xref: /openbmc/linux/drivers/nvme/host/core.c (revision b830f94f)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * NVM Express device driver
4  * Copyright (c) 2011-2014, Intel Corporation.
5  */
6 
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/delay.h>
10 #include <linux/errno.h>
11 #include <linux/hdreg.h>
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/list_sort.h>
16 #include <linux/slab.h>
17 #include <linux/types.h>
18 #include <linux/pr.h>
19 #include <linux/ptrace.h>
20 #include <linux/nvme_ioctl.h>
21 #include <linux/t10-pi.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
24 
25 #define CREATE_TRACE_POINTS
26 #include "trace.h"
27 
28 #include "nvme.h"
29 #include "fabrics.h"
30 
31 #define NVME_MINORS		(1U << MINORBITS)
32 
33 unsigned int admin_timeout = 60;
34 module_param(admin_timeout, uint, 0644);
35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
36 EXPORT_SYMBOL_GPL(admin_timeout);
37 
38 unsigned int nvme_io_timeout = 30;
39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
42 
43 static unsigned char shutdown_timeout = 5;
44 module_param(shutdown_timeout, byte, 0644);
45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
46 
47 static u8 nvme_max_retries = 5;
48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
50 
51 static unsigned long default_ps_max_latency_us = 100000;
52 module_param(default_ps_max_latency_us, ulong, 0644);
53 MODULE_PARM_DESC(default_ps_max_latency_us,
54 		 "max power saving latency for new devices; use PM QOS to change per device");
55 
56 static bool force_apst;
57 module_param(force_apst, bool, 0644);
58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
59 
60 static bool streams;
61 module_param(streams, bool, 0644);
62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
63 
64 /*
65  * nvme_wq - hosts nvme related works that are not reset or delete
66  * nvme_reset_wq - hosts nvme reset works
67  * nvme_delete_wq - hosts nvme delete works
68  *
69  * nvme_wq will host works such are scan, aen handling, fw activation,
70  * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
71  * runs reset works which also flush works hosted on nvme_wq for
72  * serialization purposes. nvme_delete_wq host controller deletion
73  * works which flush reset works for serialization.
74  */
75 struct workqueue_struct *nvme_wq;
76 EXPORT_SYMBOL_GPL(nvme_wq);
77 
78 struct workqueue_struct *nvme_reset_wq;
79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
80 
81 struct workqueue_struct *nvme_delete_wq;
82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
83 
84 static DEFINE_IDA(nvme_subsystems_ida);
85 static LIST_HEAD(nvme_subsystems);
86 static DEFINE_MUTEX(nvme_subsystems_lock);
87 
88 static DEFINE_IDA(nvme_instance_ida);
89 static dev_t nvme_chr_devt;
90 static struct class *nvme_class;
91 static struct class *nvme_subsys_class;
92 
93 static int nvme_revalidate_disk(struct gendisk *disk);
94 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
95 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
96 					   unsigned nsid);
97 
98 static void nvme_set_queue_dying(struct nvme_ns *ns)
99 {
100 	/*
101 	 * Revalidating a dead namespace sets capacity to 0. This will end
102 	 * buffered writers dirtying pages that can't be synced.
103 	 */
104 	if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
105 		return;
106 	revalidate_disk(ns->disk);
107 	blk_set_queue_dying(ns->queue);
108 	/* Forcibly unquiesce queues to avoid blocking dispatch */
109 	blk_mq_unquiesce_queue(ns->queue);
110 }
111 
112 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
113 {
114 	/*
115 	 * Only new queue scan work when admin and IO queues are both alive
116 	 */
117 	if (ctrl->state == NVME_CTRL_LIVE)
118 		queue_work(nvme_wq, &ctrl->scan_work);
119 }
120 
121 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
122 {
123 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
124 		return -EBUSY;
125 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
126 		return -EBUSY;
127 	return 0;
128 }
129 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
130 
131 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
132 {
133 	int ret;
134 
135 	ret = nvme_reset_ctrl(ctrl);
136 	if (!ret) {
137 		flush_work(&ctrl->reset_work);
138 		if (ctrl->state != NVME_CTRL_LIVE &&
139 		    ctrl->state != NVME_CTRL_ADMIN_ONLY)
140 			ret = -ENETRESET;
141 	}
142 
143 	return ret;
144 }
145 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
146 
147 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
148 {
149 	dev_info(ctrl->device,
150 		 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
151 
152 	flush_work(&ctrl->reset_work);
153 	nvme_stop_ctrl(ctrl);
154 	nvme_remove_namespaces(ctrl);
155 	ctrl->ops->delete_ctrl(ctrl);
156 	nvme_uninit_ctrl(ctrl);
157 	nvme_put_ctrl(ctrl);
158 }
159 
160 static void nvme_delete_ctrl_work(struct work_struct *work)
161 {
162 	struct nvme_ctrl *ctrl =
163 		container_of(work, struct nvme_ctrl, delete_work);
164 
165 	nvme_do_delete_ctrl(ctrl);
166 }
167 
168 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
169 {
170 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
171 		return -EBUSY;
172 	if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
173 		return -EBUSY;
174 	return 0;
175 }
176 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
177 
178 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
179 {
180 	int ret = 0;
181 
182 	/*
183 	 * Keep a reference until nvme_do_delete_ctrl() complete,
184 	 * since ->delete_ctrl can free the controller.
185 	 */
186 	nvme_get_ctrl(ctrl);
187 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
188 		ret = -EBUSY;
189 	if (!ret)
190 		nvme_do_delete_ctrl(ctrl);
191 	nvme_put_ctrl(ctrl);
192 	return ret;
193 }
194 
195 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
196 {
197 	return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
198 }
199 
200 static blk_status_t nvme_error_status(struct request *req)
201 {
202 	switch (nvme_req(req)->status & 0x7ff) {
203 	case NVME_SC_SUCCESS:
204 		return BLK_STS_OK;
205 	case NVME_SC_CAP_EXCEEDED:
206 		return BLK_STS_NOSPC;
207 	case NVME_SC_LBA_RANGE:
208 		return BLK_STS_TARGET;
209 	case NVME_SC_BAD_ATTRIBUTES:
210 	case NVME_SC_ONCS_NOT_SUPPORTED:
211 	case NVME_SC_INVALID_OPCODE:
212 	case NVME_SC_INVALID_FIELD:
213 	case NVME_SC_INVALID_NS:
214 		return BLK_STS_NOTSUPP;
215 	case NVME_SC_WRITE_FAULT:
216 	case NVME_SC_READ_ERROR:
217 	case NVME_SC_UNWRITTEN_BLOCK:
218 	case NVME_SC_ACCESS_DENIED:
219 	case NVME_SC_READ_ONLY:
220 	case NVME_SC_COMPARE_FAILED:
221 		return BLK_STS_MEDIUM;
222 	case NVME_SC_GUARD_CHECK:
223 	case NVME_SC_APPTAG_CHECK:
224 	case NVME_SC_REFTAG_CHECK:
225 	case NVME_SC_INVALID_PI:
226 		return BLK_STS_PROTECTION;
227 	case NVME_SC_RESERVATION_CONFLICT:
228 		return BLK_STS_NEXUS;
229 	default:
230 		return BLK_STS_IOERR;
231 	}
232 }
233 
234 static inline bool nvme_req_needs_retry(struct request *req)
235 {
236 	if (blk_noretry_request(req))
237 		return false;
238 	if (nvme_req(req)->status & NVME_SC_DNR)
239 		return false;
240 	if (nvme_req(req)->retries >= nvme_max_retries)
241 		return false;
242 	return true;
243 }
244 
245 static void nvme_retry_req(struct request *req)
246 {
247 	struct nvme_ns *ns = req->q->queuedata;
248 	unsigned long delay = 0;
249 	u16 crd;
250 
251 	/* The mask and shift result must be <= 3 */
252 	crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
253 	if (ns && crd)
254 		delay = ns->ctrl->crdt[crd - 1] * 100;
255 
256 	nvme_req(req)->retries++;
257 	blk_mq_requeue_request(req, false);
258 	blk_mq_delay_kick_requeue_list(req->q, delay);
259 }
260 
261 void nvme_complete_rq(struct request *req)
262 {
263 	blk_status_t status = nvme_error_status(req);
264 
265 	trace_nvme_complete_rq(req);
266 
267 	if (nvme_req(req)->ctrl->kas)
268 		nvme_req(req)->ctrl->comp_seen = true;
269 
270 	if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
271 		if ((req->cmd_flags & REQ_NVME_MPATH) &&
272 		    blk_path_error(status)) {
273 			nvme_failover_req(req);
274 			return;
275 		}
276 
277 		if (!blk_queue_dying(req->q)) {
278 			nvme_retry_req(req);
279 			return;
280 		}
281 	}
282 	blk_mq_end_request(req, status);
283 }
284 EXPORT_SYMBOL_GPL(nvme_complete_rq);
285 
286 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
287 {
288 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
289 				"Cancelling I/O %d", req->tag);
290 
291 	nvme_req(req)->status = NVME_SC_ABORT_REQ;
292 	blk_mq_complete_request_sync(req);
293 	return true;
294 }
295 EXPORT_SYMBOL_GPL(nvme_cancel_request);
296 
297 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
298 		enum nvme_ctrl_state new_state)
299 {
300 	enum nvme_ctrl_state old_state;
301 	unsigned long flags;
302 	bool changed = false;
303 
304 	spin_lock_irqsave(&ctrl->lock, flags);
305 
306 	old_state = ctrl->state;
307 	switch (new_state) {
308 	case NVME_CTRL_ADMIN_ONLY:
309 		switch (old_state) {
310 		case NVME_CTRL_CONNECTING:
311 			changed = true;
312 			/* FALLTHRU */
313 		default:
314 			break;
315 		}
316 		break;
317 	case NVME_CTRL_LIVE:
318 		switch (old_state) {
319 		case NVME_CTRL_NEW:
320 		case NVME_CTRL_RESETTING:
321 		case NVME_CTRL_CONNECTING:
322 			changed = true;
323 			/* FALLTHRU */
324 		default:
325 			break;
326 		}
327 		break;
328 	case NVME_CTRL_RESETTING:
329 		switch (old_state) {
330 		case NVME_CTRL_NEW:
331 		case NVME_CTRL_LIVE:
332 		case NVME_CTRL_ADMIN_ONLY:
333 			changed = true;
334 			/* FALLTHRU */
335 		default:
336 			break;
337 		}
338 		break;
339 	case NVME_CTRL_CONNECTING:
340 		switch (old_state) {
341 		case NVME_CTRL_NEW:
342 		case NVME_CTRL_RESETTING:
343 			changed = true;
344 			/* FALLTHRU */
345 		default:
346 			break;
347 		}
348 		break;
349 	case NVME_CTRL_DELETING:
350 		switch (old_state) {
351 		case NVME_CTRL_LIVE:
352 		case NVME_CTRL_ADMIN_ONLY:
353 		case NVME_CTRL_RESETTING:
354 		case NVME_CTRL_CONNECTING:
355 			changed = true;
356 			/* FALLTHRU */
357 		default:
358 			break;
359 		}
360 		break;
361 	case NVME_CTRL_DEAD:
362 		switch (old_state) {
363 		case NVME_CTRL_DELETING:
364 			changed = true;
365 			/* FALLTHRU */
366 		default:
367 			break;
368 		}
369 		break;
370 	default:
371 		break;
372 	}
373 
374 	if (changed)
375 		ctrl->state = new_state;
376 
377 	spin_unlock_irqrestore(&ctrl->lock, flags);
378 	if (changed && ctrl->state == NVME_CTRL_LIVE)
379 		nvme_kick_requeue_lists(ctrl);
380 	return changed;
381 }
382 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
383 
384 static void nvme_free_ns_head(struct kref *ref)
385 {
386 	struct nvme_ns_head *head =
387 		container_of(ref, struct nvme_ns_head, ref);
388 
389 	nvme_mpath_remove_disk(head);
390 	ida_simple_remove(&head->subsys->ns_ida, head->instance);
391 	list_del_init(&head->entry);
392 	cleanup_srcu_struct(&head->srcu);
393 	nvme_put_subsystem(head->subsys);
394 	kfree(head);
395 }
396 
397 static void nvme_put_ns_head(struct nvme_ns_head *head)
398 {
399 	kref_put(&head->ref, nvme_free_ns_head);
400 }
401 
402 static void nvme_free_ns(struct kref *kref)
403 {
404 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
405 
406 	if (ns->ndev)
407 		nvme_nvm_unregister(ns);
408 
409 	put_disk(ns->disk);
410 	nvme_put_ns_head(ns->head);
411 	nvme_put_ctrl(ns->ctrl);
412 	kfree(ns);
413 }
414 
415 static void nvme_put_ns(struct nvme_ns *ns)
416 {
417 	kref_put(&ns->kref, nvme_free_ns);
418 }
419 
420 static inline void nvme_clear_nvme_request(struct request *req)
421 {
422 	if (!(req->rq_flags & RQF_DONTPREP)) {
423 		nvme_req(req)->retries = 0;
424 		nvme_req(req)->flags = 0;
425 		req->rq_flags |= RQF_DONTPREP;
426 	}
427 }
428 
429 struct request *nvme_alloc_request(struct request_queue *q,
430 		struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
431 {
432 	unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
433 	struct request *req;
434 
435 	if (qid == NVME_QID_ANY) {
436 		req = blk_mq_alloc_request(q, op, flags);
437 	} else {
438 		req = blk_mq_alloc_request_hctx(q, op, flags,
439 				qid ? qid - 1 : 0);
440 	}
441 	if (IS_ERR(req))
442 		return req;
443 
444 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
445 	nvme_clear_nvme_request(req);
446 	nvme_req(req)->cmd = cmd;
447 
448 	return req;
449 }
450 EXPORT_SYMBOL_GPL(nvme_alloc_request);
451 
452 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
453 {
454 	struct nvme_command c;
455 
456 	memset(&c, 0, sizeof(c));
457 
458 	c.directive.opcode = nvme_admin_directive_send;
459 	c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
460 	c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
461 	c.directive.dtype = NVME_DIR_IDENTIFY;
462 	c.directive.tdtype = NVME_DIR_STREAMS;
463 	c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
464 
465 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
466 }
467 
468 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
469 {
470 	return nvme_toggle_streams(ctrl, false);
471 }
472 
473 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
474 {
475 	return nvme_toggle_streams(ctrl, true);
476 }
477 
478 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
479 				  struct streams_directive_params *s, u32 nsid)
480 {
481 	struct nvme_command c;
482 
483 	memset(&c, 0, sizeof(c));
484 	memset(s, 0, sizeof(*s));
485 
486 	c.directive.opcode = nvme_admin_directive_recv;
487 	c.directive.nsid = cpu_to_le32(nsid);
488 	c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
489 	c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
490 	c.directive.dtype = NVME_DIR_STREAMS;
491 
492 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
493 }
494 
495 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
496 {
497 	struct streams_directive_params s;
498 	int ret;
499 
500 	if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
501 		return 0;
502 	if (!streams)
503 		return 0;
504 
505 	ret = nvme_enable_streams(ctrl);
506 	if (ret)
507 		return ret;
508 
509 	ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
510 	if (ret)
511 		return ret;
512 
513 	ctrl->nssa = le16_to_cpu(s.nssa);
514 	if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
515 		dev_info(ctrl->device, "too few streams (%u) available\n",
516 					ctrl->nssa);
517 		nvme_disable_streams(ctrl);
518 		return 0;
519 	}
520 
521 	ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
522 	dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
523 	return 0;
524 }
525 
526 /*
527  * Check if 'req' has a write hint associated with it. If it does, assign
528  * a valid namespace stream to the write.
529  */
530 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
531 				     struct request *req, u16 *control,
532 				     u32 *dsmgmt)
533 {
534 	enum rw_hint streamid = req->write_hint;
535 
536 	if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
537 		streamid = 0;
538 	else {
539 		streamid--;
540 		if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
541 			return;
542 
543 		*control |= NVME_RW_DTYPE_STREAMS;
544 		*dsmgmt |= streamid << 16;
545 	}
546 
547 	if (streamid < ARRAY_SIZE(req->q->write_hints))
548 		req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
549 }
550 
551 static inline void nvme_setup_flush(struct nvme_ns *ns,
552 		struct nvme_command *cmnd)
553 {
554 	cmnd->common.opcode = nvme_cmd_flush;
555 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
556 }
557 
558 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
559 		struct nvme_command *cmnd)
560 {
561 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
562 	struct nvme_dsm_range *range;
563 	struct bio *bio;
564 
565 	range = kmalloc_array(segments, sizeof(*range),
566 				GFP_ATOMIC | __GFP_NOWARN);
567 	if (!range) {
568 		/*
569 		 * If we fail allocation our range, fallback to the controller
570 		 * discard page. If that's also busy, it's safe to return
571 		 * busy, as we know we can make progress once that's freed.
572 		 */
573 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
574 			return BLK_STS_RESOURCE;
575 
576 		range = page_address(ns->ctrl->discard_page);
577 	}
578 
579 	__rq_for_each_bio(bio, req) {
580 		u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
581 		u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
582 
583 		if (n < segments) {
584 			range[n].cattr = cpu_to_le32(0);
585 			range[n].nlb = cpu_to_le32(nlb);
586 			range[n].slba = cpu_to_le64(slba);
587 		}
588 		n++;
589 	}
590 
591 	if (WARN_ON_ONCE(n != segments)) {
592 		if (virt_to_page(range) == ns->ctrl->discard_page)
593 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
594 		else
595 			kfree(range);
596 		return BLK_STS_IOERR;
597 	}
598 
599 	cmnd->dsm.opcode = nvme_cmd_dsm;
600 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
601 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
602 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
603 
604 	req->special_vec.bv_page = virt_to_page(range);
605 	req->special_vec.bv_offset = offset_in_page(range);
606 	req->special_vec.bv_len = sizeof(*range) * segments;
607 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
608 
609 	return BLK_STS_OK;
610 }
611 
612 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
613 		struct request *req, struct nvme_command *cmnd)
614 {
615 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
616 		return nvme_setup_discard(ns, req, cmnd);
617 
618 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
619 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
620 	cmnd->write_zeroes.slba =
621 		cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
622 	cmnd->write_zeroes.length =
623 		cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
624 	cmnd->write_zeroes.control = 0;
625 	return BLK_STS_OK;
626 }
627 
628 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
629 		struct request *req, struct nvme_command *cmnd)
630 {
631 	struct nvme_ctrl *ctrl = ns->ctrl;
632 	u16 control = 0;
633 	u32 dsmgmt = 0;
634 
635 	if (req->cmd_flags & REQ_FUA)
636 		control |= NVME_RW_FUA;
637 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
638 		control |= NVME_RW_LR;
639 
640 	if (req->cmd_flags & REQ_RAHEAD)
641 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
642 
643 	cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
644 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
645 	cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
646 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
647 
648 	if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
649 		nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
650 
651 	if (ns->ms) {
652 		/*
653 		 * If formated with metadata, the block layer always provides a
654 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
655 		 * we enable the PRACT bit for protection information or set the
656 		 * namespace capacity to zero to prevent any I/O.
657 		 */
658 		if (!blk_integrity_rq(req)) {
659 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
660 				return BLK_STS_NOTSUPP;
661 			control |= NVME_RW_PRINFO_PRACT;
662 		} else if (req_op(req) == REQ_OP_WRITE) {
663 			t10_pi_prepare(req, ns->pi_type);
664 		}
665 
666 		switch (ns->pi_type) {
667 		case NVME_NS_DPS_PI_TYPE3:
668 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
669 			break;
670 		case NVME_NS_DPS_PI_TYPE1:
671 		case NVME_NS_DPS_PI_TYPE2:
672 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
673 					NVME_RW_PRINFO_PRCHK_REF;
674 			cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
675 			break;
676 		}
677 	}
678 
679 	cmnd->rw.control = cpu_to_le16(control);
680 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
681 	return 0;
682 }
683 
684 void nvme_cleanup_cmd(struct request *req)
685 {
686 	if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
687 	    nvme_req(req)->status == 0) {
688 		struct nvme_ns *ns = req->rq_disk->private_data;
689 
690 		t10_pi_complete(req, ns->pi_type,
691 				blk_rq_bytes(req) >> ns->lba_shift);
692 	}
693 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
694 		struct nvme_ns *ns = req->rq_disk->private_data;
695 		struct page *page = req->special_vec.bv_page;
696 
697 		if (page == ns->ctrl->discard_page)
698 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
699 		else
700 			kfree(page_address(page) + req->special_vec.bv_offset);
701 	}
702 }
703 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
704 
705 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
706 		struct nvme_command *cmd)
707 {
708 	blk_status_t ret = BLK_STS_OK;
709 
710 	nvme_clear_nvme_request(req);
711 
712 	memset(cmd, 0, sizeof(*cmd));
713 	switch (req_op(req)) {
714 	case REQ_OP_DRV_IN:
715 	case REQ_OP_DRV_OUT:
716 		memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
717 		break;
718 	case REQ_OP_FLUSH:
719 		nvme_setup_flush(ns, cmd);
720 		break;
721 	case REQ_OP_WRITE_ZEROES:
722 		ret = nvme_setup_write_zeroes(ns, req, cmd);
723 		break;
724 	case REQ_OP_DISCARD:
725 		ret = nvme_setup_discard(ns, req, cmd);
726 		break;
727 	case REQ_OP_READ:
728 	case REQ_OP_WRITE:
729 		ret = nvme_setup_rw(ns, req, cmd);
730 		break;
731 	default:
732 		WARN_ON_ONCE(1);
733 		return BLK_STS_IOERR;
734 	}
735 
736 	cmd->common.command_id = req->tag;
737 	trace_nvme_setup_cmd(req, cmd);
738 	return ret;
739 }
740 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
741 
742 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
743 {
744 	struct completion *waiting = rq->end_io_data;
745 
746 	rq->end_io_data = NULL;
747 	complete(waiting);
748 }
749 
750 static void nvme_execute_rq_polled(struct request_queue *q,
751 		struct gendisk *bd_disk, struct request *rq, int at_head)
752 {
753 	DECLARE_COMPLETION_ONSTACK(wait);
754 
755 	WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
756 
757 	rq->cmd_flags |= REQ_HIPRI;
758 	rq->end_io_data = &wait;
759 	blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
760 
761 	while (!completion_done(&wait)) {
762 		blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
763 		cond_resched();
764 	}
765 }
766 
767 /*
768  * Returns 0 on success.  If the result is negative, it's a Linux error code;
769  * if the result is positive, it's an NVM Express status code
770  */
771 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
772 		union nvme_result *result, void *buffer, unsigned bufflen,
773 		unsigned timeout, int qid, int at_head,
774 		blk_mq_req_flags_t flags, bool poll)
775 {
776 	struct request *req;
777 	int ret;
778 
779 	req = nvme_alloc_request(q, cmd, flags, qid);
780 	if (IS_ERR(req))
781 		return PTR_ERR(req);
782 
783 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
784 
785 	if (buffer && bufflen) {
786 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
787 		if (ret)
788 			goto out;
789 	}
790 
791 	if (poll)
792 		nvme_execute_rq_polled(req->q, NULL, req, at_head);
793 	else
794 		blk_execute_rq(req->q, NULL, req, at_head);
795 	if (result)
796 		*result = nvme_req(req)->result;
797 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
798 		ret = -EINTR;
799 	else
800 		ret = nvme_req(req)->status;
801  out:
802 	blk_mq_free_request(req);
803 	return ret;
804 }
805 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
806 
807 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
808 		void *buffer, unsigned bufflen)
809 {
810 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
811 			NVME_QID_ANY, 0, 0, false);
812 }
813 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
814 
815 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
816 		unsigned len, u32 seed, bool write)
817 {
818 	struct bio_integrity_payload *bip;
819 	int ret = -ENOMEM;
820 	void *buf;
821 
822 	buf = kmalloc(len, GFP_KERNEL);
823 	if (!buf)
824 		goto out;
825 
826 	ret = -EFAULT;
827 	if (write && copy_from_user(buf, ubuf, len))
828 		goto out_free_meta;
829 
830 	bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
831 	if (IS_ERR(bip)) {
832 		ret = PTR_ERR(bip);
833 		goto out_free_meta;
834 	}
835 
836 	bip->bip_iter.bi_size = len;
837 	bip->bip_iter.bi_sector = seed;
838 	ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
839 			offset_in_page(buf));
840 	if (ret == len)
841 		return buf;
842 	ret = -ENOMEM;
843 out_free_meta:
844 	kfree(buf);
845 out:
846 	return ERR_PTR(ret);
847 }
848 
849 static int nvme_submit_user_cmd(struct request_queue *q,
850 		struct nvme_command *cmd, void __user *ubuffer,
851 		unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
852 		u32 meta_seed, u32 *result, unsigned timeout)
853 {
854 	bool write = nvme_is_write(cmd);
855 	struct nvme_ns *ns = q->queuedata;
856 	struct gendisk *disk = ns ? ns->disk : NULL;
857 	struct request *req;
858 	struct bio *bio = NULL;
859 	void *meta = NULL;
860 	int ret;
861 
862 	req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
863 	if (IS_ERR(req))
864 		return PTR_ERR(req);
865 
866 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
867 	nvme_req(req)->flags |= NVME_REQ_USERCMD;
868 
869 	if (ubuffer && bufflen) {
870 		ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
871 				GFP_KERNEL);
872 		if (ret)
873 			goto out;
874 		bio = req->bio;
875 		bio->bi_disk = disk;
876 		if (disk && meta_buffer && meta_len) {
877 			meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
878 					meta_seed, write);
879 			if (IS_ERR(meta)) {
880 				ret = PTR_ERR(meta);
881 				goto out_unmap;
882 			}
883 			req->cmd_flags |= REQ_INTEGRITY;
884 		}
885 	}
886 
887 	blk_execute_rq(req->q, disk, req, 0);
888 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
889 		ret = -EINTR;
890 	else
891 		ret = nvme_req(req)->status;
892 	if (result)
893 		*result = le32_to_cpu(nvme_req(req)->result.u32);
894 	if (meta && !ret && !write) {
895 		if (copy_to_user(meta_buffer, meta, meta_len))
896 			ret = -EFAULT;
897 	}
898 	kfree(meta);
899  out_unmap:
900 	if (bio)
901 		blk_rq_unmap_user(bio);
902  out:
903 	blk_mq_free_request(req);
904 	return ret;
905 }
906 
907 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
908 {
909 	struct nvme_ctrl *ctrl = rq->end_io_data;
910 	unsigned long flags;
911 	bool startka = false;
912 
913 	blk_mq_free_request(rq);
914 
915 	if (status) {
916 		dev_err(ctrl->device,
917 			"failed nvme_keep_alive_end_io error=%d\n",
918 				status);
919 		return;
920 	}
921 
922 	ctrl->comp_seen = false;
923 	spin_lock_irqsave(&ctrl->lock, flags);
924 	if (ctrl->state == NVME_CTRL_LIVE ||
925 	    ctrl->state == NVME_CTRL_CONNECTING)
926 		startka = true;
927 	spin_unlock_irqrestore(&ctrl->lock, flags);
928 	if (startka)
929 		schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
930 }
931 
932 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
933 {
934 	struct request *rq;
935 
936 	rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
937 			NVME_QID_ANY);
938 	if (IS_ERR(rq))
939 		return PTR_ERR(rq);
940 
941 	rq->timeout = ctrl->kato * HZ;
942 	rq->end_io_data = ctrl;
943 
944 	blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
945 
946 	return 0;
947 }
948 
949 static void nvme_keep_alive_work(struct work_struct *work)
950 {
951 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
952 			struct nvme_ctrl, ka_work);
953 	bool comp_seen = ctrl->comp_seen;
954 
955 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
956 		dev_dbg(ctrl->device,
957 			"reschedule traffic based keep-alive timer\n");
958 		ctrl->comp_seen = false;
959 		schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
960 		return;
961 	}
962 
963 	if (nvme_keep_alive(ctrl)) {
964 		/* allocation failure, reset the controller */
965 		dev_err(ctrl->device, "keep-alive failed\n");
966 		nvme_reset_ctrl(ctrl);
967 		return;
968 	}
969 }
970 
971 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
972 {
973 	if (unlikely(ctrl->kato == 0))
974 		return;
975 
976 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
977 }
978 
979 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
980 {
981 	if (unlikely(ctrl->kato == 0))
982 		return;
983 
984 	cancel_delayed_work_sync(&ctrl->ka_work);
985 }
986 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
987 
988 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
989 {
990 	struct nvme_command c = { };
991 	int error;
992 
993 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
994 	c.identify.opcode = nvme_admin_identify;
995 	c.identify.cns = NVME_ID_CNS_CTRL;
996 
997 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
998 	if (!*id)
999 		return -ENOMEM;
1000 
1001 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1002 			sizeof(struct nvme_id_ctrl));
1003 	if (error)
1004 		kfree(*id);
1005 	return error;
1006 }
1007 
1008 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1009 		struct nvme_ns_ids *ids)
1010 {
1011 	struct nvme_command c = { };
1012 	int status;
1013 	void *data;
1014 	int pos;
1015 	int len;
1016 
1017 	c.identify.opcode = nvme_admin_identify;
1018 	c.identify.nsid = cpu_to_le32(nsid);
1019 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1020 
1021 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1022 	if (!data)
1023 		return -ENOMEM;
1024 
1025 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1026 				      NVME_IDENTIFY_DATA_SIZE);
1027 	if (status)
1028 		goto free_data;
1029 
1030 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1031 		struct nvme_ns_id_desc *cur = data + pos;
1032 
1033 		if (cur->nidl == 0)
1034 			break;
1035 
1036 		switch (cur->nidt) {
1037 		case NVME_NIDT_EUI64:
1038 			if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1039 				dev_warn(ctrl->device,
1040 					 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1041 					 cur->nidl);
1042 				goto free_data;
1043 			}
1044 			len = NVME_NIDT_EUI64_LEN;
1045 			memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1046 			break;
1047 		case NVME_NIDT_NGUID:
1048 			if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1049 				dev_warn(ctrl->device,
1050 					 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1051 					 cur->nidl);
1052 				goto free_data;
1053 			}
1054 			len = NVME_NIDT_NGUID_LEN;
1055 			memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1056 			break;
1057 		case NVME_NIDT_UUID:
1058 			if (cur->nidl != NVME_NIDT_UUID_LEN) {
1059 				dev_warn(ctrl->device,
1060 					 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1061 					 cur->nidl);
1062 				goto free_data;
1063 			}
1064 			len = NVME_NIDT_UUID_LEN;
1065 			uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1066 			break;
1067 		default:
1068 			/* Skip unknown types */
1069 			len = cur->nidl;
1070 			break;
1071 		}
1072 
1073 		len += sizeof(*cur);
1074 	}
1075 free_data:
1076 	kfree(data);
1077 	return status;
1078 }
1079 
1080 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1081 {
1082 	struct nvme_command c = { };
1083 
1084 	c.identify.opcode = nvme_admin_identify;
1085 	c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1086 	c.identify.nsid = cpu_to_le32(nsid);
1087 	return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1088 				    NVME_IDENTIFY_DATA_SIZE);
1089 }
1090 
1091 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
1092 		unsigned nsid)
1093 {
1094 	struct nvme_id_ns *id;
1095 	struct nvme_command c = { };
1096 	int error;
1097 
1098 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1099 	c.identify.opcode = nvme_admin_identify;
1100 	c.identify.nsid = cpu_to_le32(nsid);
1101 	c.identify.cns = NVME_ID_CNS_NS;
1102 
1103 	id = kmalloc(sizeof(*id), GFP_KERNEL);
1104 	if (!id)
1105 		return NULL;
1106 
1107 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1108 	if (error) {
1109 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1110 		kfree(id);
1111 		return NULL;
1112 	}
1113 
1114 	return id;
1115 }
1116 
1117 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1118 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1119 {
1120 	struct nvme_command c;
1121 	union nvme_result res;
1122 	int ret;
1123 
1124 	memset(&c, 0, sizeof(c));
1125 	c.features.opcode = op;
1126 	c.features.fid = cpu_to_le32(fid);
1127 	c.features.dword11 = cpu_to_le32(dword11);
1128 
1129 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1130 			buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1131 	if (ret >= 0 && result)
1132 		*result = le32_to_cpu(res.u32);
1133 	return ret;
1134 }
1135 
1136 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1137 		      unsigned int dword11, void *buffer, size_t buflen,
1138 		      u32 *result)
1139 {
1140 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1141 			     buflen, result);
1142 }
1143 EXPORT_SYMBOL_GPL(nvme_set_features);
1144 
1145 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1146 		      unsigned int dword11, void *buffer, size_t buflen,
1147 		      u32 *result)
1148 {
1149 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1150 			     buflen, result);
1151 }
1152 EXPORT_SYMBOL_GPL(nvme_get_features);
1153 
1154 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1155 {
1156 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1157 	u32 result;
1158 	int status, nr_io_queues;
1159 
1160 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1161 			&result);
1162 	if (status < 0)
1163 		return status;
1164 
1165 	/*
1166 	 * Degraded controllers might return an error when setting the queue
1167 	 * count.  We still want to be able to bring them online and offer
1168 	 * access to the admin queue, as that might be only way to fix them up.
1169 	 */
1170 	if (status > 0) {
1171 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1172 		*count = 0;
1173 	} else {
1174 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1175 		*count = min(*count, nr_io_queues);
1176 	}
1177 
1178 	return 0;
1179 }
1180 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1181 
1182 #define NVME_AEN_SUPPORTED \
1183 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1184 
1185 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1186 {
1187 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1188 	int status;
1189 
1190 	if (!supported_aens)
1191 		return;
1192 
1193 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1194 			NULL, 0, &result);
1195 	if (status)
1196 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1197 			 supported_aens);
1198 }
1199 
1200 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1201 {
1202 	struct nvme_user_io io;
1203 	struct nvme_command c;
1204 	unsigned length, meta_len;
1205 	void __user *metadata;
1206 
1207 	if (copy_from_user(&io, uio, sizeof(io)))
1208 		return -EFAULT;
1209 	if (io.flags)
1210 		return -EINVAL;
1211 
1212 	switch (io.opcode) {
1213 	case nvme_cmd_write:
1214 	case nvme_cmd_read:
1215 	case nvme_cmd_compare:
1216 		break;
1217 	default:
1218 		return -EINVAL;
1219 	}
1220 
1221 	length = (io.nblocks + 1) << ns->lba_shift;
1222 	meta_len = (io.nblocks + 1) * ns->ms;
1223 	metadata = (void __user *)(uintptr_t)io.metadata;
1224 
1225 	if (ns->ext) {
1226 		length += meta_len;
1227 		meta_len = 0;
1228 	} else if (meta_len) {
1229 		if ((io.metadata & 3) || !io.metadata)
1230 			return -EINVAL;
1231 	}
1232 
1233 	memset(&c, 0, sizeof(c));
1234 	c.rw.opcode = io.opcode;
1235 	c.rw.flags = io.flags;
1236 	c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1237 	c.rw.slba = cpu_to_le64(io.slba);
1238 	c.rw.length = cpu_to_le16(io.nblocks);
1239 	c.rw.control = cpu_to_le16(io.control);
1240 	c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1241 	c.rw.reftag = cpu_to_le32(io.reftag);
1242 	c.rw.apptag = cpu_to_le16(io.apptag);
1243 	c.rw.appmask = cpu_to_le16(io.appmask);
1244 
1245 	return nvme_submit_user_cmd(ns->queue, &c,
1246 			(void __user *)(uintptr_t)io.addr, length,
1247 			metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1248 }
1249 
1250 static u32 nvme_known_admin_effects(u8 opcode)
1251 {
1252 	switch (opcode) {
1253 	case nvme_admin_format_nvm:
1254 		return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1255 					NVME_CMD_EFFECTS_CSE_MASK;
1256 	case nvme_admin_sanitize_nvm:
1257 		return NVME_CMD_EFFECTS_CSE_MASK;
1258 	default:
1259 		break;
1260 	}
1261 	return 0;
1262 }
1263 
1264 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1265 								u8 opcode)
1266 {
1267 	u32 effects = 0;
1268 
1269 	if (ns) {
1270 		if (ctrl->effects)
1271 			effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1272 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1273 			dev_warn(ctrl->device,
1274 				 "IO command:%02x has unhandled effects:%08x\n",
1275 				 opcode, effects);
1276 		return 0;
1277 	}
1278 
1279 	if (ctrl->effects)
1280 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1281 	effects |= nvme_known_admin_effects(opcode);
1282 
1283 	/*
1284 	 * For simplicity, IO to all namespaces is quiesced even if the command
1285 	 * effects say only one namespace is affected.
1286 	 */
1287 	if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1288 		mutex_lock(&ctrl->scan_lock);
1289 		nvme_start_freeze(ctrl);
1290 		nvme_wait_freeze(ctrl);
1291 	}
1292 	return effects;
1293 }
1294 
1295 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1296 {
1297 	struct nvme_ns *ns;
1298 
1299 	down_read(&ctrl->namespaces_rwsem);
1300 	list_for_each_entry(ns, &ctrl->namespaces, list)
1301 		if (ns->disk && nvme_revalidate_disk(ns->disk))
1302 			nvme_set_queue_dying(ns);
1303 	up_read(&ctrl->namespaces_rwsem);
1304 
1305 	nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1306 }
1307 
1308 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1309 {
1310 	/*
1311 	 * Revalidate LBA changes prior to unfreezing. This is necessary to
1312 	 * prevent memory corruption if a logical block size was changed by
1313 	 * this command.
1314 	 */
1315 	if (effects & NVME_CMD_EFFECTS_LBCC)
1316 		nvme_update_formats(ctrl);
1317 	if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1318 		nvme_unfreeze(ctrl);
1319 		mutex_unlock(&ctrl->scan_lock);
1320 	}
1321 	if (effects & NVME_CMD_EFFECTS_CCC)
1322 		nvme_init_identify(ctrl);
1323 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1324 		nvme_queue_scan(ctrl);
1325 }
1326 
1327 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1328 			struct nvme_passthru_cmd __user *ucmd)
1329 {
1330 	struct nvme_passthru_cmd cmd;
1331 	struct nvme_command c;
1332 	unsigned timeout = 0;
1333 	u32 effects;
1334 	int status;
1335 
1336 	if (!capable(CAP_SYS_ADMIN))
1337 		return -EACCES;
1338 	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1339 		return -EFAULT;
1340 	if (cmd.flags)
1341 		return -EINVAL;
1342 
1343 	memset(&c, 0, sizeof(c));
1344 	c.common.opcode = cmd.opcode;
1345 	c.common.flags = cmd.flags;
1346 	c.common.nsid = cpu_to_le32(cmd.nsid);
1347 	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1348 	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1349 	c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1350 	c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1351 	c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1352 	c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1353 	c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1354 	c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1355 
1356 	if (cmd.timeout_ms)
1357 		timeout = msecs_to_jiffies(cmd.timeout_ms);
1358 
1359 	effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1360 	status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1361 			(void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1362 			(void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1363 			0, &cmd.result, timeout);
1364 	nvme_passthru_end(ctrl, effects);
1365 
1366 	if (status >= 0) {
1367 		if (put_user(cmd.result, &ucmd->result))
1368 			return -EFAULT;
1369 	}
1370 
1371 	return status;
1372 }
1373 
1374 /*
1375  * Issue ioctl requests on the first available path.  Note that unlike normal
1376  * block layer requests we will not retry failed request on another controller.
1377  */
1378 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1379 		struct nvme_ns_head **head, int *srcu_idx)
1380 {
1381 #ifdef CONFIG_NVME_MULTIPATH
1382 	if (disk->fops == &nvme_ns_head_ops) {
1383 		struct nvme_ns *ns;
1384 
1385 		*head = disk->private_data;
1386 		*srcu_idx = srcu_read_lock(&(*head)->srcu);
1387 		ns = nvme_find_path(*head);
1388 		if (!ns)
1389 			srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1390 		return ns;
1391 	}
1392 #endif
1393 	*head = NULL;
1394 	*srcu_idx = -1;
1395 	return disk->private_data;
1396 }
1397 
1398 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1399 {
1400 	if (head)
1401 		srcu_read_unlock(&head->srcu, idx);
1402 }
1403 
1404 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1405 		unsigned int cmd, unsigned long arg)
1406 {
1407 	struct nvme_ns_head *head = NULL;
1408 	void __user *argp = (void __user *)arg;
1409 	struct nvme_ns *ns;
1410 	int srcu_idx, ret;
1411 
1412 	ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1413 	if (unlikely(!ns))
1414 		return -EWOULDBLOCK;
1415 
1416 	/*
1417 	 * Handle ioctls that apply to the controller instead of the namespace
1418 	 * seperately and drop the ns SRCU reference early.  This avoids a
1419 	 * deadlock when deleting namespaces using the passthrough interface.
1420 	 */
1421 	if (cmd == NVME_IOCTL_ADMIN_CMD || is_sed_ioctl(cmd)) {
1422 		struct nvme_ctrl *ctrl = ns->ctrl;
1423 
1424 		nvme_get_ctrl(ns->ctrl);
1425 		nvme_put_ns_from_disk(head, srcu_idx);
1426 
1427 		if (cmd == NVME_IOCTL_ADMIN_CMD)
1428 			ret = nvme_user_cmd(ctrl, NULL, argp);
1429 		else
1430 			ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1431 
1432 		nvme_put_ctrl(ctrl);
1433 		return ret;
1434 	}
1435 
1436 	switch (cmd) {
1437 	case NVME_IOCTL_ID:
1438 		force_successful_syscall_return();
1439 		ret = ns->head->ns_id;
1440 		break;
1441 	case NVME_IOCTL_IO_CMD:
1442 		ret = nvme_user_cmd(ns->ctrl, ns, argp);
1443 		break;
1444 	case NVME_IOCTL_SUBMIT_IO:
1445 		ret = nvme_submit_io(ns, argp);
1446 		break;
1447 	default:
1448 		if (ns->ndev)
1449 			ret = nvme_nvm_ioctl(ns, cmd, arg);
1450 		else
1451 			ret = -ENOTTY;
1452 	}
1453 
1454 	nvme_put_ns_from_disk(head, srcu_idx);
1455 	return ret;
1456 }
1457 
1458 static int nvme_open(struct block_device *bdev, fmode_t mode)
1459 {
1460 	struct nvme_ns *ns = bdev->bd_disk->private_data;
1461 
1462 #ifdef CONFIG_NVME_MULTIPATH
1463 	/* should never be called due to GENHD_FL_HIDDEN */
1464 	if (WARN_ON_ONCE(ns->head->disk))
1465 		goto fail;
1466 #endif
1467 	if (!kref_get_unless_zero(&ns->kref))
1468 		goto fail;
1469 	if (!try_module_get(ns->ctrl->ops->module))
1470 		goto fail_put_ns;
1471 
1472 	return 0;
1473 
1474 fail_put_ns:
1475 	nvme_put_ns(ns);
1476 fail:
1477 	return -ENXIO;
1478 }
1479 
1480 static void nvme_release(struct gendisk *disk, fmode_t mode)
1481 {
1482 	struct nvme_ns *ns = disk->private_data;
1483 
1484 	module_put(ns->ctrl->ops->module);
1485 	nvme_put_ns(ns);
1486 }
1487 
1488 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1489 {
1490 	/* some standard values */
1491 	geo->heads = 1 << 6;
1492 	geo->sectors = 1 << 5;
1493 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1494 	return 0;
1495 }
1496 
1497 #ifdef CONFIG_BLK_DEV_INTEGRITY
1498 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1499 {
1500 	struct blk_integrity integrity;
1501 
1502 	memset(&integrity, 0, sizeof(integrity));
1503 	switch (pi_type) {
1504 	case NVME_NS_DPS_PI_TYPE3:
1505 		integrity.profile = &t10_pi_type3_crc;
1506 		integrity.tag_size = sizeof(u16) + sizeof(u32);
1507 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1508 		break;
1509 	case NVME_NS_DPS_PI_TYPE1:
1510 	case NVME_NS_DPS_PI_TYPE2:
1511 		integrity.profile = &t10_pi_type1_crc;
1512 		integrity.tag_size = sizeof(u16);
1513 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1514 		break;
1515 	default:
1516 		integrity.profile = NULL;
1517 		break;
1518 	}
1519 	integrity.tuple_size = ms;
1520 	blk_integrity_register(disk, &integrity);
1521 	blk_queue_max_integrity_segments(disk->queue, 1);
1522 }
1523 #else
1524 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1525 {
1526 }
1527 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1528 
1529 static void nvme_set_chunk_size(struct nvme_ns *ns)
1530 {
1531 	u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1532 	blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1533 }
1534 
1535 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1536 {
1537 	struct nvme_ctrl *ctrl = ns->ctrl;
1538 	struct request_queue *queue = disk->queue;
1539 	u32 size = queue_logical_block_size(queue);
1540 
1541 	if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1542 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1543 		return;
1544 	}
1545 
1546 	if (ctrl->nr_streams && ns->sws && ns->sgs)
1547 		size *= ns->sws * ns->sgs;
1548 
1549 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1550 			NVME_DSM_MAX_RANGES);
1551 
1552 	queue->limits.discard_alignment = 0;
1553 	queue->limits.discard_granularity = size;
1554 
1555 	/* If discard is already enabled, don't reset queue limits */
1556 	if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1557 		return;
1558 
1559 	blk_queue_max_discard_sectors(queue, UINT_MAX);
1560 	blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1561 
1562 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1563 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1564 }
1565 
1566 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1567 {
1568 	u32 max_sectors;
1569 	unsigned short bs = 1 << ns->lba_shift;
1570 
1571 	if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1572 	    (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1573 		return;
1574 	/*
1575 	 * Even though NVMe spec explicitly states that MDTS is not
1576 	 * applicable to the write-zeroes:- "The restriction does not apply to
1577 	 * commands that do not transfer data between the host and the
1578 	 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1579 	 * In order to be more cautious use controller's max_hw_sectors value
1580 	 * to configure the maximum sectors for the write-zeroes which is
1581 	 * configured based on the controller's MDTS field in the
1582 	 * nvme_init_identify() if available.
1583 	 */
1584 	if (ns->ctrl->max_hw_sectors == UINT_MAX)
1585 		max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1586 	else
1587 		max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1588 
1589 	blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1590 }
1591 
1592 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1593 		struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1594 {
1595 	memset(ids, 0, sizeof(*ids));
1596 
1597 	if (ctrl->vs >= NVME_VS(1, 1, 0))
1598 		memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1599 	if (ctrl->vs >= NVME_VS(1, 2, 0))
1600 		memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1601 	if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1602 		 /* Don't treat error as fatal we potentially
1603 		  * already have a NGUID or EUI-64
1604 		  */
1605 		if (nvme_identify_ns_descs(ctrl, nsid, ids))
1606 			dev_warn(ctrl->device,
1607 				 "%s: Identify Descriptors failed\n", __func__);
1608 	}
1609 }
1610 
1611 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1612 {
1613 	return !uuid_is_null(&ids->uuid) ||
1614 		memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1615 		memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1616 }
1617 
1618 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1619 {
1620 	return uuid_equal(&a->uuid, &b->uuid) &&
1621 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1622 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1623 }
1624 
1625 static void nvme_update_disk_info(struct gendisk *disk,
1626 		struct nvme_ns *ns, struct nvme_id_ns *id)
1627 {
1628 	sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1629 	unsigned short bs = 1 << ns->lba_shift;
1630 	u32 atomic_bs, phys_bs, io_opt;
1631 
1632 	if (ns->lba_shift > PAGE_SHIFT) {
1633 		/* unsupported block size, set capacity to 0 later */
1634 		bs = (1 << 9);
1635 	}
1636 	blk_mq_freeze_queue(disk->queue);
1637 	blk_integrity_unregister(disk);
1638 
1639 	if (id->nabo == 0) {
1640 		/*
1641 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1642 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1643 		 * 0 then AWUPF must be used instead.
1644 		 */
1645 		if (id->nsfeat & (1 << 1) && id->nawupf)
1646 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1647 		else
1648 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1649 	} else {
1650 		atomic_bs = bs;
1651 	}
1652 	phys_bs = bs;
1653 	io_opt = bs;
1654 	if (id->nsfeat & (1 << 4)) {
1655 		/* NPWG = Namespace Preferred Write Granularity */
1656 		phys_bs *= 1 + le16_to_cpu(id->npwg);
1657 		/* NOWS = Namespace Optimal Write Size */
1658 		io_opt *= 1 + le16_to_cpu(id->nows);
1659 	}
1660 
1661 	blk_queue_logical_block_size(disk->queue, bs);
1662 	/*
1663 	 * Linux filesystems assume writing a single physical block is
1664 	 * an atomic operation. Hence limit the physical block size to the
1665 	 * value of the Atomic Write Unit Power Fail parameter.
1666 	 */
1667 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1668 	blk_queue_io_min(disk->queue, phys_bs);
1669 	blk_queue_io_opt(disk->queue, io_opt);
1670 
1671 	if (ns->ms && !ns->ext &&
1672 	    (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1673 		nvme_init_integrity(disk, ns->ms, ns->pi_type);
1674 	if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1675 	    ns->lba_shift > PAGE_SHIFT)
1676 		capacity = 0;
1677 
1678 	set_capacity(disk, capacity);
1679 
1680 	nvme_config_discard(disk, ns);
1681 	nvme_config_write_zeroes(disk, ns);
1682 
1683 	if (id->nsattr & (1 << 0))
1684 		set_disk_ro(disk, true);
1685 	else
1686 		set_disk_ro(disk, false);
1687 
1688 	blk_mq_unfreeze_queue(disk->queue);
1689 }
1690 
1691 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1692 {
1693 	struct nvme_ns *ns = disk->private_data;
1694 
1695 	/*
1696 	 * If identify namespace failed, use default 512 byte block size so
1697 	 * block layer can use before failing read/write for 0 capacity.
1698 	 */
1699 	ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1700 	if (ns->lba_shift == 0)
1701 		ns->lba_shift = 9;
1702 	ns->noiob = le16_to_cpu(id->noiob);
1703 	ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1704 	ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1705 	/* the PI implementation requires metadata equal t10 pi tuple size */
1706 	if (ns->ms == sizeof(struct t10_pi_tuple))
1707 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1708 	else
1709 		ns->pi_type = 0;
1710 
1711 	if (ns->noiob)
1712 		nvme_set_chunk_size(ns);
1713 	nvme_update_disk_info(disk, ns, id);
1714 #ifdef CONFIG_NVME_MULTIPATH
1715 	if (ns->head->disk) {
1716 		nvme_update_disk_info(ns->head->disk, ns, id);
1717 		blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1718 	}
1719 #endif
1720 }
1721 
1722 static int nvme_revalidate_disk(struct gendisk *disk)
1723 {
1724 	struct nvme_ns *ns = disk->private_data;
1725 	struct nvme_ctrl *ctrl = ns->ctrl;
1726 	struct nvme_id_ns *id;
1727 	struct nvme_ns_ids ids;
1728 	int ret = 0;
1729 
1730 	if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1731 		set_capacity(disk, 0);
1732 		return -ENODEV;
1733 	}
1734 
1735 	id = nvme_identify_ns(ctrl, ns->head->ns_id);
1736 	if (!id)
1737 		return -ENODEV;
1738 
1739 	if (id->ncap == 0) {
1740 		ret = -ENODEV;
1741 		goto out;
1742 	}
1743 
1744 	__nvme_revalidate_disk(disk, id);
1745 	nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1746 	if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1747 		dev_err(ctrl->device,
1748 			"identifiers changed for nsid %d\n", ns->head->ns_id);
1749 		ret = -ENODEV;
1750 	}
1751 
1752 out:
1753 	kfree(id);
1754 	return ret;
1755 }
1756 
1757 static char nvme_pr_type(enum pr_type type)
1758 {
1759 	switch (type) {
1760 	case PR_WRITE_EXCLUSIVE:
1761 		return 1;
1762 	case PR_EXCLUSIVE_ACCESS:
1763 		return 2;
1764 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1765 		return 3;
1766 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1767 		return 4;
1768 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1769 		return 5;
1770 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1771 		return 6;
1772 	default:
1773 		return 0;
1774 	}
1775 };
1776 
1777 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1778 				u64 key, u64 sa_key, u8 op)
1779 {
1780 	struct nvme_ns_head *head = NULL;
1781 	struct nvme_ns *ns;
1782 	struct nvme_command c;
1783 	int srcu_idx, ret;
1784 	u8 data[16] = { 0, };
1785 
1786 	ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1787 	if (unlikely(!ns))
1788 		return -EWOULDBLOCK;
1789 
1790 	put_unaligned_le64(key, &data[0]);
1791 	put_unaligned_le64(sa_key, &data[8]);
1792 
1793 	memset(&c, 0, sizeof(c));
1794 	c.common.opcode = op;
1795 	c.common.nsid = cpu_to_le32(ns->head->ns_id);
1796 	c.common.cdw10 = cpu_to_le32(cdw10);
1797 
1798 	ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1799 	nvme_put_ns_from_disk(head, srcu_idx);
1800 	return ret;
1801 }
1802 
1803 static int nvme_pr_register(struct block_device *bdev, u64 old,
1804 		u64 new, unsigned flags)
1805 {
1806 	u32 cdw10;
1807 
1808 	if (flags & ~PR_FL_IGNORE_KEY)
1809 		return -EOPNOTSUPP;
1810 
1811 	cdw10 = old ? 2 : 0;
1812 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1813 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1814 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1815 }
1816 
1817 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1818 		enum pr_type type, unsigned flags)
1819 {
1820 	u32 cdw10;
1821 
1822 	if (flags & ~PR_FL_IGNORE_KEY)
1823 		return -EOPNOTSUPP;
1824 
1825 	cdw10 = nvme_pr_type(type) << 8;
1826 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1827 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1828 }
1829 
1830 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1831 		enum pr_type type, bool abort)
1832 {
1833 	u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1834 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1835 }
1836 
1837 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1838 {
1839 	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1840 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1841 }
1842 
1843 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1844 {
1845 	u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1846 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1847 }
1848 
1849 static const struct pr_ops nvme_pr_ops = {
1850 	.pr_register	= nvme_pr_register,
1851 	.pr_reserve	= nvme_pr_reserve,
1852 	.pr_release	= nvme_pr_release,
1853 	.pr_preempt	= nvme_pr_preempt,
1854 	.pr_clear	= nvme_pr_clear,
1855 };
1856 
1857 #ifdef CONFIG_BLK_SED_OPAL
1858 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1859 		bool send)
1860 {
1861 	struct nvme_ctrl *ctrl = data;
1862 	struct nvme_command cmd;
1863 
1864 	memset(&cmd, 0, sizeof(cmd));
1865 	if (send)
1866 		cmd.common.opcode = nvme_admin_security_send;
1867 	else
1868 		cmd.common.opcode = nvme_admin_security_recv;
1869 	cmd.common.nsid = 0;
1870 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1871 	cmd.common.cdw11 = cpu_to_le32(len);
1872 
1873 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1874 				      ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1875 }
1876 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1877 #endif /* CONFIG_BLK_SED_OPAL */
1878 
1879 static const struct block_device_operations nvme_fops = {
1880 	.owner		= THIS_MODULE,
1881 	.ioctl		= nvme_ioctl,
1882 	.compat_ioctl	= nvme_ioctl,
1883 	.open		= nvme_open,
1884 	.release	= nvme_release,
1885 	.getgeo		= nvme_getgeo,
1886 	.revalidate_disk= nvme_revalidate_disk,
1887 	.pr_ops		= &nvme_pr_ops,
1888 };
1889 
1890 #ifdef CONFIG_NVME_MULTIPATH
1891 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1892 {
1893 	struct nvme_ns_head *head = bdev->bd_disk->private_data;
1894 
1895 	if (!kref_get_unless_zero(&head->ref))
1896 		return -ENXIO;
1897 	return 0;
1898 }
1899 
1900 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1901 {
1902 	nvme_put_ns_head(disk->private_data);
1903 }
1904 
1905 const struct block_device_operations nvme_ns_head_ops = {
1906 	.owner		= THIS_MODULE,
1907 	.open		= nvme_ns_head_open,
1908 	.release	= nvme_ns_head_release,
1909 	.ioctl		= nvme_ioctl,
1910 	.compat_ioctl	= nvme_ioctl,
1911 	.getgeo		= nvme_getgeo,
1912 	.pr_ops		= &nvme_pr_ops,
1913 };
1914 #endif /* CONFIG_NVME_MULTIPATH */
1915 
1916 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1917 {
1918 	unsigned long timeout =
1919 		((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1920 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1921 	int ret;
1922 
1923 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1924 		if (csts == ~0)
1925 			return -ENODEV;
1926 		if ((csts & NVME_CSTS_RDY) == bit)
1927 			break;
1928 
1929 		msleep(100);
1930 		if (fatal_signal_pending(current))
1931 			return -EINTR;
1932 		if (time_after(jiffies, timeout)) {
1933 			dev_err(ctrl->device,
1934 				"Device not ready; aborting %s\n", enabled ?
1935 						"initialisation" : "reset");
1936 			return -ENODEV;
1937 		}
1938 	}
1939 
1940 	return ret;
1941 }
1942 
1943 /*
1944  * If the device has been passed off to us in an enabled state, just clear
1945  * the enabled bit.  The spec says we should set the 'shutdown notification
1946  * bits', but doing so may cause the device to complete commands to the
1947  * admin queue ... and we don't know what memory that might be pointing at!
1948  */
1949 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1950 {
1951 	int ret;
1952 
1953 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1954 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1955 
1956 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1957 	if (ret)
1958 		return ret;
1959 
1960 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1961 		msleep(NVME_QUIRK_DELAY_AMOUNT);
1962 
1963 	return nvme_wait_ready(ctrl, cap, false);
1964 }
1965 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1966 
1967 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1968 {
1969 	/*
1970 	 * Default to a 4K page size, with the intention to update this
1971 	 * path in the future to accomodate architectures with differing
1972 	 * kernel and IO page sizes.
1973 	 */
1974 	unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1975 	int ret;
1976 
1977 	if (page_shift < dev_page_min) {
1978 		dev_err(ctrl->device,
1979 			"Minimum device page size %u too large for host (%u)\n",
1980 			1 << dev_page_min, 1 << page_shift);
1981 		return -ENODEV;
1982 	}
1983 
1984 	ctrl->page_size = 1 << page_shift;
1985 
1986 	ctrl->ctrl_config = NVME_CC_CSS_NVM;
1987 	ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1988 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1989 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1990 	ctrl->ctrl_config |= NVME_CC_ENABLE;
1991 
1992 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1993 	if (ret)
1994 		return ret;
1995 	return nvme_wait_ready(ctrl, cap, true);
1996 }
1997 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1998 
1999 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2000 {
2001 	unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2002 	u32 csts;
2003 	int ret;
2004 
2005 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2006 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2007 
2008 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2009 	if (ret)
2010 		return ret;
2011 
2012 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2013 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2014 			break;
2015 
2016 		msleep(100);
2017 		if (fatal_signal_pending(current))
2018 			return -EINTR;
2019 		if (time_after(jiffies, timeout)) {
2020 			dev_err(ctrl->device,
2021 				"Device shutdown incomplete; abort shutdown\n");
2022 			return -ENODEV;
2023 		}
2024 	}
2025 
2026 	return ret;
2027 }
2028 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2029 
2030 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2031 		struct request_queue *q)
2032 {
2033 	bool vwc = false;
2034 
2035 	if (ctrl->max_hw_sectors) {
2036 		u32 max_segments =
2037 			(ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2038 
2039 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
2040 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2041 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2042 	}
2043 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2044 	    is_power_of_2(ctrl->max_hw_sectors))
2045 		blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2046 	blk_queue_virt_boundary(q, ctrl->page_size - 1);
2047 	if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2048 		vwc = true;
2049 	blk_queue_write_cache(q, vwc, vwc);
2050 }
2051 
2052 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2053 {
2054 	__le64 ts;
2055 	int ret;
2056 
2057 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2058 		return 0;
2059 
2060 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2061 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2062 			NULL);
2063 	if (ret)
2064 		dev_warn_once(ctrl->device,
2065 			"could not set timestamp (%d)\n", ret);
2066 	return ret;
2067 }
2068 
2069 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2070 {
2071 	struct nvme_feat_host_behavior *host;
2072 	int ret;
2073 
2074 	/* Don't bother enabling the feature if retry delay is not reported */
2075 	if (!ctrl->crdt[0])
2076 		return 0;
2077 
2078 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2079 	if (!host)
2080 		return 0;
2081 
2082 	host->acre = NVME_ENABLE_ACRE;
2083 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2084 				host, sizeof(*host), NULL);
2085 	kfree(host);
2086 	return ret;
2087 }
2088 
2089 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2090 {
2091 	/*
2092 	 * APST (Autonomous Power State Transition) lets us program a
2093 	 * table of power state transitions that the controller will
2094 	 * perform automatically.  We configure it with a simple
2095 	 * heuristic: we are willing to spend at most 2% of the time
2096 	 * transitioning between power states.  Therefore, when running
2097 	 * in any given state, we will enter the next lower-power
2098 	 * non-operational state after waiting 50 * (enlat + exlat)
2099 	 * microseconds, as long as that state's exit latency is under
2100 	 * the requested maximum latency.
2101 	 *
2102 	 * We will not autonomously enter any non-operational state for
2103 	 * which the total latency exceeds ps_max_latency_us.  Users
2104 	 * can set ps_max_latency_us to zero to turn off APST.
2105 	 */
2106 
2107 	unsigned apste;
2108 	struct nvme_feat_auto_pst *table;
2109 	u64 max_lat_us = 0;
2110 	int max_ps = -1;
2111 	int ret;
2112 
2113 	/*
2114 	 * If APST isn't supported or if we haven't been initialized yet,
2115 	 * then don't do anything.
2116 	 */
2117 	if (!ctrl->apsta)
2118 		return 0;
2119 
2120 	if (ctrl->npss > 31) {
2121 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2122 		return 0;
2123 	}
2124 
2125 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2126 	if (!table)
2127 		return 0;
2128 
2129 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2130 		/* Turn off APST. */
2131 		apste = 0;
2132 		dev_dbg(ctrl->device, "APST disabled\n");
2133 	} else {
2134 		__le64 target = cpu_to_le64(0);
2135 		int state;
2136 
2137 		/*
2138 		 * Walk through all states from lowest- to highest-power.
2139 		 * According to the spec, lower-numbered states use more
2140 		 * power.  NPSS, despite the name, is the index of the
2141 		 * lowest-power state, not the number of states.
2142 		 */
2143 		for (state = (int)ctrl->npss; state >= 0; state--) {
2144 			u64 total_latency_us, exit_latency_us, transition_ms;
2145 
2146 			if (target)
2147 				table->entries[state] = target;
2148 
2149 			/*
2150 			 * Don't allow transitions to the deepest state
2151 			 * if it's quirked off.
2152 			 */
2153 			if (state == ctrl->npss &&
2154 			    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2155 				continue;
2156 
2157 			/*
2158 			 * Is this state a useful non-operational state for
2159 			 * higher-power states to autonomously transition to?
2160 			 */
2161 			if (!(ctrl->psd[state].flags &
2162 			      NVME_PS_FLAGS_NON_OP_STATE))
2163 				continue;
2164 
2165 			exit_latency_us =
2166 				(u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2167 			if (exit_latency_us > ctrl->ps_max_latency_us)
2168 				continue;
2169 
2170 			total_latency_us =
2171 				exit_latency_us +
2172 				le32_to_cpu(ctrl->psd[state].entry_lat);
2173 
2174 			/*
2175 			 * This state is good.  Use it as the APST idle
2176 			 * target for higher power states.
2177 			 */
2178 			transition_ms = total_latency_us + 19;
2179 			do_div(transition_ms, 20);
2180 			if (transition_ms > (1 << 24) - 1)
2181 				transition_ms = (1 << 24) - 1;
2182 
2183 			target = cpu_to_le64((state << 3) |
2184 					     (transition_ms << 8));
2185 
2186 			if (max_ps == -1)
2187 				max_ps = state;
2188 
2189 			if (total_latency_us > max_lat_us)
2190 				max_lat_us = total_latency_us;
2191 		}
2192 
2193 		apste = 1;
2194 
2195 		if (max_ps == -1) {
2196 			dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2197 		} else {
2198 			dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2199 				max_ps, max_lat_us, (int)sizeof(*table), table);
2200 		}
2201 	}
2202 
2203 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2204 				table, sizeof(*table), NULL);
2205 	if (ret)
2206 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2207 
2208 	kfree(table);
2209 	return ret;
2210 }
2211 
2212 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2213 {
2214 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2215 	u64 latency;
2216 
2217 	switch (val) {
2218 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2219 	case PM_QOS_LATENCY_ANY:
2220 		latency = U64_MAX;
2221 		break;
2222 
2223 	default:
2224 		latency = val;
2225 	}
2226 
2227 	if (ctrl->ps_max_latency_us != latency) {
2228 		ctrl->ps_max_latency_us = latency;
2229 		nvme_configure_apst(ctrl);
2230 	}
2231 }
2232 
2233 struct nvme_core_quirk_entry {
2234 	/*
2235 	 * NVMe model and firmware strings are padded with spaces.  For
2236 	 * simplicity, strings in the quirk table are padded with NULLs
2237 	 * instead.
2238 	 */
2239 	u16 vid;
2240 	const char *mn;
2241 	const char *fr;
2242 	unsigned long quirks;
2243 };
2244 
2245 static const struct nvme_core_quirk_entry core_quirks[] = {
2246 	{
2247 		/*
2248 		 * This Toshiba device seems to die using any APST states.  See:
2249 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2250 		 */
2251 		.vid = 0x1179,
2252 		.mn = "THNSF5256GPUK TOSHIBA",
2253 		.quirks = NVME_QUIRK_NO_APST,
2254 	}
2255 };
2256 
2257 /* match is null-terminated but idstr is space-padded. */
2258 static bool string_matches(const char *idstr, const char *match, size_t len)
2259 {
2260 	size_t matchlen;
2261 
2262 	if (!match)
2263 		return true;
2264 
2265 	matchlen = strlen(match);
2266 	WARN_ON_ONCE(matchlen > len);
2267 
2268 	if (memcmp(idstr, match, matchlen))
2269 		return false;
2270 
2271 	for (; matchlen < len; matchlen++)
2272 		if (idstr[matchlen] != ' ')
2273 			return false;
2274 
2275 	return true;
2276 }
2277 
2278 static bool quirk_matches(const struct nvme_id_ctrl *id,
2279 			  const struct nvme_core_quirk_entry *q)
2280 {
2281 	return q->vid == le16_to_cpu(id->vid) &&
2282 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2283 		string_matches(id->fr, q->fr, sizeof(id->fr));
2284 }
2285 
2286 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2287 		struct nvme_id_ctrl *id)
2288 {
2289 	size_t nqnlen;
2290 	int off;
2291 
2292 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2293 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2294 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2295 			strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2296 			return;
2297 		}
2298 
2299 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2300 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2301 	}
2302 
2303 	/* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2304 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2305 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2306 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2307 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2308 	off += sizeof(id->sn);
2309 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2310 	off += sizeof(id->mn);
2311 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2312 }
2313 
2314 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2315 {
2316 	ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2317 	kfree(subsys);
2318 }
2319 
2320 static void nvme_release_subsystem(struct device *dev)
2321 {
2322 	__nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2323 }
2324 
2325 static void nvme_destroy_subsystem(struct kref *ref)
2326 {
2327 	struct nvme_subsystem *subsys =
2328 			container_of(ref, struct nvme_subsystem, ref);
2329 
2330 	mutex_lock(&nvme_subsystems_lock);
2331 	list_del(&subsys->entry);
2332 	mutex_unlock(&nvme_subsystems_lock);
2333 
2334 	ida_destroy(&subsys->ns_ida);
2335 	device_del(&subsys->dev);
2336 	put_device(&subsys->dev);
2337 }
2338 
2339 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2340 {
2341 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2342 }
2343 
2344 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2345 {
2346 	struct nvme_subsystem *subsys;
2347 
2348 	lockdep_assert_held(&nvme_subsystems_lock);
2349 
2350 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2351 		if (strcmp(subsys->subnqn, subsysnqn))
2352 			continue;
2353 		if (!kref_get_unless_zero(&subsys->ref))
2354 			continue;
2355 		return subsys;
2356 	}
2357 
2358 	return NULL;
2359 }
2360 
2361 #define SUBSYS_ATTR_RO(_name, _mode, _show)			\
2362 	struct device_attribute subsys_attr_##_name = \
2363 		__ATTR(_name, _mode, _show, NULL)
2364 
2365 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2366 				    struct device_attribute *attr,
2367 				    char *buf)
2368 {
2369 	struct nvme_subsystem *subsys =
2370 		container_of(dev, struct nvme_subsystem, dev);
2371 
2372 	return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2373 }
2374 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2375 
2376 #define nvme_subsys_show_str_function(field)				\
2377 static ssize_t subsys_##field##_show(struct device *dev,		\
2378 			    struct device_attribute *attr, char *buf)	\
2379 {									\
2380 	struct nvme_subsystem *subsys =					\
2381 		container_of(dev, struct nvme_subsystem, dev);		\
2382 	return sprintf(buf, "%.*s\n",					\
2383 		       (int)sizeof(subsys->field), subsys->field);	\
2384 }									\
2385 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2386 
2387 nvme_subsys_show_str_function(model);
2388 nvme_subsys_show_str_function(serial);
2389 nvme_subsys_show_str_function(firmware_rev);
2390 
2391 static struct attribute *nvme_subsys_attrs[] = {
2392 	&subsys_attr_model.attr,
2393 	&subsys_attr_serial.attr,
2394 	&subsys_attr_firmware_rev.attr,
2395 	&subsys_attr_subsysnqn.attr,
2396 #ifdef CONFIG_NVME_MULTIPATH
2397 	&subsys_attr_iopolicy.attr,
2398 #endif
2399 	NULL,
2400 };
2401 
2402 static struct attribute_group nvme_subsys_attrs_group = {
2403 	.attrs = nvme_subsys_attrs,
2404 };
2405 
2406 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2407 	&nvme_subsys_attrs_group,
2408 	NULL,
2409 };
2410 
2411 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2412 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2413 {
2414 	struct nvme_ctrl *tmp;
2415 
2416 	lockdep_assert_held(&nvme_subsystems_lock);
2417 
2418 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2419 		if (tmp->state == NVME_CTRL_DELETING ||
2420 		    tmp->state == NVME_CTRL_DEAD)
2421 			continue;
2422 
2423 		if (tmp->cntlid == ctrl->cntlid) {
2424 			dev_err(ctrl->device,
2425 				"Duplicate cntlid %u with %s, rejecting\n",
2426 				ctrl->cntlid, dev_name(tmp->device));
2427 			return false;
2428 		}
2429 
2430 		if ((id->cmic & (1 << 1)) ||
2431 		    (ctrl->opts && ctrl->opts->discovery_nqn))
2432 			continue;
2433 
2434 		dev_err(ctrl->device,
2435 			"Subsystem does not support multiple controllers\n");
2436 		return false;
2437 	}
2438 
2439 	return true;
2440 }
2441 
2442 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2443 {
2444 	struct nvme_subsystem *subsys, *found;
2445 	int ret;
2446 
2447 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2448 	if (!subsys)
2449 		return -ENOMEM;
2450 	ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2451 	if (ret < 0) {
2452 		kfree(subsys);
2453 		return ret;
2454 	}
2455 	subsys->instance = ret;
2456 	mutex_init(&subsys->lock);
2457 	kref_init(&subsys->ref);
2458 	INIT_LIST_HEAD(&subsys->ctrls);
2459 	INIT_LIST_HEAD(&subsys->nsheads);
2460 	nvme_init_subnqn(subsys, ctrl, id);
2461 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2462 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2463 	memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2464 	subsys->vendor_id = le16_to_cpu(id->vid);
2465 	subsys->cmic = id->cmic;
2466 	subsys->awupf = le16_to_cpu(id->awupf);
2467 #ifdef CONFIG_NVME_MULTIPATH
2468 	subsys->iopolicy = NVME_IOPOLICY_NUMA;
2469 #endif
2470 
2471 	subsys->dev.class = nvme_subsys_class;
2472 	subsys->dev.release = nvme_release_subsystem;
2473 	subsys->dev.groups = nvme_subsys_attrs_groups;
2474 	dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2475 	device_initialize(&subsys->dev);
2476 
2477 	mutex_lock(&nvme_subsystems_lock);
2478 	found = __nvme_find_get_subsystem(subsys->subnqn);
2479 	if (found) {
2480 		__nvme_release_subsystem(subsys);
2481 		subsys = found;
2482 
2483 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2484 			ret = -EINVAL;
2485 			goto out_put_subsystem;
2486 		}
2487 	} else {
2488 		ret = device_add(&subsys->dev);
2489 		if (ret) {
2490 			dev_err(ctrl->device,
2491 				"failed to register subsystem device.\n");
2492 			goto out_unlock;
2493 		}
2494 		ida_init(&subsys->ns_ida);
2495 		list_add_tail(&subsys->entry, &nvme_subsystems);
2496 	}
2497 
2498 	if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2499 			dev_name(ctrl->device))) {
2500 		dev_err(ctrl->device,
2501 			"failed to create sysfs link from subsystem.\n");
2502 		goto out_put_subsystem;
2503 	}
2504 
2505 	ctrl->subsys = subsys;
2506 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2507 	mutex_unlock(&nvme_subsystems_lock);
2508 	return 0;
2509 
2510 out_put_subsystem:
2511 	nvme_put_subsystem(subsys);
2512 out_unlock:
2513 	mutex_unlock(&nvme_subsystems_lock);
2514 	put_device(&subsys->dev);
2515 	return ret;
2516 }
2517 
2518 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2519 		void *log, size_t size, u64 offset)
2520 {
2521 	struct nvme_command c = { };
2522 	unsigned long dwlen = size / 4 - 1;
2523 
2524 	c.get_log_page.opcode = nvme_admin_get_log_page;
2525 	c.get_log_page.nsid = cpu_to_le32(nsid);
2526 	c.get_log_page.lid = log_page;
2527 	c.get_log_page.lsp = lsp;
2528 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2529 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2530 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2531 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2532 
2533 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2534 }
2535 
2536 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2537 {
2538 	int ret;
2539 
2540 	if (!ctrl->effects)
2541 		ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2542 
2543 	if (!ctrl->effects)
2544 		return 0;
2545 
2546 	ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2547 			ctrl->effects, sizeof(*ctrl->effects), 0);
2548 	if (ret) {
2549 		kfree(ctrl->effects);
2550 		ctrl->effects = NULL;
2551 	}
2552 	return ret;
2553 }
2554 
2555 /*
2556  * Initialize the cached copies of the Identify data and various controller
2557  * register in our nvme_ctrl structure.  This should be called as soon as
2558  * the admin queue is fully up and running.
2559  */
2560 int nvme_init_identify(struct nvme_ctrl *ctrl)
2561 {
2562 	struct nvme_id_ctrl *id;
2563 	u64 cap;
2564 	int ret, page_shift;
2565 	u32 max_hw_sectors;
2566 	bool prev_apst_enabled;
2567 
2568 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2569 	if (ret) {
2570 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2571 		return ret;
2572 	}
2573 
2574 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2575 	if (ret) {
2576 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2577 		return ret;
2578 	}
2579 	page_shift = NVME_CAP_MPSMIN(cap) + 12;
2580 
2581 	if (ctrl->vs >= NVME_VS(1, 1, 0))
2582 		ctrl->subsystem = NVME_CAP_NSSRC(cap);
2583 
2584 	ret = nvme_identify_ctrl(ctrl, &id);
2585 	if (ret) {
2586 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2587 		return -EIO;
2588 	}
2589 
2590 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2591 		ret = nvme_get_effects_log(ctrl);
2592 		if (ret < 0)
2593 			goto out_free;
2594 	}
2595 
2596 	if (!ctrl->identified) {
2597 		int i;
2598 
2599 		ret = nvme_init_subsystem(ctrl, id);
2600 		if (ret)
2601 			goto out_free;
2602 
2603 		/*
2604 		 * Check for quirks.  Quirk can depend on firmware version,
2605 		 * so, in principle, the set of quirks present can change
2606 		 * across a reset.  As a possible future enhancement, we
2607 		 * could re-scan for quirks every time we reinitialize
2608 		 * the device, but we'd have to make sure that the driver
2609 		 * behaves intelligently if the quirks change.
2610 		 */
2611 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2612 			if (quirk_matches(id, &core_quirks[i]))
2613 				ctrl->quirks |= core_quirks[i].quirks;
2614 		}
2615 	}
2616 
2617 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2618 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2619 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2620 	}
2621 
2622 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2623 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2624 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2625 
2626 	ctrl->oacs = le16_to_cpu(id->oacs);
2627 	ctrl->oncs = le16_to_cpu(id->oncs);
2628 	ctrl->mtfa = le16_to_cpu(id->mtfa);
2629 	ctrl->oaes = le32_to_cpu(id->oaes);
2630 	atomic_set(&ctrl->abort_limit, id->acl + 1);
2631 	ctrl->vwc = id->vwc;
2632 	if (id->mdts)
2633 		max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2634 	else
2635 		max_hw_sectors = UINT_MAX;
2636 	ctrl->max_hw_sectors =
2637 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2638 
2639 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
2640 	ctrl->sgls = le32_to_cpu(id->sgls);
2641 	ctrl->kas = le16_to_cpu(id->kas);
2642 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
2643 	ctrl->ctratt = le32_to_cpu(id->ctratt);
2644 
2645 	if (id->rtd3e) {
2646 		/* us -> s */
2647 		u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2648 
2649 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2650 						 shutdown_timeout, 60);
2651 
2652 		if (ctrl->shutdown_timeout != shutdown_timeout)
2653 			dev_info(ctrl->device,
2654 				 "Shutdown timeout set to %u seconds\n",
2655 				 ctrl->shutdown_timeout);
2656 	} else
2657 		ctrl->shutdown_timeout = shutdown_timeout;
2658 
2659 	ctrl->npss = id->npss;
2660 	ctrl->apsta = id->apsta;
2661 	prev_apst_enabled = ctrl->apst_enabled;
2662 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2663 		if (force_apst && id->apsta) {
2664 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2665 			ctrl->apst_enabled = true;
2666 		} else {
2667 			ctrl->apst_enabled = false;
2668 		}
2669 	} else {
2670 		ctrl->apst_enabled = id->apsta;
2671 	}
2672 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2673 
2674 	if (ctrl->ops->flags & NVME_F_FABRICS) {
2675 		ctrl->icdoff = le16_to_cpu(id->icdoff);
2676 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2677 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2678 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2679 
2680 		/*
2681 		 * In fabrics we need to verify the cntlid matches the
2682 		 * admin connect
2683 		 */
2684 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2685 			ret = -EINVAL;
2686 			goto out_free;
2687 		}
2688 
2689 		if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2690 			dev_err(ctrl->device,
2691 				"keep-alive support is mandatory for fabrics\n");
2692 			ret = -EINVAL;
2693 			goto out_free;
2694 		}
2695 	} else {
2696 		ctrl->cntlid = le16_to_cpu(id->cntlid);
2697 		ctrl->hmpre = le32_to_cpu(id->hmpre);
2698 		ctrl->hmmin = le32_to_cpu(id->hmmin);
2699 		ctrl->hmminds = le32_to_cpu(id->hmminds);
2700 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2701 	}
2702 
2703 	ret = nvme_mpath_init(ctrl, id);
2704 	kfree(id);
2705 
2706 	if (ret < 0)
2707 		return ret;
2708 
2709 	if (ctrl->apst_enabled && !prev_apst_enabled)
2710 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
2711 	else if (!ctrl->apst_enabled && prev_apst_enabled)
2712 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
2713 
2714 	ret = nvme_configure_apst(ctrl);
2715 	if (ret < 0)
2716 		return ret;
2717 
2718 	ret = nvme_configure_timestamp(ctrl);
2719 	if (ret < 0)
2720 		return ret;
2721 
2722 	ret = nvme_configure_directives(ctrl);
2723 	if (ret < 0)
2724 		return ret;
2725 
2726 	ret = nvme_configure_acre(ctrl);
2727 	if (ret < 0)
2728 		return ret;
2729 
2730 	ctrl->identified = true;
2731 
2732 	return 0;
2733 
2734 out_free:
2735 	kfree(id);
2736 	return ret;
2737 }
2738 EXPORT_SYMBOL_GPL(nvme_init_identify);
2739 
2740 static int nvme_dev_open(struct inode *inode, struct file *file)
2741 {
2742 	struct nvme_ctrl *ctrl =
2743 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2744 
2745 	switch (ctrl->state) {
2746 	case NVME_CTRL_LIVE:
2747 	case NVME_CTRL_ADMIN_ONLY:
2748 		break;
2749 	default:
2750 		return -EWOULDBLOCK;
2751 	}
2752 
2753 	file->private_data = ctrl;
2754 	return 0;
2755 }
2756 
2757 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2758 {
2759 	struct nvme_ns *ns;
2760 	int ret;
2761 
2762 	down_read(&ctrl->namespaces_rwsem);
2763 	if (list_empty(&ctrl->namespaces)) {
2764 		ret = -ENOTTY;
2765 		goto out_unlock;
2766 	}
2767 
2768 	ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2769 	if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2770 		dev_warn(ctrl->device,
2771 			"NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2772 		ret = -EINVAL;
2773 		goto out_unlock;
2774 	}
2775 
2776 	dev_warn(ctrl->device,
2777 		"using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2778 	kref_get(&ns->kref);
2779 	up_read(&ctrl->namespaces_rwsem);
2780 
2781 	ret = nvme_user_cmd(ctrl, ns, argp);
2782 	nvme_put_ns(ns);
2783 	return ret;
2784 
2785 out_unlock:
2786 	up_read(&ctrl->namespaces_rwsem);
2787 	return ret;
2788 }
2789 
2790 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2791 		unsigned long arg)
2792 {
2793 	struct nvme_ctrl *ctrl = file->private_data;
2794 	void __user *argp = (void __user *)arg;
2795 
2796 	switch (cmd) {
2797 	case NVME_IOCTL_ADMIN_CMD:
2798 		return nvme_user_cmd(ctrl, NULL, argp);
2799 	case NVME_IOCTL_IO_CMD:
2800 		return nvme_dev_user_cmd(ctrl, argp);
2801 	case NVME_IOCTL_RESET:
2802 		dev_warn(ctrl->device, "resetting controller\n");
2803 		return nvme_reset_ctrl_sync(ctrl);
2804 	case NVME_IOCTL_SUBSYS_RESET:
2805 		return nvme_reset_subsystem(ctrl);
2806 	case NVME_IOCTL_RESCAN:
2807 		nvme_queue_scan(ctrl);
2808 		return 0;
2809 	default:
2810 		return -ENOTTY;
2811 	}
2812 }
2813 
2814 static const struct file_operations nvme_dev_fops = {
2815 	.owner		= THIS_MODULE,
2816 	.open		= nvme_dev_open,
2817 	.unlocked_ioctl	= nvme_dev_ioctl,
2818 	.compat_ioctl	= nvme_dev_ioctl,
2819 };
2820 
2821 static ssize_t nvme_sysfs_reset(struct device *dev,
2822 				struct device_attribute *attr, const char *buf,
2823 				size_t count)
2824 {
2825 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2826 	int ret;
2827 
2828 	ret = nvme_reset_ctrl_sync(ctrl);
2829 	if (ret < 0)
2830 		return ret;
2831 	return count;
2832 }
2833 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2834 
2835 static ssize_t nvme_sysfs_rescan(struct device *dev,
2836 				struct device_attribute *attr, const char *buf,
2837 				size_t count)
2838 {
2839 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2840 
2841 	nvme_queue_scan(ctrl);
2842 	return count;
2843 }
2844 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2845 
2846 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2847 {
2848 	struct gendisk *disk = dev_to_disk(dev);
2849 
2850 	if (disk->fops == &nvme_fops)
2851 		return nvme_get_ns_from_dev(dev)->head;
2852 	else
2853 		return disk->private_data;
2854 }
2855 
2856 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2857 		char *buf)
2858 {
2859 	struct nvme_ns_head *head = dev_to_ns_head(dev);
2860 	struct nvme_ns_ids *ids = &head->ids;
2861 	struct nvme_subsystem *subsys = head->subsys;
2862 	int serial_len = sizeof(subsys->serial);
2863 	int model_len = sizeof(subsys->model);
2864 
2865 	if (!uuid_is_null(&ids->uuid))
2866 		return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2867 
2868 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2869 		return sprintf(buf, "eui.%16phN\n", ids->nguid);
2870 
2871 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2872 		return sprintf(buf, "eui.%8phN\n", ids->eui64);
2873 
2874 	while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2875 				  subsys->serial[serial_len - 1] == '\0'))
2876 		serial_len--;
2877 	while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2878 				 subsys->model[model_len - 1] == '\0'))
2879 		model_len--;
2880 
2881 	return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2882 		serial_len, subsys->serial, model_len, subsys->model,
2883 		head->ns_id);
2884 }
2885 static DEVICE_ATTR_RO(wwid);
2886 
2887 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2888 		char *buf)
2889 {
2890 	return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2891 }
2892 static DEVICE_ATTR_RO(nguid);
2893 
2894 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2895 		char *buf)
2896 {
2897 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2898 
2899 	/* For backward compatibility expose the NGUID to userspace if
2900 	 * we have no UUID set
2901 	 */
2902 	if (uuid_is_null(&ids->uuid)) {
2903 		printk_ratelimited(KERN_WARNING
2904 				   "No UUID available providing old NGUID\n");
2905 		return sprintf(buf, "%pU\n", ids->nguid);
2906 	}
2907 	return sprintf(buf, "%pU\n", &ids->uuid);
2908 }
2909 static DEVICE_ATTR_RO(uuid);
2910 
2911 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2912 		char *buf)
2913 {
2914 	return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2915 }
2916 static DEVICE_ATTR_RO(eui);
2917 
2918 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2919 		char *buf)
2920 {
2921 	return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2922 }
2923 static DEVICE_ATTR_RO(nsid);
2924 
2925 static struct attribute *nvme_ns_id_attrs[] = {
2926 	&dev_attr_wwid.attr,
2927 	&dev_attr_uuid.attr,
2928 	&dev_attr_nguid.attr,
2929 	&dev_attr_eui.attr,
2930 	&dev_attr_nsid.attr,
2931 #ifdef CONFIG_NVME_MULTIPATH
2932 	&dev_attr_ana_grpid.attr,
2933 	&dev_attr_ana_state.attr,
2934 #endif
2935 	NULL,
2936 };
2937 
2938 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2939 		struct attribute *a, int n)
2940 {
2941 	struct device *dev = container_of(kobj, struct device, kobj);
2942 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2943 
2944 	if (a == &dev_attr_uuid.attr) {
2945 		if (uuid_is_null(&ids->uuid) &&
2946 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2947 			return 0;
2948 	}
2949 	if (a == &dev_attr_nguid.attr) {
2950 		if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2951 			return 0;
2952 	}
2953 	if (a == &dev_attr_eui.attr) {
2954 		if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2955 			return 0;
2956 	}
2957 #ifdef CONFIG_NVME_MULTIPATH
2958 	if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2959 		if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2960 			return 0;
2961 		if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2962 			return 0;
2963 	}
2964 #endif
2965 	return a->mode;
2966 }
2967 
2968 static const struct attribute_group nvme_ns_id_attr_group = {
2969 	.attrs		= nvme_ns_id_attrs,
2970 	.is_visible	= nvme_ns_id_attrs_are_visible,
2971 };
2972 
2973 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2974 	&nvme_ns_id_attr_group,
2975 #ifdef CONFIG_NVM
2976 	&nvme_nvm_attr_group,
2977 #endif
2978 	NULL,
2979 };
2980 
2981 #define nvme_show_str_function(field)						\
2982 static ssize_t  field##_show(struct device *dev,				\
2983 			    struct device_attribute *attr, char *buf)		\
2984 {										\
2985         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
2986         return sprintf(buf, "%.*s\n",						\
2987 		(int)sizeof(ctrl->subsys->field), ctrl->subsys->field);		\
2988 }										\
2989 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2990 
2991 nvme_show_str_function(model);
2992 nvme_show_str_function(serial);
2993 nvme_show_str_function(firmware_rev);
2994 
2995 #define nvme_show_int_function(field)						\
2996 static ssize_t  field##_show(struct device *dev,				\
2997 			    struct device_attribute *attr, char *buf)		\
2998 {										\
2999         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3000         return sprintf(buf, "%d\n", ctrl->field);	\
3001 }										\
3002 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3003 
3004 nvme_show_int_function(cntlid);
3005 nvme_show_int_function(numa_node);
3006 
3007 static ssize_t nvme_sysfs_delete(struct device *dev,
3008 				struct device_attribute *attr, const char *buf,
3009 				size_t count)
3010 {
3011 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3012 
3013 	if (device_remove_file_self(dev, attr))
3014 		nvme_delete_ctrl_sync(ctrl);
3015 	return count;
3016 }
3017 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3018 
3019 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3020 					 struct device_attribute *attr,
3021 					 char *buf)
3022 {
3023 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3024 
3025 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3026 }
3027 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3028 
3029 static ssize_t nvme_sysfs_show_state(struct device *dev,
3030 				     struct device_attribute *attr,
3031 				     char *buf)
3032 {
3033 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3034 	static const char *const state_name[] = {
3035 		[NVME_CTRL_NEW]		= "new",
3036 		[NVME_CTRL_LIVE]	= "live",
3037 		[NVME_CTRL_ADMIN_ONLY]	= "only-admin",
3038 		[NVME_CTRL_RESETTING]	= "resetting",
3039 		[NVME_CTRL_CONNECTING]	= "connecting",
3040 		[NVME_CTRL_DELETING]	= "deleting",
3041 		[NVME_CTRL_DEAD]	= "dead",
3042 	};
3043 
3044 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3045 	    state_name[ctrl->state])
3046 		return sprintf(buf, "%s\n", state_name[ctrl->state]);
3047 
3048 	return sprintf(buf, "unknown state\n");
3049 }
3050 
3051 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3052 
3053 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3054 					 struct device_attribute *attr,
3055 					 char *buf)
3056 {
3057 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3058 
3059 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3060 }
3061 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3062 
3063 static ssize_t nvme_sysfs_show_address(struct device *dev,
3064 					 struct device_attribute *attr,
3065 					 char *buf)
3066 {
3067 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3068 
3069 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3070 }
3071 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3072 
3073 static struct attribute *nvme_dev_attrs[] = {
3074 	&dev_attr_reset_controller.attr,
3075 	&dev_attr_rescan_controller.attr,
3076 	&dev_attr_model.attr,
3077 	&dev_attr_serial.attr,
3078 	&dev_attr_firmware_rev.attr,
3079 	&dev_attr_cntlid.attr,
3080 	&dev_attr_delete_controller.attr,
3081 	&dev_attr_transport.attr,
3082 	&dev_attr_subsysnqn.attr,
3083 	&dev_attr_address.attr,
3084 	&dev_attr_state.attr,
3085 	&dev_attr_numa_node.attr,
3086 	NULL
3087 };
3088 
3089 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3090 		struct attribute *a, int n)
3091 {
3092 	struct device *dev = container_of(kobj, struct device, kobj);
3093 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3094 
3095 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3096 		return 0;
3097 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3098 		return 0;
3099 
3100 	return a->mode;
3101 }
3102 
3103 static struct attribute_group nvme_dev_attrs_group = {
3104 	.attrs		= nvme_dev_attrs,
3105 	.is_visible	= nvme_dev_attrs_are_visible,
3106 };
3107 
3108 static const struct attribute_group *nvme_dev_attr_groups[] = {
3109 	&nvme_dev_attrs_group,
3110 	NULL,
3111 };
3112 
3113 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3114 		unsigned nsid)
3115 {
3116 	struct nvme_ns_head *h;
3117 
3118 	lockdep_assert_held(&subsys->lock);
3119 
3120 	list_for_each_entry(h, &subsys->nsheads, entry) {
3121 		if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3122 			return h;
3123 	}
3124 
3125 	return NULL;
3126 }
3127 
3128 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3129 		struct nvme_ns_head *new)
3130 {
3131 	struct nvme_ns_head *h;
3132 
3133 	lockdep_assert_held(&subsys->lock);
3134 
3135 	list_for_each_entry(h, &subsys->nsheads, entry) {
3136 		if (nvme_ns_ids_valid(&new->ids) &&
3137 		    !list_empty(&h->list) &&
3138 		    nvme_ns_ids_equal(&new->ids, &h->ids))
3139 			return -EINVAL;
3140 	}
3141 
3142 	return 0;
3143 }
3144 
3145 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3146 		unsigned nsid, struct nvme_id_ns *id)
3147 {
3148 	struct nvme_ns_head *head;
3149 	size_t size = sizeof(*head);
3150 	int ret = -ENOMEM;
3151 
3152 #ifdef CONFIG_NVME_MULTIPATH
3153 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3154 #endif
3155 
3156 	head = kzalloc(size, GFP_KERNEL);
3157 	if (!head)
3158 		goto out;
3159 	ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3160 	if (ret < 0)
3161 		goto out_free_head;
3162 	head->instance = ret;
3163 	INIT_LIST_HEAD(&head->list);
3164 	ret = init_srcu_struct(&head->srcu);
3165 	if (ret)
3166 		goto out_ida_remove;
3167 	head->subsys = ctrl->subsys;
3168 	head->ns_id = nsid;
3169 	kref_init(&head->ref);
3170 
3171 	nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3172 
3173 	ret = __nvme_check_ids(ctrl->subsys, head);
3174 	if (ret) {
3175 		dev_err(ctrl->device,
3176 			"duplicate IDs for nsid %d\n", nsid);
3177 		goto out_cleanup_srcu;
3178 	}
3179 
3180 	ret = nvme_mpath_alloc_disk(ctrl, head);
3181 	if (ret)
3182 		goto out_cleanup_srcu;
3183 
3184 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3185 
3186 	kref_get(&ctrl->subsys->ref);
3187 
3188 	return head;
3189 out_cleanup_srcu:
3190 	cleanup_srcu_struct(&head->srcu);
3191 out_ida_remove:
3192 	ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3193 out_free_head:
3194 	kfree(head);
3195 out:
3196 	return ERR_PTR(ret);
3197 }
3198 
3199 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3200 		struct nvme_id_ns *id)
3201 {
3202 	struct nvme_ctrl *ctrl = ns->ctrl;
3203 	bool is_shared = id->nmic & (1 << 0);
3204 	struct nvme_ns_head *head = NULL;
3205 	int ret = 0;
3206 
3207 	mutex_lock(&ctrl->subsys->lock);
3208 	if (is_shared)
3209 		head = __nvme_find_ns_head(ctrl->subsys, nsid);
3210 	if (!head) {
3211 		head = nvme_alloc_ns_head(ctrl, nsid, id);
3212 		if (IS_ERR(head)) {
3213 			ret = PTR_ERR(head);
3214 			goto out_unlock;
3215 		}
3216 	} else {
3217 		struct nvme_ns_ids ids;
3218 
3219 		nvme_report_ns_ids(ctrl, nsid, id, &ids);
3220 		if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3221 			dev_err(ctrl->device,
3222 				"IDs don't match for shared namespace %d\n",
3223 					nsid);
3224 			ret = -EINVAL;
3225 			goto out_unlock;
3226 		}
3227 	}
3228 
3229 	list_add_tail(&ns->siblings, &head->list);
3230 	ns->head = head;
3231 
3232 out_unlock:
3233 	mutex_unlock(&ctrl->subsys->lock);
3234 	return ret;
3235 }
3236 
3237 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3238 {
3239 	struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3240 	struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3241 
3242 	return nsa->head->ns_id - nsb->head->ns_id;
3243 }
3244 
3245 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3246 {
3247 	struct nvme_ns *ns, *ret = NULL;
3248 
3249 	down_read(&ctrl->namespaces_rwsem);
3250 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3251 		if (ns->head->ns_id == nsid) {
3252 			if (!kref_get_unless_zero(&ns->kref))
3253 				continue;
3254 			ret = ns;
3255 			break;
3256 		}
3257 		if (ns->head->ns_id > nsid)
3258 			break;
3259 	}
3260 	up_read(&ctrl->namespaces_rwsem);
3261 	return ret;
3262 }
3263 
3264 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3265 {
3266 	struct streams_directive_params s;
3267 	int ret;
3268 
3269 	if (!ctrl->nr_streams)
3270 		return 0;
3271 
3272 	ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3273 	if (ret)
3274 		return ret;
3275 
3276 	ns->sws = le32_to_cpu(s.sws);
3277 	ns->sgs = le16_to_cpu(s.sgs);
3278 
3279 	if (ns->sws) {
3280 		unsigned int bs = 1 << ns->lba_shift;
3281 
3282 		blk_queue_io_min(ns->queue, bs * ns->sws);
3283 		if (ns->sgs)
3284 			blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3285 	}
3286 
3287 	return 0;
3288 }
3289 
3290 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3291 {
3292 	struct nvme_ns *ns;
3293 	struct gendisk *disk;
3294 	struct nvme_id_ns *id;
3295 	char disk_name[DISK_NAME_LEN];
3296 	int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3297 
3298 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3299 	if (!ns)
3300 		return -ENOMEM;
3301 
3302 	ns->queue = blk_mq_init_queue(ctrl->tagset);
3303 	if (IS_ERR(ns->queue)) {
3304 		ret = PTR_ERR(ns->queue);
3305 		goto out_free_ns;
3306 	}
3307 
3308 	if (ctrl->opts && ctrl->opts->data_digest)
3309 		ns->queue->backing_dev_info->capabilities
3310 			|= BDI_CAP_STABLE_WRITES;
3311 
3312 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3313 	if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3314 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3315 
3316 	ns->queue->queuedata = ns;
3317 	ns->ctrl = ctrl;
3318 
3319 	kref_init(&ns->kref);
3320 	ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3321 
3322 	blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3323 	nvme_set_queue_limits(ctrl, ns->queue);
3324 
3325 	id = nvme_identify_ns(ctrl, nsid);
3326 	if (!id) {
3327 		ret = -EIO;
3328 		goto out_free_queue;
3329 	}
3330 
3331 	if (id->ncap == 0) {
3332 		ret = -EINVAL;
3333 		goto out_free_id;
3334 	}
3335 
3336 	ret = nvme_init_ns_head(ns, nsid, id);
3337 	if (ret)
3338 		goto out_free_id;
3339 	nvme_setup_streams_ns(ctrl, ns);
3340 	nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3341 
3342 	disk = alloc_disk_node(0, node);
3343 	if (!disk) {
3344 		ret = -ENOMEM;
3345 		goto out_unlink_ns;
3346 	}
3347 
3348 	disk->fops = &nvme_fops;
3349 	disk->private_data = ns;
3350 	disk->queue = ns->queue;
3351 	disk->flags = flags;
3352 	memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3353 	ns->disk = disk;
3354 
3355 	__nvme_revalidate_disk(disk, id);
3356 
3357 	if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3358 		ret = nvme_nvm_register(ns, disk_name, node);
3359 		if (ret) {
3360 			dev_warn(ctrl->device, "LightNVM init failure\n");
3361 			goto out_put_disk;
3362 		}
3363 	}
3364 
3365 	down_write(&ctrl->namespaces_rwsem);
3366 	list_add_tail(&ns->list, &ctrl->namespaces);
3367 	up_write(&ctrl->namespaces_rwsem);
3368 
3369 	nvme_get_ctrl(ctrl);
3370 
3371 	device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3372 
3373 	nvme_mpath_add_disk(ns, id);
3374 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3375 	kfree(id);
3376 
3377 	return 0;
3378  out_put_disk:
3379 	put_disk(ns->disk);
3380  out_unlink_ns:
3381 	mutex_lock(&ctrl->subsys->lock);
3382 	list_del_rcu(&ns->siblings);
3383 	mutex_unlock(&ctrl->subsys->lock);
3384 	nvme_put_ns_head(ns->head);
3385  out_free_id:
3386 	kfree(id);
3387  out_free_queue:
3388 	blk_cleanup_queue(ns->queue);
3389  out_free_ns:
3390 	kfree(ns);
3391 	return ret;
3392 }
3393 
3394 static void nvme_ns_remove(struct nvme_ns *ns)
3395 {
3396 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3397 		return;
3398 
3399 	nvme_fault_inject_fini(&ns->fault_inject);
3400 
3401 	mutex_lock(&ns->ctrl->subsys->lock);
3402 	list_del_rcu(&ns->siblings);
3403 	mutex_unlock(&ns->ctrl->subsys->lock);
3404 	synchronize_rcu(); /* guarantee not available in head->list */
3405 	nvme_mpath_clear_current_path(ns);
3406 	synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3407 
3408 	if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3409 		del_gendisk(ns->disk);
3410 		blk_cleanup_queue(ns->queue);
3411 		if (blk_get_integrity(ns->disk))
3412 			blk_integrity_unregister(ns->disk);
3413 	}
3414 
3415 	down_write(&ns->ctrl->namespaces_rwsem);
3416 	list_del_init(&ns->list);
3417 	up_write(&ns->ctrl->namespaces_rwsem);
3418 
3419 	nvme_mpath_check_last_path(ns);
3420 	nvme_put_ns(ns);
3421 }
3422 
3423 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3424 {
3425 	struct nvme_ns *ns;
3426 
3427 	ns = nvme_find_get_ns(ctrl, nsid);
3428 	if (ns) {
3429 		if (ns->disk && revalidate_disk(ns->disk))
3430 			nvme_ns_remove(ns);
3431 		nvme_put_ns(ns);
3432 	} else
3433 		nvme_alloc_ns(ctrl, nsid);
3434 }
3435 
3436 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3437 					unsigned nsid)
3438 {
3439 	struct nvme_ns *ns, *next;
3440 	LIST_HEAD(rm_list);
3441 
3442 	down_write(&ctrl->namespaces_rwsem);
3443 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3444 		if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3445 			list_move_tail(&ns->list, &rm_list);
3446 	}
3447 	up_write(&ctrl->namespaces_rwsem);
3448 
3449 	list_for_each_entry_safe(ns, next, &rm_list, list)
3450 		nvme_ns_remove(ns);
3451 
3452 }
3453 
3454 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3455 {
3456 	struct nvme_ns *ns;
3457 	__le32 *ns_list;
3458 	unsigned i, j, nsid, prev = 0;
3459 	unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3460 	int ret = 0;
3461 
3462 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3463 	if (!ns_list)
3464 		return -ENOMEM;
3465 
3466 	for (i = 0; i < num_lists; i++) {
3467 		ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3468 		if (ret)
3469 			goto free;
3470 
3471 		for (j = 0; j < min(nn, 1024U); j++) {
3472 			nsid = le32_to_cpu(ns_list[j]);
3473 			if (!nsid)
3474 				goto out;
3475 
3476 			nvme_validate_ns(ctrl, nsid);
3477 
3478 			while (++prev < nsid) {
3479 				ns = nvme_find_get_ns(ctrl, prev);
3480 				if (ns) {
3481 					nvme_ns_remove(ns);
3482 					nvme_put_ns(ns);
3483 				}
3484 			}
3485 		}
3486 		nn -= j;
3487 	}
3488  out:
3489 	nvme_remove_invalid_namespaces(ctrl, prev);
3490  free:
3491 	kfree(ns_list);
3492 	return ret;
3493 }
3494 
3495 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3496 {
3497 	unsigned i;
3498 
3499 	for (i = 1; i <= nn; i++)
3500 		nvme_validate_ns(ctrl, i);
3501 
3502 	nvme_remove_invalid_namespaces(ctrl, nn);
3503 }
3504 
3505 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3506 {
3507 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3508 	__le32 *log;
3509 	int error;
3510 
3511 	log = kzalloc(log_size, GFP_KERNEL);
3512 	if (!log)
3513 		return;
3514 
3515 	/*
3516 	 * We need to read the log to clear the AEN, but we don't want to rely
3517 	 * on it for the changed namespace information as userspace could have
3518 	 * raced with us in reading the log page, which could cause us to miss
3519 	 * updates.
3520 	 */
3521 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3522 			log_size, 0);
3523 	if (error)
3524 		dev_warn(ctrl->device,
3525 			"reading changed ns log failed: %d\n", error);
3526 
3527 	kfree(log);
3528 }
3529 
3530 static void nvme_scan_work(struct work_struct *work)
3531 {
3532 	struct nvme_ctrl *ctrl =
3533 		container_of(work, struct nvme_ctrl, scan_work);
3534 	struct nvme_id_ctrl *id;
3535 	unsigned nn;
3536 
3537 	if (ctrl->state != NVME_CTRL_LIVE)
3538 		return;
3539 
3540 	WARN_ON_ONCE(!ctrl->tagset);
3541 
3542 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3543 		dev_info(ctrl->device, "rescanning namespaces.\n");
3544 		nvme_clear_changed_ns_log(ctrl);
3545 	}
3546 
3547 	if (nvme_identify_ctrl(ctrl, &id))
3548 		return;
3549 
3550 	mutex_lock(&ctrl->scan_lock);
3551 	nn = le32_to_cpu(id->nn);
3552 	if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3553 	    !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3554 		if (!nvme_scan_ns_list(ctrl, nn))
3555 			goto out_free_id;
3556 	}
3557 	nvme_scan_ns_sequential(ctrl, nn);
3558 out_free_id:
3559 	mutex_unlock(&ctrl->scan_lock);
3560 	kfree(id);
3561 	down_write(&ctrl->namespaces_rwsem);
3562 	list_sort(NULL, &ctrl->namespaces, ns_cmp);
3563 	up_write(&ctrl->namespaces_rwsem);
3564 }
3565 
3566 /*
3567  * This function iterates the namespace list unlocked to allow recovery from
3568  * controller failure. It is up to the caller to ensure the namespace list is
3569  * not modified by scan work while this function is executing.
3570  */
3571 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3572 {
3573 	struct nvme_ns *ns, *next;
3574 	LIST_HEAD(ns_list);
3575 
3576 	/* prevent racing with ns scanning */
3577 	flush_work(&ctrl->scan_work);
3578 
3579 	/*
3580 	 * The dead states indicates the controller was not gracefully
3581 	 * disconnected. In that case, we won't be able to flush any data while
3582 	 * removing the namespaces' disks; fail all the queues now to avoid
3583 	 * potentially having to clean up the failed sync later.
3584 	 */
3585 	if (ctrl->state == NVME_CTRL_DEAD)
3586 		nvme_kill_queues(ctrl);
3587 
3588 	down_write(&ctrl->namespaces_rwsem);
3589 	list_splice_init(&ctrl->namespaces, &ns_list);
3590 	up_write(&ctrl->namespaces_rwsem);
3591 
3592 	list_for_each_entry_safe(ns, next, &ns_list, list)
3593 		nvme_ns_remove(ns);
3594 }
3595 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3596 
3597 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3598 {
3599 	char *envp[2] = { NULL, NULL };
3600 	u32 aen_result = ctrl->aen_result;
3601 
3602 	ctrl->aen_result = 0;
3603 	if (!aen_result)
3604 		return;
3605 
3606 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3607 	if (!envp[0])
3608 		return;
3609 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3610 	kfree(envp[0]);
3611 }
3612 
3613 static void nvme_async_event_work(struct work_struct *work)
3614 {
3615 	struct nvme_ctrl *ctrl =
3616 		container_of(work, struct nvme_ctrl, async_event_work);
3617 
3618 	nvme_aen_uevent(ctrl);
3619 	ctrl->ops->submit_async_event(ctrl);
3620 }
3621 
3622 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3623 {
3624 
3625 	u32 csts;
3626 
3627 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3628 		return false;
3629 
3630 	if (csts == ~0)
3631 		return false;
3632 
3633 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3634 }
3635 
3636 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3637 {
3638 	struct nvme_fw_slot_info_log *log;
3639 
3640 	log = kmalloc(sizeof(*log), GFP_KERNEL);
3641 	if (!log)
3642 		return;
3643 
3644 	if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3645 			sizeof(*log), 0))
3646 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3647 	kfree(log);
3648 }
3649 
3650 static void nvme_fw_act_work(struct work_struct *work)
3651 {
3652 	struct nvme_ctrl *ctrl = container_of(work,
3653 				struct nvme_ctrl, fw_act_work);
3654 	unsigned long fw_act_timeout;
3655 
3656 	if (ctrl->mtfa)
3657 		fw_act_timeout = jiffies +
3658 				msecs_to_jiffies(ctrl->mtfa * 100);
3659 	else
3660 		fw_act_timeout = jiffies +
3661 				msecs_to_jiffies(admin_timeout * 1000);
3662 
3663 	nvme_stop_queues(ctrl);
3664 	while (nvme_ctrl_pp_status(ctrl)) {
3665 		if (time_after(jiffies, fw_act_timeout)) {
3666 			dev_warn(ctrl->device,
3667 				"Fw activation timeout, reset controller\n");
3668 			nvme_reset_ctrl(ctrl);
3669 			break;
3670 		}
3671 		msleep(100);
3672 	}
3673 
3674 	if (ctrl->state != NVME_CTRL_LIVE)
3675 		return;
3676 
3677 	nvme_start_queues(ctrl);
3678 	/* read FW slot information to clear the AER */
3679 	nvme_get_fw_slot_info(ctrl);
3680 }
3681 
3682 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3683 {
3684 	u32 aer_notice_type = (result & 0xff00) >> 8;
3685 
3686 	trace_nvme_async_event(ctrl, aer_notice_type);
3687 
3688 	switch (aer_notice_type) {
3689 	case NVME_AER_NOTICE_NS_CHANGED:
3690 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3691 		nvme_queue_scan(ctrl);
3692 		break;
3693 	case NVME_AER_NOTICE_FW_ACT_STARTING:
3694 		queue_work(nvme_wq, &ctrl->fw_act_work);
3695 		break;
3696 #ifdef CONFIG_NVME_MULTIPATH
3697 	case NVME_AER_NOTICE_ANA:
3698 		if (!ctrl->ana_log_buf)
3699 			break;
3700 		queue_work(nvme_wq, &ctrl->ana_work);
3701 		break;
3702 #endif
3703 	default:
3704 		dev_warn(ctrl->device, "async event result %08x\n", result);
3705 	}
3706 }
3707 
3708 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3709 		volatile union nvme_result *res)
3710 {
3711 	u32 result = le32_to_cpu(res->u32);
3712 	u32 aer_type = result & 0x07;
3713 
3714 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3715 		return;
3716 
3717 	switch (aer_type) {
3718 	case NVME_AER_NOTICE:
3719 		nvme_handle_aen_notice(ctrl, result);
3720 		break;
3721 	case NVME_AER_ERROR:
3722 	case NVME_AER_SMART:
3723 	case NVME_AER_CSS:
3724 	case NVME_AER_VS:
3725 		trace_nvme_async_event(ctrl, aer_type);
3726 		ctrl->aen_result = result;
3727 		break;
3728 	default:
3729 		break;
3730 	}
3731 	queue_work(nvme_wq, &ctrl->async_event_work);
3732 }
3733 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3734 
3735 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3736 {
3737 	nvme_mpath_stop(ctrl);
3738 	nvme_stop_keep_alive(ctrl);
3739 	flush_work(&ctrl->async_event_work);
3740 	cancel_work_sync(&ctrl->fw_act_work);
3741 }
3742 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3743 
3744 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3745 {
3746 	if (ctrl->kato)
3747 		nvme_start_keep_alive(ctrl);
3748 
3749 	if (ctrl->queue_count > 1) {
3750 		nvme_queue_scan(ctrl);
3751 		nvme_enable_aen(ctrl);
3752 		queue_work(nvme_wq, &ctrl->async_event_work);
3753 		nvme_start_queues(ctrl);
3754 	}
3755 }
3756 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3757 
3758 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3759 {
3760 	nvme_fault_inject_fini(&ctrl->fault_inject);
3761 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
3762 	cdev_device_del(&ctrl->cdev, ctrl->device);
3763 }
3764 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3765 
3766 static void nvme_free_ctrl(struct device *dev)
3767 {
3768 	struct nvme_ctrl *ctrl =
3769 		container_of(dev, struct nvme_ctrl, ctrl_device);
3770 	struct nvme_subsystem *subsys = ctrl->subsys;
3771 
3772 	ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3773 	kfree(ctrl->effects);
3774 	nvme_mpath_uninit(ctrl);
3775 	__free_page(ctrl->discard_page);
3776 
3777 	if (subsys) {
3778 		mutex_lock(&nvme_subsystems_lock);
3779 		list_del(&ctrl->subsys_entry);
3780 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3781 		mutex_unlock(&nvme_subsystems_lock);
3782 	}
3783 
3784 	ctrl->ops->free_ctrl(ctrl);
3785 
3786 	if (subsys)
3787 		nvme_put_subsystem(subsys);
3788 }
3789 
3790 /*
3791  * Initialize a NVMe controller structures.  This needs to be called during
3792  * earliest initialization so that we have the initialized structured around
3793  * during probing.
3794  */
3795 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3796 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
3797 {
3798 	int ret;
3799 
3800 	ctrl->state = NVME_CTRL_NEW;
3801 	spin_lock_init(&ctrl->lock);
3802 	mutex_init(&ctrl->scan_lock);
3803 	INIT_LIST_HEAD(&ctrl->namespaces);
3804 	init_rwsem(&ctrl->namespaces_rwsem);
3805 	ctrl->dev = dev;
3806 	ctrl->ops = ops;
3807 	ctrl->quirks = quirks;
3808 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3809 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3810 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3811 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3812 
3813 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3814 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3815 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3816 
3817 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3818 			PAGE_SIZE);
3819 	ctrl->discard_page = alloc_page(GFP_KERNEL);
3820 	if (!ctrl->discard_page) {
3821 		ret = -ENOMEM;
3822 		goto out;
3823 	}
3824 
3825 	ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3826 	if (ret < 0)
3827 		goto out;
3828 	ctrl->instance = ret;
3829 
3830 	device_initialize(&ctrl->ctrl_device);
3831 	ctrl->device = &ctrl->ctrl_device;
3832 	ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3833 	ctrl->device->class = nvme_class;
3834 	ctrl->device->parent = ctrl->dev;
3835 	ctrl->device->groups = nvme_dev_attr_groups;
3836 	ctrl->device->release = nvme_free_ctrl;
3837 	dev_set_drvdata(ctrl->device, ctrl);
3838 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3839 	if (ret)
3840 		goto out_release_instance;
3841 
3842 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
3843 	ctrl->cdev.owner = ops->module;
3844 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3845 	if (ret)
3846 		goto out_free_name;
3847 
3848 	/*
3849 	 * Initialize latency tolerance controls.  The sysfs files won't
3850 	 * be visible to userspace unless the device actually supports APST.
3851 	 */
3852 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3853 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3854 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3855 
3856 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
3857 
3858 	return 0;
3859 out_free_name:
3860 	kfree_const(ctrl->device->kobj.name);
3861 out_release_instance:
3862 	ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3863 out:
3864 	if (ctrl->discard_page)
3865 		__free_page(ctrl->discard_page);
3866 	return ret;
3867 }
3868 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3869 
3870 /**
3871  * nvme_kill_queues(): Ends all namespace queues
3872  * @ctrl: the dead controller that needs to end
3873  *
3874  * Call this function when the driver determines it is unable to get the
3875  * controller in a state capable of servicing IO.
3876  */
3877 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3878 {
3879 	struct nvme_ns *ns;
3880 
3881 	down_read(&ctrl->namespaces_rwsem);
3882 
3883 	/* Forcibly unquiesce queues to avoid blocking dispatch */
3884 	if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3885 		blk_mq_unquiesce_queue(ctrl->admin_q);
3886 
3887 	list_for_each_entry(ns, &ctrl->namespaces, list)
3888 		nvme_set_queue_dying(ns);
3889 
3890 	up_read(&ctrl->namespaces_rwsem);
3891 }
3892 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3893 
3894 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3895 {
3896 	struct nvme_ns *ns;
3897 
3898 	down_read(&ctrl->namespaces_rwsem);
3899 	list_for_each_entry(ns, &ctrl->namespaces, list)
3900 		blk_mq_unfreeze_queue(ns->queue);
3901 	up_read(&ctrl->namespaces_rwsem);
3902 }
3903 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3904 
3905 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3906 {
3907 	struct nvme_ns *ns;
3908 
3909 	down_read(&ctrl->namespaces_rwsem);
3910 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3911 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3912 		if (timeout <= 0)
3913 			break;
3914 	}
3915 	up_read(&ctrl->namespaces_rwsem);
3916 }
3917 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3918 
3919 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3920 {
3921 	struct nvme_ns *ns;
3922 
3923 	down_read(&ctrl->namespaces_rwsem);
3924 	list_for_each_entry(ns, &ctrl->namespaces, list)
3925 		blk_mq_freeze_queue_wait(ns->queue);
3926 	up_read(&ctrl->namespaces_rwsem);
3927 }
3928 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3929 
3930 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3931 {
3932 	struct nvme_ns *ns;
3933 
3934 	down_read(&ctrl->namespaces_rwsem);
3935 	list_for_each_entry(ns, &ctrl->namespaces, list)
3936 		blk_freeze_queue_start(ns->queue);
3937 	up_read(&ctrl->namespaces_rwsem);
3938 }
3939 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3940 
3941 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3942 {
3943 	struct nvme_ns *ns;
3944 
3945 	down_read(&ctrl->namespaces_rwsem);
3946 	list_for_each_entry(ns, &ctrl->namespaces, list)
3947 		blk_mq_quiesce_queue(ns->queue);
3948 	up_read(&ctrl->namespaces_rwsem);
3949 }
3950 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3951 
3952 void nvme_start_queues(struct nvme_ctrl *ctrl)
3953 {
3954 	struct nvme_ns *ns;
3955 
3956 	down_read(&ctrl->namespaces_rwsem);
3957 	list_for_each_entry(ns, &ctrl->namespaces, list)
3958 		blk_mq_unquiesce_queue(ns->queue);
3959 	up_read(&ctrl->namespaces_rwsem);
3960 }
3961 EXPORT_SYMBOL_GPL(nvme_start_queues);
3962 
3963 
3964 void nvme_sync_queues(struct nvme_ctrl *ctrl)
3965 {
3966 	struct nvme_ns *ns;
3967 
3968 	down_read(&ctrl->namespaces_rwsem);
3969 	list_for_each_entry(ns, &ctrl->namespaces, list)
3970 		blk_sync_queue(ns->queue);
3971 	up_read(&ctrl->namespaces_rwsem);
3972 }
3973 EXPORT_SYMBOL_GPL(nvme_sync_queues);
3974 
3975 /*
3976  * Check we didn't inadvertently grow the command structure sizes:
3977  */
3978 static inline void _nvme_check_size(void)
3979 {
3980 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
3981 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
3982 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
3983 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
3984 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
3985 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
3986 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
3987 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
3988 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
3989 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
3990 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
3991 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
3992 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
3993 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
3994 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
3995 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
3996 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
3997 }
3998 
3999 
4000 static int __init nvme_core_init(void)
4001 {
4002 	int result = -ENOMEM;
4003 
4004 	_nvme_check_size();
4005 
4006 	nvme_wq = alloc_workqueue("nvme-wq",
4007 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4008 	if (!nvme_wq)
4009 		goto out;
4010 
4011 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4012 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4013 	if (!nvme_reset_wq)
4014 		goto destroy_wq;
4015 
4016 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4017 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4018 	if (!nvme_delete_wq)
4019 		goto destroy_reset_wq;
4020 
4021 	result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4022 	if (result < 0)
4023 		goto destroy_delete_wq;
4024 
4025 	nvme_class = class_create(THIS_MODULE, "nvme");
4026 	if (IS_ERR(nvme_class)) {
4027 		result = PTR_ERR(nvme_class);
4028 		goto unregister_chrdev;
4029 	}
4030 
4031 	nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4032 	if (IS_ERR(nvme_subsys_class)) {
4033 		result = PTR_ERR(nvme_subsys_class);
4034 		goto destroy_class;
4035 	}
4036 	return 0;
4037 
4038 destroy_class:
4039 	class_destroy(nvme_class);
4040 unregister_chrdev:
4041 	unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4042 destroy_delete_wq:
4043 	destroy_workqueue(nvme_delete_wq);
4044 destroy_reset_wq:
4045 	destroy_workqueue(nvme_reset_wq);
4046 destroy_wq:
4047 	destroy_workqueue(nvme_wq);
4048 out:
4049 	return result;
4050 }
4051 
4052 static void __exit nvme_core_exit(void)
4053 {
4054 	ida_destroy(&nvme_subsystems_ida);
4055 	class_destroy(nvme_subsys_class);
4056 	class_destroy(nvme_class);
4057 	unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4058 	destroy_workqueue(nvme_delete_wq);
4059 	destroy_workqueue(nvme_reset_wq);
4060 	destroy_workqueue(nvme_wq);
4061 }
4062 
4063 MODULE_LICENSE("GPL");
4064 MODULE_VERSION("1.0");
4065 module_init(nvme_core_init);
4066 module_exit(nvme_core_exit);
4067