 2, 0))
1376 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1377 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1378 /* Don't treat error as fatal we potentially
1379 * already have a NGUID or EUI-64
1380 */
1381 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1382 dev_warn(ctrl->device,
1383 "%s: Identify Descriptors failed\n", __func__);
1384 }
1385 }
1386
1387 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1388 {
1389 return !uuid_is_null(&ids->uuid) ||
1390 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1391 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1392 }
1393
1394 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1395 {
1396 return uuid_equal(&a->uuid, &b->uuid) &&
1397 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1398 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1399 }
1400
1401 static void nvme_update_disk_info(struct gendisk *disk,
1402 struct nvme_ns *ns, struct nvme_id_ns *id)
1403 {
1404 sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1405 unsigned short bs = 1 << ns->lba_shift;
1406 unsigned stream_alignment = 0;
1407
1408 if (ns->ctrl->nr_streams && ns->sws && ns->sgs)
1409 stream_alignment = ns->sws * ns->sgs;
1410
1411 blk_mq_freeze_queue(disk->queue);
1412 blk_integrity_unregister(disk);
1413
1414 blk_queue_logical_block_size(disk->queue, bs);
1415 blk_queue_physical_block_size(disk->queue, bs);
1416 blk_queue_io_min(disk->queue, bs);
1417
1418 if (ns->ms && !ns->ext &&
1419 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1420 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1421 if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1422 capacity = 0;
1423 set_capacity(disk, capacity);
1424
1425 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1426 nvme_config_discard(ns->ctrl, stream_alignment, disk->queue);
1427 blk_mq_unfreeze_queue(disk->queue);
1428 }
1429
1430 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1431 {
1432 struct nvme_ns *ns = disk->private_data;
1433
1434 /*
1435 * If identify namespace failed, use default 512 byte block size so
1436 * block layer can use before failing read/write for 0 capacity.
1437 */
1438 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1439 if (ns->lba_shift == 0)
1440 ns->lba_shift = 9;
1441 ns->noiob = le16_to_cpu(id->noiob);
1442 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1443 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1444 /* the PI implementation requires metadata equal t10 pi tuple size */
1445 if (ns->ms == sizeof(struct t10_pi_tuple))
1446 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1447 else
1448 ns->pi_type = 0;
1449
1450 if (ns->noiob)
1451 nvme_set_chunk_size(ns);
1452 nvme_update_disk_info(disk, ns, id);
1453 #ifdef CONFIG_NVME_MULTIPATH
1454 if (ns->head->disk)
1455 nvme_update_disk_info(ns->head->disk, ns, id);
1456 #endif
1457 }
1458
1459 static int nvme_revalidate_disk(struct gendisk *disk)
1460 {
1461 struct nvme_ns *ns = disk->private_data;
1462 struct nvme_ctrl *ctrl = ns->ctrl;
1463 struct nvme_id_ns *id;
1464 struct nvme_ns_ids ids;
1465 int ret = 0;
1466
1467 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1468 set_capacity(disk, 0);
1469 return -ENODEV;
1470 }
1471
1472 id = nvme_identify_ns(ctrl, ns->head->ns_id);
1473 if (!id)
1474 return -ENODEV;
1475
1476 if (id->ncap == 0) {
1477 ret = -ENODEV;
1478 goto out;
1479 }
1480
1481 __nvme_revalidate_disk(disk, id);
1482 nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1483 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1484 dev_err(ctrl->device,
1485 "identifiers changed for nsid %d\n", ns->head->ns_id);
1486 ret = -ENODEV;
1487 }
1488
1489 out:
1490 kfree(id);
1491 return ret;
1492 }
1493
1494 static char nvme_pr_type(enum pr_type type)
1495 {
1496 switch (type) {
1497 case PR_WRITE_EXCLUSIVE:
1498 return 1;
1499 case PR_EXCLUSIVE_ACCESS:
1500 return 2;
1501 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1502 return 3;
1503 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1504 return 4;
1505 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1506 return 5;
1507 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1508 return 6;
1509 default:
1510 return 0;
1511 }
1512 };
1513
1514 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1515 u64 key, u64 sa_key, u8 op)
1516 {
1517 struct nvme_ns_head *head = NULL;
1518 struct nvme_ns *ns;
1519 struct nvme_command c;
1520 int srcu_idx, ret;
1521 u8 data[16] = { 0, };
1522
1523 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1524 if (unlikely(!ns))
1525 return -EWOULDBLOCK;
1526
1527 put_unaligned_le64(key, &data[0]);
1528 put_unaligned_le64(sa_key, &data[8]);
1529
1530 memset(&c, 0, sizeof(c));
1531 c.common.opcode = op;
1532 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1533 c.common.cdw10[0] = cpu_to_le32(cdw10);
1534
1535 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1536 nvme_put_ns_from_disk(head, srcu_idx);
1537 return ret;
1538 }
1539
1540 static int nvme_pr_register(struct block_device *bdev, u64 old,
1541 u64 new, unsigned flags)
1542 {
1543 u32 cdw10;
1544
1545 if (flags & ~PR_FL_IGNORE_KEY)
1546 return -EOPNOTSUPP;
1547
1548 cdw10 = old ? 2 : 0;
1549 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1550 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1551 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1552 }
1553
1554 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1555 enum pr_type type, unsigned flags)
1556 {
1557 u32 cdw10;
1558
1559 if (flags & ~PR_FL_IGNORE_KEY)
1560 return -EOPNOTSUPP;
1561
1562 cdw10 = nvme_pr_type(type) << 8;
1563 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1564 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1565 }
1566
1567 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1568 enum pr_type type, bool abort)
1569 {
1570 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1571 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1572 }
1573
1574 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1575 {
1576 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1577 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1578 }
1579
1580 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1581 {
1582 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1583 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1584 }
1585
1586 static const struct pr_ops nvme_pr_ops = {
1587 .pr_register = nvme_pr_register,
1588 .pr_reserve = nvme_pr_reserve,
1589 .pr_release = nvme_pr_release,
1590 .pr_preempt = nvme_pr_preempt,
1591 .pr_clear = nvme_pr_clear,
1592 };
1593
1594 #ifdef CONFIG_BLK_SED_OPAL
1595 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1596 bool send)
1597 {
1598 struct nvme_ctrl *ctrl = data;
1599 struct nvme_command cmd;
1600
1601 memset(&cmd, 0, sizeof(cmd));
1602 if (send)
1603 cmd.common.opcode = nvme_admin_security_send;
1604 else
1605 cmd.common.opcode = nvme_admin_security_recv;
1606 cmd.common.nsid = 0;
1607 cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1608 cmd.common.cdw10[1] = cpu_to_le32(len);
1609
1610 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1611 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1612 }
1613 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1614 #endif /* CONFIG_BLK_SED_OPAL */
1615
1616 static const struct block_device_operations nvme_fops = {
1617 .owner = THIS_MODULE,
1618 .ioctl = nvme_ioctl,
1619 .compat_ioctl = nvme_ioctl,
1620 .open = nvme_open,
1621 .release = nvme_release,
1622 .getgeo = nvme_getgeo,
1623 .revalidate_disk= nvme_revalidate_disk,
1624 .pr_ops = &nvme_pr_ops,
1625 };
1626
1627 #ifdef CONFIG_NVME_MULTIPATH
1628 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1629 {
1630 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1631
1632 if (!kref_get_unless_zero(&head->ref))
1633 return -ENXIO;
1634 return 0;
1635 }
1636
1637 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1638 {
1639 nvme_put_ns_head(disk->private_data);
1640 }
1641
1642 const struct block_device_operations nvme_ns_head_ops = {
1643 .owner = THIS_MODULE,
1644 .open = nvme_ns_head_open,
1645 .release = nvme_ns_head_release,
1646 .ioctl = nvme_ioctl,
1647 .compat_ioctl = nvme_ioctl,
1648 .getgeo = nvme_getgeo,
1649 .pr_ops = &nvme_pr_ops,
1650 };
1651 #endif /* CONFIG_NVME_MULTIPATH */
1652
1653 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1654 {
1655 unsigned long timeout =
1656 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1657 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1658 int ret;
1659
1660 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1661 if (csts == ~0)
1662 return -ENODEV;
1663 if ((csts & NVME_CSTS_RDY) == bit)
1664 break;
1665
1666 msleep(100);
1667 if (fatal_signal_pending(current))
1668 return -EINTR;
1669 if (time_after(jiffies, timeout)) {
1670 dev_err(ctrl->device,
1671 "Device not ready; aborting %s\n", enabled ?
1672 "initialisation" : "reset");
1673 return -ENODEV;
1674 }
1675 }
1676
1677 return ret;
1678 }
1679
1680 /*
1681 * If the device has been passed off to us in an enabled state, just clear
1682 * the enabled bit. The spec says we should set the 'shutdown notification
1683 * bits', but doing so may cause the device to complete commands to the
1684 * admin queue ... and we don't know what memory that might be pointing at!
1685 */
1686 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1687 {
1688 int ret;
1689
1690 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1691 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1692
1693 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1694 if (ret)
1695 return ret;
1696
1697 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1698 msleep(NVME_QUIRK_DELAY_AMOUNT);
1699
1700 return nvme_wait_ready(ctrl, cap, false);
1701 }
1702 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1703
1704 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1705 {
1706 /*
1707 * Default to a 4K page size, with the intention to update this
1708 * path in the future to accomodate architectures with differing
1709 * kernel and IO page sizes.
1710 */
1711 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1712 int ret;
1713
1714 if (page_shift < dev_page_min) {
1715 dev_err(ctrl->device,
1716 "Minimum device page size %u too large for host (%u)\n",
1717 1 << dev_page_min, 1 << page_shift);
1718 return -ENODEV;
1719 }
1720
1721 ctrl->page_size = 1 << page_shift;
1722
1723 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1724 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1725 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1726 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1727 ctrl->ctrl_config |= NVME_CC_ENABLE;
1728
1729 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1730 if (ret)
1731 return ret;
1732 return nvme_wait_ready(ctrl, cap, true);
1733 }
1734 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1735
1736 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1737 {
1738 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1739 u32 csts;
1740 int ret;
1741
1742 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1743 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1744
1745 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1746 if (ret)
1747 return ret;
1748
1749 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1750 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1751 break;
1752
1753 msleep(100);
1754 if (fatal_signal_pending(current))
1755 return -EINTR;
1756 if (time_after(jiffies, timeout)) {
1757 dev_err(ctrl->device,
1758 "Device shutdown incomplete; abort shutdown\n");
1759 return -ENODEV;
1760 }
1761 }
1762
1763 return ret;
1764 }
1765 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1766
1767 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1768 struct request_queue *q)
1769 {
1770 bool vwc = false;
1771
1772 if (ctrl->max_hw_sectors) {
1773 u32 max_segments =
1774 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1775
1776 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1777 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1778 }
1779 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1780 is_power_of_2(ctrl->max_hw_sectors))
1781 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1782 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1783 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1784 vwc = true;
1785 blk_queue_write_cache(q, vwc, vwc);
1786 }
1787
1788 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1789 {
1790 __le64 ts;
1791 int ret;
1792
1793 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1794 return 0;
1795
1796 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1797 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1798 NULL);
1799 if (ret)
1800 dev_warn_once(ctrl->device,
1801 "could not set timestamp (%d)\n", ret);
1802 return ret;
1803 }
1804
1805 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1806 {
1807 /*
1808 * APST (Autonomous Power State Transition) lets us program a
1809 * table of power state transitions that the controller will
1810 * perform automatically. We configure it with a simple
1811 * heuristic: we are willing to spend at most 2% of the time
1812 * transitioning between power states. Therefore, when running
1813 * in any given state, we will enter the next lower-power
1814 * non-operational state after waiting 50 * (enlat + exlat)
1815 * microseconds, as long as that state's exit latency is under
1816 * the requested maximum latency.
1817 *
1818 * We will not autonomously enter any non-operational state for
1819 * which the total latency exceeds ps_max_latency_us. Users
1820 * can set ps_max_latency_us to zero to turn off APST.
1821 */
1822
1823 unsigned apste;
1824 struct nvme_feat_auto_pst *table;
1825 u64 max_lat_us = 0;
1826 int max_ps = -1;
1827 int ret;
1828
1829 /*
1830 * If APST isn't supported or if we haven't been initialized yet,
1831 * then don't do anything.
1832 */
1833 if (!ctrl->apsta)
1834 return 0;
1835
1836 if (ctrl->npss > 31) {
1837 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1838 return 0;
1839 }
1840
1841 table = kzalloc(sizeof(*table), GFP_KERNEL);
1842 if (!table)
1843 return 0;
1844
1845 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1846 /* Turn off APST. */
1847 apste = 0;
1848 dev_dbg(ctrl->device, "APST disabled\n");
1849 } else {
1850 __le64 target = cpu_to_le64(0);
1851 int state;
1852
1853 /*
1854 * Walk through all states from lowest- to highest-power.
1855 * According to the spec, lower-numbered states use more
1856 * power. NPSS, despite the name, is the index of the
1857 * lowest-power state, not the number of states.
1858 */
1859 for (state = (int)ctrl->npss; state >= 0; state--) {
1860 u64 total_latency_us, exit_latency_us, transition_ms;
1861
1862 if (target)
1863 table->entries[state] = target;
1864
1865 /*
1866 * Don't allow transitions to the deepest state
1867 * if it's quirked off.
1868 */
1869 if (state == ctrl->npss &&
1870 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1871 continue;
1872
1873 /*
1874 * Is this state a useful non-operational state for
1875 * higher-power states to autonomously transition to?
1876 */
1877 if (!(ctrl->psd[state].flags &
1878 NVME_PS_FLAGS_NON_OP_STATE))
1879 continue;
1880
1881 exit_latency_us =
1882 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1883 if (exit_latency_us > ctrl->ps_max_latency_us)
1884 continue;
1885
1886 total_latency_us =
1887 exit_latency_us +
1888 le32_to_cpu(ctrl->psd[state].entry_lat);
1889
1890 /*
1891 * This state is good. Use it as the APST idle
1892 * target for higher power states.
1893 */
1894 transition_ms = total_latency_us + 19;
1895 do_div(transition_ms, 20);
1896 if (transition_ms > (1 << 24) - 1)
1897 transition_ms = (1 << 24) - 1;
1898
1899 target = cpu_to_le64((state << 3) |
1900 (transition_ms << 8));
1901
1902 if (max_ps == -1)
1903 max_ps = state;
1904
1905 if (total_latency_us > max_lat_us)
1906 max_lat_us = total_latency_us;
1907 }
1908
1909 apste = 1;
1910
1911 if (max_ps == -1) {
1912 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1913 } else {
1914 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1915 max_ps, max_lat_us, (int)sizeof(*table), table);
1916 }
1917 }
1918
1919 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1920 table, sizeof(*table), NULL);
1921 if (ret)
1922 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1923
1924 kfree(table);
1925 return ret;
1926 }
1927
1928 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1929 {
1930 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1931 u64 latency;
1932
1933 switch (val) {
1934 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1935 case PM_QOS_LATENCY_ANY:
1936 latency = U64_MAX;
1937 break;
1938
1939 default:
1940 latency = val;
1941 }
1942
1943 if (ctrl->ps_max_latency_us != latency) {
1944 ctrl->ps_max_latency_us = latency;
1945 nvme_configure_apst(ctrl);
1946 }
1947 }
1948
1949 struct nvme_core_quirk_entry {
1950 /*
1951 * NVMe model and firmware strings are padded with spaces. For
1952 * simplicity, strings in the quirk table are padded with NULLs
1953 * instead.
1954 */
1955 u16 vid;
1956 const char *mn;
1957 const char *fr;
1958 unsigned long quirks;
1959 };
1960
1961 static const struct nvme_core_quirk_entry core_quirks[] = {
1962 {
1963 /*
1964 * This Toshiba device seems to die using any APST states. See:
1965 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1966 */
1967 .vid = 0x1179,
1968 .mn = "THNSF5256GPUK TOSHIBA",
1969 .quirks = NVME_QUIRK_NO_APST,
1970 }
1971 };
1972
1973 /* match is null-terminated but idstr is space-padded. */
1974 static bool string_matches(const char *idstr, const char *match, size_t len)
1975 {
1976 size_t matchlen;
1977
1978 if (!match)
1979 return true;
1980
1981 matchlen = strlen(match);
1982 WARN_ON_ONCE(matchlen > len);
1983
1984 if (memcmp(idstr, match, matchlen))
1985 return false;
1986
1987 for (; matchlen < len; matchlen++)
1988 if (idstr[matchlen] != ' ')
1989 return false;
1990
1991 return true;
1992 }
1993
1994 static bool quirk_matches(const struct nvme_id_ctrl *id,
1995 const struct nvme_core_quirk_entry *q)
1996 {
1997 return q->vid == le16_to_cpu(id->vid) &&
1998 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1999 string_matches(id->fr, q->fr, sizeof(id->fr));
2000 }
2001
2002 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2003 struct nvme_id_ctrl *id)
2004 {
2005 size_t nqnlen;
2006 int off;
2007
2008 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2009 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2010 strncpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2011 return;
2012 }
2013
2014 if (ctrl->vs >= NVME_VS(1, 2, 1))
2015 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2016
2017 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2018 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2019 "nqn.2014.08.org.nvmexpress:%4x%4x",
2020 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2021 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2022 off += sizeof(id->sn);
2023 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2024 off += sizeof(id->mn);
2025 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2026 }
2027
2028 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2029 {
2030 ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2031 kfree(subsys);
2032 }
2033
2034 static void nvme_release_subsystem(struct device *dev)
2035 {
2036 __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2037 }
2038
2039 static void nvme_destroy_subsystem(struct kref *ref)
2040 {
2041 struct nvme_subsystem *subsys =
2042 container_of(ref, struct nvme_subsystem, ref);
2043
2044 mutex_lock(&nvme_subsystems_lock);
2045 list_del(&subsys->entry);
2046 mutex_unlock(&nvme_subsystems_lock);
2047
2048 ida_destroy(&subsys->ns_ida);
2049 device_del(&subsys->dev);
2050 put_device(&subsys->dev);
2051 }
2052
2053 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2054 {
2055 kref_put(&subsys->ref, nvme_destroy_subsystem);
2056 }
2057
2058 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2059 {
2060 struct nvme_subsystem *subsys;
2061
2062 lockdep_assert_held(&nvme_subsystems_lock);
2063
2064 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2065 if (strcmp(subsys->subnqn, subsysnqn))
2066 continue;
2067 if (!kref_get_unless_zero(&subsys->ref))
2068 continue;
2069 return subsys;
2070 }
2071
2072 return NULL;
2073 }
2074
2075 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2076 struct device_attribute subsys_attr_##_name = \
2077 __ATTR(_name, _mode, _show, NULL)
2078
2079 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2080 struct device_attribute *attr,
2081 char *buf)
2082 {
2083 struct nvme_subsystem *subsys =
2084 container_of(dev, struct nvme_subsystem, dev);
2085
2086 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2087 }
2088 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2089
2090 #define nvme_subsys_show_str_function(field) \
2091 static ssize_t subsys_##field##_show(struct device *dev, \
2092 struct device_attribute *attr, char *buf) \
2093 { \
2094 struct nvme_subsystem *subsys = \
2095 container_of(dev, struct nvme_subsystem, dev); \
2096 return sprintf(buf, "%.*s\n", \
2097 (int)sizeof(subsys->field), subsys->field); \
2098 } \
2099 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2100
2101 nvme_subsys_show_str_function(model);
2102 nvme_subsys_show_str_function(serial);
2103 nvme_subsys_show_str_function(firmware_rev);
2104
2105 static struct attribute *nvme_subsys_attrs[] = {
2106 &subsys_attr_model.attr,
2107 &subsys_attr_serial.attr,
2108 &subsys_attr_firmware_rev.attr,
2109 &subsys_attr_subsysnqn.attr,
2110 NULL,
2111 };
2112
2113 static struct attribute_group nvme_subsys_attrs_group = {
2114 .attrs = nvme_subsys_attrs,
2115 };
2116
2117 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2118 &nvme_subsys_attrs_group,
2119 NULL,
2120 };
2121
2122 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2123 {
2124 int count = 0;
2125 struct nvme_ctrl *ctrl;
2126
2127 mutex_lock(&subsys->lock);
2128 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2129 if (ctrl->state != NVME_CTRL_DELETING &&
2130 ctrl->state != NVME_CTRL_DEAD)
2131 count++;
2132 }
2133 mutex_unlock(&subsys->lock);
2134
2135 return count;
2136 }
2137
2138 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2139 {
2140 struct nvme_subsystem *subsys, *found;
2141 int ret;
2142
2143 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2144 if (!subsys)
2145 return -ENOMEM;
2146 ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2147 if (ret < 0) {
2148 kfree(subsys);
2149 return ret;
2150 }
2151 subsys->instance = ret;
2152 mutex_init(&subsys->lock);
2153 kref_init(&subsys->ref);
2154 INIT_LIST_HEAD(&subsys->ctrls);
2155 INIT_LIST_HEAD(&subsys->nsheads);
2156 nvme_init_subnqn(subsys, ctrl, id);
2157 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2158 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2159 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2160 subsys->vendor_id = le16_to_cpu(id->vid);
2161 subsys->cmic = id->cmic;
2162
2163 subsys->dev.class = nvme_subsys_class;
2164 subsys->dev.release = nvme_release_subsystem;
2165 subsys->dev.groups = nvme_subsys_attrs_groups;
2166 dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2167 device_initialize(&subsys->dev);
2168
2169 mutex_lock(&nvme_subsystems_lock);
2170 found = __nvme_find_get_subsystem(subsys->subnqn);
2171 if (found) {
2172 /*
2173 * Verify that the subsystem actually supports multiple
2174 * controllers, else bail out.
2175 */
2176 if (nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2177 dev_err(ctrl->device,
2178 "ignoring ctrl due to duplicate subnqn (%s).\n",
2179 found->subnqn);
2180 nvme_put_subsystem(found);
2181 ret = -EINVAL;
2182 goto out_unlock;
2183 }
2184
2185 __nvme_release_subsystem(subsys);
2186 subsys = found;
2187 } else {
2188 ret = device_add(&subsys->dev);
2189 if (ret) {
2190 dev_err(ctrl->device,
2191 "failed to register subsystem device.\n");
2192 goto out_unlock;
2193 }
2194 ida_init(&subsys->ns_ida);
2195 list_add_tail(&subsys->entry, &nvme_subsystems);
2196 }
2197
2198 ctrl->subsys = subsys;
2199 mutex_unlock(&nvme_subsystems_lock);
2200
2201 if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2202 dev_name(ctrl->device))) {
2203 dev_err(ctrl->device,
2204 "failed to create sysfs link from subsystem.\n");
2205 /* the transport driver will eventually put the subsystem */
2206 return -EINVAL;
2207 }
2208
2209 mutex_lock(&subsys->lock);
2210 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2211 mutex_unlock(&subsys->lock);
2212
2213 return 0;
2214
2215 out_unlock:
2216 mutex_unlock(&nvme_subsystems_lock);
2217 put_device(&subsys->dev);
2218 return ret;
2219 }
2220
2221 static int nvme_get_log(struct nvme_ctrl *ctrl, u8 log_page, void *log,
2222 size_t size)
2223 {
2224 struct nvme_command c = { };
2225
2226 c.common.opcode = nvme_admin_get_log_page;
2227 c.common.nsid = cpu_to_le32(NVME_NSID_ALL);
2228 c.common.cdw10[0] = nvme_get_log_dw10(log_page, size);
2229
2230 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2231 }
2232
2233 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2234 {
2235 int ret;
2236
2237 if (!ctrl->effects)
2238 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2239
2240 if (!ctrl->effects)
2241 return 0;
2242
2243 ret = nvme_get_log(ctrl, NVME_LOG_CMD_EFFECTS, ctrl->effects,
2244 sizeof(*ctrl->effects));
2245 if (ret) {
2246 kfree(ctrl->effects);
2247 ctrl->effects = NULL;
2248 }
2249 return ret;
2250 }
2251
2252 /*
2253 * Initialize the cached copies of the Identify data and various controller
2254 * register in our nvme_ctrl structure. This should be called as soon as
2255 * the admin queue is fully up and running.
2256 */
2257 int nvme_init_identify(struct nvme_ctrl *ctrl)
2258 {
2259 struct nvme_id_ctrl *id;
2260 u64 cap;
2261 int ret, page_shift;
2262 u32 max_hw_sectors;
2263 bool prev_apst_enabled;
2264
2265 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2266 if (ret) {
2267 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2268 return ret;
2269 }
2270
2271 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2272 if (ret) {
2273 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2274 return ret;
2275 }
2276 page_shift = NVME_CAP_MPSMIN(cap) + 12;
2277
2278 if (ctrl->vs >= NVME_VS(1, 1, 0))
2279 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2280
2281 ret = nvme_identify_ctrl(ctrl, &id);
2282 if (ret) {
2283 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2284 return -EIO;
2285 }
2286
2287 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2288 ret = nvme_get_effects_log(ctrl);
2289 if (ret < 0)
2290 return ret;
2291 }
2292
2293 if (!ctrl->identified) {
2294 int i;
2295
2296 ret = nvme_init_subsystem(ctrl, id);
2297 if (ret)
2298 goto out_free;
2299
2300 /*
2301 * Check for quirks. Quirk can depend on firmware version,
2302 * so, in principle, the set of quirks present can change
2303 * across a reset. As a possible future enhancement, we
2304 * could re-scan for quirks every time we reinitialize
2305 * the device, but we'd have to make sure that the driver
2306 * behaves intelligently if the quirks change.
2307 */
2308 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2309 if (quirk_matches(id, &core_quirks[i]))
2310 ctrl->quirks |= core_quirks[i].quirks;
2311 }
2312 }
2313
2314 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2315 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2316 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2317 }
2318
2319 ctrl->oacs = le16_to_cpu(id->oacs);
2320 ctrl->oncs = le16_to_cpup(&id->oncs);
2321 atomic_set(&ctrl->abort_limit, id->acl + 1);
2322 ctrl->vwc = id->vwc;
2323 ctrl->cntlid = le16_to_cpup(&id->cntlid);
2324 if (id->mdts)
2325 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2326 else
2327 max_hw_sectors = UINT_MAX;
2328 ctrl->max_hw_sectors =
2329 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2330
2331 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2332 ctrl->sgls = le32_to_cpu(id->sgls);
2333 ctrl->kas = le16_to_cpu(id->kas);
2334
2335 if (id->rtd3e) {
2336 /* us -> s */
2337 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2338
2339 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2340 shutdown_timeout, 60);
2341
2342 if (ctrl->shutdown_timeout != shutdown_timeout)
2343 dev_info(ctrl->device,
2344 "Shutdown timeout set to %u seconds\n",
2345 ctrl->shutdown_timeout);
2346 } else
2347 ctrl->shutdown_timeout = shutdown_timeout;
2348
2349 ctrl->npss = id->npss;
2350 ctrl->apsta = id->apsta;
2351 prev_apst_enabled = ctrl->apst_enabled;
2352 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2353 if (force_apst && id->apsta) {
2354 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2355 ctrl->apst_enabled = true;
2356 } else {
2357 ctrl->apst_enabled = false;
2358 }
2359 } else {
2360 ctrl->apst_enabled = id->apsta;
2361 }
2362 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2363
2364 if (ctrl->ops->flags & NVME_F_FABRICS) {
2365 ctrl->icdoff = le16_to_cpu(id->icdoff);
2366 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2367 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2368 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2369
2370 /*
2371 * In fabrics we need to verify the cntlid matches the
2372 * admin connect
2373 */
2374 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2375 ret = -EINVAL;
2376 goto out_free;
2377 }
2378
2379 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2380 dev_err(ctrl->device,
2381 "keep-alive support is mandatory for fabrics\n");
2382 ret = -EINVAL;
2383 goto out_free;
2384 }
2385 } else {
2386 ctrl->cntlid = le16_to_cpu(id->cntlid);
2387 ctrl->hmpre = le32_to_cpu(id->hmpre);
2388 ctrl->hmmin = le32_to_cpu(id->hmmin);
2389 ctrl->hmminds = le32_to_cpu(id->hmminds);
2390 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2391 }
2392
2393 kfree(id);
2394
2395 if (ctrl->apst_enabled && !prev_apst_enabled)
2396 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2397 else if (!ctrl->apst_enabled && prev_apst_enabled)
2398 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2399
2400 ret = nvme_configure_apst(ctrl);
2401 if (ret < 0)
2402 return ret;
2403
2404 ret = nvme_configure_timestamp(ctrl);
2405 if (ret < 0)
2406 return ret;
2407
2408 ret = nvme_configure_directives(ctrl);
2409 if (ret < 0)
2410 return ret;
2411
2412 ctrl->identified = true;
2413
2414 return 0;
2415
2416 out_free:
2417 kfree(id);
2418 return ret;
2419 }
2420 EXPORT_SYMBOL_GPL(nvme_init_identify);
2421
2422 static int nvme_dev_open(struct inode *inode, struct file *file)
2423 {
2424 struct nvme_ctrl *ctrl =
2425 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2426
2427 switch (ctrl->state) {
2428 case NVME_CTRL_LIVE:
2429 case NVME_CTRL_ADMIN_ONLY:
2430 break;
2431 default:
2432 return -EWOULDBLOCK;
2433 }
2434
2435 file->private_data = ctrl;
2436 return 0;
2437 }
2438
2439 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2440 {
2441 struct nvme_ns *ns;
2442 int ret;
2443
2444 mutex_lock(&ctrl->namespaces_mutex);
2445 if (list_empty(&ctrl->namespaces)) {
2446 ret = -ENOTTY;
2447 goto out_unlock;
2448 }
2449
2450 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2451 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2452 dev_warn(ctrl->device,
2453 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2454 ret = -EINVAL;
2455 goto out_unlock;
2456 }
2457
2458 dev_warn(ctrl->device,
2459 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2460 kref_get(&ns->kref);
2461 mutex_unlock(&ctrl->namespaces_mutex);
2462
2463 ret = nvme_user_cmd(ctrl, ns, argp);
2464 nvme_put_ns(ns);
2465 return ret;
2466
2467 out_unlock:
2468 mutex_unlock(&ctrl->namespaces_mutex);
2469 return ret;
2470 }
2471
2472 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2473 unsigned long arg)
2474 {
2475 struct nvme_ctrl *ctrl = file->private_data;
2476 void __user *argp = (void __user *)arg;
2477
2478 switch (cmd) {
2479 case NVME_IOCTL_ADMIN_CMD:
2480 return nvme_user_cmd(ctrl, NULL, argp);
2481 case NVME_IOCTL_IO_CMD:
2482 return nvme_dev_user_cmd(ctrl, argp);
2483 case NVME_IOCTL_RESET:
2484 dev_warn(ctrl->device, "resetting controller\n");
2485 return nvme_reset_ctrl_sync(ctrl);
2486 case NVME_IOCTL_SUBSYS_RESET:
2487 return nvme_reset_subsystem(ctrl);
2488 case NVME_IOCTL_RESCAN:
2489 nvme_queue_scan(ctrl);
2490 return 0;
2491 default:
2492 return -ENOTTY;
2493 }
2494 }
2495
2496 static const struct file_operations nvme_dev_fops = {
2497 .owner = THIS_MODULE,
2498 .open = nvme_dev_open,
2499 .unlocked_ioctl = nvme_dev_ioctl,
2500 .compat_ioctl = nvme_dev_ioctl,
2501 };
2502
2503 static ssize_t nvme_sysfs_reset(struct device *dev,
2504 struct device_attribute *attr, const char *buf,
2505 size_t count)
2506 {
2507 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2508 int ret;
2509
2510 ret = nvme_reset_ctrl_sync(ctrl);
2511 if (ret < 0)
2512 return ret;
2513 return count;
2514 }
2515 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2516
2517 static ssize_t nvme_sysfs_rescan(struct device *dev,
2518 struct device_attribute *attr, const char *buf,
2519 size_t count)
2520 {
2521 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2522
2523 nvme_queue_scan(ctrl);
2524 return count;
2525 }
2526 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2527
2528 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2529 {
2530 struct gendisk *disk = dev_to_disk(dev);
2531
2532 if (disk->fops == &nvme_fops)
2533 return nvme_get_ns_from_dev(dev)->head;
2534 else
2535 return disk->private_data;
2536 }
2537
2538 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2539 char *buf)
2540 {
2541 struct nvme_ns_head *head = dev_to_ns_head(dev);
2542 struct nvme_ns_ids *ids = &head->ids;
2543 struct nvme_subsystem *subsys = head->subsys;
2544 int serial_len = sizeof(subsys->serial);
2545 int model_len = sizeof(subsys->model);
2546
2547 if (!uuid_is_null(&ids->uuid))
2548 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2549
2550 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2551 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2552
2553 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2554 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2555
2556 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2557 subsys->serial[serial_len - 1] == '\0'))
2558 serial_len--;
2559 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2560 subsys->model[model_len - 1] == '\0'))
2561 model_len--;
2562
2563 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2564 serial_len, subsys->serial, model_len, subsys->model,
2565 head->ns_id);
2566 }
2567 static DEVICE_ATTR_RO(wwid);
2568
2569 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2570 char *buf)
2571 {
2572 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2573 }
2574 static DEVICE_ATTR_RO(nguid);
2575
2576 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2577 char *buf)
2578 {
2579 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2580
2581 /* For backward compatibility expose the NGUID to userspace if
2582 * we have no UUID set
2583 */
2584 if (uuid_is_null(&ids->uuid)) {
2585 printk_ratelimited(KERN_WARNING
2586 "No UUID available providing old NGUID\n");
2587 return sprintf(buf, "%pU\n", ids->nguid);
2588 }
2589 return sprintf(buf, "%pU\n", &ids->uuid);
2590 }
2591 static DEVICE_ATTR_RO(uuid);
2592
2593 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2594 char *buf)
2595 {
2596 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2597 }
2598 static DEVICE_ATTR_RO(eui);
2599
2600 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2601 char *buf)
2602 {
2603 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2604 }
2605 static DEVICE_ATTR_RO(nsid);
2606
2607 static struct attribute *nvme_ns_id_attrs[] = {
2608 &dev_attr_wwid.attr,
2609 &dev_attr_uuid.attr,
2610 &dev_attr_nguid.attr,
2611 &dev_attr_eui.attr,
2612 &dev_attr_nsid.attr,
2613 NULL,
2614 };
2615
2616 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2617 struct attribute *a, int n)
2618 {
2619 struct device *dev = container_of(kobj, struct device, kobj);
2620 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2621
2622 if (a == &dev_attr_uuid.attr) {
2623 if (uuid_is_null(&ids->uuid) &&
2624 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2625 return 0;
2626 }
2627 if (a == &dev_attr_nguid.attr) {
2628 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2629 return 0;
2630 }
2631 if (a == &dev_attr_eui.attr) {
2632 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2633 return 0;
2634 }
2635 return a->mode;
2636 }
2637
2638 const struct attribute_group nvme_ns_id_attr_group = {
2639 .attrs = nvme_ns_id_attrs,
2640 .is_visible = nvme_ns_id_attrs_are_visible,
2641 };
2642
2643 #define nvme_show_str_function(field) \
2644 static ssize_t field##_show(struct device *dev, \
2645 struct device_attribute *attr, char *buf) \
2646 { \
2647 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2648 return sprintf(buf, "%.*s\n", \
2649 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
2650 } \
2651 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2652
2653 nvme_show_str_function(model);
2654 nvme_show_str_function(serial);
2655 nvme_show_str_function(firmware_rev);
2656
2657 #define nvme_show_int_function(field) \
2658 static ssize_t field##_show(struct device *dev, \
2659 struct device_attribute *attr, char *buf) \
2660 { \
2661 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2662 return sprintf(buf, "%d\n", ctrl->field); \
2663 } \
2664 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2665
2666 nvme_show_int_function(cntlid);
2667
2668 static ssize_t nvme_sysfs_delete(struct device *dev,
2669 struct device_attribute *attr, const char *buf,
2670 size_t count)
2671 {
2672 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2673
2674 if (device_remove_file_self(dev, attr))
2675 nvme_delete_ctrl_sync(ctrl);
2676 return count;
2677 }
2678 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2679
2680 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2681 struct device_attribute *attr,
2682 char *buf)
2683 {
2684 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2685
2686 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2687 }
2688 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2689
2690 static ssize_t nvme_sysfs_show_state(struct device *dev,
2691 struct device_attribute *attr,
2692 char *buf)
2693 {
2694 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2695 static const char *const state_name[] = {
2696 [NVME_CTRL_NEW] = "new",
2697 [NVME_CTRL_LIVE] = "live",
2698 [NVME_CTRL_ADMIN_ONLY] = "only-admin",
2699 [NVME_CTRL_RESETTING] = "resetting",
2700 [NVME_CTRL_CONNECTING] = "connecting",
2701 [NVME_CTRL_DELETING] = "deleting",
2702 [NVME_CTRL_DEAD] = "dead",
2703 };
2704
2705 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2706 state_name[ctrl->state])
2707 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2708
2709 return sprintf(buf, "unknown state\n");
2710 }
2711
2712 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2713
2714 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2715 struct device_attribute *attr,
2716 char *buf)
2717 {
2718 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2719
2720 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2721 }
2722 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2723
2724 static ssize_t nvme_sysfs_show_address(struct device *dev,
2725 struct device_attribute *attr,
2726 char *buf)
2727 {
2728 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2729
2730 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2731 }
2732 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2733
2734 static struct attribute *nvme_dev_attrs[] = {
2735 &dev_attr_reset_controller.attr,
2736 &dev_attr_rescan_controller.attr,
2737 &dev_attr_model.attr,
2738 &dev_attr_serial.attr,
2739 &dev_attr_firmware_rev.attr,
2740 &dev_attr_cntlid.attr,
2741 &dev_attr_delete_controller.attr,
2742 &dev_attr_transport.attr,
2743 &dev_attr_subsysnqn.attr,
2744 &dev_attr_address.attr,
2745 &dev_attr_state.attr,
2746 NULL
2747 };
2748
2749 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2750 struct attribute *a, int n)
2751 {
2752 struct device *dev = container_of(kobj, struct device, kobj);
2753 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2754
2755 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2756 return 0;
2757 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2758 return 0;
2759
2760 return a->mode;
2761 }
2762
2763 static struct attribute_group nvme_dev_attrs_group = {
2764 .attrs = nvme_dev_attrs,
2765 .is_visible = nvme_dev_attrs_are_visible,
2766 };
2767
2768 static const struct attribute_group *nvme_dev_attr_groups[] = {
2769 &nvme_dev_attrs_group,
2770 NULL,
2771 };
2772
2773 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2774 unsigned nsid)
2775 {
2776 struct nvme_ns_head *h;
2777
2778 lockdep_assert_held(&subsys->lock);
2779
2780 list_for_each_entry(h, &subsys->nsheads, entry) {
2781 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2782 return h;
2783 }
2784
2785 return NULL;
2786 }
2787
2788 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2789 struct nvme_ns_head *new)
2790 {
2791 struct nvme_ns_head *h;
2792
2793 lockdep_assert_held(&subsys->lock);
2794
2795 list_for_each_entry(h, &subsys->nsheads, entry) {
2796 if (nvme_ns_ids_valid(&new->ids) &&
2797 !list_empty(&h->list) &&
2798 nvme_ns_ids_equal(&new->ids, &h->ids))
2799 return -EINVAL;
2800 }
2801
2802 return 0;
2803 }
2804
2805 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2806 unsigned nsid, struct nvme_id_ns *id)
2807 {
2808 struct nvme_ns_head *head;
2809 int ret = -ENOMEM;
2810
2811 head = kzalloc(sizeof(*head), GFP_KERNEL);
2812 if (!head)
2813 goto out;
2814 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
2815 if (ret < 0)
2816 goto out_free_head;
2817 head->instance = ret;
2818 INIT_LIST_HEAD(&head->list);
2819 init_srcu_struct(&head->srcu);
2820 head->subsys = ctrl->subsys;
2821 head->ns_id = nsid;
2822 kref_init(&head->ref);
2823
2824 nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
2825
2826 ret = __nvme_check_ids(ctrl->subsys, head);
2827 if (ret) {
2828 dev_err(ctrl->device,
2829 "duplicate IDs for nsid %d\n", nsid);
2830 goto out_cleanup_srcu;
2831 }
2832
2833 ret = nvme_mpath_alloc_disk(ctrl, head);
2834 if (ret)
2835 goto out_cleanup_srcu;
2836
2837 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
2838 return head;
2839 out_cleanup_srcu:
2840 cleanup_srcu_struct(&head->srcu);
2841 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
2842 out_free_head:
2843 kfree(head);
2844 out:
2845 return ERR_PTR(ret);
2846 }
2847
2848 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
2849 struct nvme_id_ns *id)
2850 {
2851 struct nvme_ctrl *ctrl = ns->ctrl;
2852 bool is_shared = id->nmic & (1 << 0);
2853 struct nvme_ns_head *head = NULL;
2854 int ret = 0;
2855
2856 mutex_lock(&ctrl->subsys->lock);
2857 if (is_shared)
2858 head = __nvme_find_ns_head(ctrl->subsys, nsid);
2859 if (!head) {
2860 head = nvme_alloc_ns_head(ctrl, nsid, id);
2861 if (IS_ERR(head)) {
2862 ret = PTR_ERR(head);
2863 goto out_unlock;
2864 }
2865 } else {
2866 struct nvme_ns_ids ids;
2867
2868 nvme_report_ns_ids(ctrl, nsid, id, &ids);
2869 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
2870 dev_err(ctrl->device,
2871 "IDs don't match for shared namespace %d\n",
2872 nsid);
2873 ret = -EINVAL;
2874 goto out_unlock;
2875 }
2876 }
2877
2878 list_add_tail(&ns->siblings, &head->list);
2879 ns->head = head;
2880
2881 out_unlock:
2882 mutex_unlock(&ctrl->subsys->lock);
2883 return ret;
2884 }
2885
2886 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2887 {
2888 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2889 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2890
2891 return nsa->head->ns_id - nsb->head->ns_id;
2892 }
2893
2894 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2895 {
2896 struct nvme_ns *ns, *ret = NULL;
2897
2898 mutex_lock(&ctrl->namespaces_mutex);
2899 list_for_each_entry(ns, &ctrl->namespaces, list) {
2900 if (ns->head->ns_id == nsid) {
2901 if (!kref_get_unless_zero(&ns->kref))
2902 continue;
2903 ret = ns;
2904 break;
2905 }
2906 if (ns->head->ns_id > nsid)
2907 break;
2908 }
2909 mutex_unlock(&ctrl->namespaces_mutex);
2910 return ret;
2911 }
2912
2913 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
2914 {
2915 struct streams_directive_params s;
2916 int ret;
2917
2918 if (!ctrl->nr_streams)
2919 return 0;
2920
2921 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
2922 if (ret)
2923 return ret;
2924
2925 ns->sws = le32_to_cpu(s.sws);
2926 ns->sgs = le16_to_cpu(s.sgs);
2927
2928 if (ns->sws) {
2929 unsigned int bs = 1 << ns->lba_shift;
2930
2931 blk_queue_io_min(ns->queue, bs * ns->sws);
2932 if (ns->sgs)
2933 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
2934 }
2935
2936 return 0;
2937 }
2938
2939 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2940 {
2941 struct nvme_ns *ns;
2942 struct gendisk *disk;
2943 struct nvme_id_ns *id;
2944 char disk_name[DISK_NAME_LEN];
2945 int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
2946
2947 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2948 if (!ns)
2949 return;
2950
2951 ns->queue = blk_mq_init_queue(ctrl->tagset);
2952 if (IS_ERR(ns->queue))
2953 goto out_free_ns;
2954 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2955 ns->queue->queuedata = ns;
2956 ns->ctrl = ctrl;
2957
2958 kref_init(&ns->kref);
2959 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2960
2961 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2962 nvme_set_queue_limits(ctrl, ns->queue);
2963
2964 id = nvme_identify_ns(ctrl, nsid);
2965 if (!id)
2966 goto out_free_queue;
2967
2968 if (id->ncap == 0)
2969 goto out_free_id;
2970
2971 if (nvme_init_ns_head(ns, nsid, id))
2972 goto out_free_id;
2973 nvme_setup_streams_ns(ctrl, ns);
2974
2975 #ifdef CONFIG_NVME_MULTIPATH
2976 /*
2977 * If multipathing is enabled we need to always use the subsystem
2978 * instance number for numbering our devices to avoid conflicts
2979 * between subsystems that have multiple controllers and thus use
2980 * the multipath-aware subsystem node and those that have a single
2981 * controller and use the controller node directly.
2982 */
2983 if (ns->head->disk) {
2984 sprintf(disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
2985 ctrl->cntlid, ns->head->instance);
2986 flags = GENHD_FL_HIDDEN;
2987 } else {
2988 sprintf(disk_name, "nvme%dn%d", ctrl->subsys->instance,
2989 ns->head->instance);
2990 }
2991 #else
2992 /*
2993 * But without the multipath code enabled, multiple controller per
2994 * subsystems are visible as devices and thus we cannot use the
2995 * subsystem instance.
2996 */
2997 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
2998 #endif
2999
3000 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3001 if (nvme_nvm_register(ns, disk_name, node)) {
3002 dev_warn(ctrl->device, "LightNVM init failure\n");
3003 goto out_unlink_ns;
3004 }
3005 }
3006
3007 disk = alloc_disk_node(0, node);
3008 if (!disk)
3009 goto out_unlink_ns;
3010
3011 disk->fops = &nvme_fops;
3012 disk->private_data = ns;
3013 disk->queue = ns->queue;
3014 disk->flags = flags;
3015 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3016 ns->disk = disk;
3017
3018 __nvme_revalidate_disk(disk, id);
3019
3020 mutex_lock(&ctrl->namespaces_mutex);
3021 list_add_tail(&ns->list, &ctrl->namespaces);
3022 mutex_unlock(&ctrl->namespaces_mutex);
3023
3024 nvme_get_ctrl(ctrl);
3025
3026 kfree(id);
3027
3028 device_add_disk(ctrl->device, ns->disk);
3029 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
3030 &nvme_ns_id_attr_group))
3031 pr_warn("%s: failed to create sysfs group for identification\n",
3032 ns->disk->disk_name);
3033 if (ns->ndev && nvme_nvm_register_sysfs(ns))
3034 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
3035 ns->disk->disk_name);
3036
3037 nvme_mpath_add_disk(ns->head);
3038 return;
3039 out_unlink_ns:
3040 mutex_lock(&ctrl->subsys->lock);
3041 list_del_rcu(&ns->siblings);
3042 mutex_unlock(&ctrl->subsys->lock);
3043 out_free_id:
3044 kfree(id);
3045 out_free_queue:
3046 blk_cleanup_queue(ns->queue);
3047 out_free_ns:
3048 kfree(ns);
3049 }
3050
3051 static void nvme_ns_remove(struct nvme_ns *ns)
3052 {
3053 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3054 return;
3055
3056 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3057 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
3058 &nvme_ns_id_attr_group);
3059 if (ns->ndev)
3060 nvme_nvm_unregister_sysfs(ns);
3061 del_gendisk(ns->disk);
3062 blk_cleanup_queue(ns->queue);
3063 if (blk_get_integrity(ns->disk))
3064 blk_integrity_unregister(ns->disk);
3065 }
3066
3067 mutex_lock(&ns->ctrl->subsys->lock);
3068 nvme_mpath_clear_current_path(ns);
3069 list_del_rcu(&ns->siblings);
3070 mutex_unlock(&ns->ctrl->subsys->lock);
3071
3072 mutex_lock(&ns->ctrl->namespaces_mutex);
3073 list_del_init(&ns->list);
3074 mutex_unlock(&ns->ctrl->namespaces_mutex);
3075
3076 synchronize_srcu(&ns->head->srcu);
3077 nvme_mpath_check_last_path(ns);
3078 nvme_put_ns(ns);
3079 }
3080
3081 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3082 {
3083 struct nvme_ns *ns;
3084
3085 ns = nvme_find_get_ns(ctrl, nsid);
3086 if (ns) {
3087 if (ns->disk && revalidate_disk(ns->disk))
3088 nvme_ns_remove(ns);
3089 nvme_put_ns(ns);
3090 } else
3091 nvme_alloc_ns(ctrl, nsid);
3092 }
3093
3094 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3095 unsigned nsid)
3096 {
3097 struct nvme_ns *ns, *next;
3098
3099 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3100 if (ns->head->ns_id > nsid)
3101 nvme_ns_remove(ns);
3102 }
3103 }
3104
3105 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3106 {
3107 struct nvme_ns *ns;
3108 __le32 *ns_list;
3109 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3110 int ret = 0;
3111
3112 ns_list = kzalloc(0x1000, GFP_KERNEL);
3113 if (!ns_list)
3114 return -ENOMEM;
3115
3116 for (i = 0; i < num_lists; i++) {
3117 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3118 if (ret)
3119 goto free;
3120
3121 for (j = 0; j < min(nn, 1024U); j++) {
3122 nsid = le32_to_cpu(ns_list[j]);
3123 if (!nsid)
3124 goto out;
3125
3126 nvme_validate_ns(ctrl, nsid);
3127
3128 while (++prev < nsid) {
3129 ns = nvme_find_get_ns(ctrl, prev);
3130 if (ns) {
3131 nvme_ns_remove(ns);
3132 nvme_put_ns(ns);
3133 }
3134 }
3135 }
3136 nn -= j;
3137 }
3138 out:
3139 nvme_remove_invalid_namespaces(ctrl, prev);
3140 free:
3141 kfree(ns_list);
3142 return ret;
3143 }
3144
3145 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3146 {
3147 unsigned i;
3148
3149 for (i = 1; i <= nn; i++)
3150 nvme_validate_ns(ctrl, i);
3151
3152 nvme_remove_invalid_namespaces(ctrl, nn);
3153 }
3154
3155 static void nvme_scan_work(struct work_struct *work)
3156 {
3157 struct nvme_ctrl *ctrl =
3158 container_of(work, struct nvme_ctrl, scan_work);
3159 struct nvme_id_ctrl *id;
3160 unsigned nn;
3161
3162 if (ctrl->state != NVME_CTRL_LIVE)
3163 return;
3164
3165 WARN_ON_ONCE(!ctrl->tagset);
3166
3167 if (nvme_identify_ctrl(ctrl, &id))
3168 return;
3169
3170 nn = le32_to_cpu(id->nn);
3171 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3172 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3173 if (!nvme_scan_ns_list(ctrl, nn))
3174 goto done;
3175 }
3176 nvme_scan_ns_sequential(ctrl, nn);
3177 done:
3178 mutex_lock(&ctrl->namespaces_mutex);
3179 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3180 mutex_unlock(&ctrl->namespaces_mutex);
3181 kfree(id);
3182 }
3183
3184 void nvme_queue_scan(struct nvme_ctrl *ctrl)
3185 {
3186 /*
3187 * Only new queue scan work when admin and IO queues are both alive
3188 */
3189 if (ctrl->state == NVME_CTRL_LIVE)
3190 queue_work(nvme_wq, &ctrl->scan_work);
3191 }
3192 EXPORT_SYMBOL_GPL(nvme_queue_scan);
3193
3194 /*
3195 * This function iterates the namespace list unlocked to allow recovery from
3196 * controller failure. It is up to the caller to ensure the namespace list is
3197 * not modified by scan work while this function is executing.
3198 */
3199 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3200 {
3201 struct nvme_ns *ns, *next;
3202
3203 /*
3204 * The dead states indicates the controller was not gracefully
3205 * disconnected. In that case, we won't be able to flush any data while
3206 * removing the namespaces' disks; fail all the queues now to avoid
3207 * potentially having to clean up the failed sync later.
3208 */
3209 if (ctrl->state == NVME_CTRL_DEAD)
3210 nvme_kill_queues(ctrl);
3211
3212 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
3213 nvme_ns_remove(ns);
3214 }
3215 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3216
3217 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3218 {
3219 char *envp[2] = { NULL, NULL };
3220 u32 aen_result = ctrl->aen_result;
3221
3222 ctrl->aen_result = 0;
3223 if (!aen_result)
3224 return;
3225
3226 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3227 if (!envp[0])
3228 return;
3229 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3230 kfree(envp[0]);
3231 }
3232
3233 static void nvme_async_event_work(struct work_struct *work)
3234 {
3235 struct nvme_ctrl *ctrl =
3236 container_of(work, struct nvme_ctrl, async_event_work);
3237
3238 nvme_aen_uevent(ctrl);
3239 ctrl->ops->submit_async_event(ctrl);
3240 }
3241
3242 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3243 {
3244
3245 u32 csts;
3246
3247 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3248 return false;
3249
3250 if (csts == ~0)
3251 return false;
3252
3253 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3254 }
3255
3256 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3257 {
3258 struct nvme_fw_slot_info_log *log;
3259
3260 log = kmalloc(sizeof(*log), GFP_KERNEL);
3261 if (!log)
3262 return;
3263
3264 if (nvme_get_log(ctrl, NVME_LOG_FW_SLOT, log, sizeof(*log)))
3265 dev_warn(ctrl->device,
3266 "Get FW SLOT INFO log error\n");
3267 kfree(log);
3268 }
3269
3270 static void nvme_fw_act_work(struct work_struct *work)
3271 {
3272 struct nvme_ctrl *ctrl = container_of(work,
3273 struct nvme_ctrl, fw_act_work);
3274 unsigned long fw_act_timeout;
3275
3276 if (ctrl->mtfa)
3277 fw_act_timeout = jiffies +
3278 msecs_to_jiffies(ctrl->mtfa * 100);
3279 else
3280 fw_act_timeout = jiffies +
3281 msecs_to_jiffies(admin_timeout * 1000);
3282
3283 nvme_stop_queues(ctrl);
3284 while (nvme_ctrl_pp_status(ctrl)) {
3285 if (time_after(jiffies, fw_act_timeout)) {
3286 dev_warn(ctrl->device,
3287 "Fw activation timeout, reset controller\n");
3288 nvme_reset_ctrl(ctrl);
3289 break;
3290 }
3291 msleep(100);
3292 }
3293
3294 if (ctrl->state != NVME_CTRL_LIVE)
3295 return;
3296
3297 nvme_start_queues(ctrl);
3298 /* read FW slot information to clear the AER */
3299 nvme_get_fw_slot_info(ctrl);
3300 }
3301
3302 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3303 union nvme_result *res)
3304 {
3305 u32 result = le32_to_cpu(res->u32);
3306
3307 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3308 return;
3309
3310 switch (result & 0x7) {
3311 case NVME_AER_ERROR:
3312 case NVME_AER_SMART:
3313 case NVME_AER_CSS:
3314 case NVME_AER_VS:
3315 ctrl->aen_result = result;
3316 break;
3317 default:
3318 break;
3319 }
3320
3321 switch (result & 0xff07) {
3322 case NVME_AER_NOTICE_NS_CHANGED:
3323 dev_info(ctrl->device, "rescanning\n");
3324 nvme_queue_scan(ctrl);
3325 break;
3326 case NVME_AER_NOTICE_FW_ACT_STARTING:
3327 queue_work(nvme_wq, &ctrl->fw_act_work);
3328 break;
3329 default:
3330 dev_warn(ctrl->device, "async event result %08x\n", result);
3331 }
3332 queue_work(nvme_wq, &ctrl->async_event_work);
3333 }
3334 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3335
3336 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3337 {
3338 nvme_stop_keep_alive(ctrl);
3339 flush_work(&ctrl->async_event_work);
3340 flush_work(&ctrl->scan_work);
3341 cancel_work_sync(&ctrl->fw_act_work);
3342 }
3343 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3344
3345 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3346 {
3347 if (ctrl->kato)
3348 nvme_start_keep_alive(ctrl);
3349
3350 if (ctrl->queue_count > 1) {
3351 nvme_queue_scan(ctrl);
3352 queue_work(nvme_wq, &ctrl->async_event_work);
3353 nvme_start_queues(ctrl);
3354 }
3355 }
3356 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3357
3358 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3359 {
3360 cdev_device_del(&ctrl->cdev, ctrl->device);
3361 }
3362 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3363
3364 static void nvme_free_ctrl(struct device *dev)
3365 {
3366 struct nvme_ctrl *ctrl =
3367 container_of(dev, struct nvme_ctrl, ctrl_device);
3368 struct nvme_subsystem *subsys = ctrl->subsys;
3369
3370 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3371 kfree(ctrl->effects);
3372
3373 if (subsys) {
3374 mutex_lock(&subsys->lock);
3375 list_del(&ctrl->subsys_entry);
3376 mutex_unlock(&subsys->lock);
3377 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3378 }
3379
3380 ctrl->ops->free_ctrl(ctrl);
3381
3382 if (subsys)
3383 nvme_put_subsystem(subsys);
3384 }
3385
3386 /*
3387 * Initialize a NVMe controller structures. This needs to be called during
3388 * earliest initialization so that we have the initialized structured around
3389 * during probing.
3390 */
3391 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3392 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3393 {
3394 int ret;
3395
3396 ctrl->state = NVME_CTRL_NEW;
3397 spin_lock_init(&ctrl->lock);
3398 INIT_LIST_HEAD(&ctrl->namespaces);
3399 mutex_init(&ctrl->namespaces_mutex);
3400 ctrl->dev = dev;
3401 ctrl->ops = ops;
3402 ctrl->quirks = quirks;
3403 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3404 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3405 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3406 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3407
3408 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3409 if (ret < 0)
3410 goto out;
3411 ctrl->instance = ret;
3412
3413 device_initialize(&ctrl->ctrl_device);
3414 ctrl->device = &ctrl->ctrl_device;
3415 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3416 ctrl->device->class = nvme_class;
3417 ctrl->device->parent = ctrl->dev;
3418 ctrl->device->groups = nvme_dev_attr_groups;
3419 ctrl->device->release = nvme_free_ctrl;
3420 dev_set_drvdata(ctrl->device, ctrl);
3421 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3422 if (ret)
3423 goto out_release_instance;
3424
3425 cdev_init(&ctrl->cdev, &nvme_dev_fops);
3426 ctrl->cdev.owner = ops->module;
3427 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3428 if (ret)
3429 goto out_free_name;
3430
3431 /*
3432 * Initialize latency tolerance controls. The sysfs files won't
3433 * be visible to userspace unless the device actually supports APST.
3434 */
3435 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3436 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3437 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3438
3439 return 0;
3440 out_free_name:
3441 kfree_const(dev->kobj.name);
3442 out_release_instance:
3443 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3444 out:
3445 return ret;
3446 }
3447 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3448
3449 /**
3450 * nvme_kill_queues(): Ends all namespace queues
3451 * @ctrl: the dead controller that needs to end
3452 *
3453 * Call this function when the driver determines it is unable to get the
3454 * controller in a state capable of servicing IO.
3455 */
3456 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3457 {
3458 struct nvme_ns *ns;
3459
3460 mutex_lock(&ctrl->namespaces_mutex);
3461
3462 /* Forcibly unquiesce queues to avoid blocking dispatch */
3463 if (ctrl->admin_q)
3464 blk_mq_unquiesce_queue(ctrl->admin_q);
3465
3466 list_for_each_entry(ns, &ctrl->namespaces, list) {
3467 /*
3468 * Revalidating a dead namespace sets capacity to 0. This will
3469 * end buffered writers dirtying pages that can't be synced.
3470 */
3471 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
3472 continue;
3473 revalidate_disk(ns->disk);
3474 blk_set_queue_dying(ns->queue);
3475
3476 /* Forcibly unquiesce queues to avoid blocking dispatch */
3477 blk_mq_unquiesce_queue(ns->queue);
3478 }
3479 mutex_unlock(&ctrl->namespaces_mutex);
3480 }
3481 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3482
3483 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3484 {
3485 struct nvme_ns *ns;
3486
3487 mutex_lock(&ctrl->namespaces_mutex);
3488 list_for_each_entry(ns, &ctrl->namespaces, list)
3489 blk_mq_unfreeze_queue(ns->queue);
3490 mutex_unlock(&ctrl->namespaces_mutex);
3491 }
3492 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3493
3494 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3495 {
3496 struct nvme_ns *ns;
3497
3498 mutex_lock(&ctrl->namespaces_mutex);
3499 list_for_each_entry(ns, &ctrl->namespaces, list) {
3500 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3501 if (timeout <= 0)
3502 break;
3503 }
3504 mutex_unlock(&ctrl->namespaces_mutex);
3505 }
3506 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3507
3508 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3509 {
3510 struct nvme_ns *ns;
3511
3512 mutex_lock(&ctrl->namespaces_mutex);
3513 list_for_each_entry(ns, &ctrl->namespaces, list)
3514 blk_mq_freeze_queue_wait(ns->queue);
3515 mutex_unlock(&ctrl->namespaces_mutex);
3516 }
3517 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3518
3519 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3520 {
3521 struct nvme_ns *ns;
3522
3523 mutex_lock(&ctrl->namespaces_mutex);
3524 list_for_each_entry(ns, &ctrl->namespaces, list)
3525 blk_freeze_queue_start(ns->queue);
3526 mutex_unlock(&ctrl->namespaces_mutex);
3527 }
3528 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3529
3530 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3531 {
3532 struct nvme_ns *ns;
3533
3534 mutex_lock(&ctrl->namespaces_mutex);
3535 list_for_each_entry(ns, &ctrl->namespaces, list)
3536 blk_mq_quiesce_queue(ns->queue);
3537 mutex_unlock(&ctrl->namespaces_mutex);
3538 }
3539 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3540
3541 void nvme_start_queues(struct nvme_ctrl *ctrl)
3542 {
3543 struct nvme_ns *ns;
3544
3545 mutex_lock(&ctrl->namespaces_mutex);
3546 list_for_each_entry(ns, &ctrl->namespaces, list)
3547 blk_mq_unquiesce_queue(ns->queue);
3548 mutex_unlock(&ctrl->namespaces_mutex);
3549 }
3550 EXPORT_SYMBOL_GPL(nvme_start_queues);
3551
3552 int nvme_reinit_tagset(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set)
3553 {
3554 if (!ctrl->ops->reinit_request)
3555 return 0;
3556
3557 return blk_mq_tagset_iter(set, set->driver_data,
3558 ctrl->ops->reinit_request);
3559 }
3560 EXPORT_SYMBOL_GPL(nvme_reinit_tagset);
3561
3562 int __init nvme_core_init(void)
3563 {
3564 int result = -ENOMEM;
3565
3566 nvme_wq = alloc_workqueue("nvme-wq",
3567 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3568 if (!nvme_wq)
3569 goto out;
3570
3571 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3572 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3573 if (!nvme_reset_wq)
3574 goto destroy_wq;
3575
3576 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3577 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3578 if (!nvme_delete_wq)
3579 goto destroy_reset_wq;
3580
3581 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3582 if (result < 0)
3583 goto destroy_delete_wq;
3584
3585 nvme_class = class_create(THIS_MODULE, "nvme");
3586 if (IS_ERR(nvme_class)) {
3587 result = PTR_ERR(nvme_class);
3588 goto unregister_chrdev;
3589 }
3590
3591 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3592 if (IS_ERR(nvme_subsys_class)) {
3593 result = PTR_ERR(nvme_subsys_class);
3594 goto destroy_class;
3595 }
3596 return 0;
3597
3598 destroy_class:
3599 class_destroy(nvme_class);
3600 unregister_chrdev:
3601 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3602 destroy_delete_wq:
3603 destroy_workqueue(nvme_delete_wq);
3604 destroy_reset_wq:
3605 destroy_workqueue(nvme_reset_wq);
3606 destroy_wq:
3607 destroy_workqueue(nvme_wq);
3608 out:
3609 return result;
3610 }
3611
3612 void nvme_core_exit(void)
3613 {
3614 ida_destroy(&nvme_subsystems_ida);
3615 class_destroy(nvme_subsys_class);
3616 class_destroy(nvme_class);
3617 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3618 destroy_workqueue(nvme_delete_wq);
3619 destroy_workqueue(nvme_reset_wq);
3620 destroy_workqueue(nvme_wq);
3621 }
3622
3623 MODULE_LICENSE("GPL");
3624 MODULE_VERSION("1.0");
3625 module_init(nvme_core_init);
3626 module_exit(nvme_core_exit);
Linux with added FPC-III support