xref: /openbmc/linux/drivers/nvme/host/core.c (revision 8781e5df)
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 
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 
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  */
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 
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 
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 
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 
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 
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 
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 
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 
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 
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 
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 
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 
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_PATH_ERROR;
317 	blk_mq_complete_request(req);
318 	return true;
319 }
320 EXPORT_SYMBOL_GPL(nvme_cancel_request);
321 
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  */
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  */
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 
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 
446 static void nvme_put_ns_head(struct nvme_ns_head *head)
447 {
448 	kref_put(&head->ref, nvme_free_ns_head);
449 }
450 
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 
464 static void nvme_put_ns(struct nvme_ns *ns)
465 {
466 	kref_put(&ns->kref, nvme_free_ns);
467 }
468 
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 
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 
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 
517 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
518 {
519 	return nvme_toggle_streams(ctrl, false);
520 }
521 
522 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
523 {
524 	return nvme_toggle_streams(ctrl, true);
525 }
526 
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 
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  */
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 
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 
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 	range = kmalloc_array(segments, sizeof(*range),
615 				GFP_ATOMIC | __GFP_NOWARN);
616 	if (!range) {
617 		/*
618 		 * If we fail allocation our range, fallback to the controller
619 		 * discard page. If that's also busy, it's safe to return
620 		 * busy, as we know we can make progress once that's freed.
621 		 */
622 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
623 			return BLK_STS_RESOURCE;
624 
625 		range = page_address(ns->ctrl->discard_page);
626 	}
627 
628 	__rq_for_each_bio(bio, req) {
629 		u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
630 		u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
631 
632 		if (n < segments) {
633 			range[n].cattr = cpu_to_le32(0);
634 			range[n].nlb = cpu_to_le32(nlb);
635 			range[n].slba = cpu_to_le64(slba);
636 		}
637 		n++;
638 	}
639 
640 	if (WARN_ON_ONCE(n != segments)) {
641 		if (virt_to_page(range) == ns->ctrl->discard_page)
642 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
643 		else
644 			kfree(range);
645 		return BLK_STS_IOERR;
646 	}
647 
648 	cmnd->dsm.opcode = nvme_cmd_dsm;
649 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
650 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
651 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
652 
653 	req->special_vec.bv_page = virt_to_page(range);
654 	req->special_vec.bv_offset = offset_in_page(range);
655 	req->special_vec.bv_len = sizeof(*range) * segments;
656 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
657 
658 	return BLK_STS_OK;
659 }
660 
661 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
662 		struct request *req, struct nvme_command *cmnd)
663 {
664 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
665 		return nvme_setup_discard(ns, req, cmnd);
666 
667 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
668 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
669 	cmnd->write_zeroes.slba =
670 		cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
671 	cmnd->write_zeroes.length =
672 		cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
673 	cmnd->write_zeroes.control = 0;
674 	return BLK_STS_OK;
675 }
676 
677 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
678 		struct request *req, struct nvme_command *cmnd)
679 {
680 	struct nvme_ctrl *ctrl = ns->ctrl;
681 	u16 control = 0;
682 	u32 dsmgmt = 0;
683 
684 	if (req->cmd_flags & REQ_FUA)
685 		control |= NVME_RW_FUA;
686 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
687 		control |= NVME_RW_LR;
688 
689 	if (req->cmd_flags & REQ_RAHEAD)
690 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
691 
692 	cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
693 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
694 	cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
695 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
696 
697 	if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
698 		nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
699 
700 	if (ns->ms) {
701 		/*
702 		 * If formated with metadata, the block layer always provides a
703 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
704 		 * we enable the PRACT bit for protection information or set the
705 		 * namespace capacity to zero to prevent any I/O.
706 		 */
707 		if (!blk_integrity_rq(req)) {
708 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
709 				return BLK_STS_NOTSUPP;
710 			control |= NVME_RW_PRINFO_PRACT;
711 		}
712 
713 		switch (ns->pi_type) {
714 		case NVME_NS_DPS_PI_TYPE3:
715 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
716 			break;
717 		case NVME_NS_DPS_PI_TYPE1:
718 		case NVME_NS_DPS_PI_TYPE2:
719 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
720 					NVME_RW_PRINFO_PRCHK_REF;
721 			cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
722 			break;
723 		}
724 	}
725 
726 	cmnd->rw.control = cpu_to_le16(control);
727 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
728 	return 0;
729 }
730 
731 void nvme_cleanup_cmd(struct request *req)
732 {
733 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
734 		struct nvme_ns *ns = req->rq_disk->private_data;
735 		struct page *page = req->special_vec.bv_page;
736 
737 		if (page == ns->ctrl->discard_page)
738 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
739 		else
740 			kfree(page_address(page) + req->special_vec.bv_offset);
741 	}
742 }
743 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
744 
745 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
746 		struct nvme_command *cmd)
747 {
748 	blk_status_t ret = BLK_STS_OK;
749 
750 	nvme_clear_nvme_request(req);
751 
752 	memset(cmd, 0, sizeof(*cmd));
753 	switch (req_op(req)) {
754 	case REQ_OP_DRV_IN:
755 	case REQ_OP_DRV_OUT:
756 		memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
757 		break;
758 	case REQ_OP_FLUSH:
759 		nvme_setup_flush(ns, cmd);
760 		break;
761 	case REQ_OP_WRITE_ZEROES:
762 		ret = nvme_setup_write_zeroes(ns, req, cmd);
763 		break;
764 	case REQ_OP_DISCARD:
765 		ret = nvme_setup_discard(ns, req, cmd);
766 		break;
767 	case REQ_OP_READ:
768 	case REQ_OP_WRITE:
769 		ret = nvme_setup_rw(ns, req, cmd);
770 		break;
771 	default:
772 		WARN_ON_ONCE(1);
773 		return BLK_STS_IOERR;
774 	}
775 
776 	cmd->common.command_id = req->tag;
777 	trace_nvme_setup_cmd(req, cmd);
778 	return ret;
779 }
780 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
781 
782 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
783 {
784 	struct completion *waiting = rq->end_io_data;
785 
786 	rq->end_io_data = NULL;
787 	complete(waiting);
788 }
789 
790 static void nvme_execute_rq_polled(struct request_queue *q,
791 		struct gendisk *bd_disk, struct request *rq, int at_head)
792 {
793 	DECLARE_COMPLETION_ONSTACK(wait);
794 
795 	WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
796 
797 	rq->cmd_flags |= REQ_HIPRI;
798 	rq->end_io_data = &wait;
799 	blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
800 
801 	while (!completion_done(&wait)) {
802 		blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
803 		cond_resched();
804 	}
805 }
806 
807 /*
808  * Returns 0 on success.  If the result is negative, it's a Linux error code;
809  * if the result is positive, it's an NVM Express status code
810  */
811 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
812 		union nvme_result *result, void *buffer, unsigned bufflen,
813 		unsigned timeout, int qid, int at_head,
814 		blk_mq_req_flags_t flags, bool poll)
815 {
816 	struct request *req;
817 	int ret;
818 
819 	req = nvme_alloc_request(q, cmd, flags, qid);
820 	if (IS_ERR(req))
821 		return PTR_ERR(req);
822 
823 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
824 
825 	if (buffer && bufflen) {
826 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
827 		if (ret)
828 			goto out;
829 	}
830 
831 	if (poll)
832 		nvme_execute_rq_polled(req->q, NULL, req, at_head);
833 	else
834 		blk_execute_rq(req->q, NULL, req, at_head);
835 	if (result)
836 		*result = nvme_req(req)->result;
837 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
838 		ret = -EINTR;
839 	else
840 		ret = nvme_req(req)->status;
841  out:
842 	blk_mq_free_request(req);
843 	return ret;
844 }
845 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
846 
847 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
848 		void *buffer, unsigned bufflen)
849 {
850 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
851 			NVME_QID_ANY, 0, 0, false);
852 }
853 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
854 
855 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
856 		unsigned len, u32 seed, bool write)
857 {
858 	struct bio_integrity_payload *bip;
859 	int ret = -ENOMEM;
860 	void *buf;
861 
862 	buf = kmalloc(len, GFP_KERNEL);
863 	if (!buf)
864 		goto out;
865 
866 	ret = -EFAULT;
867 	if (write && copy_from_user(buf, ubuf, len))
868 		goto out_free_meta;
869 
870 	bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
871 	if (IS_ERR(bip)) {
872 		ret = PTR_ERR(bip);
873 		goto out_free_meta;
874 	}
875 
876 	bip->bip_iter.bi_size = len;
877 	bip->bip_iter.bi_sector = seed;
878 	ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
879 			offset_in_page(buf));
880 	if (ret == len)
881 		return buf;
882 	ret = -ENOMEM;
883 out_free_meta:
884 	kfree(buf);
885 out:
886 	return ERR_PTR(ret);
887 }
888 
889 static int nvme_submit_user_cmd(struct request_queue *q,
890 		struct nvme_command *cmd, void __user *ubuffer,
891 		unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
892 		u32 meta_seed, u64 *result, unsigned timeout)
893 {
894 	bool write = nvme_is_write(cmd);
895 	struct nvme_ns *ns = q->queuedata;
896 	struct gendisk *disk = ns ? ns->disk : NULL;
897 	struct request *req;
898 	struct bio *bio = NULL;
899 	void *meta = NULL;
900 	int ret;
901 
902 	req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
903 	if (IS_ERR(req))
904 		return PTR_ERR(req);
905 
906 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
907 	nvme_req(req)->flags |= NVME_REQ_USERCMD;
908 
909 	if (ubuffer && bufflen) {
910 		ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
911 				GFP_KERNEL);
912 		if (ret)
913 			goto out;
914 		bio = req->bio;
915 		bio->bi_disk = disk;
916 		if (disk && meta_buffer && meta_len) {
917 			meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
918 					meta_seed, write);
919 			if (IS_ERR(meta)) {
920 				ret = PTR_ERR(meta);
921 				goto out_unmap;
922 			}
923 			req->cmd_flags |= REQ_INTEGRITY;
924 		}
925 	}
926 
927 	blk_execute_rq(req->q, disk, req, 0);
928 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
929 		ret = -EINTR;
930 	else
931 		ret = nvme_req(req)->status;
932 	if (result)
933 		*result = le64_to_cpu(nvme_req(req)->result.u64);
934 	if (meta && !ret && !write) {
935 		if (copy_to_user(meta_buffer, meta, meta_len))
936 			ret = -EFAULT;
937 	}
938 	kfree(meta);
939  out_unmap:
940 	if (bio)
941 		blk_rq_unmap_user(bio);
942  out:
943 	blk_mq_free_request(req);
944 	return ret;
945 }
946 
947 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
948 {
949 	struct nvme_ctrl *ctrl = rq->end_io_data;
950 	unsigned long flags;
951 	bool startka = false;
952 
953 	blk_mq_free_request(rq);
954 
955 	if (status) {
956 		dev_err(ctrl->device,
957 			"failed nvme_keep_alive_end_io error=%d\n",
958 				status);
959 		return;
960 	}
961 
962 	ctrl->comp_seen = false;
963 	spin_lock_irqsave(&ctrl->lock, flags);
964 	if (ctrl->state == NVME_CTRL_LIVE ||
965 	    ctrl->state == NVME_CTRL_CONNECTING)
966 		startka = true;
967 	spin_unlock_irqrestore(&ctrl->lock, flags);
968 	if (startka)
969 		schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
970 }
971 
972 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
973 {
974 	struct request *rq;
975 
976 	rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
977 			NVME_QID_ANY);
978 	if (IS_ERR(rq))
979 		return PTR_ERR(rq);
980 
981 	rq->timeout = ctrl->kato * HZ;
982 	rq->end_io_data = ctrl;
983 
984 	blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
985 
986 	return 0;
987 }
988 
989 static void nvme_keep_alive_work(struct work_struct *work)
990 {
991 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
992 			struct nvme_ctrl, ka_work);
993 	bool comp_seen = ctrl->comp_seen;
994 
995 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
996 		dev_dbg(ctrl->device,
997 			"reschedule traffic based keep-alive timer\n");
998 		ctrl->comp_seen = false;
999 		schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
1000 		return;
1001 	}
1002 
1003 	if (nvme_keep_alive(ctrl)) {
1004 		/* allocation failure, reset the controller */
1005 		dev_err(ctrl->device, "keep-alive failed\n");
1006 		nvme_reset_ctrl(ctrl);
1007 		return;
1008 	}
1009 }
1010 
1011 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1012 {
1013 	if (unlikely(ctrl->kato == 0))
1014 		return;
1015 
1016 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
1017 }
1018 
1019 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1020 {
1021 	if (unlikely(ctrl->kato == 0))
1022 		return;
1023 
1024 	cancel_delayed_work_sync(&ctrl->ka_work);
1025 }
1026 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1027 
1028 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1029 {
1030 	struct nvme_command c = { };
1031 	int error;
1032 
1033 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1034 	c.identify.opcode = nvme_admin_identify;
1035 	c.identify.cns = NVME_ID_CNS_CTRL;
1036 
1037 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1038 	if (!*id)
1039 		return -ENOMEM;
1040 
1041 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1042 			sizeof(struct nvme_id_ctrl));
1043 	if (error)
1044 		kfree(*id);
1045 	return error;
1046 }
1047 
1048 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1049 		struct nvme_ns_ids *ids)
1050 {
1051 	struct nvme_command c = { };
1052 	int status;
1053 	void *data;
1054 	int pos;
1055 	int len;
1056 
1057 	c.identify.opcode = nvme_admin_identify;
1058 	c.identify.nsid = cpu_to_le32(nsid);
1059 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1060 
1061 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1062 	if (!data)
1063 		return -ENOMEM;
1064 
1065 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1066 				      NVME_IDENTIFY_DATA_SIZE);
1067 	if (status)
1068 		goto free_data;
1069 
1070 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1071 		struct nvme_ns_id_desc *cur = data + pos;
1072 
1073 		if (cur->nidl == 0)
1074 			break;
1075 
1076 		switch (cur->nidt) {
1077 		case NVME_NIDT_EUI64:
1078 			if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1079 				dev_warn(ctrl->device,
1080 					 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1081 					 cur->nidl);
1082 				goto free_data;
1083 			}
1084 			len = NVME_NIDT_EUI64_LEN;
1085 			memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1086 			break;
1087 		case NVME_NIDT_NGUID:
1088 			if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1089 				dev_warn(ctrl->device,
1090 					 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1091 					 cur->nidl);
1092 				goto free_data;
1093 			}
1094 			len = NVME_NIDT_NGUID_LEN;
1095 			memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1096 			break;
1097 		case NVME_NIDT_UUID:
1098 			if (cur->nidl != NVME_NIDT_UUID_LEN) {
1099 				dev_warn(ctrl->device,
1100 					 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1101 					 cur->nidl);
1102 				goto free_data;
1103 			}
1104 			len = NVME_NIDT_UUID_LEN;
1105 			uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1106 			break;
1107 		default:
1108 			/* Skip unknown types */
1109 			len = cur->nidl;
1110 			break;
1111 		}
1112 
1113 		len += sizeof(*cur);
1114 	}
1115 free_data:
1116 	kfree(data);
1117 	return status;
1118 }
1119 
1120 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1121 {
1122 	struct nvme_command c = { };
1123 
1124 	c.identify.opcode = nvme_admin_identify;
1125 	c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1126 	c.identify.nsid = cpu_to_le32(nsid);
1127 	return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1128 				    NVME_IDENTIFY_DATA_SIZE);
1129 }
1130 
1131 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1132 		unsigned nsid, struct nvme_id_ns **id)
1133 {
1134 	struct nvme_command c = { };
1135 	int error;
1136 
1137 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1138 	c.identify.opcode = nvme_admin_identify;
1139 	c.identify.nsid = cpu_to_le32(nsid);
1140 	c.identify.cns = NVME_ID_CNS_NS;
1141 
1142 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1143 	if (!*id)
1144 		return -ENOMEM;
1145 
1146 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1147 	if (error) {
1148 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1149 		kfree(*id);
1150 	}
1151 
1152 	return error;
1153 }
1154 
1155 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1156 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1157 {
1158 	struct nvme_command c;
1159 	union nvme_result res;
1160 	int ret;
1161 
1162 	memset(&c, 0, sizeof(c));
1163 	c.features.opcode = op;
1164 	c.features.fid = cpu_to_le32(fid);
1165 	c.features.dword11 = cpu_to_le32(dword11);
1166 
1167 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1168 			buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1169 	if (ret >= 0 && result)
1170 		*result = le32_to_cpu(res.u32);
1171 	return ret;
1172 }
1173 
1174 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1175 		      unsigned int dword11, void *buffer, size_t buflen,
1176 		      u32 *result)
1177 {
1178 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1179 			     buflen, result);
1180 }
1181 EXPORT_SYMBOL_GPL(nvme_set_features);
1182 
1183 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1184 		      unsigned int dword11, void *buffer, size_t buflen,
1185 		      u32 *result)
1186 {
1187 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1188 			     buflen, result);
1189 }
1190 EXPORT_SYMBOL_GPL(nvme_get_features);
1191 
1192 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1193 {
1194 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1195 	u32 result;
1196 	int status, nr_io_queues;
1197 
1198 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1199 			&result);
1200 	if (status < 0)
1201 		return status;
1202 
1203 	/*
1204 	 * Degraded controllers might return an error when setting the queue
1205 	 * count.  We still want to be able to bring them online and offer
1206 	 * access to the admin queue, as that might be only way to fix them up.
1207 	 */
1208 	if (status > 0) {
1209 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1210 		*count = 0;
1211 	} else {
1212 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1213 		*count = min(*count, nr_io_queues);
1214 	}
1215 
1216 	return 0;
1217 }
1218 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1219 
1220 #define NVME_AEN_SUPPORTED \
1221 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1222 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1223 
1224 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1225 {
1226 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1227 	int status;
1228 
1229 	if (!supported_aens)
1230 		return;
1231 
1232 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1233 			NULL, 0, &result);
1234 	if (status)
1235 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1236 			 supported_aens);
1237 
1238 	queue_work(nvme_wq, &ctrl->async_event_work);
1239 }
1240 
1241 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1242 {
1243 	struct nvme_user_io io;
1244 	struct nvme_command c;
1245 	unsigned length, meta_len;
1246 	void __user *metadata;
1247 
1248 	if (copy_from_user(&io, uio, sizeof(io)))
1249 		return -EFAULT;
1250 	if (io.flags)
1251 		return -EINVAL;
1252 
1253 	switch (io.opcode) {
1254 	case nvme_cmd_write:
1255 	case nvme_cmd_read:
1256 	case nvme_cmd_compare:
1257 		break;
1258 	default:
1259 		return -EINVAL;
1260 	}
1261 
1262 	length = (io.nblocks + 1) << ns->lba_shift;
1263 	meta_len = (io.nblocks + 1) * ns->ms;
1264 	metadata = (void __user *)(uintptr_t)io.metadata;
1265 
1266 	if (ns->ext) {
1267 		length += meta_len;
1268 		meta_len = 0;
1269 	} else if (meta_len) {
1270 		if ((io.metadata & 3) || !io.metadata)
1271 			return -EINVAL;
1272 	}
1273 
1274 	memset(&c, 0, sizeof(c));
1275 	c.rw.opcode = io.opcode;
1276 	c.rw.flags = io.flags;
1277 	c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1278 	c.rw.slba = cpu_to_le64(io.slba);
1279 	c.rw.length = cpu_to_le16(io.nblocks);
1280 	c.rw.control = cpu_to_le16(io.control);
1281 	c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1282 	c.rw.reftag = cpu_to_le32(io.reftag);
1283 	c.rw.apptag = cpu_to_le16(io.apptag);
1284 	c.rw.appmask = cpu_to_le16(io.appmask);
1285 
1286 	return nvme_submit_user_cmd(ns->queue, &c,
1287 			(void __user *)(uintptr_t)io.addr, length,
1288 			metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1289 }
1290 
1291 static u32 nvme_known_admin_effects(u8 opcode)
1292 {
1293 	switch (opcode) {
1294 	case nvme_admin_format_nvm:
1295 		return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1296 					NVME_CMD_EFFECTS_CSE_MASK;
1297 	case nvme_admin_sanitize_nvm:
1298 		return NVME_CMD_EFFECTS_CSE_MASK;
1299 	default:
1300 		break;
1301 	}
1302 	return 0;
1303 }
1304 
1305 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1306 								u8 opcode)
1307 {
1308 	u32 effects = 0;
1309 
1310 	if (ns) {
1311 		if (ctrl->effects)
1312 			effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1313 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1314 			dev_warn(ctrl->device,
1315 				 "IO command:%02x has unhandled effects:%08x\n",
1316 				 opcode, effects);
1317 		return 0;
1318 	}
1319 
1320 	if (ctrl->effects)
1321 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1322 	effects |= nvme_known_admin_effects(opcode);
1323 
1324 	/*
1325 	 * For simplicity, IO to all namespaces is quiesced even if the command
1326 	 * effects say only one namespace is affected.
1327 	 */
1328 	if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1329 		mutex_lock(&ctrl->scan_lock);
1330 		mutex_lock(&ctrl->subsys->lock);
1331 		nvme_mpath_start_freeze(ctrl->subsys);
1332 		nvme_mpath_wait_freeze(ctrl->subsys);
1333 		nvme_start_freeze(ctrl);
1334 		nvme_wait_freeze(ctrl);
1335 	}
1336 	return effects;
1337 }
1338 
1339 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1340 {
1341 	struct nvme_ns *ns;
1342 
1343 	down_read(&ctrl->namespaces_rwsem);
1344 	list_for_each_entry(ns, &ctrl->namespaces, list)
1345 		if (ns->disk && nvme_revalidate_disk(ns->disk))
1346 			nvme_set_queue_dying(ns);
1347 	up_read(&ctrl->namespaces_rwsem);
1348 }
1349 
1350 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1351 {
1352 	/*
1353 	 * Revalidate LBA changes prior to unfreezing. This is necessary to
1354 	 * prevent memory corruption if a logical block size was changed by
1355 	 * this command.
1356 	 */
1357 	if (effects & NVME_CMD_EFFECTS_LBCC)
1358 		nvme_update_formats(ctrl);
1359 	if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1360 		nvme_unfreeze(ctrl);
1361 		nvme_mpath_unfreeze(ctrl->subsys);
1362 		mutex_unlock(&ctrl->subsys->lock);
1363 		nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1364 		mutex_unlock(&ctrl->scan_lock);
1365 	}
1366 	if (effects & NVME_CMD_EFFECTS_CCC)
1367 		nvme_init_identify(ctrl);
1368 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1369 		nvme_queue_scan(ctrl);
1370 }
1371 
1372 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1373 			struct nvme_passthru_cmd __user *ucmd)
1374 {
1375 	struct nvme_passthru_cmd cmd;
1376 	struct nvme_command c;
1377 	unsigned timeout = 0;
1378 	u32 effects;
1379 	u64 result;
1380 	int status;
1381 
1382 	if (!capable(CAP_SYS_ADMIN))
1383 		return -EACCES;
1384 	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1385 		return -EFAULT;
1386 	if (cmd.flags)
1387 		return -EINVAL;
1388 
1389 	memset(&c, 0, sizeof(c));
1390 	c.common.opcode = cmd.opcode;
1391 	c.common.flags = cmd.flags;
1392 	c.common.nsid = cpu_to_le32(cmd.nsid);
1393 	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1394 	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1395 	c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1396 	c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1397 	c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1398 	c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1399 	c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1400 	c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1401 
1402 	if (cmd.timeout_ms)
1403 		timeout = msecs_to_jiffies(cmd.timeout_ms);
1404 
1405 	effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1406 	status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1407 			(void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1408 			(void __user *)(uintptr_t)cmd.metadata,
1409 			cmd.metadata_len, 0, &result, timeout);
1410 	nvme_passthru_end(ctrl, effects);
1411 
1412 	if (status >= 0) {
1413 		if (put_user(result, &ucmd->result))
1414 			return -EFAULT;
1415 	}
1416 
1417 	return status;
1418 }
1419 
1420 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1421 			struct nvme_passthru_cmd64 __user *ucmd)
1422 {
1423 	struct nvme_passthru_cmd64 cmd;
1424 	struct nvme_command c;
1425 	unsigned timeout = 0;
1426 	u32 effects;
1427 	int status;
1428 
1429 	if (!capable(CAP_SYS_ADMIN))
1430 		return -EACCES;
1431 	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1432 		return -EFAULT;
1433 	if (cmd.flags)
1434 		return -EINVAL;
1435 
1436 	memset(&c, 0, sizeof(c));
1437 	c.common.opcode = cmd.opcode;
1438 	c.common.flags = cmd.flags;
1439 	c.common.nsid = cpu_to_le32(cmd.nsid);
1440 	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1441 	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1442 	c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1443 	c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1444 	c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1445 	c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1446 	c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1447 	c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1448 
1449 	if (cmd.timeout_ms)
1450 		timeout = msecs_to_jiffies(cmd.timeout_ms);
1451 
1452 	effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1453 	status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1454 			(void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1455 			(void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1456 			0, &cmd.result, timeout);
1457 	nvme_passthru_end(ctrl, effects);
1458 
1459 	if (status >= 0) {
1460 		if (put_user(cmd.result, &ucmd->result))
1461 			return -EFAULT;
1462 	}
1463 
1464 	return status;
1465 }
1466 
1467 /*
1468  * Issue ioctl requests on the first available path.  Note that unlike normal
1469  * block layer requests we will not retry failed request on another controller.
1470  */
1471 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1472 		struct nvme_ns_head **head, int *srcu_idx)
1473 {
1474 #ifdef CONFIG_NVME_MULTIPATH
1475 	if (disk->fops == &nvme_ns_head_ops) {
1476 		struct nvme_ns *ns;
1477 
1478 		*head = disk->private_data;
1479 		*srcu_idx = srcu_read_lock(&(*head)->srcu);
1480 		ns = nvme_find_path(*head);
1481 		if (!ns)
1482 			srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1483 		return ns;
1484 	}
1485 #endif
1486 	*head = NULL;
1487 	*srcu_idx = -1;
1488 	return disk->private_data;
1489 }
1490 
1491 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1492 {
1493 	if (head)
1494 		srcu_read_unlock(&head->srcu, idx);
1495 }
1496 
1497 static bool is_ctrl_ioctl(unsigned int cmd)
1498 {
1499 	if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1500 		return true;
1501 	if (is_sed_ioctl(cmd))
1502 		return true;
1503 	return false;
1504 }
1505 
1506 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1507 				  void __user *argp,
1508 				  struct nvme_ns_head *head,
1509 				  int srcu_idx)
1510 {
1511 	struct nvme_ctrl *ctrl = ns->ctrl;
1512 	int ret;
1513 
1514 	nvme_get_ctrl(ns->ctrl);
1515 	nvme_put_ns_from_disk(head, srcu_idx);
1516 
1517 	switch (cmd) {
1518 	case NVME_IOCTL_ADMIN_CMD:
1519 		ret = nvme_user_cmd(ctrl, NULL, argp);
1520 		break;
1521 	case NVME_IOCTL_ADMIN64_CMD:
1522 		ret = nvme_user_cmd64(ctrl, NULL, argp);
1523 		break;
1524 	default:
1525 		ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1526 		break;
1527 	}
1528 	nvme_put_ctrl(ctrl);
1529 	return ret;
1530 }
1531 
1532 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1533 		unsigned int cmd, unsigned long arg)
1534 {
1535 	struct nvme_ns_head *head = NULL;
1536 	void __user *argp = (void __user *)arg;
1537 	struct nvme_ns *ns;
1538 	int srcu_idx, ret;
1539 
1540 	ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1541 	if (unlikely(!ns))
1542 		return -EWOULDBLOCK;
1543 
1544 	/*
1545 	 * Handle ioctls that apply to the controller instead of the namespace
1546 	 * seperately and drop the ns SRCU reference early.  This avoids a
1547 	 * deadlock when deleting namespaces using the passthrough interface.
1548 	 */
1549 	if (is_ctrl_ioctl(cmd))
1550 		return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1551 
1552 	switch (cmd) {
1553 	case NVME_IOCTL_ID:
1554 		force_successful_syscall_return();
1555 		ret = ns->head->ns_id;
1556 		break;
1557 	case NVME_IOCTL_IO_CMD:
1558 		ret = nvme_user_cmd(ns->ctrl, ns, argp);
1559 		break;
1560 	case NVME_IOCTL_SUBMIT_IO:
1561 		ret = nvme_submit_io(ns, argp);
1562 		break;
1563 	case NVME_IOCTL_IO64_CMD:
1564 		ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1565 		break;
1566 	default:
1567 		if (ns->ndev)
1568 			ret = nvme_nvm_ioctl(ns, cmd, arg);
1569 		else
1570 			ret = -ENOTTY;
1571 	}
1572 
1573 	nvme_put_ns_from_disk(head, srcu_idx);
1574 	return ret;
1575 }
1576 
1577 static int nvme_open(struct block_device *bdev, fmode_t mode)
1578 {
1579 	struct nvme_ns *ns = bdev->bd_disk->private_data;
1580 
1581 #ifdef CONFIG_NVME_MULTIPATH
1582 	/* should never be called due to GENHD_FL_HIDDEN */
1583 	if (WARN_ON_ONCE(ns->head->disk))
1584 		goto fail;
1585 #endif
1586 	if (!kref_get_unless_zero(&ns->kref))
1587 		goto fail;
1588 	if (!try_module_get(ns->ctrl->ops->module))
1589 		goto fail_put_ns;
1590 
1591 	return 0;
1592 
1593 fail_put_ns:
1594 	nvme_put_ns(ns);
1595 fail:
1596 	return -ENXIO;
1597 }
1598 
1599 static void nvme_release(struct gendisk *disk, fmode_t mode)
1600 {
1601 	struct nvme_ns *ns = disk->private_data;
1602 
1603 	module_put(ns->ctrl->ops->module);
1604 	nvme_put_ns(ns);
1605 }
1606 
1607 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1608 {
1609 	/* some standard values */
1610 	geo->heads = 1 << 6;
1611 	geo->sectors = 1 << 5;
1612 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1613 	return 0;
1614 }
1615 
1616 #ifdef CONFIG_BLK_DEV_INTEGRITY
1617 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1618 {
1619 	struct blk_integrity integrity;
1620 
1621 	memset(&integrity, 0, sizeof(integrity));
1622 	switch (pi_type) {
1623 	case NVME_NS_DPS_PI_TYPE3:
1624 		integrity.profile = &t10_pi_type3_crc;
1625 		integrity.tag_size = sizeof(u16) + sizeof(u32);
1626 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1627 		break;
1628 	case NVME_NS_DPS_PI_TYPE1:
1629 	case NVME_NS_DPS_PI_TYPE2:
1630 		integrity.profile = &t10_pi_type1_crc;
1631 		integrity.tag_size = sizeof(u16);
1632 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1633 		break;
1634 	default:
1635 		integrity.profile = NULL;
1636 		break;
1637 	}
1638 	integrity.tuple_size = ms;
1639 	blk_integrity_register(disk, &integrity);
1640 	blk_queue_max_integrity_segments(disk->queue, 1);
1641 }
1642 #else
1643 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1644 {
1645 }
1646 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1647 
1648 static void nvme_set_chunk_size(struct nvme_ns *ns)
1649 {
1650 	u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1651 	blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1652 }
1653 
1654 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1655 {
1656 	struct nvme_ctrl *ctrl = ns->ctrl;
1657 	struct request_queue *queue = disk->queue;
1658 	u32 size = queue_logical_block_size(queue);
1659 
1660 	if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1661 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1662 		return;
1663 	}
1664 
1665 	if (ctrl->nr_streams && ns->sws && ns->sgs)
1666 		size *= ns->sws * ns->sgs;
1667 
1668 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1669 			NVME_DSM_MAX_RANGES);
1670 
1671 	queue->limits.discard_alignment = 0;
1672 	queue->limits.discard_granularity = size;
1673 
1674 	/* If discard is already enabled, don't reset queue limits */
1675 	if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1676 		return;
1677 
1678 	blk_queue_max_discard_sectors(queue, UINT_MAX);
1679 	blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1680 
1681 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1682 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1683 }
1684 
1685 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1686 {
1687 	u32 max_sectors;
1688 	unsigned short bs = 1 << ns->lba_shift;
1689 
1690 	if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1691 	    (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1692 		return;
1693 	/*
1694 	 * Even though NVMe spec explicitly states that MDTS is not
1695 	 * applicable to the write-zeroes:- "The restriction does not apply to
1696 	 * commands that do not transfer data between the host and the
1697 	 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1698 	 * In order to be more cautious use controller's max_hw_sectors value
1699 	 * to configure the maximum sectors for the write-zeroes which is
1700 	 * configured based on the controller's MDTS field in the
1701 	 * nvme_init_identify() if available.
1702 	 */
1703 	if (ns->ctrl->max_hw_sectors == UINT_MAX)
1704 		max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1705 	else
1706 		max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1707 
1708 	blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1709 }
1710 
1711 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1712 		struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1713 {
1714 	int ret = 0;
1715 
1716 	memset(ids, 0, sizeof(*ids));
1717 
1718 	if (ctrl->vs >= NVME_VS(1, 1, 0))
1719 		memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1720 	if (ctrl->vs >= NVME_VS(1, 2, 0))
1721 		memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1722 	if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1723 		 /* Don't treat error as fatal we potentially
1724 		  * already have a NGUID or EUI-64
1725 		  */
1726 		ret = nvme_identify_ns_descs(ctrl, nsid, ids);
1727 		if (ret)
1728 			dev_warn(ctrl->device,
1729 				 "Identify Descriptors failed (%d)\n", ret);
1730 	}
1731 	return ret;
1732 }
1733 
1734 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1735 {
1736 	return !uuid_is_null(&ids->uuid) ||
1737 		memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1738 		memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1739 }
1740 
1741 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1742 {
1743 	return uuid_equal(&a->uuid, &b->uuid) &&
1744 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1745 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1746 }
1747 
1748 static void nvme_update_disk_info(struct gendisk *disk,
1749 		struct nvme_ns *ns, struct nvme_id_ns *id)
1750 {
1751 	sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1752 	unsigned short bs = 1 << ns->lba_shift;
1753 	u32 atomic_bs, phys_bs, io_opt;
1754 
1755 	if (ns->lba_shift > PAGE_SHIFT) {
1756 		/* unsupported block size, set capacity to 0 later */
1757 		bs = (1 << 9);
1758 	}
1759 	blk_mq_freeze_queue(disk->queue);
1760 	blk_integrity_unregister(disk);
1761 
1762 	if (id->nabo == 0) {
1763 		/*
1764 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1765 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1766 		 * 0 then AWUPF must be used instead.
1767 		 */
1768 		if (id->nsfeat & (1 << 1) && id->nawupf)
1769 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1770 		else
1771 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1772 	} else {
1773 		atomic_bs = bs;
1774 	}
1775 	phys_bs = bs;
1776 	io_opt = bs;
1777 	if (id->nsfeat & (1 << 4)) {
1778 		/* NPWG = Namespace Preferred Write Granularity */
1779 		phys_bs *= 1 + le16_to_cpu(id->npwg);
1780 		/* NOWS = Namespace Optimal Write Size */
1781 		io_opt *= 1 + le16_to_cpu(id->nows);
1782 	}
1783 
1784 	blk_queue_logical_block_size(disk->queue, bs);
1785 	/*
1786 	 * Linux filesystems assume writing a single physical block is
1787 	 * an atomic operation. Hence limit the physical block size to the
1788 	 * value of the Atomic Write Unit Power Fail parameter.
1789 	 */
1790 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1791 	blk_queue_io_min(disk->queue, phys_bs);
1792 	blk_queue_io_opt(disk->queue, io_opt);
1793 
1794 	if (ns->ms && !ns->ext &&
1795 	    (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1796 		nvme_init_integrity(disk, ns->ms, ns->pi_type);
1797 	if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1798 	    ns->lba_shift > PAGE_SHIFT)
1799 		capacity = 0;
1800 
1801 	set_capacity(disk, capacity);
1802 
1803 	nvme_config_discard(disk, ns);
1804 	nvme_config_write_zeroes(disk, ns);
1805 
1806 	if (id->nsattr & (1 << 0))
1807 		set_disk_ro(disk, true);
1808 	else
1809 		set_disk_ro(disk, false);
1810 
1811 	blk_mq_unfreeze_queue(disk->queue);
1812 }
1813 
1814 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1815 {
1816 	struct nvme_ns *ns = disk->private_data;
1817 
1818 	/*
1819 	 * If identify namespace failed, use default 512 byte block size so
1820 	 * block layer can use before failing read/write for 0 capacity.
1821 	 */
1822 	ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1823 	if (ns->lba_shift == 0)
1824 		ns->lba_shift = 9;
1825 	ns->noiob = le16_to_cpu(id->noiob);
1826 	ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1827 	ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1828 	/* the PI implementation requires metadata equal t10 pi tuple size */
1829 	if (ns->ms == sizeof(struct t10_pi_tuple))
1830 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1831 	else
1832 		ns->pi_type = 0;
1833 
1834 	if (ns->noiob)
1835 		nvme_set_chunk_size(ns);
1836 	nvme_update_disk_info(disk, ns, id);
1837 #ifdef CONFIG_NVME_MULTIPATH
1838 	if (ns->head->disk) {
1839 		nvme_update_disk_info(ns->head->disk, ns, id);
1840 		blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1841 		revalidate_disk(ns->head->disk);
1842 	}
1843 #endif
1844 }
1845 
1846 static int nvme_revalidate_disk(struct gendisk *disk)
1847 {
1848 	struct nvme_ns *ns = disk->private_data;
1849 	struct nvme_ctrl *ctrl = ns->ctrl;
1850 	struct nvme_id_ns *id;
1851 	struct nvme_ns_ids ids;
1852 	int ret = 0;
1853 
1854 	if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1855 		set_capacity(disk, 0);
1856 		return -ENODEV;
1857 	}
1858 
1859 	ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1860 	if (ret)
1861 		goto out;
1862 
1863 	if (id->ncap == 0) {
1864 		ret = -ENODEV;
1865 		goto free_id;
1866 	}
1867 
1868 	__nvme_revalidate_disk(disk, id);
1869 	ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1870 	if (ret)
1871 		goto free_id;
1872 
1873 	if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1874 		dev_err(ctrl->device,
1875 			"identifiers changed for nsid %d\n", ns->head->ns_id);
1876 		ret = -ENODEV;
1877 	}
1878 
1879 free_id:
1880 	kfree(id);
1881 out:
1882 	/*
1883 	 * Only fail the function if we got a fatal error back from the
1884 	 * device, otherwise ignore the error and just move on.
1885 	 */
1886 	if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1887 		ret = 0;
1888 	else if (ret > 0)
1889 		ret = blk_status_to_errno(nvme_error_status(ret));
1890 	return ret;
1891 }
1892 
1893 static char nvme_pr_type(enum pr_type type)
1894 {
1895 	switch (type) {
1896 	case PR_WRITE_EXCLUSIVE:
1897 		return 1;
1898 	case PR_EXCLUSIVE_ACCESS:
1899 		return 2;
1900 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1901 		return 3;
1902 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1903 		return 4;
1904 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1905 		return 5;
1906 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1907 		return 6;
1908 	default:
1909 		return 0;
1910 	}
1911 };
1912 
1913 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1914 				u64 key, u64 sa_key, u8 op)
1915 {
1916 	struct nvme_ns_head *head = NULL;
1917 	struct nvme_ns *ns;
1918 	struct nvme_command c;
1919 	int srcu_idx, ret;
1920 	u8 data[16] = { 0, };
1921 
1922 	ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1923 	if (unlikely(!ns))
1924 		return -EWOULDBLOCK;
1925 
1926 	put_unaligned_le64(key, &data[0]);
1927 	put_unaligned_le64(sa_key, &data[8]);
1928 
1929 	memset(&c, 0, sizeof(c));
1930 	c.common.opcode = op;
1931 	c.common.nsid = cpu_to_le32(ns->head->ns_id);
1932 	c.common.cdw10 = cpu_to_le32(cdw10);
1933 
1934 	ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1935 	nvme_put_ns_from_disk(head, srcu_idx);
1936 	return ret;
1937 }
1938 
1939 static int nvme_pr_register(struct block_device *bdev, u64 old,
1940 		u64 new, unsigned flags)
1941 {
1942 	u32 cdw10;
1943 
1944 	if (flags & ~PR_FL_IGNORE_KEY)
1945 		return -EOPNOTSUPP;
1946 
1947 	cdw10 = old ? 2 : 0;
1948 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1949 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1950 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1951 }
1952 
1953 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1954 		enum pr_type type, unsigned flags)
1955 {
1956 	u32 cdw10;
1957 
1958 	if (flags & ~PR_FL_IGNORE_KEY)
1959 		return -EOPNOTSUPP;
1960 
1961 	cdw10 = nvme_pr_type(type) << 8;
1962 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1963 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1964 }
1965 
1966 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1967 		enum pr_type type, bool abort)
1968 {
1969 	u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1970 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1971 }
1972 
1973 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1974 {
1975 	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1976 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1977 }
1978 
1979 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1980 {
1981 	u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1982 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1983 }
1984 
1985 static const struct pr_ops nvme_pr_ops = {
1986 	.pr_register	= nvme_pr_register,
1987 	.pr_reserve	= nvme_pr_reserve,
1988 	.pr_release	= nvme_pr_release,
1989 	.pr_preempt	= nvme_pr_preempt,
1990 	.pr_clear	= nvme_pr_clear,
1991 };
1992 
1993 #ifdef CONFIG_BLK_SED_OPAL
1994 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1995 		bool send)
1996 {
1997 	struct nvme_ctrl *ctrl = data;
1998 	struct nvme_command cmd;
1999 
2000 	memset(&cmd, 0, sizeof(cmd));
2001 	if (send)
2002 		cmd.common.opcode = nvme_admin_security_send;
2003 	else
2004 		cmd.common.opcode = nvme_admin_security_recv;
2005 	cmd.common.nsid = 0;
2006 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2007 	cmd.common.cdw11 = cpu_to_le32(len);
2008 
2009 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2010 				      ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2011 }
2012 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2013 #endif /* CONFIG_BLK_SED_OPAL */
2014 
2015 static const struct block_device_operations nvme_fops = {
2016 	.owner		= THIS_MODULE,
2017 	.ioctl		= nvme_ioctl,
2018 	.compat_ioctl	= nvme_ioctl,
2019 	.open		= nvme_open,
2020 	.release	= nvme_release,
2021 	.getgeo		= nvme_getgeo,
2022 	.revalidate_disk= nvme_revalidate_disk,
2023 	.pr_ops		= &nvme_pr_ops,
2024 };
2025 
2026 #ifdef CONFIG_NVME_MULTIPATH
2027 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2028 {
2029 	struct nvme_ns_head *head = bdev->bd_disk->private_data;
2030 
2031 	if (!kref_get_unless_zero(&head->ref))
2032 		return -ENXIO;
2033 	return 0;
2034 }
2035 
2036 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2037 {
2038 	nvme_put_ns_head(disk->private_data);
2039 }
2040 
2041 const struct block_device_operations nvme_ns_head_ops = {
2042 	.owner		= THIS_MODULE,
2043 	.open		= nvme_ns_head_open,
2044 	.release	= nvme_ns_head_release,
2045 	.ioctl		= nvme_ioctl,
2046 	.compat_ioctl	= nvme_ioctl,
2047 	.getgeo		= nvme_getgeo,
2048 	.pr_ops		= &nvme_pr_ops,
2049 };
2050 #endif /* CONFIG_NVME_MULTIPATH */
2051 
2052 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2053 {
2054 	unsigned long timeout =
2055 		((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2056 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2057 	int ret;
2058 
2059 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2060 		if (csts == ~0)
2061 			return -ENODEV;
2062 		if ((csts & NVME_CSTS_RDY) == bit)
2063 			break;
2064 
2065 		msleep(100);
2066 		if (fatal_signal_pending(current))
2067 			return -EINTR;
2068 		if (time_after(jiffies, timeout)) {
2069 			dev_err(ctrl->device,
2070 				"Device not ready; aborting %s\n", enabled ?
2071 						"initialisation" : "reset");
2072 			return -ENODEV;
2073 		}
2074 	}
2075 
2076 	return ret;
2077 }
2078 
2079 /*
2080  * If the device has been passed off to us in an enabled state, just clear
2081  * the enabled bit.  The spec says we should set the 'shutdown notification
2082  * bits', but doing so may cause the device to complete commands to the
2083  * admin queue ... and we don't know what memory that might be pointing at!
2084  */
2085 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2086 {
2087 	int ret;
2088 
2089 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2090 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2091 
2092 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2093 	if (ret)
2094 		return ret;
2095 
2096 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2097 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2098 
2099 	return nvme_wait_ready(ctrl, ctrl->cap, false);
2100 }
2101 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2102 
2103 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2104 {
2105 	/*
2106 	 * Default to a 4K page size, with the intention to update this
2107 	 * path in the future to accomodate architectures with differing
2108 	 * kernel and IO page sizes.
2109 	 */
2110 	unsigned dev_page_min, page_shift = 12;
2111 	int ret;
2112 
2113 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2114 	if (ret) {
2115 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2116 		return ret;
2117 	}
2118 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2119 
2120 	if (page_shift < dev_page_min) {
2121 		dev_err(ctrl->device,
2122 			"Minimum device page size %u too large for host (%u)\n",
2123 			1 << dev_page_min, 1 << page_shift);
2124 		return -ENODEV;
2125 	}
2126 
2127 	ctrl->page_size = 1 << page_shift;
2128 
2129 	ctrl->ctrl_config = NVME_CC_CSS_NVM;
2130 	ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2131 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2132 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2133 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2134 
2135 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2136 	if (ret)
2137 		return ret;
2138 	return nvme_wait_ready(ctrl, ctrl->cap, true);
2139 }
2140 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2141 
2142 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2143 {
2144 	unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2145 	u32 csts;
2146 	int ret;
2147 
2148 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2149 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2150 
2151 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2152 	if (ret)
2153 		return ret;
2154 
2155 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2156 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2157 			break;
2158 
2159 		msleep(100);
2160 		if (fatal_signal_pending(current))
2161 			return -EINTR;
2162 		if (time_after(jiffies, timeout)) {
2163 			dev_err(ctrl->device,
2164 				"Device shutdown incomplete; abort shutdown\n");
2165 			return -ENODEV;
2166 		}
2167 	}
2168 
2169 	return ret;
2170 }
2171 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2172 
2173 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2174 		struct request_queue *q)
2175 {
2176 	bool vwc = false;
2177 
2178 	if (ctrl->max_hw_sectors) {
2179 		u32 max_segments =
2180 			(ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2181 
2182 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
2183 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2184 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2185 	}
2186 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2187 	    is_power_of_2(ctrl->max_hw_sectors))
2188 		blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2189 	blk_queue_virt_boundary(q, ctrl->page_size - 1);
2190 	if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2191 		vwc = true;
2192 	blk_queue_write_cache(q, vwc, vwc);
2193 }
2194 
2195 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2196 {
2197 	__le64 ts;
2198 	int ret;
2199 
2200 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2201 		return 0;
2202 
2203 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2204 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2205 			NULL);
2206 	if (ret)
2207 		dev_warn_once(ctrl->device,
2208 			"could not set timestamp (%d)\n", ret);
2209 	return ret;
2210 }
2211 
2212 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2213 {
2214 	struct nvme_feat_host_behavior *host;
2215 	int ret;
2216 
2217 	/* Don't bother enabling the feature if retry delay is not reported */
2218 	if (!ctrl->crdt[0])
2219 		return 0;
2220 
2221 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2222 	if (!host)
2223 		return 0;
2224 
2225 	host->acre = NVME_ENABLE_ACRE;
2226 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2227 				host, sizeof(*host), NULL);
2228 	kfree(host);
2229 	return ret;
2230 }
2231 
2232 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2233 {
2234 	/*
2235 	 * APST (Autonomous Power State Transition) lets us program a
2236 	 * table of power state transitions that the controller will
2237 	 * perform automatically.  We configure it with a simple
2238 	 * heuristic: we are willing to spend at most 2% of the time
2239 	 * transitioning between power states.  Therefore, when running
2240 	 * in any given state, we will enter the next lower-power
2241 	 * non-operational state after waiting 50 * (enlat + exlat)
2242 	 * microseconds, as long as that state's exit latency is under
2243 	 * the requested maximum latency.
2244 	 *
2245 	 * We will not autonomously enter any non-operational state for
2246 	 * which the total latency exceeds ps_max_latency_us.  Users
2247 	 * can set ps_max_latency_us to zero to turn off APST.
2248 	 */
2249 
2250 	unsigned apste;
2251 	struct nvme_feat_auto_pst *table;
2252 	u64 max_lat_us = 0;
2253 	int max_ps = -1;
2254 	int ret;
2255 
2256 	/*
2257 	 * If APST isn't supported or if we haven't been initialized yet,
2258 	 * then don't do anything.
2259 	 */
2260 	if (!ctrl->apsta)
2261 		return 0;
2262 
2263 	if (ctrl->npss > 31) {
2264 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2265 		return 0;
2266 	}
2267 
2268 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2269 	if (!table)
2270 		return 0;
2271 
2272 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2273 		/* Turn off APST. */
2274 		apste = 0;
2275 		dev_dbg(ctrl->device, "APST disabled\n");
2276 	} else {
2277 		__le64 target = cpu_to_le64(0);
2278 		int state;
2279 
2280 		/*
2281 		 * Walk through all states from lowest- to highest-power.
2282 		 * According to the spec, lower-numbered states use more
2283 		 * power.  NPSS, despite the name, is the index of the
2284 		 * lowest-power state, not the number of states.
2285 		 */
2286 		for (state = (int)ctrl->npss; state >= 0; state--) {
2287 			u64 total_latency_us, exit_latency_us, transition_ms;
2288 
2289 			if (target)
2290 				table->entries[state] = target;
2291 
2292 			/*
2293 			 * Don't allow transitions to the deepest state
2294 			 * if it's quirked off.
2295 			 */
2296 			if (state == ctrl->npss &&
2297 			    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2298 				continue;
2299 
2300 			/*
2301 			 * Is this state a useful non-operational state for
2302 			 * higher-power states to autonomously transition to?
2303 			 */
2304 			if (!(ctrl->psd[state].flags &
2305 			      NVME_PS_FLAGS_NON_OP_STATE))
2306 				continue;
2307 
2308 			exit_latency_us =
2309 				(u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2310 			if (exit_latency_us > ctrl->ps_max_latency_us)
2311 				continue;
2312 
2313 			total_latency_us =
2314 				exit_latency_us +
2315 				le32_to_cpu(ctrl->psd[state].entry_lat);
2316 
2317 			/*
2318 			 * This state is good.  Use it as the APST idle
2319 			 * target for higher power states.
2320 			 */
2321 			transition_ms = total_latency_us + 19;
2322 			do_div(transition_ms, 20);
2323 			if (transition_ms > (1 << 24) - 1)
2324 				transition_ms = (1 << 24) - 1;
2325 
2326 			target = cpu_to_le64((state << 3) |
2327 					     (transition_ms << 8));
2328 
2329 			if (max_ps == -1)
2330 				max_ps = state;
2331 
2332 			if (total_latency_us > max_lat_us)
2333 				max_lat_us = total_latency_us;
2334 		}
2335 
2336 		apste = 1;
2337 
2338 		if (max_ps == -1) {
2339 			dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2340 		} else {
2341 			dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2342 				max_ps, max_lat_us, (int)sizeof(*table), table);
2343 		}
2344 	}
2345 
2346 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2347 				table, sizeof(*table), NULL);
2348 	if (ret)
2349 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2350 
2351 	kfree(table);
2352 	return ret;
2353 }
2354 
2355 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2356 {
2357 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2358 	u64 latency;
2359 
2360 	switch (val) {
2361 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2362 	case PM_QOS_LATENCY_ANY:
2363 		latency = U64_MAX;
2364 		break;
2365 
2366 	default:
2367 		latency = val;
2368 	}
2369 
2370 	if (ctrl->ps_max_latency_us != latency) {
2371 		ctrl->ps_max_latency_us = latency;
2372 		nvme_configure_apst(ctrl);
2373 	}
2374 }
2375 
2376 struct nvme_core_quirk_entry {
2377 	/*
2378 	 * NVMe model and firmware strings are padded with spaces.  For
2379 	 * simplicity, strings in the quirk table are padded with NULLs
2380 	 * instead.
2381 	 */
2382 	u16 vid;
2383 	const char *mn;
2384 	const char *fr;
2385 	unsigned long quirks;
2386 };
2387 
2388 static const struct nvme_core_quirk_entry core_quirks[] = {
2389 	{
2390 		/*
2391 		 * This Toshiba device seems to die using any APST states.  See:
2392 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2393 		 */
2394 		.vid = 0x1179,
2395 		.mn = "THNSF5256GPUK TOSHIBA",
2396 		.quirks = NVME_QUIRK_NO_APST,
2397 	},
2398 	{
2399 		/*
2400 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2401 		 * condition associated with actions related to suspend to idle
2402 		 * LiteON has resolved the problem in future firmware
2403 		 */
2404 		.vid = 0x14a4,
2405 		.fr = "22301111",
2406 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2407 	},
2408 	{
2409 		/*
2410 		 * This Kingston E8FK11.T firmware version has no interrupt
2411 		 * after resume with actions related to suspend to idle
2412 		 * https://bugzilla.kernel.org/show_bug.cgi?id=204887
2413 		 */
2414 		.vid = 0x2646,
2415 		.fr = "E8FK11.T",
2416 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2417 	}
2418 };
2419 
2420 /* match is null-terminated but idstr is space-padded. */
2421 static bool string_matches(const char *idstr, const char *match, size_t len)
2422 {
2423 	size_t matchlen;
2424 
2425 	if (!match)
2426 		return true;
2427 
2428 	matchlen = strlen(match);
2429 	WARN_ON_ONCE(matchlen > len);
2430 
2431 	if (memcmp(idstr, match, matchlen))
2432 		return false;
2433 
2434 	for (; matchlen < len; matchlen++)
2435 		if (idstr[matchlen] != ' ')
2436 			return false;
2437 
2438 	return true;
2439 }
2440 
2441 static bool quirk_matches(const struct nvme_id_ctrl *id,
2442 			  const struct nvme_core_quirk_entry *q)
2443 {
2444 	return q->vid == le16_to_cpu(id->vid) &&
2445 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2446 		string_matches(id->fr, q->fr, sizeof(id->fr));
2447 }
2448 
2449 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2450 		struct nvme_id_ctrl *id)
2451 {
2452 	size_t nqnlen;
2453 	int off;
2454 
2455 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2456 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2457 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2458 			strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2459 			return;
2460 		}
2461 
2462 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2463 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2464 	}
2465 
2466 	/* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2467 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2468 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2469 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2470 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2471 	off += sizeof(id->sn);
2472 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2473 	off += sizeof(id->mn);
2474 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2475 }
2476 
2477 static void nvme_release_subsystem(struct device *dev)
2478 {
2479 	struct nvme_subsystem *subsys =
2480 		container_of(dev, struct nvme_subsystem, dev);
2481 
2482 	if (subsys->instance >= 0)
2483 		ida_simple_remove(&nvme_instance_ida, subsys->instance);
2484 	kfree(subsys);
2485 }
2486 
2487 static void nvme_destroy_subsystem(struct kref *ref)
2488 {
2489 	struct nvme_subsystem *subsys =
2490 			container_of(ref, struct nvme_subsystem, ref);
2491 
2492 	mutex_lock(&nvme_subsystems_lock);
2493 	list_del(&subsys->entry);
2494 	mutex_unlock(&nvme_subsystems_lock);
2495 
2496 	ida_destroy(&subsys->ns_ida);
2497 	device_del(&subsys->dev);
2498 	put_device(&subsys->dev);
2499 }
2500 
2501 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2502 {
2503 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2504 }
2505 
2506 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2507 {
2508 	struct nvme_subsystem *subsys;
2509 
2510 	lockdep_assert_held(&nvme_subsystems_lock);
2511 
2512 	/*
2513 	 * Fail matches for discovery subsystems. This results
2514 	 * in each discovery controller bound to a unique subsystem.
2515 	 * This avoids issues with validating controller values
2516 	 * that can only be true when there is a single unique subsystem.
2517 	 * There may be multiple and completely independent entities
2518 	 * that provide discovery controllers.
2519 	 */
2520 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2521 		return NULL;
2522 
2523 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2524 		if (strcmp(subsys->subnqn, subsysnqn))
2525 			continue;
2526 		if (!kref_get_unless_zero(&subsys->ref))
2527 			continue;
2528 		return subsys;
2529 	}
2530 
2531 	return NULL;
2532 }
2533 
2534 #define SUBSYS_ATTR_RO(_name, _mode, _show)			\
2535 	struct device_attribute subsys_attr_##_name = \
2536 		__ATTR(_name, _mode, _show, NULL)
2537 
2538 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2539 				    struct device_attribute *attr,
2540 				    char *buf)
2541 {
2542 	struct nvme_subsystem *subsys =
2543 		container_of(dev, struct nvme_subsystem, dev);
2544 
2545 	return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2546 }
2547 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2548 
2549 #define nvme_subsys_show_str_function(field)				\
2550 static ssize_t subsys_##field##_show(struct device *dev,		\
2551 			    struct device_attribute *attr, char *buf)	\
2552 {									\
2553 	struct nvme_subsystem *subsys =					\
2554 		container_of(dev, struct nvme_subsystem, dev);		\
2555 	return sprintf(buf, "%.*s\n",					\
2556 		       (int)sizeof(subsys->field), subsys->field);	\
2557 }									\
2558 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2559 
2560 nvme_subsys_show_str_function(model);
2561 nvme_subsys_show_str_function(serial);
2562 nvme_subsys_show_str_function(firmware_rev);
2563 
2564 static struct attribute *nvme_subsys_attrs[] = {
2565 	&subsys_attr_model.attr,
2566 	&subsys_attr_serial.attr,
2567 	&subsys_attr_firmware_rev.attr,
2568 	&subsys_attr_subsysnqn.attr,
2569 #ifdef CONFIG_NVME_MULTIPATH
2570 	&subsys_attr_iopolicy.attr,
2571 #endif
2572 	NULL,
2573 };
2574 
2575 static struct attribute_group nvme_subsys_attrs_group = {
2576 	.attrs = nvme_subsys_attrs,
2577 };
2578 
2579 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2580 	&nvme_subsys_attrs_group,
2581 	NULL,
2582 };
2583 
2584 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2585 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2586 {
2587 	struct nvme_ctrl *tmp;
2588 
2589 	lockdep_assert_held(&nvme_subsystems_lock);
2590 
2591 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2592 		if (tmp->state == NVME_CTRL_DELETING ||
2593 		    tmp->state == NVME_CTRL_DEAD)
2594 			continue;
2595 
2596 		if (tmp->cntlid == ctrl->cntlid) {
2597 			dev_err(ctrl->device,
2598 				"Duplicate cntlid %u with %s, rejecting\n",
2599 				ctrl->cntlid, dev_name(tmp->device));
2600 			return false;
2601 		}
2602 
2603 		if ((id->cmic & (1 << 1)) ||
2604 		    (ctrl->opts && ctrl->opts->discovery_nqn))
2605 			continue;
2606 
2607 		dev_err(ctrl->device,
2608 			"Subsystem does not support multiple controllers\n");
2609 		return false;
2610 	}
2611 
2612 	return true;
2613 }
2614 
2615 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2616 {
2617 	struct nvme_subsystem *subsys, *found;
2618 	int ret;
2619 
2620 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2621 	if (!subsys)
2622 		return -ENOMEM;
2623 
2624 	subsys->instance = -1;
2625 	mutex_init(&subsys->lock);
2626 	kref_init(&subsys->ref);
2627 	INIT_LIST_HEAD(&subsys->ctrls);
2628 	INIT_LIST_HEAD(&subsys->nsheads);
2629 	nvme_init_subnqn(subsys, ctrl, id);
2630 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2631 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2632 	memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2633 	subsys->vendor_id = le16_to_cpu(id->vid);
2634 	subsys->cmic = id->cmic;
2635 	subsys->awupf = le16_to_cpu(id->awupf);
2636 #ifdef CONFIG_NVME_MULTIPATH
2637 	subsys->iopolicy = NVME_IOPOLICY_NUMA;
2638 #endif
2639 
2640 	subsys->dev.class = nvme_subsys_class;
2641 	subsys->dev.release = nvme_release_subsystem;
2642 	subsys->dev.groups = nvme_subsys_attrs_groups;
2643 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2644 	device_initialize(&subsys->dev);
2645 
2646 	mutex_lock(&nvme_subsystems_lock);
2647 	found = __nvme_find_get_subsystem(subsys->subnqn);
2648 	if (found) {
2649 		put_device(&subsys->dev);
2650 		subsys = found;
2651 
2652 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2653 			ret = -EINVAL;
2654 			goto out_put_subsystem;
2655 		}
2656 	} else {
2657 		ret = device_add(&subsys->dev);
2658 		if (ret) {
2659 			dev_err(ctrl->device,
2660 				"failed to register subsystem device.\n");
2661 			put_device(&subsys->dev);
2662 			goto out_unlock;
2663 		}
2664 		ida_init(&subsys->ns_ida);
2665 		list_add_tail(&subsys->entry, &nvme_subsystems);
2666 	}
2667 
2668 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2669 				dev_name(ctrl->device));
2670 	if (ret) {
2671 		dev_err(ctrl->device,
2672 			"failed to create sysfs link from subsystem.\n");
2673 		goto out_put_subsystem;
2674 	}
2675 
2676 	if (!found)
2677 		subsys->instance = ctrl->instance;
2678 	ctrl->subsys = subsys;
2679 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2680 	mutex_unlock(&nvme_subsystems_lock);
2681 	return 0;
2682 
2683 out_put_subsystem:
2684 	nvme_put_subsystem(subsys);
2685 out_unlock:
2686 	mutex_unlock(&nvme_subsystems_lock);
2687 	return ret;
2688 }
2689 
2690 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2691 		void *log, size_t size, u64 offset)
2692 {
2693 	struct nvme_command c = { };
2694 	unsigned long dwlen = size / 4 - 1;
2695 
2696 	c.get_log_page.opcode = nvme_admin_get_log_page;
2697 	c.get_log_page.nsid = cpu_to_le32(nsid);
2698 	c.get_log_page.lid = log_page;
2699 	c.get_log_page.lsp = lsp;
2700 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2701 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2702 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2703 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2704 
2705 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2706 }
2707 
2708 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2709 {
2710 	int ret;
2711 
2712 	if (!ctrl->effects)
2713 		ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2714 
2715 	if (!ctrl->effects)
2716 		return 0;
2717 
2718 	ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2719 			ctrl->effects, sizeof(*ctrl->effects), 0);
2720 	if (ret) {
2721 		kfree(ctrl->effects);
2722 		ctrl->effects = NULL;
2723 	}
2724 	return ret;
2725 }
2726 
2727 /*
2728  * Initialize the cached copies of the Identify data and various controller
2729  * register in our nvme_ctrl structure.  This should be called as soon as
2730  * the admin queue is fully up and running.
2731  */
2732 int nvme_init_identify(struct nvme_ctrl *ctrl)
2733 {
2734 	struct nvme_id_ctrl *id;
2735 	int ret, page_shift;
2736 	u32 max_hw_sectors;
2737 	bool prev_apst_enabled;
2738 
2739 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2740 	if (ret) {
2741 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2742 		return ret;
2743 	}
2744 	page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2745 	ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2746 
2747 	if (ctrl->vs >= NVME_VS(1, 1, 0))
2748 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2749 
2750 	ret = nvme_identify_ctrl(ctrl, &id);
2751 	if (ret) {
2752 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2753 		return -EIO;
2754 	}
2755 
2756 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2757 		ret = nvme_get_effects_log(ctrl);
2758 		if (ret < 0)
2759 			goto out_free;
2760 	}
2761 
2762 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
2763 		ctrl->cntlid = le16_to_cpu(id->cntlid);
2764 
2765 	if (!ctrl->identified) {
2766 		int i;
2767 
2768 		ret = nvme_init_subsystem(ctrl, id);
2769 		if (ret)
2770 			goto out_free;
2771 
2772 		/*
2773 		 * Check for quirks.  Quirk can depend on firmware version,
2774 		 * so, in principle, the set of quirks present can change
2775 		 * across a reset.  As a possible future enhancement, we
2776 		 * could re-scan for quirks every time we reinitialize
2777 		 * the device, but we'd have to make sure that the driver
2778 		 * behaves intelligently if the quirks change.
2779 		 */
2780 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2781 			if (quirk_matches(id, &core_quirks[i]))
2782 				ctrl->quirks |= core_quirks[i].quirks;
2783 		}
2784 	}
2785 
2786 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2787 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2788 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2789 	}
2790 
2791 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2792 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2793 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2794 
2795 	ctrl->oacs = le16_to_cpu(id->oacs);
2796 	ctrl->oncs = le16_to_cpu(id->oncs);
2797 	ctrl->mtfa = le16_to_cpu(id->mtfa);
2798 	ctrl->oaes = le32_to_cpu(id->oaes);
2799 	atomic_set(&ctrl->abort_limit, id->acl + 1);
2800 	ctrl->vwc = id->vwc;
2801 	if (id->mdts)
2802 		max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2803 	else
2804 		max_hw_sectors = UINT_MAX;
2805 	ctrl->max_hw_sectors =
2806 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2807 
2808 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
2809 	ctrl->sgls = le32_to_cpu(id->sgls);
2810 	ctrl->kas = le16_to_cpu(id->kas);
2811 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
2812 	ctrl->ctratt = le32_to_cpu(id->ctratt);
2813 
2814 	if (id->rtd3e) {
2815 		/* us -> s */
2816 		u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2817 
2818 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2819 						 shutdown_timeout, 60);
2820 
2821 		if (ctrl->shutdown_timeout != shutdown_timeout)
2822 			dev_info(ctrl->device,
2823 				 "Shutdown timeout set to %u seconds\n",
2824 				 ctrl->shutdown_timeout);
2825 	} else
2826 		ctrl->shutdown_timeout = shutdown_timeout;
2827 
2828 	ctrl->npss = id->npss;
2829 	ctrl->apsta = id->apsta;
2830 	prev_apst_enabled = ctrl->apst_enabled;
2831 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2832 		if (force_apst && id->apsta) {
2833 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2834 			ctrl->apst_enabled = true;
2835 		} else {
2836 			ctrl->apst_enabled = false;
2837 		}
2838 	} else {
2839 		ctrl->apst_enabled = id->apsta;
2840 	}
2841 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2842 
2843 	if (ctrl->ops->flags & NVME_F_FABRICS) {
2844 		ctrl->icdoff = le16_to_cpu(id->icdoff);
2845 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2846 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2847 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2848 
2849 		/*
2850 		 * In fabrics we need to verify the cntlid matches the
2851 		 * admin connect
2852 		 */
2853 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2854 			ret = -EINVAL;
2855 			goto out_free;
2856 		}
2857 
2858 		if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2859 			dev_err(ctrl->device,
2860 				"keep-alive support is mandatory for fabrics\n");
2861 			ret = -EINVAL;
2862 			goto out_free;
2863 		}
2864 	} else {
2865 		ctrl->hmpre = le32_to_cpu(id->hmpre);
2866 		ctrl->hmmin = le32_to_cpu(id->hmmin);
2867 		ctrl->hmminds = le32_to_cpu(id->hmminds);
2868 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2869 	}
2870 
2871 	ret = nvme_mpath_init(ctrl, id);
2872 	kfree(id);
2873 
2874 	if (ret < 0)
2875 		return ret;
2876 
2877 	if (ctrl->apst_enabled && !prev_apst_enabled)
2878 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
2879 	else if (!ctrl->apst_enabled && prev_apst_enabled)
2880 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
2881 
2882 	ret = nvme_configure_apst(ctrl);
2883 	if (ret < 0)
2884 		return ret;
2885 
2886 	ret = nvme_configure_timestamp(ctrl);
2887 	if (ret < 0)
2888 		return ret;
2889 
2890 	ret = nvme_configure_directives(ctrl);
2891 	if (ret < 0)
2892 		return ret;
2893 
2894 	ret = nvme_configure_acre(ctrl);
2895 	if (ret < 0)
2896 		return ret;
2897 
2898 	ctrl->identified = true;
2899 
2900 	return 0;
2901 
2902 out_free:
2903 	kfree(id);
2904 	return ret;
2905 }
2906 EXPORT_SYMBOL_GPL(nvme_init_identify);
2907 
2908 static int nvme_dev_open(struct inode *inode, struct file *file)
2909 {
2910 	struct nvme_ctrl *ctrl =
2911 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2912 
2913 	switch (ctrl->state) {
2914 	case NVME_CTRL_LIVE:
2915 		break;
2916 	default:
2917 		return -EWOULDBLOCK;
2918 	}
2919 
2920 	file->private_data = ctrl;
2921 	return 0;
2922 }
2923 
2924 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2925 {
2926 	struct nvme_ns *ns;
2927 	int ret;
2928 
2929 	down_read(&ctrl->namespaces_rwsem);
2930 	if (list_empty(&ctrl->namespaces)) {
2931 		ret = -ENOTTY;
2932 		goto out_unlock;
2933 	}
2934 
2935 	ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2936 	if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2937 		dev_warn(ctrl->device,
2938 			"NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2939 		ret = -EINVAL;
2940 		goto out_unlock;
2941 	}
2942 
2943 	dev_warn(ctrl->device,
2944 		"using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2945 	kref_get(&ns->kref);
2946 	up_read(&ctrl->namespaces_rwsem);
2947 
2948 	ret = nvme_user_cmd(ctrl, ns, argp);
2949 	nvme_put_ns(ns);
2950 	return ret;
2951 
2952 out_unlock:
2953 	up_read(&ctrl->namespaces_rwsem);
2954 	return ret;
2955 }
2956 
2957 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2958 		unsigned long arg)
2959 {
2960 	struct nvme_ctrl *ctrl = file->private_data;
2961 	void __user *argp = (void __user *)arg;
2962 
2963 	switch (cmd) {
2964 	case NVME_IOCTL_ADMIN_CMD:
2965 		return nvme_user_cmd(ctrl, NULL, argp);
2966 	case NVME_IOCTL_ADMIN64_CMD:
2967 		return nvme_user_cmd64(ctrl, NULL, argp);
2968 	case NVME_IOCTL_IO_CMD:
2969 		return nvme_dev_user_cmd(ctrl, argp);
2970 	case NVME_IOCTL_RESET:
2971 		dev_warn(ctrl->device, "resetting controller\n");
2972 		return nvme_reset_ctrl_sync(ctrl);
2973 	case NVME_IOCTL_SUBSYS_RESET:
2974 		return nvme_reset_subsystem(ctrl);
2975 	case NVME_IOCTL_RESCAN:
2976 		nvme_queue_scan(ctrl);
2977 		return 0;
2978 	default:
2979 		return -ENOTTY;
2980 	}
2981 }
2982 
2983 static const struct file_operations nvme_dev_fops = {
2984 	.owner		= THIS_MODULE,
2985 	.open		= nvme_dev_open,
2986 	.unlocked_ioctl	= nvme_dev_ioctl,
2987 	.compat_ioctl	= nvme_dev_ioctl,
2988 };
2989 
2990 static ssize_t nvme_sysfs_reset(struct device *dev,
2991 				struct device_attribute *attr, const char *buf,
2992 				size_t count)
2993 {
2994 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2995 	int ret;
2996 
2997 	ret = nvme_reset_ctrl_sync(ctrl);
2998 	if (ret < 0)
2999 		return ret;
3000 	return count;
3001 }
3002 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3003 
3004 static ssize_t nvme_sysfs_rescan(struct device *dev,
3005 				struct device_attribute *attr, const char *buf,
3006 				size_t count)
3007 {
3008 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3009 
3010 	nvme_queue_scan(ctrl);
3011 	return count;
3012 }
3013 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3014 
3015 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3016 {
3017 	struct gendisk *disk = dev_to_disk(dev);
3018 
3019 	if (disk->fops == &nvme_fops)
3020 		return nvme_get_ns_from_dev(dev)->head;
3021 	else
3022 		return disk->private_data;
3023 }
3024 
3025 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3026 		char *buf)
3027 {
3028 	struct nvme_ns_head *head = dev_to_ns_head(dev);
3029 	struct nvme_ns_ids *ids = &head->ids;
3030 	struct nvme_subsystem *subsys = head->subsys;
3031 	int serial_len = sizeof(subsys->serial);
3032 	int model_len = sizeof(subsys->model);
3033 
3034 	if (!uuid_is_null(&ids->uuid))
3035 		return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3036 
3037 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3038 		return sprintf(buf, "eui.%16phN\n", ids->nguid);
3039 
3040 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3041 		return sprintf(buf, "eui.%8phN\n", ids->eui64);
3042 
3043 	while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3044 				  subsys->serial[serial_len - 1] == '\0'))
3045 		serial_len--;
3046 	while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3047 				 subsys->model[model_len - 1] == '\0'))
3048 		model_len--;
3049 
3050 	return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3051 		serial_len, subsys->serial, model_len, subsys->model,
3052 		head->ns_id);
3053 }
3054 static DEVICE_ATTR_RO(wwid);
3055 
3056 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3057 		char *buf)
3058 {
3059 	return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3060 }
3061 static DEVICE_ATTR_RO(nguid);
3062 
3063 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3064 		char *buf)
3065 {
3066 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3067 
3068 	/* For backward compatibility expose the NGUID to userspace if
3069 	 * we have no UUID set
3070 	 */
3071 	if (uuid_is_null(&ids->uuid)) {
3072 		printk_ratelimited(KERN_WARNING
3073 				   "No UUID available providing old NGUID\n");
3074 		return sprintf(buf, "%pU\n", ids->nguid);
3075 	}
3076 	return sprintf(buf, "%pU\n", &ids->uuid);
3077 }
3078 static DEVICE_ATTR_RO(uuid);
3079 
3080 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3081 		char *buf)
3082 {
3083 	return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3084 }
3085 static DEVICE_ATTR_RO(eui);
3086 
3087 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3088 		char *buf)
3089 {
3090 	return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3091 }
3092 static DEVICE_ATTR_RO(nsid);
3093 
3094 static struct attribute *nvme_ns_id_attrs[] = {
3095 	&dev_attr_wwid.attr,
3096 	&dev_attr_uuid.attr,
3097 	&dev_attr_nguid.attr,
3098 	&dev_attr_eui.attr,
3099 	&dev_attr_nsid.attr,
3100 #ifdef CONFIG_NVME_MULTIPATH
3101 	&dev_attr_ana_grpid.attr,
3102 	&dev_attr_ana_state.attr,
3103 #endif
3104 	NULL,
3105 };
3106 
3107 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3108 		struct attribute *a, int n)
3109 {
3110 	struct device *dev = container_of(kobj, struct device, kobj);
3111 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3112 
3113 	if (a == &dev_attr_uuid.attr) {
3114 		if (uuid_is_null(&ids->uuid) &&
3115 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3116 			return 0;
3117 	}
3118 	if (a == &dev_attr_nguid.attr) {
3119 		if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3120 			return 0;
3121 	}
3122 	if (a == &dev_attr_eui.attr) {
3123 		if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3124 			return 0;
3125 	}
3126 #ifdef CONFIG_NVME_MULTIPATH
3127 	if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3128 		if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3129 			return 0;
3130 		if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3131 			return 0;
3132 	}
3133 #endif
3134 	return a->mode;
3135 }
3136 
3137 static const struct attribute_group nvme_ns_id_attr_group = {
3138 	.attrs		= nvme_ns_id_attrs,
3139 	.is_visible	= nvme_ns_id_attrs_are_visible,
3140 };
3141 
3142 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3143 	&nvme_ns_id_attr_group,
3144 #ifdef CONFIG_NVM
3145 	&nvme_nvm_attr_group,
3146 #endif
3147 	NULL,
3148 };
3149 
3150 #define nvme_show_str_function(field)						\
3151 static ssize_t  field##_show(struct device *dev,				\
3152 			    struct device_attribute *attr, char *buf)		\
3153 {										\
3154         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3155         return sprintf(buf, "%.*s\n",						\
3156 		(int)sizeof(ctrl->subsys->field), ctrl->subsys->field);		\
3157 }										\
3158 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3159 
3160 nvme_show_str_function(model);
3161 nvme_show_str_function(serial);
3162 nvme_show_str_function(firmware_rev);
3163 
3164 #define nvme_show_int_function(field)						\
3165 static ssize_t  field##_show(struct device *dev,				\
3166 			    struct device_attribute *attr, char *buf)		\
3167 {										\
3168         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3169         return sprintf(buf, "%d\n", ctrl->field);	\
3170 }										\
3171 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3172 
3173 nvme_show_int_function(cntlid);
3174 nvme_show_int_function(numa_node);
3175 nvme_show_int_function(queue_count);
3176 nvme_show_int_function(sqsize);
3177 
3178 static ssize_t nvme_sysfs_delete(struct device *dev,
3179 				struct device_attribute *attr, const char *buf,
3180 				size_t count)
3181 {
3182 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3183 
3184 	if (device_remove_file_self(dev, attr))
3185 		nvme_delete_ctrl_sync(ctrl);
3186 	return count;
3187 }
3188 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3189 
3190 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3191 					 struct device_attribute *attr,
3192 					 char *buf)
3193 {
3194 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3195 
3196 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3197 }
3198 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3199 
3200 static ssize_t nvme_sysfs_show_state(struct device *dev,
3201 				     struct device_attribute *attr,
3202 				     char *buf)
3203 {
3204 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3205 	static const char *const state_name[] = {
3206 		[NVME_CTRL_NEW]		= "new",
3207 		[NVME_CTRL_LIVE]	= "live",
3208 		[NVME_CTRL_RESETTING]	= "resetting",
3209 		[NVME_CTRL_CONNECTING]	= "connecting",
3210 		[NVME_CTRL_DELETING]	= "deleting",
3211 		[NVME_CTRL_DEAD]	= "dead",
3212 	};
3213 
3214 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3215 	    state_name[ctrl->state])
3216 		return sprintf(buf, "%s\n", state_name[ctrl->state]);
3217 
3218 	return sprintf(buf, "unknown state\n");
3219 }
3220 
3221 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3222 
3223 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3224 					 struct device_attribute *attr,
3225 					 char *buf)
3226 {
3227 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3228 
3229 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3230 }
3231 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3232 
3233 static ssize_t nvme_sysfs_show_address(struct device *dev,
3234 					 struct device_attribute *attr,
3235 					 char *buf)
3236 {
3237 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3238 
3239 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3240 }
3241 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3242 
3243 static struct attribute *nvme_dev_attrs[] = {
3244 	&dev_attr_reset_controller.attr,
3245 	&dev_attr_rescan_controller.attr,
3246 	&dev_attr_model.attr,
3247 	&dev_attr_serial.attr,
3248 	&dev_attr_firmware_rev.attr,
3249 	&dev_attr_cntlid.attr,
3250 	&dev_attr_delete_controller.attr,
3251 	&dev_attr_transport.attr,
3252 	&dev_attr_subsysnqn.attr,
3253 	&dev_attr_address.attr,
3254 	&dev_attr_state.attr,
3255 	&dev_attr_numa_node.attr,
3256 	&dev_attr_queue_count.attr,
3257 	&dev_attr_sqsize.attr,
3258 	NULL
3259 };
3260 
3261 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3262 		struct attribute *a, int n)
3263 {
3264 	struct device *dev = container_of(kobj, struct device, kobj);
3265 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3266 
3267 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3268 		return 0;
3269 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3270 		return 0;
3271 
3272 	return a->mode;
3273 }
3274 
3275 static struct attribute_group nvme_dev_attrs_group = {
3276 	.attrs		= nvme_dev_attrs,
3277 	.is_visible	= nvme_dev_attrs_are_visible,
3278 };
3279 
3280 static const struct attribute_group *nvme_dev_attr_groups[] = {
3281 	&nvme_dev_attrs_group,
3282 	NULL,
3283 };
3284 
3285 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3286 		unsigned nsid)
3287 {
3288 	struct nvme_ns_head *h;
3289 
3290 	lockdep_assert_held(&subsys->lock);
3291 
3292 	list_for_each_entry(h, &subsys->nsheads, entry) {
3293 		if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3294 			return h;
3295 	}
3296 
3297 	return NULL;
3298 }
3299 
3300 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3301 		struct nvme_ns_head *new)
3302 {
3303 	struct nvme_ns_head *h;
3304 
3305 	lockdep_assert_held(&subsys->lock);
3306 
3307 	list_for_each_entry(h, &subsys->nsheads, entry) {
3308 		if (nvme_ns_ids_valid(&new->ids) &&
3309 		    !list_empty(&h->list) &&
3310 		    nvme_ns_ids_equal(&new->ids, &h->ids))
3311 			return -EINVAL;
3312 	}
3313 
3314 	return 0;
3315 }
3316 
3317 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3318 		unsigned nsid, struct nvme_id_ns *id)
3319 {
3320 	struct nvme_ns_head *head;
3321 	size_t size = sizeof(*head);
3322 	int ret = -ENOMEM;
3323 
3324 #ifdef CONFIG_NVME_MULTIPATH
3325 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3326 #endif
3327 
3328 	head = kzalloc(size, GFP_KERNEL);
3329 	if (!head)
3330 		goto out;
3331 	ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3332 	if (ret < 0)
3333 		goto out_free_head;
3334 	head->instance = ret;
3335 	INIT_LIST_HEAD(&head->list);
3336 	ret = init_srcu_struct(&head->srcu);
3337 	if (ret)
3338 		goto out_ida_remove;
3339 	head->subsys = ctrl->subsys;
3340 	head->ns_id = nsid;
3341 	kref_init(&head->ref);
3342 
3343 	ret = nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3344 	if (ret)
3345 		goto out_cleanup_srcu;
3346 
3347 	ret = __nvme_check_ids(ctrl->subsys, head);
3348 	if (ret) {
3349 		dev_err(ctrl->device,
3350 			"duplicate IDs for nsid %d\n", nsid);
3351 		goto out_cleanup_srcu;
3352 	}
3353 
3354 	ret = nvme_mpath_alloc_disk(ctrl, head);
3355 	if (ret)
3356 		goto out_cleanup_srcu;
3357 
3358 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3359 
3360 	kref_get(&ctrl->subsys->ref);
3361 
3362 	return head;
3363 out_cleanup_srcu:
3364 	cleanup_srcu_struct(&head->srcu);
3365 out_ida_remove:
3366 	ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3367 out_free_head:
3368 	kfree(head);
3369 out:
3370 	if (ret > 0)
3371 		ret = blk_status_to_errno(nvme_error_status(ret));
3372 	return ERR_PTR(ret);
3373 }
3374 
3375 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3376 		struct nvme_id_ns *id)
3377 {
3378 	struct nvme_ctrl *ctrl = ns->ctrl;
3379 	bool is_shared = id->nmic & (1 << 0);
3380 	struct nvme_ns_head *head = NULL;
3381 	int ret = 0;
3382 
3383 	mutex_lock(&ctrl->subsys->lock);
3384 	if (is_shared)
3385 		head = __nvme_find_ns_head(ctrl->subsys, nsid);
3386 	if (!head) {
3387 		head = nvme_alloc_ns_head(ctrl, nsid, id);
3388 		if (IS_ERR(head)) {
3389 			ret = PTR_ERR(head);
3390 			goto out_unlock;
3391 		}
3392 	} else {
3393 		struct nvme_ns_ids ids;
3394 
3395 		ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3396 		if (ret)
3397 			goto out_unlock;
3398 
3399 		if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3400 			dev_err(ctrl->device,
3401 				"IDs don't match for shared namespace %d\n",
3402 					nsid);
3403 			ret = -EINVAL;
3404 			goto out_unlock;
3405 		}
3406 	}
3407 
3408 	list_add_tail(&ns->siblings, &head->list);
3409 	ns->head = head;
3410 
3411 out_unlock:
3412 	mutex_unlock(&ctrl->subsys->lock);
3413 	if (ret > 0)
3414 		ret = blk_status_to_errno(nvme_error_status(ret));
3415 	return ret;
3416 }
3417 
3418 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3419 {
3420 	struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3421 	struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3422 
3423 	return nsa->head->ns_id - nsb->head->ns_id;
3424 }
3425 
3426 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3427 {
3428 	struct nvme_ns *ns, *ret = NULL;
3429 
3430 	down_read(&ctrl->namespaces_rwsem);
3431 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3432 		if (ns->head->ns_id == nsid) {
3433 			if (!kref_get_unless_zero(&ns->kref))
3434 				continue;
3435 			ret = ns;
3436 			break;
3437 		}
3438 		if (ns->head->ns_id > nsid)
3439 			break;
3440 	}
3441 	up_read(&ctrl->namespaces_rwsem);
3442 	return ret;
3443 }
3444 
3445 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3446 {
3447 	struct streams_directive_params s;
3448 	int ret;
3449 
3450 	if (!ctrl->nr_streams)
3451 		return 0;
3452 
3453 	ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3454 	if (ret)
3455 		return ret;
3456 
3457 	ns->sws = le32_to_cpu(s.sws);
3458 	ns->sgs = le16_to_cpu(s.sgs);
3459 
3460 	if (ns->sws) {
3461 		unsigned int bs = 1 << ns->lba_shift;
3462 
3463 		blk_queue_io_min(ns->queue, bs * ns->sws);
3464 		if (ns->sgs)
3465 			blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3466 	}
3467 
3468 	return 0;
3469 }
3470 
3471 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3472 {
3473 	struct nvme_ns *ns;
3474 	struct gendisk *disk;
3475 	struct nvme_id_ns *id;
3476 	char disk_name[DISK_NAME_LEN];
3477 	int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3478 
3479 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3480 	if (!ns)
3481 		return -ENOMEM;
3482 
3483 	ns->queue = blk_mq_init_queue(ctrl->tagset);
3484 	if (IS_ERR(ns->queue)) {
3485 		ret = PTR_ERR(ns->queue);
3486 		goto out_free_ns;
3487 	}
3488 
3489 	if (ctrl->opts && ctrl->opts->data_digest)
3490 		ns->queue->backing_dev_info->capabilities
3491 			|= BDI_CAP_STABLE_WRITES;
3492 
3493 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3494 	if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3495 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3496 
3497 	ns->queue->queuedata = ns;
3498 	ns->ctrl = ctrl;
3499 
3500 	kref_init(&ns->kref);
3501 	ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3502 
3503 	blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3504 	nvme_set_queue_limits(ctrl, ns->queue);
3505 
3506 	ret = nvme_identify_ns(ctrl, nsid, &id);
3507 	if (ret)
3508 		goto out_free_queue;
3509 
3510 	if (id->ncap == 0) {
3511 		ret = -EINVAL;
3512 		goto out_free_id;
3513 	}
3514 
3515 	ret = nvme_init_ns_head(ns, nsid, id);
3516 	if (ret)
3517 		goto out_free_id;
3518 	nvme_setup_streams_ns(ctrl, ns);
3519 	nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3520 
3521 	disk = alloc_disk_node(0, node);
3522 	if (!disk) {
3523 		ret = -ENOMEM;
3524 		goto out_unlink_ns;
3525 	}
3526 
3527 	disk->fops = &nvme_fops;
3528 	disk->private_data = ns;
3529 	disk->queue = ns->queue;
3530 	disk->flags = flags;
3531 	memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3532 	ns->disk = disk;
3533 
3534 	__nvme_revalidate_disk(disk, id);
3535 
3536 	if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3537 		ret = nvme_nvm_register(ns, disk_name, node);
3538 		if (ret) {
3539 			dev_warn(ctrl->device, "LightNVM init failure\n");
3540 			goto out_put_disk;
3541 		}
3542 	}
3543 
3544 	down_write(&ctrl->namespaces_rwsem);
3545 	list_add_tail(&ns->list, &ctrl->namespaces);
3546 	up_write(&ctrl->namespaces_rwsem);
3547 
3548 	nvme_get_ctrl(ctrl);
3549 
3550 	device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3551 
3552 	nvme_mpath_add_disk(ns, id);
3553 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3554 	kfree(id);
3555 
3556 	return 0;
3557  out_put_disk:
3558 	put_disk(ns->disk);
3559  out_unlink_ns:
3560 	mutex_lock(&ctrl->subsys->lock);
3561 	list_del_rcu(&ns->siblings);
3562 	mutex_unlock(&ctrl->subsys->lock);
3563 	nvme_put_ns_head(ns->head);
3564  out_free_id:
3565 	kfree(id);
3566  out_free_queue:
3567 	blk_cleanup_queue(ns->queue);
3568  out_free_ns:
3569 	kfree(ns);
3570 	if (ret > 0)
3571 		ret = blk_status_to_errno(nvme_error_status(ret));
3572 	return ret;
3573 }
3574 
3575 static void nvme_ns_remove(struct nvme_ns *ns)
3576 {
3577 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3578 		return;
3579 
3580 	nvme_fault_inject_fini(&ns->fault_inject);
3581 
3582 	mutex_lock(&ns->ctrl->subsys->lock);
3583 	list_del_rcu(&ns->siblings);
3584 	mutex_unlock(&ns->ctrl->subsys->lock);
3585 	synchronize_rcu(); /* guarantee not available in head->list */
3586 	nvme_mpath_clear_current_path(ns);
3587 	synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3588 
3589 	if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3590 		del_gendisk(ns->disk);
3591 		blk_cleanup_queue(ns->queue);
3592 		if (blk_get_integrity(ns->disk))
3593 			blk_integrity_unregister(ns->disk);
3594 	}
3595 
3596 	down_write(&ns->ctrl->namespaces_rwsem);
3597 	list_del_init(&ns->list);
3598 	up_write(&ns->ctrl->namespaces_rwsem);
3599 
3600 	nvme_mpath_check_last_path(ns);
3601 	nvme_put_ns(ns);
3602 }
3603 
3604 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3605 {
3606 	struct nvme_ns *ns;
3607 
3608 	ns = nvme_find_get_ns(ctrl, nsid);
3609 	if (ns) {
3610 		if (ns->disk && revalidate_disk(ns->disk))
3611 			nvme_ns_remove(ns);
3612 		nvme_put_ns(ns);
3613 	} else
3614 		nvme_alloc_ns(ctrl, nsid);
3615 }
3616 
3617 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3618 					unsigned nsid)
3619 {
3620 	struct nvme_ns *ns, *next;
3621 	LIST_HEAD(rm_list);
3622 
3623 	down_write(&ctrl->namespaces_rwsem);
3624 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3625 		if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3626 			list_move_tail(&ns->list, &rm_list);
3627 	}
3628 	up_write(&ctrl->namespaces_rwsem);
3629 
3630 	list_for_each_entry_safe(ns, next, &rm_list, list)
3631 		nvme_ns_remove(ns);
3632 
3633 }
3634 
3635 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3636 {
3637 	struct nvme_ns *ns;
3638 	__le32 *ns_list;
3639 	unsigned i, j, nsid, prev = 0;
3640 	unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3641 	int ret = 0;
3642 
3643 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3644 	if (!ns_list)
3645 		return -ENOMEM;
3646 
3647 	for (i = 0; i < num_lists; i++) {
3648 		ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3649 		if (ret)
3650 			goto free;
3651 
3652 		for (j = 0; j < min(nn, 1024U); j++) {
3653 			nsid = le32_to_cpu(ns_list[j]);
3654 			if (!nsid)
3655 				goto out;
3656 
3657 			nvme_validate_ns(ctrl, nsid);
3658 
3659 			while (++prev < nsid) {
3660 				ns = nvme_find_get_ns(ctrl, prev);
3661 				if (ns) {
3662 					nvme_ns_remove(ns);
3663 					nvme_put_ns(ns);
3664 				}
3665 			}
3666 		}
3667 		nn -= j;
3668 	}
3669  out:
3670 	nvme_remove_invalid_namespaces(ctrl, prev);
3671  free:
3672 	kfree(ns_list);
3673 	return ret;
3674 }
3675 
3676 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3677 {
3678 	unsigned i;
3679 
3680 	for (i = 1; i <= nn; i++)
3681 		nvme_validate_ns(ctrl, i);
3682 
3683 	nvme_remove_invalid_namespaces(ctrl, nn);
3684 }
3685 
3686 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3687 {
3688 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3689 	__le32 *log;
3690 	int error;
3691 
3692 	log = kzalloc(log_size, GFP_KERNEL);
3693 	if (!log)
3694 		return;
3695 
3696 	/*
3697 	 * We need to read the log to clear the AEN, but we don't want to rely
3698 	 * on it for the changed namespace information as userspace could have
3699 	 * raced with us in reading the log page, which could cause us to miss
3700 	 * updates.
3701 	 */
3702 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3703 			log_size, 0);
3704 	if (error)
3705 		dev_warn(ctrl->device,
3706 			"reading changed ns log failed: %d\n", error);
3707 
3708 	kfree(log);
3709 }
3710 
3711 static void nvme_scan_work(struct work_struct *work)
3712 {
3713 	struct nvme_ctrl *ctrl =
3714 		container_of(work, struct nvme_ctrl, scan_work);
3715 	struct nvme_id_ctrl *id;
3716 	unsigned nn;
3717 
3718 	/* No tagset on a live ctrl means IO queues could not created */
3719 	if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3720 		return;
3721 
3722 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3723 		dev_info(ctrl->device, "rescanning namespaces.\n");
3724 		nvme_clear_changed_ns_log(ctrl);
3725 	}
3726 
3727 	if (nvme_identify_ctrl(ctrl, &id))
3728 		return;
3729 
3730 	mutex_lock(&ctrl->scan_lock);
3731 	nn = le32_to_cpu(id->nn);
3732 	if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3733 	    !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3734 		if (!nvme_scan_ns_list(ctrl, nn))
3735 			goto out_free_id;
3736 	}
3737 	nvme_scan_ns_sequential(ctrl, nn);
3738 out_free_id:
3739 	mutex_unlock(&ctrl->scan_lock);
3740 	kfree(id);
3741 	down_write(&ctrl->namespaces_rwsem);
3742 	list_sort(NULL, &ctrl->namespaces, ns_cmp);
3743 	up_write(&ctrl->namespaces_rwsem);
3744 }
3745 
3746 /*
3747  * This function iterates the namespace list unlocked to allow recovery from
3748  * controller failure. It is up to the caller to ensure the namespace list is
3749  * not modified by scan work while this function is executing.
3750  */
3751 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3752 {
3753 	struct nvme_ns *ns, *next;
3754 	LIST_HEAD(ns_list);
3755 
3756 	/*
3757 	 * make sure to requeue I/O to all namespaces as these
3758 	 * might result from the scan itself and must complete
3759 	 * for the scan_work to make progress
3760 	 */
3761 	nvme_mpath_clear_ctrl_paths(ctrl);
3762 
3763 	/* prevent racing with ns scanning */
3764 	flush_work(&ctrl->scan_work);
3765 
3766 	/*
3767 	 * The dead states indicates the controller was not gracefully
3768 	 * disconnected. In that case, we won't be able to flush any data while
3769 	 * removing the namespaces' disks; fail all the queues now to avoid
3770 	 * potentially having to clean up the failed sync later.
3771 	 */
3772 	if (ctrl->state == NVME_CTRL_DEAD)
3773 		nvme_kill_queues(ctrl);
3774 
3775 	down_write(&ctrl->namespaces_rwsem);
3776 	list_splice_init(&ctrl->namespaces, &ns_list);
3777 	up_write(&ctrl->namespaces_rwsem);
3778 
3779 	list_for_each_entry_safe(ns, next, &ns_list, list)
3780 		nvme_ns_remove(ns);
3781 }
3782 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3783 
3784 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3785 {
3786 	struct nvme_ctrl *ctrl =
3787 		container_of(dev, struct nvme_ctrl, ctrl_device);
3788 	struct nvmf_ctrl_options *opts = ctrl->opts;
3789 	int ret;
3790 
3791 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3792 	if (ret)
3793 		return ret;
3794 
3795 	if (opts) {
3796 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3797 		if (ret)
3798 			return ret;
3799 
3800 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3801 				opts->trsvcid ?: "none");
3802 		if (ret)
3803 			return ret;
3804 
3805 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3806 				opts->host_traddr ?: "none");
3807 	}
3808 	return ret;
3809 }
3810 
3811 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3812 {
3813 	char *envp[2] = { NULL, NULL };
3814 	u32 aen_result = ctrl->aen_result;
3815 
3816 	ctrl->aen_result = 0;
3817 	if (!aen_result)
3818 		return;
3819 
3820 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3821 	if (!envp[0])
3822 		return;
3823 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3824 	kfree(envp[0]);
3825 }
3826 
3827 static void nvme_async_event_work(struct work_struct *work)
3828 {
3829 	struct nvme_ctrl *ctrl =
3830 		container_of(work, struct nvme_ctrl, async_event_work);
3831 
3832 	nvme_aen_uevent(ctrl);
3833 	ctrl->ops->submit_async_event(ctrl);
3834 }
3835 
3836 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3837 {
3838 
3839 	u32 csts;
3840 
3841 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3842 		return false;
3843 
3844 	if (csts == ~0)
3845 		return false;
3846 
3847 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3848 }
3849 
3850 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3851 {
3852 	struct nvme_fw_slot_info_log *log;
3853 
3854 	log = kmalloc(sizeof(*log), GFP_KERNEL);
3855 	if (!log)
3856 		return;
3857 
3858 	if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3859 			sizeof(*log), 0))
3860 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3861 	kfree(log);
3862 }
3863 
3864 static void nvme_fw_act_work(struct work_struct *work)
3865 {
3866 	struct nvme_ctrl *ctrl = container_of(work,
3867 				struct nvme_ctrl, fw_act_work);
3868 	unsigned long fw_act_timeout;
3869 
3870 	if (ctrl->mtfa)
3871 		fw_act_timeout = jiffies +
3872 				msecs_to_jiffies(ctrl->mtfa * 100);
3873 	else
3874 		fw_act_timeout = jiffies +
3875 				msecs_to_jiffies(admin_timeout * 1000);
3876 
3877 	nvme_stop_queues(ctrl);
3878 	while (nvme_ctrl_pp_status(ctrl)) {
3879 		if (time_after(jiffies, fw_act_timeout)) {
3880 			dev_warn(ctrl->device,
3881 				"Fw activation timeout, reset controller\n");
3882 			nvme_try_sched_reset(ctrl);
3883 			return;
3884 		}
3885 		msleep(100);
3886 	}
3887 
3888 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
3889 		return;
3890 
3891 	nvme_start_queues(ctrl);
3892 	/* read FW slot information to clear the AER */
3893 	nvme_get_fw_slot_info(ctrl);
3894 }
3895 
3896 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3897 {
3898 	u32 aer_notice_type = (result & 0xff00) >> 8;
3899 
3900 	trace_nvme_async_event(ctrl, aer_notice_type);
3901 
3902 	switch (aer_notice_type) {
3903 	case NVME_AER_NOTICE_NS_CHANGED:
3904 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3905 		nvme_queue_scan(ctrl);
3906 		break;
3907 	case NVME_AER_NOTICE_FW_ACT_STARTING:
3908 		/*
3909 		 * We are (ab)using the RESETTING state to prevent subsequent
3910 		 * recovery actions from interfering with the controller's
3911 		 * firmware activation.
3912 		 */
3913 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
3914 			queue_work(nvme_wq, &ctrl->fw_act_work);
3915 		break;
3916 #ifdef CONFIG_NVME_MULTIPATH
3917 	case NVME_AER_NOTICE_ANA:
3918 		if (!ctrl->ana_log_buf)
3919 			break;
3920 		queue_work(nvme_wq, &ctrl->ana_work);
3921 		break;
3922 #endif
3923 	case NVME_AER_NOTICE_DISC_CHANGED:
3924 		ctrl->aen_result = result;
3925 		break;
3926 	default:
3927 		dev_warn(ctrl->device, "async event result %08x\n", result);
3928 	}
3929 }
3930 
3931 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3932 		volatile union nvme_result *res)
3933 {
3934 	u32 result = le32_to_cpu(res->u32);
3935 	u32 aer_type = result & 0x07;
3936 
3937 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3938 		return;
3939 
3940 	switch (aer_type) {
3941 	case NVME_AER_NOTICE:
3942 		nvme_handle_aen_notice(ctrl, result);
3943 		break;
3944 	case NVME_AER_ERROR:
3945 	case NVME_AER_SMART:
3946 	case NVME_AER_CSS:
3947 	case NVME_AER_VS:
3948 		trace_nvme_async_event(ctrl, aer_type);
3949 		ctrl->aen_result = result;
3950 		break;
3951 	default:
3952 		break;
3953 	}
3954 	queue_work(nvme_wq, &ctrl->async_event_work);
3955 }
3956 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3957 
3958 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3959 {
3960 	nvme_mpath_stop(ctrl);
3961 	nvme_stop_keep_alive(ctrl);
3962 	flush_work(&ctrl->async_event_work);
3963 	cancel_work_sync(&ctrl->fw_act_work);
3964 }
3965 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3966 
3967 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3968 {
3969 	if (ctrl->kato)
3970 		nvme_start_keep_alive(ctrl);
3971 
3972 	nvme_enable_aen(ctrl);
3973 
3974 	if (ctrl->queue_count > 1) {
3975 		nvme_queue_scan(ctrl);
3976 		nvme_start_queues(ctrl);
3977 	}
3978 }
3979 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3980 
3981 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3982 {
3983 	nvme_fault_inject_fini(&ctrl->fault_inject);
3984 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
3985 	cdev_device_del(&ctrl->cdev, ctrl->device);
3986 }
3987 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3988 
3989 static void nvme_free_ctrl(struct device *dev)
3990 {
3991 	struct nvme_ctrl *ctrl =
3992 		container_of(dev, struct nvme_ctrl, ctrl_device);
3993 	struct nvme_subsystem *subsys = ctrl->subsys;
3994 
3995 	if (subsys && ctrl->instance != subsys->instance)
3996 		ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3997 
3998 	kfree(ctrl->effects);
3999 	nvme_mpath_uninit(ctrl);
4000 	__free_page(ctrl->discard_page);
4001 
4002 	if (subsys) {
4003 		mutex_lock(&nvme_subsystems_lock);
4004 		list_del(&ctrl->subsys_entry);
4005 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4006 		mutex_unlock(&nvme_subsystems_lock);
4007 	}
4008 
4009 	ctrl->ops->free_ctrl(ctrl);
4010 
4011 	if (subsys)
4012 		nvme_put_subsystem(subsys);
4013 }
4014 
4015 /*
4016  * Initialize a NVMe controller structures.  This needs to be called during
4017  * earliest initialization so that we have the initialized structured around
4018  * during probing.
4019  */
4020 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4021 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
4022 {
4023 	int ret;
4024 
4025 	ctrl->state = NVME_CTRL_NEW;
4026 	spin_lock_init(&ctrl->lock);
4027 	mutex_init(&ctrl->scan_lock);
4028 	INIT_LIST_HEAD(&ctrl->namespaces);
4029 	init_rwsem(&ctrl->namespaces_rwsem);
4030 	ctrl->dev = dev;
4031 	ctrl->ops = ops;
4032 	ctrl->quirks = quirks;
4033 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4034 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4035 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4036 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4037 	init_waitqueue_head(&ctrl->state_wq);
4038 
4039 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4040 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4041 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4042 
4043 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4044 			PAGE_SIZE);
4045 	ctrl->discard_page = alloc_page(GFP_KERNEL);
4046 	if (!ctrl->discard_page) {
4047 		ret = -ENOMEM;
4048 		goto out;
4049 	}
4050 
4051 	ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4052 	if (ret < 0)
4053 		goto out;
4054 	ctrl->instance = ret;
4055 
4056 	device_initialize(&ctrl->ctrl_device);
4057 	ctrl->device = &ctrl->ctrl_device;
4058 	ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4059 	ctrl->device->class = nvme_class;
4060 	ctrl->device->parent = ctrl->dev;
4061 	ctrl->device->groups = nvme_dev_attr_groups;
4062 	ctrl->device->release = nvme_free_ctrl;
4063 	dev_set_drvdata(ctrl->device, ctrl);
4064 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4065 	if (ret)
4066 		goto out_release_instance;
4067 
4068 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
4069 	ctrl->cdev.owner = ops->module;
4070 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4071 	if (ret)
4072 		goto out_free_name;
4073 
4074 	/*
4075 	 * Initialize latency tolerance controls.  The sysfs files won't
4076 	 * be visible to userspace unless the device actually supports APST.
4077 	 */
4078 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4079 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4080 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4081 
4082 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4083 
4084 	return 0;
4085 out_free_name:
4086 	kfree_const(ctrl->device->kobj.name);
4087 out_release_instance:
4088 	ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4089 out:
4090 	if (ctrl->discard_page)
4091 		__free_page(ctrl->discard_page);
4092 	return ret;
4093 }
4094 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4095 
4096 /**
4097  * nvme_kill_queues(): Ends all namespace queues
4098  * @ctrl: the dead controller that needs to end
4099  *
4100  * Call this function when the driver determines it is unable to get the
4101  * controller in a state capable of servicing IO.
4102  */
4103 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4104 {
4105 	struct nvme_ns *ns;
4106 
4107 	down_read(&ctrl->namespaces_rwsem);
4108 
4109 	/* Forcibly unquiesce queues to avoid blocking dispatch */
4110 	if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4111 		blk_mq_unquiesce_queue(ctrl->admin_q);
4112 
4113 	list_for_each_entry(ns, &ctrl->namespaces, list)
4114 		nvme_set_queue_dying(ns);
4115 
4116 	up_read(&ctrl->namespaces_rwsem);
4117 }
4118 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4119 
4120 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4121 {
4122 	struct nvme_ns *ns;
4123 
4124 	down_read(&ctrl->namespaces_rwsem);
4125 	list_for_each_entry(ns, &ctrl->namespaces, list)
4126 		blk_mq_unfreeze_queue(ns->queue);
4127 	up_read(&ctrl->namespaces_rwsem);
4128 }
4129 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4130 
4131 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4132 {
4133 	struct nvme_ns *ns;
4134 
4135 	down_read(&ctrl->namespaces_rwsem);
4136 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4137 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4138 		if (timeout <= 0)
4139 			break;
4140 	}
4141 	up_read(&ctrl->namespaces_rwsem);
4142 }
4143 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4144 
4145 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4146 {
4147 	struct nvme_ns *ns;
4148 
4149 	down_read(&ctrl->namespaces_rwsem);
4150 	list_for_each_entry(ns, &ctrl->namespaces, list)
4151 		blk_mq_freeze_queue_wait(ns->queue);
4152 	up_read(&ctrl->namespaces_rwsem);
4153 }
4154 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4155 
4156 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4157 {
4158 	struct nvme_ns *ns;
4159 
4160 	down_read(&ctrl->namespaces_rwsem);
4161 	list_for_each_entry(ns, &ctrl->namespaces, list)
4162 		blk_freeze_queue_start(ns->queue);
4163 	up_read(&ctrl->namespaces_rwsem);
4164 }
4165 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4166 
4167 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4168 {
4169 	struct nvme_ns *ns;
4170 
4171 	down_read(&ctrl->namespaces_rwsem);
4172 	list_for_each_entry(ns, &ctrl->namespaces, list)
4173 		blk_mq_quiesce_queue(ns->queue);
4174 	up_read(&ctrl->namespaces_rwsem);
4175 }
4176 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4177 
4178 void nvme_start_queues(struct nvme_ctrl *ctrl)
4179 {
4180 	struct nvme_ns *ns;
4181 
4182 	down_read(&ctrl->namespaces_rwsem);
4183 	list_for_each_entry(ns, &ctrl->namespaces, list)
4184 		blk_mq_unquiesce_queue(ns->queue);
4185 	up_read(&ctrl->namespaces_rwsem);
4186 }
4187 EXPORT_SYMBOL_GPL(nvme_start_queues);
4188 
4189 
4190 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4191 {
4192 	struct nvme_ns *ns;
4193 
4194 	down_read(&ctrl->namespaces_rwsem);
4195 	list_for_each_entry(ns, &ctrl->namespaces, list)
4196 		blk_sync_queue(ns->queue);
4197 	up_read(&ctrl->namespaces_rwsem);
4198 
4199 	if (ctrl->admin_q)
4200 		blk_sync_queue(ctrl->admin_q);
4201 }
4202 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4203 
4204 /*
4205  * Check we didn't inadvertently grow the command structure sizes:
4206  */
4207 static inline void _nvme_check_size(void)
4208 {
4209 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4210 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4211 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4212 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4213 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4214 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4215 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4216 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4217 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4218 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4219 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4220 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4221 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4222 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4223 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4224 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4225 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4226 }
4227 
4228 
4229 static int __init nvme_core_init(void)
4230 {
4231 	int result = -ENOMEM;
4232 
4233 	_nvme_check_size();
4234 
4235 	nvme_wq = alloc_workqueue("nvme-wq",
4236 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4237 	if (!nvme_wq)
4238 		goto out;
4239 
4240 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4241 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4242 	if (!nvme_reset_wq)
4243 		goto destroy_wq;
4244 
4245 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4246 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4247 	if (!nvme_delete_wq)
4248 		goto destroy_reset_wq;
4249 
4250 	result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4251 	if (result < 0)
4252 		goto destroy_delete_wq;
4253 
4254 	nvme_class = class_create(THIS_MODULE, "nvme");
4255 	if (IS_ERR(nvme_class)) {
4256 		result = PTR_ERR(nvme_class);
4257 		goto unregister_chrdev;
4258 	}
4259 	nvme_class->dev_uevent = nvme_class_uevent;
4260 
4261 	nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4262 	if (IS_ERR(nvme_subsys_class)) {
4263 		result = PTR_ERR(nvme_subsys_class);
4264 		goto destroy_class;
4265 	}
4266 	return 0;
4267 
4268 destroy_class:
4269 	class_destroy(nvme_class);
4270 unregister_chrdev:
4271 	unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4272 destroy_delete_wq:
4273 	destroy_workqueue(nvme_delete_wq);
4274 destroy_reset_wq:
4275 	destroy_workqueue(nvme_reset_wq);
4276 destroy_wq:
4277 	destroy_workqueue(nvme_wq);
4278 out:
4279 	return result;
4280 }
4281 
4282 static void __exit nvme_core_exit(void)
4283 {
4284 	class_destroy(nvme_subsys_class);
4285 	class_destroy(nvme_class);
4286 	unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4287 	destroy_workqueue(nvme_delete_wq);
4288 	destroy_workqueue(nvme_reset_wq);
4289 	destroy_workqueue(nvme_wq);
4290 }
4291 
4292 MODULE_LICENSE("GPL");
4293 MODULE_VERSION("1.0");
4294 module_init(nvme_core_init);
4295 module_exit(nvme_core_exit);
4296