xref: /openbmc/linux/drivers/nvme/host/core.c (revision 95298d63)
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/compat.h>
10 #include <linux/delay.h>
11 #include <linux/errno.h>
12 #include <linux/hdreg.h>
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/list_sort.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
19 #include <linux/pr.h>
20 #include <linux/ptrace.h>
21 #include <linux/nvme_ioctl.h>
22 #include <linux/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 as scan, aen handling, fw activation,
70  * keep-alive, 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 }
175 
176 static void nvme_delete_ctrl_work(struct work_struct *work)
177 {
178 	struct nvme_ctrl *ctrl =
179 		container_of(work, struct nvme_ctrl, delete_work);
180 
181 	nvme_do_delete_ctrl(ctrl);
182 }
183 
184 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
185 {
186 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
187 		return -EBUSY;
188 	if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
189 		return -EBUSY;
190 	return 0;
191 }
192 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
193 
194 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
195 {
196 	/*
197 	 * Keep a reference until nvme_do_delete_ctrl() complete,
198 	 * since ->delete_ctrl can free the controller.
199 	 */
200 	nvme_get_ctrl(ctrl);
201 	if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
202 		nvme_do_delete_ctrl(ctrl);
203 	nvme_put_ctrl(ctrl);
204 }
205 
206 static blk_status_t nvme_error_status(u16 status)
207 {
208 	switch (status & 0x7ff) {
209 	case NVME_SC_SUCCESS:
210 		return BLK_STS_OK;
211 	case NVME_SC_CAP_EXCEEDED:
212 		return BLK_STS_NOSPC;
213 	case NVME_SC_LBA_RANGE:
214 	case NVME_SC_CMD_INTERRUPTED:
215 	case NVME_SC_NS_NOT_READY:
216 		return BLK_STS_TARGET;
217 	case NVME_SC_BAD_ATTRIBUTES:
218 	case NVME_SC_ONCS_NOT_SUPPORTED:
219 	case NVME_SC_INVALID_OPCODE:
220 	case NVME_SC_INVALID_FIELD:
221 	case NVME_SC_INVALID_NS:
222 		return BLK_STS_NOTSUPP;
223 	case NVME_SC_WRITE_FAULT:
224 	case NVME_SC_READ_ERROR:
225 	case NVME_SC_UNWRITTEN_BLOCK:
226 	case NVME_SC_ACCESS_DENIED:
227 	case NVME_SC_READ_ONLY:
228 	case NVME_SC_COMPARE_FAILED:
229 		return BLK_STS_MEDIUM;
230 	case NVME_SC_GUARD_CHECK:
231 	case NVME_SC_APPTAG_CHECK:
232 	case NVME_SC_REFTAG_CHECK:
233 	case NVME_SC_INVALID_PI:
234 		return BLK_STS_PROTECTION;
235 	case NVME_SC_RESERVATION_CONFLICT:
236 		return BLK_STS_NEXUS;
237 	case NVME_SC_HOST_PATH_ERROR:
238 		return BLK_STS_TRANSPORT;
239 	default:
240 		return BLK_STS_IOERR;
241 	}
242 }
243 
244 static inline bool nvme_req_needs_retry(struct request *req)
245 {
246 	if (blk_noretry_request(req))
247 		return false;
248 	if (nvme_req(req)->status & NVME_SC_DNR)
249 		return false;
250 	if (nvme_req(req)->retries >= nvme_max_retries)
251 		return false;
252 	return true;
253 }
254 
255 static void nvme_retry_req(struct request *req)
256 {
257 	struct nvme_ns *ns = req->q->queuedata;
258 	unsigned long delay = 0;
259 	u16 crd;
260 
261 	/* The mask and shift result must be <= 3 */
262 	crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
263 	if (ns && crd)
264 		delay = ns->ctrl->crdt[crd - 1] * 100;
265 
266 	nvme_req(req)->retries++;
267 	blk_mq_requeue_request(req, false);
268 	blk_mq_delay_kick_requeue_list(req->q, delay);
269 }
270 
271 void nvme_complete_rq(struct request *req)
272 {
273 	blk_status_t status = nvme_error_status(nvme_req(req)->status);
274 
275 	trace_nvme_complete_rq(req);
276 
277 	nvme_cleanup_cmd(req);
278 
279 	if (nvme_req(req)->ctrl->kas)
280 		nvme_req(req)->ctrl->comp_seen = true;
281 
282 	if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
283 		if ((req->cmd_flags & REQ_NVME_MPATH) && nvme_failover_req(req))
284 			return;
285 
286 		if (!blk_queue_dying(req->q)) {
287 			nvme_retry_req(req);
288 			return;
289 		}
290 	}
291 
292 	nvme_trace_bio_complete(req, status);
293 	blk_mq_end_request(req, status);
294 }
295 EXPORT_SYMBOL_GPL(nvme_complete_rq);
296 
297 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
298 {
299 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
300 				"Cancelling I/O %d", req->tag);
301 
302 	/* don't abort one completed request */
303 	if (blk_mq_request_completed(req))
304 		return true;
305 
306 	nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
307 	blk_mq_force_complete_rq(req);
308 	return true;
309 }
310 EXPORT_SYMBOL_GPL(nvme_cancel_request);
311 
312 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
313 		enum nvme_ctrl_state new_state)
314 {
315 	enum nvme_ctrl_state old_state;
316 	unsigned long flags;
317 	bool changed = false;
318 
319 	spin_lock_irqsave(&ctrl->lock, flags);
320 
321 	old_state = ctrl->state;
322 	switch (new_state) {
323 	case NVME_CTRL_LIVE:
324 		switch (old_state) {
325 		case NVME_CTRL_NEW:
326 		case NVME_CTRL_RESETTING:
327 		case NVME_CTRL_CONNECTING:
328 			changed = true;
329 			/* FALLTHRU */
330 		default:
331 			break;
332 		}
333 		break;
334 	case NVME_CTRL_RESETTING:
335 		switch (old_state) {
336 		case NVME_CTRL_NEW:
337 		case NVME_CTRL_LIVE:
338 			changed = true;
339 			/* FALLTHRU */
340 		default:
341 			break;
342 		}
343 		break;
344 	case NVME_CTRL_CONNECTING:
345 		switch (old_state) {
346 		case NVME_CTRL_NEW:
347 		case NVME_CTRL_RESETTING:
348 			changed = true;
349 			/* FALLTHRU */
350 		default:
351 			break;
352 		}
353 		break;
354 	case NVME_CTRL_DELETING:
355 		switch (old_state) {
356 		case NVME_CTRL_LIVE:
357 		case NVME_CTRL_RESETTING:
358 		case NVME_CTRL_CONNECTING:
359 			changed = true;
360 			/* FALLTHRU */
361 		default:
362 			break;
363 		}
364 		break;
365 	case NVME_CTRL_DEAD:
366 		switch (old_state) {
367 		case NVME_CTRL_DELETING:
368 			changed = true;
369 			/* FALLTHRU */
370 		default:
371 			break;
372 		}
373 		break;
374 	default:
375 		break;
376 	}
377 
378 	if (changed) {
379 		ctrl->state = new_state;
380 		wake_up_all(&ctrl->state_wq);
381 	}
382 
383 	spin_unlock_irqrestore(&ctrl->lock, flags);
384 	if (changed && ctrl->state == NVME_CTRL_LIVE)
385 		nvme_kick_requeue_lists(ctrl);
386 	return changed;
387 }
388 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
389 
390 /*
391  * Returns true for sink states that can't ever transition back to live.
392  */
393 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
394 {
395 	switch (ctrl->state) {
396 	case NVME_CTRL_NEW:
397 	case NVME_CTRL_LIVE:
398 	case NVME_CTRL_RESETTING:
399 	case NVME_CTRL_CONNECTING:
400 		return false;
401 	case NVME_CTRL_DELETING:
402 	case NVME_CTRL_DEAD:
403 		return true;
404 	default:
405 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
406 		return true;
407 	}
408 }
409 
410 /*
411  * Waits for the controller state to be resetting, or returns false if it is
412  * not possible to ever transition to that state.
413  */
414 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
415 {
416 	wait_event(ctrl->state_wq,
417 		   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
418 		   nvme_state_terminal(ctrl));
419 	return ctrl->state == NVME_CTRL_RESETTING;
420 }
421 EXPORT_SYMBOL_GPL(nvme_wait_reset);
422 
423 static void nvme_free_ns_head(struct kref *ref)
424 {
425 	struct nvme_ns_head *head =
426 		container_of(ref, struct nvme_ns_head, ref);
427 
428 	nvme_mpath_remove_disk(head);
429 	ida_simple_remove(&head->subsys->ns_ida, head->instance);
430 	cleanup_srcu_struct(&head->srcu);
431 	nvme_put_subsystem(head->subsys);
432 	kfree(head);
433 }
434 
435 static void nvme_put_ns_head(struct nvme_ns_head *head)
436 {
437 	kref_put(&head->ref, nvme_free_ns_head);
438 }
439 
440 static void nvme_free_ns(struct kref *kref)
441 {
442 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
443 
444 	if (ns->ndev)
445 		nvme_nvm_unregister(ns);
446 
447 	put_disk(ns->disk);
448 	nvme_put_ns_head(ns->head);
449 	nvme_put_ctrl(ns->ctrl);
450 	kfree(ns);
451 }
452 
453 static void nvme_put_ns(struct nvme_ns *ns)
454 {
455 	kref_put(&ns->kref, nvme_free_ns);
456 }
457 
458 static inline void nvme_clear_nvme_request(struct request *req)
459 {
460 	if (!(req->rq_flags & RQF_DONTPREP)) {
461 		nvme_req(req)->retries = 0;
462 		nvme_req(req)->flags = 0;
463 		req->rq_flags |= RQF_DONTPREP;
464 	}
465 }
466 
467 struct request *nvme_alloc_request(struct request_queue *q,
468 		struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
469 {
470 	unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
471 	struct request *req;
472 
473 	if (qid == NVME_QID_ANY) {
474 		req = blk_mq_alloc_request(q, op, flags);
475 	} else {
476 		req = blk_mq_alloc_request_hctx(q, op, flags,
477 				qid ? qid - 1 : 0);
478 	}
479 	if (IS_ERR(req))
480 		return req;
481 
482 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
483 	nvme_clear_nvme_request(req);
484 	nvme_req(req)->cmd = cmd;
485 
486 	return req;
487 }
488 EXPORT_SYMBOL_GPL(nvme_alloc_request);
489 
490 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
491 {
492 	struct nvme_command c;
493 
494 	memset(&c, 0, sizeof(c));
495 
496 	c.directive.opcode = nvme_admin_directive_send;
497 	c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
498 	c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
499 	c.directive.dtype = NVME_DIR_IDENTIFY;
500 	c.directive.tdtype = NVME_DIR_STREAMS;
501 	c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
502 
503 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
504 }
505 
506 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
507 {
508 	return nvme_toggle_streams(ctrl, false);
509 }
510 
511 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
512 {
513 	return nvme_toggle_streams(ctrl, true);
514 }
515 
516 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
517 				  struct streams_directive_params *s, u32 nsid)
518 {
519 	struct nvme_command c;
520 
521 	memset(&c, 0, sizeof(c));
522 	memset(s, 0, sizeof(*s));
523 
524 	c.directive.opcode = nvme_admin_directive_recv;
525 	c.directive.nsid = cpu_to_le32(nsid);
526 	c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
527 	c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
528 	c.directive.dtype = NVME_DIR_STREAMS;
529 
530 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
531 }
532 
533 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
534 {
535 	struct streams_directive_params s;
536 	int ret;
537 
538 	if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
539 		return 0;
540 	if (!streams)
541 		return 0;
542 
543 	ret = nvme_enable_streams(ctrl);
544 	if (ret)
545 		return ret;
546 
547 	ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
548 	if (ret)
549 		goto out_disable_stream;
550 
551 	ctrl->nssa = le16_to_cpu(s.nssa);
552 	if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
553 		dev_info(ctrl->device, "too few streams (%u) available\n",
554 					ctrl->nssa);
555 		goto out_disable_stream;
556 	}
557 
558 	ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
559 	dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
560 	return 0;
561 
562 out_disable_stream:
563 	nvme_disable_streams(ctrl);
564 	return ret;
565 }
566 
567 /*
568  * Check if 'req' has a write hint associated with it. If it does, assign
569  * a valid namespace stream to the write.
570  */
571 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
572 				     struct request *req, u16 *control,
573 				     u32 *dsmgmt)
574 {
575 	enum rw_hint streamid = req->write_hint;
576 
577 	if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
578 		streamid = 0;
579 	else {
580 		streamid--;
581 		if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
582 			return;
583 
584 		*control |= NVME_RW_DTYPE_STREAMS;
585 		*dsmgmt |= streamid << 16;
586 	}
587 
588 	if (streamid < ARRAY_SIZE(req->q->write_hints))
589 		req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
590 }
591 
592 static inline void nvme_setup_flush(struct nvme_ns *ns,
593 		struct nvme_command *cmnd)
594 {
595 	cmnd->common.opcode = nvme_cmd_flush;
596 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
597 }
598 
599 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
600 		struct nvme_command *cmnd)
601 {
602 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
603 	struct nvme_dsm_range *range;
604 	struct bio *bio;
605 
606 	/*
607 	 * Some devices do not consider the DSM 'Number of Ranges' field when
608 	 * determining how much data to DMA. Always allocate memory for maximum
609 	 * number of segments to prevent device reading beyond end of buffer.
610 	 */
611 	static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
612 
613 	range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
614 	if (!range) {
615 		/*
616 		 * If we fail allocation our range, fallback to the controller
617 		 * discard page. If that's also busy, it's safe to return
618 		 * busy, as we know we can make progress once that's freed.
619 		 */
620 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
621 			return BLK_STS_RESOURCE;
622 
623 		range = page_address(ns->ctrl->discard_page);
624 	}
625 
626 	__rq_for_each_bio(bio, req) {
627 		u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
628 		u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
629 
630 		if (n < segments) {
631 			range[n].cattr = cpu_to_le32(0);
632 			range[n].nlb = cpu_to_le32(nlb);
633 			range[n].slba = cpu_to_le64(slba);
634 		}
635 		n++;
636 	}
637 
638 	if (WARN_ON_ONCE(n != segments)) {
639 		if (virt_to_page(range) == ns->ctrl->discard_page)
640 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
641 		else
642 			kfree(range);
643 		return BLK_STS_IOERR;
644 	}
645 
646 	cmnd->dsm.opcode = nvme_cmd_dsm;
647 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
648 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
649 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
650 
651 	req->special_vec.bv_page = virt_to_page(range);
652 	req->special_vec.bv_offset = offset_in_page(range);
653 	req->special_vec.bv_len = alloc_size;
654 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
655 
656 	return BLK_STS_OK;
657 }
658 
659 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
660 		struct request *req, struct nvme_command *cmnd)
661 {
662 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
663 		return nvme_setup_discard(ns, req, cmnd);
664 
665 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
666 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
667 	cmnd->write_zeroes.slba =
668 		cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
669 	cmnd->write_zeroes.length =
670 		cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
671 	cmnd->write_zeroes.control = 0;
672 	return BLK_STS_OK;
673 }
674 
675 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
676 		struct request *req, struct nvme_command *cmnd)
677 {
678 	struct nvme_ctrl *ctrl = ns->ctrl;
679 	u16 control = 0;
680 	u32 dsmgmt = 0;
681 
682 	if (req->cmd_flags & REQ_FUA)
683 		control |= NVME_RW_FUA;
684 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
685 		control |= NVME_RW_LR;
686 
687 	if (req->cmd_flags & REQ_RAHEAD)
688 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
689 
690 	cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
691 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
692 	cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
693 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
694 
695 	if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
696 		nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
697 
698 	if (ns->ms) {
699 		/*
700 		 * If formated with metadata, the block layer always provides a
701 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
702 		 * we enable the PRACT bit for protection information or set the
703 		 * namespace capacity to zero to prevent any I/O.
704 		 */
705 		if (!blk_integrity_rq(req)) {
706 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
707 				return BLK_STS_NOTSUPP;
708 			control |= NVME_RW_PRINFO_PRACT;
709 		}
710 
711 		switch (ns->pi_type) {
712 		case NVME_NS_DPS_PI_TYPE3:
713 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
714 			break;
715 		case NVME_NS_DPS_PI_TYPE1:
716 		case NVME_NS_DPS_PI_TYPE2:
717 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
718 					NVME_RW_PRINFO_PRCHK_REF;
719 			cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
720 			break;
721 		}
722 	}
723 
724 	cmnd->rw.control = cpu_to_le16(control);
725 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
726 	return 0;
727 }
728 
729 void nvme_cleanup_cmd(struct request *req)
730 {
731 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
732 		struct nvme_ns *ns = req->rq_disk->private_data;
733 		struct page *page = req->special_vec.bv_page;
734 
735 		if (page == ns->ctrl->discard_page)
736 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
737 		else
738 			kfree(page_address(page) + req->special_vec.bv_offset);
739 	}
740 }
741 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
742 
743 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
744 		struct nvme_command *cmd)
745 {
746 	blk_status_t ret = BLK_STS_OK;
747 
748 	nvme_clear_nvme_request(req);
749 
750 	memset(cmd, 0, sizeof(*cmd));
751 	switch (req_op(req)) {
752 	case REQ_OP_DRV_IN:
753 	case REQ_OP_DRV_OUT:
754 		memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
755 		break;
756 	case REQ_OP_FLUSH:
757 		nvme_setup_flush(ns, cmd);
758 		break;
759 	case REQ_OP_WRITE_ZEROES:
760 		ret = nvme_setup_write_zeroes(ns, req, cmd);
761 		break;
762 	case REQ_OP_DISCARD:
763 		ret = nvme_setup_discard(ns, req, cmd);
764 		break;
765 	case REQ_OP_READ:
766 	case REQ_OP_WRITE:
767 		ret = nvme_setup_rw(ns, req, cmd);
768 		break;
769 	default:
770 		WARN_ON_ONCE(1);
771 		return BLK_STS_IOERR;
772 	}
773 
774 	cmd->common.command_id = req->tag;
775 	trace_nvme_setup_cmd(req, cmd);
776 	return ret;
777 }
778 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
779 
780 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
781 {
782 	struct completion *waiting = rq->end_io_data;
783 
784 	rq->end_io_data = NULL;
785 	complete(waiting);
786 }
787 
788 static void nvme_execute_rq_polled(struct request_queue *q,
789 		struct gendisk *bd_disk, struct request *rq, int at_head)
790 {
791 	DECLARE_COMPLETION_ONSTACK(wait);
792 
793 	WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
794 
795 	rq->cmd_flags |= REQ_HIPRI;
796 	rq->end_io_data = &wait;
797 	blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
798 
799 	while (!completion_done(&wait)) {
800 		blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
801 		cond_resched();
802 	}
803 }
804 
805 /*
806  * Returns 0 on success.  If the result is negative, it's a Linux error code;
807  * if the result is positive, it's an NVM Express status code
808  */
809 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
810 		union nvme_result *result, void *buffer, unsigned bufflen,
811 		unsigned timeout, int qid, int at_head,
812 		blk_mq_req_flags_t flags, bool poll)
813 {
814 	struct request *req;
815 	int ret;
816 
817 	req = nvme_alloc_request(q, cmd, flags, qid);
818 	if (IS_ERR(req))
819 		return PTR_ERR(req);
820 
821 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
822 
823 	if (buffer && bufflen) {
824 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
825 		if (ret)
826 			goto out;
827 	}
828 
829 	if (poll)
830 		nvme_execute_rq_polled(req->q, NULL, req, at_head);
831 	else
832 		blk_execute_rq(req->q, NULL, req, at_head);
833 	if (result)
834 		*result = nvme_req(req)->result;
835 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
836 		ret = -EINTR;
837 	else
838 		ret = nvme_req(req)->status;
839  out:
840 	blk_mq_free_request(req);
841 	return ret;
842 }
843 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
844 
845 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
846 		void *buffer, unsigned bufflen)
847 {
848 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
849 			NVME_QID_ANY, 0, 0, false);
850 }
851 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
852 
853 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
854 		unsigned len, u32 seed, bool write)
855 {
856 	struct bio_integrity_payload *bip;
857 	int ret = -ENOMEM;
858 	void *buf;
859 
860 	buf = kmalloc(len, GFP_KERNEL);
861 	if (!buf)
862 		goto out;
863 
864 	ret = -EFAULT;
865 	if (write && copy_from_user(buf, ubuf, len))
866 		goto out_free_meta;
867 
868 	bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
869 	if (IS_ERR(bip)) {
870 		ret = PTR_ERR(bip);
871 		goto out_free_meta;
872 	}
873 
874 	bip->bip_iter.bi_size = len;
875 	bip->bip_iter.bi_sector = seed;
876 	ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
877 			offset_in_page(buf));
878 	if (ret == len)
879 		return buf;
880 	ret = -ENOMEM;
881 out_free_meta:
882 	kfree(buf);
883 out:
884 	return ERR_PTR(ret);
885 }
886 
887 static int nvme_submit_user_cmd(struct request_queue *q,
888 		struct nvme_command *cmd, void __user *ubuffer,
889 		unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
890 		u32 meta_seed, u64 *result, unsigned timeout)
891 {
892 	bool write = nvme_is_write(cmd);
893 	struct nvme_ns *ns = q->queuedata;
894 	struct gendisk *disk = ns ? ns->disk : NULL;
895 	struct request *req;
896 	struct bio *bio = NULL;
897 	void *meta = NULL;
898 	int ret;
899 
900 	req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
901 	if (IS_ERR(req))
902 		return PTR_ERR(req);
903 
904 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
905 	nvme_req(req)->flags |= NVME_REQ_USERCMD;
906 
907 	if (ubuffer && bufflen) {
908 		ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
909 				GFP_KERNEL);
910 		if (ret)
911 			goto out;
912 		bio = req->bio;
913 		bio->bi_disk = disk;
914 		if (disk && meta_buffer && meta_len) {
915 			meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
916 					meta_seed, write);
917 			if (IS_ERR(meta)) {
918 				ret = PTR_ERR(meta);
919 				goto out_unmap;
920 			}
921 			req->cmd_flags |= REQ_INTEGRITY;
922 		}
923 	}
924 
925 	blk_execute_rq(req->q, disk, req, 0);
926 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
927 		ret = -EINTR;
928 	else
929 		ret = nvme_req(req)->status;
930 	if (result)
931 		*result = le64_to_cpu(nvme_req(req)->result.u64);
932 	if (meta && !ret && !write) {
933 		if (copy_to_user(meta_buffer, meta, meta_len))
934 			ret = -EFAULT;
935 	}
936 	kfree(meta);
937  out_unmap:
938 	if (bio)
939 		blk_rq_unmap_user(bio);
940  out:
941 	blk_mq_free_request(req);
942 	return ret;
943 }
944 
945 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
946 {
947 	struct nvme_ctrl *ctrl = rq->end_io_data;
948 	unsigned long flags;
949 	bool startka = false;
950 
951 	blk_mq_free_request(rq);
952 
953 	if (status) {
954 		dev_err(ctrl->device,
955 			"failed nvme_keep_alive_end_io error=%d\n",
956 				status);
957 		return;
958 	}
959 
960 	ctrl->comp_seen = false;
961 	spin_lock_irqsave(&ctrl->lock, flags);
962 	if (ctrl->state == NVME_CTRL_LIVE ||
963 	    ctrl->state == NVME_CTRL_CONNECTING)
964 		startka = true;
965 	spin_unlock_irqrestore(&ctrl->lock, flags);
966 	if (startka)
967 		queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
968 }
969 
970 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
971 {
972 	struct request *rq;
973 
974 	rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
975 			NVME_QID_ANY);
976 	if (IS_ERR(rq))
977 		return PTR_ERR(rq);
978 
979 	rq->timeout = ctrl->kato * HZ;
980 	rq->end_io_data = ctrl;
981 
982 	blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
983 
984 	return 0;
985 }
986 
987 static void nvme_keep_alive_work(struct work_struct *work)
988 {
989 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
990 			struct nvme_ctrl, ka_work);
991 	bool comp_seen = ctrl->comp_seen;
992 
993 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
994 		dev_dbg(ctrl->device,
995 			"reschedule traffic based keep-alive timer\n");
996 		ctrl->comp_seen = false;
997 		queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
998 		return;
999 	}
1000 
1001 	if (nvme_keep_alive(ctrl)) {
1002 		/* allocation failure, reset the controller */
1003 		dev_err(ctrl->device, "keep-alive failed\n");
1004 		nvme_reset_ctrl(ctrl);
1005 		return;
1006 	}
1007 }
1008 
1009 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1010 {
1011 	if (unlikely(ctrl->kato == 0))
1012 		return;
1013 
1014 	queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1015 }
1016 
1017 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1018 {
1019 	if (unlikely(ctrl->kato == 0))
1020 		return;
1021 
1022 	cancel_delayed_work_sync(&ctrl->ka_work);
1023 }
1024 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1025 
1026 /*
1027  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1028  * flag, thus sending any new CNS opcodes has a big chance of not working.
1029  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1030  * (but not for any later version).
1031  */
1032 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1033 {
1034 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1035 		return ctrl->vs < NVME_VS(1, 2, 0);
1036 	return ctrl->vs < NVME_VS(1, 1, 0);
1037 }
1038 
1039 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1040 {
1041 	struct nvme_command c = { };
1042 	int error;
1043 
1044 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1045 	c.identify.opcode = nvme_admin_identify;
1046 	c.identify.cns = NVME_ID_CNS_CTRL;
1047 
1048 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1049 	if (!*id)
1050 		return -ENOMEM;
1051 
1052 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1053 			sizeof(struct nvme_id_ctrl));
1054 	if (error)
1055 		kfree(*id);
1056 	return error;
1057 }
1058 
1059 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1060 		struct nvme_ns_id_desc *cur)
1061 {
1062 	const char *warn_str = "ctrl returned bogus length:";
1063 	void *data = cur;
1064 
1065 	switch (cur->nidt) {
1066 	case NVME_NIDT_EUI64:
1067 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1068 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1069 				 warn_str, cur->nidl);
1070 			return -1;
1071 		}
1072 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1073 		return NVME_NIDT_EUI64_LEN;
1074 	case NVME_NIDT_NGUID:
1075 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1076 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1077 				 warn_str, cur->nidl);
1078 			return -1;
1079 		}
1080 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1081 		return NVME_NIDT_NGUID_LEN;
1082 	case NVME_NIDT_UUID:
1083 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1084 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1085 				 warn_str, cur->nidl);
1086 			return -1;
1087 		}
1088 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1089 		return NVME_NIDT_UUID_LEN;
1090 	default:
1091 		/* Skip unknown types */
1092 		return cur->nidl;
1093 	}
1094 }
1095 
1096 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1097 		struct nvme_ns_ids *ids)
1098 {
1099 	struct nvme_command c = { };
1100 	int status;
1101 	void *data;
1102 	int pos;
1103 	int len;
1104 
1105 	c.identify.opcode = nvme_admin_identify;
1106 	c.identify.nsid = cpu_to_le32(nsid);
1107 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1108 
1109 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1110 	if (!data)
1111 		return -ENOMEM;
1112 
1113 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1114 				      NVME_IDENTIFY_DATA_SIZE);
1115 	if (status) {
1116 		dev_warn(ctrl->device,
1117 			"Identify Descriptors failed (%d)\n", status);
1118 		 /*
1119 		  * Don't treat an error as fatal, as we potentially already
1120 		  * have a NGUID or EUI-64.
1121 		  */
1122 		if (status > 0 && !(status & NVME_SC_DNR))
1123 			status = 0;
1124 		goto free_data;
1125 	}
1126 
1127 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1128 		struct nvme_ns_id_desc *cur = data + pos;
1129 
1130 		if (cur->nidl == 0)
1131 			break;
1132 
1133 		len = nvme_process_ns_desc(ctrl, ids, cur);
1134 		if (len < 0)
1135 			goto free_data;
1136 
1137 		len += sizeof(*cur);
1138 	}
1139 free_data:
1140 	kfree(data);
1141 	return status;
1142 }
1143 
1144 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1145 {
1146 	struct nvme_command c = { };
1147 
1148 	c.identify.opcode = nvme_admin_identify;
1149 	c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1150 	c.identify.nsid = cpu_to_le32(nsid);
1151 	return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1152 				    NVME_IDENTIFY_DATA_SIZE);
1153 }
1154 
1155 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1156 		unsigned nsid, struct nvme_id_ns **id)
1157 {
1158 	struct nvme_command c = { };
1159 	int error;
1160 
1161 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1162 	c.identify.opcode = nvme_admin_identify;
1163 	c.identify.nsid = cpu_to_le32(nsid);
1164 	c.identify.cns = NVME_ID_CNS_NS;
1165 
1166 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1167 	if (!*id)
1168 		return -ENOMEM;
1169 
1170 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1171 	if (error) {
1172 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1173 		kfree(*id);
1174 	}
1175 
1176 	return error;
1177 }
1178 
1179 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1180 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1181 {
1182 	union nvme_result res = { 0 };
1183 	struct nvme_command c;
1184 	int ret;
1185 
1186 	memset(&c, 0, sizeof(c));
1187 	c.features.opcode = op;
1188 	c.features.fid = cpu_to_le32(fid);
1189 	c.features.dword11 = cpu_to_le32(dword11);
1190 
1191 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1192 			buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1193 	if (ret >= 0 && result)
1194 		*result = le32_to_cpu(res.u32);
1195 	return ret;
1196 }
1197 
1198 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1199 		      unsigned int dword11, void *buffer, size_t buflen,
1200 		      u32 *result)
1201 {
1202 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1203 			     buflen, result);
1204 }
1205 EXPORT_SYMBOL_GPL(nvme_set_features);
1206 
1207 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1208 		      unsigned int dword11, void *buffer, size_t buflen,
1209 		      u32 *result)
1210 {
1211 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1212 			     buflen, result);
1213 }
1214 EXPORT_SYMBOL_GPL(nvme_get_features);
1215 
1216 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1217 {
1218 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1219 	u32 result;
1220 	int status, nr_io_queues;
1221 
1222 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1223 			&result);
1224 	if (status < 0)
1225 		return status;
1226 
1227 	/*
1228 	 * Degraded controllers might return an error when setting the queue
1229 	 * count.  We still want to be able to bring them online and offer
1230 	 * access to the admin queue, as that might be only way to fix them up.
1231 	 */
1232 	if (status > 0) {
1233 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1234 		*count = 0;
1235 	} else {
1236 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1237 		*count = min(*count, nr_io_queues);
1238 	}
1239 
1240 	return 0;
1241 }
1242 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1243 
1244 #define NVME_AEN_SUPPORTED \
1245 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1246 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1247 
1248 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1249 {
1250 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1251 	int status;
1252 
1253 	if (!supported_aens)
1254 		return;
1255 
1256 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1257 			NULL, 0, &result);
1258 	if (status)
1259 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1260 			 supported_aens);
1261 
1262 	queue_work(nvme_wq, &ctrl->async_event_work);
1263 }
1264 
1265 /*
1266  * Convert integer values from ioctl structures to user pointers, silently
1267  * ignoring the upper bits in the compat case to match behaviour of 32-bit
1268  * kernels.
1269  */
1270 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1271 {
1272 	if (in_compat_syscall())
1273 		ptrval = (compat_uptr_t)ptrval;
1274 	return (void __user *)ptrval;
1275 }
1276 
1277 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1278 {
1279 	struct nvme_user_io io;
1280 	struct nvme_command c;
1281 	unsigned length, meta_len;
1282 	void __user *metadata;
1283 
1284 	if (copy_from_user(&io, uio, sizeof(io)))
1285 		return -EFAULT;
1286 	if (io.flags)
1287 		return -EINVAL;
1288 
1289 	switch (io.opcode) {
1290 	case nvme_cmd_write:
1291 	case nvme_cmd_read:
1292 	case nvme_cmd_compare:
1293 		break;
1294 	default:
1295 		return -EINVAL;
1296 	}
1297 
1298 	length = (io.nblocks + 1) << ns->lba_shift;
1299 	meta_len = (io.nblocks + 1) * ns->ms;
1300 	metadata = nvme_to_user_ptr(io.metadata);
1301 
1302 	if (ns->features & NVME_NS_EXT_LBAS) {
1303 		length += meta_len;
1304 		meta_len = 0;
1305 	} else if (meta_len) {
1306 		if ((io.metadata & 3) || !io.metadata)
1307 			return -EINVAL;
1308 	}
1309 
1310 	memset(&c, 0, sizeof(c));
1311 	c.rw.opcode = io.opcode;
1312 	c.rw.flags = io.flags;
1313 	c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1314 	c.rw.slba = cpu_to_le64(io.slba);
1315 	c.rw.length = cpu_to_le16(io.nblocks);
1316 	c.rw.control = cpu_to_le16(io.control);
1317 	c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1318 	c.rw.reftag = cpu_to_le32(io.reftag);
1319 	c.rw.apptag = cpu_to_le16(io.apptag);
1320 	c.rw.appmask = cpu_to_le16(io.appmask);
1321 
1322 	return nvme_submit_user_cmd(ns->queue, &c,
1323 			nvme_to_user_ptr(io.addr), length,
1324 			metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1325 }
1326 
1327 static u32 nvme_known_admin_effects(u8 opcode)
1328 {
1329 	switch (opcode) {
1330 	case nvme_admin_format_nvm:
1331 		return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1332 					NVME_CMD_EFFECTS_CSE_MASK;
1333 	case nvme_admin_sanitize_nvm:
1334 		return NVME_CMD_EFFECTS_CSE_MASK;
1335 	default:
1336 		break;
1337 	}
1338 	return 0;
1339 }
1340 
1341 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1342 								u8 opcode)
1343 {
1344 	u32 effects = 0;
1345 
1346 	if (ns) {
1347 		if (ctrl->effects)
1348 			effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1349 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1350 			dev_warn(ctrl->device,
1351 				 "IO command:%02x has unhandled effects:%08x\n",
1352 				 opcode, effects);
1353 		return 0;
1354 	}
1355 
1356 	if (ctrl->effects)
1357 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1358 	effects |= nvme_known_admin_effects(opcode);
1359 
1360 	/*
1361 	 * For simplicity, IO to all namespaces is quiesced even if the command
1362 	 * effects say only one namespace is affected.
1363 	 */
1364 	if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1365 		mutex_lock(&ctrl->scan_lock);
1366 		mutex_lock(&ctrl->subsys->lock);
1367 		nvme_mpath_start_freeze(ctrl->subsys);
1368 		nvme_mpath_wait_freeze(ctrl->subsys);
1369 		nvme_start_freeze(ctrl);
1370 		nvme_wait_freeze(ctrl);
1371 	}
1372 	return effects;
1373 }
1374 
1375 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1376 {
1377 	struct nvme_ns *ns;
1378 
1379 	down_read(&ctrl->namespaces_rwsem);
1380 	list_for_each_entry(ns, &ctrl->namespaces, list)
1381 		if (ns->disk && nvme_revalidate_disk(ns->disk))
1382 			nvme_set_queue_dying(ns);
1383 	up_read(&ctrl->namespaces_rwsem);
1384 }
1385 
1386 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1387 {
1388 	/*
1389 	 * Revalidate LBA changes prior to unfreezing. This is necessary to
1390 	 * prevent memory corruption if a logical block size was changed by
1391 	 * this command.
1392 	 */
1393 	if (effects & NVME_CMD_EFFECTS_LBCC)
1394 		nvme_update_formats(ctrl);
1395 	if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1396 		nvme_unfreeze(ctrl);
1397 		nvme_mpath_unfreeze(ctrl->subsys);
1398 		mutex_unlock(&ctrl->subsys->lock);
1399 		nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1400 		mutex_unlock(&ctrl->scan_lock);
1401 	}
1402 	if (effects & NVME_CMD_EFFECTS_CCC)
1403 		nvme_init_identify(ctrl);
1404 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1405 		nvme_queue_scan(ctrl);
1406 		flush_work(&ctrl->scan_work);
1407 	}
1408 }
1409 
1410 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1411 			struct nvme_passthru_cmd __user *ucmd)
1412 {
1413 	struct nvme_passthru_cmd cmd;
1414 	struct nvme_command c;
1415 	unsigned timeout = 0;
1416 	u32 effects;
1417 	u64 result;
1418 	int status;
1419 
1420 	if (!capable(CAP_SYS_ADMIN))
1421 		return -EACCES;
1422 	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1423 		return -EFAULT;
1424 	if (cmd.flags)
1425 		return -EINVAL;
1426 
1427 	memset(&c, 0, sizeof(c));
1428 	c.common.opcode = cmd.opcode;
1429 	c.common.flags = cmd.flags;
1430 	c.common.nsid = cpu_to_le32(cmd.nsid);
1431 	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1432 	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1433 	c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1434 	c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1435 	c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1436 	c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1437 	c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1438 	c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1439 
1440 	if (cmd.timeout_ms)
1441 		timeout = msecs_to_jiffies(cmd.timeout_ms);
1442 
1443 	effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1444 	status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1445 			nvme_to_user_ptr(cmd.addr), cmd.data_len,
1446 			nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1447 			0, &result, timeout);
1448 	nvme_passthru_end(ctrl, effects);
1449 
1450 	if (status >= 0) {
1451 		if (put_user(result, &ucmd->result))
1452 			return -EFAULT;
1453 	}
1454 
1455 	return status;
1456 }
1457 
1458 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1459 			struct nvme_passthru_cmd64 __user *ucmd)
1460 {
1461 	struct nvme_passthru_cmd64 cmd;
1462 	struct nvme_command c;
1463 	unsigned timeout = 0;
1464 	u32 effects;
1465 	int status;
1466 
1467 	if (!capable(CAP_SYS_ADMIN))
1468 		return -EACCES;
1469 	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1470 		return -EFAULT;
1471 	if (cmd.flags)
1472 		return -EINVAL;
1473 
1474 	memset(&c, 0, sizeof(c));
1475 	c.common.opcode = cmd.opcode;
1476 	c.common.flags = cmd.flags;
1477 	c.common.nsid = cpu_to_le32(cmd.nsid);
1478 	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1479 	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1480 	c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1481 	c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1482 	c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1483 	c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1484 	c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1485 	c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1486 
1487 	if (cmd.timeout_ms)
1488 		timeout = msecs_to_jiffies(cmd.timeout_ms);
1489 
1490 	effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1491 	status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1492 			nvme_to_user_ptr(cmd.addr), cmd.data_len,
1493 			nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1494 			0, &cmd.result, timeout);
1495 	nvme_passthru_end(ctrl, effects);
1496 
1497 	if (status >= 0) {
1498 		if (put_user(cmd.result, &ucmd->result))
1499 			return -EFAULT;
1500 	}
1501 
1502 	return status;
1503 }
1504 
1505 /*
1506  * Issue ioctl requests on the first available path.  Note that unlike normal
1507  * block layer requests we will not retry failed request on another controller.
1508  */
1509 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1510 		struct nvme_ns_head **head, int *srcu_idx)
1511 {
1512 #ifdef CONFIG_NVME_MULTIPATH
1513 	if (disk->fops == &nvme_ns_head_ops) {
1514 		struct nvme_ns *ns;
1515 
1516 		*head = disk->private_data;
1517 		*srcu_idx = srcu_read_lock(&(*head)->srcu);
1518 		ns = nvme_find_path(*head);
1519 		if (!ns)
1520 			srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1521 		return ns;
1522 	}
1523 #endif
1524 	*head = NULL;
1525 	*srcu_idx = -1;
1526 	return disk->private_data;
1527 }
1528 
1529 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1530 {
1531 	if (head)
1532 		srcu_read_unlock(&head->srcu, idx);
1533 }
1534 
1535 static bool is_ctrl_ioctl(unsigned int cmd)
1536 {
1537 	if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1538 		return true;
1539 	if (is_sed_ioctl(cmd))
1540 		return true;
1541 	return false;
1542 }
1543 
1544 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1545 				  void __user *argp,
1546 				  struct nvme_ns_head *head,
1547 				  int srcu_idx)
1548 {
1549 	struct nvme_ctrl *ctrl = ns->ctrl;
1550 	int ret;
1551 
1552 	nvme_get_ctrl(ns->ctrl);
1553 	nvme_put_ns_from_disk(head, srcu_idx);
1554 
1555 	switch (cmd) {
1556 	case NVME_IOCTL_ADMIN_CMD:
1557 		ret = nvme_user_cmd(ctrl, NULL, argp);
1558 		break;
1559 	case NVME_IOCTL_ADMIN64_CMD:
1560 		ret = nvme_user_cmd64(ctrl, NULL, argp);
1561 		break;
1562 	default:
1563 		ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1564 		break;
1565 	}
1566 	nvme_put_ctrl(ctrl);
1567 	return ret;
1568 }
1569 
1570 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1571 		unsigned int cmd, unsigned long arg)
1572 {
1573 	struct nvme_ns_head *head = NULL;
1574 	void __user *argp = (void __user *)arg;
1575 	struct nvme_ns *ns;
1576 	int srcu_idx, ret;
1577 
1578 	ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1579 	if (unlikely(!ns))
1580 		return -EWOULDBLOCK;
1581 
1582 	/*
1583 	 * Handle ioctls that apply to the controller instead of the namespace
1584 	 * seperately and drop the ns SRCU reference early.  This avoids a
1585 	 * deadlock when deleting namespaces using the passthrough interface.
1586 	 */
1587 	if (is_ctrl_ioctl(cmd))
1588 		return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1589 
1590 	switch (cmd) {
1591 	case NVME_IOCTL_ID:
1592 		force_successful_syscall_return();
1593 		ret = ns->head->ns_id;
1594 		break;
1595 	case NVME_IOCTL_IO_CMD:
1596 		ret = nvme_user_cmd(ns->ctrl, ns, argp);
1597 		break;
1598 	case NVME_IOCTL_SUBMIT_IO:
1599 		ret = nvme_submit_io(ns, argp);
1600 		break;
1601 	case NVME_IOCTL_IO64_CMD:
1602 		ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1603 		break;
1604 	default:
1605 		if (ns->ndev)
1606 			ret = nvme_nvm_ioctl(ns, cmd, arg);
1607 		else
1608 			ret = -ENOTTY;
1609 	}
1610 
1611 	nvme_put_ns_from_disk(head, srcu_idx);
1612 	return ret;
1613 }
1614 
1615 #ifdef CONFIG_COMPAT
1616 struct nvme_user_io32 {
1617 	__u8	opcode;
1618 	__u8	flags;
1619 	__u16	control;
1620 	__u16	nblocks;
1621 	__u16	rsvd;
1622 	__u64	metadata;
1623 	__u64	addr;
1624 	__u64	slba;
1625 	__u32	dsmgmt;
1626 	__u32	reftag;
1627 	__u16	apptag;
1628 	__u16	appmask;
1629 } __attribute__((__packed__));
1630 
1631 #define NVME_IOCTL_SUBMIT_IO32	_IOW('N', 0x42, struct nvme_user_io32)
1632 
1633 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1634 		unsigned int cmd, unsigned long arg)
1635 {
1636 	/*
1637 	 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1638 	 * between 32 bit programs and 64 bit kernel.
1639 	 * The cause is that the results of sizeof(struct nvme_user_io),
1640 	 * which is used to define NVME_IOCTL_SUBMIT_IO,
1641 	 * are not same between 32 bit compiler and 64 bit compiler.
1642 	 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1643 	 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1644 	 * Other IOCTL numbers are same between 32 bit and 64 bit.
1645 	 * So there is nothing to do regarding to other IOCTL numbers.
1646 	 */
1647 	if (cmd == NVME_IOCTL_SUBMIT_IO32)
1648 		return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1649 
1650 	return nvme_ioctl(bdev, mode, cmd, arg);
1651 }
1652 #else
1653 #define nvme_compat_ioctl	NULL
1654 #endif /* CONFIG_COMPAT */
1655 
1656 static int nvme_open(struct block_device *bdev, fmode_t mode)
1657 {
1658 	struct nvme_ns *ns = bdev->bd_disk->private_data;
1659 
1660 #ifdef CONFIG_NVME_MULTIPATH
1661 	/* should never be called due to GENHD_FL_HIDDEN */
1662 	if (WARN_ON_ONCE(ns->head->disk))
1663 		goto fail;
1664 #endif
1665 	if (!kref_get_unless_zero(&ns->kref))
1666 		goto fail;
1667 	if (!try_module_get(ns->ctrl->ops->module))
1668 		goto fail_put_ns;
1669 
1670 	return 0;
1671 
1672 fail_put_ns:
1673 	nvme_put_ns(ns);
1674 fail:
1675 	return -ENXIO;
1676 }
1677 
1678 static void nvme_release(struct gendisk *disk, fmode_t mode)
1679 {
1680 	struct nvme_ns *ns = disk->private_data;
1681 
1682 	module_put(ns->ctrl->ops->module);
1683 	nvme_put_ns(ns);
1684 }
1685 
1686 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1687 {
1688 	/* some standard values */
1689 	geo->heads = 1 << 6;
1690 	geo->sectors = 1 << 5;
1691 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1692 	return 0;
1693 }
1694 
1695 #ifdef CONFIG_BLK_DEV_INTEGRITY
1696 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1697 				u32 max_integrity_segments)
1698 {
1699 	struct blk_integrity integrity;
1700 
1701 	memset(&integrity, 0, sizeof(integrity));
1702 	switch (pi_type) {
1703 	case NVME_NS_DPS_PI_TYPE3:
1704 		integrity.profile = &t10_pi_type3_crc;
1705 		integrity.tag_size = sizeof(u16) + sizeof(u32);
1706 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1707 		break;
1708 	case NVME_NS_DPS_PI_TYPE1:
1709 	case NVME_NS_DPS_PI_TYPE2:
1710 		integrity.profile = &t10_pi_type1_crc;
1711 		integrity.tag_size = sizeof(u16);
1712 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1713 		break;
1714 	default:
1715 		integrity.profile = NULL;
1716 		break;
1717 	}
1718 	integrity.tuple_size = ms;
1719 	blk_integrity_register(disk, &integrity);
1720 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1721 }
1722 #else
1723 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1724 				u32 max_integrity_segments)
1725 {
1726 }
1727 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1728 
1729 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1730 {
1731 	struct nvme_ctrl *ctrl = ns->ctrl;
1732 	struct request_queue *queue = disk->queue;
1733 	u32 size = queue_logical_block_size(queue);
1734 
1735 	if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1736 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1737 		return;
1738 	}
1739 
1740 	if (ctrl->nr_streams && ns->sws && ns->sgs)
1741 		size *= ns->sws * ns->sgs;
1742 
1743 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1744 			NVME_DSM_MAX_RANGES);
1745 
1746 	queue->limits.discard_alignment = 0;
1747 	queue->limits.discard_granularity = size;
1748 
1749 	/* If discard is already enabled, don't reset queue limits */
1750 	if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1751 		return;
1752 
1753 	blk_queue_max_discard_sectors(queue, UINT_MAX);
1754 	blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1755 
1756 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1757 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1758 }
1759 
1760 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1761 {
1762 	u64 max_blocks;
1763 
1764 	if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1765 	    (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1766 		return;
1767 	/*
1768 	 * Even though NVMe spec explicitly states that MDTS is not
1769 	 * applicable to the write-zeroes:- "The restriction does not apply to
1770 	 * commands that do not transfer data between the host and the
1771 	 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1772 	 * In order to be more cautious use controller's max_hw_sectors value
1773 	 * to configure the maximum sectors for the write-zeroes which is
1774 	 * configured based on the controller's MDTS field in the
1775 	 * nvme_init_identify() if available.
1776 	 */
1777 	if (ns->ctrl->max_hw_sectors == UINT_MAX)
1778 		max_blocks = (u64)USHRT_MAX + 1;
1779 	else
1780 		max_blocks = ns->ctrl->max_hw_sectors + 1;
1781 
1782 	blk_queue_max_write_zeroes_sectors(disk->queue,
1783 					   nvme_lba_to_sect(ns, max_blocks));
1784 }
1785 
1786 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1787 		struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1788 {
1789 	memset(ids, 0, sizeof(*ids));
1790 
1791 	if (ctrl->vs >= NVME_VS(1, 1, 0))
1792 		memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1793 	if (ctrl->vs >= NVME_VS(1, 2, 0))
1794 		memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1795 	if (ctrl->vs >= NVME_VS(1, 3, 0))
1796 		return nvme_identify_ns_descs(ctrl, nsid, ids);
1797 	return 0;
1798 }
1799 
1800 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1801 {
1802 	return !uuid_is_null(&ids->uuid) ||
1803 		memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1804 		memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1805 }
1806 
1807 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1808 {
1809 	return uuid_equal(&a->uuid, &b->uuid) &&
1810 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1811 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1812 }
1813 
1814 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1815 				 u32 *phys_bs, u32 *io_opt)
1816 {
1817 	struct streams_directive_params s;
1818 	int ret;
1819 
1820 	if (!ctrl->nr_streams)
1821 		return 0;
1822 
1823 	ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1824 	if (ret)
1825 		return ret;
1826 
1827 	ns->sws = le32_to_cpu(s.sws);
1828 	ns->sgs = le16_to_cpu(s.sgs);
1829 
1830 	if (ns->sws) {
1831 		*phys_bs = ns->sws * (1 << ns->lba_shift);
1832 		if (ns->sgs)
1833 			*io_opt = *phys_bs * ns->sgs;
1834 	}
1835 
1836 	return 0;
1837 }
1838 
1839 static void nvme_update_disk_info(struct gendisk *disk,
1840 		struct nvme_ns *ns, struct nvme_id_ns *id)
1841 {
1842 	sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1843 	unsigned short bs = 1 << ns->lba_shift;
1844 	u32 atomic_bs, phys_bs, io_opt = 0;
1845 
1846 	if (ns->lba_shift > PAGE_SHIFT) {
1847 		/* unsupported block size, set capacity to 0 later */
1848 		bs = (1 << 9);
1849 	}
1850 	blk_mq_freeze_queue(disk->queue);
1851 	blk_integrity_unregister(disk);
1852 
1853 	atomic_bs = phys_bs = bs;
1854 	nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1855 	if (id->nabo == 0) {
1856 		/*
1857 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1858 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1859 		 * 0 then AWUPF must be used instead.
1860 		 */
1861 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1862 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1863 		else
1864 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1865 	}
1866 
1867 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1868 		/* NPWG = Namespace Preferred Write Granularity */
1869 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1870 		/* NOWS = Namespace Optimal Write Size */
1871 		io_opt = bs * (1 + le16_to_cpu(id->nows));
1872 	}
1873 
1874 	blk_queue_logical_block_size(disk->queue, bs);
1875 	/*
1876 	 * Linux filesystems assume writing a single physical block is
1877 	 * an atomic operation. Hence limit the physical block size to the
1878 	 * value of the Atomic Write Unit Power Fail parameter.
1879 	 */
1880 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1881 	blk_queue_io_min(disk->queue, phys_bs);
1882 	blk_queue_io_opt(disk->queue, io_opt);
1883 
1884 	/*
1885 	 * The block layer can't support LBA sizes larger than the page size
1886 	 * yet, so catch this early and don't allow block I/O.
1887 	 */
1888 	if (ns->lba_shift > PAGE_SHIFT)
1889 		capacity = 0;
1890 
1891 	/*
1892 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
1893 	 * metadata masquerading as Type 0 if supported, otherwise reject block
1894 	 * I/O to namespaces with metadata except when the namespace supports
1895 	 * PI, as it can strip/insert in that case.
1896 	 */
1897 	if (ns->ms) {
1898 		if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1899 		    (ns->features & NVME_NS_METADATA_SUPPORTED))
1900 			nvme_init_integrity(disk, ns->ms, ns->pi_type,
1901 					    ns->ctrl->max_integrity_segments);
1902 		else if (!nvme_ns_has_pi(ns))
1903 			capacity = 0;
1904 	}
1905 
1906 	set_capacity_revalidate_and_notify(disk, capacity, false);
1907 
1908 	nvme_config_discard(disk, ns);
1909 	nvme_config_write_zeroes(disk, ns);
1910 
1911 	if (id->nsattr & NVME_NS_ATTR_RO)
1912 		set_disk_ro(disk, true);
1913 	else
1914 		set_disk_ro(disk, false);
1915 
1916 	blk_mq_unfreeze_queue(disk->queue);
1917 }
1918 
1919 static int __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1920 {
1921 	struct nvme_ns *ns = disk->private_data;
1922 	struct nvme_ctrl *ctrl = ns->ctrl;
1923 	u32 iob;
1924 
1925 	/*
1926 	 * If identify namespace failed, use default 512 byte block size so
1927 	 * block layer can use before failing read/write for 0 capacity.
1928 	 */
1929 	ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1930 	if (ns->lba_shift == 0)
1931 		ns->lba_shift = 9;
1932 
1933 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1934 	    is_power_of_2(ctrl->max_hw_sectors))
1935 		iob = ctrl->max_hw_sectors;
1936 	else
1937 		iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1938 
1939 	ns->features = 0;
1940 	ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1941 	/* the PI implementation requires metadata equal t10 pi tuple size */
1942 	if (ns->ms == sizeof(struct t10_pi_tuple))
1943 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1944 	else
1945 		ns->pi_type = 0;
1946 
1947 	if (ns->ms) {
1948 		/*
1949 		 * For PCIe only the separate metadata pointer is supported,
1950 		 * as the block layer supplies metadata in a separate bio_vec
1951 		 * chain. For Fabrics, only metadata as part of extended data
1952 		 * LBA is supported on the wire per the Fabrics specification,
1953 		 * but the HBA/HCA will do the remapping from the separate
1954 		 * metadata buffers for us.
1955 		 */
1956 		if (id->flbas & NVME_NS_FLBAS_META_EXT) {
1957 			ns->features |= NVME_NS_EXT_LBAS;
1958 			if ((ctrl->ops->flags & NVME_F_FABRICS) &&
1959 			    (ctrl->ops->flags & NVME_F_METADATA_SUPPORTED) &&
1960 			    ctrl->max_integrity_segments)
1961 				ns->features |= NVME_NS_METADATA_SUPPORTED;
1962 		} else {
1963 			if (WARN_ON_ONCE(ctrl->ops->flags & NVME_F_FABRICS))
1964 				return -EINVAL;
1965 			if (ctrl->ops->flags & NVME_F_METADATA_SUPPORTED)
1966 				ns->features |= NVME_NS_METADATA_SUPPORTED;
1967 		}
1968 	}
1969 
1970 	if (iob)
1971 		blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(iob));
1972 	nvme_update_disk_info(disk, ns, id);
1973 #ifdef CONFIG_NVME_MULTIPATH
1974 	if (ns->head->disk) {
1975 		nvme_update_disk_info(ns->head->disk, ns, id);
1976 		blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1977 		revalidate_disk(ns->head->disk);
1978 	}
1979 #endif
1980 	return 0;
1981 }
1982 
1983 static int nvme_revalidate_disk(struct gendisk *disk)
1984 {
1985 	struct nvme_ns *ns = disk->private_data;
1986 	struct nvme_ctrl *ctrl = ns->ctrl;
1987 	struct nvme_id_ns *id;
1988 	struct nvme_ns_ids ids;
1989 	int ret = 0;
1990 
1991 	if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1992 		set_capacity(disk, 0);
1993 		return -ENODEV;
1994 	}
1995 
1996 	ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1997 	if (ret)
1998 		goto out;
1999 
2000 	if (id->ncap == 0) {
2001 		ret = -ENODEV;
2002 		goto free_id;
2003 	}
2004 
2005 	ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
2006 	if (ret)
2007 		goto free_id;
2008 
2009 	if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
2010 		dev_err(ctrl->device,
2011 			"identifiers changed for nsid %d\n", ns->head->ns_id);
2012 		ret = -ENODEV;
2013 		goto free_id;
2014 	}
2015 
2016 	ret = __nvme_revalidate_disk(disk, id);
2017 free_id:
2018 	kfree(id);
2019 out:
2020 	/*
2021 	 * Only fail the function if we got a fatal error back from the
2022 	 * device, otherwise ignore the error and just move on.
2023 	 */
2024 	if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
2025 		ret = 0;
2026 	else if (ret > 0)
2027 		ret = blk_status_to_errno(nvme_error_status(ret));
2028 	return ret;
2029 }
2030 
2031 static char nvme_pr_type(enum pr_type type)
2032 {
2033 	switch (type) {
2034 	case PR_WRITE_EXCLUSIVE:
2035 		return 1;
2036 	case PR_EXCLUSIVE_ACCESS:
2037 		return 2;
2038 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
2039 		return 3;
2040 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2041 		return 4;
2042 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
2043 		return 5;
2044 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2045 		return 6;
2046 	default:
2047 		return 0;
2048 	}
2049 };
2050 
2051 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2052 				u64 key, u64 sa_key, u8 op)
2053 {
2054 	struct nvme_ns_head *head = NULL;
2055 	struct nvme_ns *ns;
2056 	struct nvme_command c;
2057 	int srcu_idx, ret;
2058 	u8 data[16] = { 0, };
2059 
2060 	ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2061 	if (unlikely(!ns))
2062 		return -EWOULDBLOCK;
2063 
2064 	put_unaligned_le64(key, &data[0]);
2065 	put_unaligned_le64(sa_key, &data[8]);
2066 
2067 	memset(&c, 0, sizeof(c));
2068 	c.common.opcode = op;
2069 	c.common.nsid = cpu_to_le32(ns->head->ns_id);
2070 	c.common.cdw10 = cpu_to_le32(cdw10);
2071 
2072 	ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2073 	nvme_put_ns_from_disk(head, srcu_idx);
2074 	return ret;
2075 }
2076 
2077 static int nvme_pr_register(struct block_device *bdev, u64 old,
2078 		u64 new, unsigned flags)
2079 {
2080 	u32 cdw10;
2081 
2082 	if (flags & ~PR_FL_IGNORE_KEY)
2083 		return -EOPNOTSUPP;
2084 
2085 	cdw10 = old ? 2 : 0;
2086 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2087 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2088 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2089 }
2090 
2091 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2092 		enum pr_type type, unsigned flags)
2093 {
2094 	u32 cdw10;
2095 
2096 	if (flags & ~PR_FL_IGNORE_KEY)
2097 		return -EOPNOTSUPP;
2098 
2099 	cdw10 = nvme_pr_type(type) << 8;
2100 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2101 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2102 }
2103 
2104 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2105 		enum pr_type type, bool abort)
2106 {
2107 	u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2108 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2109 }
2110 
2111 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2112 {
2113 	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2114 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2115 }
2116 
2117 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2118 {
2119 	u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2120 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2121 }
2122 
2123 static const struct pr_ops nvme_pr_ops = {
2124 	.pr_register	= nvme_pr_register,
2125 	.pr_reserve	= nvme_pr_reserve,
2126 	.pr_release	= nvme_pr_release,
2127 	.pr_preempt	= nvme_pr_preempt,
2128 	.pr_clear	= nvme_pr_clear,
2129 };
2130 
2131 #ifdef CONFIG_BLK_SED_OPAL
2132 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2133 		bool send)
2134 {
2135 	struct nvme_ctrl *ctrl = data;
2136 	struct nvme_command cmd;
2137 
2138 	memset(&cmd, 0, sizeof(cmd));
2139 	if (send)
2140 		cmd.common.opcode = nvme_admin_security_send;
2141 	else
2142 		cmd.common.opcode = nvme_admin_security_recv;
2143 	cmd.common.nsid = 0;
2144 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2145 	cmd.common.cdw11 = cpu_to_le32(len);
2146 
2147 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2148 				      ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2149 }
2150 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2151 #endif /* CONFIG_BLK_SED_OPAL */
2152 
2153 static const struct block_device_operations nvme_fops = {
2154 	.owner		= THIS_MODULE,
2155 	.ioctl		= nvme_ioctl,
2156 	.compat_ioctl	= nvme_compat_ioctl,
2157 	.open		= nvme_open,
2158 	.release	= nvme_release,
2159 	.getgeo		= nvme_getgeo,
2160 	.revalidate_disk= nvme_revalidate_disk,
2161 	.pr_ops		= &nvme_pr_ops,
2162 };
2163 
2164 #ifdef CONFIG_NVME_MULTIPATH
2165 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2166 {
2167 	struct nvme_ns_head *head = bdev->bd_disk->private_data;
2168 
2169 	if (!kref_get_unless_zero(&head->ref))
2170 		return -ENXIO;
2171 	return 0;
2172 }
2173 
2174 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2175 {
2176 	nvme_put_ns_head(disk->private_data);
2177 }
2178 
2179 const struct block_device_operations nvme_ns_head_ops = {
2180 	.owner		= THIS_MODULE,
2181 	.open		= nvme_ns_head_open,
2182 	.release	= nvme_ns_head_release,
2183 	.ioctl		= nvme_ioctl,
2184 	.compat_ioctl	= nvme_compat_ioctl,
2185 	.getgeo		= nvme_getgeo,
2186 	.pr_ops		= &nvme_pr_ops,
2187 };
2188 #endif /* CONFIG_NVME_MULTIPATH */
2189 
2190 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2191 {
2192 	unsigned long timeout =
2193 		((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2194 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2195 	int ret;
2196 
2197 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2198 		if (csts == ~0)
2199 			return -ENODEV;
2200 		if ((csts & NVME_CSTS_RDY) == bit)
2201 			break;
2202 
2203 		usleep_range(1000, 2000);
2204 		if (fatal_signal_pending(current))
2205 			return -EINTR;
2206 		if (time_after(jiffies, timeout)) {
2207 			dev_err(ctrl->device,
2208 				"Device not ready; aborting %s, CSTS=0x%x\n",
2209 				enabled ? "initialisation" : "reset", csts);
2210 			return -ENODEV;
2211 		}
2212 	}
2213 
2214 	return ret;
2215 }
2216 
2217 /*
2218  * If the device has been passed off to us in an enabled state, just clear
2219  * the enabled bit.  The spec says we should set the 'shutdown notification
2220  * bits', but doing so may cause the device to complete commands to the
2221  * admin queue ... and we don't know what memory that might be pointing at!
2222  */
2223 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2224 {
2225 	int ret;
2226 
2227 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2228 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2229 
2230 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2231 	if (ret)
2232 		return ret;
2233 
2234 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2235 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2236 
2237 	return nvme_wait_ready(ctrl, ctrl->cap, false);
2238 }
2239 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2240 
2241 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2242 {
2243 	/*
2244 	 * Default to a 4K page size, with the intention to update this
2245 	 * path in the future to accomodate architectures with differing
2246 	 * kernel and IO page sizes.
2247 	 */
2248 	unsigned dev_page_min, page_shift = 12;
2249 	int ret;
2250 
2251 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2252 	if (ret) {
2253 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2254 		return ret;
2255 	}
2256 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2257 
2258 	if (page_shift < dev_page_min) {
2259 		dev_err(ctrl->device,
2260 			"Minimum device page size %u too large for host (%u)\n",
2261 			1 << dev_page_min, 1 << page_shift);
2262 		return -ENODEV;
2263 	}
2264 
2265 	ctrl->page_size = 1 << page_shift;
2266 
2267 	ctrl->ctrl_config = NVME_CC_CSS_NVM;
2268 	ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2269 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2270 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2271 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2272 
2273 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2274 	if (ret)
2275 		return ret;
2276 	return nvme_wait_ready(ctrl, ctrl->cap, true);
2277 }
2278 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2279 
2280 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2281 {
2282 	unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2283 	u32 csts;
2284 	int ret;
2285 
2286 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2287 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2288 
2289 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2290 	if (ret)
2291 		return ret;
2292 
2293 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2294 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2295 			break;
2296 
2297 		msleep(100);
2298 		if (fatal_signal_pending(current))
2299 			return -EINTR;
2300 		if (time_after(jiffies, timeout)) {
2301 			dev_err(ctrl->device,
2302 				"Device shutdown incomplete; abort shutdown\n");
2303 			return -ENODEV;
2304 		}
2305 	}
2306 
2307 	return ret;
2308 }
2309 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2310 
2311 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2312 		struct request_queue *q)
2313 {
2314 	bool vwc = false;
2315 
2316 	if (ctrl->max_hw_sectors) {
2317 		u32 max_segments =
2318 			(ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2319 
2320 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
2321 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2322 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2323 	}
2324 	blk_queue_virt_boundary(q, ctrl->page_size - 1);
2325 	blk_queue_dma_alignment(q, 7);
2326 	if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2327 		vwc = true;
2328 	blk_queue_write_cache(q, vwc, vwc);
2329 }
2330 
2331 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2332 {
2333 	__le64 ts;
2334 	int ret;
2335 
2336 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2337 		return 0;
2338 
2339 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2340 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2341 			NULL);
2342 	if (ret)
2343 		dev_warn_once(ctrl->device,
2344 			"could not set timestamp (%d)\n", ret);
2345 	return ret;
2346 }
2347 
2348 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2349 {
2350 	struct nvme_feat_host_behavior *host;
2351 	int ret;
2352 
2353 	/* Don't bother enabling the feature if retry delay is not reported */
2354 	if (!ctrl->crdt[0])
2355 		return 0;
2356 
2357 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2358 	if (!host)
2359 		return 0;
2360 
2361 	host->acre = NVME_ENABLE_ACRE;
2362 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2363 				host, sizeof(*host), NULL);
2364 	kfree(host);
2365 	return ret;
2366 }
2367 
2368 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2369 {
2370 	/*
2371 	 * APST (Autonomous Power State Transition) lets us program a
2372 	 * table of power state transitions that the controller will
2373 	 * perform automatically.  We configure it with a simple
2374 	 * heuristic: we are willing to spend at most 2% of the time
2375 	 * transitioning between power states.  Therefore, when running
2376 	 * in any given state, we will enter the next lower-power
2377 	 * non-operational state after waiting 50 * (enlat + exlat)
2378 	 * microseconds, as long as that state's exit latency is under
2379 	 * the requested maximum latency.
2380 	 *
2381 	 * We will not autonomously enter any non-operational state for
2382 	 * which the total latency exceeds ps_max_latency_us.  Users
2383 	 * can set ps_max_latency_us to zero to turn off APST.
2384 	 */
2385 
2386 	unsigned apste;
2387 	struct nvme_feat_auto_pst *table;
2388 	u64 max_lat_us = 0;
2389 	int max_ps = -1;
2390 	int ret;
2391 
2392 	/*
2393 	 * If APST isn't supported or if we haven't been initialized yet,
2394 	 * then don't do anything.
2395 	 */
2396 	if (!ctrl->apsta)
2397 		return 0;
2398 
2399 	if (ctrl->npss > 31) {
2400 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2401 		return 0;
2402 	}
2403 
2404 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2405 	if (!table)
2406 		return 0;
2407 
2408 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2409 		/* Turn off APST. */
2410 		apste = 0;
2411 		dev_dbg(ctrl->device, "APST disabled\n");
2412 	} else {
2413 		__le64 target = cpu_to_le64(0);
2414 		int state;
2415 
2416 		/*
2417 		 * Walk through all states from lowest- to highest-power.
2418 		 * According to the spec, lower-numbered states use more
2419 		 * power.  NPSS, despite the name, is the index of the
2420 		 * lowest-power state, not the number of states.
2421 		 */
2422 		for (state = (int)ctrl->npss; state >= 0; state--) {
2423 			u64 total_latency_us, exit_latency_us, transition_ms;
2424 
2425 			if (target)
2426 				table->entries[state] = target;
2427 
2428 			/*
2429 			 * Don't allow transitions to the deepest state
2430 			 * if it's quirked off.
2431 			 */
2432 			if (state == ctrl->npss &&
2433 			    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2434 				continue;
2435 
2436 			/*
2437 			 * Is this state a useful non-operational state for
2438 			 * higher-power states to autonomously transition to?
2439 			 */
2440 			if (!(ctrl->psd[state].flags &
2441 			      NVME_PS_FLAGS_NON_OP_STATE))
2442 				continue;
2443 
2444 			exit_latency_us =
2445 				(u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2446 			if (exit_latency_us > ctrl->ps_max_latency_us)
2447 				continue;
2448 
2449 			total_latency_us =
2450 				exit_latency_us +
2451 				le32_to_cpu(ctrl->psd[state].entry_lat);
2452 
2453 			/*
2454 			 * This state is good.  Use it as the APST idle
2455 			 * target for higher power states.
2456 			 */
2457 			transition_ms = total_latency_us + 19;
2458 			do_div(transition_ms, 20);
2459 			if (transition_ms > (1 << 24) - 1)
2460 				transition_ms = (1 << 24) - 1;
2461 
2462 			target = cpu_to_le64((state << 3) |
2463 					     (transition_ms << 8));
2464 
2465 			if (max_ps == -1)
2466 				max_ps = state;
2467 
2468 			if (total_latency_us > max_lat_us)
2469 				max_lat_us = total_latency_us;
2470 		}
2471 
2472 		apste = 1;
2473 
2474 		if (max_ps == -1) {
2475 			dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2476 		} else {
2477 			dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2478 				max_ps, max_lat_us, (int)sizeof(*table), table);
2479 		}
2480 	}
2481 
2482 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2483 				table, sizeof(*table), NULL);
2484 	if (ret)
2485 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2486 
2487 	kfree(table);
2488 	return ret;
2489 }
2490 
2491 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2492 {
2493 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2494 	u64 latency;
2495 
2496 	switch (val) {
2497 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2498 	case PM_QOS_LATENCY_ANY:
2499 		latency = U64_MAX;
2500 		break;
2501 
2502 	default:
2503 		latency = val;
2504 	}
2505 
2506 	if (ctrl->ps_max_latency_us != latency) {
2507 		ctrl->ps_max_latency_us = latency;
2508 		nvme_configure_apst(ctrl);
2509 	}
2510 }
2511 
2512 struct nvme_core_quirk_entry {
2513 	/*
2514 	 * NVMe model and firmware strings are padded with spaces.  For
2515 	 * simplicity, strings in the quirk table are padded with NULLs
2516 	 * instead.
2517 	 */
2518 	u16 vid;
2519 	const char *mn;
2520 	const char *fr;
2521 	unsigned long quirks;
2522 };
2523 
2524 static const struct nvme_core_quirk_entry core_quirks[] = {
2525 	{
2526 		/*
2527 		 * This Toshiba device seems to die using any APST states.  See:
2528 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2529 		 */
2530 		.vid = 0x1179,
2531 		.mn = "THNSF5256GPUK TOSHIBA",
2532 		.quirks = NVME_QUIRK_NO_APST,
2533 	},
2534 	{
2535 		/*
2536 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2537 		 * condition associated with actions related to suspend to idle
2538 		 * LiteON has resolved the problem in future firmware
2539 		 */
2540 		.vid = 0x14a4,
2541 		.fr = "22301111",
2542 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2543 	}
2544 };
2545 
2546 /* match is null-terminated but idstr is space-padded. */
2547 static bool string_matches(const char *idstr, const char *match, size_t len)
2548 {
2549 	size_t matchlen;
2550 
2551 	if (!match)
2552 		return true;
2553 
2554 	matchlen = strlen(match);
2555 	WARN_ON_ONCE(matchlen > len);
2556 
2557 	if (memcmp(idstr, match, matchlen))
2558 		return false;
2559 
2560 	for (; matchlen < len; matchlen++)
2561 		if (idstr[matchlen] != ' ')
2562 			return false;
2563 
2564 	return true;
2565 }
2566 
2567 static bool quirk_matches(const struct nvme_id_ctrl *id,
2568 			  const struct nvme_core_quirk_entry *q)
2569 {
2570 	return q->vid == le16_to_cpu(id->vid) &&
2571 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2572 		string_matches(id->fr, q->fr, sizeof(id->fr));
2573 }
2574 
2575 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2576 		struct nvme_id_ctrl *id)
2577 {
2578 	size_t nqnlen;
2579 	int off;
2580 
2581 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2582 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2583 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2584 			strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2585 			return;
2586 		}
2587 
2588 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2589 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2590 	}
2591 
2592 	/* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2593 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2594 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2595 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2596 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2597 	off += sizeof(id->sn);
2598 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2599 	off += sizeof(id->mn);
2600 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2601 }
2602 
2603 static void nvme_release_subsystem(struct device *dev)
2604 {
2605 	struct nvme_subsystem *subsys =
2606 		container_of(dev, struct nvme_subsystem, dev);
2607 
2608 	if (subsys->instance >= 0)
2609 		ida_simple_remove(&nvme_instance_ida, subsys->instance);
2610 	kfree(subsys);
2611 }
2612 
2613 static void nvme_destroy_subsystem(struct kref *ref)
2614 {
2615 	struct nvme_subsystem *subsys =
2616 			container_of(ref, struct nvme_subsystem, ref);
2617 
2618 	mutex_lock(&nvme_subsystems_lock);
2619 	list_del(&subsys->entry);
2620 	mutex_unlock(&nvme_subsystems_lock);
2621 
2622 	ida_destroy(&subsys->ns_ida);
2623 	device_del(&subsys->dev);
2624 	put_device(&subsys->dev);
2625 }
2626 
2627 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2628 {
2629 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2630 }
2631 
2632 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2633 {
2634 	struct nvme_subsystem *subsys;
2635 
2636 	lockdep_assert_held(&nvme_subsystems_lock);
2637 
2638 	/*
2639 	 * Fail matches for discovery subsystems. This results
2640 	 * in each discovery controller bound to a unique subsystem.
2641 	 * This avoids issues with validating controller values
2642 	 * that can only be true when there is a single unique subsystem.
2643 	 * There may be multiple and completely independent entities
2644 	 * that provide discovery controllers.
2645 	 */
2646 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2647 		return NULL;
2648 
2649 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2650 		if (strcmp(subsys->subnqn, subsysnqn))
2651 			continue;
2652 		if (!kref_get_unless_zero(&subsys->ref))
2653 			continue;
2654 		return subsys;
2655 	}
2656 
2657 	return NULL;
2658 }
2659 
2660 #define SUBSYS_ATTR_RO(_name, _mode, _show)			\
2661 	struct device_attribute subsys_attr_##_name = \
2662 		__ATTR(_name, _mode, _show, NULL)
2663 
2664 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2665 				    struct device_attribute *attr,
2666 				    char *buf)
2667 {
2668 	struct nvme_subsystem *subsys =
2669 		container_of(dev, struct nvme_subsystem, dev);
2670 
2671 	return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2672 }
2673 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2674 
2675 #define nvme_subsys_show_str_function(field)				\
2676 static ssize_t subsys_##field##_show(struct device *dev,		\
2677 			    struct device_attribute *attr, char *buf)	\
2678 {									\
2679 	struct nvme_subsystem *subsys =					\
2680 		container_of(dev, struct nvme_subsystem, dev);		\
2681 	return sprintf(buf, "%.*s\n",					\
2682 		       (int)sizeof(subsys->field), subsys->field);	\
2683 }									\
2684 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2685 
2686 nvme_subsys_show_str_function(model);
2687 nvme_subsys_show_str_function(serial);
2688 nvme_subsys_show_str_function(firmware_rev);
2689 
2690 static struct attribute *nvme_subsys_attrs[] = {
2691 	&subsys_attr_model.attr,
2692 	&subsys_attr_serial.attr,
2693 	&subsys_attr_firmware_rev.attr,
2694 	&subsys_attr_subsysnqn.attr,
2695 #ifdef CONFIG_NVME_MULTIPATH
2696 	&subsys_attr_iopolicy.attr,
2697 #endif
2698 	NULL,
2699 };
2700 
2701 static struct attribute_group nvme_subsys_attrs_group = {
2702 	.attrs = nvme_subsys_attrs,
2703 };
2704 
2705 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2706 	&nvme_subsys_attrs_group,
2707 	NULL,
2708 };
2709 
2710 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2711 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2712 {
2713 	struct nvme_ctrl *tmp;
2714 
2715 	lockdep_assert_held(&nvme_subsystems_lock);
2716 
2717 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2718 		if (nvme_state_terminal(tmp))
2719 			continue;
2720 
2721 		if (tmp->cntlid == ctrl->cntlid) {
2722 			dev_err(ctrl->device,
2723 				"Duplicate cntlid %u with %s, rejecting\n",
2724 				ctrl->cntlid, dev_name(tmp->device));
2725 			return false;
2726 		}
2727 
2728 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2729 		    (ctrl->opts && ctrl->opts->discovery_nqn))
2730 			continue;
2731 
2732 		dev_err(ctrl->device,
2733 			"Subsystem does not support multiple controllers\n");
2734 		return false;
2735 	}
2736 
2737 	return true;
2738 }
2739 
2740 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2741 {
2742 	struct nvme_subsystem *subsys, *found;
2743 	int ret;
2744 
2745 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2746 	if (!subsys)
2747 		return -ENOMEM;
2748 
2749 	subsys->instance = -1;
2750 	mutex_init(&subsys->lock);
2751 	kref_init(&subsys->ref);
2752 	INIT_LIST_HEAD(&subsys->ctrls);
2753 	INIT_LIST_HEAD(&subsys->nsheads);
2754 	nvme_init_subnqn(subsys, ctrl, id);
2755 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2756 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2757 	memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2758 	subsys->vendor_id = le16_to_cpu(id->vid);
2759 	subsys->cmic = id->cmic;
2760 	subsys->awupf = le16_to_cpu(id->awupf);
2761 #ifdef CONFIG_NVME_MULTIPATH
2762 	subsys->iopolicy = NVME_IOPOLICY_NUMA;
2763 #endif
2764 
2765 	subsys->dev.class = nvme_subsys_class;
2766 	subsys->dev.release = nvme_release_subsystem;
2767 	subsys->dev.groups = nvme_subsys_attrs_groups;
2768 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2769 	device_initialize(&subsys->dev);
2770 
2771 	mutex_lock(&nvme_subsystems_lock);
2772 	found = __nvme_find_get_subsystem(subsys->subnqn);
2773 	if (found) {
2774 		put_device(&subsys->dev);
2775 		subsys = found;
2776 
2777 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2778 			ret = -EINVAL;
2779 			goto out_put_subsystem;
2780 		}
2781 	} else {
2782 		ret = device_add(&subsys->dev);
2783 		if (ret) {
2784 			dev_err(ctrl->device,
2785 				"failed to register subsystem device.\n");
2786 			put_device(&subsys->dev);
2787 			goto out_unlock;
2788 		}
2789 		ida_init(&subsys->ns_ida);
2790 		list_add_tail(&subsys->entry, &nvme_subsystems);
2791 	}
2792 
2793 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2794 				dev_name(ctrl->device));
2795 	if (ret) {
2796 		dev_err(ctrl->device,
2797 			"failed to create sysfs link from subsystem.\n");
2798 		goto out_put_subsystem;
2799 	}
2800 
2801 	if (!found)
2802 		subsys->instance = ctrl->instance;
2803 	ctrl->subsys = subsys;
2804 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2805 	mutex_unlock(&nvme_subsystems_lock);
2806 	return 0;
2807 
2808 out_put_subsystem:
2809 	nvme_put_subsystem(subsys);
2810 out_unlock:
2811 	mutex_unlock(&nvme_subsystems_lock);
2812 	return ret;
2813 }
2814 
2815 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2816 		void *log, size_t size, u64 offset)
2817 {
2818 	struct nvme_command c = { };
2819 	u32 dwlen = nvme_bytes_to_numd(size);
2820 
2821 	c.get_log_page.opcode = nvme_admin_get_log_page;
2822 	c.get_log_page.nsid = cpu_to_le32(nsid);
2823 	c.get_log_page.lid = log_page;
2824 	c.get_log_page.lsp = lsp;
2825 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2826 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2827 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2828 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2829 
2830 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2831 }
2832 
2833 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2834 {
2835 	int ret;
2836 
2837 	if (!ctrl->effects)
2838 		ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2839 
2840 	if (!ctrl->effects)
2841 		return 0;
2842 
2843 	ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2844 			ctrl->effects, sizeof(*ctrl->effects), 0);
2845 	if (ret) {
2846 		kfree(ctrl->effects);
2847 		ctrl->effects = NULL;
2848 	}
2849 	return ret;
2850 }
2851 
2852 /*
2853  * Initialize the cached copies of the Identify data and various controller
2854  * register in our nvme_ctrl structure.  This should be called as soon as
2855  * the admin queue is fully up and running.
2856  */
2857 int nvme_init_identify(struct nvme_ctrl *ctrl)
2858 {
2859 	struct nvme_id_ctrl *id;
2860 	int ret, page_shift;
2861 	u32 max_hw_sectors;
2862 	bool prev_apst_enabled;
2863 
2864 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2865 	if (ret) {
2866 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2867 		return ret;
2868 	}
2869 	page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2870 	ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2871 
2872 	if (ctrl->vs >= NVME_VS(1, 1, 0))
2873 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2874 
2875 	ret = nvme_identify_ctrl(ctrl, &id);
2876 	if (ret) {
2877 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2878 		return -EIO;
2879 	}
2880 
2881 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2882 		ret = nvme_get_effects_log(ctrl);
2883 		if (ret < 0)
2884 			goto out_free;
2885 	}
2886 
2887 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
2888 		ctrl->cntlid = le16_to_cpu(id->cntlid);
2889 
2890 	if (!ctrl->identified) {
2891 		int i;
2892 
2893 		ret = nvme_init_subsystem(ctrl, id);
2894 		if (ret)
2895 			goto out_free;
2896 
2897 		/*
2898 		 * Check for quirks.  Quirk can depend on firmware version,
2899 		 * so, in principle, the set of quirks present can change
2900 		 * across a reset.  As a possible future enhancement, we
2901 		 * could re-scan for quirks every time we reinitialize
2902 		 * the device, but we'd have to make sure that the driver
2903 		 * behaves intelligently if the quirks change.
2904 		 */
2905 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2906 			if (quirk_matches(id, &core_quirks[i]))
2907 				ctrl->quirks |= core_quirks[i].quirks;
2908 		}
2909 	}
2910 
2911 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2912 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2913 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2914 	}
2915 
2916 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2917 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2918 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2919 
2920 	ctrl->oacs = le16_to_cpu(id->oacs);
2921 	ctrl->oncs = le16_to_cpu(id->oncs);
2922 	ctrl->mtfa = le16_to_cpu(id->mtfa);
2923 	ctrl->oaes = le32_to_cpu(id->oaes);
2924 	ctrl->wctemp = le16_to_cpu(id->wctemp);
2925 	ctrl->cctemp = le16_to_cpu(id->cctemp);
2926 
2927 	atomic_set(&ctrl->abort_limit, id->acl + 1);
2928 	ctrl->vwc = id->vwc;
2929 	if (id->mdts)
2930 		max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2931 	else
2932 		max_hw_sectors = UINT_MAX;
2933 	ctrl->max_hw_sectors =
2934 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2935 
2936 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
2937 	ctrl->sgls = le32_to_cpu(id->sgls);
2938 	ctrl->kas = le16_to_cpu(id->kas);
2939 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
2940 	ctrl->ctratt = le32_to_cpu(id->ctratt);
2941 
2942 	if (id->rtd3e) {
2943 		/* us -> s */
2944 		u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2945 
2946 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2947 						 shutdown_timeout, 60);
2948 
2949 		if (ctrl->shutdown_timeout != shutdown_timeout)
2950 			dev_info(ctrl->device,
2951 				 "Shutdown timeout set to %u seconds\n",
2952 				 ctrl->shutdown_timeout);
2953 	} else
2954 		ctrl->shutdown_timeout = shutdown_timeout;
2955 
2956 	ctrl->npss = id->npss;
2957 	ctrl->apsta = id->apsta;
2958 	prev_apst_enabled = ctrl->apst_enabled;
2959 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2960 		if (force_apst && id->apsta) {
2961 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2962 			ctrl->apst_enabled = true;
2963 		} else {
2964 			ctrl->apst_enabled = false;
2965 		}
2966 	} else {
2967 		ctrl->apst_enabled = id->apsta;
2968 	}
2969 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2970 
2971 	if (ctrl->ops->flags & NVME_F_FABRICS) {
2972 		ctrl->icdoff = le16_to_cpu(id->icdoff);
2973 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2974 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2975 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2976 
2977 		/*
2978 		 * In fabrics we need to verify the cntlid matches the
2979 		 * admin connect
2980 		 */
2981 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2982 			dev_err(ctrl->device,
2983 				"Mismatching cntlid: Connect %u vs Identify "
2984 				"%u, rejecting\n",
2985 				ctrl->cntlid, le16_to_cpu(id->cntlid));
2986 			ret = -EINVAL;
2987 			goto out_free;
2988 		}
2989 
2990 		if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2991 			dev_err(ctrl->device,
2992 				"keep-alive support is mandatory for fabrics\n");
2993 			ret = -EINVAL;
2994 			goto out_free;
2995 		}
2996 	} else {
2997 		ctrl->hmpre = le32_to_cpu(id->hmpre);
2998 		ctrl->hmmin = le32_to_cpu(id->hmmin);
2999 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3000 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3001 	}
3002 
3003 	ret = nvme_mpath_init(ctrl, id);
3004 	kfree(id);
3005 
3006 	if (ret < 0)
3007 		return ret;
3008 
3009 	if (ctrl->apst_enabled && !prev_apst_enabled)
3010 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3011 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3012 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3013 
3014 	ret = nvme_configure_apst(ctrl);
3015 	if (ret < 0)
3016 		return ret;
3017 
3018 	ret = nvme_configure_timestamp(ctrl);
3019 	if (ret < 0)
3020 		return ret;
3021 
3022 	ret = nvme_configure_directives(ctrl);
3023 	if (ret < 0)
3024 		return ret;
3025 
3026 	ret = nvme_configure_acre(ctrl);
3027 	if (ret < 0)
3028 		return ret;
3029 
3030 	if (!ctrl->identified)
3031 		nvme_hwmon_init(ctrl);
3032 
3033 	ctrl->identified = true;
3034 
3035 	return 0;
3036 
3037 out_free:
3038 	kfree(id);
3039 	return ret;
3040 }
3041 EXPORT_SYMBOL_GPL(nvme_init_identify);
3042 
3043 static int nvme_dev_open(struct inode *inode, struct file *file)
3044 {
3045 	struct nvme_ctrl *ctrl =
3046 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3047 
3048 	switch (ctrl->state) {
3049 	case NVME_CTRL_LIVE:
3050 		break;
3051 	default:
3052 		return -EWOULDBLOCK;
3053 	}
3054 
3055 	file->private_data = ctrl;
3056 	return 0;
3057 }
3058 
3059 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3060 {
3061 	struct nvme_ns *ns;
3062 	int ret;
3063 
3064 	down_read(&ctrl->namespaces_rwsem);
3065 	if (list_empty(&ctrl->namespaces)) {
3066 		ret = -ENOTTY;
3067 		goto out_unlock;
3068 	}
3069 
3070 	ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3071 	if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3072 		dev_warn(ctrl->device,
3073 			"NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3074 		ret = -EINVAL;
3075 		goto out_unlock;
3076 	}
3077 
3078 	dev_warn(ctrl->device,
3079 		"using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3080 	kref_get(&ns->kref);
3081 	up_read(&ctrl->namespaces_rwsem);
3082 
3083 	ret = nvme_user_cmd(ctrl, ns, argp);
3084 	nvme_put_ns(ns);
3085 	return ret;
3086 
3087 out_unlock:
3088 	up_read(&ctrl->namespaces_rwsem);
3089 	return ret;
3090 }
3091 
3092 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3093 		unsigned long arg)
3094 {
3095 	struct nvme_ctrl *ctrl = file->private_data;
3096 	void __user *argp = (void __user *)arg;
3097 
3098 	switch (cmd) {
3099 	case NVME_IOCTL_ADMIN_CMD:
3100 		return nvme_user_cmd(ctrl, NULL, argp);
3101 	case NVME_IOCTL_ADMIN64_CMD:
3102 		return nvme_user_cmd64(ctrl, NULL, argp);
3103 	case NVME_IOCTL_IO_CMD:
3104 		return nvme_dev_user_cmd(ctrl, argp);
3105 	case NVME_IOCTL_RESET:
3106 		dev_warn(ctrl->device, "resetting controller\n");
3107 		return nvme_reset_ctrl_sync(ctrl);
3108 	case NVME_IOCTL_SUBSYS_RESET:
3109 		return nvme_reset_subsystem(ctrl);
3110 	case NVME_IOCTL_RESCAN:
3111 		nvme_queue_scan(ctrl);
3112 		return 0;
3113 	default:
3114 		return -ENOTTY;
3115 	}
3116 }
3117 
3118 static const struct file_operations nvme_dev_fops = {
3119 	.owner		= THIS_MODULE,
3120 	.open		= nvme_dev_open,
3121 	.unlocked_ioctl	= nvme_dev_ioctl,
3122 	.compat_ioctl	= compat_ptr_ioctl,
3123 };
3124 
3125 static ssize_t nvme_sysfs_reset(struct device *dev,
3126 				struct device_attribute *attr, const char *buf,
3127 				size_t count)
3128 {
3129 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3130 	int ret;
3131 
3132 	ret = nvme_reset_ctrl_sync(ctrl);
3133 	if (ret < 0)
3134 		return ret;
3135 	return count;
3136 }
3137 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3138 
3139 static ssize_t nvme_sysfs_rescan(struct device *dev,
3140 				struct device_attribute *attr, const char *buf,
3141 				size_t count)
3142 {
3143 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3144 
3145 	nvme_queue_scan(ctrl);
3146 	return count;
3147 }
3148 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3149 
3150 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3151 {
3152 	struct gendisk *disk = dev_to_disk(dev);
3153 
3154 	if (disk->fops == &nvme_fops)
3155 		return nvme_get_ns_from_dev(dev)->head;
3156 	else
3157 		return disk->private_data;
3158 }
3159 
3160 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3161 		char *buf)
3162 {
3163 	struct nvme_ns_head *head = dev_to_ns_head(dev);
3164 	struct nvme_ns_ids *ids = &head->ids;
3165 	struct nvme_subsystem *subsys = head->subsys;
3166 	int serial_len = sizeof(subsys->serial);
3167 	int model_len = sizeof(subsys->model);
3168 
3169 	if (!uuid_is_null(&ids->uuid))
3170 		return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3171 
3172 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3173 		return sprintf(buf, "eui.%16phN\n", ids->nguid);
3174 
3175 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3176 		return sprintf(buf, "eui.%8phN\n", ids->eui64);
3177 
3178 	while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3179 				  subsys->serial[serial_len - 1] == '\0'))
3180 		serial_len--;
3181 	while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3182 				 subsys->model[model_len - 1] == '\0'))
3183 		model_len--;
3184 
3185 	return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3186 		serial_len, subsys->serial, model_len, subsys->model,
3187 		head->ns_id);
3188 }
3189 static DEVICE_ATTR_RO(wwid);
3190 
3191 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3192 		char *buf)
3193 {
3194 	return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3195 }
3196 static DEVICE_ATTR_RO(nguid);
3197 
3198 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3199 		char *buf)
3200 {
3201 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3202 
3203 	/* For backward compatibility expose the NGUID to userspace if
3204 	 * we have no UUID set
3205 	 */
3206 	if (uuid_is_null(&ids->uuid)) {
3207 		printk_ratelimited(KERN_WARNING
3208 				   "No UUID available providing old NGUID\n");
3209 		return sprintf(buf, "%pU\n", ids->nguid);
3210 	}
3211 	return sprintf(buf, "%pU\n", &ids->uuid);
3212 }
3213 static DEVICE_ATTR_RO(uuid);
3214 
3215 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3216 		char *buf)
3217 {
3218 	return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3219 }
3220 static DEVICE_ATTR_RO(eui);
3221 
3222 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3223 		char *buf)
3224 {
3225 	return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3226 }
3227 static DEVICE_ATTR_RO(nsid);
3228 
3229 static struct attribute *nvme_ns_id_attrs[] = {
3230 	&dev_attr_wwid.attr,
3231 	&dev_attr_uuid.attr,
3232 	&dev_attr_nguid.attr,
3233 	&dev_attr_eui.attr,
3234 	&dev_attr_nsid.attr,
3235 #ifdef CONFIG_NVME_MULTIPATH
3236 	&dev_attr_ana_grpid.attr,
3237 	&dev_attr_ana_state.attr,
3238 #endif
3239 	NULL,
3240 };
3241 
3242 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3243 		struct attribute *a, int n)
3244 {
3245 	struct device *dev = container_of(kobj, struct device, kobj);
3246 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3247 
3248 	if (a == &dev_attr_uuid.attr) {
3249 		if (uuid_is_null(&ids->uuid) &&
3250 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3251 			return 0;
3252 	}
3253 	if (a == &dev_attr_nguid.attr) {
3254 		if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3255 			return 0;
3256 	}
3257 	if (a == &dev_attr_eui.attr) {
3258 		if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3259 			return 0;
3260 	}
3261 #ifdef CONFIG_NVME_MULTIPATH
3262 	if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3263 		if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3264 			return 0;
3265 		if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3266 			return 0;
3267 	}
3268 #endif
3269 	return a->mode;
3270 }
3271 
3272 static const struct attribute_group nvme_ns_id_attr_group = {
3273 	.attrs		= nvme_ns_id_attrs,
3274 	.is_visible	= nvme_ns_id_attrs_are_visible,
3275 };
3276 
3277 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3278 	&nvme_ns_id_attr_group,
3279 #ifdef CONFIG_NVM
3280 	&nvme_nvm_attr_group,
3281 #endif
3282 	NULL,
3283 };
3284 
3285 #define nvme_show_str_function(field)						\
3286 static ssize_t  field##_show(struct device *dev,				\
3287 			    struct device_attribute *attr, char *buf)		\
3288 {										\
3289         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3290         return sprintf(buf, "%.*s\n",						\
3291 		(int)sizeof(ctrl->subsys->field), ctrl->subsys->field);		\
3292 }										\
3293 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3294 
3295 nvme_show_str_function(model);
3296 nvme_show_str_function(serial);
3297 nvme_show_str_function(firmware_rev);
3298 
3299 #define nvme_show_int_function(field)						\
3300 static ssize_t  field##_show(struct device *dev,				\
3301 			    struct device_attribute *attr, char *buf)		\
3302 {										\
3303         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3304         return sprintf(buf, "%d\n", ctrl->field);	\
3305 }										\
3306 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3307 
3308 nvme_show_int_function(cntlid);
3309 nvme_show_int_function(numa_node);
3310 nvme_show_int_function(queue_count);
3311 nvme_show_int_function(sqsize);
3312 
3313 static ssize_t nvme_sysfs_delete(struct device *dev,
3314 				struct device_attribute *attr, const char *buf,
3315 				size_t count)
3316 {
3317 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3318 
3319 	/* Can't delete non-created controllers */
3320 	if (!ctrl->created)
3321 		return -EBUSY;
3322 
3323 	if (device_remove_file_self(dev, attr))
3324 		nvme_delete_ctrl_sync(ctrl);
3325 	return count;
3326 }
3327 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3328 
3329 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3330 					 struct device_attribute *attr,
3331 					 char *buf)
3332 {
3333 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3334 
3335 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3336 }
3337 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3338 
3339 static ssize_t nvme_sysfs_show_state(struct device *dev,
3340 				     struct device_attribute *attr,
3341 				     char *buf)
3342 {
3343 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3344 	static const char *const state_name[] = {
3345 		[NVME_CTRL_NEW]		= "new",
3346 		[NVME_CTRL_LIVE]	= "live",
3347 		[NVME_CTRL_RESETTING]	= "resetting",
3348 		[NVME_CTRL_CONNECTING]	= "connecting",
3349 		[NVME_CTRL_DELETING]	= "deleting",
3350 		[NVME_CTRL_DEAD]	= "dead",
3351 	};
3352 
3353 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3354 	    state_name[ctrl->state])
3355 		return sprintf(buf, "%s\n", state_name[ctrl->state]);
3356 
3357 	return sprintf(buf, "unknown state\n");
3358 }
3359 
3360 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3361 
3362 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3363 					 struct device_attribute *attr,
3364 					 char *buf)
3365 {
3366 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3367 
3368 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3369 }
3370 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3371 
3372 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3373 					struct device_attribute *attr,
3374 					char *buf)
3375 {
3376 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3377 
3378 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3379 }
3380 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3381 
3382 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3383 					struct device_attribute *attr,
3384 					char *buf)
3385 {
3386 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3387 
3388 	return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3389 }
3390 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3391 
3392 static ssize_t nvme_sysfs_show_address(struct device *dev,
3393 					 struct device_attribute *attr,
3394 					 char *buf)
3395 {
3396 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3397 
3398 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3399 }
3400 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3401 
3402 static struct attribute *nvme_dev_attrs[] = {
3403 	&dev_attr_reset_controller.attr,
3404 	&dev_attr_rescan_controller.attr,
3405 	&dev_attr_model.attr,
3406 	&dev_attr_serial.attr,
3407 	&dev_attr_firmware_rev.attr,
3408 	&dev_attr_cntlid.attr,
3409 	&dev_attr_delete_controller.attr,
3410 	&dev_attr_transport.attr,
3411 	&dev_attr_subsysnqn.attr,
3412 	&dev_attr_address.attr,
3413 	&dev_attr_state.attr,
3414 	&dev_attr_numa_node.attr,
3415 	&dev_attr_queue_count.attr,
3416 	&dev_attr_sqsize.attr,
3417 	&dev_attr_hostnqn.attr,
3418 	&dev_attr_hostid.attr,
3419 	NULL
3420 };
3421 
3422 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3423 		struct attribute *a, int n)
3424 {
3425 	struct device *dev = container_of(kobj, struct device, kobj);
3426 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3427 
3428 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3429 		return 0;
3430 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3431 		return 0;
3432 	if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3433 		return 0;
3434 	if (a == &dev_attr_hostid.attr && !ctrl->opts)
3435 		return 0;
3436 
3437 	return a->mode;
3438 }
3439 
3440 static struct attribute_group nvme_dev_attrs_group = {
3441 	.attrs		= nvme_dev_attrs,
3442 	.is_visible	= nvme_dev_attrs_are_visible,
3443 };
3444 
3445 static const struct attribute_group *nvme_dev_attr_groups[] = {
3446 	&nvme_dev_attrs_group,
3447 	NULL,
3448 };
3449 
3450 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3451 		unsigned nsid)
3452 {
3453 	struct nvme_ns_head *h;
3454 
3455 	lockdep_assert_held(&subsys->lock);
3456 
3457 	list_for_each_entry(h, &subsys->nsheads, entry) {
3458 		if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3459 			return h;
3460 	}
3461 
3462 	return NULL;
3463 }
3464 
3465 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3466 		struct nvme_ns_head *new)
3467 {
3468 	struct nvme_ns_head *h;
3469 
3470 	lockdep_assert_held(&subsys->lock);
3471 
3472 	list_for_each_entry(h, &subsys->nsheads, entry) {
3473 		if (nvme_ns_ids_valid(&new->ids) &&
3474 		    nvme_ns_ids_equal(&new->ids, &h->ids))
3475 			return -EINVAL;
3476 	}
3477 
3478 	return 0;
3479 }
3480 
3481 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3482 		unsigned nsid, struct nvme_ns_ids *ids)
3483 {
3484 	struct nvme_ns_head *head;
3485 	size_t size = sizeof(*head);
3486 	int ret = -ENOMEM;
3487 
3488 #ifdef CONFIG_NVME_MULTIPATH
3489 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3490 #endif
3491 
3492 	head = kzalloc(size, GFP_KERNEL);
3493 	if (!head)
3494 		goto out;
3495 	ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3496 	if (ret < 0)
3497 		goto out_free_head;
3498 	head->instance = ret;
3499 	INIT_LIST_HEAD(&head->list);
3500 	ret = init_srcu_struct(&head->srcu);
3501 	if (ret)
3502 		goto out_ida_remove;
3503 	head->subsys = ctrl->subsys;
3504 	head->ns_id = nsid;
3505 	head->ids = *ids;
3506 	kref_init(&head->ref);
3507 
3508 	ret = __nvme_check_ids(ctrl->subsys, head);
3509 	if (ret) {
3510 		dev_err(ctrl->device,
3511 			"duplicate IDs for nsid %d\n", nsid);
3512 		goto out_cleanup_srcu;
3513 	}
3514 
3515 	ret = nvme_mpath_alloc_disk(ctrl, head);
3516 	if (ret)
3517 		goto out_cleanup_srcu;
3518 
3519 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3520 
3521 	kref_get(&ctrl->subsys->ref);
3522 
3523 	return head;
3524 out_cleanup_srcu:
3525 	cleanup_srcu_struct(&head->srcu);
3526 out_ida_remove:
3527 	ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3528 out_free_head:
3529 	kfree(head);
3530 out:
3531 	if (ret > 0)
3532 		ret = blk_status_to_errno(nvme_error_status(ret));
3533 	return ERR_PTR(ret);
3534 }
3535 
3536 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3537 		struct nvme_id_ns *id)
3538 {
3539 	struct nvme_ctrl *ctrl = ns->ctrl;
3540 	bool is_shared = id->nmic & NVME_NS_NMIC_SHARED;
3541 	struct nvme_ns_head *head = NULL;
3542 	struct nvme_ns_ids ids;
3543 	int ret = 0;
3544 
3545 	ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3546 	if (ret) {
3547 		if (ret < 0)
3548 			return ret;
3549 		return blk_status_to_errno(nvme_error_status(ret));
3550 	}
3551 
3552 	mutex_lock(&ctrl->subsys->lock);
3553 	head = nvme_find_ns_head(ctrl->subsys, nsid);
3554 	if (!head) {
3555 		head = nvme_alloc_ns_head(ctrl, nsid, &ids);
3556 		if (IS_ERR(head)) {
3557 			ret = PTR_ERR(head);
3558 			goto out_unlock;
3559 		}
3560 		head->shared = is_shared;
3561 	} else {
3562 		ret = -EINVAL;
3563 		if (!is_shared || !head->shared) {
3564 			dev_err(ctrl->device,
3565 				"Duplicate unshared namespace %d\n", nsid);
3566 			goto out_put_ns_head;
3567 		}
3568 		if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3569 			dev_err(ctrl->device,
3570 				"IDs don't match for shared namespace %d\n",
3571 					nsid);
3572 			goto out_put_ns_head;
3573 		}
3574 	}
3575 
3576 	list_add_tail(&ns->siblings, &head->list);
3577 	ns->head = head;
3578 	mutex_unlock(&ctrl->subsys->lock);
3579 	return 0;
3580 
3581 out_put_ns_head:
3582 	nvme_put_ns_head(head);
3583 out_unlock:
3584 	mutex_unlock(&ctrl->subsys->lock);
3585 	return ret;
3586 }
3587 
3588 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3589 {
3590 	struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3591 	struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3592 
3593 	return nsa->head->ns_id - nsb->head->ns_id;
3594 }
3595 
3596 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3597 {
3598 	struct nvme_ns *ns, *ret = NULL;
3599 
3600 	down_read(&ctrl->namespaces_rwsem);
3601 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3602 		if (ns->head->ns_id == nsid) {
3603 			if (!kref_get_unless_zero(&ns->kref))
3604 				continue;
3605 			ret = ns;
3606 			break;
3607 		}
3608 		if (ns->head->ns_id > nsid)
3609 			break;
3610 	}
3611 	up_read(&ctrl->namespaces_rwsem);
3612 	return ret;
3613 }
3614 
3615 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3616 {
3617 	struct nvme_ns *ns;
3618 	struct gendisk *disk;
3619 	struct nvme_id_ns *id;
3620 	char disk_name[DISK_NAME_LEN];
3621 	int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3622 
3623 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3624 	if (!ns)
3625 		return;
3626 
3627 	ns->queue = blk_mq_init_queue(ctrl->tagset);
3628 	if (IS_ERR(ns->queue))
3629 		goto out_free_ns;
3630 
3631 	if (ctrl->opts && ctrl->opts->data_digest)
3632 		ns->queue->backing_dev_info->capabilities
3633 			|= BDI_CAP_STABLE_WRITES;
3634 
3635 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3636 	if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3637 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3638 
3639 	ns->queue->queuedata = ns;
3640 	ns->ctrl = ctrl;
3641 
3642 	kref_init(&ns->kref);
3643 	ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3644 
3645 	blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3646 	nvme_set_queue_limits(ctrl, ns->queue);
3647 
3648 	ret = nvme_identify_ns(ctrl, nsid, &id);
3649 	if (ret)
3650 		goto out_free_queue;
3651 
3652 	if (id->ncap == 0)	/* no namespace (legacy quirk) */
3653 		goto out_free_id;
3654 
3655 	ret = nvme_init_ns_head(ns, nsid, id);
3656 	if (ret)
3657 		goto out_free_id;
3658 	nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3659 
3660 	disk = alloc_disk_node(0, node);
3661 	if (!disk)
3662 		goto out_unlink_ns;
3663 
3664 	disk->fops = &nvme_fops;
3665 	disk->private_data = ns;
3666 	disk->queue = ns->queue;
3667 	disk->flags = flags;
3668 	memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3669 	ns->disk = disk;
3670 
3671 	if (__nvme_revalidate_disk(disk, id))
3672 		goto out_put_disk;
3673 
3674 	if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3675 		ret = nvme_nvm_register(ns, disk_name, node);
3676 		if (ret) {
3677 			dev_warn(ctrl->device, "LightNVM init failure\n");
3678 			goto out_put_disk;
3679 		}
3680 	}
3681 
3682 	down_write(&ctrl->namespaces_rwsem);
3683 	list_add_tail(&ns->list, &ctrl->namespaces);
3684 	up_write(&ctrl->namespaces_rwsem);
3685 
3686 	nvme_get_ctrl(ctrl);
3687 
3688 	device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3689 
3690 	nvme_mpath_add_disk(ns, id);
3691 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3692 	kfree(id);
3693 
3694 	return;
3695  out_put_disk:
3696 	/* prevent double queue cleanup */
3697 	ns->disk->queue = NULL;
3698 	put_disk(ns->disk);
3699  out_unlink_ns:
3700 	mutex_lock(&ctrl->subsys->lock);
3701 	list_del_rcu(&ns->siblings);
3702 	if (list_empty(&ns->head->list))
3703 		list_del_init(&ns->head->entry);
3704 	mutex_unlock(&ctrl->subsys->lock);
3705 	nvme_put_ns_head(ns->head);
3706  out_free_id:
3707 	kfree(id);
3708  out_free_queue:
3709 	blk_cleanup_queue(ns->queue);
3710  out_free_ns:
3711 	kfree(ns);
3712 }
3713 
3714 static void nvme_ns_remove(struct nvme_ns *ns)
3715 {
3716 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3717 		return;
3718 
3719 	nvme_fault_inject_fini(&ns->fault_inject);
3720 
3721 	mutex_lock(&ns->ctrl->subsys->lock);
3722 	list_del_rcu(&ns->siblings);
3723 	if (list_empty(&ns->head->list))
3724 		list_del_init(&ns->head->entry);
3725 	mutex_unlock(&ns->ctrl->subsys->lock);
3726 
3727 	synchronize_rcu(); /* guarantee not available in head->list */
3728 	nvme_mpath_clear_current_path(ns);
3729 	synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3730 
3731 	if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3732 		del_gendisk(ns->disk);
3733 		blk_cleanup_queue(ns->queue);
3734 		if (blk_get_integrity(ns->disk))
3735 			blk_integrity_unregister(ns->disk);
3736 	}
3737 
3738 	down_write(&ns->ctrl->namespaces_rwsem);
3739 	list_del_init(&ns->list);
3740 	up_write(&ns->ctrl->namespaces_rwsem);
3741 
3742 	nvme_mpath_check_last_path(ns);
3743 	nvme_put_ns(ns);
3744 }
3745 
3746 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3747 {
3748 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3749 
3750 	if (ns) {
3751 		nvme_ns_remove(ns);
3752 		nvme_put_ns(ns);
3753 	}
3754 }
3755 
3756 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3757 {
3758 	struct nvme_ns *ns;
3759 
3760 	ns = nvme_find_get_ns(ctrl, nsid);
3761 	if (ns) {
3762 		if (ns->disk && revalidate_disk(ns->disk))
3763 			nvme_ns_remove(ns);
3764 		nvme_put_ns(ns);
3765 	} else
3766 		nvme_alloc_ns(ctrl, nsid);
3767 }
3768 
3769 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3770 					unsigned nsid)
3771 {
3772 	struct nvme_ns *ns, *next;
3773 	LIST_HEAD(rm_list);
3774 
3775 	down_write(&ctrl->namespaces_rwsem);
3776 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3777 		if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3778 			list_move_tail(&ns->list, &rm_list);
3779 	}
3780 	up_write(&ctrl->namespaces_rwsem);
3781 
3782 	list_for_each_entry_safe(ns, next, &rm_list, list)
3783 		nvme_ns_remove(ns);
3784 
3785 }
3786 
3787 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3788 {
3789 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3790 	__le32 *ns_list;
3791 	u32 prev = 0;
3792 	int ret = 0, i;
3793 
3794 	if (nvme_ctrl_limited_cns(ctrl))
3795 		return -EOPNOTSUPP;
3796 
3797 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3798 	if (!ns_list)
3799 		return -ENOMEM;
3800 
3801 	for (;;) {
3802 		ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3803 		if (ret)
3804 			goto free;
3805 
3806 		for (i = 0; i < nr_entries; i++) {
3807 			u32 nsid = le32_to_cpu(ns_list[i]);
3808 
3809 			if (!nsid)	/* end of the list? */
3810 				goto out;
3811 			nvme_validate_ns(ctrl, nsid);
3812 			while (++prev < nsid)
3813 				nvme_ns_remove_by_nsid(ctrl, prev);
3814 		}
3815 	}
3816  out:
3817 	nvme_remove_invalid_namespaces(ctrl, prev);
3818  free:
3819 	kfree(ns_list);
3820 	return ret;
3821 }
3822 
3823 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3824 {
3825 	struct nvme_id_ctrl *id;
3826 	u32 nn, i;
3827 
3828 	if (nvme_identify_ctrl(ctrl, &id))
3829 		return;
3830 	nn = le32_to_cpu(id->nn);
3831 	kfree(id);
3832 
3833 	for (i = 1; i <= nn; i++)
3834 		nvme_validate_ns(ctrl, i);
3835 
3836 	nvme_remove_invalid_namespaces(ctrl, nn);
3837 }
3838 
3839 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3840 {
3841 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3842 	__le32 *log;
3843 	int error;
3844 
3845 	log = kzalloc(log_size, GFP_KERNEL);
3846 	if (!log)
3847 		return;
3848 
3849 	/*
3850 	 * We need to read the log to clear the AEN, but we don't want to rely
3851 	 * on it for the changed namespace information as userspace could have
3852 	 * raced with us in reading the log page, which could cause us to miss
3853 	 * updates.
3854 	 */
3855 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3856 			log_size, 0);
3857 	if (error)
3858 		dev_warn(ctrl->device,
3859 			"reading changed ns log failed: %d\n", error);
3860 
3861 	kfree(log);
3862 }
3863 
3864 static void nvme_scan_work(struct work_struct *work)
3865 {
3866 	struct nvme_ctrl *ctrl =
3867 		container_of(work, struct nvme_ctrl, scan_work);
3868 
3869 	/* No tagset on a live ctrl means IO queues could not created */
3870 	if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3871 		return;
3872 
3873 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3874 		dev_info(ctrl->device, "rescanning namespaces.\n");
3875 		nvme_clear_changed_ns_log(ctrl);
3876 	}
3877 
3878 	mutex_lock(&ctrl->scan_lock);
3879 	if (nvme_scan_ns_list(ctrl) != 0)
3880 		nvme_scan_ns_sequential(ctrl);
3881 	mutex_unlock(&ctrl->scan_lock);
3882 
3883 	down_write(&ctrl->namespaces_rwsem);
3884 	list_sort(NULL, &ctrl->namespaces, ns_cmp);
3885 	up_write(&ctrl->namespaces_rwsem);
3886 }
3887 
3888 /*
3889  * This function iterates the namespace list unlocked to allow recovery from
3890  * controller failure. It is up to the caller to ensure the namespace list is
3891  * not modified by scan work while this function is executing.
3892  */
3893 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3894 {
3895 	struct nvme_ns *ns, *next;
3896 	LIST_HEAD(ns_list);
3897 
3898 	/*
3899 	 * make sure to requeue I/O to all namespaces as these
3900 	 * might result from the scan itself and must complete
3901 	 * for the scan_work to make progress
3902 	 */
3903 	nvme_mpath_clear_ctrl_paths(ctrl);
3904 
3905 	/* prevent racing with ns scanning */
3906 	flush_work(&ctrl->scan_work);
3907 
3908 	/*
3909 	 * The dead states indicates the controller was not gracefully
3910 	 * disconnected. In that case, we won't be able to flush any data while
3911 	 * removing the namespaces' disks; fail all the queues now to avoid
3912 	 * potentially having to clean up the failed sync later.
3913 	 */
3914 	if (ctrl->state == NVME_CTRL_DEAD)
3915 		nvme_kill_queues(ctrl);
3916 
3917 	down_write(&ctrl->namespaces_rwsem);
3918 	list_splice_init(&ctrl->namespaces, &ns_list);
3919 	up_write(&ctrl->namespaces_rwsem);
3920 
3921 	list_for_each_entry_safe(ns, next, &ns_list, list)
3922 		nvme_ns_remove(ns);
3923 }
3924 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3925 
3926 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3927 {
3928 	struct nvme_ctrl *ctrl =
3929 		container_of(dev, struct nvme_ctrl, ctrl_device);
3930 	struct nvmf_ctrl_options *opts = ctrl->opts;
3931 	int ret;
3932 
3933 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3934 	if (ret)
3935 		return ret;
3936 
3937 	if (opts) {
3938 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3939 		if (ret)
3940 			return ret;
3941 
3942 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3943 				opts->trsvcid ?: "none");
3944 		if (ret)
3945 			return ret;
3946 
3947 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3948 				opts->host_traddr ?: "none");
3949 	}
3950 	return ret;
3951 }
3952 
3953 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3954 {
3955 	char *envp[2] = { NULL, NULL };
3956 	u32 aen_result = ctrl->aen_result;
3957 
3958 	ctrl->aen_result = 0;
3959 	if (!aen_result)
3960 		return;
3961 
3962 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3963 	if (!envp[0])
3964 		return;
3965 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3966 	kfree(envp[0]);
3967 }
3968 
3969 static void nvme_async_event_work(struct work_struct *work)
3970 {
3971 	struct nvme_ctrl *ctrl =
3972 		container_of(work, struct nvme_ctrl, async_event_work);
3973 
3974 	nvme_aen_uevent(ctrl);
3975 	ctrl->ops->submit_async_event(ctrl);
3976 }
3977 
3978 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3979 {
3980 
3981 	u32 csts;
3982 
3983 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3984 		return false;
3985 
3986 	if (csts == ~0)
3987 		return false;
3988 
3989 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3990 }
3991 
3992 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3993 {
3994 	struct nvme_fw_slot_info_log *log;
3995 
3996 	log = kmalloc(sizeof(*log), GFP_KERNEL);
3997 	if (!log)
3998 		return;
3999 
4000 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, log,
4001 			sizeof(*log), 0))
4002 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4003 	kfree(log);
4004 }
4005 
4006 static void nvme_fw_act_work(struct work_struct *work)
4007 {
4008 	struct nvme_ctrl *ctrl = container_of(work,
4009 				struct nvme_ctrl, fw_act_work);
4010 	unsigned long fw_act_timeout;
4011 
4012 	if (ctrl->mtfa)
4013 		fw_act_timeout = jiffies +
4014 				msecs_to_jiffies(ctrl->mtfa * 100);
4015 	else
4016 		fw_act_timeout = jiffies +
4017 				msecs_to_jiffies(admin_timeout * 1000);
4018 
4019 	nvme_stop_queues(ctrl);
4020 	while (nvme_ctrl_pp_status(ctrl)) {
4021 		if (time_after(jiffies, fw_act_timeout)) {
4022 			dev_warn(ctrl->device,
4023 				"Fw activation timeout, reset controller\n");
4024 			nvme_try_sched_reset(ctrl);
4025 			return;
4026 		}
4027 		msleep(100);
4028 	}
4029 
4030 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4031 		return;
4032 
4033 	nvme_start_queues(ctrl);
4034 	/* read FW slot information to clear the AER */
4035 	nvme_get_fw_slot_info(ctrl);
4036 }
4037 
4038 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4039 {
4040 	u32 aer_notice_type = (result & 0xff00) >> 8;
4041 
4042 	trace_nvme_async_event(ctrl, aer_notice_type);
4043 
4044 	switch (aer_notice_type) {
4045 	case NVME_AER_NOTICE_NS_CHANGED:
4046 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4047 		nvme_queue_scan(ctrl);
4048 		break;
4049 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4050 		/*
4051 		 * We are (ab)using the RESETTING state to prevent subsequent
4052 		 * recovery actions from interfering with the controller's
4053 		 * firmware activation.
4054 		 */
4055 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4056 			queue_work(nvme_wq, &ctrl->fw_act_work);
4057 		break;
4058 #ifdef CONFIG_NVME_MULTIPATH
4059 	case NVME_AER_NOTICE_ANA:
4060 		if (!ctrl->ana_log_buf)
4061 			break;
4062 		queue_work(nvme_wq, &ctrl->ana_work);
4063 		break;
4064 #endif
4065 	case NVME_AER_NOTICE_DISC_CHANGED:
4066 		ctrl->aen_result = result;
4067 		break;
4068 	default:
4069 		dev_warn(ctrl->device, "async event result %08x\n", result);
4070 	}
4071 }
4072 
4073 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4074 		volatile union nvme_result *res)
4075 {
4076 	u32 result = le32_to_cpu(res->u32);
4077 	u32 aer_type = result & 0x07;
4078 
4079 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4080 		return;
4081 
4082 	switch (aer_type) {
4083 	case NVME_AER_NOTICE:
4084 		nvme_handle_aen_notice(ctrl, result);
4085 		break;
4086 	case NVME_AER_ERROR:
4087 	case NVME_AER_SMART:
4088 	case NVME_AER_CSS:
4089 	case NVME_AER_VS:
4090 		trace_nvme_async_event(ctrl, aer_type);
4091 		ctrl->aen_result = result;
4092 		break;
4093 	default:
4094 		break;
4095 	}
4096 	queue_work(nvme_wq, &ctrl->async_event_work);
4097 }
4098 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4099 
4100 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4101 {
4102 	nvme_mpath_stop(ctrl);
4103 	nvme_stop_keep_alive(ctrl);
4104 	flush_work(&ctrl->async_event_work);
4105 	cancel_work_sync(&ctrl->fw_act_work);
4106 }
4107 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4108 
4109 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4110 {
4111 	if (ctrl->kato)
4112 		nvme_start_keep_alive(ctrl);
4113 
4114 	nvme_enable_aen(ctrl);
4115 
4116 	if (ctrl->queue_count > 1) {
4117 		nvme_queue_scan(ctrl);
4118 		nvme_start_queues(ctrl);
4119 	}
4120 	ctrl->created = true;
4121 }
4122 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4123 
4124 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4125 {
4126 	nvme_fault_inject_fini(&ctrl->fault_inject);
4127 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4128 	cdev_device_del(&ctrl->cdev, ctrl->device);
4129 	nvme_put_ctrl(ctrl);
4130 }
4131 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4132 
4133 static void nvme_free_ctrl(struct device *dev)
4134 {
4135 	struct nvme_ctrl *ctrl =
4136 		container_of(dev, struct nvme_ctrl, ctrl_device);
4137 	struct nvme_subsystem *subsys = ctrl->subsys;
4138 
4139 	if (subsys && ctrl->instance != subsys->instance)
4140 		ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4141 
4142 	kfree(ctrl->effects);
4143 	nvme_mpath_uninit(ctrl);
4144 	__free_page(ctrl->discard_page);
4145 
4146 	if (subsys) {
4147 		mutex_lock(&nvme_subsystems_lock);
4148 		list_del(&ctrl->subsys_entry);
4149 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4150 		mutex_unlock(&nvme_subsystems_lock);
4151 	}
4152 
4153 	ctrl->ops->free_ctrl(ctrl);
4154 
4155 	if (subsys)
4156 		nvme_put_subsystem(subsys);
4157 }
4158 
4159 /*
4160  * Initialize a NVMe controller structures.  This needs to be called during
4161  * earliest initialization so that we have the initialized structured around
4162  * during probing.
4163  */
4164 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4165 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
4166 {
4167 	int ret;
4168 
4169 	ctrl->state = NVME_CTRL_NEW;
4170 	spin_lock_init(&ctrl->lock);
4171 	mutex_init(&ctrl->scan_lock);
4172 	INIT_LIST_HEAD(&ctrl->namespaces);
4173 	init_rwsem(&ctrl->namespaces_rwsem);
4174 	ctrl->dev = dev;
4175 	ctrl->ops = ops;
4176 	ctrl->quirks = quirks;
4177 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4178 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4179 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4180 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4181 	init_waitqueue_head(&ctrl->state_wq);
4182 
4183 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4184 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4185 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4186 
4187 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4188 			PAGE_SIZE);
4189 	ctrl->discard_page = alloc_page(GFP_KERNEL);
4190 	if (!ctrl->discard_page) {
4191 		ret = -ENOMEM;
4192 		goto out;
4193 	}
4194 
4195 	ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4196 	if (ret < 0)
4197 		goto out;
4198 	ctrl->instance = ret;
4199 
4200 	device_initialize(&ctrl->ctrl_device);
4201 	ctrl->device = &ctrl->ctrl_device;
4202 	ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4203 	ctrl->device->class = nvme_class;
4204 	ctrl->device->parent = ctrl->dev;
4205 	ctrl->device->groups = nvme_dev_attr_groups;
4206 	ctrl->device->release = nvme_free_ctrl;
4207 	dev_set_drvdata(ctrl->device, ctrl);
4208 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4209 	if (ret)
4210 		goto out_release_instance;
4211 
4212 	nvme_get_ctrl(ctrl);
4213 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
4214 	ctrl->cdev.owner = ops->module;
4215 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4216 	if (ret)
4217 		goto out_free_name;
4218 
4219 	/*
4220 	 * Initialize latency tolerance controls.  The sysfs files won't
4221 	 * be visible to userspace unless the device actually supports APST.
4222 	 */
4223 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4224 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4225 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4226 
4227 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4228 
4229 	return 0;
4230 out_free_name:
4231 	nvme_put_ctrl(ctrl);
4232 	kfree_const(ctrl->device->kobj.name);
4233 out_release_instance:
4234 	ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4235 out:
4236 	if (ctrl->discard_page)
4237 		__free_page(ctrl->discard_page);
4238 	return ret;
4239 }
4240 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4241 
4242 /**
4243  * nvme_kill_queues(): Ends all namespace queues
4244  * @ctrl: the dead controller that needs to end
4245  *
4246  * Call this function when the driver determines it is unable to get the
4247  * controller in a state capable of servicing IO.
4248  */
4249 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4250 {
4251 	struct nvme_ns *ns;
4252 
4253 	down_read(&ctrl->namespaces_rwsem);
4254 
4255 	/* Forcibly unquiesce queues to avoid blocking dispatch */
4256 	if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4257 		blk_mq_unquiesce_queue(ctrl->admin_q);
4258 
4259 	list_for_each_entry(ns, &ctrl->namespaces, list)
4260 		nvme_set_queue_dying(ns);
4261 
4262 	up_read(&ctrl->namespaces_rwsem);
4263 }
4264 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4265 
4266 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4267 {
4268 	struct nvme_ns *ns;
4269 
4270 	down_read(&ctrl->namespaces_rwsem);
4271 	list_for_each_entry(ns, &ctrl->namespaces, list)
4272 		blk_mq_unfreeze_queue(ns->queue);
4273 	up_read(&ctrl->namespaces_rwsem);
4274 }
4275 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4276 
4277 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4278 {
4279 	struct nvme_ns *ns;
4280 
4281 	down_read(&ctrl->namespaces_rwsem);
4282 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4283 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4284 		if (timeout <= 0)
4285 			break;
4286 	}
4287 	up_read(&ctrl->namespaces_rwsem);
4288 }
4289 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4290 
4291 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4292 {
4293 	struct nvme_ns *ns;
4294 
4295 	down_read(&ctrl->namespaces_rwsem);
4296 	list_for_each_entry(ns, &ctrl->namespaces, list)
4297 		blk_mq_freeze_queue_wait(ns->queue);
4298 	up_read(&ctrl->namespaces_rwsem);
4299 }
4300 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4301 
4302 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4303 {
4304 	struct nvme_ns *ns;
4305 
4306 	down_read(&ctrl->namespaces_rwsem);
4307 	list_for_each_entry(ns, &ctrl->namespaces, list)
4308 		blk_freeze_queue_start(ns->queue);
4309 	up_read(&ctrl->namespaces_rwsem);
4310 }
4311 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4312 
4313 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4314 {
4315 	struct nvme_ns *ns;
4316 
4317 	down_read(&ctrl->namespaces_rwsem);
4318 	list_for_each_entry(ns, &ctrl->namespaces, list)
4319 		blk_mq_quiesce_queue(ns->queue);
4320 	up_read(&ctrl->namespaces_rwsem);
4321 }
4322 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4323 
4324 void nvme_start_queues(struct nvme_ctrl *ctrl)
4325 {
4326 	struct nvme_ns *ns;
4327 
4328 	down_read(&ctrl->namespaces_rwsem);
4329 	list_for_each_entry(ns, &ctrl->namespaces, list)
4330 		blk_mq_unquiesce_queue(ns->queue);
4331 	up_read(&ctrl->namespaces_rwsem);
4332 }
4333 EXPORT_SYMBOL_GPL(nvme_start_queues);
4334 
4335 
4336 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4337 {
4338 	struct nvme_ns *ns;
4339 
4340 	down_read(&ctrl->namespaces_rwsem);
4341 	list_for_each_entry(ns, &ctrl->namespaces, list)
4342 		blk_sync_queue(ns->queue);
4343 	up_read(&ctrl->namespaces_rwsem);
4344 
4345 	if (ctrl->admin_q)
4346 		blk_sync_queue(ctrl->admin_q);
4347 }
4348 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4349 
4350 /*
4351  * Check we didn't inadvertently grow the command structure sizes:
4352  */
4353 static inline void _nvme_check_size(void)
4354 {
4355 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4356 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4357 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4358 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4359 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4360 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4361 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4362 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4363 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4364 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4365 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4366 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4367 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4368 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4369 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4370 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4371 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4372 }
4373 
4374 
4375 static int __init nvme_core_init(void)
4376 {
4377 	int result = -ENOMEM;
4378 
4379 	_nvme_check_size();
4380 
4381 	nvme_wq = alloc_workqueue("nvme-wq",
4382 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4383 	if (!nvme_wq)
4384 		goto out;
4385 
4386 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4387 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4388 	if (!nvme_reset_wq)
4389 		goto destroy_wq;
4390 
4391 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4392 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4393 	if (!nvme_delete_wq)
4394 		goto destroy_reset_wq;
4395 
4396 	result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4397 	if (result < 0)
4398 		goto destroy_delete_wq;
4399 
4400 	nvme_class = class_create(THIS_MODULE, "nvme");
4401 	if (IS_ERR(nvme_class)) {
4402 		result = PTR_ERR(nvme_class);
4403 		goto unregister_chrdev;
4404 	}
4405 	nvme_class->dev_uevent = nvme_class_uevent;
4406 
4407 	nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4408 	if (IS_ERR(nvme_subsys_class)) {
4409 		result = PTR_ERR(nvme_subsys_class);
4410 		goto destroy_class;
4411 	}
4412 	return 0;
4413 
4414 destroy_class:
4415 	class_destroy(nvme_class);
4416 unregister_chrdev:
4417 	unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4418 destroy_delete_wq:
4419 	destroy_workqueue(nvme_delete_wq);
4420 destroy_reset_wq:
4421 	destroy_workqueue(nvme_reset_wq);
4422 destroy_wq:
4423 	destroy_workqueue(nvme_wq);
4424 out:
4425 	return result;
4426 }
4427 
4428 static void __exit nvme_core_exit(void)
4429 {
4430 	class_destroy(nvme_subsys_class);
4431 	class_destroy(nvme_class);
4432 	unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4433 	destroy_workqueue(nvme_delete_wq);
4434 	destroy_workqueue(nvme_reset_wq);
4435 	destroy_workqueue(nvme_wq);
4436 	ida_destroy(&nvme_instance_ida);
4437 }
4438 
4439 MODULE_LICENSE("GPL");
4440 MODULE_VERSION("1.0");
4441 module_init(nvme_core_init);
4442 module_exit(nvme_core_exit);
4443