xref: /openbmc/linux/drivers/nvme/host/core.c (revision 7a2eb736)
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 		mutex_lock(&ctrl->subsys->lock);
1290 		nvme_mpath_start_freeze(ctrl->subsys);
1291 		nvme_mpath_wait_freeze(ctrl->subsys);
1292 		nvme_start_freeze(ctrl);
1293 		nvme_wait_freeze(ctrl);
1294 	}
1295 	return effects;
1296 }
1297 
1298 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1299 {
1300 	struct nvme_ns *ns;
1301 
1302 	down_read(&ctrl->namespaces_rwsem);
1303 	list_for_each_entry(ns, &ctrl->namespaces, list)
1304 		if (ns->disk && nvme_revalidate_disk(ns->disk))
1305 			nvme_set_queue_dying(ns);
1306 	up_read(&ctrl->namespaces_rwsem);
1307 
1308 	nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1309 }
1310 
1311 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1312 {
1313 	/*
1314 	 * Revalidate LBA changes prior to unfreezing. This is necessary to
1315 	 * prevent memory corruption if a logical block size was changed by
1316 	 * this command.
1317 	 */
1318 	if (effects & NVME_CMD_EFFECTS_LBCC)
1319 		nvme_update_formats(ctrl);
1320 	if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1321 		nvme_unfreeze(ctrl);
1322 		nvme_mpath_unfreeze(ctrl->subsys);
1323 		mutex_unlock(&ctrl->subsys->lock);
1324 		mutex_unlock(&ctrl->scan_lock);
1325 	}
1326 	if (effects & NVME_CMD_EFFECTS_CCC)
1327 		nvme_init_identify(ctrl);
1328 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1329 		nvme_queue_scan(ctrl);
1330 }
1331 
1332 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1333 			struct nvme_passthru_cmd __user *ucmd)
1334 {
1335 	struct nvme_passthru_cmd cmd;
1336 	struct nvme_command c;
1337 	unsigned timeout = 0;
1338 	u32 effects;
1339 	int status;
1340 
1341 	if (!capable(CAP_SYS_ADMIN))
1342 		return -EACCES;
1343 	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1344 		return -EFAULT;
1345 	if (cmd.flags)
1346 		return -EINVAL;
1347 
1348 	memset(&c, 0, sizeof(c));
1349 	c.common.opcode = cmd.opcode;
1350 	c.common.flags = cmd.flags;
1351 	c.common.nsid = cpu_to_le32(cmd.nsid);
1352 	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1353 	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1354 	c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1355 	c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1356 	c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1357 	c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1358 	c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1359 	c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1360 
1361 	if (cmd.timeout_ms)
1362 		timeout = msecs_to_jiffies(cmd.timeout_ms);
1363 
1364 	effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1365 	status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1366 			(void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1367 			(void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1368 			0, &cmd.result, timeout);
1369 	nvme_passthru_end(ctrl, effects);
1370 
1371 	if (status >= 0) {
1372 		if (put_user(cmd.result, &ucmd->result))
1373 			return -EFAULT;
1374 	}
1375 
1376 	return status;
1377 }
1378 
1379 /*
1380  * Issue ioctl requests on the first available path.  Note that unlike normal
1381  * block layer requests we will not retry failed request on another controller.
1382  */
1383 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1384 		struct nvme_ns_head **head, int *srcu_idx)
1385 {
1386 #ifdef CONFIG_NVME_MULTIPATH
1387 	if (disk->fops == &nvme_ns_head_ops) {
1388 		struct nvme_ns *ns;
1389 
1390 		*head = disk->private_data;
1391 		*srcu_idx = srcu_read_lock(&(*head)->srcu);
1392 		ns = nvme_find_path(*head);
1393 		if (!ns)
1394 			srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1395 		return ns;
1396 	}
1397 #endif
1398 	*head = NULL;
1399 	*srcu_idx = -1;
1400 	return disk->private_data;
1401 }
1402 
1403 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1404 {
1405 	if (head)
1406 		srcu_read_unlock(&head->srcu, idx);
1407 }
1408 
1409 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1410 		unsigned int cmd, unsigned long arg)
1411 {
1412 	struct nvme_ns_head *head = NULL;
1413 	void __user *argp = (void __user *)arg;
1414 	struct nvme_ns *ns;
1415 	int srcu_idx, ret;
1416 
1417 	ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1418 	if (unlikely(!ns))
1419 		return -EWOULDBLOCK;
1420 
1421 	/*
1422 	 * Handle ioctls that apply to the controller instead of the namespace
1423 	 * seperately and drop the ns SRCU reference early.  This avoids a
1424 	 * deadlock when deleting namespaces using the passthrough interface.
1425 	 */
1426 	if (cmd == NVME_IOCTL_ADMIN_CMD || is_sed_ioctl(cmd)) {
1427 		struct nvme_ctrl *ctrl = ns->ctrl;
1428 
1429 		nvme_get_ctrl(ns->ctrl);
1430 		nvme_put_ns_from_disk(head, srcu_idx);
1431 
1432 		if (cmd == NVME_IOCTL_ADMIN_CMD)
1433 			ret = nvme_user_cmd(ctrl, NULL, argp);
1434 		else
1435 			ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1436 
1437 		nvme_put_ctrl(ctrl);
1438 		return ret;
1439 	}
1440 
1441 	switch (cmd) {
1442 	case NVME_IOCTL_ID:
1443 		force_successful_syscall_return();
1444 		ret = ns->head->ns_id;
1445 		break;
1446 	case NVME_IOCTL_IO_CMD:
1447 		ret = nvme_user_cmd(ns->ctrl, ns, argp);
1448 		break;
1449 	case NVME_IOCTL_SUBMIT_IO:
1450 		ret = nvme_submit_io(ns, argp);
1451 		break;
1452 	default:
1453 		if (ns->ndev)
1454 			ret = nvme_nvm_ioctl(ns, cmd, arg);
1455 		else
1456 			ret = -ENOTTY;
1457 	}
1458 
1459 	nvme_put_ns_from_disk(head, srcu_idx);
1460 	return ret;
1461 }
1462 
1463 static int nvme_open(struct block_device *bdev, fmode_t mode)
1464 {
1465 	struct nvme_ns *ns = bdev->bd_disk->private_data;
1466 
1467 #ifdef CONFIG_NVME_MULTIPATH
1468 	/* should never be called due to GENHD_FL_HIDDEN */
1469 	if (WARN_ON_ONCE(ns->head->disk))
1470 		goto fail;
1471 #endif
1472 	if (!kref_get_unless_zero(&ns->kref))
1473 		goto fail;
1474 	if (!try_module_get(ns->ctrl->ops->module))
1475 		goto fail_put_ns;
1476 
1477 	return 0;
1478 
1479 fail_put_ns:
1480 	nvme_put_ns(ns);
1481 fail:
1482 	return -ENXIO;
1483 }
1484 
1485 static void nvme_release(struct gendisk *disk, fmode_t mode)
1486 {
1487 	struct nvme_ns *ns = disk->private_data;
1488 
1489 	module_put(ns->ctrl->ops->module);
1490 	nvme_put_ns(ns);
1491 }
1492 
1493 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1494 {
1495 	/* some standard values */
1496 	geo->heads = 1 << 6;
1497 	geo->sectors = 1 << 5;
1498 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1499 	return 0;
1500 }
1501 
1502 #ifdef CONFIG_BLK_DEV_INTEGRITY
1503 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1504 {
1505 	struct blk_integrity integrity;
1506 
1507 	memset(&integrity, 0, sizeof(integrity));
1508 	switch (pi_type) {
1509 	case NVME_NS_DPS_PI_TYPE3:
1510 		integrity.profile = &t10_pi_type3_crc;
1511 		integrity.tag_size = sizeof(u16) + sizeof(u32);
1512 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1513 		break;
1514 	case NVME_NS_DPS_PI_TYPE1:
1515 	case NVME_NS_DPS_PI_TYPE2:
1516 		integrity.profile = &t10_pi_type1_crc;
1517 		integrity.tag_size = sizeof(u16);
1518 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1519 		break;
1520 	default:
1521 		integrity.profile = NULL;
1522 		break;
1523 	}
1524 	integrity.tuple_size = ms;
1525 	blk_integrity_register(disk, &integrity);
1526 	blk_queue_max_integrity_segments(disk->queue, 1);
1527 }
1528 #else
1529 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1530 {
1531 }
1532 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1533 
1534 static void nvme_set_chunk_size(struct nvme_ns *ns)
1535 {
1536 	u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1537 	blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1538 }
1539 
1540 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1541 {
1542 	struct nvme_ctrl *ctrl = ns->ctrl;
1543 	struct request_queue *queue = disk->queue;
1544 	u32 size = queue_logical_block_size(queue);
1545 
1546 	if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1547 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1548 		return;
1549 	}
1550 
1551 	if (ctrl->nr_streams && ns->sws && ns->sgs)
1552 		size *= ns->sws * ns->sgs;
1553 
1554 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1555 			NVME_DSM_MAX_RANGES);
1556 
1557 	queue->limits.discard_alignment = 0;
1558 	queue->limits.discard_granularity = size;
1559 
1560 	/* If discard is already enabled, don't reset queue limits */
1561 	if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1562 		return;
1563 
1564 	blk_queue_max_discard_sectors(queue, UINT_MAX);
1565 	blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1566 
1567 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1568 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1569 }
1570 
1571 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1572 {
1573 	u32 max_sectors;
1574 	unsigned short bs = 1 << ns->lba_shift;
1575 
1576 	if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1577 	    (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1578 		return;
1579 	/*
1580 	 * Even though NVMe spec explicitly states that MDTS is not
1581 	 * applicable to the write-zeroes:- "The restriction does not apply to
1582 	 * commands that do not transfer data between the host and the
1583 	 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1584 	 * In order to be more cautious use controller's max_hw_sectors value
1585 	 * to configure the maximum sectors for the write-zeroes which is
1586 	 * configured based on the controller's MDTS field in the
1587 	 * nvme_init_identify() if available.
1588 	 */
1589 	if (ns->ctrl->max_hw_sectors == UINT_MAX)
1590 		max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1591 	else
1592 		max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1593 
1594 	blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1595 }
1596 
1597 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1598 		struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1599 {
1600 	memset(ids, 0, sizeof(*ids));
1601 
1602 	if (ctrl->vs >= NVME_VS(1, 1, 0))
1603 		memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1604 	if (ctrl->vs >= NVME_VS(1, 2, 0))
1605 		memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1606 	if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1607 		 /* Don't treat error as fatal we potentially
1608 		  * already have a NGUID or EUI-64
1609 		  */
1610 		if (nvme_identify_ns_descs(ctrl, nsid, ids))
1611 			dev_warn(ctrl->device,
1612 				 "%s: Identify Descriptors failed\n", __func__);
1613 	}
1614 }
1615 
1616 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1617 {
1618 	return !uuid_is_null(&ids->uuid) ||
1619 		memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1620 		memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1621 }
1622 
1623 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1624 {
1625 	return uuid_equal(&a->uuid, &b->uuid) &&
1626 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1627 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1628 }
1629 
1630 static void nvme_update_disk_info(struct gendisk *disk,
1631 		struct nvme_ns *ns, struct nvme_id_ns *id)
1632 {
1633 	sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1634 	unsigned short bs = 1 << ns->lba_shift;
1635 	u32 atomic_bs, phys_bs, io_opt;
1636 
1637 	if (ns->lba_shift > PAGE_SHIFT) {
1638 		/* unsupported block size, set capacity to 0 later */
1639 		bs = (1 << 9);
1640 	}
1641 	blk_mq_freeze_queue(disk->queue);
1642 	blk_integrity_unregister(disk);
1643 
1644 	if (id->nabo == 0) {
1645 		/*
1646 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1647 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1648 		 * 0 then AWUPF must be used instead.
1649 		 */
1650 		if (id->nsfeat & (1 << 1) && id->nawupf)
1651 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1652 		else
1653 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1654 	} else {
1655 		atomic_bs = bs;
1656 	}
1657 	phys_bs = bs;
1658 	io_opt = bs;
1659 	if (id->nsfeat & (1 << 4)) {
1660 		/* NPWG = Namespace Preferred Write Granularity */
1661 		phys_bs *= 1 + le16_to_cpu(id->npwg);
1662 		/* NOWS = Namespace Optimal Write Size */
1663 		io_opt *= 1 + le16_to_cpu(id->nows);
1664 	}
1665 
1666 	blk_queue_logical_block_size(disk->queue, bs);
1667 	/*
1668 	 * Linux filesystems assume writing a single physical block is
1669 	 * an atomic operation. Hence limit the physical block size to the
1670 	 * value of the Atomic Write Unit Power Fail parameter.
1671 	 */
1672 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1673 	blk_queue_io_min(disk->queue, phys_bs);
1674 	blk_queue_io_opt(disk->queue, io_opt);
1675 
1676 	if (ns->ms && !ns->ext &&
1677 	    (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1678 		nvme_init_integrity(disk, ns->ms, ns->pi_type);
1679 	if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1680 	    ns->lba_shift > PAGE_SHIFT)
1681 		capacity = 0;
1682 
1683 	set_capacity(disk, capacity);
1684 
1685 	nvme_config_discard(disk, ns);
1686 	nvme_config_write_zeroes(disk, ns);
1687 
1688 	if (id->nsattr & (1 << 0))
1689 		set_disk_ro(disk, true);
1690 	else
1691 		set_disk_ro(disk, false);
1692 
1693 	blk_mq_unfreeze_queue(disk->queue);
1694 }
1695 
1696 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1697 {
1698 	struct nvme_ns *ns = disk->private_data;
1699 
1700 	/*
1701 	 * If identify namespace failed, use default 512 byte block size so
1702 	 * block layer can use before failing read/write for 0 capacity.
1703 	 */
1704 	ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1705 	if (ns->lba_shift == 0)
1706 		ns->lba_shift = 9;
1707 	ns->noiob = le16_to_cpu(id->noiob);
1708 	ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1709 	ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1710 	/* the PI implementation requires metadata equal t10 pi tuple size */
1711 	if (ns->ms == sizeof(struct t10_pi_tuple))
1712 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1713 	else
1714 		ns->pi_type = 0;
1715 
1716 	if (ns->noiob)
1717 		nvme_set_chunk_size(ns);
1718 	nvme_update_disk_info(disk, ns, id);
1719 #ifdef CONFIG_NVME_MULTIPATH
1720 	if (ns->head->disk) {
1721 		nvme_update_disk_info(ns->head->disk, ns, id);
1722 		blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1723 		revalidate_disk(ns->head->disk);
1724 	}
1725 #endif
1726 }
1727 
1728 static int nvme_revalidate_disk(struct gendisk *disk)
1729 {
1730 	struct nvme_ns *ns = disk->private_data;
1731 	struct nvme_ctrl *ctrl = ns->ctrl;
1732 	struct nvme_id_ns *id;
1733 	struct nvme_ns_ids ids;
1734 	int ret = 0;
1735 
1736 	if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1737 		set_capacity(disk, 0);
1738 		return -ENODEV;
1739 	}
1740 
1741 	id = nvme_identify_ns(ctrl, ns->head->ns_id);
1742 	if (!id)
1743 		return -ENODEV;
1744 
1745 	if (id->ncap == 0) {
1746 		ret = -ENODEV;
1747 		goto out;
1748 	}
1749 
1750 	__nvme_revalidate_disk(disk, id);
1751 	nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1752 	if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1753 		dev_err(ctrl->device,
1754 			"identifiers changed for nsid %d\n", ns->head->ns_id);
1755 		ret = -ENODEV;
1756 	}
1757 
1758 out:
1759 	kfree(id);
1760 	return ret;
1761 }
1762 
1763 static char nvme_pr_type(enum pr_type type)
1764 {
1765 	switch (type) {
1766 	case PR_WRITE_EXCLUSIVE:
1767 		return 1;
1768 	case PR_EXCLUSIVE_ACCESS:
1769 		return 2;
1770 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1771 		return 3;
1772 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1773 		return 4;
1774 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1775 		return 5;
1776 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1777 		return 6;
1778 	default:
1779 		return 0;
1780 	}
1781 };
1782 
1783 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1784 				u64 key, u64 sa_key, u8 op)
1785 {
1786 	struct nvme_ns_head *head = NULL;
1787 	struct nvme_ns *ns;
1788 	struct nvme_command c;
1789 	int srcu_idx, ret;
1790 	u8 data[16] = { 0, };
1791 
1792 	ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1793 	if (unlikely(!ns))
1794 		return -EWOULDBLOCK;
1795 
1796 	put_unaligned_le64(key, &data[0]);
1797 	put_unaligned_le64(sa_key, &data[8]);
1798 
1799 	memset(&c, 0, sizeof(c));
1800 	c.common.opcode = op;
1801 	c.common.nsid = cpu_to_le32(ns->head->ns_id);
1802 	c.common.cdw10 = cpu_to_le32(cdw10);
1803 
1804 	ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1805 	nvme_put_ns_from_disk(head, srcu_idx);
1806 	return ret;
1807 }
1808 
1809 static int nvme_pr_register(struct block_device *bdev, u64 old,
1810 		u64 new, unsigned flags)
1811 {
1812 	u32 cdw10;
1813 
1814 	if (flags & ~PR_FL_IGNORE_KEY)
1815 		return -EOPNOTSUPP;
1816 
1817 	cdw10 = old ? 2 : 0;
1818 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1819 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1820 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1821 }
1822 
1823 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1824 		enum pr_type type, unsigned flags)
1825 {
1826 	u32 cdw10;
1827 
1828 	if (flags & ~PR_FL_IGNORE_KEY)
1829 		return -EOPNOTSUPP;
1830 
1831 	cdw10 = nvme_pr_type(type) << 8;
1832 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1833 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1834 }
1835 
1836 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1837 		enum pr_type type, bool abort)
1838 {
1839 	u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1840 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1841 }
1842 
1843 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1844 {
1845 	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1846 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1847 }
1848 
1849 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1850 {
1851 	u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1852 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1853 }
1854 
1855 static const struct pr_ops nvme_pr_ops = {
1856 	.pr_register	= nvme_pr_register,
1857 	.pr_reserve	= nvme_pr_reserve,
1858 	.pr_release	= nvme_pr_release,
1859 	.pr_preempt	= nvme_pr_preempt,
1860 	.pr_clear	= nvme_pr_clear,
1861 };
1862 
1863 #ifdef CONFIG_BLK_SED_OPAL
1864 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1865 		bool send)
1866 {
1867 	struct nvme_ctrl *ctrl = data;
1868 	struct nvme_command cmd;
1869 
1870 	memset(&cmd, 0, sizeof(cmd));
1871 	if (send)
1872 		cmd.common.opcode = nvme_admin_security_send;
1873 	else
1874 		cmd.common.opcode = nvme_admin_security_recv;
1875 	cmd.common.nsid = 0;
1876 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1877 	cmd.common.cdw11 = cpu_to_le32(len);
1878 
1879 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1880 				      ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1881 }
1882 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1883 #endif /* CONFIG_BLK_SED_OPAL */
1884 
1885 static const struct block_device_operations nvme_fops = {
1886 	.owner		= THIS_MODULE,
1887 	.ioctl		= nvme_ioctl,
1888 	.compat_ioctl	= nvme_ioctl,
1889 	.open		= nvme_open,
1890 	.release	= nvme_release,
1891 	.getgeo		= nvme_getgeo,
1892 	.revalidate_disk= nvme_revalidate_disk,
1893 	.pr_ops		= &nvme_pr_ops,
1894 };
1895 
1896 #ifdef CONFIG_NVME_MULTIPATH
1897 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1898 {
1899 	struct nvme_ns_head *head = bdev->bd_disk->private_data;
1900 
1901 	if (!kref_get_unless_zero(&head->ref))
1902 		return -ENXIO;
1903 	return 0;
1904 }
1905 
1906 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1907 {
1908 	nvme_put_ns_head(disk->private_data);
1909 }
1910 
1911 const struct block_device_operations nvme_ns_head_ops = {
1912 	.owner		= THIS_MODULE,
1913 	.open		= nvme_ns_head_open,
1914 	.release	= nvme_ns_head_release,
1915 	.ioctl		= nvme_ioctl,
1916 	.compat_ioctl	= nvme_ioctl,
1917 	.getgeo		= nvme_getgeo,
1918 	.pr_ops		= &nvme_pr_ops,
1919 };
1920 #endif /* CONFIG_NVME_MULTIPATH */
1921 
1922 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1923 {
1924 	unsigned long timeout =
1925 		((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1926 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1927 	int ret;
1928 
1929 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1930 		if (csts == ~0)
1931 			return -ENODEV;
1932 		if ((csts & NVME_CSTS_RDY) == bit)
1933 			break;
1934 
1935 		msleep(100);
1936 		if (fatal_signal_pending(current))
1937 			return -EINTR;
1938 		if (time_after(jiffies, timeout)) {
1939 			dev_err(ctrl->device,
1940 				"Device not ready; aborting %s\n", enabled ?
1941 						"initialisation" : "reset");
1942 			return -ENODEV;
1943 		}
1944 	}
1945 
1946 	return ret;
1947 }
1948 
1949 /*
1950  * If the device has been passed off to us in an enabled state, just clear
1951  * the enabled bit.  The spec says we should set the 'shutdown notification
1952  * bits', but doing so may cause the device to complete commands to the
1953  * admin queue ... and we don't know what memory that might be pointing at!
1954  */
1955 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1956 {
1957 	int ret;
1958 
1959 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1960 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1961 
1962 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1963 	if (ret)
1964 		return ret;
1965 
1966 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1967 		msleep(NVME_QUIRK_DELAY_AMOUNT);
1968 
1969 	return nvme_wait_ready(ctrl, cap, false);
1970 }
1971 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1972 
1973 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1974 {
1975 	/*
1976 	 * Default to a 4K page size, with the intention to update this
1977 	 * path in the future to accomodate architectures with differing
1978 	 * kernel and IO page sizes.
1979 	 */
1980 	unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1981 	int ret;
1982 
1983 	if (page_shift < dev_page_min) {
1984 		dev_err(ctrl->device,
1985 			"Minimum device page size %u too large for host (%u)\n",
1986 			1 << dev_page_min, 1 << page_shift);
1987 		return -ENODEV;
1988 	}
1989 
1990 	ctrl->page_size = 1 << page_shift;
1991 
1992 	ctrl->ctrl_config = NVME_CC_CSS_NVM;
1993 	ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1994 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1995 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1996 	ctrl->ctrl_config |= NVME_CC_ENABLE;
1997 
1998 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1999 	if (ret)
2000 		return ret;
2001 	return nvme_wait_ready(ctrl, cap, true);
2002 }
2003 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2004 
2005 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2006 {
2007 	unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2008 	u32 csts;
2009 	int ret;
2010 
2011 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2012 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2013 
2014 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2015 	if (ret)
2016 		return ret;
2017 
2018 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2019 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2020 			break;
2021 
2022 		msleep(100);
2023 		if (fatal_signal_pending(current))
2024 			return -EINTR;
2025 		if (time_after(jiffies, timeout)) {
2026 			dev_err(ctrl->device,
2027 				"Device shutdown incomplete; abort shutdown\n");
2028 			return -ENODEV;
2029 		}
2030 	}
2031 
2032 	return ret;
2033 }
2034 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2035 
2036 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2037 		struct request_queue *q)
2038 {
2039 	bool vwc = false;
2040 
2041 	if (ctrl->max_hw_sectors) {
2042 		u32 max_segments =
2043 			(ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2044 
2045 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
2046 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2047 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2048 	}
2049 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2050 	    is_power_of_2(ctrl->max_hw_sectors))
2051 		blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2052 	blk_queue_virt_boundary(q, ctrl->page_size - 1);
2053 	if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2054 		vwc = true;
2055 	blk_queue_write_cache(q, vwc, vwc);
2056 }
2057 
2058 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2059 {
2060 	__le64 ts;
2061 	int ret;
2062 
2063 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2064 		return 0;
2065 
2066 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2067 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2068 			NULL);
2069 	if (ret)
2070 		dev_warn_once(ctrl->device,
2071 			"could not set timestamp (%d)\n", ret);
2072 	return ret;
2073 }
2074 
2075 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2076 {
2077 	struct nvme_feat_host_behavior *host;
2078 	int ret;
2079 
2080 	/* Don't bother enabling the feature if retry delay is not reported */
2081 	if (!ctrl->crdt[0])
2082 		return 0;
2083 
2084 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2085 	if (!host)
2086 		return 0;
2087 
2088 	host->acre = NVME_ENABLE_ACRE;
2089 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2090 				host, sizeof(*host), NULL);
2091 	kfree(host);
2092 	return ret;
2093 }
2094 
2095 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2096 {
2097 	/*
2098 	 * APST (Autonomous Power State Transition) lets us program a
2099 	 * table of power state transitions that the controller will
2100 	 * perform automatically.  We configure it with a simple
2101 	 * heuristic: we are willing to spend at most 2% of the time
2102 	 * transitioning between power states.  Therefore, when running
2103 	 * in any given state, we will enter the next lower-power
2104 	 * non-operational state after waiting 50 * (enlat + exlat)
2105 	 * microseconds, as long as that state's exit latency is under
2106 	 * the requested maximum latency.
2107 	 *
2108 	 * We will not autonomously enter any non-operational state for
2109 	 * which the total latency exceeds ps_max_latency_us.  Users
2110 	 * can set ps_max_latency_us to zero to turn off APST.
2111 	 */
2112 
2113 	unsigned apste;
2114 	struct nvme_feat_auto_pst *table;
2115 	u64 max_lat_us = 0;
2116 	int max_ps = -1;
2117 	int ret;
2118 
2119 	/*
2120 	 * If APST isn't supported or if we haven't been initialized yet,
2121 	 * then don't do anything.
2122 	 */
2123 	if (!ctrl->apsta)
2124 		return 0;
2125 
2126 	if (ctrl->npss > 31) {
2127 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2128 		return 0;
2129 	}
2130 
2131 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2132 	if (!table)
2133 		return 0;
2134 
2135 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2136 		/* Turn off APST. */
2137 		apste = 0;
2138 		dev_dbg(ctrl->device, "APST disabled\n");
2139 	} else {
2140 		__le64 target = cpu_to_le64(0);
2141 		int state;
2142 
2143 		/*
2144 		 * Walk through all states from lowest- to highest-power.
2145 		 * According to the spec, lower-numbered states use more
2146 		 * power.  NPSS, despite the name, is the index of the
2147 		 * lowest-power state, not the number of states.
2148 		 */
2149 		for (state = (int)ctrl->npss; state >= 0; state--) {
2150 			u64 total_latency_us, exit_latency_us, transition_ms;
2151 
2152 			if (target)
2153 				table->entries[state] = target;
2154 
2155 			/*
2156 			 * Don't allow transitions to the deepest state
2157 			 * if it's quirked off.
2158 			 */
2159 			if (state == ctrl->npss &&
2160 			    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2161 				continue;
2162 
2163 			/*
2164 			 * Is this state a useful non-operational state for
2165 			 * higher-power states to autonomously transition to?
2166 			 */
2167 			if (!(ctrl->psd[state].flags &
2168 			      NVME_PS_FLAGS_NON_OP_STATE))
2169 				continue;
2170 
2171 			exit_latency_us =
2172 				(u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2173 			if (exit_latency_us > ctrl->ps_max_latency_us)
2174 				continue;
2175 
2176 			total_latency_us =
2177 				exit_latency_us +
2178 				le32_to_cpu(ctrl->psd[state].entry_lat);
2179 
2180 			/*
2181 			 * This state is good.  Use it as the APST idle
2182 			 * target for higher power states.
2183 			 */
2184 			transition_ms = total_latency_us + 19;
2185 			do_div(transition_ms, 20);
2186 			if (transition_ms > (1 << 24) - 1)
2187 				transition_ms = (1 << 24) - 1;
2188 
2189 			target = cpu_to_le64((state << 3) |
2190 					     (transition_ms << 8));
2191 
2192 			if (max_ps == -1)
2193 				max_ps = state;
2194 
2195 			if (total_latency_us > max_lat_us)
2196 				max_lat_us = total_latency_us;
2197 		}
2198 
2199 		apste = 1;
2200 
2201 		if (max_ps == -1) {
2202 			dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2203 		} else {
2204 			dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2205 				max_ps, max_lat_us, (int)sizeof(*table), table);
2206 		}
2207 	}
2208 
2209 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2210 				table, sizeof(*table), NULL);
2211 	if (ret)
2212 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2213 
2214 	kfree(table);
2215 	return ret;
2216 }
2217 
2218 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2219 {
2220 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2221 	u64 latency;
2222 
2223 	switch (val) {
2224 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2225 	case PM_QOS_LATENCY_ANY:
2226 		latency = U64_MAX;
2227 		break;
2228 
2229 	default:
2230 		latency = val;
2231 	}
2232 
2233 	if (ctrl->ps_max_latency_us != latency) {
2234 		ctrl->ps_max_latency_us = latency;
2235 		nvme_configure_apst(ctrl);
2236 	}
2237 }
2238 
2239 struct nvme_core_quirk_entry {
2240 	/*
2241 	 * NVMe model and firmware strings are padded with spaces.  For
2242 	 * simplicity, strings in the quirk table are padded with NULLs
2243 	 * instead.
2244 	 */
2245 	u16 vid;
2246 	const char *mn;
2247 	const char *fr;
2248 	unsigned long quirks;
2249 };
2250 
2251 static const struct nvme_core_quirk_entry core_quirks[] = {
2252 	{
2253 		/*
2254 		 * This Toshiba device seems to die using any APST states.  See:
2255 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2256 		 */
2257 		.vid = 0x1179,
2258 		.mn = "THNSF5256GPUK TOSHIBA",
2259 		.quirks = NVME_QUIRK_NO_APST,
2260 	},
2261 	{
2262 		/*
2263 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2264 		 * condition associated with actions related to suspend to idle
2265 		 * LiteON has resolved the problem in future firmware
2266 		 */
2267 		.vid = 0x14a4,
2268 		.fr = "22301111",
2269 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2270 	}
2271 };
2272 
2273 /* match is null-terminated but idstr is space-padded. */
2274 static bool string_matches(const char *idstr, const char *match, size_t len)
2275 {
2276 	size_t matchlen;
2277 
2278 	if (!match)
2279 		return true;
2280 
2281 	matchlen = strlen(match);
2282 	WARN_ON_ONCE(matchlen > len);
2283 
2284 	if (memcmp(idstr, match, matchlen))
2285 		return false;
2286 
2287 	for (; matchlen < len; matchlen++)
2288 		if (idstr[matchlen] != ' ')
2289 			return false;
2290 
2291 	return true;
2292 }
2293 
2294 static bool quirk_matches(const struct nvme_id_ctrl *id,
2295 			  const struct nvme_core_quirk_entry *q)
2296 {
2297 	return q->vid == le16_to_cpu(id->vid) &&
2298 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2299 		string_matches(id->fr, q->fr, sizeof(id->fr));
2300 }
2301 
2302 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2303 		struct nvme_id_ctrl *id)
2304 {
2305 	size_t nqnlen;
2306 	int off;
2307 
2308 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2309 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2310 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2311 			strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2312 			return;
2313 		}
2314 
2315 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2316 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2317 	}
2318 
2319 	/* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2320 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2321 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2322 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2323 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2324 	off += sizeof(id->sn);
2325 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2326 	off += sizeof(id->mn);
2327 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2328 }
2329 
2330 static void nvme_release_subsystem(struct device *dev)
2331 {
2332 	struct nvme_subsystem *subsys =
2333 		container_of(dev, struct nvme_subsystem, dev);
2334 
2335 	ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2336 	kfree(subsys);
2337 }
2338 
2339 static void nvme_destroy_subsystem(struct kref *ref)
2340 {
2341 	struct nvme_subsystem *subsys =
2342 			container_of(ref, struct nvme_subsystem, ref);
2343 
2344 	mutex_lock(&nvme_subsystems_lock);
2345 	list_del(&subsys->entry);
2346 	mutex_unlock(&nvme_subsystems_lock);
2347 
2348 	ida_destroy(&subsys->ns_ida);
2349 	device_del(&subsys->dev);
2350 	put_device(&subsys->dev);
2351 }
2352 
2353 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2354 {
2355 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2356 }
2357 
2358 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2359 {
2360 	struct nvme_subsystem *subsys;
2361 
2362 	lockdep_assert_held(&nvme_subsystems_lock);
2363 
2364 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2365 		if (strcmp(subsys->subnqn, subsysnqn))
2366 			continue;
2367 		if (!kref_get_unless_zero(&subsys->ref))
2368 			continue;
2369 		return subsys;
2370 	}
2371 
2372 	return NULL;
2373 }
2374 
2375 #define SUBSYS_ATTR_RO(_name, _mode, _show)			\
2376 	struct device_attribute subsys_attr_##_name = \
2377 		__ATTR(_name, _mode, _show, NULL)
2378 
2379 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2380 				    struct device_attribute *attr,
2381 				    char *buf)
2382 {
2383 	struct nvme_subsystem *subsys =
2384 		container_of(dev, struct nvme_subsystem, dev);
2385 
2386 	return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2387 }
2388 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2389 
2390 #define nvme_subsys_show_str_function(field)				\
2391 static ssize_t subsys_##field##_show(struct device *dev,		\
2392 			    struct device_attribute *attr, char *buf)	\
2393 {									\
2394 	struct nvme_subsystem *subsys =					\
2395 		container_of(dev, struct nvme_subsystem, dev);		\
2396 	return sprintf(buf, "%.*s\n",					\
2397 		       (int)sizeof(subsys->field), subsys->field);	\
2398 }									\
2399 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2400 
2401 nvme_subsys_show_str_function(model);
2402 nvme_subsys_show_str_function(serial);
2403 nvme_subsys_show_str_function(firmware_rev);
2404 
2405 static struct attribute *nvme_subsys_attrs[] = {
2406 	&subsys_attr_model.attr,
2407 	&subsys_attr_serial.attr,
2408 	&subsys_attr_firmware_rev.attr,
2409 	&subsys_attr_subsysnqn.attr,
2410 #ifdef CONFIG_NVME_MULTIPATH
2411 	&subsys_attr_iopolicy.attr,
2412 #endif
2413 	NULL,
2414 };
2415 
2416 static struct attribute_group nvme_subsys_attrs_group = {
2417 	.attrs = nvme_subsys_attrs,
2418 };
2419 
2420 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2421 	&nvme_subsys_attrs_group,
2422 	NULL,
2423 };
2424 
2425 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2426 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2427 {
2428 	struct nvme_ctrl *tmp;
2429 
2430 	lockdep_assert_held(&nvme_subsystems_lock);
2431 
2432 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2433 		if (tmp->state == NVME_CTRL_DELETING ||
2434 		    tmp->state == NVME_CTRL_DEAD)
2435 			continue;
2436 
2437 		if (tmp->cntlid == ctrl->cntlid) {
2438 			dev_err(ctrl->device,
2439 				"Duplicate cntlid %u with %s, rejecting\n",
2440 				ctrl->cntlid, dev_name(tmp->device));
2441 			return false;
2442 		}
2443 
2444 		if ((id->cmic & (1 << 1)) ||
2445 		    (ctrl->opts && ctrl->opts->discovery_nqn))
2446 			continue;
2447 
2448 		dev_err(ctrl->device,
2449 			"Subsystem does not support multiple controllers\n");
2450 		return false;
2451 	}
2452 
2453 	return true;
2454 }
2455 
2456 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2457 {
2458 	struct nvme_subsystem *subsys, *found;
2459 	int ret;
2460 
2461 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2462 	if (!subsys)
2463 		return -ENOMEM;
2464 	ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2465 	if (ret < 0) {
2466 		kfree(subsys);
2467 		return ret;
2468 	}
2469 	subsys->instance = ret;
2470 	mutex_init(&subsys->lock);
2471 	kref_init(&subsys->ref);
2472 	INIT_LIST_HEAD(&subsys->ctrls);
2473 	INIT_LIST_HEAD(&subsys->nsheads);
2474 	nvme_init_subnqn(subsys, ctrl, id);
2475 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2476 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2477 	memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2478 	subsys->vendor_id = le16_to_cpu(id->vid);
2479 	subsys->cmic = id->cmic;
2480 	subsys->awupf = le16_to_cpu(id->awupf);
2481 #ifdef CONFIG_NVME_MULTIPATH
2482 	subsys->iopolicy = NVME_IOPOLICY_NUMA;
2483 #endif
2484 
2485 	subsys->dev.class = nvme_subsys_class;
2486 	subsys->dev.release = nvme_release_subsystem;
2487 	subsys->dev.groups = nvme_subsys_attrs_groups;
2488 	dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2489 	device_initialize(&subsys->dev);
2490 
2491 	mutex_lock(&nvme_subsystems_lock);
2492 	found = __nvme_find_get_subsystem(subsys->subnqn);
2493 	if (found) {
2494 		put_device(&subsys->dev);
2495 		subsys = found;
2496 
2497 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2498 			ret = -EINVAL;
2499 			goto out_put_subsystem;
2500 		}
2501 	} else {
2502 		ret = device_add(&subsys->dev);
2503 		if (ret) {
2504 			dev_err(ctrl->device,
2505 				"failed to register subsystem device.\n");
2506 			put_device(&subsys->dev);
2507 			goto out_unlock;
2508 		}
2509 		ida_init(&subsys->ns_ida);
2510 		list_add_tail(&subsys->entry, &nvme_subsystems);
2511 	}
2512 
2513 	if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2514 			dev_name(ctrl->device))) {
2515 		dev_err(ctrl->device,
2516 			"failed to create sysfs link from subsystem.\n");
2517 		goto out_put_subsystem;
2518 	}
2519 
2520 	ctrl->subsys = subsys;
2521 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2522 	mutex_unlock(&nvme_subsystems_lock);
2523 	return 0;
2524 
2525 out_put_subsystem:
2526 	nvme_put_subsystem(subsys);
2527 out_unlock:
2528 	mutex_unlock(&nvme_subsystems_lock);
2529 	return ret;
2530 }
2531 
2532 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2533 		void *log, size_t size, u64 offset)
2534 {
2535 	struct nvme_command c = { };
2536 	unsigned long dwlen = size / 4 - 1;
2537 
2538 	c.get_log_page.opcode = nvme_admin_get_log_page;
2539 	c.get_log_page.nsid = cpu_to_le32(nsid);
2540 	c.get_log_page.lid = log_page;
2541 	c.get_log_page.lsp = lsp;
2542 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2543 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2544 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2545 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2546 
2547 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2548 }
2549 
2550 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2551 {
2552 	int ret;
2553 
2554 	if (!ctrl->effects)
2555 		ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2556 
2557 	if (!ctrl->effects)
2558 		return 0;
2559 
2560 	ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2561 			ctrl->effects, sizeof(*ctrl->effects), 0);
2562 	if (ret) {
2563 		kfree(ctrl->effects);
2564 		ctrl->effects = NULL;
2565 	}
2566 	return ret;
2567 }
2568 
2569 /*
2570  * Initialize the cached copies of the Identify data and various controller
2571  * register in our nvme_ctrl structure.  This should be called as soon as
2572  * the admin queue is fully up and running.
2573  */
2574 int nvme_init_identify(struct nvme_ctrl *ctrl)
2575 {
2576 	struct nvme_id_ctrl *id;
2577 	u64 cap;
2578 	int ret, page_shift;
2579 	u32 max_hw_sectors;
2580 	bool prev_apst_enabled;
2581 
2582 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2583 	if (ret) {
2584 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2585 		return ret;
2586 	}
2587 
2588 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2589 	if (ret) {
2590 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2591 		return ret;
2592 	}
2593 	page_shift = NVME_CAP_MPSMIN(cap) + 12;
2594 
2595 	if (ctrl->vs >= NVME_VS(1, 1, 0))
2596 		ctrl->subsystem = NVME_CAP_NSSRC(cap);
2597 
2598 	ret = nvme_identify_ctrl(ctrl, &id);
2599 	if (ret) {
2600 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2601 		return -EIO;
2602 	}
2603 
2604 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2605 		ret = nvme_get_effects_log(ctrl);
2606 		if (ret < 0)
2607 			goto out_free;
2608 	}
2609 
2610 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
2611 		ctrl->cntlid = le16_to_cpu(id->cntlid);
2612 
2613 	if (!ctrl->identified) {
2614 		int i;
2615 
2616 		ret = nvme_init_subsystem(ctrl, id);
2617 		if (ret)
2618 			goto out_free;
2619 
2620 		/*
2621 		 * Check for quirks.  Quirk can depend on firmware version,
2622 		 * so, in principle, the set of quirks present can change
2623 		 * across a reset.  As a possible future enhancement, we
2624 		 * could re-scan for quirks every time we reinitialize
2625 		 * the device, but we'd have to make sure that the driver
2626 		 * behaves intelligently if the quirks change.
2627 		 */
2628 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2629 			if (quirk_matches(id, &core_quirks[i]))
2630 				ctrl->quirks |= core_quirks[i].quirks;
2631 		}
2632 	}
2633 
2634 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2635 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2636 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2637 	}
2638 
2639 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2640 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2641 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2642 
2643 	ctrl->oacs = le16_to_cpu(id->oacs);
2644 	ctrl->oncs = le16_to_cpu(id->oncs);
2645 	ctrl->mtfa = le16_to_cpu(id->mtfa);
2646 	ctrl->oaes = le32_to_cpu(id->oaes);
2647 	atomic_set(&ctrl->abort_limit, id->acl + 1);
2648 	ctrl->vwc = id->vwc;
2649 	if (id->mdts)
2650 		max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2651 	else
2652 		max_hw_sectors = UINT_MAX;
2653 	ctrl->max_hw_sectors =
2654 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2655 
2656 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
2657 	ctrl->sgls = le32_to_cpu(id->sgls);
2658 	ctrl->kas = le16_to_cpu(id->kas);
2659 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
2660 	ctrl->ctratt = le32_to_cpu(id->ctratt);
2661 
2662 	if (id->rtd3e) {
2663 		/* us -> s */
2664 		u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2665 
2666 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2667 						 shutdown_timeout, 60);
2668 
2669 		if (ctrl->shutdown_timeout != shutdown_timeout)
2670 			dev_info(ctrl->device,
2671 				 "Shutdown timeout set to %u seconds\n",
2672 				 ctrl->shutdown_timeout);
2673 	} else
2674 		ctrl->shutdown_timeout = shutdown_timeout;
2675 
2676 	ctrl->npss = id->npss;
2677 	ctrl->apsta = id->apsta;
2678 	prev_apst_enabled = ctrl->apst_enabled;
2679 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2680 		if (force_apst && id->apsta) {
2681 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2682 			ctrl->apst_enabled = true;
2683 		} else {
2684 			ctrl->apst_enabled = false;
2685 		}
2686 	} else {
2687 		ctrl->apst_enabled = id->apsta;
2688 	}
2689 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2690 
2691 	if (ctrl->ops->flags & NVME_F_FABRICS) {
2692 		ctrl->icdoff = le16_to_cpu(id->icdoff);
2693 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2694 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2695 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2696 
2697 		/*
2698 		 * In fabrics we need to verify the cntlid matches the
2699 		 * admin connect
2700 		 */
2701 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2702 			ret = -EINVAL;
2703 			goto out_free;
2704 		}
2705 
2706 		if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2707 			dev_err(ctrl->device,
2708 				"keep-alive support is mandatory for fabrics\n");
2709 			ret = -EINVAL;
2710 			goto out_free;
2711 		}
2712 	} else {
2713 		ctrl->hmpre = le32_to_cpu(id->hmpre);
2714 		ctrl->hmmin = le32_to_cpu(id->hmmin);
2715 		ctrl->hmminds = le32_to_cpu(id->hmminds);
2716 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2717 	}
2718 
2719 	ret = nvme_mpath_init(ctrl, id);
2720 	kfree(id);
2721 
2722 	if (ret < 0)
2723 		return ret;
2724 
2725 	if (ctrl->apst_enabled && !prev_apst_enabled)
2726 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
2727 	else if (!ctrl->apst_enabled && prev_apst_enabled)
2728 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
2729 
2730 	ret = nvme_configure_apst(ctrl);
2731 	if (ret < 0)
2732 		return ret;
2733 
2734 	ret = nvme_configure_timestamp(ctrl);
2735 	if (ret < 0)
2736 		return ret;
2737 
2738 	ret = nvme_configure_directives(ctrl);
2739 	if (ret < 0)
2740 		return ret;
2741 
2742 	ret = nvme_configure_acre(ctrl);
2743 	if (ret < 0)
2744 		return ret;
2745 
2746 	ctrl->identified = true;
2747 
2748 	return 0;
2749 
2750 out_free:
2751 	kfree(id);
2752 	return ret;
2753 }
2754 EXPORT_SYMBOL_GPL(nvme_init_identify);
2755 
2756 static int nvme_dev_open(struct inode *inode, struct file *file)
2757 {
2758 	struct nvme_ctrl *ctrl =
2759 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2760 
2761 	switch (ctrl->state) {
2762 	case NVME_CTRL_LIVE:
2763 	case NVME_CTRL_ADMIN_ONLY:
2764 		break;
2765 	default:
2766 		return -EWOULDBLOCK;
2767 	}
2768 
2769 	file->private_data = ctrl;
2770 	return 0;
2771 }
2772 
2773 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2774 {
2775 	struct nvme_ns *ns;
2776 	int ret;
2777 
2778 	down_read(&ctrl->namespaces_rwsem);
2779 	if (list_empty(&ctrl->namespaces)) {
2780 		ret = -ENOTTY;
2781 		goto out_unlock;
2782 	}
2783 
2784 	ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2785 	if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2786 		dev_warn(ctrl->device,
2787 			"NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2788 		ret = -EINVAL;
2789 		goto out_unlock;
2790 	}
2791 
2792 	dev_warn(ctrl->device,
2793 		"using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2794 	kref_get(&ns->kref);
2795 	up_read(&ctrl->namespaces_rwsem);
2796 
2797 	ret = nvme_user_cmd(ctrl, ns, argp);
2798 	nvme_put_ns(ns);
2799 	return ret;
2800 
2801 out_unlock:
2802 	up_read(&ctrl->namespaces_rwsem);
2803 	return ret;
2804 }
2805 
2806 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2807 		unsigned long arg)
2808 {
2809 	struct nvme_ctrl *ctrl = file->private_data;
2810 	void __user *argp = (void __user *)arg;
2811 
2812 	switch (cmd) {
2813 	case NVME_IOCTL_ADMIN_CMD:
2814 		return nvme_user_cmd(ctrl, NULL, argp);
2815 	case NVME_IOCTL_IO_CMD:
2816 		return nvme_dev_user_cmd(ctrl, argp);
2817 	case NVME_IOCTL_RESET:
2818 		dev_warn(ctrl->device, "resetting controller\n");
2819 		return nvme_reset_ctrl_sync(ctrl);
2820 	case NVME_IOCTL_SUBSYS_RESET:
2821 		return nvme_reset_subsystem(ctrl);
2822 	case NVME_IOCTL_RESCAN:
2823 		nvme_queue_scan(ctrl);
2824 		return 0;
2825 	default:
2826 		return -ENOTTY;
2827 	}
2828 }
2829 
2830 static const struct file_operations nvme_dev_fops = {
2831 	.owner		= THIS_MODULE,
2832 	.open		= nvme_dev_open,
2833 	.unlocked_ioctl	= nvme_dev_ioctl,
2834 	.compat_ioctl	= nvme_dev_ioctl,
2835 };
2836 
2837 static ssize_t nvme_sysfs_reset(struct device *dev,
2838 				struct device_attribute *attr, const char *buf,
2839 				size_t count)
2840 {
2841 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2842 	int ret;
2843 
2844 	ret = nvme_reset_ctrl_sync(ctrl);
2845 	if (ret < 0)
2846 		return ret;
2847 	return count;
2848 }
2849 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2850 
2851 static ssize_t nvme_sysfs_rescan(struct device *dev,
2852 				struct device_attribute *attr, const char *buf,
2853 				size_t count)
2854 {
2855 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2856 
2857 	nvme_queue_scan(ctrl);
2858 	return count;
2859 }
2860 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2861 
2862 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2863 {
2864 	struct gendisk *disk = dev_to_disk(dev);
2865 
2866 	if (disk->fops == &nvme_fops)
2867 		return nvme_get_ns_from_dev(dev)->head;
2868 	else
2869 		return disk->private_data;
2870 }
2871 
2872 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2873 		char *buf)
2874 {
2875 	struct nvme_ns_head *head = dev_to_ns_head(dev);
2876 	struct nvme_ns_ids *ids = &head->ids;
2877 	struct nvme_subsystem *subsys = head->subsys;
2878 	int serial_len = sizeof(subsys->serial);
2879 	int model_len = sizeof(subsys->model);
2880 
2881 	if (!uuid_is_null(&ids->uuid))
2882 		return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2883 
2884 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2885 		return sprintf(buf, "eui.%16phN\n", ids->nguid);
2886 
2887 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2888 		return sprintf(buf, "eui.%8phN\n", ids->eui64);
2889 
2890 	while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2891 				  subsys->serial[serial_len - 1] == '\0'))
2892 		serial_len--;
2893 	while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2894 				 subsys->model[model_len - 1] == '\0'))
2895 		model_len--;
2896 
2897 	return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2898 		serial_len, subsys->serial, model_len, subsys->model,
2899 		head->ns_id);
2900 }
2901 static DEVICE_ATTR_RO(wwid);
2902 
2903 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2904 		char *buf)
2905 {
2906 	return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2907 }
2908 static DEVICE_ATTR_RO(nguid);
2909 
2910 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2911 		char *buf)
2912 {
2913 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2914 
2915 	/* For backward compatibility expose the NGUID to userspace if
2916 	 * we have no UUID set
2917 	 */
2918 	if (uuid_is_null(&ids->uuid)) {
2919 		printk_ratelimited(KERN_WARNING
2920 				   "No UUID available providing old NGUID\n");
2921 		return sprintf(buf, "%pU\n", ids->nguid);
2922 	}
2923 	return sprintf(buf, "%pU\n", &ids->uuid);
2924 }
2925 static DEVICE_ATTR_RO(uuid);
2926 
2927 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2928 		char *buf)
2929 {
2930 	return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2931 }
2932 static DEVICE_ATTR_RO(eui);
2933 
2934 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2935 		char *buf)
2936 {
2937 	return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2938 }
2939 static DEVICE_ATTR_RO(nsid);
2940 
2941 static struct attribute *nvme_ns_id_attrs[] = {
2942 	&dev_attr_wwid.attr,
2943 	&dev_attr_uuid.attr,
2944 	&dev_attr_nguid.attr,
2945 	&dev_attr_eui.attr,
2946 	&dev_attr_nsid.attr,
2947 #ifdef CONFIG_NVME_MULTIPATH
2948 	&dev_attr_ana_grpid.attr,
2949 	&dev_attr_ana_state.attr,
2950 #endif
2951 	NULL,
2952 };
2953 
2954 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2955 		struct attribute *a, int n)
2956 {
2957 	struct device *dev = container_of(kobj, struct device, kobj);
2958 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2959 
2960 	if (a == &dev_attr_uuid.attr) {
2961 		if (uuid_is_null(&ids->uuid) &&
2962 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2963 			return 0;
2964 	}
2965 	if (a == &dev_attr_nguid.attr) {
2966 		if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2967 			return 0;
2968 	}
2969 	if (a == &dev_attr_eui.attr) {
2970 		if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2971 			return 0;
2972 	}
2973 #ifdef CONFIG_NVME_MULTIPATH
2974 	if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2975 		if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2976 			return 0;
2977 		if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2978 			return 0;
2979 	}
2980 #endif
2981 	return a->mode;
2982 }
2983 
2984 static const struct attribute_group nvme_ns_id_attr_group = {
2985 	.attrs		= nvme_ns_id_attrs,
2986 	.is_visible	= nvme_ns_id_attrs_are_visible,
2987 };
2988 
2989 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2990 	&nvme_ns_id_attr_group,
2991 #ifdef CONFIG_NVM
2992 	&nvme_nvm_attr_group,
2993 #endif
2994 	NULL,
2995 };
2996 
2997 #define nvme_show_str_function(field)						\
2998 static ssize_t  field##_show(struct device *dev,				\
2999 			    struct device_attribute *attr, char *buf)		\
3000 {										\
3001         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3002         return sprintf(buf, "%.*s\n",						\
3003 		(int)sizeof(ctrl->subsys->field), ctrl->subsys->field);		\
3004 }										\
3005 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3006 
3007 nvme_show_str_function(model);
3008 nvme_show_str_function(serial);
3009 nvme_show_str_function(firmware_rev);
3010 
3011 #define nvme_show_int_function(field)						\
3012 static ssize_t  field##_show(struct device *dev,				\
3013 			    struct device_attribute *attr, char *buf)		\
3014 {										\
3015         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3016         return sprintf(buf, "%d\n", ctrl->field);	\
3017 }										\
3018 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3019 
3020 nvme_show_int_function(cntlid);
3021 nvme_show_int_function(numa_node);
3022 
3023 static ssize_t nvme_sysfs_delete(struct device *dev,
3024 				struct device_attribute *attr, const char *buf,
3025 				size_t count)
3026 {
3027 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3028 
3029 	if (device_remove_file_self(dev, attr))
3030 		nvme_delete_ctrl_sync(ctrl);
3031 	return count;
3032 }
3033 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3034 
3035 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3036 					 struct device_attribute *attr,
3037 					 char *buf)
3038 {
3039 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3040 
3041 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3042 }
3043 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3044 
3045 static ssize_t nvme_sysfs_show_state(struct device *dev,
3046 				     struct device_attribute *attr,
3047 				     char *buf)
3048 {
3049 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3050 	static const char *const state_name[] = {
3051 		[NVME_CTRL_NEW]		= "new",
3052 		[NVME_CTRL_LIVE]	= "live",
3053 		[NVME_CTRL_ADMIN_ONLY]	= "only-admin",
3054 		[NVME_CTRL_RESETTING]	= "resetting",
3055 		[NVME_CTRL_CONNECTING]	= "connecting",
3056 		[NVME_CTRL_DELETING]	= "deleting",
3057 		[NVME_CTRL_DEAD]	= "dead",
3058 	};
3059 
3060 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3061 	    state_name[ctrl->state])
3062 		return sprintf(buf, "%s\n", state_name[ctrl->state]);
3063 
3064 	return sprintf(buf, "unknown state\n");
3065 }
3066 
3067 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3068 
3069 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3070 					 struct device_attribute *attr,
3071 					 char *buf)
3072 {
3073 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3074 
3075 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3076 }
3077 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3078 
3079 static ssize_t nvme_sysfs_show_address(struct device *dev,
3080 					 struct device_attribute *attr,
3081 					 char *buf)
3082 {
3083 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3084 
3085 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3086 }
3087 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3088 
3089 static struct attribute *nvme_dev_attrs[] = {
3090 	&dev_attr_reset_controller.attr,
3091 	&dev_attr_rescan_controller.attr,
3092 	&dev_attr_model.attr,
3093 	&dev_attr_serial.attr,
3094 	&dev_attr_firmware_rev.attr,
3095 	&dev_attr_cntlid.attr,
3096 	&dev_attr_delete_controller.attr,
3097 	&dev_attr_transport.attr,
3098 	&dev_attr_subsysnqn.attr,
3099 	&dev_attr_address.attr,
3100 	&dev_attr_state.attr,
3101 	&dev_attr_numa_node.attr,
3102 	NULL
3103 };
3104 
3105 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3106 		struct attribute *a, int n)
3107 {
3108 	struct device *dev = container_of(kobj, struct device, kobj);
3109 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3110 
3111 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3112 		return 0;
3113 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3114 		return 0;
3115 
3116 	return a->mode;
3117 }
3118 
3119 static struct attribute_group nvme_dev_attrs_group = {
3120 	.attrs		= nvme_dev_attrs,
3121 	.is_visible	= nvme_dev_attrs_are_visible,
3122 };
3123 
3124 static const struct attribute_group *nvme_dev_attr_groups[] = {
3125 	&nvme_dev_attrs_group,
3126 	NULL,
3127 };
3128 
3129 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3130 		unsigned nsid)
3131 {
3132 	struct nvme_ns_head *h;
3133 
3134 	lockdep_assert_held(&subsys->lock);
3135 
3136 	list_for_each_entry(h, &subsys->nsheads, entry) {
3137 		if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3138 			return h;
3139 	}
3140 
3141 	return NULL;
3142 }
3143 
3144 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3145 		struct nvme_ns_head *new)
3146 {
3147 	struct nvme_ns_head *h;
3148 
3149 	lockdep_assert_held(&subsys->lock);
3150 
3151 	list_for_each_entry(h, &subsys->nsheads, entry) {
3152 		if (nvme_ns_ids_valid(&new->ids) &&
3153 		    !list_empty(&h->list) &&
3154 		    nvme_ns_ids_equal(&new->ids, &h->ids))
3155 			return -EINVAL;
3156 	}
3157 
3158 	return 0;
3159 }
3160 
3161 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3162 		unsigned nsid, struct nvme_id_ns *id)
3163 {
3164 	struct nvme_ns_head *head;
3165 	size_t size = sizeof(*head);
3166 	int ret = -ENOMEM;
3167 
3168 #ifdef CONFIG_NVME_MULTIPATH
3169 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3170 #endif
3171 
3172 	head = kzalloc(size, GFP_KERNEL);
3173 	if (!head)
3174 		goto out;
3175 	ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3176 	if (ret < 0)
3177 		goto out_free_head;
3178 	head->instance = ret;
3179 	INIT_LIST_HEAD(&head->list);
3180 	ret = init_srcu_struct(&head->srcu);
3181 	if (ret)
3182 		goto out_ida_remove;
3183 	head->subsys = ctrl->subsys;
3184 	head->ns_id = nsid;
3185 	kref_init(&head->ref);
3186 
3187 	nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3188 
3189 	ret = __nvme_check_ids(ctrl->subsys, head);
3190 	if (ret) {
3191 		dev_err(ctrl->device,
3192 			"duplicate IDs for nsid %d\n", nsid);
3193 		goto out_cleanup_srcu;
3194 	}
3195 
3196 	ret = nvme_mpath_alloc_disk(ctrl, head);
3197 	if (ret)
3198 		goto out_cleanup_srcu;
3199 
3200 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3201 
3202 	kref_get(&ctrl->subsys->ref);
3203 
3204 	return head;
3205 out_cleanup_srcu:
3206 	cleanup_srcu_struct(&head->srcu);
3207 out_ida_remove:
3208 	ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3209 out_free_head:
3210 	kfree(head);
3211 out:
3212 	return ERR_PTR(ret);
3213 }
3214 
3215 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3216 		struct nvme_id_ns *id)
3217 {
3218 	struct nvme_ctrl *ctrl = ns->ctrl;
3219 	bool is_shared = id->nmic & (1 << 0);
3220 	struct nvme_ns_head *head = NULL;
3221 	int ret = 0;
3222 
3223 	mutex_lock(&ctrl->subsys->lock);
3224 	if (is_shared)
3225 		head = __nvme_find_ns_head(ctrl->subsys, nsid);
3226 	if (!head) {
3227 		head = nvme_alloc_ns_head(ctrl, nsid, id);
3228 		if (IS_ERR(head)) {
3229 			ret = PTR_ERR(head);
3230 			goto out_unlock;
3231 		}
3232 	} else {
3233 		struct nvme_ns_ids ids;
3234 
3235 		nvme_report_ns_ids(ctrl, nsid, id, &ids);
3236 		if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3237 			dev_err(ctrl->device,
3238 				"IDs don't match for shared namespace %d\n",
3239 					nsid);
3240 			ret = -EINVAL;
3241 			goto out_unlock;
3242 		}
3243 	}
3244 
3245 	list_add_tail(&ns->siblings, &head->list);
3246 	ns->head = head;
3247 
3248 out_unlock:
3249 	mutex_unlock(&ctrl->subsys->lock);
3250 	return ret;
3251 }
3252 
3253 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3254 {
3255 	struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3256 	struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3257 
3258 	return nsa->head->ns_id - nsb->head->ns_id;
3259 }
3260 
3261 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3262 {
3263 	struct nvme_ns *ns, *ret = NULL;
3264 
3265 	down_read(&ctrl->namespaces_rwsem);
3266 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3267 		if (ns->head->ns_id == nsid) {
3268 			if (!kref_get_unless_zero(&ns->kref))
3269 				continue;
3270 			ret = ns;
3271 			break;
3272 		}
3273 		if (ns->head->ns_id > nsid)
3274 			break;
3275 	}
3276 	up_read(&ctrl->namespaces_rwsem);
3277 	return ret;
3278 }
3279 
3280 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3281 {
3282 	struct streams_directive_params s;
3283 	int ret;
3284 
3285 	if (!ctrl->nr_streams)
3286 		return 0;
3287 
3288 	ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3289 	if (ret)
3290 		return ret;
3291 
3292 	ns->sws = le32_to_cpu(s.sws);
3293 	ns->sgs = le16_to_cpu(s.sgs);
3294 
3295 	if (ns->sws) {
3296 		unsigned int bs = 1 << ns->lba_shift;
3297 
3298 		blk_queue_io_min(ns->queue, bs * ns->sws);
3299 		if (ns->sgs)
3300 			blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3301 	}
3302 
3303 	return 0;
3304 }
3305 
3306 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3307 {
3308 	struct nvme_ns *ns;
3309 	struct gendisk *disk;
3310 	struct nvme_id_ns *id;
3311 	char disk_name[DISK_NAME_LEN];
3312 	int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3313 
3314 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3315 	if (!ns)
3316 		return -ENOMEM;
3317 
3318 	ns->queue = blk_mq_init_queue(ctrl->tagset);
3319 	if (IS_ERR(ns->queue)) {
3320 		ret = PTR_ERR(ns->queue);
3321 		goto out_free_ns;
3322 	}
3323 
3324 	if (ctrl->opts && ctrl->opts->data_digest)
3325 		ns->queue->backing_dev_info->capabilities
3326 			|= BDI_CAP_STABLE_WRITES;
3327 
3328 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3329 	if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3330 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3331 
3332 	ns->queue->queuedata = ns;
3333 	ns->ctrl = ctrl;
3334 
3335 	kref_init(&ns->kref);
3336 	ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3337 
3338 	blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3339 	nvme_set_queue_limits(ctrl, ns->queue);
3340 
3341 	id = nvme_identify_ns(ctrl, nsid);
3342 	if (!id) {
3343 		ret = -EIO;
3344 		goto out_free_queue;
3345 	}
3346 
3347 	if (id->ncap == 0) {
3348 		ret = -EINVAL;
3349 		goto out_free_id;
3350 	}
3351 
3352 	ret = nvme_init_ns_head(ns, nsid, id);
3353 	if (ret)
3354 		goto out_free_id;
3355 	nvme_setup_streams_ns(ctrl, ns);
3356 	nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3357 
3358 	disk = alloc_disk_node(0, node);
3359 	if (!disk) {
3360 		ret = -ENOMEM;
3361 		goto out_unlink_ns;
3362 	}
3363 
3364 	disk->fops = &nvme_fops;
3365 	disk->private_data = ns;
3366 	disk->queue = ns->queue;
3367 	disk->flags = flags;
3368 	memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3369 	ns->disk = disk;
3370 
3371 	__nvme_revalidate_disk(disk, id);
3372 
3373 	if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3374 		ret = nvme_nvm_register(ns, disk_name, node);
3375 		if (ret) {
3376 			dev_warn(ctrl->device, "LightNVM init failure\n");
3377 			goto out_put_disk;
3378 		}
3379 	}
3380 
3381 	down_write(&ctrl->namespaces_rwsem);
3382 	list_add_tail(&ns->list, &ctrl->namespaces);
3383 	up_write(&ctrl->namespaces_rwsem);
3384 
3385 	nvme_get_ctrl(ctrl);
3386 
3387 	device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3388 
3389 	nvme_mpath_add_disk(ns, id);
3390 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3391 	kfree(id);
3392 
3393 	return 0;
3394  out_put_disk:
3395 	put_disk(ns->disk);
3396  out_unlink_ns:
3397 	mutex_lock(&ctrl->subsys->lock);
3398 	list_del_rcu(&ns->siblings);
3399 	mutex_unlock(&ctrl->subsys->lock);
3400 	nvme_put_ns_head(ns->head);
3401  out_free_id:
3402 	kfree(id);
3403  out_free_queue:
3404 	blk_cleanup_queue(ns->queue);
3405  out_free_ns:
3406 	kfree(ns);
3407 	return ret;
3408 }
3409 
3410 static void nvme_ns_remove(struct nvme_ns *ns)
3411 {
3412 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3413 		return;
3414 
3415 	nvme_fault_inject_fini(&ns->fault_inject);
3416 
3417 	mutex_lock(&ns->ctrl->subsys->lock);
3418 	list_del_rcu(&ns->siblings);
3419 	mutex_unlock(&ns->ctrl->subsys->lock);
3420 	synchronize_rcu(); /* guarantee not available in head->list */
3421 	nvme_mpath_clear_current_path(ns);
3422 	synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3423 
3424 	if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3425 		del_gendisk(ns->disk);
3426 		blk_cleanup_queue(ns->queue);
3427 		if (blk_get_integrity(ns->disk))
3428 			blk_integrity_unregister(ns->disk);
3429 	}
3430 
3431 	down_write(&ns->ctrl->namespaces_rwsem);
3432 	list_del_init(&ns->list);
3433 	up_write(&ns->ctrl->namespaces_rwsem);
3434 
3435 	nvme_mpath_check_last_path(ns);
3436 	nvme_put_ns(ns);
3437 }
3438 
3439 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3440 {
3441 	struct nvme_ns *ns;
3442 
3443 	ns = nvme_find_get_ns(ctrl, nsid);
3444 	if (ns) {
3445 		if (ns->disk && revalidate_disk(ns->disk))
3446 			nvme_ns_remove(ns);
3447 		nvme_put_ns(ns);
3448 	} else
3449 		nvme_alloc_ns(ctrl, nsid);
3450 }
3451 
3452 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3453 					unsigned nsid)
3454 {
3455 	struct nvme_ns *ns, *next;
3456 	LIST_HEAD(rm_list);
3457 
3458 	down_write(&ctrl->namespaces_rwsem);
3459 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3460 		if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3461 			list_move_tail(&ns->list, &rm_list);
3462 	}
3463 	up_write(&ctrl->namespaces_rwsem);
3464 
3465 	list_for_each_entry_safe(ns, next, &rm_list, list)
3466 		nvme_ns_remove(ns);
3467 
3468 }
3469 
3470 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3471 {
3472 	struct nvme_ns *ns;
3473 	__le32 *ns_list;
3474 	unsigned i, j, nsid, prev = 0;
3475 	unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3476 	int ret = 0;
3477 
3478 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3479 	if (!ns_list)
3480 		return -ENOMEM;
3481 
3482 	for (i = 0; i < num_lists; i++) {
3483 		ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3484 		if (ret)
3485 			goto free;
3486 
3487 		for (j = 0; j < min(nn, 1024U); j++) {
3488 			nsid = le32_to_cpu(ns_list[j]);
3489 			if (!nsid)
3490 				goto out;
3491 
3492 			nvme_validate_ns(ctrl, nsid);
3493 
3494 			while (++prev < nsid) {
3495 				ns = nvme_find_get_ns(ctrl, prev);
3496 				if (ns) {
3497 					nvme_ns_remove(ns);
3498 					nvme_put_ns(ns);
3499 				}
3500 			}
3501 		}
3502 		nn -= j;
3503 	}
3504  out:
3505 	nvme_remove_invalid_namespaces(ctrl, prev);
3506  free:
3507 	kfree(ns_list);
3508 	return ret;
3509 }
3510 
3511 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3512 {
3513 	unsigned i;
3514 
3515 	for (i = 1; i <= nn; i++)
3516 		nvme_validate_ns(ctrl, i);
3517 
3518 	nvme_remove_invalid_namespaces(ctrl, nn);
3519 }
3520 
3521 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3522 {
3523 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3524 	__le32 *log;
3525 	int error;
3526 
3527 	log = kzalloc(log_size, GFP_KERNEL);
3528 	if (!log)
3529 		return;
3530 
3531 	/*
3532 	 * We need to read the log to clear the AEN, but we don't want to rely
3533 	 * on it for the changed namespace information as userspace could have
3534 	 * raced with us in reading the log page, which could cause us to miss
3535 	 * updates.
3536 	 */
3537 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3538 			log_size, 0);
3539 	if (error)
3540 		dev_warn(ctrl->device,
3541 			"reading changed ns log failed: %d\n", error);
3542 
3543 	kfree(log);
3544 }
3545 
3546 static void nvme_scan_work(struct work_struct *work)
3547 {
3548 	struct nvme_ctrl *ctrl =
3549 		container_of(work, struct nvme_ctrl, scan_work);
3550 	struct nvme_id_ctrl *id;
3551 	unsigned nn;
3552 
3553 	if (ctrl->state != NVME_CTRL_LIVE)
3554 		return;
3555 
3556 	WARN_ON_ONCE(!ctrl->tagset);
3557 
3558 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3559 		dev_info(ctrl->device, "rescanning namespaces.\n");
3560 		nvme_clear_changed_ns_log(ctrl);
3561 	}
3562 
3563 	if (nvme_identify_ctrl(ctrl, &id))
3564 		return;
3565 
3566 	mutex_lock(&ctrl->scan_lock);
3567 	nn = le32_to_cpu(id->nn);
3568 	if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3569 	    !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3570 		if (!nvme_scan_ns_list(ctrl, nn))
3571 			goto out_free_id;
3572 	}
3573 	nvme_scan_ns_sequential(ctrl, nn);
3574 out_free_id:
3575 	mutex_unlock(&ctrl->scan_lock);
3576 	kfree(id);
3577 	down_write(&ctrl->namespaces_rwsem);
3578 	list_sort(NULL, &ctrl->namespaces, ns_cmp);
3579 	up_write(&ctrl->namespaces_rwsem);
3580 }
3581 
3582 /*
3583  * This function iterates the namespace list unlocked to allow recovery from
3584  * controller failure. It is up to the caller to ensure the namespace list is
3585  * not modified by scan work while this function is executing.
3586  */
3587 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3588 {
3589 	struct nvme_ns *ns, *next;
3590 	LIST_HEAD(ns_list);
3591 
3592 	/*
3593 	 * make sure to requeue I/O to all namespaces as these
3594 	 * might result from the scan itself and must complete
3595 	 * for the scan_work to make progress
3596 	 */
3597 	nvme_mpath_clear_ctrl_paths(ctrl);
3598 
3599 	/* prevent racing with ns scanning */
3600 	flush_work(&ctrl->scan_work);
3601 
3602 	/*
3603 	 * The dead states indicates the controller was not gracefully
3604 	 * disconnected. In that case, we won't be able to flush any data while
3605 	 * removing the namespaces' disks; fail all the queues now to avoid
3606 	 * potentially having to clean up the failed sync later.
3607 	 */
3608 	if (ctrl->state == NVME_CTRL_DEAD)
3609 		nvme_kill_queues(ctrl);
3610 
3611 	down_write(&ctrl->namespaces_rwsem);
3612 	list_splice_init(&ctrl->namespaces, &ns_list);
3613 	up_write(&ctrl->namespaces_rwsem);
3614 
3615 	list_for_each_entry_safe(ns, next, &ns_list, list)
3616 		nvme_ns_remove(ns);
3617 }
3618 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3619 
3620 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3621 {
3622 	char *envp[2] = { NULL, NULL };
3623 	u32 aen_result = ctrl->aen_result;
3624 
3625 	ctrl->aen_result = 0;
3626 	if (!aen_result)
3627 		return;
3628 
3629 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3630 	if (!envp[0])
3631 		return;
3632 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3633 	kfree(envp[0]);
3634 }
3635 
3636 static void nvme_async_event_work(struct work_struct *work)
3637 {
3638 	struct nvme_ctrl *ctrl =
3639 		container_of(work, struct nvme_ctrl, async_event_work);
3640 
3641 	nvme_aen_uevent(ctrl);
3642 	ctrl->ops->submit_async_event(ctrl);
3643 }
3644 
3645 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3646 {
3647 
3648 	u32 csts;
3649 
3650 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3651 		return false;
3652 
3653 	if (csts == ~0)
3654 		return false;
3655 
3656 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3657 }
3658 
3659 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3660 {
3661 	struct nvme_fw_slot_info_log *log;
3662 
3663 	log = kmalloc(sizeof(*log), GFP_KERNEL);
3664 	if (!log)
3665 		return;
3666 
3667 	if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3668 			sizeof(*log), 0))
3669 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3670 	kfree(log);
3671 }
3672 
3673 static void nvme_fw_act_work(struct work_struct *work)
3674 {
3675 	struct nvme_ctrl *ctrl = container_of(work,
3676 				struct nvme_ctrl, fw_act_work);
3677 	unsigned long fw_act_timeout;
3678 
3679 	if (ctrl->mtfa)
3680 		fw_act_timeout = jiffies +
3681 				msecs_to_jiffies(ctrl->mtfa * 100);
3682 	else
3683 		fw_act_timeout = jiffies +
3684 				msecs_to_jiffies(admin_timeout * 1000);
3685 
3686 	nvme_stop_queues(ctrl);
3687 	while (nvme_ctrl_pp_status(ctrl)) {
3688 		if (time_after(jiffies, fw_act_timeout)) {
3689 			dev_warn(ctrl->device,
3690 				"Fw activation timeout, reset controller\n");
3691 			nvme_reset_ctrl(ctrl);
3692 			break;
3693 		}
3694 		msleep(100);
3695 	}
3696 
3697 	if (ctrl->state != NVME_CTRL_LIVE)
3698 		return;
3699 
3700 	nvme_start_queues(ctrl);
3701 	/* read FW slot information to clear the AER */
3702 	nvme_get_fw_slot_info(ctrl);
3703 }
3704 
3705 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3706 {
3707 	u32 aer_notice_type = (result & 0xff00) >> 8;
3708 
3709 	trace_nvme_async_event(ctrl, aer_notice_type);
3710 
3711 	switch (aer_notice_type) {
3712 	case NVME_AER_NOTICE_NS_CHANGED:
3713 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3714 		nvme_queue_scan(ctrl);
3715 		break;
3716 	case NVME_AER_NOTICE_FW_ACT_STARTING:
3717 		queue_work(nvme_wq, &ctrl->fw_act_work);
3718 		break;
3719 #ifdef CONFIG_NVME_MULTIPATH
3720 	case NVME_AER_NOTICE_ANA:
3721 		if (!ctrl->ana_log_buf)
3722 			break;
3723 		queue_work(nvme_wq, &ctrl->ana_work);
3724 		break;
3725 #endif
3726 	default:
3727 		dev_warn(ctrl->device, "async event result %08x\n", result);
3728 	}
3729 }
3730 
3731 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3732 		volatile union nvme_result *res)
3733 {
3734 	u32 result = le32_to_cpu(res->u32);
3735 	u32 aer_type = result & 0x07;
3736 
3737 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3738 		return;
3739 
3740 	switch (aer_type) {
3741 	case NVME_AER_NOTICE:
3742 		nvme_handle_aen_notice(ctrl, result);
3743 		break;
3744 	case NVME_AER_ERROR:
3745 	case NVME_AER_SMART:
3746 	case NVME_AER_CSS:
3747 	case NVME_AER_VS:
3748 		trace_nvme_async_event(ctrl, aer_type);
3749 		ctrl->aen_result = result;
3750 		break;
3751 	default:
3752 		break;
3753 	}
3754 	queue_work(nvme_wq, &ctrl->async_event_work);
3755 }
3756 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3757 
3758 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3759 {
3760 	nvme_mpath_stop(ctrl);
3761 	nvme_stop_keep_alive(ctrl);
3762 	flush_work(&ctrl->async_event_work);
3763 	cancel_work_sync(&ctrl->fw_act_work);
3764 }
3765 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3766 
3767 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3768 {
3769 	if (ctrl->kato)
3770 		nvme_start_keep_alive(ctrl);
3771 
3772 	if (ctrl->queue_count > 1) {
3773 		nvme_queue_scan(ctrl);
3774 		nvme_enable_aen(ctrl);
3775 		queue_work(nvme_wq, &ctrl->async_event_work);
3776 		nvme_start_queues(ctrl);
3777 	}
3778 }
3779 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3780 
3781 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3782 {
3783 	nvme_fault_inject_fini(&ctrl->fault_inject);
3784 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
3785 	cdev_device_del(&ctrl->cdev, ctrl->device);
3786 }
3787 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3788 
3789 static void nvme_free_ctrl(struct device *dev)
3790 {
3791 	struct nvme_ctrl *ctrl =
3792 		container_of(dev, struct nvme_ctrl, ctrl_device);
3793 	struct nvme_subsystem *subsys = ctrl->subsys;
3794 
3795 	ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3796 	kfree(ctrl->effects);
3797 	nvme_mpath_uninit(ctrl);
3798 	__free_page(ctrl->discard_page);
3799 
3800 	if (subsys) {
3801 		mutex_lock(&nvme_subsystems_lock);
3802 		list_del(&ctrl->subsys_entry);
3803 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3804 		mutex_unlock(&nvme_subsystems_lock);
3805 	}
3806 
3807 	ctrl->ops->free_ctrl(ctrl);
3808 
3809 	if (subsys)
3810 		nvme_put_subsystem(subsys);
3811 }
3812 
3813 /*
3814  * Initialize a NVMe controller structures.  This needs to be called during
3815  * earliest initialization so that we have the initialized structured around
3816  * during probing.
3817  */
3818 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3819 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
3820 {
3821 	int ret;
3822 
3823 	ctrl->state = NVME_CTRL_NEW;
3824 	spin_lock_init(&ctrl->lock);
3825 	mutex_init(&ctrl->scan_lock);
3826 	INIT_LIST_HEAD(&ctrl->namespaces);
3827 	init_rwsem(&ctrl->namespaces_rwsem);
3828 	ctrl->dev = dev;
3829 	ctrl->ops = ops;
3830 	ctrl->quirks = quirks;
3831 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3832 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3833 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3834 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3835 
3836 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3837 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3838 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3839 
3840 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3841 			PAGE_SIZE);
3842 	ctrl->discard_page = alloc_page(GFP_KERNEL);
3843 	if (!ctrl->discard_page) {
3844 		ret = -ENOMEM;
3845 		goto out;
3846 	}
3847 
3848 	ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3849 	if (ret < 0)
3850 		goto out;
3851 	ctrl->instance = ret;
3852 
3853 	device_initialize(&ctrl->ctrl_device);
3854 	ctrl->device = &ctrl->ctrl_device;
3855 	ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3856 	ctrl->device->class = nvme_class;
3857 	ctrl->device->parent = ctrl->dev;
3858 	ctrl->device->groups = nvme_dev_attr_groups;
3859 	ctrl->device->release = nvme_free_ctrl;
3860 	dev_set_drvdata(ctrl->device, ctrl);
3861 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3862 	if (ret)
3863 		goto out_release_instance;
3864 
3865 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
3866 	ctrl->cdev.owner = ops->module;
3867 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3868 	if (ret)
3869 		goto out_free_name;
3870 
3871 	/*
3872 	 * Initialize latency tolerance controls.  The sysfs files won't
3873 	 * be visible to userspace unless the device actually supports APST.
3874 	 */
3875 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3876 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3877 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3878 
3879 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
3880 
3881 	return 0;
3882 out_free_name:
3883 	kfree_const(ctrl->device->kobj.name);
3884 out_release_instance:
3885 	ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3886 out:
3887 	if (ctrl->discard_page)
3888 		__free_page(ctrl->discard_page);
3889 	return ret;
3890 }
3891 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3892 
3893 /**
3894  * nvme_kill_queues(): Ends all namespace queues
3895  * @ctrl: the dead controller that needs to end
3896  *
3897  * Call this function when the driver determines it is unable to get the
3898  * controller in a state capable of servicing IO.
3899  */
3900 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3901 {
3902 	struct nvme_ns *ns;
3903 
3904 	down_read(&ctrl->namespaces_rwsem);
3905 
3906 	/* Forcibly unquiesce queues to avoid blocking dispatch */
3907 	if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3908 		blk_mq_unquiesce_queue(ctrl->admin_q);
3909 
3910 	list_for_each_entry(ns, &ctrl->namespaces, list)
3911 		nvme_set_queue_dying(ns);
3912 
3913 	up_read(&ctrl->namespaces_rwsem);
3914 }
3915 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3916 
3917 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3918 {
3919 	struct nvme_ns *ns;
3920 
3921 	down_read(&ctrl->namespaces_rwsem);
3922 	list_for_each_entry(ns, &ctrl->namespaces, list)
3923 		blk_mq_unfreeze_queue(ns->queue);
3924 	up_read(&ctrl->namespaces_rwsem);
3925 }
3926 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3927 
3928 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3929 {
3930 	struct nvme_ns *ns;
3931 
3932 	down_read(&ctrl->namespaces_rwsem);
3933 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3934 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3935 		if (timeout <= 0)
3936 			break;
3937 	}
3938 	up_read(&ctrl->namespaces_rwsem);
3939 }
3940 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3941 
3942 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3943 {
3944 	struct nvme_ns *ns;
3945 
3946 	down_read(&ctrl->namespaces_rwsem);
3947 	list_for_each_entry(ns, &ctrl->namespaces, list)
3948 		blk_mq_freeze_queue_wait(ns->queue);
3949 	up_read(&ctrl->namespaces_rwsem);
3950 }
3951 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3952 
3953 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3954 {
3955 	struct nvme_ns *ns;
3956 
3957 	down_read(&ctrl->namespaces_rwsem);
3958 	list_for_each_entry(ns, &ctrl->namespaces, list)
3959 		blk_freeze_queue_start(ns->queue);
3960 	up_read(&ctrl->namespaces_rwsem);
3961 }
3962 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3963 
3964 void nvme_stop_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_mq_quiesce_queue(ns->queue);
3971 	up_read(&ctrl->namespaces_rwsem);
3972 }
3973 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3974 
3975 void nvme_start_queues(struct nvme_ctrl *ctrl)
3976 {
3977 	struct nvme_ns *ns;
3978 
3979 	down_read(&ctrl->namespaces_rwsem);
3980 	list_for_each_entry(ns, &ctrl->namespaces, list)
3981 		blk_mq_unquiesce_queue(ns->queue);
3982 	up_read(&ctrl->namespaces_rwsem);
3983 }
3984 EXPORT_SYMBOL_GPL(nvme_start_queues);
3985 
3986 
3987 void nvme_sync_queues(struct nvme_ctrl *ctrl)
3988 {
3989 	struct nvme_ns *ns;
3990 
3991 	down_read(&ctrl->namespaces_rwsem);
3992 	list_for_each_entry(ns, &ctrl->namespaces, list)
3993 		blk_sync_queue(ns->queue);
3994 	up_read(&ctrl->namespaces_rwsem);
3995 }
3996 EXPORT_SYMBOL_GPL(nvme_sync_queues);
3997 
3998 /*
3999  * Check we didn't inadvertently grow the command structure sizes:
4000  */
4001 static inline void _nvme_check_size(void)
4002 {
4003 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4004 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4005 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4006 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4007 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4008 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4009 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4010 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4011 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4012 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4013 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4014 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4015 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4016 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4017 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4018 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4019 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4020 }
4021 
4022 
4023 static int __init nvme_core_init(void)
4024 {
4025 	int result = -ENOMEM;
4026 
4027 	_nvme_check_size();
4028 
4029 	nvme_wq = alloc_workqueue("nvme-wq",
4030 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4031 	if (!nvme_wq)
4032 		goto out;
4033 
4034 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4035 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4036 	if (!nvme_reset_wq)
4037 		goto destroy_wq;
4038 
4039 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4040 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4041 	if (!nvme_delete_wq)
4042 		goto destroy_reset_wq;
4043 
4044 	result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4045 	if (result < 0)
4046 		goto destroy_delete_wq;
4047 
4048 	nvme_class = class_create(THIS_MODULE, "nvme");
4049 	if (IS_ERR(nvme_class)) {
4050 		result = PTR_ERR(nvme_class);
4051 		goto unregister_chrdev;
4052 	}
4053 
4054 	nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4055 	if (IS_ERR(nvme_subsys_class)) {
4056 		result = PTR_ERR(nvme_subsys_class);
4057 		goto destroy_class;
4058 	}
4059 	return 0;
4060 
4061 destroy_class:
4062 	class_destroy(nvme_class);
4063 unregister_chrdev:
4064 	unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4065 destroy_delete_wq:
4066 	destroy_workqueue(nvme_delete_wq);
4067 destroy_reset_wq:
4068 	destroy_workqueue(nvme_reset_wq);
4069 destroy_wq:
4070 	destroy_workqueue(nvme_wq);
4071 out:
4072 	return result;
4073 }
4074 
4075 static void __exit nvme_core_exit(void)
4076 {
4077 	ida_destroy(&nvme_subsystems_ida);
4078 	class_destroy(nvme_subsys_class);
4079 	class_destroy(nvme_class);
4080 	unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4081 	destroy_workqueue(nvme_delete_wq);
4082 	destroy_workqueue(nvme_reset_wq);
4083 	destroy_workqueue(nvme_wq);
4084 }
4085 
4086 MODULE_LICENSE("GPL");
4087 MODULE_VERSION("1.0");
4088 module_init(nvme_core_init);
4089 module_exit(nvme_core_exit);
4090