1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * NVM Express device driver
4 * Copyright (c) 2011-2014, Intel Corporation.
5 */
6
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/delay.h>
10 #include <linux/errno.h>
11 #include <linux/hdreg.h>
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/list_sort.h>
16 #include <linux/slab.h>
17 #include <linux/types.h>
18 #include <linux/pr.h>
19 #include <linux/ptrace.h>
20 #include <linux/nvme_ioctl.h>
21 #include <linux/t10-pi.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
24
25 #include "nvme.h"
26 #include "fabrics.h"
27
28 #define CREATE_TRACE_POINTS
29 #include "trace.h"
30
31 #define NVME_MINORS (1U << MINORBITS)
32
33 unsigned int admin_timeout = 60;
34 module_param(admin_timeout, uint, 0644);
35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
36 EXPORT_SYMBOL_GPL(admin_timeout);
37
38 unsigned int nvme_io_timeout = 30;
39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
42
43 static unsigned char shutdown_timeout = 5;
44 module_param(shutdown_timeout, byte, 0644);
45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
46
47 static u8 nvme_max_retries = 5;
48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
50
51 static unsigned long default_ps_max_latency_us = 100000;
52 module_param(default_ps_max_latency_us, ulong, 0644);
53 MODULE_PARM_DESC(default_ps_max_latency_us,
54 "max power saving latency for new devices; use PM QOS to change per device");
55
56 static bool force_apst;
57 module_param(force_apst, bool, 0644);
58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
59
60 static bool streams;
61 module_param(streams, bool, 0644);
62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
63
64 /*
65 * nvme_wq - hosts nvme related works that are not reset or delete
66 * nvme_reset_wq - hosts nvme reset works
67 * nvme_delete_wq - hosts nvme delete works
68 *
69 * nvme_wq will host works such are scan, aen handling, fw activation,
70 * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
71 * runs reset works which also flush works hosted on nvme_wq for
72 * serialization purposes. nvme_delete_wq host controller deletion
73 * works which flush reset works for serialization.
74 */
75 struct workqueue_struct *nvme_wq;
76 EXPORT_SYMBOL_GPL(nvme_wq);
77
78 struct workqueue_struct *nvme_reset_wq;
79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
80
81 struct workqueue_struct *nvme_delete_wq;
82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
83
84 static LIST_HEAD(nvme_subsystems);
85 static DEFINE_MUTEX(nvme_subsystems_lock);
86
87 static DEFINE_IDA(nvme_instance_ida);
88 static dev_t nvme_chr_devt;
89 static struct class *nvme_class;
90 static struct class *nvme_subsys_class;
91
92 static int nvme_revalidate_disk(struct gendisk *disk);
93 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
94 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
95 unsigned nsid);
96
nvme_set_queue_dying(struct nvme_ns * ns)97 static void nvme_set_queue_dying(struct nvme_ns *ns)
98 {
99 /*
100 * Revalidating a dead namespace sets capacity to 0. This will end
101 * buffered writers dirtying pages that can't be synced.
102 */
103 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
104 return;
105 blk_set_queue_dying(ns->queue);
106 /* Forcibly unquiesce queues to avoid blocking dispatch */
107 blk_mq_unquiesce_queue(ns->queue);
108 /*
109 * Revalidate after unblocking dispatchers that may be holding bd_butex
110 */
111 revalidate_disk(ns->disk);
112 }
113
nvme_queue_scan(struct nvme_ctrl * ctrl)114 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
115 {
116 /*
117 * Only new queue scan work when admin and IO queues are both alive
118 */
119 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
120 queue_work(nvme_wq, &ctrl->scan_work);
121 }
122
123 /*
124 * Use this function to proceed with scheduling reset_work for a controller
125 * that had previously been set to the resetting state. This is intended for
126 * code paths that can't be interrupted by other reset attempts. A hot removal
127 * may prevent this from succeeding.
128 */
nvme_try_sched_reset(struct nvme_ctrl * ctrl)129 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
130 {
131 if (ctrl->state != NVME_CTRL_RESETTING)
132 return -EBUSY;
133 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
134 return -EBUSY;
135 return 0;
136 }
137 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
138
nvme_reset_ctrl(struct nvme_ctrl * ctrl)139 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
140 {
141 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
142 return -EBUSY;
143 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
144 return -EBUSY;
145 return 0;
146 }
147 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
148
nvme_reset_ctrl_sync(struct nvme_ctrl * ctrl)149 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
150 {
151 int ret;
152
153 ret = nvme_reset_ctrl(ctrl);
154 if (!ret) {
155 flush_work(&ctrl->reset_work);
156 if (ctrl->state != NVME_CTRL_LIVE)
157 ret = -ENETRESET;
158 }
159
160 return ret;
161 }
162 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
163
nvme_do_delete_ctrl(struct nvme_ctrl * ctrl)164 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
165 {
166 dev_info(ctrl->device,
167 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
168
169 flush_work(&ctrl->reset_work);
170 nvme_stop_ctrl(ctrl);
171 nvme_remove_namespaces(ctrl);
172 ctrl->ops->delete_ctrl(ctrl);
173 nvme_uninit_ctrl(ctrl);
174 nvme_put_ctrl(ctrl);
175 }
176
nvme_delete_ctrl_work(struct work_struct * work)177 static void nvme_delete_ctrl_work(struct work_struct *work)
178 {
179 struct nvme_ctrl *ctrl =
180 container_of(work, struct nvme_ctrl, delete_work);
181
182 nvme_do_delete_ctrl(ctrl);
183 }
184
nvme_delete_ctrl(struct nvme_ctrl * ctrl)185 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
186 {
187 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
188 return -EBUSY;
189 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
190 return -EBUSY;
191 return 0;
192 }
193 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
194
nvme_delete_ctrl_sync(struct nvme_ctrl * ctrl)195 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
196 {
197 int ret = 0;
198
199 /*
200 * Keep a reference until nvme_do_delete_ctrl() complete,
201 * since ->delete_ctrl can free the controller.
202 */
203 nvme_get_ctrl(ctrl);
204 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
205 ret = -EBUSY;
206 if (!ret)
207 nvme_do_delete_ctrl(ctrl);
208 nvme_put_ctrl(ctrl);
209 return ret;
210 }
211
nvme_ns_has_pi(struct nvme_ns * ns)212 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
213 {
214 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
215 }
216
nvme_error_status(u16 status)217 static blk_status_t nvme_error_status(u16 status)
218 {
219 switch (status & 0x7ff) {
220 case NVME_SC_SUCCESS:
221 return BLK_STS_OK;
222 case NVME_SC_CAP_EXCEEDED:
223 return BLK_STS_NOSPC;
224 case NVME_SC_LBA_RANGE:
225 return BLK_STS_TARGET;
226 case NVME_SC_BAD_ATTRIBUTES:
227 case NVME_SC_ONCS_NOT_SUPPORTED:
228 case NVME_SC_INVALID_OPCODE:
229 case NVME_SC_INVALID_FIELD:
230 case NVME_SC_INVALID_NS:
231 return BLK_STS_NOTSUPP;
232 case NVME_SC_WRITE_FAULT:
233 case NVME_SC_READ_ERROR:
234 case NVME_SC_UNWRITTEN_BLOCK:
235 case NVME_SC_ACCESS_DENIED:
236 case NVME_SC_READ_ONLY:
237 case NVME_SC_COMPARE_FAILED:
238 return BLK_STS_MEDIUM;
239 case NVME_SC_GUARD_CHECK:
240 case NVME_SC_APPTAG_CHECK:
241 case NVME_SC_REFTAG_CHECK:
242 case NVME_SC_INVALID_PI:
243 return BLK_STS_PROTECTION;
244 case NVME_SC_RESERVATION_CONFLICT:
245 return BLK_STS_NEXUS;
246 case NVME_SC_HOST_PATH_ERROR:
247 return BLK_STS_TRANSPORT;
248 default:
249 return BLK_STS_IOERR;
250 }
251 }
252
nvme_req_needs_retry(struct request * req)253 static inline bool nvme_req_needs_retry(struct request *req)
254 {
255 if (blk_noretry_request(req))
256 return false;
257 if (nvme_req(req)->status & NVME_SC_DNR)
258 return false;
259 if (nvme_req(req)->retries >= nvme_max_retries)
260 return false;
261 return true;
262 }
263
nvme_retry_req(struct request * req)264 static void nvme_retry_req(struct request *req)
265 {
266 struct nvme_ns *ns = req->q->queuedata;
267 unsigned long delay = 0;
268 u16 crd;
269
270 /* The mask and shift result must be <= 3 */
271 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
272 if (ns && crd)
273 delay = ns->ctrl->crdt[crd - 1] * 100;
274
275 nvme_req(req)->retries++;
276 blk_mq_requeue_request(req, false);
277 blk_mq_delay_kick_requeue_list(req->q, delay);
278 }
279
nvme_complete_rq(struct request * req)280 void nvme_complete_rq(struct request *req)
281 {
282 blk_status_t status = nvme_error_status(nvme_req(req)->status);
283
284 trace_nvme_complete_rq(req);
285
286 if (nvme_req(req)->ctrl->kas)
287 nvme_req(req)->ctrl->comp_seen = true;
288
289 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
290 if ((req->cmd_flags & REQ_NVME_MPATH) &&
291 blk_path_error(status)) {
292 nvme_failover_req(req);
293 return;
294 }
295
296 if (!blk_queue_dying(req->q)) {
297 nvme_retry_req(req);
298 return;
299 }
300 }
301
302 nvme_trace_bio_complete(req, status);
303 blk_mq_end_request(req, status);
304 }
305 EXPORT_SYMBOL_GPL(nvme_complete_rq);
306
nvme_cancel_request(struct request * req,void * data,bool reserved)307 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
308 {
309 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
310 "Cancelling I/O %d", req->tag);
311
312 /* don't abort one completed request */
313 if (blk_mq_request_completed(req))
314 return true;
315
316 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
317 blk_mq_complete_request(req);
318 return true;
319 }
320 EXPORT_SYMBOL_GPL(nvme_cancel_request);
321
nvme_change_ctrl_state(struct nvme_ctrl * ctrl,enum nvme_ctrl_state new_state)322 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
323 enum nvme_ctrl_state new_state)
324 {
325 enum nvme_ctrl_state old_state;
326 unsigned long flags;
327 bool changed = false;
328
329 spin_lock_irqsave(&ctrl->lock, flags);
330
331 old_state = ctrl->state;
332 switch (new_state) {
333 case NVME_CTRL_LIVE:
334 switch (old_state) {
335 case NVME_CTRL_NEW:
336 case NVME_CTRL_RESETTING:
337 case NVME_CTRL_CONNECTING:
338 changed = true;
339 /* FALLTHRU */
340 default:
341 break;
342 }
343 break;
344 case NVME_CTRL_RESETTING:
345 switch (old_state) {
346 case NVME_CTRL_NEW:
347 case NVME_CTRL_LIVE:
348 changed = true;
349 /* FALLTHRU */
350 default:
351 break;
352 }
353 break;
354 case NVME_CTRL_CONNECTING:
355 switch (old_state) {
356 case NVME_CTRL_NEW:
357 case NVME_CTRL_RESETTING:
358 changed = true;
359 /* FALLTHRU */
360 default:
361 break;
362 }
363 break;
364 case NVME_CTRL_DELETING:
365 switch (old_state) {
366 case NVME_CTRL_LIVE:
367 case NVME_CTRL_RESETTING:
368 case NVME_CTRL_CONNECTING:
369 changed = true;
370 /* FALLTHRU */
371 default:
372 break;
373 }
374 break;
375 case NVME_CTRL_DEAD:
376 switch (old_state) {
377 case NVME_CTRL_DELETING:
378 changed = true;
379 /* FALLTHRU */
380 default:
381 break;
382 }
383 break;
384 default:
385 break;
386 }
387
388 if (changed) {
389 ctrl->state = new_state;
390 wake_up_all(&ctrl->state_wq);
391 }
392
393 spin_unlock_irqrestore(&ctrl->lock, flags);
394 if (changed && ctrl->state == NVME_CTRL_LIVE)
395 nvme_kick_requeue_lists(ctrl);
396 return changed;
397 }
398 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
399
400 /*
401 * Returns true for sink states that can't ever transition back to live.
402 */
nvme_state_terminal(struct nvme_ctrl * ctrl)403 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
404 {
405 switch (ctrl->state) {
406 case NVME_CTRL_NEW:
407 case NVME_CTRL_LIVE:
408 case NVME_CTRL_RESETTING:
409 case NVME_CTRL_CONNECTING:
410 return false;
411 case NVME_CTRL_DELETING:
412 case NVME_CTRL_DEAD:
413 return true;
414 default:
415 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
416 return true;
417 }
418 }
419
420 /*
421 * Waits for the controller state to be resetting, or returns false if it is
422 * not possible to ever transition to that state.
423 */
nvme_wait_reset(struct nvme_ctrl * ctrl)424 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
425 {
426 wait_event(ctrl->state_wq,
427 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
428 nvme_state_terminal(ctrl));
429 return ctrl->state == NVME_CTRL_RESETTING;
430 }
431 EXPORT_SYMBOL_GPL(nvme_wait_reset);
432
nvme_free_ns_head(struct kref * ref)433 static void nvme_free_ns_head(struct kref *ref)
434 {
435 struct nvme_ns_head *head =
436 container_of(ref, struct nvme_ns_head, ref);
437
438 nvme_mpath_remove_disk(head);
439 ida_simple_remove(&head->subsys->ns_ida, head->instance);
440 list_del_init(&head->entry);
441 cleanup_srcu_struct(&head->srcu);
442 nvme_put_subsystem(head->subsys);
443 kfree(head);
444 }
445
nvme_put_ns_head(struct nvme_ns_head * head)446 static void nvme_put_ns_head(struct nvme_ns_head *head)
447 {
448 kref_put(&head->ref, nvme_free_ns_head);
449 }
450
nvme_free_ns(struct kref * kref)451 static void nvme_free_ns(struct kref *kref)
452 {
453 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
454
455 if (ns->ndev)
456 nvme_nvm_unregister(ns);
457
458 put_disk(ns->disk);
459 nvme_put_ns_head(ns->head);
460 nvme_put_ctrl(ns->ctrl);
461 kfree(ns);
462 }
463
nvme_put_ns(struct nvme_ns * ns)464 static void nvme_put_ns(struct nvme_ns *ns)
465 {
466 kref_put(&ns->kref, nvme_free_ns);
467 }
468
nvme_clear_nvme_request(struct request * req)469 static inline void nvme_clear_nvme_request(struct request *req)
470 {
471 if (!(req->rq_flags & RQF_DONTPREP)) {
472 nvme_req(req)->retries = 0;
473 nvme_req(req)->flags = 0;
474 req->rq_flags |= RQF_DONTPREP;
475 }
476 }
477
nvme_alloc_request(struct request_queue * q,struct nvme_command * cmd,blk_mq_req_flags_t flags,int qid)478 struct request *nvme_alloc_request(struct request_queue *q,
479 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
480 {
481 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
482 struct request *req;
483
484 if (qid == NVME_QID_ANY) {
485 req = blk_mq_alloc_request(q, op, flags);
486 } else {
487 req = blk_mq_alloc_request_hctx(q, op, flags,
488 qid ? qid - 1 : 0);
489 }
490 if (IS_ERR(req))
491 return req;
492
493 req->cmd_flags |= REQ_FAILFAST_DRIVER;
494 nvme_clear_nvme_request(req);
495 nvme_req(req)->cmd = cmd;
496
497 return req;
498 }
499 EXPORT_SYMBOL_GPL(nvme_alloc_request);
500
nvme_toggle_streams(struct nvme_ctrl * ctrl,bool enable)501 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
502 {
503 struct nvme_command c;
504
505 memset(&c, 0, sizeof(c));
506
507 c.directive.opcode = nvme_admin_directive_send;
508 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
509 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
510 c.directive.dtype = NVME_DIR_IDENTIFY;
511 c.directive.tdtype = NVME_DIR_STREAMS;
512 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
513
514 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
515 }
516
nvme_disable_streams(struct nvme_ctrl * ctrl)517 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
518 {
519 return nvme_toggle_streams(ctrl, false);
520 }
521
nvme_enable_streams(struct nvme_ctrl * ctrl)522 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
523 {
524 return nvme_toggle_streams(ctrl, true);
525 }
526
nvme_get_stream_params(struct nvme_ctrl * ctrl,struct streams_directive_params * s,u32 nsid)527 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
528 struct streams_directive_params *s, u32 nsid)
529 {
530 struct nvme_command c;
531
532 memset(&c, 0, sizeof(c));
533 memset(s, 0, sizeof(*s));
534
535 c.directive.opcode = nvme_admin_directive_recv;
536 c.directive.nsid = cpu_to_le32(nsid);
537 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
538 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
539 c.directive.dtype = NVME_DIR_STREAMS;
540
541 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
542 }
543
nvme_configure_directives(struct nvme_ctrl * ctrl)544 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
545 {
546 struct streams_directive_params s;
547 int ret;
548
549 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
550 return 0;
551 if (!streams)
552 return 0;
553
554 ret = nvme_enable_streams(ctrl);
555 if (ret)
556 return ret;
557
558 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
559 if (ret)
560 return ret;
561
562 ctrl->nssa = le16_to_cpu(s.nssa);
563 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
564 dev_info(ctrl->device, "too few streams (%u) available\n",
565 ctrl->nssa);
566 nvme_disable_streams(ctrl);
567 return 0;
568 }
569
570 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
571 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
572 return 0;
573 }
574
575 /*
576 * Check if 'req' has a write hint associated with it. If it does, assign
577 * a valid namespace stream to the write.
578 */
nvme_assign_write_stream(struct nvme_ctrl * ctrl,struct request * req,u16 * control,u32 * dsmgmt)579 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
580 struct request *req, u16 *control,
581 u32 *dsmgmt)
582 {
583 enum rw_hint streamid = req->write_hint;
584
585 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
586 streamid = 0;
587 else {
588 streamid--;
589 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
590 return;
591
592 *control |= NVME_RW_DTYPE_STREAMS;
593 *dsmgmt |= streamid << 16;
594 }
595
596 if (streamid < ARRAY_SIZE(req->q->write_hints))
597 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
598 }
599
nvme_setup_flush(struct nvme_ns * ns,struct nvme_command * cmnd)600 static inline void nvme_setup_flush(struct nvme_ns *ns,
601 struct nvme_command *cmnd)
602 {
603 cmnd->common.opcode = nvme_cmd_flush;
604 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
605 }
606
nvme_setup_discard(struct nvme_ns * ns,struct request * req,struct nvme_command * cmnd)607 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
608 struct nvme_command *cmnd)
609 {
610 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
611 struct nvme_dsm_range *range;
612 struct bio *bio;
613
614 /*
615 * Some devices do not consider the DSM 'Number of Ranges' field when
616 * determining how much data to DMA. Always allocate memory for maximum
617 * number of segments to prevent device reading beyond end of buffer.
618 */
619 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
620
621 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
622 if (!range) {
623 /*
624 * If we fail allocation our range, fallback to the controller
625 * discard page. If that's also busy, it's safe to return
626 * busy, as we know we can make progress once that's freed.
627 */
628 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
629 return BLK_STS_RESOURCE;
630
631 range = page_address(ns->ctrl->discard_page);
632 }
633
634 __rq_for_each_bio(bio, req) {
635 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
636 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
637
638 if (n < segments) {
639 range[n].cattr = cpu_to_le32(0);
640 range[n].nlb = cpu_to_le32(nlb);
641 range[n].slba = cpu_to_le64(slba);
642 }
643 n++;
644 }
645
646 if (WARN_ON_ONCE(n != segments)) {
647 if (virt_to_page(range) == ns->ctrl->discard_page)
648 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
649 else
650 kfree(range);
651 return BLK_STS_IOERR;
652 }
653
654 cmnd->dsm.opcode = nvme_cmd_dsm;
655 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
656 cmnd->dsm.nr = cpu_to_le32(segments - 1);
657 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
658
659 req->special_vec.bv_page = virt_to_page(range);
660 req->special_vec.bv_offset = offset_in_page(range);
661 req->special_vec.bv_len = alloc_size;
662 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
663
664 return BLK_STS_OK;
665 }
666
nvme_setup_write_zeroes(struct nvme_ns * ns,struct request * req,struct nvme_command * cmnd)667 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
668 struct request *req, struct nvme_command *cmnd)
669 {
670 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
671 return nvme_setup_discard(ns, req, cmnd);
672
673 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
674 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
675 cmnd->write_zeroes.slba =
676 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
677 cmnd->write_zeroes.length =
678 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
679 cmnd->write_zeroes.control = 0;
680 return BLK_STS_OK;
681 }
682
nvme_setup_rw(struct nvme_ns * ns,struct request * req,struct nvme_command * cmnd)683 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
684 struct request *req, struct nvme_command *cmnd)
685 {
686 struct nvme_ctrl *ctrl = ns->ctrl;
687 u16 control = 0;
688 u32 dsmgmt = 0;
689
690 if (req->cmd_flags & REQ_FUA)
691 control |= NVME_RW_FUA;
692 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
693 control |= NVME_RW_LR;
694
695 if (req->cmd_flags & REQ_RAHEAD)
696 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
697
698 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
699 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
700 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
701 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
702
703 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
704 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
705
706 if (ns->ms) {
707 /*
708 * If formated with metadata, the block layer always provides a
709 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
710 * we enable the PRACT bit for protection information or set the
711 * namespace capacity to zero to prevent any I/O.
712 */
713 if (!blk_integrity_rq(req)) {
714 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
715 return BLK_STS_NOTSUPP;
716 control |= NVME_RW_PRINFO_PRACT;
717 }
718
719 switch (ns->pi_type) {
720 case NVME_NS_DPS_PI_TYPE3:
721 control |= NVME_RW_PRINFO_PRCHK_GUARD;
722 break;
723 case NVME_NS_DPS_PI_TYPE1:
724 case NVME_NS_DPS_PI_TYPE2:
725 control |= NVME_RW_PRINFO_PRCHK_GUARD |
726 NVME_RW_PRINFO_PRCHK_REF;
727 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
728 break;
729 }
730 }
731
732 cmnd->rw.control = cpu_to_le16(control);
733 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
734 return 0;
735 }
736
nvme_cleanup_cmd(struct request * req)737 void nvme_cleanup_cmd(struct request *req)
738 {
739 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
740 struct nvme_ns *ns = req->rq_disk->private_data;
741 struct page *page = req->special_vec.bv_page;
742
743 if (page == ns->ctrl->discard_page)
744 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
745 else
746 kfree(page_address(page) + req->special_vec.bv_offset);
747 }
748 }
749 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
750
nvme_setup_cmd(struct nvme_ns * ns,struct request * req,struct nvme_command * cmd)751 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
752 struct nvme_command *cmd)
753 {
754 blk_status_t ret = BLK_STS_OK;
755
756 nvme_clear_nvme_request(req);
757
758 memset(cmd, 0, sizeof(*cmd));
759 switch (req_op(req)) {
760 case REQ_OP_DRV_IN:
761 case REQ_OP_DRV_OUT:
762 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
763 break;
764 case REQ_OP_FLUSH:
765 nvme_setup_flush(ns, cmd);
766 break;
767 case REQ_OP_WRITE_ZEROES:
768 ret = nvme_setup_write_zeroes(ns, req, cmd);
769 break;
770 case REQ_OP_DISCARD:
771 ret = nvme_setup_discard(ns, req, cmd);
772 break;
773 case REQ_OP_READ:
774 case REQ_OP_WRITE:
775 ret = nvme_setup_rw(ns, req, cmd);
776 break;
777 default:
778 WARN_ON_ONCE(1);
779 return BLK_STS_IOERR;
780 }
781
782 cmd->common.command_id = req->tag;
783 trace_nvme_setup_cmd(req, cmd);
784 return ret;
785 }
786 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
787
nvme_end_sync_rq(struct request * rq,blk_status_t error)788 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
789 {
790 struct completion *waiting = rq->end_io_data;
791
792 rq->end_io_data = NULL;
793 complete(waiting);
794 }
795
nvme_execute_rq_polled(struct request_queue * q,struct gendisk * bd_disk,struct request * rq,int at_head)796 static void nvme_execute_rq_polled(struct request_queue *q,
797 struct gendisk *bd_disk, struct request *rq, int at_head)
798 {
799 DECLARE_COMPLETION_ONSTACK(wait);
800
801 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
802
803 rq->cmd_flags |= REQ_HIPRI;
804 rq->end_io_data = &wait;
805 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
806
807 while (!completion_done(&wait)) {
808 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
809 cond_resched();
810 }
811 }
812
813 /*
814 * Returns 0 on success. If the result is negative, it's a Linux error code;
815 * if the result is positive, it's an NVM Express status code
816 */
__nvme_submit_sync_cmd(struct request_queue * q,struct nvme_command * cmd,union nvme_result * result,void * buffer,unsigned bufflen,unsigned timeout,int qid,int at_head,blk_mq_req_flags_t flags,bool poll)817 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
818 union nvme_result *result, void *buffer, unsigned bufflen,
819 unsigned timeout, int qid, int at_head,
820 blk_mq_req_flags_t flags, bool poll)
821 {
822 struct request *req;
823 int ret;
824
825 req = nvme_alloc_request(q, cmd, flags, qid);
826 if (IS_ERR(req))
827 return PTR_ERR(req);
828
829 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
830
831 if (buffer && bufflen) {
832 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
833 if (ret)
834 goto out;
835 }
836
837 if (poll)
838 nvme_execute_rq_polled(req->q, NULL, req, at_head);
839 else
840 blk_execute_rq(req->q, NULL, req, at_head);
841 if (result)
842 *result = nvme_req(req)->result;
843 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
844 ret = -EINTR;
845 else
846 ret = nvme_req(req)->status;
847 out:
848 blk_mq_free_request(req);
849 return ret;
850 }
851 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
852
nvme_submit_sync_cmd(struct request_queue * q,struct nvme_command * cmd,void * buffer,unsigned bufflen)853 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
854 void *buffer, unsigned bufflen)
855 {
856 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
857 NVME_QID_ANY, 0, 0, false);
858 }
859 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
860
nvme_add_user_metadata(struct bio * bio,void __user * ubuf,unsigned len,u32 seed,bool write)861 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
862 unsigned len, u32 seed, bool write)
863 {
864 struct bio_integrity_payload *bip;
865 int ret = -ENOMEM;
866 void *buf;
867
868 buf = kmalloc(len, GFP_KERNEL);
869 if (!buf)
870 goto out;
871
872 ret = -EFAULT;
873 if (write && copy_from_user(buf, ubuf, len))
874 goto out_free_meta;
875
876 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
877 if (IS_ERR(bip)) {
878 ret = PTR_ERR(bip);
879 goto out_free_meta;
880 }
881
882 bip->bip_iter.bi_size = len;
883 bip->bip_iter.bi_sector = seed;
884 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
885 offset_in_page(buf));
886 if (ret == len)
887 return buf;
888 ret = -ENOMEM;
889 out_free_meta:
890 kfree(buf);
891 out:
892 return ERR_PTR(ret);
893 }
894
nvme_submit_user_cmd(struct request_queue * q,struct nvme_command * cmd,void __user * ubuffer,unsigned bufflen,void __user * meta_buffer,unsigned meta_len,u32 meta_seed,u64 * result,unsigned timeout)895 static int nvme_submit_user_cmd(struct request_queue *q,
896 struct nvme_command *cmd, void __user *ubuffer,
897 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
898 u32 meta_seed, u64 *result, unsigned timeout)
899 {
900 bool write = nvme_is_write(cmd);
901 struct nvme_ns *ns = q->queuedata;
902 struct gendisk *disk = ns ? ns->disk : NULL;
903 struct request *req;
904 struct bio *bio = NULL;
905 void *meta = NULL;
906 int ret;
907
908 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
909 if (IS_ERR(req))
910 return PTR_ERR(req);
911
912 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
913 nvme_req(req)->flags |= NVME_REQ_USERCMD;
914
915 if (ubuffer && bufflen) {
916 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
917 GFP_KERNEL);
918 if (ret)
919 goto out;
920 bio = req->bio;
921 bio->bi_disk = disk;
922 if (disk && meta_buffer && meta_len) {
923 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
924 meta_seed, write);
925 if (IS_ERR(meta)) {
926 ret = PTR_ERR(meta);
927 goto out_unmap;
928 }
929 req->cmd_flags |= REQ_INTEGRITY;
930 }
931 }
932
933 blk_execute_rq(req->q, disk, req, 0);
934 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
935 ret = -EINTR;
936 else
937 ret = nvme_req(req)->status;
938 if (result)
939 *result = le64_to_cpu(nvme_req(req)->result.u64);
940 if (meta && !ret && !write) {
941 if (copy_to_user(meta_buffer, meta, meta_len))
942 ret = -EFAULT;
943 }
944 kfree(meta);
945 out_unmap:
946 if (bio)
947 blk_rq_unmap_user(bio);
948 out:
949 blk_mq_free_request(req);
950 return ret;
951 }
952
nvme_keep_alive_end_io(struct request * rq,blk_status_t status)953 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
954 {
955 struct nvme_ctrl *ctrl = rq->end_io_data;
956 unsigned long flags;
957 bool startka = false;
958
959 blk_mq_free_request(rq);
960
961 if (status) {
962 dev_err(ctrl->device,
963 "failed nvme_keep_alive_end_io error=%d\n",
964 status);
965 return;
966 }
967
968 ctrl->comp_seen = false;
969 spin_lock_irqsave(&ctrl->lock, flags);
970 if (ctrl->state == NVME_CTRL_LIVE ||
971 ctrl->state == NVME_CTRL_CONNECTING)
972 startka = true;
973 spin_unlock_irqrestore(&ctrl->lock, flags);
974 if (startka)
975 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
976 }
977
nvme_keep_alive(struct nvme_ctrl * ctrl)978 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
979 {
980 struct request *rq;
981
982 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
983 NVME_QID_ANY);
984 if (IS_ERR(rq))
985 return PTR_ERR(rq);
986
987 rq->timeout = ctrl->kato * HZ;
988 rq->end_io_data = ctrl;
989
990 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
991
992 return 0;
993 }
994
nvme_keep_alive_work(struct work_struct * work)995 static void nvme_keep_alive_work(struct work_struct *work)
996 {
997 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
998 struct nvme_ctrl, ka_work);
999 bool comp_seen = ctrl->comp_seen;
1000
1001 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1002 dev_dbg(ctrl->device,
1003 "reschedule traffic based keep-alive timer\n");
1004 ctrl->comp_seen = false;
1005 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
1006 return;
1007 }
1008
1009 if (nvme_keep_alive(ctrl)) {
1010 /* allocation failure, reset the controller */
1011 dev_err(ctrl->device, "keep-alive failed\n");
1012 nvme_reset_ctrl(ctrl);
1013 return;
1014 }
1015 }
1016
nvme_start_keep_alive(struct nvme_ctrl * ctrl)1017 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1018 {
1019 if (unlikely(ctrl->kato == 0))
1020 return;
1021
1022 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
1023 }
1024
nvme_stop_keep_alive(struct nvme_ctrl * ctrl)1025 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1026 {
1027 if (unlikely(ctrl->kato == 0))
1028 return;
1029
1030 cancel_delayed_work_sync(&ctrl->ka_work);
1031 }
1032 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1033
nvme_identify_ctrl(struct nvme_ctrl * dev,struct nvme_id_ctrl ** id)1034 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1035 {
1036 struct nvme_command c = { };
1037 int error;
1038
1039 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1040 c.identify.opcode = nvme_admin_identify;
1041 c.identify.cns = NVME_ID_CNS_CTRL;
1042
1043 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1044 if (!*id)
1045 return -ENOMEM;
1046
1047 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1048 sizeof(struct nvme_id_ctrl));
1049 if (error)
1050 kfree(*id);
1051 return error;
1052 }
1053
nvme_identify_ns_descs(struct nvme_ctrl * ctrl,unsigned nsid,struct nvme_ns_ids * ids)1054 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1055 struct nvme_ns_ids *ids)
1056 {
1057 struct nvme_command c = { };
1058 int status;
1059 void *data;
1060 int pos;
1061 int len;
1062
1063 c.identify.opcode = nvme_admin_identify;
1064 c.identify.nsid = cpu_to_le32(nsid);
1065 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1066
1067 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1068 if (!data)
1069 return -ENOMEM;
1070
1071 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1072 NVME_IDENTIFY_DATA_SIZE);
1073 if (status)
1074 goto free_data;
1075
1076 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1077 struct nvme_ns_id_desc *cur = data + pos;
1078
1079 if (cur->nidl == 0)
1080 break;
1081
1082 switch (cur->nidt) {
1083 case NVME_NIDT_EUI64:
1084 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1085 dev_warn(ctrl->device,
1086 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1087 cur->nidl);
1088 goto free_data;
1089 }
1090 len = NVME_NIDT_EUI64_LEN;
1091 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1092 break;
1093 case NVME_NIDT_NGUID:
1094 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1095 dev_warn(ctrl->device,
1096 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1097 cur->nidl);
1098 goto free_data;
1099 }
1100 len = NVME_NIDT_NGUID_LEN;
1101 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1102 break;
1103 case NVME_NIDT_UUID:
1104 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1105 dev_warn(ctrl->device,
1106 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1107 cur->nidl);
1108 goto free_data;
1109 }
1110 len = NVME_NIDT_UUID_LEN;
1111 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1112 break;
1113 default:
1114 /* Skip unknown types */
1115 len = cur->nidl;
1116 break;
1117 }
1118
1119 len += sizeof(*cur);
1120 }
1121 free_data:
1122 kfree(data);
1123 return status;
1124 }
1125
nvme_identify_ns_list(struct nvme_ctrl * dev,unsigned nsid,__le32 * ns_list)1126 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1127 {
1128 struct nvme_command c = { };
1129
1130 c.identify.opcode = nvme_admin_identify;
1131 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1132 c.identify.nsid = cpu_to_le32(nsid);
1133 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1134 NVME_IDENTIFY_DATA_SIZE);
1135 }
1136
nvme_identify_ns(struct nvme_ctrl * ctrl,unsigned nsid,struct nvme_id_ns ** id)1137 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1138 unsigned nsid, struct nvme_id_ns **id)
1139 {
1140 struct nvme_command c = { };
1141 int error;
1142
1143 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1144 c.identify.opcode = nvme_admin_identify;
1145 c.identify.nsid = cpu_to_le32(nsid);
1146 c.identify.cns = NVME_ID_CNS_NS;
1147
1148 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1149 if (!*id)
1150 return -ENOMEM;
1151
1152 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1153 if (error) {
1154 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1155 kfree(*id);
1156 }
1157
1158 return error;
1159 }
1160
nvme_features(struct nvme_ctrl * dev,u8 op,unsigned int fid,unsigned int dword11,void * buffer,size_t buflen,u32 * result)1161 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1162 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1163 {
1164 struct nvme_command c;
1165 union nvme_result res;
1166 int ret;
1167
1168 memset(&c, 0, sizeof(c));
1169 c.features.opcode = op;
1170 c.features.fid = cpu_to_le32(fid);
1171 c.features.dword11 = cpu_to_le32(dword11);
1172
1173 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1174 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1175 if (ret >= 0 && result)
1176 *result = le32_to_cpu(res.u32);
1177 return ret;
1178 }
1179
nvme_set_features(struct nvme_ctrl * dev,unsigned int fid,unsigned int dword11,void * buffer,size_t buflen,u32 * result)1180 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1181 unsigned int dword11, void *buffer, size_t buflen,
1182 u32 *result)
1183 {
1184 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1185 buflen, result);
1186 }
1187 EXPORT_SYMBOL_GPL(nvme_set_features);
1188
nvme_get_features(struct nvme_ctrl * dev,unsigned int fid,unsigned int dword11,void * buffer,size_t buflen,u32 * result)1189 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1190 unsigned int dword11, void *buffer, size_t buflen,
1191 u32 *result)
1192 {
1193 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1194 buflen, result);
1195 }
1196 EXPORT_SYMBOL_GPL(nvme_get_features);
1197
nvme_set_queue_count(struct nvme_ctrl * ctrl,int * count)1198 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1199 {
1200 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1201 u32 result;
1202 int status, nr_io_queues;
1203
1204 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1205 &result);
1206 if (status < 0)
1207 return status;
1208
1209 /*
1210 * Degraded controllers might return an error when setting the queue
1211 * count. We still want to be able to bring them online and offer
1212 * access to the admin queue, as that might be only way to fix them up.
1213 */
1214 if (status > 0) {
1215 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1216 *count = 0;
1217 } else {
1218 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1219 *count = min(*count, nr_io_queues);
1220 }
1221
1222 return 0;
1223 }
1224 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1225
1226 #define NVME_AEN_SUPPORTED \
1227 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1228 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1229
nvme_enable_aen(struct nvme_ctrl * ctrl)1230 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1231 {
1232 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1233 int status;
1234
1235 if (!supported_aens)
1236 return;
1237
1238 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1239 NULL, 0, &result);
1240 if (status)
1241 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1242 supported_aens);
1243
1244 queue_work(nvme_wq, &ctrl->async_event_work);
1245 }
1246
nvme_submit_io(struct nvme_ns * ns,struct nvme_user_io __user * uio)1247 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1248 {
1249 struct nvme_user_io io;
1250 struct nvme_command c;
1251 unsigned length, meta_len;
1252 void __user *metadata;
1253
1254 if (copy_from_user(&io, uio, sizeof(io)))
1255 return -EFAULT;
1256 if (io.flags)
1257 return -EINVAL;
1258
1259 switch (io.opcode) {
1260 case nvme_cmd_write:
1261 case nvme_cmd_read:
1262 case nvme_cmd_compare:
1263 break;
1264 default:
1265 return -EINVAL;
1266 }
1267
1268 length = (io.nblocks + 1) << ns->lba_shift;
1269 meta_len = (io.nblocks + 1) * ns->ms;
1270 metadata = (void __user *)(uintptr_t)io.metadata;
1271
1272 if (ns->ext) {
1273 length += meta_len;
1274 meta_len = 0;
1275 } else if (meta_len) {
1276 if ((io.metadata & 3) || !io.metadata)
1277 return -EINVAL;
1278 }
1279
1280 memset(&c, 0, sizeof(c));
1281 c.rw.opcode = io.opcode;
1282 c.rw.flags = io.flags;
1283 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1284 c.rw.slba = cpu_to_le64(io.slba);
1285 c.rw.length = cpu_to_le16(io.nblocks);
1286 c.rw.control = cpu_to_le16(io.control);
1287 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1288 c.rw.reftag = cpu_to_le32(io.reftag);
1289 c.rw.apptag = cpu_to_le16(io.apptag);
1290 c.rw.appmask = cpu_to_le16(io.appmask);
1291
1292 return nvme_submit_user_cmd(ns->queue, &c,
1293 (void __user *)(uintptr_t)io.addr, length,
1294 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1295 }
1296
nvme_known_admin_effects(u8 opcode)1297 static u32 nvme_known_admin_effects(u8 opcode)
1298 {
1299 switch (opcode) {
1300 case nvme_admin_format_nvm:
1301 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1302 NVME_CMD_EFFECTS_CSE_MASK;
1303 case nvme_admin_sanitize_nvm:
1304 return NVME_CMD_EFFECTS_CSE_MASK;
1305 default:
1306 break;
1307 }
1308 return 0;
1309 }
1310
nvme_passthru_start(struct nvme_ctrl * ctrl,struct nvme_ns * ns,u8 opcode)1311 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1312 u8 opcode)
1313 {
1314 u32 effects = 0;
1315
1316 if (ns) {
1317 if (ctrl->effects)
1318 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1319 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1320 dev_warn(ctrl->device,
1321 "IO command:%02x has unhandled effects:%08x\n",
1322 opcode, effects);
1323 return 0;
1324 }
1325
1326 if (ctrl->effects)
1327 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1328 effects |= nvme_known_admin_effects(opcode);
1329
1330 /*
1331 * For simplicity, IO to all namespaces is quiesced even if the command
1332 * effects say only one namespace is affected.
1333 */
1334 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1335 mutex_lock(&ctrl->scan_lock);
1336 mutex_lock(&ctrl->subsys->lock);
1337 nvme_mpath_start_freeze(ctrl->subsys);
1338 nvme_mpath_wait_freeze(ctrl->subsys);
1339 nvme_start_freeze(ctrl);
1340 nvme_wait_freeze(ctrl);
1341 }
1342 return effects;
1343 }
1344
nvme_update_formats(struct nvme_ctrl * ctrl)1345 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1346 {
1347 struct nvme_ns *ns;
1348
1349 down_read(&ctrl->namespaces_rwsem);
1350 list_for_each_entry(ns, &ctrl->namespaces, list)
1351 if (ns->disk && nvme_revalidate_disk(ns->disk))
1352 nvme_set_queue_dying(ns);
1353 up_read(&ctrl->namespaces_rwsem);
1354 }
1355
nvme_passthru_end(struct nvme_ctrl * ctrl,u32 effects)1356 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1357 {
1358 /*
1359 * Revalidate LBA changes prior to unfreezing. This is necessary to
1360 * prevent memory corruption if a logical block size was changed by
1361 * this command.
1362 */
1363 if (effects & NVME_CMD_EFFECTS_LBCC)
1364 nvme_update_formats(ctrl);
1365 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1366 nvme_unfreeze(ctrl);
1367 nvme_mpath_unfreeze(ctrl->subsys);
1368 mutex_unlock(&ctrl->subsys->lock);
1369 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1370 mutex_unlock(&ctrl->scan_lock);
1371 }
1372 if (effects & NVME_CMD_EFFECTS_CCC)
1373 nvme_init_identify(ctrl);
1374 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1375 nvme_queue_scan(ctrl);
1376 }
1377
nvme_user_cmd(struct nvme_ctrl * ctrl,struct nvme_ns * ns,struct nvme_passthru_cmd __user * ucmd)1378 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1379 struct nvme_passthru_cmd __user *ucmd)
1380 {
1381 struct nvme_passthru_cmd cmd;
1382 struct nvme_command c;
1383 unsigned timeout = 0;
1384 u32 effects;
1385 u64 result;
1386 int status;
1387
1388 if (!capable(CAP_SYS_ADMIN))
1389 return -EACCES;
1390 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1391 return -EFAULT;
1392 if (cmd.flags)
1393 return -EINVAL;
1394
1395 memset(&c, 0, sizeof(c));
1396 c.common.opcode = cmd.opcode;
1397 c.common.flags = cmd.flags;
1398 c.common.nsid = cpu_to_le32(cmd.nsid);
1399 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1400 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1401 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1402 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1403 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1404 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1405 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1406 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1407
1408 if (cmd.timeout_ms)
1409 timeout = msecs_to_jiffies(cmd.timeout_ms);
1410
1411 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1412 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1413 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1414 (void __user *)(uintptr_t)cmd.metadata,
1415 cmd.metadata_len, 0, &result, timeout);
1416 nvme_passthru_end(ctrl, effects);
1417
1418 if (status >= 0) {
1419 if (put_user(result, &ucmd->result))
1420 return -EFAULT;
1421 }
1422
1423 return status;
1424 }
1425
nvme_user_cmd64(struct nvme_ctrl * ctrl,struct nvme_ns * ns,struct nvme_passthru_cmd64 __user * ucmd)1426 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1427 struct nvme_passthru_cmd64 __user *ucmd)
1428 {
1429 struct nvme_passthru_cmd64 cmd;
1430 struct nvme_command c;
1431 unsigned timeout = 0;
1432 u32 effects;
1433 int status;
1434
1435 if (!capable(CAP_SYS_ADMIN))
1436 return -EACCES;
1437 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1438 return -EFAULT;
1439 if (cmd.flags)
1440 return -EINVAL;
1441
1442 memset(&c, 0, sizeof(c));
1443 c.common.opcode = cmd.opcode;
1444 c.common.flags = cmd.flags;
1445 c.common.nsid = cpu_to_le32(cmd.nsid);
1446 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1447 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1448 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1449 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1450 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1451 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1452 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1453 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1454
1455 if (cmd.timeout_ms)
1456 timeout = msecs_to_jiffies(cmd.timeout_ms);
1457
1458 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1459 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1460 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1461 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1462 0, &cmd.result, timeout);
1463 nvme_passthru_end(ctrl, effects);
1464
1465 if (status >= 0) {
1466 if (put_user(cmd.result, &ucmd->result))
1467 return -EFAULT;
1468 }
1469
1470 return status;
1471 }
1472
1473 /*
1474 * Issue ioctl requests on the first available path. Note that unlike normal
1475 * block layer requests we will not retry failed request on another controller.
1476 */
nvme_get_ns_from_disk(struct gendisk * disk,struct nvme_ns_head ** head,int * srcu_idx)1477 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1478 struct nvme_ns_head **head, int *srcu_idx)
1479 {
1480 #ifdef CONFIG_NVME_MULTIPATH
1481 if (disk->fops == &nvme_ns_head_ops) {
1482 struct nvme_ns *ns;
1483
1484 *head = disk->private_data;
1485 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1486 ns = nvme_find_path(*head);
1487 if (!ns)
1488 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1489 return ns;
1490 }
1491 #endif
1492 *head = NULL;
1493 *srcu_idx = -1;
1494 return disk->private_data;
1495 }
1496
nvme_put_ns_from_disk(struct nvme_ns_head * head,int idx)1497 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1498 {
1499 if (head)
1500 srcu_read_unlock(&head->srcu, idx);
1501 }
1502
is_ctrl_ioctl(unsigned int cmd)1503 static bool is_ctrl_ioctl(unsigned int cmd)
1504 {
1505 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1506 return true;
1507 if (is_sed_ioctl(cmd))
1508 return true;
1509 return false;
1510 }
1511
nvme_handle_ctrl_ioctl(struct nvme_ns * ns,unsigned int cmd,void __user * argp,struct nvme_ns_head * head,int srcu_idx)1512 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1513 void __user *argp,
1514 struct nvme_ns_head *head,
1515 int srcu_idx)
1516 {
1517 struct nvme_ctrl *ctrl = ns->ctrl;
1518 int ret;
1519
1520 nvme_get_ctrl(ns->ctrl);
1521 nvme_put_ns_from_disk(head, srcu_idx);
1522
1523 switch (cmd) {
1524 case NVME_IOCTL_ADMIN_CMD:
1525 ret = nvme_user_cmd(ctrl, NULL, argp);
1526 break;
1527 case NVME_IOCTL_ADMIN64_CMD:
1528 ret = nvme_user_cmd64(ctrl, NULL, argp);
1529 break;
1530 default:
1531 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1532 break;
1533 }
1534 nvme_put_ctrl(ctrl);
1535 return ret;
1536 }
1537
nvme_ioctl(struct block_device * bdev,fmode_t mode,unsigned int cmd,unsigned long arg)1538 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1539 unsigned int cmd, unsigned long arg)
1540 {
1541 struct nvme_ns_head *head = NULL;
1542 void __user *argp = (void __user *)arg;
1543 struct nvme_ns *ns;
1544 int srcu_idx, ret;
1545
1546 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1547 if (unlikely(!ns))
1548 return -EWOULDBLOCK;
1549
1550 /*
1551 * Handle ioctls that apply to the controller instead of the namespace
1552 * seperately and drop the ns SRCU reference early. This avoids a
1553 * deadlock when deleting namespaces using the passthrough interface.
1554 */
1555 if (is_ctrl_ioctl(cmd))
1556 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1557
1558 switch (cmd) {
1559 case NVME_IOCTL_ID:
1560 force_successful_syscall_return();
1561 ret = ns->head->ns_id;
1562 break;
1563 case NVME_IOCTL_IO_CMD:
1564 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1565 break;
1566 case NVME_IOCTL_SUBMIT_IO:
1567 ret = nvme_submit_io(ns, argp);
1568 break;
1569 case NVME_IOCTL_IO64_CMD:
1570 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1571 break;
1572 default:
1573 if (ns->ndev)
1574 ret = nvme_nvm_ioctl(ns, cmd, arg);
1575 else
1576 ret = -ENOTTY;
1577 }
1578
1579 nvme_put_ns_from_disk(head, srcu_idx);
1580 return ret;
1581 }
1582
nvme_open(struct block_device * bdev,fmode_t mode)1583 static int nvme_open(struct block_device *bdev, fmode_t mode)
1584 {
1585 struct nvme_ns *ns = bdev->bd_disk->private_data;
1586
1587 #ifdef CONFIG_NVME_MULTIPATH
1588 /* should never be called due to GENHD_FL_HIDDEN */
1589 if (WARN_ON_ONCE(ns->head->disk))
1590 goto fail;
1591 #endif
1592 if (!kref_get_unless_zero(&ns->kref))
1593 goto fail;
1594 if (!try_module_get(ns->ctrl->ops->module))
1595 goto fail_put_ns;
1596
1597 return 0;
1598
1599 fail_put_ns:
1600 nvme_put_ns(ns);
1601 fail:
1602 return -ENXIO;
1603 }
1604
nvme_release(struct gendisk * disk,fmode_t mode)1605 static void nvme_release(struct gendisk *disk, fmode_t mode)
1606 {
1607 struct nvme_ns *ns = disk->private_data;
1608
1609 module_put(ns->ctrl->ops->module);
1610 nvme_put_ns(ns);
1611 }
1612
nvme_getgeo(struct block_device * bdev,struct hd_geometry * geo)1613 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1614 {
1615 /* some standard values */
1616 geo->heads = 1 << 6;
1617 geo->sectors = 1 << 5;
1618 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1619 return 0;
1620 }
1621
1622 #ifdef CONFIG_BLK_DEV_INTEGRITY
nvme_init_integrity(struct gendisk * disk,u16 ms,u8 pi_type)1623 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1624 {
1625 struct blk_integrity integrity;
1626
1627 memset(&integrity, 0, sizeof(integrity));
1628 switch (pi_type) {
1629 case NVME_NS_DPS_PI_TYPE3:
1630 integrity.profile = &t10_pi_type3_crc;
1631 integrity.tag_size = sizeof(u16) + sizeof(u32);
1632 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1633 break;
1634 case NVME_NS_DPS_PI_TYPE1:
1635 case NVME_NS_DPS_PI_TYPE2:
1636 integrity.profile = &t10_pi_type1_crc;
1637 integrity.tag_size = sizeof(u16);
1638 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1639 break;
1640 default:
1641 integrity.profile = NULL;
1642 break;
1643 }
1644 integrity.tuple_size = ms;
1645 blk_integrity_register(disk, &integrity);
1646 blk_queue_max_integrity_segments(disk->queue, 1);
1647 }
1648 #else
nvme_init_integrity(struct gendisk * disk,u16 ms,u8 pi_type)1649 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1650 {
1651 }
1652 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1653
nvme_set_chunk_size(struct nvme_ns * ns)1654 static void nvme_set_chunk_size(struct nvme_ns *ns)
1655 {
1656 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1657 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1658 }
1659
nvme_config_discard(struct gendisk * disk,struct nvme_ns * ns)1660 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1661 {
1662 struct nvme_ctrl *ctrl = ns->ctrl;
1663 struct request_queue *queue = disk->queue;
1664 u32 size = queue_logical_block_size(queue);
1665
1666 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1667 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1668 return;
1669 }
1670
1671 if (ctrl->nr_streams && ns->sws && ns->sgs)
1672 size *= ns->sws * ns->sgs;
1673
1674 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1675 NVME_DSM_MAX_RANGES);
1676
1677 queue->limits.discard_alignment = 0;
1678 queue->limits.discard_granularity = size;
1679
1680 /* If discard is already enabled, don't reset queue limits */
1681 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1682 return;
1683
1684 blk_queue_max_discard_sectors(queue, UINT_MAX);
1685 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1686
1687 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1688 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1689 }
1690
nvme_config_write_zeroes(struct gendisk * disk,struct nvme_ns * ns)1691 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1692 {
1693 u32 max_sectors;
1694 unsigned short bs = 1 << ns->lba_shift;
1695
1696 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1697 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1698 return;
1699 /*
1700 * Even though NVMe spec explicitly states that MDTS is not
1701 * applicable to the write-zeroes:- "The restriction does not apply to
1702 * commands that do not transfer data between the host and the
1703 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1704 * In order to be more cautious use controller's max_hw_sectors value
1705 * to configure the maximum sectors for the write-zeroes which is
1706 * configured based on the controller's MDTS field in the
1707 * nvme_init_identify() if available.
1708 */
1709 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1710 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1711 else
1712 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1713
1714 blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1715 }
1716
nvme_report_ns_ids(struct nvme_ctrl * ctrl,unsigned int nsid,struct nvme_id_ns * id,struct nvme_ns_ids * ids)1717 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1718 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1719 {
1720 int ret = 0;
1721
1722 memset(ids, 0, sizeof(*ids));
1723
1724 if (ctrl->vs >= NVME_VS(1, 1, 0))
1725 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1726 if (ctrl->vs >= NVME_VS(1, 2, 0))
1727 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1728 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1729 /* Don't treat error as fatal we potentially
1730 * already have a NGUID or EUI-64
1731 */
1732 ret = nvme_identify_ns_descs(ctrl, nsid, ids);
1733 if (ret)
1734 dev_warn(ctrl->device,
1735 "Identify Descriptors failed (%d)\n", ret);
1736 if (ret > 0)
1737 ret = 0;
1738 }
1739 return ret;
1740 }
1741
nvme_ns_ids_valid(struct nvme_ns_ids * ids)1742 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1743 {
1744 return !uuid_is_null(&ids->uuid) ||
1745 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1746 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1747 }
1748
nvme_ns_ids_equal(struct nvme_ns_ids * a,struct nvme_ns_ids * b)1749 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1750 {
1751 return uuid_equal(&a->uuid, &b->uuid) &&
1752 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1753 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1754 }
1755
nvme_update_disk_info(struct gendisk * disk,struct nvme_ns * ns,struct nvme_id_ns * id)1756 static void nvme_update_disk_info(struct gendisk *disk,
1757 struct nvme_ns *ns, struct nvme_id_ns *id)
1758 {
1759 sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1760 unsigned short bs = 1 << ns->lba_shift;
1761 u32 atomic_bs, phys_bs, io_opt;
1762
1763 if (ns->lba_shift > PAGE_SHIFT) {
1764 /* unsupported block size, set capacity to 0 later */
1765 bs = (1 << 9);
1766 }
1767 blk_mq_freeze_queue(disk->queue);
1768 blk_integrity_unregister(disk);
1769
1770 if (id->nabo == 0) {
1771 /*
1772 * Bit 1 indicates whether NAWUPF is defined for this namespace
1773 * and whether it should be used instead of AWUPF. If NAWUPF ==
1774 * 0 then AWUPF must be used instead.
1775 */
1776 if (id->nsfeat & (1 << 1) && id->nawupf)
1777 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1778 else
1779 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1780 } else {
1781 atomic_bs = bs;
1782 }
1783 phys_bs = bs;
1784 io_opt = bs;
1785 if (id->nsfeat & (1 << 4)) {
1786 /* NPWG = Namespace Preferred Write Granularity */
1787 phys_bs *= 1 + le16_to_cpu(id->npwg);
1788 /* NOWS = Namespace Optimal Write Size */
1789 io_opt *= 1 + le16_to_cpu(id->nows);
1790 }
1791
1792 blk_queue_logical_block_size(disk->queue, bs);
1793 /*
1794 * Linux filesystems assume writing a single physical block is
1795 * an atomic operation. Hence limit the physical block size to the
1796 * value of the Atomic Write Unit Power Fail parameter.
1797 */
1798 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1799 blk_queue_io_min(disk->queue, phys_bs);
1800 blk_queue_io_opt(disk->queue, io_opt);
1801
1802 if (ns->ms && !ns->ext &&
1803 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1804 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1805 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1806 ns->lba_shift > PAGE_SHIFT)
1807 capacity = 0;
1808
1809 set_capacity(disk, capacity);
1810
1811 nvme_config_discard(disk, ns);
1812 nvme_config_write_zeroes(disk, ns);
1813
1814 if (id->nsattr & (1 << 0))
1815 set_disk_ro(disk, true);
1816 else
1817 set_disk_ro(disk, false);
1818
1819 blk_mq_unfreeze_queue(disk->queue);
1820 }
1821
__nvme_revalidate_disk(struct gendisk * disk,struct nvme_id_ns * id)1822 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1823 {
1824 struct nvme_ns *ns = disk->private_data;
1825
1826 /*
1827 * If identify namespace failed, use default 512 byte block size so
1828 * block layer can use before failing read/write for 0 capacity.
1829 */
1830 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1831 if (ns->lba_shift == 0)
1832 ns->lba_shift = 9;
1833 ns->noiob = le16_to_cpu(id->noiob);
1834 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1835 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1836 /* the PI implementation requires metadata equal t10 pi tuple size */
1837 if (ns->ms == sizeof(struct t10_pi_tuple))
1838 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1839 else
1840 ns->pi_type = 0;
1841
1842 if (ns->noiob)
1843 nvme_set_chunk_size(ns);
1844 nvme_update_disk_info(disk, ns, id);
1845 #ifdef CONFIG_NVME_MULTIPATH
1846 if (ns->head->disk) {
1847 nvme_update_disk_info(ns->head->disk, ns, id);
1848 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1849 revalidate_disk(ns->head->disk);
1850 }
1851 #endif
1852 }
1853
nvme_revalidate_disk(struct gendisk * disk)1854 static int nvme_revalidate_disk(struct gendisk *disk)
1855 {
1856 struct nvme_ns *ns = disk->private_data;
1857 struct nvme_ctrl *ctrl = ns->ctrl;
1858 struct nvme_id_ns *id;
1859 struct nvme_ns_ids ids;
1860 int ret = 0;
1861
1862 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1863 set_capacity(disk, 0);
1864 return -ENODEV;
1865 }
1866
1867 ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1868 if (ret)
1869 goto out;
1870
1871 if (id->ncap == 0) {
1872 ret = -ENODEV;
1873 goto free_id;
1874 }
1875
1876 __nvme_revalidate_disk(disk, id);
1877 ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1878 if (ret)
1879 goto free_id;
1880
1881 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1882 dev_err(ctrl->device,
1883 "identifiers changed for nsid %d\n", ns->head->ns_id);
1884 ret = -ENODEV;
1885 }
1886
1887 free_id:
1888 kfree(id);
1889 out:
1890 /*
1891 * Only fail the function if we got a fatal error back from the
1892 * device, otherwise ignore the error and just move on.
1893 */
1894 if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1895 ret = 0;
1896 else if (ret > 0)
1897 ret = blk_status_to_errno(nvme_error_status(ret));
1898 return ret;
1899 }
1900
nvme_pr_type(enum pr_type type)1901 static char nvme_pr_type(enum pr_type type)
1902 {
1903 switch (type) {
1904 case PR_WRITE_EXCLUSIVE:
1905 return 1;
1906 case PR_EXCLUSIVE_ACCESS:
1907 return 2;
1908 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1909 return 3;
1910 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1911 return 4;
1912 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1913 return 5;
1914 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1915 return 6;
1916 default:
1917 return 0;
1918 }
1919 };
1920
nvme_pr_command(struct block_device * bdev,u32 cdw10,u64 key,u64 sa_key,u8 op)1921 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1922 u64 key, u64 sa_key, u8 op)
1923 {
1924 struct nvme_ns_head *head = NULL;
1925 struct nvme_ns *ns;
1926 struct nvme_command c;
1927 int srcu_idx, ret;
1928 u8 data[16] = { 0, };
1929
1930 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1931 if (unlikely(!ns))
1932 return -EWOULDBLOCK;
1933
1934 put_unaligned_le64(key, &data[0]);
1935 put_unaligned_le64(sa_key, &data[8]);
1936
1937 memset(&c, 0, sizeof(c));
1938 c.common.opcode = op;
1939 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1940 c.common.cdw10 = cpu_to_le32(cdw10);
1941
1942 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1943 nvme_put_ns_from_disk(head, srcu_idx);
1944 return ret;
1945 }
1946
nvme_pr_register(struct block_device * bdev,u64 old,u64 new,unsigned flags)1947 static int nvme_pr_register(struct block_device *bdev, u64 old,
1948 u64 new, unsigned flags)
1949 {
1950 u32 cdw10;
1951
1952 if (flags & ~PR_FL_IGNORE_KEY)
1953 return -EOPNOTSUPP;
1954
1955 cdw10 = old ? 2 : 0;
1956 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1957 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1958 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1959 }
1960
nvme_pr_reserve(struct block_device * bdev,u64 key,enum pr_type type,unsigned flags)1961 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1962 enum pr_type type, unsigned flags)
1963 {
1964 u32 cdw10;
1965
1966 if (flags & ~PR_FL_IGNORE_KEY)
1967 return -EOPNOTSUPP;
1968
1969 cdw10 = nvme_pr_type(type) << 8;
1970 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1971 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1972 }
1973
nvme_pr_preempt(struct block_device * bdev,u64 old,u64 new,enum pr_type type,bool abort)1974 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1975 enum pr_type type, bool abort)
1976 {
1977 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1978 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1979 }
1980
nvme_pr_clear(struct block_device * bdev,u64 key)1981 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1982 {
1983 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1984 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1985 }
1986
nvme_pr_release(struct block_device * bdev,u64 key,enum pr_type type)1987 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1988 {
1989 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1990 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1991 }
1992
1993 static const struct pr_ops nvme_pr_ops = {
1994 .pr_register = nvme_pr_register,
1995 .pr_reserve = nvme_pr_reserve,
1996 .pr_release = nvme_pr_release,
1997 .pr_preempt = nvme_pr_preempt,
1998 .pr_clear = nvme_pr_clear,
1999 };
2000
2001 #ifdef CONFIG_BLK_SED_OPAL
nvme_sec_submit(void * data,u16 spsp,u8 secp,void * buffer,size_t len,bool send)2002 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2003 bool send)
2004 {
2005 struct nvme_ctrl *ctrl = data;
2006 struct nvme_command cmd;
2007
2008 memset(&cmd, 0, sizeof(cmd));
2009 if (send)
2010 cmd.common.opcode = nvme_admin_security_send;
2011 else
2012 cmd.common.opcode = nvme_admin_security_recv;
2013 cmd.common.nsid = 0;
2014 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2015 cmd.common.cdw11 = cpu_to_le32(len);
2016
2017 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2018 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2019 }
2020 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2021 #endif /* CONFIG_BLK_SED_OPAL */
2022
2023 static const struct block_device_operations nvme_fops = {
2024 .owner = THIS_MODULE,
2025 .ioctl = nvme_ioctl,
2026 .compat_ioctl = nvme_ioctl,
2027 .open = nvme_open,
2028 .release = nvme_release,
2029 .getgeo = nvme_getgeo,
2030 .revalidate_disk= nvme_revalidate_disk,
2031 .pr_ops = &nvme_pr_ops,
2032 };
2033
2034 #ifdef CONFIG_NVME_MULTIPATH
nvme_ns_head_open(struct block_device * bdev,fmode_t mode)2035 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2036 {
2037 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2038
2039 if (!kref_get_unless_zero(&head->ref))
2040 return -ENXIO;
2041 return 0;
2042 }
2043
nvme_ns_head_release(struct gendisk * disk,fmode_t mode)2044 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2045 {
2046 nvme_put_ns_head(disk->private_data);
2047 }
2048
2049 const struct block_device_operations nvme_ns_head_ops = {
2050 .owner = THIS_MODULE,
2051 .open = nvme_ns_head_open,
2052 .release = nvme_ns_head_release,
2053 .ioctl = nvme_ioctl,
2054 .compat_ioctl = nvme_ioctl,
2055 .getgeo = nvme_getgeo,
2056 .pr_ops = &nvme_pr_ops,
2057 };
2058 #endif /* CONFIG_NVME_MULTIPATH */
2059
nvme_wait_ready(struct nvme_ctrl * ctrl,u64 cap,bool enabled)2060 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2061 {
2062 unsigned long timeout =
2063 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2064 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2065 int ret;
2066
2067 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2068 if (csts == ~0)
2069 return -ENODEV;
2070 if ((csts & NVME_CSTS_RDY) == bit)
2071 break;
2072
2073 msleep(100);
2074 if (fatal_signal_pending(current))
2075 return -EINTR;
2076 if (time_after(jiffies, timeout)) {
2077 dev_err(ctrl->device,
2078 "Device not ready; aborting %s\n", enabled ?
2079 "initialisation" : "reset");
2080 return -ENODEV;
2081 }
2082 }
2083
2084 return ret;
2085 }
2086
2087 /*
2088 * If the device has been passed off to us in an enabled state, just clear
2089 * the enabled bit. The spec says we should set the 'shutdown notification
2090 * bits', but doing so may cause the device to complete commands to the
2091 * admin queue ... and we don't know what memory that might be pointing at!
2092 */
nvme_disable_ctrl(struct nvme_ctrl * ctrl)2093 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2094 {
2095 int ret;
2096
2097 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2098 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2099
2100 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2101 if (ret)
2102 return ret;
2103
2104 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2105 msleep(NVME_QUIRK_DELAY_AMOUNT);
2106
2107 return nvme_wait_ready(ctrl, ctrl->cap, false);
2108 }
2109 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2110
nvme_enable_ctrl(struct nvme_ctrl * ctrl)2111 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2112 {
2113 /*
2114 * Default to a 4K page size, with the intention to update this
2115 * path in the future to accomodate architectures with differing
2116 * kernel and IO page sizes.
2117 */
2118 unsigned dev_page_min, page_shift = 12;
2119 int ret;
2120
2121 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2122 if (ret) {
2123 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2124 return ret;
2125 }
2126 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2127
2128 if (page_shift < dev_page_min) {
2129 dev_err(ctrl->device,
2130 "Minimum device page size %u too large for host (%u)\n",
2131 1 << dev_page_min, 1 << page_shift);
2132 return -ENODEV;
2133 }
2134
2135 ctrl->page_size = 1 << page_shift;
2136
2137 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2138 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2139 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2140 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2141 ctrl->ctrl_config |= NVME_CC_ENABLE;
2142
2143 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2144 if (ret)
2145 return ret;
2146 return nvme_wait_ready(ctrl, ctrl->cap, true);
2147 }
2148 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2149
nvme_shutdown_ctrl(struct nvme_ctrl * ctrl)2150 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2151 {
2152 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2153 u32 csts;
2154 int ret;
2155
2156 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2157 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2158
2159 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2160 if (ret)
2161 return ret;
2162
2163 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2164 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2165 break;
2166
2167 msleep(100);
2168 if (fatal_signal_pending(current))
2169 return -EINTR;
2170 if (time_after(jiffies, timeout)) {
2171 dev_err(ctrl->device,
2172 "Device shutdown incomplete; abort shutdown\n");
2173 return -ENODEV;
2174 }
2175 }
2176
2177 return ret;
2178 }
2179 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2180
nvme_set_queue_limits(struct nvme_ctrl * ctrl,struct request_queue * q)2181 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2182 struct request_queue *q)
2183 {
2184 bool vwc = false;
2185
2186 if (ctrl->max_hw_sectors) {
2187 u32 max_segments =
2188 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2189
2190 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2191 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2192 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2193 }
2194 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2195 is_power_of_2(ctrl->max_hw_sectors))
2196 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2197 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2198 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2199 vwc = true;
2200 blk_queue_write_cache(q, vwc, vwc);
2201 }
2202
nvme_configure_timestamp(struct nvme_ctrl * ctrl)2203 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2204 {
2205 __le64 ts;
2206 int ret;
2207
2208 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2209 return 0;
2210
2211 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2212 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2213 NULL);
2214 if (ret)
2215 dev_warn_once(ctrl->device,
2216 "could not set timestamp (%d)\n", ret);
2217 return ret;
2218 }
2219
nvme_configure_acre(struct nvme_ctrl * ctrl)2220 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2221 {
2222 struct nvme_feat_host_behavior *host;
2223 int ret;
2224
2225 /* Don't bother enabling the feature if retry delay is not reported */
2226 if (!ctrl->crdt[0])
2227 return 0;
2228
2229 host = kzalloc(sizeof(*host), GFP_KERNEL);
2230 if (!host)
2231 return 0;
2232
2233 host->acre = NVME_ENABLE_ACRE;
2234 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2235 host, sizeof(*host), NULL);
2236 kfree(host);
2237 return ret;
2238 }
2239
nvme_configure_apst(struct nvme_ctrl * ctrl)2240 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2241 {
2242 /*
2243 * APST (Autonomous Power State Transition) lets us program a
2244 * table of power state transitions that the controller will
2245 * perform automatically. We configure it with a simple
2246 * heuristic: we are willing to spend at most 2% of the time
2247 * transitioning between power states. Therefore, when running
2248 * in any given state, we will enter the next lower-power
2249 * non-operational state after waiting 50 * (enlat + exlat)
2250 * microseconds, as long as that state's exit latency is under
2251 * the requested maximum latency.
2252 *
2253 * We will not autonomously enter any non-operational state for
2254 * which the total latency exceeds ps_max_latency_us. Users
2255 * can set ps_max_latency_us to zero to turn off APST.
2256 */
2257
2258 unsigned apste;
2259 struct nvme_feat_auto_pst *table;
2260 u64 max_lat_us = 0;
2261 int max_ps = -1;
2262 int ret;
2263
2264 /*
2265 * If APST isn't supported or if we haven't been initialized yet,
2266 * then don't do anything.
2267 */
2268 if (!ctrl->apsta)
2269 return 0;
2270
2271 if (ctrl->npss > 31) {
2272 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2273 return 0;
2274 }
2275
2276 table = kzalloc(sizeof(*table), GFP_KERNEL);
2277 if (!table)
2278 return 0;
2279
2280 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2281 /* Turn off APST. */
2282 apste = 0;
2283 dev_dbg(ctrl->device, "APST disabled\n");
2284 } else {
2285 __le64 target = cpu_to_le64(0);
2286 int state;
2287
2288 /*
2289 * Walk through all states from lowest- to highest-power.
2290 * According to the spec, lower-numbered states use more
2291 * power. NPSS, despite the name, is the index of the
2292 * lowest-power state, not the number of states.
2293 */
2294 for (state = (int)ctrl->npss; state >= 0; state--) {
2295 u64 total_latency_us, exit_latency_us, transition_ms;
2296
2297 if (target)
2298 table->entries[state] = target;
2299
2300 /*
2301 * Don't allow transitions to the deepest state
2302 * if it's quirked off.
2303 */
2304 if (state == ctrl->npss &&
2305 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2306 continue;
2307
2308 /*
2309 * Is this state a useful non-operational state for
2310 * higher-power states to autonomously transition to?
2311 */
2312 if (!(ctrl->psd[state].flags &
2313 NVME_PS_FLAGS_NON_OP_STATE))
2314 continue;
2315
2316 exit_latency_us =
2317 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2318 if (exit_latency_us > ctrl->ps_max_latency_us)
2319 continue;
2320
2321 total_latency_us =
2322 exit_latency_us +
2323 le32_to_cpu(ctrl->psd[state].entry_lat);
2324
2325 /*
2326 * This state is good. Use it as the APST idle
2327 * target for higher power states.
2328 */
2329 transition_ms = total_latency_us + 19;
2330 do_div(transition_ms, 20);
2331 if (transition_ms > (1 << 24) - 1)
2332 transition_ms = (1 << 24) - 1;
2333
2334 target = cpu_to_le64((state << 3) |
2335 (transition_ms << 8));
2336
2337 if (max_ps == -1)
2338 max_ps = state;
2339
2340 if (total_latency_us > max_lat_us)
2341 max_lat_us = total_latency_us;
2342 }
2343
2344 apste = 1;
2345
2346 if (max_ps == -1) {
2347 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2348 } else {
2349 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2350 max_ps, max_lat_us, (int)sizeof(*table), table);
2351 }
2352 }
2353
2354 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2355 table, sizeof(*table), NULL);
2356 if (ret)
2357 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2358
2359 kfree(table);
2360 return ret;
2361 }
2362
nvme_set_latency_tolerance(struct device * dev,s32 val)2363 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2364 {
2365 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2366 u64 latency;
2367
2368 switch (val) {
2369 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2370 case PM_QOS_LATENCY_ANY:
2371 latency = U64_MAX;
2372 break;
2373
2374 default:
2375 latency = val;
2376 }
2377
2378 if (ctrl->ps_max_latency_us != latency) {
2379 ctrl->ps_max_latency_us = latency;
2380 nvme_configure_apst(ctrl);
2381 }
2382 }
2383
2384 struct nvme_core_quirk_entry {
2385 /*
2386 * NVMe model and firmware strings are padded with spaces. For
2387 * simplicity, strings in the quirk table are padded with NULLs
2388 * instead.
2389 */
2390 u16 vid;
2391 const char *mn;
2392 const char *fr;
2393 unsigned long quirks;
2394 };
2395
2396 static const struct nvme_core_quirk_entry core_quirks[] = {
2397 {
2398 /*
2399 * This Toshiba device seems to die using any APST states. See:
2400 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2401 */
2402 .vid = 0x1179,
2403 .mn = "THNSF5256GPUK TOSHIBA",
2404 .quirks = NVME_QUIRK_NO_APST,
2405 },
2406 {
2407 /*
2408 * This LiteON CL1-3D*-Q11 firmware version has a race
2409 * condition associated with actions related to suspend to idle
2410 * LiteON has resolved the problem in future firmware
2411 */
2412 .vid = 0x14a4,
2413 .fr = "22301111",
2414 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2415 }
2416 };
2417
2418 /* match is null-terminated but idstr is space-padded. */
string_matches(const char * idstr,const char * match,size_t len)2419 static bool string_matches(const char *idstr, const char *match, size_t len)
2420 {
2421 size_t matchlen;
2422
2423 if (!match)
2424 return true;
2425
2426 matchlen = strlen(match);
2427 WARN_ON_ONCE(matchlen > len);
2428
2429 if (memcmp(idstr, match, matchlen))
2430 return false;
2431
2432 for (; matchlen < len; matchlen++)
2433 if (idstr[matchlen] != ' ')
2434 return false;
2435
2436 return true;
2437 }
2438
quirk_matches(const struct nvme_id_ctrl * id,const struct nvme_core_quirk_entry * q)2439 static bool quirk_matches(const struct nvme_id_ctrl *id,
2440 const struct nvme_core_quirk_entry *q)
2441 {
2442 return q->vid == le16_to_cpu(id->vid) &&
2443 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2444 string_matches(id->fr, q->fr, sizeof(id->fr));
2445 }
2446
nvme_init_subnqn(struct nvme_subsystem * subsys,struct nvme_ctrl * ctrl,struct nvme_id_ctrl * id)2447 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2448 struct nvme_id_ctrl *id)
2449 {
2450 size_t nqnlen;
2451 int off;
2452
2453 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2454 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2455 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2456 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2457 return;
2458 }
2459
2460 if (ctrl->vs >= NVME_VS(1, 2, 1))
2461 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2462 }
2463
2464 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2465 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2466 "nqn.2014.08.org.nvmexpress:%04x%04x",
2467 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2468 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2469 off += sizeof(id->sn);
2470 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2471 off += sizeof(id->mn);
2472 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2473 }
2474
nvme_release_subsystem(struct device * dev)2475 static void nvme_release_subsystem(struct device *dev)
2476 {
2477 struct nvme_subsystem *subsys =
2478 container_of(dev, struct nvme_subsystem, dev);
2479
2480 if (subsys->instance >= 0)
2481 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2482 kfree(subsys);
2483 }
2484
nvme_destroy_subsystem(struct kref * ref)2485 static void nvme_destroy_subsystem(struct kref *ref)
2486 {
2487 struct nvme_subsystem *subsys =
2488 container_of(ref, struct nvme_subsystem, ref);
2489
2490 mutex_lock(&nvme_subsystems_lock);
2491 list_del(&subsys->entry);
2492 mutex_unlock(&nvme_subsystems_lock);
2493
2494 ida_destroy(&subsys->ns_ida);
2495 device_del(&subsys->dev);
2496 put_device(&subsys->dev);
2497 }
2498
nvme_put_subsystem(struct nvme_subsystem * subsys)2499 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2500 {
2501 kref_put(&subsys->ref, nvme_destroy_subsystem);
2502 }
2503
__nvme_find_get_subsystem(const char * subsysnqn)2504 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2505 {
2506 struct nvme_subsystem *subsys;
2507
2508 lockdep_assert_held(&nvme_subsystems_lock);
2509
2510 /*
2511 * Fail matches for discovery subsystems. This results
2512 * in each discovery controller bound to a unique subsystem.
2513 * This avoids issues with validating controller values
2514 * that can only be true when there is a single unique subsystem.
2515 * There may be multiple and completely independent entities
2516 * that provide discovery controllers.
2517 */
2518 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2519 return NULL;
2520
2521 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2522 if (strcmp(subsys->subnqn, subsysnqn))
2523 continue;
2524 if (!kref_get_unless_zero(&subsys->ref))
2525 continue;
2526 return subsys;
2527 }
2528
2529 return NULL;
2530 }
2531
2532 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2533 struct device_attribute subsys_attr_##_name = \
2534 __ATTR(_name, _mode, _show, NULL)
2535
nvme_subsys_show_nqn(struct device * dev,struct device_attribute * attr,char * buf)2536 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2537 struct device_attribute *attr,
2538 char *buf)
2539 {
2540 struct nvme_subsystem *subsys =
2541 container_of(dev, struct nvme_subsystem, dev);
2542
2543 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2544 }
2545 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2546
2547 #define nvme_subsys_show_str_function(field) \
2548 static ssize_t subsys_##field##_show(struct device *dev, \
2549 struct device_attribute *attr, char *buf) \
2550 { \
2551 struct nvme_subsystem *subsys = \
2552 container_of(dev, struct nvme_subsystem, dev); \
2553 return sprintf(buf, "%.*s\n", \
2554 (int)sizeof(subsys->field), subsys->field); \
2555 } \
2556 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2557
2558 nvme_subsys_show_str_function(model);
2559 nvme_subsys_show_str_function(serial);
2560 nvme_subsys_show_str_function(firmware_rev);
2561
2562 static struct attribute *nvme_subsys_attrs[] = {
2563 &subsys_attr_model.attr,
2564 &subsys_attr_serial.attr,
2565 &subsys_attr_firmware_rev.attr,
2566 &subsys_attr_subsysnqn.attr,
2567 #ifdef CONFIG_NVME_MULTIPATH
2568 &subsys_attr_iopolicy.attr,
2569 #endif
2570 NULL,
2571 };
2572
2573 static struct attribute_group nvme_subsys_attrs_group = {
2574 .attrs = nvme_subsys_attrs,
2575 };
2576
2577 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2578 &nvme_subsys_attrs_group,
2579 NULL,
2580 };
2581
nvme_validate_cntlid(struct nvme_subsystem * subsys,struct nvme_ctrl * ctrl,struct nvme_id_ctrl * id)2582 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2583 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2584 {
2585 struct nvme_ctrl *tmp;
2586
2587 lockdep_assert_held(&nvme_subsystems_lock);
2588
2589 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2590 if (tmp->state == NVME_CTRL_DELETING ||
2591 tmp->state == NVME_CTRL_DEAD)
2592 continue;
2593
2594 if (tmp->cntlid == ctrl->cntlid) {
2595 dev_err(ctrl->device,
2596 "Duplicate cntlid %u with %s, rejecting\n",
2597 ctrl->cntlid, dev_name(tmp->device));
2598 return false;
2599 }
2600
2601 if ((id->cmic & (1 << 1)) ||
2602 (ctrl->opts && ctrl->opts->discovery_nqn))
2603 continue;
2604
2605 dev_err(ctrl->device,
2606 "Subsystem does not support multiple controllers\n");
2607 return false;
2608 }
2609
2610 return true;
2611 }
2612
nvme_init_subsystem(struct nvme_ctrl * ctrl,struct nvme_id_ctrl * id)2613 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2614 {
2615 struct nvme_subsystem *subsys, *found;
2616 int ret;
2617
2618 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2619 if (!subsys)
2620 return -ENOMEM;
2621
2622 subsys->instance = -1;
2623 mutex_init(&subsys->lock);
2624 kref_init(&subsys->ref);
2625 INIT_LIST_HEAD(&subsys->ctrls);
2626 INIT_LIST_HEAD(&subsys->nsheads);
2627 nvme_init_subnqn(subsys, ctrl, id);
2628 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2629 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2630 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2631 subsys->vendor_id = le16_to_cpu(id->vid);
2632 subsys->cmic = id->cmic;
2633 subsys->awupf = le16_to_cpu(id->awupf);
2634 #ifdef CONFIG_NVME_MULTIPATH
2635 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2636 #endif
2637
2638 subsys->dev.class = nvme_subsys_class;
2639 subsys->dev.release = nvme_release_subsystem;
2640 subsys->dev.groups = nvme_subsys_attrs_groups;
2641 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2642 device_initialize(&subsys->dev);
2643
2644 mutex_lock(&nvme_subsystems_lock);
2645 found = __nvme_find_get_subsystem(subsys->subnqn);
2646 if (found) {
2647 put_device(&subsys->dev);
2648 subsys = found;
2649
2650 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2651 ret = -EINVAL;
2652 goto out_put_subsystem;
2653 }
2654 } else {
2655 ret = device_add(&subsys->dev);
2656 if (ret) {
2657 dev_err(ctrl->device,
2658 "failed to register subsystem device.\n");
2659 put_device(&subsys->dev);
2660 goto out_unlock;
2661 }
2662 ida_init(&subsys->ns_ida);
2663 list_add_tail(&subsys->entry, &nvme_subsystems);
2664 }
2665
2666 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2667 dev_name(ctrl->device));
2668 if (ret) {
2669 dev_err(ctrl->device,
2670 "failed to create sysfs link from subsystem.\n");
2671 goto out_put_subsystem;
2672 }
2673
2674 if (!found)
2675 subsys->instance = ctrl->instance;
2676 ctrl->subsys = subsys;
2677 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2678 mutex_unlock(&nvme_subsystems_lock);
2679 return 0;
2680
2681 out_put_subsystem:
2682 nvme_put_subsystem(subsys);
2683 out_unlock:
2684 mutex_unlock(&nvme_subsystems_lock);
2685 return ret;
2686 }
2687
nvme_get_log(struct nvme_ctrl * ctrl,u32 nsid,u8 log_page,u8 lsp,void * log,size_t size,u64 offset)2688 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2689 void *log, size_t size, u64 offset)
2690 {
2691 struct nvme_command c = { };
2692 unsigned long dwlen = size / 4 - 1;
2693
2694 c.get_log_page.opcode = nvme_admin_get_log_page;
2695 c.get_log_page.nsid = cpu_to_le32(nsid);
2696 c.get_log_page.lid = log_page;
2697 c.get_log_page.lsp = lsp;
2698 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2699 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2700 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2701 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2702
2703 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2704 }
2705
nvme_get_effects_log(struct nvme_ctrl * ctrl)2706 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2707 {
2708 int ret;
2709
2710 if (!ctrl->effects)
2711 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2712
2713 if (!ctrl->effects)
2714 return 0;
2715
2716 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2717 ctrl->effects, sizeof(*ctrl->effects), 0);
2718 if (ret) {
2719 kfree(ctrl->effects);
2720 ctrl->effects = NULL;
2721 }
2722 return ret;
2723 }
2724
2725 /*
2726 * Initialize the cached copies of the Identify data and various controller
2727 * register in our nvme_ctrl structure. This should be called as soon as
2728 * the admin queue is fully up and running.
2729 */
nvme_init_identify(struct nvme_ctrl * ctrl)2730 int nvme_init_identify(struct nvme_ctrl *ctrl)
2731 {
2732 struct nvme_id_ctrl *id;
2733 int ret, page_shift;
2734 u32 max_hw_sectors;
2735 bool prev_apst_enabled;
2736
2737 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2738 if (ret) {
2739 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2740 return ret;
2741 }
2742 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2743 ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2744
2745 if (ctrl->vs >= NVME_VS(1, 1, 0))
2746 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2747
2748 ret = nvme_identify_ctrl(ctrl, &id);
2749 if (ret) {
2750 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2751 return -EIO;
2752 }
2753
2754 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2755 ret = nvme_get_effects_log(ctrl);
2756 if (ret < 0)
2757 goto out_free;
2758 }
2759
2760 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2761 ctrl->cntlid = le16_to_cpu(id->cntlid);
2762
2763 if (!ctrl->identified) {
2764 int i;
2765
2766 ret = nvme_init_subsystem(ctrl, id);
2767 if (ret)
2768 goto out_free;
2769
2770 /*
2771 * Check for quirks. Quirk can depend on firmware version,
2772 * so, in principle, the set of quirks present can change
2773 * across a reset. As a possible future enhancement, we
2774 * could re-scan for quirks every time we reinitialize
2775 * the device, but we'd have to make sure that the driver
2776 * behaves intelligently if the quirks change.
2777 */
2778 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2779 if (quirk_matches(id, &core_quirks[i]))
2780 ctrl->quirks |= core_quirks[i].quirks;
2781 }
2782 }
2783
2784 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2785 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2786 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2787 }
2788
2789 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2790 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2791 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2792
2793 ctrl->oacs = le16_to_cpu(id->oacs);
2794 ctrl->oncs = le16_to_cpu(id->oncs);
2795 ctrl->mtfa = le16_to_cpu(id->mtfa);
2796 ctrl->oaes = le32_to_cpu(id->oaes);
2797 atomic_set(&ctrl->abort_limit, id->acl + 1);
2798 ctrl->vwc = id->vwc;
2799 if (id->mdts)
2800 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2801 else
2802 max_hw_sectors = UINT_MAX;
2803 ctrl->max_hw_sectors =
2804 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2805
2806 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2807 ctrl->sgls = le32_to_cpu(id->sgls);
2808 ctrl->kas = le16_to_cpu(id->kas);
2809 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2810 ctrl->ctratt = le32_to_cpu(id->ctratt);
2811
2812 if (id->rtd3e) {
2813 /* us -> s */
2814 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2815
2816 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2817 shutdown_timeout, 60);
2818
2819 if (ctrl->shutdown_timeout != shutdown_timeout)
2820 dev_info(ctrl->device,
2821 "Shutdown timeout set to %u seconds\n",
2822 ctrl->shutdown_timeout);
2823 } else
2824 ctrl->shutdown_timeout = shutdown_timeout;
2825
2826 ctrl->npss = id->npss;
2827 ctrl->apsta = id->apsta;
2828 prev_apst_enabled = ctrl->apst_enabled;
2829 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2830 if (force_apst && id->apsta) {
2831 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2832 ctrl->apst_enabled = true;
2833 } else {
2834 ctrl->apst_enabled = false;
2835 }
2836 } else {
2837 ctrl->apst_enabled = id->apsta;
2838 }
2839 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2840
2841 if (ctrl->ops->flags & NVME_F_FABRICS) {
2842 ctrl->icdoff = le16_to_cpu(id->icdoff);
2843 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2844 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2845 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2846
2847 /*
2848 * In fabrics we need to verify the cntlid matches the
2849 * admin connect
2850 */
2851 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2852 ret = -EINVAL;
2853 goto out_free;
2854 }
2855
2856 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2857 dev_err(ctrl->device,
2858 "keep-alive support is mandatory for fabrics\n");
2859 ret = -EINVAL;
2860 goto out_free;
2861 }
2862 } else {
2863 ctrl->hmpre = le32_to_cpu(id->hmpre);
2864 ctrl->hmmin = le32_to_cpu(id->hmmin);
2865 ctrl->hmminds = le32_to_cpu(id->hmminds);
2866 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2867 }
2868
2869 ret = nvme_mpath_init(ctrl, id);
2870 kfree(id);
2871
2872 if (ret < 0)
2873 return ret;
2874
2875 if (ctrl->apst_enabled && !prev_apst_enabled)
2876 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2877 else if (!ctrl->apst_enabled && prev_apst_enabled)
2878 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2879
2880 ret = nvme_configure_apst(ctrl);
2881 if (ret < 0)
2882 return ret;
2883
2884 ret = nvme_configure_timestamp(ctrl);
2885 if (ret < 0)
2886 return ret;
2887
2888 ret = nvme_configure_directives(ctrl);
2889 if (ret < 0)
2890 return ret;
2891
2892 ret = nvme_configure_acre(ctrl);
2893 if (ret < 0)
2894 return ret;
2895
2896 ctrl->identified = true;
2897
2898 return 0;
2899
2900 out_free:
2901 kfree(id);
2902 return ret;
2903 }
2904 EXPORT_SYMBOL_GPL(nvme_init_identify);
2905
nvme_dev_open(struct inode * inode,struct file * file)2906 static int nvme_dev_open(struct inode *inode, struct file *file)
2907 {
2908 struct nvme_ctrl *ctrl =
2909 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2910
2911 switch (ctrl->state) {
2912 case NVME_CTRL_LIVE:
2913 break;
2914 default:
2915 return -EWOULDBLOCK;
2916 }
2917
2918 file->private_data = ctrl;
2919 return 0;
2920 }
2921
nvme_dev_user_cmd(struct nvme_ctrl * ctrl,void __user * argp)2922 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2923 {
2924 struct nvme_ns *ns;
2925 int ret;
2926
2927 down_read(&ctrl->namespaces_rwsem);
2928 if (list_empty(&ctrl->namespaces)) {
2929 ret = -ENOTTY;
2930 goto out_unlock;
2931 }
2932
2933 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2934 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2935 dev_warn(ctrl->device,
2936 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2937 ret = -EINVAL;
2938 goto out_unlock;
2939 }
2940
2941 dev_warn(ctrl->device,
2942 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2943 kref_get(&ns->kref);
2944 up_read(&ctrl->namespaces_rwsem);
2945
2946 ret = nvme_user_cmd(ctrl, ns, argp);
2947 nvme_put_ns(ns);
2948 return ret;
2949
2950 out_unlock:
2951 up_read(&ctrl->namespaces_rwsem);
2952 return ret;
2953 }
2954
nvme_dev_ioctl(struct file * file,unsigned int cmd,unsigned long arg)2955 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2956 unsigned long arg)
2957 {
2958 struct nvme_ctrl *ctrl = file->private_data;
2959 void __user *argp = (void __user *)arg;
2960
2961 switch (cmd) {
2962 case NVME_IOCTL_ADMIN_CMD:
2963 return nvme_user_cmd(ctrl, NULL, argp);
2964 case NVME_IOCTL_ADMIN64_CMD:
2965 return nvme_user_cmd64(ctrl, NULL, argp);
2966 case NVME_IOCTL_IO_CMD:
2967 return nvme_dev_user_cmd(ctrl, argp);
2968 case NVME_IOCTL_RESET:
2969 dev_warn(ctrl->device, "resetting controller\n");
2970 return nvme_reset_ctrl_sync(ctrl);
2971 case NVME_IOCTL_SUBSYS_RESET:
2972 return nvme_reset_subsystem(ctrl);
2973 case NVME_IOCTL_RESCAN:
2974 nvme_queue_scan(ctrl);
2975 return 0;
2976 default:
2977 return -ENOTTY;
2978 }
2979 }
2980
2981 static const struct file_operations nvme_dev_fops = {
2982 .owner = THIS_MODULE,
2983 .open = nvme_dev_open,
2984 .unlocked_ioctl = nvme_dev_ioctl,
2985 .compat_ioctl = nvme_dev_ioctl,
2986 };
2987
nvme_sysfs_reset(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)2988 static ssize_t nvme_sysfs_reset(struct device *dev,
2989 struct device_attribute *attr, const char *buf,
2990 size_t count)
2991 {
2992 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2993 int ret;
2994
2995 ret = nvme_reset_ctrl_sync(ctrl);
2996 if (ret < 0)
2997 return ret;
2998 return count;
2999 }
3000 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3001
nvme_sysfs_rescan(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)3002 static ssize_t nvme_sysfs_rescan(struct device *dev,
3003 struct device_attribute *attr, const char *buf,
3004 size_t count)
3005 {
3006 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3007
3008 nvme_queue_scan(ctrl);
3009 return count;
3010 }
3011 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3012
dev_to_ns_head(struct device * dev)3013 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3014 {
3015 struct gendisk *disk = dev_to_disk(dev);
3016
3017 if (disk->fops == &nvme_fops)
3018 return nvme_get_ns_from_dev(dev)->head;
3019 else
3020 return disk->private_data;
3021 }
3022
wwid_show(struct device * dev,struct device_attribute * attr,char * buf)3023 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3024 char *buf)
3025 {
3026 struct nvme_ns_head *head = dev_to_ns_head(dev);
3027 struct nvme_ns_ids *ids = &head->ids;
3028 struct nvme_subsystem *subsys = head->subsys;
3029 int serial_len = sizeof(subsys->serial);
3030 int model_len = sizeof(subsys->model);
3031
3032 if (!uuid_is_null(&ids->uuid))
3033 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3034
3035 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3036 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3037
3038 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3039 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3040
3041 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3042 subsys->serial[serial_len - 1] == '\0'))
3043 serial_len--;
3044 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3045 subsys->model[model_len - 1] == '\0'))
3046 model_len--;
3047
3048 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3049 serial_len, subsys->serial, model_len, subsys->model,
3050 head->ns_id);
3051 }
3052 static DEVICE_ATTR_RO(wwid);
3053
nguid_show(struct device * dev,struct device_attribute * attr,char * buf)3054 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3055 char *buf)
3056 {
3057 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3058 }
3059 static DEVICE_ATTR_RO(nguid);
3060
uuid_show(struct device * dev,struct device_attribute * attr,char * buf)3061 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3062 char *buf)
3063 {
3064 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3065
3066 /* For backward compatibility expose the NGUID to userspace if
3067 * we have no UUID set
3068 */
3069 if (uuid_is_null(&ids->uuid)) {
3070 printk_ratelimited(KERN_WARNING
3071 "No UUID available providing old NGUID\n");
3072 return sprintf(buf, "%pU\n", ids->nguid);
3073 }
3074 return sprintf(buf, "%pU\n", &ids->uuid);
3075 }
3076 static DEVICE_ATTR_RO(uuid);
3077
eui_show(struct device * dev,struct device_attribute * attr,char * buf)3078 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3079 char *buf)
3080 {
3081 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3082 }
3083 static DEVICE_ATTR_RO(eui);
3084
nsid_show(struct device * dev,struct device_attribute * attr,char * buf)3085 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3086 char *buf)
3087 {
3088 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3089 }
3090 static DEVICE_ATTR_RO(nsid);
3091
3092 static struct attribute *nvme_ns_id_attrs[] = {
3093 &dev_attr_wwid.attr,
3094 &dev_attr_uuid.attr,
3095 &dev_attr_nguid.attr,
3096 &dev_attr_eui.attr,
3097 &dev_attr_nsid.attr,
3098 #ifdef CONFIG_NVME_MULTIPATH
3099 &dev_attr_ana_grpid.attr,
3100 &dev_attr_ana_state.attr,
3101 #endif
3102 NULL,
3103 };
3104
nvme_ns_id_attrs_are_visible(struct kobject * kobj,struct attribute * a,int n)3105 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3106 struct attribute *a, int n)
3107 {
3108 struct device *dev = container_of(kobj, struct device, kobj);
3109 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3110
3111 if (a == &dev_attr_uuid.attr) {
3112 if (uuid_is_null(&ids->uuid) &&
3113 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3114 return 0;
3115 }
3116 if (a == &dev_attr_nguid.attr) {
3117 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3118 return 0;
3119 }
3120 if (a == &dev_attr_eui.attr) {
3121 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3122 return 0;
3123 }
3124 #ifdef CONFIG_NVME_MULTIPATH
3125 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3126 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3127 return 0;
3128 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3129 return 0;
3130 }
3131 #endif
3132 return a->mode;
3133 }
3134
3135 static const struct attribute_group nvme_ns_id_attr_group = {
3136 .attrs = nvme_ns_id_attrs,
3137 .is_visible = nvme_ns_id_attrs_are_visible,
3138 };
3139
3140 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3141 &nvme_ns_id_attr_group,
3142 #ifdef CONFIG_NVM
3143 &nvme_nvm_attr_group,
3144 #endif
3145 NULL,
3146 };
3147
3148 #define nvme_show_str_function(field) \
3149 static ssize_t field##_show(struct device *dev, \
3150 struct device_attribute *attr, char *buf) \
3151 { \
3152 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3153 return sprintf(buf, "%.*s\n", \
3154 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3155 } \
3156 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3157
3158 nvme_show_str_function(model);
3159 nvme_show_str_function(serial);
3160 nvme_show_str_function(firmware_rev);
3161
3162 #define nvme_show_int_function(field) \
3163 static ssize_t field##_show(struct device *dev, \
3164 struct device_attribute *attr, char *buf) \
3165 { \
3166 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3167 return sprintf(buf, "%d\n", ctrl->field); \
3168 } \
3169 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3170
3171 nvme_show_int_function(cntlid);
3172 nvme_show_int_function(numa_node);
3173 nvme_show_int_function(queue_count);
3174 nvme_show_int_function(sqsize);
3175
nvme_sysfs_delete(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)3176 static ssize_t nvme_sysfs_delete(struct device *dev,
3177 struct device_attribute *attr, const char *buf,
3178 size_t count)
3179 {
3180 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3181
3182 if (device_remove_file_self(dev, attr))
3183 nvme_delete_ctrl_sync(ctrl);
3184 return count;
3185 }
3186 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3187
nvme_sysfs_show_transport(struct device * dev,struct device_attribute * attr,char * buf)3188 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3189 struct device_attribute *attr,
3190 char *buf)
3191 {
3192 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3193
3194 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3195 }
3196 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3197
nvme_sysfs_show_state(struct device * dev,struct device_attribute * attr,char * buf)3198 static ssize_t nvme_sysfs_show_state(struct device *dev,
3199 struct device_attribute *attr,
3200 char *buf)
3201 {
3202 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3203 static const char *const state_name[] = {
3204 [NVME_CTRL_NEW] = "new",
3205 [NVME_CTRL_LIVE] = "live",
3206 [NVME_CTRL_RESETTING] = "resetting",
3207 [NVME_CTRL_CONNECTING] = "connecting",
3208 [NVME_CTRL_DELETING] = "deleting",
3209 [NVME_CTRL_DEAD] = "dead",
3210 };
3211
3212 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3213 state_name[ctrl->state])
3214 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3215
3216 return sprintf(buf, "unknown state\n");
3217 }
3218
3219 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3220
nvme_sysfs_show_subsysnqn(struct device * dev,struct device_attribute * attr,char * buf)3221 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3222 struct device_attribute *attr,
3223 char *buf)
3224 {
3225 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3226
3227 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3228 }
3229 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3230
nvme_sysfs_show_address(struct device * dev,struct device_attribute * attr,char * buf)3231 static ssize_t nvme_sysfs_show_address(struct device *dev,
3232 struct device_attribute *attr,
3233 char *buf)
3234 {
3235 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3236
3237 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3238 }
3239 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3240
3241 static struct attribute *nvme_dev_attrs[] = {
3242 &dev_attr_reset_controller.attr,
3243 &dev_attr_rescan_controller.attr,
3244 &dev_attr_model.attr,
3245 &dev_attr_serial.attr,
3246 &dev_attr_firmware_rev.attr,
3247 &dev_attr_cntlid.attr,
3248 &dev_attr_delete_controller.attr,
3249 &dev_attr_transport.attr,
3250 &dev_attr_subsysnqn.attr,
3251 &dev_attr_address.attr,
3252 &dev_attr_state.attr,
3253 &dev_attr_numa_node.attr,
3254 &dev_attr_queue_count.attr,
3255 &dev_attr_sqsize.attr,
3256 NULL
3257 };
3258
nvme_dev_attrs_are_visible(struct kobject * kobj,struct attribute * a,int n)3259 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3260 struct attribute *a, int n)
3261 {
3262 struct device *dev = container_of(kobj, struct device, kobj);
3263 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3264
3265 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3266 return 0;
3267 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3268 return 0;
3269
3270 return a->mode;
3271 }
3272
3273 static struct attribute_group nvme_dev_attrs_group = {
3274 .attrs = nvme_dev_attrs,
3275 .is_visible = nvme_dev_attrs_are_visible,
3276 };
3277
3278 static const struct attribute_group *nvme_dev_attr_groups[] = {
3279 &nvme_dev_attrs_group,
3280 NULL,
3281 };
3282
__nvme_find_ns_head(struct nvme_subsystem * subsys,unsigned nsid)3283 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3284 unsigned nsid)
3285 {
3286 struct nvme_ns_head *h;
3287
3288 lockdep_assert_held(&subsys->lock);
3289
3290 list_for_each_entry(h, &subsys->nsheads, entry) {
3291 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3292 return h;
3293 }
3294
3295 return NULL;
3296 }
3297
__nvme_check_ids(struct nvme_subsystem * subsys,struct nvme_ns_head * new)3298 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3299 struct nvme_ns_head *new)
3300 {
3301 struct nvme_ns_head *h;
3302
3303 lockdep_assert_held(&subsys->lock);
3304
3305 list_for_each_entry(h, &subsys->nsheads, entry) {
3306 if (nvme_ns_ids_valid(&new->ids) &&
3307 !list_empty(&h->list) &&
3308 nvme_ns_ids_equal(&new->ids, &h->ids))
3309 return -EINVAL;
3310 }
3311
3312 return 0;
3313 }
3314
nvme_alloc_ns_head(struct nvme_ctrl * ctrl,unsigned nsid,struct nvme_id_ns * id)3315 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3316 unsigned nsid, struct nvme_id_ns *id)
3317 {
3318 struct nvme_ns_head *head;
3319 size_t size = sizeof(*head);
3320 int ret = -ENOMEM;
3321
3322 #ifdef CONFIG_NVME_MULTIPATH
3323 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3324 #endif
3325
3326 head = kzalloc(size, GFP_KERNEL);
3327 if (!head)
3328 goto out;
3329 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3330 if (ret < 0)
3331 goto out_free_head;
3332 head->instance = ret;
3333 INIT_LIST_HEAD(&head->list);
3334 ret = init_srcu_struct(&head->srcu);
3335 if (ret)
3336 goto out_ida_remove;
3337 head->subsys = ctrl->subsys;
3338 head->ns_id = nsid;
3339 kref_init(&head->ref);
3340
3341 ret = nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3342 if (ret)
3343 goto out_cleanup_srcu;
3344
3345 ret = __nvme_check_ids(ctrl->subsys, head);
3346 if (ret) {
3347 dev_err(ctrl->device,
3348 "duplicate IDs for nsid %d\n", nsid);
3349 goto out_cleanup_srcu;
3350 }
3351
3352 ret = nvme_mpath_alloc_disk(ctrl, head);
3353 if (ret)
3354 goto out_cleanup_srcu;
3355
3356 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3357
3358 kref_get(&ctrl->subsys->ref);
3359
3360 return head;
3361 out_cleanup_srcu:
3362 cleanup_srcu_struct(&head->srcu);
3363 out_ida_remove:
3364 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3365 out_free_head:
3366 kfree(head);
3367 out:
3368 if (ret > 0)
3369 ret = blk_status_to_errno(nvme_error_status(ret));
3370 return ERR_PTR(ret);
3371 }
3372
nvme_init_ns_head(struct nvme_ns * ns,unsigned nsid,struct nvme_id_ns * id)3373 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3374 struct nvme_id_ns *id)
3375 {
3376 struct nvme_ctrl *ctrl = ns->ctrl;
3377 bool is_shared = id->nmic & (1 << 0);
3378 struct nvme_ns_head *head = NULL;
3379 int ret = 0;
3380
3381 mutex_lock(&ctrl->subsys->lock);
3382 if (is_shared)
3383 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3384 if (!head) {
3385 head = nvme_alloc_ns_head(ctrl, nsid, id);
3386 if (IS_ERR(head)) {
3387 ret = PTR_ERR(head);
3388 goto out_unlock;
3389 }
3390 } else {
3391 struct nvme_ns_ids ids;
3392
3393 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3394 if (ret)
3395 goto out_unlock;
3396
3397 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3398 dev_err(ctrl->device,
3399 "IDs don't match for shared namespace %d\n",
3400 nsid);
3401 ret = -EINVAL;
3402 goto out_unlock;
3403 }
3404 }
3405
3406 list_add_tail(&ns->siblings, &head->list);
3407 ns->head = head;
3408
3409 out_unlock:
3410 mutex_unlock(&ctrl->subsys->lock);
3411 if (ret > 0)
3412 ret = blk_status_to_errno(nvme_error_status(ret));
3413 return ret;
3414 }
3415
ns_cmp(void * priv,struct list_head * a,struct list_head * b)3416 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3417 {
3418 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3419 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3420
3421 return nsa->head->ns_id - nsb->head->ns_id;
3422 }
3423
nvme_find_get_ns(struct nvme_ctrl * ctrl,unsigned nsid)3424 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3425 {
3426 struct nvme_ns *ns, *ret = NULL;
3427
3428 down_read(&ctrl->namespaces_rwsem);
3429 list_for_each_entry(ns, &ctrl->namespaces, list) {
3430 if (ns->head->ns_id == nsid) {
3431 if (!kref_get_unless_zero(&ns->kref))
3432 continue;
3433 ret = ns;
3434 break;
3435 }
3436 if (ns->head->ns_id > nsid)
3437 break;
3438 }
3439 up_read(&ctrl->namespaces_rwsem);
3440 return ret;
3441 }
3442
nvme_setup_streams_ns(struct nvme_ctrl * ctrl,struct nvme_ns * ns)3443 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3444 {
3445 struct streams_directive_params s;
3446 int ret;
3447
3448 if (!ctrl->nr_streams)
3449 return 0;
3450
3451 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3452 if (ret)
3453 return ret;
3454
3455 ns->sws = le32_to_cpu(s.sws);
3456 ns->sgs = le16_to_cpu(s.sgs);
3457
3458 if (ns->sws) {
3459 unsigned int bs = 1 << ns->lba_shift;
3460
3461 blk_queue_io_min(ns->queue, bs * ns->sws);
3462 if (ns->sgs)
3463 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3464 }
3465
3466 return 0;
3467 }
3468
nvme_alloc_ns(struct nvme_ctrl * ctrl,unsigned nsid)3469 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3470 {
3471 struct nvme_ns *ns;
3472 struct gendisk *disk;
3473 struct nvme_id_ns *id;
3474 char disk_name[DISK_NAME_LEN];
3475 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3476
3477 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3478 if (!ns)
3479 return -ENOMEM;
3480
3481 ns->queue = blk_mq_init_queue(ctrl->tagset);
3482 if (IS_ERR(ns->queue)) {
3483 ret = PTR_ERR(ns->queue);
3484 goto out_free_ns;
3485 }
3486
3487 if (ctrl->opts && ctrl->opts->data_digest)
3488 ns->queue->backing_dev_info->capabilities
3489 |= BDI_CAP_STABLE_WRITES;
3490
3491 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3492 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3493 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3494
3495 ns->queue->queuedata = ns;
3496 ns->ctrl = ctrl;
3497
3498 kref_init(&ns->kref);
3499 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3500
3501 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3502 nvme_set_queue_limits(ctrl, ns->queue);
3503
3504 ret = nvme_identify_ns(ctrl, nsid, &id);
3505 if (ret)
3506 goto out_free_queue;
3507
3508 if (id->ncap == 0) {
3509 ret = -EINVAL;
3510 goto out_free_id;
3511 }
3512
3513 ret = nvme_init_ns_head(ns, nsid, id);
3514 if (ret)
3515 goto out_free_id;
3516 nvme_setup_streams_ns(ctrl, ns);
3517 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3518
3519 disk = alloc_disk_node(0, node);
3520 if (!disk) {
3521 ret = -ENOMEM;
3522 goto out_unlink_ns;
3523 }
3524
3525 disk->fops = &nvme_fops;
3526 disk->private_data = ns;
3527 disk->queue = ns->queue;
3528 disk->flags = flags;
3529 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3530 ns->disk = disk;
3531
3532 __nvme_revalidate_disk(disk, id);
3533
3534 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3535 ret = nvme_nvm_register(ns, disk_name, node);
3536 if (ret) {
3537 dev_warn(ctrl->device, "LightNVM init failure\n");
3538 goto out_put_disk;
3539 }
3540 }
3541
3542 down_write(&ctrl->namespaces_rwsem);
3543 list_add_tail(&ns->list, &ctrl->namespaces);
3544 up_write(&ctrl->namespaces_rwsem);
3545
3546 nvme_get_ctrl(ctrl);
3547
3548 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3549
3550 nvme_mpath_add_disk(ns, id);
3551 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3552 kfree(id);
3553
3554 return 0;
3555 out_put_disk:
3556 put_disk(ns->disk);
3557 out_unlink_ns:
3558 mutex_lock(&ctrl->subsys->lock);
3559 list_del_rcu(&ns->siblings);
3560 mutex_unlock(&ctrl->subsys->lock);
3561 nvme_put_ns_head(ns->head);
3562 out_free_id:
3563 kfree(id);
3564 out_free_queue:
3565 blk_cleanup_queue(ns->queue);
3566 out_free_ns:
3567 kfree(ns);
3568 if (ret > 0)
3569 ret = blk_status_to_errno(nvme_error_status(ret));
3570 return ret;
3571 }
3572
nvme_ns_remove(struct nvme_ns * ns)3573 static void nvme_ns_remove(struct nvme_ns *ns)
3574 {
3575 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3576 return;
3577
3578 nvme_fault_inject_fini(&ns->fault_inject);
3579
3580 mutex_lock(&ns->ctrl->subsys->lock);
3581 list_del_rcu(&ns->siblings);
3582 mutex_unlock(&ns->ctrl->subsys->lock);
3583 synchronize_rcu(); /* guarantee not available in head->list */
3584 nvme_mpath_clear_current_path(ns);
3585 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3586
3587 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3588 del_gendisk(ns->disk);
3589 blk_cleanup_queue(ns->queue);
3590 if (blk_get_integrity(ns->disk))
3591 blk_integrity_unregister(ns->disk);
3592 }
3593
3594 down_write(&ns->ctrl->namespaces_rwsem);
3595 list_del_init(&ns->list);
3596 up_write(&ns->ctrl->namespaces_rwsem);
3597
3598 nvme_mpath_check_last_path(ns);
3599 nvme_put_ns(ns);
3600 }
3601
nvme_validate_ns(struct nvme_ctrl * ctrl,unsigned nsid)3602 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3603 {
3604 struct nvme_ns *ns;
3605
3606 ns = nvme_find_get_ns(ctrl, nsid);
3607 if (ns) {
3608 if (ns->disk && revalidate_disk(ns->disk))
3609 nvme_ns_remove(ns);
3610 nvme_put_ns(ns);
3611 } else
3612 nvme_alloc_ns(ctrl, nsid);
3613 }
3614
nvme_remove_invalid_namespaces(struct nvme_ctrl * ctrl,unsigned nsid)3615 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3616 unsigned nsid)
3617 {
3618 struct nvme_ns *ns, *next;
3619 LIST_HEAD(rm_list);
3620
3621 down_write(&ctrl->namespaces_rwsem);
3622 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3623 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3624 list_move_tail(&ns->list, &rm_list);
3625 }
3626 up_write(&ctrl->namespaces_rwsem);
3627
3628 list_for_each_entry_safe(ns, next, &rm_list, list)
3629 nvme_ns_remove(ns);
3630
3631 }
3632
nvme_scan_ns_list(struct nvme_ctrl * ctrl,unsigned nn)3633 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3634 {
3635 struct nvme_ns *ns;
3636 __le32 *ns_list;
3637 unsigned i, j, nsid, prev = 0;
3638 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3639 int ret = 0;
3640
3641 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3642 if (!ns_list)
3643 return -ENOMEM;
3644
3645 for (i = 0; i < num_lists; i++) {
3646 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3647 if (ret)
3648 goto free;
3649
3650 for (j = 0; j < min(nn, 1024U); j++) {
3651 nsid = le32_to_cpu(ns_list[j]);
3652 if (!nsid)
3653 goto out;
3654
3655 nvme_validate_ns(ctrl, nsid);
3656
3657 while (++prev < nsid) {
3658 ns = nvme_find_get_ns(ctrl, prev);
3659 if (ns) {
3660 nvme_ns_remove(ns);
3661 nvme_put_ns(ns);
3662 }
3663 }
3664 }
3665 nn -= j;
3666 }
3667 out:
3668 nvme_remove_invalid_namespaces(ctrl, prev);
3669 free:
3670 kfree(ns_list);
3671 return ret;
3672 }
3673
nvme_scan_ns_sequential(struct nvme_ctrl * ctrl,unsigned nn)3674 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3675 {
3676 unsigned i;
3677
3678 for (i = 1; i <= nn; i++)
3679 nvme_validate_ns(ctrl, i);
3680
3681 nvme_remove_invalid_namespaces(ctrl, nn);
3682 }
3683
nvme_clear_changed_ns_log(struct nvme_ctrl * ctrl)3684 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3685 {
3686 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3687 __le32 *log;
3688 int error;
3689
3690 log = kzalloc(log_size, GFP_KERNEL);
3691 if (!log)
3692 return;
3693
3694 /*
3695 * We need to read the log to clear the AEN, but we don't want to rely
3696 * on it for the changed namespace information as userspace could have
3697 * raced with us in reading the log page, which could cause us to miss
3698 * updates.
3699 */
3700 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3701 log_size, 0);
3702 if (error)
3703 dev_warn(ctrl->device,
3704 "reading changed ns log failed: %d\n", error);
3705
3706 kfree(log);
3707 }
3708
nvme_scan_work(struct work_struct * work)3709 static void nvme_scan_work(struct work_struct *work)
3710 {
3711 struct nvme_ctrl *ctrl =
3712 container_of(work, struct nvme_ctrl, scan_work);
3713 struct nvme_id_ctrl *id;
3714 unsigned nn;
3715
3716 /* No tagset on a live ctrl means IO queues could not created */
3717 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3718 return;
3719
3720 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3721 dev_info(ctrl->device, "rescanning namespaces.\n");
3722 nvme_clear_changed_ns_log(ctrl);
3723 }
3724
3725 if (nvme_identify_ctrl(ctrl, &id))
3726 return;
3727
3728 mutex_lock(&ctrl->scan_lock);
3729 nn = le32_to_cpu(id->nn);
3730 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3731 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3732 if (!nvme_scan_ns_list(ctrl, nn))
3733 goto out_free_id;
3734 }
3735 nvme_scan_ns_sequential(ctrl, nn);
3736 out_free_id:
3737 mutex_unlock(&ctrl->scan_lock);
3738 kfree(id);
3739 down_write(&ctrl->namespaces_rwsem);
3740 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3741 up_write(&ctrl->namespaces_rwsem);
3742 }
3743
3744 /*
3745 * This function iterates the namespace list unlocked to allow recovery from
3746 * controller failure. It is up to the caller to ensure the namespace list is
3747 * not modified by scan work while this function is executing.
3748 */
nvme_remove_namespaces(struct nvme_ctrl * ctrl)3749 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3750 {
3751 struct nvme_ns *ns, *next;
3752 LIST_HEAD(ns_list);
3753
3754 /*
3755 * make sure to requeue I/O to all namespaces as these
3756 * might result from the scan itself and must complete
3757 * for the scan_work to make progress
3758 */
3759 nvme_mpath_clear_ctrl_paths(ctrl);
3760
3761 /* prevent racing with ns scanning */
3762 flush_work(&ctrl->scan_work);
3763
3764 /*
3765 * The dead states indicates the controller was not gracefully
3766 * disconnected. In that case, we won't be able to flush any data while
3767 * removing the namespaces' disks; fail all the queues now to avoid
3768 * potentially having to clean up the failed sync later.
3769 */
3770 if (ctrl->state == NVME_CTRL_DEAD)
3771 nvme_kill_queues(ctrl);
3772
3773 down_write(&ctrl->namespaces_rwsem);
3774 list_splice_init(&ctrl->namespaces, &ns_list);
3775 up_write(&ctrl->namespaces_rwsem);
3776
3777 list_for_each_entry_safe(ns, next, &ns_list, list)
3778 nvme_ns_remove(ns);
3779 }
3780 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3781
nvme_class_uevent(struct device * dev,struct kobj_uevent_env * env)3782 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3783 {
3784 struct nvme_ctrl *ctrl =
3785 container_of(dev, struct nvme_ctrl, ctrl_device);
3786 struct nvmf_ctrl_options *opts = ctrl->opts;
3787 int ret;
3788
3789 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3790 if (ret)
3791 return ret;
3792
3793 if (opts) {
3794 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3795 if (ret)
3796 return ret;
3797
3798 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3799 opts->trsvcid ?: "none");
3800 if (ret)
3801 return ret;
3802
3803 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3804 opts->host_traddr ?: "none");
3805 }
3806 return ret;
3807 }
3808
nvme_aen_uevent(struct nvme_ctrl * ctrl)3809 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3810 {
3811 char *envp[2] = { NULL, NULL };
3812 u32 aen_result = ctrl->aen_result;
3813
3814 ctrl->aen_result = 0;
3815 if (!aen_result)
3816 return;
3817
3818 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3819 if (!envp[0])
3820 return;
3821 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3822 kfree(envp[0]);
3823 }
3824
nvme_async_event_work(struct work_struct * work)3825 static void nvme_async_event_work(struct work_struct *work)
3826 {
3827 struct nvme_ctrl *ctrl =
3828 container_of(work, struct nvme_ctrl, async_event_work);
3829
3830 nvme_aen_uevent(ctrl);
3831 ctrl->ops->submit_async_event(ctrl);
3832 }
3833
nvme_ctrl_pp_status(struct nvme_ctrl * ctrl)3834 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3835 {
3836
3837 u32 csts;
3838
3839 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3840 return false;
3841
3842 if (csts == ~0)
3843 return false;
3844
3845 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3846 }
3847
nvme_get_fw_slot_info(struct nvme_ctrl * ctrl)3848 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3849 {
3850 struct nvme_fw_slot_info_log *log;
3851
3852 log = kmalloc(sizeof(*log), GFP_KERNEL);
3853 if (!log)
3854 return;
3855
3856 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3857 sizeof(*log), 0))
3858 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3859 kfree(log);
3860 }
3861
nvme_fw_act_work(struct work_struct * work)3862 static void nvme_fw_act_work(struct work_struct *work)
3863 {
3864 struct nvme_ctrl *ctrl = container_of(work,
3865 struct nvme_ctrl, fw_act_work);
3866 unsigned long fw_act_timeout;
3867
3868 if (ctrl->mtfa)
3869 fw_act_timeout = jiffies +
3870 msecs_to_jiffies(ctrl->mtfa * 100);
3871 else
3872 fw_act_timeout = jiffies +
3873 msecs_to_jiffies(admin_timeout * 1000);
3874
3875 nvme_stop_queues(ctrl);
3876 while (nvme_ctrl_pp_status(ctrl)) {
3877 if (time_after(jiffies, fw_act_timeout)) {
3878 dev_warn(ctrl->device,
3879 "Fw activation timeout, reset controller\n");
3880 nvme_try_sched_reset(ctrl);
3881 return;
3882 }
3883 msleep(100);
3884 }
3885
3886 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
3887 return;
3888
3889 nvme_start_queues(ctrl);
3890 /* read FW slot information to clear the AER */
3891 nvme_get_fw_slot_info(ctrl);
3892 }
3893
nvme_handle_aen_notice(struct nvme_ctrl * ctrl,u32 result)3894 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3895 {
3896 u32 aer_notice_type = (result & 0xff00) >> 8;
3897
3898 trace_nvme_async_event(ctrl, aer_notice_type);
3899
3900 switch (aer_notice_type) {
3901 case NVME_AER_NOTICE_NS_CHANGED:
3902 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3903 nvme_queue_scan(ctrl);
3904 break;
3905 case NVME_AER_NOTICE_FW_ACT_STARTING:
3906 /*
3907 * We are (ab)using the RESETTING state to prevent subsequent
3908 * recovery actions from interfering with the controller's
3909 * firmware activation.
3910 */
3911 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
3912 queue_work(nvme_wq, &ctrl->fw_act_work);
3913 break;
3914 #ifdef CONFIG_NVME_MULTIPATH
3915 case NVME_AER_NOTICE_ANA:
3916 if (!ctrl->ana_log_buf)
3917 break;
3918 queue_work(nvme_wq, &ctrl->ana_work);
3919 break;
3920 #endif
3921 case NVME_AER_NOTICE_DISC_CHANGED:
3922 ctrl->aen_result = result;
3923 break;
3924 default:
3925 dev_warn(ctrl->device, "async event result %08x\n", result);
3926 }
3927 }
3928
nvme_complete_async_event(struct nvme_ctrl * ctrl,__le16 status,volatile union nvme_result * res)3929 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3930 volatile union nvme_result *res)
3931 {
3932 u32 result = le32_to_cpu(res->u32);
3933 u32 aer_type = result & 0x07;
3934
3935 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3936 return;
3937
3938 switch (aer_type) {
3939 case NVME_AER_NOTICE:
3940 nvme_handle_aen_notice(ctrl, result);
3941 break;
3942 case NVME_AER_ERROR:
3943 case NVME_AER_SMART:
3944 case NVME_AER_CSS:
3945 case NVME_AER_VS:
3946 trace_nvme_async_event(ctrl, aer_type);
3947 ctrl->aen_result = result;
3948 break;
3949 default:
3950 break;
3951 }
3952 queue_work(nvme_wq, &ctrl->async_event_work);
3953 }
3954 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3955
nvme_stop_ctrl(struct nvme_ctrl * ctrl)3956 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3957 {
3958 nvme_mpath_stop(ctrl);
3959 nvme_stop_keep_alive(ctrl);
3960 flush_work(&ctrl->async_event_work);
3961 cancel_work_sync(&ctrl->fw_act_work);
3962 }
3963 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3964
nvme_start_ctrl(struct nvme_ctrl * ctrl)3965 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3966 {
3967 if (ctrl->kato)
3968 nvme_start_keep_alive(ctrl);
3969
3970 nvme_enable_aen(ctrl);
3971
3972 if (ctrl->queue_count > 1) {
3973 nvme_queue_scan(ctrl);
3974 nvme_start_queues(ctrl);
3975 }
3976 }
3977 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3978
nvme_uninit_ctrl(struct nvme_ctrl * ctrl)3979 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3980 {
3981 nvme_fault_inject_fini(&ctrl->fault_inject);
3982 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3983 cdev_device_del(&ctrl->cdev, ctrl->device);
3984 }
3985 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3986
nvme_free_ctrl(struct device * dev)3987 static void nvme_free_ctrl(struct device *dev)
3988 {
3989 struct nvme_ctrl *ctrl =
3990 container_of(dev, struct nvme_ctrl, ctrl_device);
3991 struct nvme_subsystem *subsys = ctrl->subsys;
3992
3993 if (subsys && ctrl->instance != subsys->instance)
3994 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3995
3996 kfree(ctrl->effects);
3997 nvme_mpath_uninit(ctrl);
3998 __free_page(ctrl->discard_page);
3999
4000 if (subsys) {
4001 mutex_lock(&nvme_subsystems_lock);
4002 list_del(&ctrl->subsys_entry);
4003 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4004 mutex_unlock(&nvme_subsystems_lock);
4005 }
4006
4007 ctrl->ops->free_ctrl(ctrl);
4008
4009 if (subsys)
4010 nvme_put_subsystem(subsys);
4011 }
4012
4013 /*
4014 * Initialize a NVMe controller structures. This needs to be called during
4015 * earliest initialization so that we have the initialized structured around
4016 * during probing.
4017 */
nvme_init_ctrl(struct nvme_ctrl * ctrl,struct device * dev,const struct nvme_ctrl_ops * ops,unsigned long quirks)4018 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4019 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4020 {
4021 int ret;
4022
4023 ctrl->state = NVME_CTRL_NEW;
4024 spin_lock_init(&ctrl->lock);
4025 mutex_init(&ctrl->scan_lock);
4026 INIT_LIST_HEAD(&ctrl->namespaces);
4027 init_rwsem(&ctrl->namespaces_rwsem);
4028 ctrl->dev = dev;
4029 ctrl->ops = ops;
4030 ctrl->quirks = quirks;
4031 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4032 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4033 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4034 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4035 init_waitqueue_head(&ctrl->state_wq);
4036
4037 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4038 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4039 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4040
4041 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4042 PAGE_SIZE);
4043 ctrl->discard_page = alloc_page(GFP_KERNEL);
4044 if (!ctrl->discard_page) {
4045 ret = -ENOMEM;
4046 goto out;
4047 }
4048
4049 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4050 if (ret < 0)
4051 goto out;
4052 ctrl->instance = ret;
4053
4054 device_initialize(&ctrl->ctrl_device);
4055 ctrl->device = &ctrl->ctrl_device;
4056 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4057 ctrl->device->class = nvme_class;
4058 ctrl->device->parent = ctrl->dev;
4059 ctrl->device->groups = nvme_dev_attr_groups;
4060 ctrl->device->release = nvme_free_ctrl;
4061 dev_set_drvdata(ctrl->device, ctrl);
4062 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4063 if (ret)
4064 goto out_release_instance;
4065
4066 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4067 ctrl->cdev.owner = ops->module;
4068 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4069 if (ret)
4070 goto out_free_name;
4071
4072 /*
4073 * Initialize latency tolerance controls. The sysfs files won't
4074 * be visible to userspace unless the device actually supports APST.
4075 */
4076 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4077 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4078 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4079
4080 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4081
4082 return 0;
4083 out_free_name:
4084 kfree_const(ctrl->device->kobj.name);
4085 out_release_instance:
4086 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4087 out:
4088 if (ctrl->discard_page)
4089 __free_page(ctrl->discard_page);
4090 return ret;
4091 }
4092 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4093
4094 /**
4095 * nvme_kill_queues(): Ends all namespace queues
4096 * @ctrl: the dead controller that needs to end
4097 *
4098 * Call this function when the driver determines it is unable to get the
4099 * controller in a state capable of servicing IO.
4100 */
nvme_kill_queues(struct nvme_ctrl * ctrl)4101 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4102 {
4103 struct nvme_ns *ns;
4104
4105 down_read(&ctrl->namespaces_rwsem);
4106
4107 /* Forcibly unquiesce queues to avoid blocking dispatch */
4108 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4109 blk_mq_unquiesce_queue(ctrl->admin_q);
4110
4111 list_for_each_entry(ns, &ctrl->namespaces, list)
4112 nvme_set_queue_dying(ns);
4113
4114 up_read(&ctrl->namespaces_rwsem);
4115 }
4116 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4117
nvme_unfreeze(struct nvme_ctrl * ctrl)4118 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4119 {
4120 struct nvme_ns *ns;
4121
4122 down_read(&ctrl->namespaces_rwsem);
4123 list_for_each_entry(ns, &ctrl->namespaces, list)
4124 blk_mq_unfreeze_queue(ns->queue);
4125 up_read(&ctrl->namespaces_rwsem);
4126 }
4127 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4128
nvme_wait_freeze_timeout(struct nvme_ctrl * ctrl,long timeout)4129 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4130 {
4131 struct nvme_ns *ns;
4132
4133 down_read(&ctrl->namespaces_rwsem);
4134 list_for_each_entry(ns, &ctrl->namespaces, list) {
4135 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4136 if (timeout <= 0)
4137 break;
4138 }
4139 up_read(&ctrl->namespaces_rwsem);
4140 }
4141 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4142
nvme_wait_freeze(struct nvme_ctrl * ctrl)4143 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4144 {
4145 struct nvme_ns *ns;
4146
4147 down_read(&ctrl->namespaces_rwsem);
4148 list_for_each_entry(ns, &ctrl->namespaces, list)
4149 blk_mq_freeze_queue_wait(ns->queue);
4150 up_read(&ctrl->namespaces_rwsem);
4151 }
4152 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4153
nvme_start_freeze(struct nvme_ctrl * ctrl)4154 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4155 {
4156 struct nvme_ns *ns;
4157
4158 down_read(&ctrl->namespaces_rwsem);
4159 list_for_each_entry(ns, &ctrl->namespaces, list)
4160 blk_freeze_queue_start(ns->queue);
4161 up_read(&ctrl->namespaces_rwsem);
4162 }
4163 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4164
nvme_stop_queues(struct nvme_ctrl * ctrl)4165 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4166 {
4167 struct nvme_ns *ns;
4168
4169 down_read(&ctrl->namespaces_rwsem);
4170 list_for_each_entry(ns, &ctrl->namespaces, list)
4171 blk_mq_quiesce_queue(ns->queue);
4172 up_read(&ctrl->namespaces_rwsem);
4173 }
4174 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4175
nvme_start_queues(struct nvme_ctrl * ctrl)4176 void nvme_start_queues(struct nvme_ctrl *ctrl)
4177 {
4178 struct nvme_ns *ns;
4179
4180 down_read(&ctrl->namespaces_rwsem);
4181 list_for_each_entry(ns, &ctrl->namespaces, list)
4182 blk_mq_unquiesce_queue(ns->queue);
4183 up_read(&ctrl->namespaces_rwsem);
4184 }
4185 EXPORT_SYMBOL_GPL(nvme_start_queues);
4186
4187
nvme_sync_queues(struct nvme_ctrl * ctrl)4188 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4189 {
4190 struct nvme_ns *ns;
4191
4192 down_read(&ctrl->namespaces_rwsem);
4193 list_for_each_entry(ns, &ctrl->namespaces, list)
4194 blk_sync_queue(ns->queue);
4195 up_read(&ctrl->namespaces_rwsem);
4196
4197 if (ctrl->admin_q)
4198 blk_sync_queue(ctrl->admin_q);
4199 }
4200 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4201
4202 /*
4203 * Check we didn't inadvertently grow the command structure sizes:
4204 */
_nvme_check_size(void)4205 static inline void _nvme_check_size(void)
4206 {
4207 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4208 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4209 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4210 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4211 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4212 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4213 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4214 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4215 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4216 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4217 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4218 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4219 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4220 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4221 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4222 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4223 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4224 }
4225
4226
nvme_core_init(void)4227 static int __init nvme_core_init(void)
4228 {
4229 int result = -ENOMEM;
4230
4231 _nvme_check_size();
4232
4233 nvme_wq = alloc_workqueue("nvme-wq",
4234 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4235 if (!nvme_wq)
4236 goto out;
4237
4238 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4239 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4240 if (!nvme_reset_wq)
4241 goto destroy_wq;
4242
4243 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4244 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4245 if (!nvme_delete_wq)
4246 goto destroy_reset_wq;
4247
4248 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4249 if (result < 0)
4250 goto destroy_delete_wq;
4251
4252 nvme_class = class_create(THIS_MODULE, "nvme");
4253 if (IS_ERR(nvme_class)) {
4254 result = PTR_ERR(nvme_class);
4255 goto unregister_chrdev;
4256 }
4257 nvme_class->dev_uevent = nvme_class_uevent;
4258
4259 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4260 if (IS_ERR(nvme_subsys_class)) {
4261 result = PTR_ERR(nvme_subsys_class);
4262 goto destroy_class;
4263 }
4264 return 0;
4265
4266 destroy_class:
4267 class_destroy(nvme_class);
4268 unregister_chrdev:
4269 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4270 destroy_delete_wq:
4271 destroy_workqueue(nvme_delete_wq);
4272 destroy_reset_wq:
4273 destroy_workqueue(nvme_reset_wq);
4274 destroy_wq:
4275 destroy_workqueue(nvme_wq);
4276 out:
4277 return result;
4278 }
4279
nvme_core_exit(void)4280 static void __exit nvme_core_exit(void)
4281 {
4282 class_destroy(nvme_subsys_class);
4283 class_destroy(nvme_class);
4284 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4285 destroy_workqueue(nvme_delete_wq);
4286 destroy_workqueue(nvme_reset_wq);
4287 destroy_workqueue(nvme_wq);
4288 }
4289
4290 MODULE_LICENSE("GPL");
4291 MODULE_VERSION("1.0");
4292 module_init(nvme_core_init);
4293 module_exit(nvme_core_exit);
4294