xref: /openbmc/linux/drivers/nvme/host/core.c (revision 5cd17f0e)
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/blk-integrity.h>
10 #include <linux/compat.h>
11 #include <linux/delay.h>
12 #include <linux/errno.h>
13 #include <linux/hdreg.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
19 #include <linux/pr.h>
20 #include <linux/ptrace.h>
21 #include <linux/nvme_ioctl.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
24 
25 #include "nvme.h"
26 #include "fabrics.h"
27 #include <linux/nvme-auth.h>
28 
29 #define CREATE_TRACE_POINTS
30 #include "trace.h"
31 
32 #define NVME_MINORS		(1U << MINORBITS)
33 
34 struct nvme_ns_info {
35 	struct nvme_ns_ids ids;
36 	u32 nsid;
37 	__le32 anagrpid;
38 	bool is_shared;
39 	bool is_readonly;
40 	bool is_ready;
41 	bool is_removed;
42 };
43 
44 unsigned int admin_timeout = 60;
45 module_param(admin_timeout, uint, 0644);
46 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
47 EXPORT_SYMBOL_GPL(admin_timeout);
48 
49 unsigned int nvme_io_timeout = 30;
50 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
51 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
52 EXPORT_SYMBOL_GPL(nvme_io_timeout);
53 
54 static unsigned char shutdown_timeout = 5;
55 module_param(shutdown_timeout, byte, 0644);
56 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
57 
58 static u8 nvme_max_retries = 5;
59 module_param_named(max_retries, nvme_max_retries, byte, 0644);
60 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
61 
62 static unsigned long default_ps_max_latency_us = 100000;
63 module_param(default_ps_max_latency_us, ulong, 0644);
64 MODULE_PARM_DESC(default_ps_max_latency_us,
65 		 "max power saving latency for new devices; use PM QOS to change per device");
66 
67 static bool force_apst;
68 module_param(force_apst, bool, 0644);
69 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
70 
71 static unsigned long apst_primary_timeout_ms = 100;
72 module_param(apst_primary_timeout_ms, ulong, 0644);
73 MODULE_PARM_DESC(apst_primary_timeout_ms,
74 	"primary APST timeout in ms");
75 
76 static unsigned long apst_secondary_timeout_ms = 2000;
77 module_param(apst_secondary_timeout_ms, ulong, 0644);
78 MODULE_PARM_DESC(apst_secondary_timeout_ms,
79 	"secondary APST timeout in ms");
80 
81 static unsigned long apst_primary_latency_tol_us = 15000;
82 module_param(apst_primary_latency_tol_us, ulong, 0644);
83 MODULE_PARM_DESC(apst_primary_latency_tol_us,
84 	"primary APST latency tolerance in us");
85 
86 static unsigned long apst_secondary_latency_tol_us = 100000;
87 module_param(apst_secondary_latency_tol_us, ulong, 0644);
88 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
89 	"secondary APST latency tolerance in us");
90 
91 /*
92  * nvme_wq - hosts nvme related works that are not reset or delete
93  * nvme_reset_wq - hosts nvme reset works
94  * nvme_delete_wq - hosts nvme delete works
95  *
96  * nvme_wq will host works such as scan, aen handling, fw activation,
97  * keep-alive, periodic reconnects etc. nvme_reset_wq
98  * runs reset works which also flush works hosted on nvme_wq for
99  * serialization purposes. nvme_delete_wq host controller deletion
100  * works which flush reset works for serialization.
101  */
102 struct workqueue_struct *nvme_wq;
103 EXPORT_SYMBOL_GPL(nvme_wq);
104 
105 struct workqueue_struct *nvme_reset_wq;
106 EXPORT_SYMBOL_GPL(nvme_reset_wq);
107 
108 struct workqueue_struct *nvme_delete_wq;
109 EXPORT_SYMBOL_GPL(nvme_delete_wq);
110 
111 static LIST_HEAD(nvme_subsystems);
112 static DEFINE_MUTEX(nvme_subsystems_lock);
113 
114 static DEFINE_IDA(nvme_instance_ida);
115 static dev_t nvme_ctrl_base_chr_devt;
116 static struct class *nvme_class;
117 static struct class *nvme_subsys_class;
118 
119 static DEFINE_IDA(nvme_ns_chr_minor_ida);
120 static dev_t nvme_ns_chr_devt;
121 static struct class *nvme_ns_chr_class;
122 
123 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
124 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
125 					   unsigned nsid);
126 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
127 				   struct nvme_command *cmd);
128 
129 void nvme_queue_scan(struct nvme_ctrl *ctrl)
130 {
131 	/*
132 	 * Only new queue scan work when admin and IO queues are both alive
133 	 */
134 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE && ctrl->tagset)
135 		queue_work(nvme_wq, &ctrl->scan_work);
136 }
137 
138 /*
139  * Use this function to proceed with scheduling reset_work for a controller
140  * that had previously been set to the resetting state. This is intended for
141  * code paths that can't be interrupted by other reset attempts. A hot removal
142  * may prevent this from succeeding.
143  */
144 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
145 {
146 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_RESETTING)
147 		return -EBUSY;
148 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
149 		return -EBUSY;
150 	return 0;
151 }
152 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
153 
154 static void nvme_failfast_work(struct work_struct *work)
155 {
156 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
157 			struct nvme_ctrl, failfast_work);
158 
159 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_CONNECTING)
160 		return;
161 
162 	set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
163 	dev_info(ctrl->device, "failfast expired\n");
164 	nvme_kick_requeue_lists(ctrl);
165 }
166 
167 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
168 {
169 	if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
170 		return;
171 
172 	schedule_delayed_work(&ctrl->failfast_work,
173 			      ctrl->opts->fast_io_fail_tmo * HZ);
174 }
175 
176 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
177 {
178 	if (!ctrl->opts)
179 		return;
180 
181 	cancel_delayed_work_sync(&ctrl->failfast_work);
182 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
183 }
184 
185 
186 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
187 {
188 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
189 		return -EBUSY;
190 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
191 		return -EBUSY;
192 	return 0;
193 }
194 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
195 
196 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
197 {
198 	int ret;
199 
200 	ret = nvme_reset_ctrl(ctrl);
201 	if (!ret) {
202 		flush_work(&ctrl->reset_work);
203 		if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE)
204 			ret = -ENETRESET;
205 	}
206 
207 	return ret;
208 }
209 
210 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
211 {
212 	dev_info(ctrl->device,
213 		 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
214 
215 	flush_work(&ctrl->reset_work);
216 	nvme_stop_ctrl(ctrl);
217 	nvme_remove_namespaces(ctrl);
218 	ctrl->ops->delete_ctrl(ctrl);
219 	nvme_uninit_ctrl(ctrl);
220 }
221 
222 static void nvme_delete_ctrl_work(struct work_struct *work)
223 {
224 	struct nvme_ctrl *ctrl =
225 		container_of(work, struct nvme_ctrl, delete_work);
226 
227 	nvme_do_delete_ctrl(ctrl);
228 }
229 
230 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
231 {
232 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
233 		return -EBUSY;
234 	if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
235 		return -EBUSY;
236 	return 0;
237 }
238 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
239 
240 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
241 {
242 	/*
243 	 * Keep a reference until nvme_do_delete_ctrl() complete,
244 	 * since ->delete_ctrl can free the controller.
245 	 */
246 	nvme_get_ctrl(ctrl);
247 	if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
248 		nvme_do_delete_ctrl(ctrl);
249 	nvme_put_ctrl(ctrl);
250 }
251 
252 static blk_status_t nvme_error_status(u16 status)
253 {
254 	switch (status & 0x7ff) {
255 	case NVME_SC_SUCCESS:
256 		return BLK_STS_OK;
257 	case NVME_SC_CAP_EXCEEDED:
258 		return BLK_STS_NOSPC;
259 	case NVME_SC_LBA_RANGE:
260 	case NVME_SC_CMD_INTERRUPTED:
261 	case NVME_SC_NS_NOT_READY:
262 		return BLK_STS_TARGET;
263 	case NVME_SC_BAD_ATTRIBUTES:
264 	case NVME_SC_ONCS_NOT_SUPPORTED:
265 	case NVME_SC_INVALID_OPCODE:
266 	case NVME_SC_INVALID_FIELD:
267 	case NVME_SC_INVALID_NS:
268 		return BLK_STS_NOTSUPP;
269 	case NVME_SC_WRITE_FAULT:
270 	case NVME_SC_READ_ERROR:
271 	case NVME_SC_UNWRITTEN_BLOCK:
272 	case NVME_SC_ACCESS_DENIED:
273 	case NVME_SC_READ_ONLY:
274 	case NVME_SC_COMPARE_FAILED:
275 		return BLK_STS_MEDIUM;
276 	case NVME_SC_GUARD_CHECK:
277 	case NVME_SC_APPTAG_CHECK:
278 	case NVME_SC_REFTAG_CHECK:
279 	case NVME_SC_INVALID_PI:
280 		return BLK_STS_PROTECTION;
281 	case NVME_SC_RESERVATION_CONFLICT:
282 		return BLK_STS_RESV_CONFLICT;
283 	case NVME_SC_HOST_PATH_ERROR:
284 		return BLK_STS_TRANSPORT;
285 	case NVME_SC_ZONE_TOO_MANY_ACTIVE:
286 		return BLK_STS_ZONE_ACTIVE_RESOURCE;
287 	case NVME_SC_ZONE_TOO_MANY_OPEN:
288 		return BLK_STS_ZONE_OPEN_RESOURCE;
289 	default:
290 		return BLK_STS_IOERR;
291 	}
292 }
293 
294 static void nvme_retry_req(struct request *req)
295 {
296 	unsigned long delay = 0;
297 	u16 crd;
298 
299 	/* The mask and shift result must be <= 3 */
300 	crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
301 	if (crd)
302 		delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
303 
304 	nvme_req(req)->retries++;
305 	blk_mq_requeue_request(req, false);
306 	blk_mq_delay_kick_requeue_list(req->q, delay);
307 }
308 
309 static void nvme_log_error(struct request *req)
310 {
311 	struct nvme_ns *ns = req->q->queuedata;
312 	struct nvme_request *nr = nvme_req(req);
313 
314 	if (ns) {
315 		pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %llu blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
316 		       ns->disk ? ns->disk->disk_name : "?",
317 		       nvme_get_opcode_str(nr->cmd->common.opcode),
318 		       nr->cmd->common.opcode,
319 		       (unsigned long long)nvme_sect_to_lba(ns, blk_rq_pos(req)),
320 		       (unsigned long long)blk_rq_bytes(req) >> ns->lba_shift,
321 		       nvme_get_error_status_str(nr->status),
322 		       nr->status >> 8 & 7,	/* Status Code Type */
323 		       nr->status & 0xff,	/* Status Code */
324 		       nr->status & NVME_SC_MORE ? "MORE " : "",
325 		       nr->status & NVME_SC_DNR  ? "DNR "  : "");
326 		return;
327 	}
328 
329 	pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
330 			   dev_name(nr->ctrl->device),
331 			   nvme_get_admin_opcode_str(nr->cmd->common.opcode),
332 			   nr->cmd->common.opcode,
333 			   nvme_get_error_status_str(nr->status),
334 			   nr->status >> 8 & 7,	/* Status Code Type */
335 			   nr->status & 0xff,	/* Status Code */
336 			   nr->status & NVME_SC_MORE ? "MORE " : "",
337 			   nr->status & NVME_SC_DNR  ? "DNR "  : "");
338 }
339 
340 enum nvme_disposition {
341 	COMPLETE,
342 	RETRY,
343 	FAILOVER,
344 	AUTHENTICATE,
345 };
346 
347 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
348 {
349 	if (likely(nvme_req(req)->status == 0))
350 		return COMPLETE;
351 
352 	if ((nvme_req(req)->status & 0x7ff) == NVME_SC_AUTH_REQUIRED)
353 		return AUTHENTICATE;
354 
355 	if (blk_noretry_request(req) ||
356 	    (nvme_req(req)->status & NVME_SC_DNR) ||
357 	    nvme_req(req)->retries >= nvme_max_retries)
358 		return COMPLETE;
359 
360 	if (req->cmd_flags & REQ_NVME_MPATH) {
361 		if (nvme_is_path_error(nvme_req(req)->status) ||
362 		    blk_queue_dying(req->q))
363 			return FAILOVER;
364 	} else {
365 		if (blk_queue_dying(req->q))
366 			return COMPLETE;
367 	}
368 
369 	return RETRY;
370 }
371 
372 static inline void nvme_end_req_zoned(struct request *req)
373 {
374 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
375 	    req_op(req) == REQ_OP_ZONE_APPEND)
376 		req->__sector = nvme_lba_to_sect(req->q->queuedata,
377 			le64_to_cpu(nvme_req(req)->result.u64));
378 }
379 
380 static inline void nvme_end_req(struct request *req)
381 {
382 	blk_status_t status = nvme_error_status(nvme_req(req)->status);
383 
384 	if (unlikely(nvme_req(req)->status && !(req->rq_flags & RQF_QUIET)))
385 		nvme_log_error(req);
386 	nvme_end_req_zoned(req);
387 	nvme_trace_bio_complete(req);
388 	if (req->cmd_flags & REQ_NVME_MPATH)
389 		nvme_mpath_end_request(req);
390 	blk_mq_end_request(req, status);
391 }
392 
393 void nvme_complete_rq(struct request *req)
394 {
395 	struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
396 
397 	trace_nvme_complete_rq(req);
398 	nvme_cleanup_cmd(req);
399 
400 	/*
401 	 * Completions of long-running commands should not be able to
402 	 * defer sending of periodic keep alives, since the controller
403 	 * may have completed processing such commands a long time ago
404 	 * (arbitrarily close to command submission time).
405 	 * req->deadline - req->timeout is the command submission time
406 	 * in jiffies.
407 	 */
408 	if (ctrl->kas &&
409 	    req->deadline - req->timeout >= ctrl->ka_last_check_time)
410 		ctrl->comp_seen = true;
411 
412 	switch (nvme_decide_disposition(req)) {
413 	case COMPLETE:
414 		nvme_end_req(req);
415 		return;
416 	case RETRY:
417 		nvme_retry_req(req);
418 		return;
419 	case FAILOVER:
420 		nvme_failover_req(req);
421 		return;
422 	case AUTHENTICATE:
423 #ifdef CONFIG_NVME_AUTH
424 		queue_work(nvme_wq, &ctrl->dhchap_auth_work);
425 		nvme_retry_req(req);
426 #else
427 		nvme_end_req(req);
428 #endif
429 		return;
430 	}
431 }
432 EXPORT_SYMBOL_GPL(nvme_complete_rq);
433 
434 void nvme_complete_batch_req(struct request *req)
435 {
436 	trace_nvme_complete_rq(req);
437 	nvme_cleanup_cmd(req);
438 	nvme_end_req_zoned(req);
439 }
440 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
441 
442 /*
443  * Called to unwind from ->queue_rq on a failed command submission so that the
444  * multipathing code gets called to potentially failover to another path.
445  * The caller needs to unwind all transport specific resource allocations and
446  * must return propagate the return value.
447  */
448 blk_status_t nvme_host_path_error(struct request *req)
449 {
450 	nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
451 	blk_mq_set_request_complete(req);
452 	nvme_complete_rq(req);
453 	return BLK_STS_OK;
454 }
455 EXPORT_SYMBOL_GPL(nvme_host_path_error);
456 
457 bool nvme_cancel_request(struct request *req, void *data)
458 {
459 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
460 				"Cancelling I/O %d", req->tag);
461 
462 	/* don't abort one completed or idle request */
463 	if (blk_mq_rq_state(req) != MQ_RQ_IN_FLIGHT)
464 		return true;
465 
466 	nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
467 	nvme_req(req)->flags |= NVME_REQ_CANCELLED;
468 	blk_mq_complete_request(req);
469 	return true;
470 }
471 EXPORT_SYMBOL_GPL(nvme_cancel_request);
472 
473 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
474 {
475 	if (ctrl->tagset) {
476 		blk_mq_tagset_busy_iter(ctrl->tagset,
477 				nvme_cancel_request, ctrl);
478 		blk_mq_tagset_wait_completed_request(ctrl->tagset);
479 	}
480 }
481 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
482 
483 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
484 {
485 	if (ctrl->admin_tagset) {
486 		blk_mq_tagset_busy_iter(ctrl->admin_tagset,
487 				nvme_cancel_request, ctrl);
488 		blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
489 	}
490 }
491 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
492 
493 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
494 		enum nvme_ctrl_state new_state)
495 {
496 	enum nvme_ctrl_state old_state;
497 	unsigned long flags;
498 	bool changed = false;
499 
500 	spin_lock_irqsave(&ctrl->lock, flags);
501 
502 	old_state = nvme_ctrl_state(ctrl);
503 	switch (new_state) {
504 	case NVME_CTRL_LIVE:
505 		switch (old_state) {
506 		case NVME_CTRL_NEW:
507 		case NVME_CTRL_RESETTING:
508 		case NVME_CTRL_CONNECTING:
509 			changed = true;
510 			fallthrough;
511 		default:
512 			break;
513 		}
514 		break;
515 	case NVME_CTRL_RESETTING:
516 		switch (old_state) {
517 		case NVME_CTRL_NEW:
518 		case NVME_CTRL_LIVE:
519 			changed = true;
520 			fallthrough;
521 		default:
522 			break;
523 		}
524 		break;
525 	case NVME_CTRL_CONNECTING:
526 		switch (old_state) {
527 		case NVME_CTRL_NEW:
528 		case NVME_CTRL_RESETTING:
529 			changed = true;
530 			fallthrough;
531 		default:
532 			break;
533 		}
534 		break;
535 	case NVME_CTRL_DELETING:
536 		switch (old_state) {
537 		case NVME_CTRL_LIVE:
538 		case NVME_CTRL_RESETTING:
539 		case NVME_CTRL_CONNECTING:
540 			changed = true;
541 			fallthrough;
542 		default:
543 			break;
544 		}
545 		break;
546 	case NVME_CTRL_DELETING_NOIO:
547 		switch (old_state) {
548 		case NVME_CTRL_DELETING:
549 		case NVME_CTRL_DEAD:
550 			changed = true;
551 			fallthrough;
552 		default:
553 			break;
554 		}
555 		break;
556 	case NVME_CTRL_DEAD:
557 		switch (old_state) {
558 		case NVME_CTRL_DELETING:
559 			changed = true;
560 			fallthrough;
561 		default:
562 			break;
563 		}
564 		break;
565 	default:
566 		break;
567 	}
568 
569 	if (changed) {
570 		WRITE_ONCE(ctrl->state, new_state);
571 		wake_up_all(&ctrl->state_wq);
572 	}
573 
574 	spin_unlock_irqrestore(&ctrl->lock, flags);
575 	if (!changed)
576 		return false;
577 
578 	if (new_state == NVME_CTRL_LIVE) {
579 		if (old_state == NVME_CTRL_CONNECTING)
580 			nvme_stop_failfast_work(ctrl);
581 		nvme_kick_requeue_lists(ctrl);
582 	} else if (new_state == NVME_CTRL_CONNECTING &&
583 		old_state == NVME_CTRL_RESETTING) {
584 		nvme_start_failfast_work(ctrl);
585 	}
586 	return changed;
587 }
588 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
589 
590 /*
591  * Returns true for sink states that can't ever transition back to live.
592  */
593 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
594 {
595 	switch (nvme_ctrl_state(ctrl)) {
596 	case NVME_CTRL_NEW:
597 	case NVME_CTRL_LIVE:
598 	case NVME_CTRL_RESETTING:
599 	case NVME_CTRL_CONNECTING:
600 		return false;
601 	case NVME_CTRL_DELETING:
602 	case NVME_CTRL_DELETING_NOIO:
603 	case NVME_CTRL_DEAD:
604 		return true;
605 	default:
606 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
607 		return true;
608 	}
609 }
610 
611 /*
612  * Waits for the controller state to be resetting, or returns false if it is
613  * not possible to ever transition to that state.
614  */
615 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
616 {
617 	wait_event(ctrl->state_wq,
618 		   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
619 		   nvme_state_terminal(ctrl));
620 	return nvme_ctrl_state(ctrl) == NVME_CTRL_RESETTING;
621 }
622 EXPORT_SYMBOL_GPL(nvme_wait_reset);
623 
624 static void nvme_free_ns_head(struct kref *ref)
625 {
626 	struct nvme_ns_head *head =
627 		container_of(ref, struct nvme_ns_head, ref);
628 
629 	nvme_mpath_remove_disk(head);
630 	ida_free(&head->subsys->ns_ida, head->instance);
631 	cleanup_srcu_struct(&head->srcu);
632 	nvme_put_subsystem(head->subsys);
633 	kfree(head);
634 }
635 
636 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
637 {
638 	return kref_get_unless_zero(&head->ref);
639 }
640 
641 void nvme_put_ns_head(struct nvme_ns_head *head)
642 {
643 	kref_put(&head->ref, nvme_free_ns_head);
644 }
645 
646 static void nvme_free_ns(struct kref *kref)
647 {
648 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
649 
650 	put_disk(ns->disk);
651 	nvme_put_ns_head(ns->head);
652 	nvme_put_ctrl(ns->ctrl);
653 	kfree(ns);
654 }
655 
656 static inline bool nvme_get_ns(struct nvme_ns *ns)
657 {
658 	return kref_get_unless_zero(&ns->kref);
659 }
660 
661 void nvme_put_ns(struct nvme_ns *ns)
662 {
663 	kref_put(&ns->kref, nvme_free_ns);
664 }
665 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
666 
667 static inline void nvme_clear_nvme_request(struct request *req)
668 {
669 	nvme_req(req)->status = 0;
670 	nvme_req(req)->retries = 0;
671 	nvme_req(req)->flags = 0;
672 	req->rq_flags |= RQF_DONTPREP;
673 }
674 
675 /* initialize a passthrough request */
676 void nvme_init_request(struct request *req, struct nvme_command *cmd)
677 {
678 	if (req->q->queuedata)
679 		req->timeout = NVME_IO_TIMEOUT;
680 	else /* no queuedata implies admin queue */
681 		req->timeout = NVME_ADMIN_TIMEOUT;
682 
683 	/* passthru commands should let the driver set the SGL flags */
684 	cmd->common.flags &= ~NVME_CMD_SGL_ALL;
685 
686 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
687 	if (req->mq_hctx->type == HCTX_TYPE_POLL)
688 		req->cmd_flags |= REQ_POLLED;
689 	nvme_clear_nvme_request(req);
690 	req->rq_flags |= RQF_QUIET;
691 	memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
692 }
693 EXPORT_SYMBOL_GPL(nvme_init_request);
694 
695 /*
696  * For something we're not in a state to send to the device the default action
697  * is to busy it and retry it after the controller state is recovered.  However,
698  * if the controller is deleting or if anything is marked for failfast or
699  * nvme multipath it is immediately failed.
700  *
701  * Note: commands used to initialize the controller will be marked for failfast.
702  * Note: nvme cli/ioctl commands are marked for failfast.
703  */
704 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
705 		struct request *rq)
706 {
707 	enum nvme_ctrl_state state = nvme_ctrl_state(ctrl);
708 
709 	if (state != NVME_CTRL_DELETING_NOIO &&
710 	    state != NVME_CTRL_DELETING &&
711 	    state != NVME_CTRL_DEAD &&
712 	    !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
713 	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
714 		return BLK_STS_RESOURCE;
715 	return nvme_host_path_error(rq);
716 }
717 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
718 
719 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
720 		bool queue_live)
721 {
722 	struct nvme_request *req = nvme_req(rq);
723 
724 	/*
725 	 * currently we have a problem sending passthru commands
726 	 * on the admin_q if the controller is not LIVE because we can't
727 	 * make sure that they are going out after the admin connect,
728 	 * controller enable and/or other commands in the initialization
729 	 * sequence. until the controller will be LIVE, fail with
730 	 * BLK_STS_RESOURCE so that they will be rescheduled.
731 	 */
732 	if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
733 		return false;
734 
735 	if (ctrl->ops->flags & NVME_F_FABRICS) {
736 		/*
737 		 * Only allow commands on a live queue, except for the connect
738 		 * command, which is require to set the queue live in the
739 		 * appropinquate states.
740 		 */
741 		switch (nvme_ctrl_state(ctrl)) {
742 		case NVME_CTRL_CONNECTING:
743 			if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
744 			    (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
745 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
746 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
747 				return true;
748 			break;
749 		default:
750 			break;
751 		case NVME_CTRL_DEAD:
752 			return false;
753 		}
754 	}
755 
756 	return queue_live;
757 }
758 EXPORT_SYMBOL_GPL(__nvme_check_ready);
759 
760 static inline void nvme_setup_flush(struct nvme_ns *ns,
761 		struct nvme_command *cmnd)
762 {
763 	memset(cmnd, 0, sizeof(*cmnd));
764 	cmnd->common.opcode = nvme_cmd_flush;
765 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
766 }
767 
768 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
769 		struct nvme_command *cmnd)
770 {
771 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
772 	struct nvme_dsm_range *range;
773 	struct bio *bio;
774 
775 	/*
776 	 * Some devices do not consider the DSM 'Number of Ranges' field when
777 	 * determining how much data to DMA. Always allocate memory for maximum
778 	 * number of segments to prevent device reading beyond end of buffer.
779 	 */
780 	static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
781 
782 	range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
783 	if (!range) {
784 		/*
785 		 * If we fail allocation our range, fallback to the controller
786 		 * discard page. If that's also busy, it's safe to return
787 		 * busy, as we know we can make progress once that's freed.
788 		 */
789 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
790 			return BLK_STS_RESOURCE;
791 
792 		range = page_address(ns->ctrl->discard_page);
793 	}
794 
795 	if (queue_max_discard_segments(req->q) == 1) {
796 		u64 slba = nvme_sect_to_lba(ns, blk_rq_pos(req));
797 		u32 nlb = blk_rq_sectors(req) >> (ns->lba_shift - 9);
798 
799 		range[0].cattr = cpu_to_le32(0);
800 		range[0].nlb = cpu_to_le32(nlb);
801 		range[0].slba = cpu_to_le64(slba);
802 		n = 1;
803 	} else {
804 		__rq_for_each_bio(bio, req) {
805 			u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
806 			u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
807 
808 			if (n < segments) {
809 				range[n].cattr = cpu_to_le32(0);
810 				range[n].nlb = cpu_to_le32(nlb);
811 				range[n].slba = cpu_to_le64(slba);
812 			}
813 			n++;
814 		}
815 	}
816 
817 	if (WARN_ON_ONCE(n != segments)) {
818 		if (virt_to_page(range) == ns->ctrl->discard_page)
819 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
820 		else
821 			kfree(range);
822 		return BLK_STS_IOERR;
823 	}
824 
825 	memset(cmnd, 0, sizeof(*cmnd));
826 	cmnd->dsm.opcode = nvme_cmd_dsm;
827 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
828 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
829 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
830 
831 	bvec_set_virt(&req->special_vec, range, alloc_size);
832 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
833 
834 	return BLK_STS_OK;
835 }
836 
837 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
838 			      struct request *req)
839 {
840 	u32 upper, lower;
841 	u64 ref48;
842 
843 	/* both rw and write zeroes share the same reftag format */
844 	switch (ns->guard_type) {
845 	case NVME_NVM_NS_16B_GUARD:
846 		cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
847 		break;
848 	case NVME_NVM_NS_64B_GUARD:
849 		ref48 = ext_pi_ref_tag(req);
850 		lower = lower_32_bits(ref48);
851 		upper = upper_32_bits(ref48);
852 
853 		cmnd->rw.reftag = cpu_to_le32(lower);
854 		cmnd->rw.cdw3 = cpu_to_le32(upper);
855 		break;
856 	default:
857 		break;
858 	}
859 }
860 
861 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
862 		struct request *req, struct nvme_command *cmnd)
863 {
864 	memset(cmnd, 0, sizeof(*cmnd));
865 
866 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
867 		return nvme_setup_discard(ns, req, cmnd);
868 
869 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
870 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
871 	cmnd->write_zeroes.slba =
872 		cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
873 	cmnd->write_zeroes.length =
874 		cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
875 
876 	if (!(req->cmd_flags & REQ_NOUNMAP) && (ns->features & NVME_NS_DEAC))
877 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_WZ_DEAC);
878 
879 	if (nvme_ns_has_pi(ns)) {
880 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_RW_PRINFO_PRACT);
881 
882 		switch (ns->pi_type) {
883 		case NVME_NS_DPS_PI_TYPE1:
884 		case NVME_NS_DPS_PI_TYPE2:
885 			nvme_set_ref_tag(ns, cmnd, req);
886 			break;
887 		}
888 	}
889 
890 	return BLK_STS_OK;
891 }
892 
893 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
894 		struct request *req, struct nvme_command *cmnd,
895 		enum nvme_opcode op)
896 {
897 	u16 control = 0;
898 	u32 dsmgmt = 0;
899 
900 	if (req->cmd_flags & REQ_FUA)
901 		control |= NVME_RW_FUA;
902 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
903 		control |= NVME_RW_LR;
904 
905 	if (req->cmd_flags & REQ_RAHEAD)
906 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
907 
908 	cmnd->rw.opcode = op;
909 	cmnd->rw.flags = 0;
910 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
911 	cmnd->rw.cdw2 = 0;
912 	cmnd->rw.cdw3 = 0;
913 	cmnd->rw.metadata = 0;
914 	cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
915 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
916 	cmnd->rw.reftag = 0;
917 	cmnd->rw.apptag = 0;
918 	cmnd->rw.appmask = 0;
919 
920 	if (ns->ms) {
921 		/*
922 		 * If formated with metadata, the block layer always provides a
923 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
924 		 * we enable the PRACT bit for protection information or set the
925 		 * namespace capacity to zero to prevent any I/O.
926 		 */
927 		if (!blk_integrity_rq(req)) {
928 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
929 				return BLK_STS_NOTSUPP;
930 			control |= NVME_RW_PRINFO_PRACT;
931 		}
932 
933 		switch (ns->pi_type) {
934 		case NVME_NS_DPS_PI_TYPE3:
935 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
936 			break;
937 		case NVME_NS_DPS_PI_TYPE1:
938 		case NVME_NS_DPS_PI_TYPE2:
939 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
940 					NVME_RW_PRINFO_PRCHK_REF;
941 			if (op == nvme_cmd_zone_append)
942 				control |= NVME_RW_APPEND_PIREMAP;
943 			nvme_set_ref_tag(ns, cmnd, req);
944 			break;
945 		}
946 	}
947 
948 	cmnd->rw.control = cpu_to_le16(control);
949 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
950 	return 0;
951 }
952 
953 void nvme_cleanup_cmd(struct request *req)
954 {
955 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
956 		struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
957 
958 		if (req->special_vec.bv_page == ctrl->discard_page)
959 			clear_bit_unlock(0, &ctrl->discard_page_busy);
960 		else
961 			kfree(bvec_virt(&req->special_vec));
962 	}
963 }
964 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
965 
966 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
967 {
968 	struct nvme_command *cmd = nvme_req(req)->cmd;
969 	blk_status_t ret = BLK_STS_OK;
970 
971 	if (!(req->rq_flags & RQF_DONTPREP))
972 		nvme_clear_nvme_request(req);
973 
974 	switch (req_op(req)) {
975 	case REQ_OP_DRV_IN:
976 	case REQ_OP_DRV_OUT:
977 		/* these are setup prior to execution in nvme_init_request() */
978 		break;
979 	case REQ_OP_FLUSH:
980 		nvme_setup_flush(ns, cmd);
981 		break;
982 	case REQ_OP_ZONE_RESET_ALL:
983 	case REQ_OP_ZONE_RESET:
984 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
985 		break;
986 	case REQ_OP_ZONE_OPEN:
987 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
988 		break;
989 	case REQ_OP_ZONE_CLOSE:
990 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
991 		break;
992 	case REQ_OP_ZONE_FINISH:
993 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
994 		break;
995 	case REQ_OP_WRITE_ZEROES:
996 		ret = nvme_setup_write_zeroes(ns, req, cmd);
997 		break;
998 	case REQ_OP_DISCARD:
999 		ret = nvme_setup_discard(ns, req, cmd);
1000 		break;
1001 	case REQ_OP_READ:
1002 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1003 		break;
1004 	case REQ_OP_WRITE:
1005 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1006 		break;
1007 	case REQ_OP_ZONE_APPEND:
1008 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1009 		break;
1010 	default:
1011 		WARN_ON_ONCE(1);
1012 		return BLK_STS_IOERR;
1013 	}
1014 
1015 	cmd->common.command_id = nvme_cid(req);
1016 	trace_nvme_setup_cmd(req, cmd);
1017 	return ret;
1018 }
1019 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1020 
1021 /*
1022  * Return values:
1023  * 0:  success
1024  * >0: nvme controller's cqe status response
1025  * <0: kernel error in lieu of controller response
1026  */
1027 int nvme_execute_rq(struct request *rq, bool at_head)
1028 {
1029 	blk_status_t status;
1030 
1031 	status = blk_execute_rq(rq, at_head);
1032 	if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1033 		return -EINTR;
1034 	if (nvme_req(rq)->status)
1035 		return nvme_req(rq)->status;
1036 	return blk_status_to_errno(status);
1037 }
1038 EXPORT_SYMBOL_NS_GPL(nvme_execute_rq, NVME_TARGET_PASSTHRU);
1039 
1040 /*
1041  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1042  * if the result is positive, it's an NVM Express status code
1043  */
1044 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1045 		union nvme_result *result, void *buffer, unsigned bufflen,
1046 		int qid, int at_head, blk_mq_req_flags_t flags)
1047 {
1048 	struct request *req;
1049 	int ret;
1050 
1051 	if (qid == NVME_QID_ANY)
1052 		req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1053 	else
1054 		req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1055 						qid - 1);
1056 
1057 	if (IS_ERR(req))
1058 		return PTR_ERR(req);
1059 	nvme_init_request(req, cmd);
1060 
1061 	if (buffer && bufflen) {
1062 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1063 		if (ret)
1064 			goto out;
1065 	}
1066 
1067 	ret = nvme_execute_rq(req, at_head);
1068 	if (result && ret >= 0)
1069 		*result = nvme_req(req)->result;
1070  out:
1071 	blk_mq_free_request(req);
1072 	return ret;
1073 }
1074 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1075 
1076 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1077 		void *buffer, unsigned bufflen)
1078 {
1079 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1080 			NVME_QID_ANY, 0, 0);
1081 }
1082 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1083 
1084 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1085 {
1086 	u32 effects = 0;
1087 
1088 	if (ns) {
1089 		effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1090 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1091 			dev_warn_once(ctrl->device,
1092 				"IO command:%02x has unusual effects:%08x\n",
1093 				opcode, effects);
1094 
1095 		/*
1096 		 * NVME_CMD_EFFECTS_CSE_MASK causes a freeze all I/O queues,
1097 		 * which would deadlock when done on an I/O command.  Note that
1098 		 * We already warn about an unusual effect above.
1099 		 */
1100 		effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1101 	} else {
1102 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1103 	}
1104 
1105 	return effects;
1106 }
1107 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1108 
1109 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1110 {
1111 	u32 effects = nvme_command_effects(ctrl, ns, opcode);
1112 
1113 	/*
1114 	 * For simplicity, IO to all namespaces is quiesced even if the command
1115 	 * effects say only one namespace is affected.
1116 	 */
1117 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1118 		mutex_lock(&ctrl->scan_lock);
1119 		mutex_lock(&ctrl->subsys->lock);
1120 		nvme_mpath_start_freeze(ctrl->subsys);
1121 		nvme_mpath_wait_freeze(ctrl->subsys);
1122 		nvme_start_freeze(ctrl);
1123 		nvme_wait_freeze(ctrl);
1124 	}
1125 	return effects;
1126 }
1127 EXPORT_SYMBOL_NS_GPL(nvme_passthru_start, NVME_TARGET_PASSTHRU);
1128 
1129 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1130 		       struct nvme_command *cmd, int status)
1131 {
1132 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1133 		nvme_unfreeze(ctrl);
1134 		nvme_mpath_unfreeze(ctrl->subsys);
1135 		mutex_unlock(&ctrl->subsys->lock);
1136 		mutex_unlock(&ctrl->scan_lock);
1137 	}
1138 	if (effects & NVME_CMD_EFFECTS_CCC) {
1139 		if (!test_and_set_bit(NVME_CTRL_DIRTY_CAPABILITY,
1140 				      &ctrl->flags)) {
1141 			dev_info(ctrl->device,
1142 "controller capabilities changed, reset may be required to take effect.\n");
1143 		}
1144 	}
1145 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1146 		nvme_queue_scan(ctrl);
1147 		flush_work(&ctrl->scan_work);
1148 	}
1149 	if (ns)
1150 		return;
1151 
1152 	switch (cmd->common.opcode) {
1153 	case nvme_admin_set_features:
1154 		switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1155 		case NVME_FEAT_KATO:
1156 			/*
1157 			 * Keep alive commands interval on the host should be
1158 			 * updated when KATO is modified by Set Features
1159 			 * commands.
1160 			 */
1161 			if (!status)
1162 				nvme_update_keep_alive(ctrl, cmd);
1163 			break;
1164 		default:
1165 			break;
1166 		}
1167 		break;
1168 	default:
1169 		break;
1170 	}
1171 }
1172 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1173 
1174 /*
1175  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1176  *
1177  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1178  *   accounting for transport roundtrip times [..].
1179  */
1180 static unsigned long nvme_keep_alive_work_period(struct nvme_ctrl *ctrl)
1181 {
1182 	unsigned long delay = ctrl->kato * HZ / 2;
1183 
1184 	/*
1185 	 * When using Traffic Based Keep Alive, we need to run
1186 	 * nvme_keep_alive_work at twice the normal frequency, as one
1187 	 * command completion can postpone sending a keep alive command
1188 	 * by up to twice the delay between runs.
1189 	 */
1190 	if (ctrl->ctratt & NVME_CTRL_ATTR_TBKAS)
1191 		delay /= 2;
1192 	return delay;
1193 }
1194 
1195 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1196 {
1197 	queue_delayed_work(nvme_wq, &ctrl->ka_work,
1198 			   nvme_keep_alive_work_period(ctrl));
1199 }
1200 
1201 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1202 						 blk_status_t status)
1203 {
1204 	struct nvme_ctrl *ctrl = rq->end_io_data;
1205 	unsigned long flags;
1206 	bool startka = false;
1207 	unsigned long rtt = jiffies - (rq->deadline - rq->timeout);
1208 	unsigned long delay = nvme_keep_alive_work_period(ctrl);
1209 
1210 	/*
1211 	 * Subtract off the keepalive RTT so nvme_keep_alive_work runs
1212 	 * at the desired frequency.
1213 	 */
1214 	if (rtt <= delay) {
1215 		delay -= rtt;
1216 	} else {
1217 		dev_warn(ctrl->device, "long keepalive RTT (%u ms)\n",
1218 			 jiffies_to_msecs(rtt));
1219 		delay = 0;
1220 	}
1221 
1222 	blk_mq_free_request(rq);
1223 
1224 	if (status) {
1225 		dev_err(ctrl->device,
1226 			"failed nvme_keep_alive_end_io error=%d\n",
1227 				status);
1228 		return RQ_END_IO_NONE;
1229 	}
1230 
1231 	ctrl->ka_last_check_time = jiffies;
1232 	ctrl->comp_seen = false;
1233 	spin_lock_irqsave(&ctrl->lock, flags);
1234 	if (ctrl->state == NVME_CTRL_LIVE ||
1235 	    ctrl->state == NVME_CTRL_CONNECTING)
1236 		startka = true;
1237 	spin_unlock_irqrestore(&ctrl->lock, flags);
1238 	if (startka)
1239 		queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1240 	return RQ_END_IO_NONE;
1241 }
1242 
1243 static void nvme_keep_alive_work(struct work_struct *work)
1244 {
1245 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1246 			struct nvme_ctrl, ka_work);
1247 	bool comp_seen = ctrl->comp_seen;
1248 	struct request *rq;
1249 
1250 	ctrl->ka_last_check_time = jiffies;
1251 
1252 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1253 		dev_dbg(ctrl->device,
1254 			"reschedule traffic based keep-alive timer\n");
1255 		ctrl->comp_seen = false;
1256 		nvme_queue_keep_alive_work(ctrl);
1257 		return;
1258 	}
1259 
1260 	rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1261 				  BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1262 	if (IS_ERR(rq)) {
1263 		/* allocation failure, reset the controller */
1264 		dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1265 		nvme_reset_ctrl(ctrl);
1266 		return;
1267 	}
1268 	nvme_init_request(rq, &ctrl->ka_cmd);
1269 
1270 	rq->timeout = ctrl->kato * HZ;
1271 	rq->end_io = nvme_keep_alive_end_io;
1272 	rq->end_io_data = ctrl;
1273 	blk_execute_rq_nowait(rq, false);
1274 }
1275 
1276 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1277 {
1278 	if (unlikely(ctrl->kato == 0))
1279 		return;
1280 
1281 	nvme_queue_keep_alive_work(ctrl);
1282 }
1283 
1284 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1285 {
1286 	if (unlikely(ctrl->kato == 0))
1287 		return;
1288 
1289 	cancel_delayed_work_sync(&ctrl->ka_work);
1290 }
1291 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1292 
1293 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1294 				   struct nvme_command *cmd)
1295 {
1296 	unsigned int new_kato =
1297 		DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1298 
1299 	dev_info(ctrl->device,
1300 		 "keep alive interval updated from %u ms to %u ms\n",
1301 		 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1302 
1303 	nvme_stop_keep_alive(ctrl);
1304 	ctrl->kato = new_kato;
1305 	nvme_start_keep_alive(ctrl);
1306 }
1307 
1308 /*
1309  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1310  * flag, thus sending any new CNS opcodes has a big chance of not working.
1311  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1312  * (but not for any later version).
1313  */
1314 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1315 {
1316 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1317 		return ctrl->vs < NVME_VS(1, 2, 0);
1318 	return ctrl->vs < NVME_VS(1, 1, 0);
1319 }
1320 
1321 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1322 {
1323 	struct nvme_command c = { };
1324 	int error;
1325 
1326 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1327 	c.identify.opcode = nvme_admin_identify;
1328 	c.identify.cns = NVME_ID_CNS_CTRL;
1329 
1330 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1331 	if (!*id)
1332 		return -ENOMEM;
1333 
1334 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1335 			sizeof(struct nvme_id_ctrl));
1336 	if (error)
1337 		kfree(*id);
1338 	return error;
1339 }
1340 
1341 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1342 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1343 {
1344 	const char *warn_str = "ctrl returned bogus length:";
1345 	void *data = cur;
1346 
1347 	switch (cur->nidt) {
1348 	case NVME_NIDT_EUI64:
1349 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1350 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1351 				 warn_str, cur->nidl);
1352 			return -1;
1353 		}
1354 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1355 			return NVME_NIDT_EUI64_LEN;
1356 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1357 		return NVME_NIDT_EUI64_LEN;
1358 	case NVME_NIDT_NGUID:
1359 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1360 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1361 				 warn_str, cur->nidl);
1362 			return -1;
1363 		}
1364 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1365 			return NVME_NIDT_NGUID_LEN;
1366 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1367 		return NVME_NIDT_NGUID_LEN;
1368 	case NVME_NIDT_UUID:
1369 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1370 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1371 				 warn_str, cur->nidl);
1372 			return -1;
1373 		}
1374 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1375 			return NVME_NIDT_UUID_LEN;
1376 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1377 		return NVME_NIDT_UUID_LEN;
1378 	case NVME_NIDT_CSI:
1379 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1380 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1381 				 warn_str, cur->nidl);
1382 			return -1;
1383 		}
1384 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1385 		*csi_seen = true;
1386 		return NVME_NIDT_CSI_LEN;
1387 	default:
1388 		/* Skip unknown types */
1389 		return cur->nidl;
1390 	}
1391 }
1392 
1393 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1394 		struct nvme_ns_info *info)
1395 {
1396 	struct nvme_command c = { };
1397 	bool csi_seen = false;
1398 	int status, pos, len;
1399 	void *data;
1400 
1401 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1402 		return 0;
1403 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1404 		return 0;
1405 
1406 	c.identify.opcode = nvme_admin_identify;
1407 	c.identify.nsid = cpu_to_le32(info->nsid);
1408 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1409 
1410 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1411 	if (!data)
1412 		return -ENOMEM;
1413 
1414 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1415 				      NVME_IDENTIFY_DATA_SIZE);
1416 	if (status) {
1417 		dev_warn(ctrl->device,
1418 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1419 			info->nsid, status);
1420 		goto free_data;
1421 	}
1422 
1423 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1424 		struct nvme_ns_id_desc *cur = data + pos;
1425 
1426 		if (cur->nidl == 0)
1427 			break;
1428 
1429 		len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1430 		if (len < 0)
1431 			break;
1432 
1433 		len += sizeof(*cur);
1434 	}
1435 
1436 	if (nvme_multi_css(ctrl) && !csi_seen) {
1437 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1438 			 info->nsid);
1439 		status = -EINVAL;
1440 	}
1441 
1442 free_data:
1443 	kfree(data);
1444 	return status;
1445 }
1446 
1447 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1448 			struct nvme_id_ns **id)
1449 {
1450 	struct nvme_command c = { };
1451 	int error;
1452 
1453 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1454 	c.identify.opcode = nvme_admin_identify;
1455 	c.identify.nsid = cpu_to_le32(nsid);
1456 	c.identify.cns = NVME_ID_CNS_NS;
1457 
1458 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1459 	if (!*id)
1460 		return -ENOMEM;
1461 
1462 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1463 	if (error) {
1464 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1465 		kfree(*id);
1466 	}
1467 	return error;
1468 }
1469 
1470 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1471 		struct nvme_ns_info *info)
1472 {
1473 	struct nvme_ns_ids *ids = &info->ids;
1474 	struct nvme_id_ns *id;
1475 	int ret;
1476 
1477 	ret = nvme_identify_ns(ctrl, info->nsid, &id);
1478 	if (ret)
1479 		return ret;
1480 
1481 	if (id->ncap == 0) {
1482 		/* namespace not allocated or attached */
1483 		info->is_removed = true;
1484 		ret = -ENODEV;
1485 		goto error;
1486 	}
1487 
1488 	info->anagrpid = id->anagrpid;
1489 	info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1490 	info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1491 	info->is_ready = true;
1492 	if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1493 		dev_info(ctrl->device,
1494 			 "Ignoring bogus Namespace Identifiers\n");
1495 	} else {
1496 		if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1497 		    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1498 			memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1499 		if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1500 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1501 			memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1502 	}
1503 
1504 error:
1505 	kfree(id);
1506 	return ret;
1507 }
1508 
1509 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1510 		struct nvme_ns_info *info)
1511 {
1512 	struct nvme_id_ns_cs_indep *id;
1513 	struct nvme_command c = {
1514 		.identify.opcode	= nvme_admin_identify,
1515 		.identify.nsid		= cpu_to_le32(info->nsid),
1516 		.identify.cns		= NVME_ID_CNS_NS_CS_INDEP,
1517 	};
1518 	int ret;
1519 
1520 	id = kmalloc(sizeof(*id), GFP_KERNEL);
1521 	if (!id)
1522 		return -ENOMEM;
1523 
1524 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1525 	if (!ret) {
1526 		info->anagrpid = id->anagrpid;
1527 		info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1528 		info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1529 		info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1530 	}
1531 	kfree(id);
1532 	return ret;
1533 }
1534 
1535 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1536 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1537 {
1538 	union nvme_result res = { 0 };
1539 	struct nvme_command c = { };
1540 	int ret;
1541 
1542 	c.features.opcode = op;
1543 	c.features.fid = cpu_to_le32(fid);
1544 	c.features.dword11 = cpu_to_le32(dword11);
1545 
1546 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1547 			buffer, buflen, NVME_QID_ANY, 0, 0);
1548 	if (ret >= 0 && result)
1549 		*result = le32_to_cpu(res.u32);
1550 	return ret;
1551 }
1552 
1553 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1554 		      unsigned int dword11, void *buffer, size_t buflen,
1555 		      u32 *result)
1556 {
1557 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1558 			     buflen, result);
1559 }
1560 EXPORT_SYMBOL_GPL(nvme_set_features);
1561 
1562 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1563 		      unsigned int dword11, void *buffer, size_t buflen,
1564 		      u32 *result)
1565 {
1566 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1567 			     buflen, result);
1568 }
1569 EXPORT_SYMBOL_GPL(nvme_get_features);
1570 
1571 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1572 {
1573 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1574 	u32 result;
1575 	int status, nr_io_queues;
1576 
1577 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1578 			&result);
1579 	if (status < 0)
1580 		return status;
1581 
1582 	/*
1583 	 * Degraded controllers might return an error when setting the queue
1584 	 * count.  We still want to be able to bring them online and offer
1585 	 * access to the admin queue, as that might be only way to fix them up.
1586 	 */
1587 	if (status > 0) {
1588 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1589 		*count = 0;
1590 	} else {
1591 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1592 		*count = min(*count, nr_io_queues);
1593 	}
1594 
1595 	return 0;
1596 }
1597 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1598 
1599 #define NVME_AEN_SUPPORTED \
1600 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1601 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1602 
1603 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1604 {
1605 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1606 	int status;
1607 
1608 	if (!supported_aens)
1609 		return;
1610 
1611 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1612 			NULL, 0, &result);
1613 	if (status)
1614 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1615 			 supported_aens);
1616 
1617 	queue_work(nvme_wq, &ctrl->async_event_work);
1618 }
1619 
1620 static int nvme_ns_open(struct nvme_ns *ns)
1621 {
1622 
1623 	/* should never be called due to GENHD_FL_HIDDEN */
1624 	if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1625 		goto fail;
1626 	if (!nvme_get_ns(ns))
1627 		goto fail;
1628 	if (!try_module_get(ns->ctrl->ops->module))
1629 		goto fail_put_ns;
1630 
1631 	return 0;
1632 
1633 fail_put_ns:
1634 	nvme_put_ns(ns);
1635 fail:
1636 	return -ENXIO;
1637 }
1638 
1639 static void nvme_ns_release(struct nvme_ns *ns)
1640 {
1641 
1642 	module_put(ns->ctrl->ops->module);
1643 	nvme_put_ns(ns);
1644 }
1645 
1646 static int nvme_open(struct gendisk *disk, blk_mode_t mode)
1647 {
1648 	return nvme_ns_open(disk->private_data);
1649 }
1650 
1651 static void nvme_release(struct gendisk *disk)
1652 {
1653 	nvme_ns_release(disk->private_data);
1654 }
1655 
1656 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1657 {
1658 	/* some standard values */
1659 	geo->heads = 1 << 6;
1660 	geo->sectors = 1 << 5;
1661 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1662 	return 0;
1663 }
1664 
1665 #ifdef CONFIG_BLK_DEV_INTEGRITY
1666 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1667 				u32 max_integrity_segments)
1668 {
1669 	struct blk_integrity integrity = { };
1670 
1671 	switch (ns->pi_type) {
1672 	case NVME_NS_DPS_PI_TYPE3:
1673 		switch (ns->guard_type) {
1674 		case NVME_NVM_NS_16B_GUARD:
1675 			integrity.profile = &t10_pi_type3_crc;
1676 			integrity.tag_size = sizeof(u16) + sizeof(u32);
1677 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1678 			break;
1679 		case NVME_NVM_NS_64B_GUARD:
1680 			integrity.profile = &ext_pi_type3_crc64;
1681 			integrity.tag_size = sizeof(u16) + 6;
1682 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1683 			break;
1684 		default:
1685 			integrity.profile = NULL;
1686 			break;
1687 		}
1688 		break;
1689 	case NVME_NS_DPS_PI_TYPE1:
1690 	case NVME_NS_DPS_PI_TYPE2:
1691 		switch (ns->guard_type) {
1692 		case NVME_NVM_NS_16B_GUARD:
1693 			integrity.profile = &t10_pi_type1_crc;
1694 			integrity.tag_size = sizeof(u16);
1695 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1696 			break;
1697 		case NVME_NVM_NS_64B_GUARD:
1698 			integrity.profile = &ext_pi_type1_crc64;
1699 			integrity.tag_size = sizeof(u16);
1700 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1701 			break;
1702 		default:
1703 			integrity.profile = NULL;
1704 			break;
1705 		}
1706 		break;
1707 	default:
1708 		integrity.profile = NULL;
1709 		break;
1710 	}
1711 
1712 	integrity.tuple_size = ns->ms;
1713 	blk_integrity_register(disk, &integrity);
1714 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1715 }
1716 #else
1717 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1718 				u32 max_integrity_segments)
1719 {
1720 }
1721 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1722 
1723 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1724 {
1725 	struct nvme_ctrl *ctrl = ns->ctrl;
1726 	struct request_queue *queue = disk->queue;
1727 	u32 size = queue_logical_block_size(queue);
1728 
1729 	if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1730 		ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1731 
1732 	if (ctrl->max_discard_sectors == 0) {
1733 		blk_queue_max_discard_sectors(queue, 0);
1734 		return;
1735 	}
1736 
1737 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1738 			NVME_DSM_MAX_RANGES);
1739 
1740 	queue->limits.discard_granularity = size;
1741 
1742 	/* If discard is already enabled, don't reset queue limits */
1743 	if (queue->limits.max_discard_sectors)
1744 		return;
1745 
1746 	blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1747 	blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1748 
1749 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1750 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1751 }
1752 
1753 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1754 {
1755 	return uuid_equal(&a->uuid, &b->uuid) &&
1756 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1757 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1758 		a->csi == b->csi;
1759 }
1760 
1761 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1762 {
1763 	bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1764 	unsigned lbaf = nvme_lbaf_index(id->flbas);
1765 	struct nvme_ctrl *ctrl = ns->ctrl;
1766 	struct nvme_command c = { };
1767 	struct nvme_id_ns_nvm *nvm;
1768 	int ret = 0;
1769 	u32 elbaf;
1770 
1771 	ns->pi_size = 0;
1772 	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1773 	if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1774 		ns->pi_size = sizeof(struct t10_pi_tuple);
1775 		ns->guard_type = NVME_NVM_NS_16B_GUARD;
1776 		goto set_pi;
1777 	}
1778 
1779 	nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1780 	if (!nvm)
1781 		return -ENOMEM;
1782 
1783 	c.identify.opcode = nvme_admin_identify;
1784 	c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1785 	c.identify.cns = NVME_ID_CNS_CS_NS;
1786 	c.identify.csi = NVME_CSI_NVM;
1787 
1788 	ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1789 	if (ret)
1790 		goto free_data;
1791 
1792 	elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1793 
1794 	/* no support for storage tag formats right now */
1795 	if (nvme_elbaf_sts(elbaf))
1796 		goto free_data;
1797 
1798 	ns->guard_type = nvme_elbaf_guard_type(elbaf);
1799 	switch (ns->guard_type) {
1800 	case NVME_NVM_NS_64B_GUARD:
1801 		ns->pi_size = sizeof(struct crc64_pi_tuple);
1802 		break;
1803 	case NVME_NVM_NS_16B_GUARD:
1804 		ns->pi_size = sizeof(struct t10_pi_tuple);
1805 		break;
1806 	default:
1807 		break;
1808 	}
1809 
1810 free_data:
1811 	kfree(nvm);
1812 set_pi:
1813 	if (ns->pi_size && (first || ns->ms == ns->pi_size))
1814 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1815 	else
1816 		ns->pi_type = 0;
1817 
1818 	return ret;
1819 }
1820 
1821 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1822 {
1823 	struct nvme_ctrl *ctrl = ns->ctrl;
1824 	int ret;
1825 
1826 	ret = nvme_init_ms(ns, id);
1827 	if (ret)
1828 		return ret;
1829 
1830 	ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1831 	if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1832 		return 0;
1833 
1834 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1835 		/*
1836 		 * The NVMe over Fabrics specification only supports metadata as
1837 		 * part of the extended data LBA.  We rely on HCA/HBA support to
1838 		 * remap the separate metadata buffer from the block layer.
1839 		 */
1840 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1841 			return 0;
1842 
1843 		ns->features |= NVME_NS_EXT_LBAS;
1844 
1845 		/*
1846 		 * The current fabrics transport drivers support namespace
1847 		 * metadata formats only if nvme_ns_has_pi() returns true.
1848 		 * Suppress support for all other formats so the namespace will
1849 		 * have a 0 capacity and not be usable through the block stack.
1850 		 *
1851 		 * Note, this check will need to be modified if any drivers
1852 		 * gain the ability to use other metadata formats.
1853 		 */
1854 		if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1855 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1856 	} else {
1857 		/*
1858 		 * For PCIe controllers, we can't easily remap the separate
1859 		 * metadata buffer from the block layer and thus require a
1860 		 * separate metadata buffer for block layer metadata/PI support.
1861 		 * We allow extended LBAs for the passthrough interface, though.
1862 		 */
1863 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
1864 			ns->features |= NVME_NS_EXT_LBAS;
1865 		else
1866 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1867 	}
1868 	return 0;
1869 }
1870 
1871 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1872 		struct request_queue *q)
1873 {
1874 	bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1875 
1876 	if (ctrl->max_hw_sectors) {
1877 		u32 max_segments =
1878 			(ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1879 
1880 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
1881 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1882 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1883 	}
1884 	blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1885 	blk_queue_dma_alignment(q, 3);
1886 	blk_queue_write_cache(q, vwc, vwc);
1887 }
1888 
1889 static void nvme_update_disk_info(struct gendisk *disk,
1890 		struct nvme_ns *ns, struct nvme_id_ns *id)
1891 {
1892 	sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1893 	u32 bs = 1U << ns->lba_shift;
1894 	u32 atomic_bs, phys_bs, io_opt = 0;
1895 
1896 	/*
1897 	 * The block layer can't support LBA sizes larger than the page size
1898 	 * or smaller than a sector size yet, so catch this early and don't
1899 	 * allow block I/O.
1900 	 */
1901 	if (ns->lba_shift > PAGE_SHIFT || ns->lba_shift < SECTOR_SHIFT) {
1902 		capacity = 0;
1903 		bs = (1 << 9);
1904 	}
1905 
1906 	blk_integrity_unregister(disk);
1907 
1908 	atomic_bs = phys_bs = bs;
1909 	if (id->nabo == 0) {
1910 		/*
1911 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1912 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1913 		 * 0 then AWUPF must be used instead.
1914 		 */
1915 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1916 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1917 		else
1918 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1919 	}
1920 
1921 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1922 		/* NPWG = Namespace Preferred Write Granularity */
1923 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1924 		/* NOWS = Namespace Optimal Write Size */
1925 		io_opt = bs * (1 + le16_to_cpu(id->nows));
1926 	}
1927 
1928 	blk_queue_logical_block_size(disk->queue, bs);
1929 	/*
1930 	 * Linux filesystems assume writing a single physical block is
1931 	 * an atomic operation. Hence limit the physical block size to the
1932 	 * value of the Atomic Write Unit Power Fail parameter.
1933 	 */
1934 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1935 	blk_queue_io_min(disk->queue, phys_bs);
1936 	blk_queue_io_opt(disk->queue, io_opt);
1937 
1938 	/*
1939 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
1940 	 * metadata masquerading as Type 0 if supported, otherwise reject block
1941 	 * I/O to namespaces with metadata except when the namespace supports
1942 	 * PI, as it can strip/insert in that case.
1943 	 */
1944 	if (ns->ms) {
1945 		if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1946 		    (ns->features & NVME_NS_METADATA_SUPPORTED))
1947 			nvme_init_integrity(disk, ns,
1948 					    ns->ctrl->max_integrity_segments);
1949 		else if (!nvme_ns_has_pi(ns))
1950 			capacity = 0;
1951 	}
1952 
1953 	set_capacity_and_notify(disk, capacity);
1954 
1955 	nvme_config_discard(disk, ns);
1956 	blk_queue_max_write_zeroes_sectors(disk->queue,
1957 					   ns->ctrl->max_zeroes_sectors);
1958 }
1959 
1960 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
1961 {
1962 	return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
1963 }
1964 
1965 static inline bool nvme_first_scan(struct gendisk *disk)
1966 {
1967 	/* nvme_alloc_ns() scans the disk prior to adding it */
1968 	return !disk_live(disk);
1969 }
1970 
1971 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1972 {
1973 	struct nvme_ctrl *ctrl = ns->ctrl;
1974 	u32 iob;
1975 
1976 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1977 	    is_power_of_2(ctrl->max_hw_sectors))
1978 		iob = ctrl->max_hw_sectors;
1979 	else
1980 		iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1981 
1982 	if (!iob)
1983 		return;
1984 
1985 	if (!is_power_of_2(iob)) {
1986 		if (nvme_first_scan(ns->disk))
1987 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1988 				ns->disk->disk_name, iob);
1989 		return;
1990 	}
1991 
1992 	if (blk_queue_is_zoned(ns->disk->queue)) {
1993 		if (nvme_first_scan(ns->disk))
1994 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
1995 				ns->disk->disk_name);
1996 		return;
1997 	}
1998 
1999 	blk_queue_chunk_sectors(ns->queue, iob);
2000 }
2001 
2002 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
2003 		struct nvme_ns_info *info)
2004 {
2005 	blk_mq_freeze_queue(ns->disk->queue);
2006 	nvme_set_queue_limits(ns->ctrl, ns->queue);
2007 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2008 	blk_mq_unfreeze_queue(ns->disk->queue);
2009 
2010 	if (nvme_ns_head_multipath(ns->head)) {
2011 		blk_mq_freeze_queue(ns->head->disk->queue);
2012 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2013 		nvme_mpath_revalidate_paths(ns);
2014 		blk_stack_limits(&ns->head->disk->queue->limits,
2015 				 &ns->queue->limits, 0);
2016 		ns->head->disk->flags |= GENHD_FL_HIDDEN;
2017 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2018 	}
2019 
2020 	/* Hide the block-interface for these devices */
2021 	ns->disk->flags |= GENHD_FL_HIDDEN;
2022 	set_bit(NVME_NS_READY, &ns->flags);
2023 
2024 	return 0;
2025 }
2026 
2027 static int nvme_update_ns_info_block(struct nvme_ns *ns,
2028 		struct nvme_ns_info *info)
2029 {
2030 	struct nvme_id_ns *id;
2031 	unsigned lbaf;
2032 	int ret;
2033 
2034 	ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
2035 	if (ret)
2036 		return ret;
2037 
2038 	if (id->ncap == 0) {
2039 		/* namespace not allocated or attached */
2040 		info->is_removed = true;
2041 		ret = -ENODEV;
2042 		goto error;
2043 	}
2044 
2045 	blk_mq_freeze_queue(ns->disk->queue);
2046 	lbaf = nvme_lbaf_index(id->flbas);
2047 	ns->lba_shift = id->lbaf[lbaf].ds;
2048 	nvme_set_queue_limits(ns->ctrl, ns->queue);
2049 
2050 	ret = nvme_configure_metadata(ns, id);
2051 	if (ret < 0) {
2052 		blk_mq_unfreeze_queue(ns->disk->queue);
2053 		goto out;
2054 	}
2055 	nvme_set_chunk_sectors(ns, id);
2056 	nvme_update_disk_info(ns->disk, ns, id);
2057 
2058 	if (ns->head->ids.csi == NVME_CSI_ZNS) {
2059 		ret = nvme_update_zone_info(ns, lbaf);
2060 		if (ret) {
2061 			blk_mq_unfreeze_queue(ns->disk->queue);
2062 			goto out;
2063 		}
2064 	}
2065 
2066 	/*
2067 	 * Only set the DEAC bit if the device guarantees that reads from
2068 	 * deallocated data return zeroes.  While the DEAC bit does not
2069 	 * require that, it must be a no-op if reads from deallocated data
2070 	 * do not return zeroes.
2071 	 */
2072 	if ((id->dlfeat & 0x7) == 0x1 && (id->dlfeat & (1 << 3)))
2073 		ns->features |= NVME_NS_DEAC;
2074 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2075 	set_bit(NVME_NS_READY, &ns->flags);
2076 	blk_mq_unfreeze_queue(ns->disk->queue);
2077 
2078 	if (blk_queue_is_zoned(ns->queue)) {
2079 		ret = nvme_revalidate_zones(ns);
2080 		if (ret && !nvme_first_scan(ns->disk))
2081 			goto out;
2082 	}
2083 
2084 	if (nvme_ns_head_multipath(ns->head)) {
2085 		blk_mq_freeze_queue(ns->head->disk->queue);
2086 		nvme_update_disk_info(ns->head->disk, ns, id);
2087 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2088 		nvme_mpath_revalidate_paths(ns);
2089 		blk_stack_limits(&ns->head->disk->queue->limits,
2090 				 &ns->queue->limits, 0);
2091 		disk_update_readahead(ns->head->disk);
2092 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2093 	}
2094 
2095 	ret = 0;
2096 out:
2097 	/*
2098 	 * If probing fails due an unsupported feature, hide the block device,
2099 	 * but still allow other access.
2100 	 */
2101 	if (ret == -ENODEV) {
2102 		ns->disk->flags |= GENHD_FL_HIDDEN;
2103 		set_bit(NVME_NS_READY, &ns->flags);
2104 		ret = 0;
2105 	}
2106 
2107 error:
2108 	kfree(id);
2109 	return ret;
2110 }
2111 
2112 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2113 {
2114 	switch (info->ids.csi) {
2115 	case NVME_CSI_ZNS:
2116 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2117 			dev_info(ns->ctrl->device,
2118 	"block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2119 				info->nsid);
2120 			return nvme_update_ns_info_generic(ns, info);
2121 		}
2122 		return nvme_update_ns_info_block(ns, info);
2123 	case NVME_CSI_NVM:
2124 		return nvme_update_ns_info_block(ns, info);
2125 	default:
2126 		dev_info(ns->ctrl->device,
2127 			"block device for nsid %u not supported (csi %u)\n",
2128 			info->nsid, info->ids.csi);
2129 		return nvme_update_ns_info_generic(ns, info);
2130 	}
2131 }
2132 
2133 #ifdef CONFIG_BLK_SED_OPAL
2134 static int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2135 		bool send)
2136 {
2137 	struct nvme_ctrl *ctrl = data;
2138 	struct nvme_command cmd = { };
2139 
2140 	if (send)
2141 		cmd.common.opcode = nvme_admin_security_send;
2142 	else
2143 		cmd.common.opcode = nvme_admin_security_recv;
2144 	cmd.common.nsid = 0;
2145 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2146 	cmd.common.cdw11 = cpu_to_le32(len);
2147 
2148 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2149 			NVME_QID_ANY, 1, 0);
2150 }
2151 
2152 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2153 {
2154 	if (ctrl->oacs & NVME_CTRL_OACS_SEC_SUPP) {
2155 		if (!ctrl->opal_dev)
2156 			ctrl->opal_dev = init_opal_dev(ctrl, &nvme_sec_submit);
2157 		else if (was_suspended)
2158 			opal_unlock_from_suspend(ctrl->opal_dev);
2159 	} else {
2160 		free_opal_dev(ctrl->opal_dev);
2161 		ctrl->opal_dev = NULL;
2162 	}
2163 }
2164 #else
2165 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2166 {
2167 }
2168 #endif /* CONFIG_BLK_SED_OPAL */
2169 
2170 #ifdef CONFIG_BLK_DEV_ZONED
2171 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2172 		unsigned int nr_zones, report_zones_cb cb, void *data)
2173 {
2174 	return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2175 			data);
2176 }
2177 #else
2178 #define nvme_report_zones	NULL
2179 #endif /* CONFIG_BLK_DEV_ZONED */
2180 
2181 const struct block_device_operations nvme_bdev_ops = {
2182 	.owner		= THIS_MODULE,
2183 	.ioctl		= nvme_ioctl,
2184 	.compat_ioctl	= blkdev_compat_ptr_ioctl,
2185 	.open		= nvme_open,
2186 	.release	= nvme_release,
2187 	.getgeo		= nvme_getgeo,
2188 	.report_zones	= nvme_report_zones,
2189 	.pr_ops		= &nvme_pr_ops,
2190 };
2191 
2192 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 mask, u32 val,
2193 		u32 timeout, const char *op)
2194 {
2195 	unsigned long timeout_jiffies = jiffies + timeout * HZ;
2196 	u32 csts;
2197 	int ret;
2198 
2199 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2200 		if (csts == ~0)
2201 			return -ENODEV;
2202 		if ((csts & mask) == val)
2203 			break;
2204 
2205 		usleep_range(1000, 2000);
2206 		if (fatal_signal_pending(current))
2207 			return -EINTR;
2208 		if (time_after(jiffies, timeout_jiffies)) {
2209 			dev_err(ctrl->device,
2210 				"Device not ready; aborting %s, CSTS=0x%x\n",
2211 				op, csts);
2212 			return -ENODEV;
2213 		}
2214 	}
2215 
2216 	return ret;
2217 }
2218 
2219 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown)
2220 {
2221 	int ret;
2222 
2223 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2224 	if (shutdown)
2225 		ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2226 	else
2227 		ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2228 
2229 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2230 	if (ret)
2231 		return ret;
2232 
2233 	if (shutdown) {
2234 		return nvme_wait_ready(ctrl, NVME_CSTS_SHST_MASK,
2235 				       NVME_CSTS_SHST_CMPLT,
2236 				       ctrl->shutdown_timeout, "shutdown");
2237 	}
2238 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2239 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2240 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, 0,
2241 			       (NVME_CAP_TIMEOUT(ctrl->cap) + 1) / 2, "reset");
2242 }
2243 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2244 
2245 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2246 {
2247 	unsigned dev_page_min;
2248 	u32 timeout;
2249 	int ret;
2250 
2251 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2252 	if (ret) {
2253 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2254 		return ret;
2255 	}
2256 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2257 
2258 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2259 		dev_err(ctrl->device,
2260 			"Minimum device page size %u too large for host (%u)\n",
2261 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2262 		return -ENODEV;
2263 	}
2264 
2265 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2266 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2267 	else
2268 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2269 
2270 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS && ctrl->cap & NVME_CAP_CRMS_CRIMS)
2271 		ctrl->ctrl_config |= NVME_CC_CRIME;
2272 
2273 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2274 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2275 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2276 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2277 	if (ret)
2278 		return ret;
2279 
2280 	/* Flush write to device (required if transport is PCI) */
2281 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2282 	if (ret)
2283 		return ret;
2284 
2285 	/* CAP value may change after initial CC write */
2286 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2287 	if (ret)
2288 		return ret;
2289 
2290 	timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2291 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2292 		u32 crto, ready_timeout;
2293 
2294 		ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2295 		if (ret) {
2296 			dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2297 				ret);
2298 			return ret;
2299 		}
2300 
2301 		/*
2302 		 * CRTO should always be greater or equal to CAP.TO, but some
2303 		 * devices are known to get this wrong. Use the larger of the
2304 		 * two values.
2305 		 */
2306 		if (ctrl->ctrl_config & NVME_CC_CRIME)
2307 			ready_timeout = NVME_CRTO_CRIMT(crto);
2308 		else
2309 			ready_timeout = NVME_CRTO_CRWMT(crto);
2310 
2311 		if (ready_timeout < timeout)
2312 			dev_warn_once(ctrl->device, "bad crto:%x cap:%llx\n",
2313 				      crto, ctrl->cap);
2314 		else
2315 			timeout = ready_timeout;
2316 	}
2317 
2318 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2319 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2320 	if (ret)
2321 		return ret;
2322 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, NVME_CSTS_RDY,
2323 			       (timeout + 1) / 2, "initialisation");
2324 }
2325 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2326 
2327 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2328 {
2329 	__le64 ts;
2330 	int ret;
2331 
2332 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2333 		return 0;
2334 
2335 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2336 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2337 			NULL);
2338 	if (ret)
2339 		dev_warn_once(ctrl->device,
2340 			"could not set timestamp (%d)\n", ret);
2341 	return ret;
2342 }
2343 
2344 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2345 {
2346 	struct nvme_feat_host_behavior *host;
2347 	u8 acre = 0, lbafee = 0;
2348 	int ret;
2349 
2350 	/* Don't bother enabling the feature if retry delay is not reported */
2351 	if (ctrl->crdt[0])
2352 		acre = NVME_ENABLE_ACRE;
2353 	if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2354 		lbafee = NVME_ENABLE_LBAFEE;
2355 
2356 	if (!acre && !lbafee)
2357 		return 0;
2358 
2359 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2360 	if (!host)
2361 		return 0;
2362 
2363 	host->acre = acre;
2364 	host->lbafee = lbafee;
2365 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2366 				host, sizeof(*host), NULL);
2367 	kfree(host);
2368 	return ret;
2369 }
2370 
2371 /*
2372  * The function checks whether the given total (exlat + enlat) latency of
2373  * a power state allows the latter to be used as an APST transition target.
2374  * It does so by comparing the latency to the primary and secondary latency
2375  * tolerances defined by module params. If there's a match, the corresponding
2376  * timeout value is returned and the matching tolerance index (1 or 2) is
2377  * reported.
2378  */
2379 static bool nvme_apst_get_transition_time(u64 total_latency,
2380 		u64 *transition_time, unsigned *last_index)
2381 {
2382 	if (total_latency <= apst_primary_latency_tol_us) {
2383 		if (*last_index == 1)
2384 			return false;
2385 		*last_index = 1;
2386 		*transition_time = apst_primary_timeout_ms;
2387 		return true;
2388 	}
2389 	if (apst_secondary_timeout_ms &&
2390 		total_latency <= apst_secondary_latency_tol_us) {
2391 		if (*last_index <= 2)
2392 			return false;
2393 		*last_index = 2;
2394 		*transition_time = apst_secondary_timeout_ms;
2395 		return true;
2396 	}
2397 	return false;
2398 }
2399 
2400 /*
2401  * APST (Autonomous Power State Transition) lets us program a table of power
2402  * state transitions that the controller will perform automatically.
2403  *
2404  * Depending on module params, one of the two supported techniques will be used:
2405  *
2406  * - If the parameters provide explicit timeouts and tolerances, they will be
2407  *   used to build a table with up to 2 non-operational states to transition to.
2408  *   The default parameter values were selected based on the values used by
2409  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2410  *   regeneration of the APST table in the event of switching between external
2411  *   and battery power, the timeouts and tolerances reflect a compromise
2412  *   between values used by Microsoft for AC and battery scenarios.
2413  * - If not, we'll configure the table with a simple heuristic: we are willing
2414  *   to spend at most 2% of the time transitioning between power states.
2415  *   Therefore, when running in any given state, we will enter the next
2416  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2417  *   microseconds, as long as that state's exit latency is under the requested
2418  *   maximum latency.
2419  *
2420  * We will not autonomously enter any non-operational state for which the total
2421  * latency exceeds ps_max_latency_us.
2422  *
2423  * Users can set ps_max_latency_us to zero to turn off APST.
2424  */
2425 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2426 {
2427 	struct nvme_feat_auto_pst *table;
2428 	unsigned apste = 0;
2429 	u64 max_lat_us = 0;
2430 	__le64 target = 0;
2431 	int max_ps = -1;
2432 	int state;
2433 	int ret;
2434 	unsigned last_lt_index = UINT_MAX;
2435 
2436 	/*
2437 	 * If APST isn't supported or if we haven't been initialized yet,
2438 	 * then don't do anything.
2439 	 */
2440 	if (!ctrl->apsta)
2441 		return 0;
2442 
2443 	if (ctrl->npss > 31) {
2444 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2445 		return 0;
2446 	}
2447 
2448 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2449 	if (!table)
2450 		return 0;
2451 
2452 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2453 		/* Turn off APST. */
2454 		dev_dbg(ctrl->device, "APST disabled\n");
2455 		goto done;
2456 	}
2457 
2458 	/*
2459 	 * Walk through all states from lowest- to highest-power.
2460 	 * According to the spec, lower-numbered states use more power.  NPSS,
2461 	 * despite the name, is the index of the lowest-power state, not the
2462 	 * number of states.
2463 	 */
2464 	for (state = (int)ctrl->npss; state >= 0; state--) {
2465 		u64 total_latency_us, exit_latency_us, transition_ms;
2466 
2467 		if (target)
2468 			table->entries[state] = target;
2469 
2470 		/*
2471 		 * Don't allow transitions to the deepest state if it's quirked
2472 		 * off.
2473 		 */
2474 		if (state == ctrl->npss &&
2475 		    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2476 			continue;
2477 
2478 		/*
2479 		 * Is this state a useful non-operational state for higher-power
2480 		 * states to autonomously transition to?
2481 		 */
2482 		if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2483 			continue;
2484 
2485 		exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2486 		if (exit_latency_us > ctrl->ps_max_latency_us)
2487 			continue;
2488 
2489 		total_latency_us = exit_latency_us +
2490 			le32_to_cpu(ctrl->psd[state].entry_lat);
2491 
2492 		/*
2493 		 * This state is good. It can be used as the APST idle target
2494 		 * for higher power states.
2495 		 */
2496 		if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2497 			if (!nvme_apst_get_transition_time(total_latency_us,
2498 					&transition_ms, &last_lt_index))
2499 				continue;
2500 		} else {
2501 			transition_ms = total_latency_us + 19;
2502 			do_div(transition_ms, 20);
2503 			if (transition_ms > (1 << 24) - 1)
2504 				transition_ms = (1 << 24) - 1;
2505 		}
2506 
2507 		target = cpu_to_le64((state << 3) | (transition_ms << 8));
2508 		if (max_ps == -1)
2509 			max_ps = state;
2510 		if (total_latency_us > max_lat_us)
2511 			max_lat_us = total_latency_us;
2512 	}
2513 
2514 	if (max_ps == -1)
2515 		dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2516 	else
2517 		dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2518 			max_ps, max_lat_us, (int)sizeof(*table), table);
2519 	apste = 1;
2520 
2521 done:
2522 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2523 				table, sizeof(*table), NULL);
2524 	if (ret)
2525 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2526 	kfree(table);
2527 	return ret;
2528 }
2529 
2530 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2531 {
2532 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2533 	u64 latency;
2534 
2535 	switch (val) {
2536 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2537 	case PM_QOS_LATENCY_ANY:
2538 		latency = U64_MAX;
2539 		break;
2540 
2541 	default:
2542 		latency = val;
2543 	}
2544 
2545 	if (ctrl->ps_max_latency_us != latency) {
2546 		ctrl->ps_max_latency_us = latency;
2547 		if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
2548 			nvme_configure_apst(ctrl);
2549 	}
2550 }
2551 
2552 struct nvme_core_quirk_entry {
2553 	/*
2554 	 * NVMe model and firmware strings are padded with spaces.  For
2555 	 * simplicity, strings in the quirk table are padded with NULLs
2556 	 * instead.
2557 	 */
2558 	u16 vid;
2559 	const char *mn;
2560 	const char *fr;
2561 	unsigned long quirks;
2562 };
2563 
2564 static const struct nvme_core_quirk_entry core_quirks[] = {
2565 	{
2566 		/*
2567 		 * This Toshiba device seems to die using any APST states.  See:
2568 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2569 		 */
2570 		.vid = 0x1179,
2571 		.mn = "THNSF5256GPUK TOSHIBA",
2572 		.quirks = NVME_QUIRK_NO_APST,
2573 	},
2574 	{
2575 		/*
2576 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2577 		 * condition associated with actions related to suspend to idle
2578 		 * LiteON has resolved the problem in future firmware
2579 		 */
2580 		.vid = 0x14a4,
2581 		.fr = "22301111",
2582 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2583 	},
2584 	{
2585 		/*
2586 		 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2587 		 * aborts I/O during any load, but more easily reproducible
2588 		 * with discards (fstrim).
2589 		 *
2590 		 * The device is left in a state where it is also not possible
2591 		 * to use "nvme set-feature" to disable APST, but booting with
2592 		 * nvme_core.default_ps_max_latency=0 works.
2593 		 */
2594 		.vid = 0x1e0f,
2595 		.mn = "KCD6XVUL6T40",
2596 		.quirks = NVME_QUIRK_NO_APST,
2597 	},
2598 	{
2599 		/*
2600 		 * The external Samsung X5 SSD fails initialization without a
2601 		 * delay before checking if it is ready and has a whole set of
2602 		 * other problems.  To make this even more interesting, it
2603 		 * shares the PCI ID with internal Samsung 970 Evo Plus that
2604 		 * does not need or want these quirks.
2605 		 */
2606 		.vid = 0x144d,
2607 		.mn = "Samsung Portable SSD X5",
2608 		.quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2609 			  NVME_QUIRK_NO_DEEPEST_PS |
2610 			  NVME_QUIRK_IGNORE_DEV_SUBNQN,
2611 	}
2612 };
2613 
2614 /* match is null-terminated but idstr is space-padded. */
2615 static bool string_matches(const char *idstr, const char *match, size_t len)
2616 {
2617 	size_t matchlen;
2618 
2619 	if (!match)
2620 		return true;
2621 
2622 	matchlen = strlen(match);
2623 	WARN_ON_ONCE(matchlen > len);
2624 
2625 	if (memcmp(idstr, match, matchlen))
2626 		return false;
2627 
2628 	for (; matchlen < len; matchlen++)
2629 		if (idstr[matchlen] != ' ')
2630 			return false;
2631 
2632 	return true;
2633 }
2634 
2635 static bool quirk_matches(const struct nvme_id_ctrl *id,
2636 			  const struct nvme_core_quirk_entry *q)
2637 {
2638 	return q->vid == le16_to_cpu(id->vid) &&
2639 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2640 		string_matches(id->fr, q->fr, sizeof(id->fr));
2641 }
2642 
2643 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2644 		struct nvme_id_ctrl *id)
2645 {
2646 	size_t nqnlen;
2647 	int off;
2648 
2649 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2650 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2651 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2652 			strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2653 			return;
2654 		}
2655 
2656 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2657 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2658 	}
2659 
2660 	/*
2661 	 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2662 	 * Base Specification 2.0.  It is slightly different from the format
2663 	 * specified there due to historic reasons, and we can't change it now.
2664 	 */
2665 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2666 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2667 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2668 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2669 	off += sizeof(id->sn);
2670 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2671 	off += sizeof(id->mn);
2672 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2673 }
2674 
2675 static void nvme_release_subsystem(struct device *dev)
2676 {
2677 	struct nvme_subsystem *subsys =
2678 		container_of(dev, struct nvme_subsystem, dev);
2679 
2680 	if (subsys->instance >= 0)
2681 		ida_free(&nvme_instance_ida, subsys->instance);
2682 	kfree(subsys);
2683 }
2684 
2685 static void nvme_destroy_subsystem(struct kref *ref)
2686 {
2687 	struct nvme_subsystem *subsys =
2688 			container_of(ref, struct nvme_subsystem, ref);
2689 
2690 	mutex_lock(&nvme_subsystems_lock);
2691 	list_del(&subsys->entry);
2692 	mutex_unlock(&nvme_subsystems_lock);
2693 
2694 	ida_destroy(&subsys->ns_ida);
2695 	device_del(&subsys->dev);
2696 	put_device(&subsys->dev);
2697 }
2698 
2699 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2700 {
2701 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2702 }
2703 
2704 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2705 {
2706 	struct nvme_subsystem *subsys;
2707 
2708 	lockdep_assert_held(&nvme_subsystems_lock);
2709 
2710 	/*
2711 	 * Fail matches for discovery subsystems. This results
2712 	 * in each discovery controller bound to a unique subsystem.
2713 	 * This avoids issues with validating controller values
2714 	 * that can only be true when there is a single unique subsystem.
2715 	 * There may be multiple and completely independent entities
2716 	 * that provide discovery controllers.
2717 	 */
2718 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2719 		return NULL;
2720 
2721 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2722 		if (strcmp(subsys->subnqn, subsysnqn))
2723 			continue;
2724 		if (!kref_get_unless_zero(&subsys->ref))
2725 			continue;
2726 		return subsys;
2727 	}
2728 
2729 	return NULL;
2730 }
2731 
2732 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2733 {
2734 	return ctrl->opts && ctrl->opts->discovery_nqn;
2735 }
2736 
2737 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2738 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2739 {
2740 	struct nvme_ctrl *tmp;
2741 
2742 	lockdep_assert_held(&nvme_subsystems_lock);
2743 
2744 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2745 		if (nvme_state_terminal(tmp))
2746 			continue;
2747 
2748 		if (tmp->cntlid == ctrl->cntlid) {
2749 			dev_err(ctrl->device,
2750 				"Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2751 				ctrl->cntlid, dev_name(tmp->device),
2752 				subsys->subnqn);
2753 			return false;
2754 		}
2755 
2756 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2757 		    nvme_discovery_ctrl(ctrl))
2758 			continue;
2759 
2760 		dev_err(ctrl->device,
2761 			"Subsystem does not support multiple controllers\n");
2762 		return false;
2763 	}
2764 
2765 	return true;
2766 }
2767 
2768 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2769 {
2770 	struct nvme_subsystem *subsys, *found;
2771 	int ret;
2772 
2773 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2774 	if (!subsys)
2775 		return -ENOMEM;
2776 
2777 	subsys->instance = -1;
2778 	mutex_init(&subsys->lock);
2779 	kref_init(&subsys->ref);
2780 	INIT_LIST_HEAD(&subsys->ctrls);
2781 	INIT_LIST_HEAD(&subsys->nsheads);
2782 	nvme_init_subnqn(subsys, ctrl, id);
2783 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2784 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2785 	subsys->vendor_id = le16_to_cpu(id->vid);
2786 	subsys->cmic = id->cmic;
2787 
2788 	/* Versions prior to 1.4 don't necessarily report a valid type */
2789 	if (id->cntrltype == NVME_CTRL_DISC ||
2790 	    !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2791 		subsys->subtype = NVME_NQN_DISC;
2792 	else
2793 		subsys->subtype = NVME_NQN_NVME;
2794 
2795 	if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2796 		dev_err(ctrl->device,
2797 			"Subsystem %s is not a discovery controller",
2798 			subsys->subnqn);
2799 		kfree(subsys);
2800 		return -EINVAL;
2801 	}
2802 	subsys->awupf = le16_to_cpu(id->awupf);
2803 	nvme_mpath_default_iopolicy(subsys);
2804 
2805 	subsys->dev.class = nvme_subsys_class;
2806 	subsys->dev.release = nvme_release_subsystem;
2807 	subsys->dev.groups = nvme_subsys_attrs_groups;
2808 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2809 	device_initialize(&subsys->dev);
2810 
2811 	mutex_lock(&nvme_subsystems_lock);
2812 	found = __nvme_find_get_subsystem(subsys->subnqn);
2813 	if (found) {
2814 		put_device(&subsys->dev);
2815 		subsys = found;
2816 
2817 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2818 			ret = -EINVAL;
2819 			goto out_put_subsystem;
2820 		}
2821 	} else {
2822 		ret = device_add(&subsys->dev);
2823 		if (ret) {
2824 			dev_err(ctrl->device,
2825 				"failed to register subsystem device.\n");
2826 			put_device(&subsys->dev);
2827 			goto out_unlock;
2828 		}
2829 		ida_init(&subsys->ns_ida);
2830 		list_add_tail(&subsys->entry, &nvme_subsystems);
2831 	}
2832 
2833 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2834 				dev_name(ctrl->device));
2835 	if (ret) {
2836 		dev_err(ctrl->device,
2837 			"failed to create sysfs link from subsystem.\n");
2838 		goto out_put_subsystem;
2839 	}
2840 
2841 	if (!found)
2842 		subsys->instance = ctrl->instance;
2843 	ctrl->subsys = subsys;
2844 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2845 	mutex_unlock(&nvme_subsystems_lock);
2846 	return 0;
2847 
2848 out_put_subsystem:
2849 	nvme_put_subsystem(subsys);
2850 out_unlock:
2851 	mutex_unlock(&nvme_subsystems_lock);
2852 	return ret;
2853 }
2854 
2855 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2856 		void *log, size_t size, u64 offset)
2857 {
2858 	struct nvme_command c = { };
2859 	u32 dwlen = nvme_bytes_to_numd(size);
2860 
2861 	c.get_log_page.opcode = nvme_admin_get_log_page;
2862 	c.get_log_page.nsid = cpu_to_le32(nsid);
2863 	c.get_log_page.lid = log_page;
2864 	c.get_log_page.lsp = lsp;
2865 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2866 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2867 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2868 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2869 	c.get_log_page.csi = csi;
2870 
2871 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2872 }
2873 
2874 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2875 				struct nvme_effects_log **log)
2876 {
2877 	struct nvme_effects_log	*cel = xa_load(&ctrl->cels, csi);
2878 	int ret;
2879 
2880 	if (cel)
2881 		goto out;
2882 
2883 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2884 	if (!cel)
2885 		return -ENOMEM;
2886 
2887 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2888 			cel, sizeof(*cel), 0);
2889 	if (ret) {
2890 		kfree(cel);
2891 		return ret;
2892 	}
2893 
2894 	xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2895 out:
2896 	*log = cel;
2897 	return 0;
2898 }
2899 
2900 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2901 {
2902 	u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2903 
2904 	if (check_shl_overflow(1U, units + page_shift - 9, &val))
2905 		return UINT_MAX;
2906 	return val;
2907 }
2908 
2909 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2910 {
2911 	struct nvme_command c = { };
2912 	struct nvme_id_ctrl_nvm *id;
2913 	int ret;
2914 
2915 	if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2916 		ctrl->max_discard_sectors = UINT_MAX;
2917 		ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2918 	} else {
2919 		ctrl->max_discard_sectors = 0;
2920 		ctrl->max_discard_segments = 0;
2921 	}
2922 
2923 	/*
2924 	 * Even though NVMe spec explicitly states that MDTS is not applicable
2925 	 * to the write-zeroes, we are cautious and limit the size to the
2926 	 * controllers max_hw_sectors value, which is based on the MDTS field
2927 	 * and possibly other limiting factors.
2928 	 */
2929 	if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2930 	    !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2931 		ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2932 	else
2933 		ctrl->max_zeroes_sectors = 0;
2934 
2935 	if (ctrl->subsys->subtype != NVME_NQN_NVME ||
2936 	    nvme_ctrl_limited_cns(ctrl) ||
2937 	    test_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags))
2938 		return 0;
2939 
2940 	id = kzalloc(sizeof(*id), GFP_KERNEL);
2941 	if (!id)
2942 		return -ENOMEM;
2943 
2944 	c.identify.opcode = nvme_admin_identify;
2945 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
2946 	c.identify.csi = NVME_CSI_NVM;
2947 
2948 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2949 	if (ret)
2950 		goto free_data;
2951 
2952 	if (id->dmrl)
2953 		ctrl->max_discard_segments = id->dmrl;
2954 	ctrl->dmrsl = le32_to_cpu(id->dmrsl);
2955 	if (id->wzsl)
2956 		ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2957 
2958 free_data:
2959 	if (ret > 0)
2960 		set_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags);
2961 	kfree(id);
2962 	return ret;
2963 }
2964 
2965 static void nvme_init_known_nvm_effects(struct nvme_ctrl *ctrl)
2966 {
2967 	struct nvme_effects_log	*log = ctrl->effects;
2968 
2969 	log->acs[nvme_admin_format_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
2970 						NVME_CMD_EFFECTS_NCC |
2971 						NVME_CMD_EFFECTS_CSE_MASK);
2972 	log->acs[nvme_admin_sanitize_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
2973 						NVME_CMD_EFFECTS_CSE_MASK);
2974 
2975 	/*
2976 	 * The spec says the result of a security receive command depends on
2977 	 * the previous security send command. As such, many vendors log this
2978 	 * command as one to submitted only when no other commands to the same
2979 	 * namespace are outstanding. The intention is to tell the host to
2980 	 * prevent mixing security send and receive.
2981 	 *
2982 	 * This driver can only enforce such exclusive access against IO
2983 	 * queues, though. We are not readily able to enforce such a rule for
2984 	 * two commands to the admin queue, which is the only queue that
2985 	 * matters for this command.
2986 	 *
2987 	 * Rather than blindly freezing the IO queues for this effect that
2988 	 * doesn't even apply to IO, mask it off.
2989 	 */
2990 	log->acs[nvme_admin_security_recv] &= cpu_to_le32(~NVME_CMD_EFFECTS_CSE_MASK);
2991 
2992 	log->iocs[nvme_cmd_write] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
2993 	log->iocs[nvme_cmd_write_zeroes] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
2994 	log->iocs[nvme_cmd_write_uncor] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
2995 }
2996 
2997 static int nvme_init_effects(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2998 {
2999 	int ret = 0;
3000 
3001 	if (ctrl->effects)
3002 		return 0;
3003 
3004 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3005 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3006 		if (ret < 0)
3007 			return ret;
3008 	}
3009 
3010 	if (!ctrl->effects) {
3011 		ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
3012 		if (!ctrl->effects)
3013 			return -ENOMEM;
3014 		xa_store(&ctrl->cels, NVME_CSI_NVM, ctrl->effects, GFP_KERNEL);
3015 	}
3016 
3017 	nvme_init_known_nvm_effects(ctrl);
3018 	return 0;
3019 }
3020 
3021 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3022 {
3023 	struct nvme_id_ctrl *id;
3024 	u32 max_hw_sectors;
3025 	bool prev_apst_enabled;
3026 	int ret;
3027 
3028 	ret = nvme_identify_ctrl(ctrl, &id);
3029 	if (ret) {
3030 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3031 		return -EIO;
3032 	}
3033 
3034 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
3035 		ctrl->cntlid = le16_to_cpu(id->cntlid);
3036 
3037 	if (!ctrl->identified) {
3038 		unsigned int i;
3039 
3040 		/*
3041 		 * Check for quirks.  Quirk can depend on firmware version,
3042 		 * so, in principle, the set of quirks present can change
3043 		 * across a reset.  As a possible future enhancement, we
3044 		 * could re-scan for quirks every time we reinitialize
3045 		 * the device, but we'd have to make sure that the driver
3046 		 * behaves intelligently if the quirks change.
3047 		 */
3048 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3049 			if (quirk_matches(id, &core_quirks[i]))
3050 				ctrl->quirks |= core_quirks[i].quirks;
3051 		}
3052 
3053 		ret = nvme_init_subsystem(ctrl, id);
3054 		if (ret)
3055 			goto out_free;
3056 
3057 		ret = nvme_init_effects(ctrl, id);
3058 		if (ret)
3059 			goto out_free;
3060 	}
3061 	memcpy(ctrl->subsys->firmware_rev, id->fr,
3062 	       sizeof(ctrl->subsys->firmware_rev));
3063 
3064 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3065 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3066 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3067 	}
3068 
3069 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3070 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3071 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3072 
3073 	ctrl->oacs = le16_to_cpu(id->oacs);
3074 	ctrl->oncs = le16_to_cpu(id->oncs);
3075 	ctrl->mtfa = le16_to_cpu(id->mtfa);
3076 	ctrl->oaes = le32_to_cpu(id->oaes);
3077 	ctrl->wctemp = le16_to_cpu(id->wctemp);
3078 	ctrl->cctemp = le16_to_cpu(id->cctemp);
3079 
3080 	atomic_set(&ctrl->abort_limit, id->acl + 1);
3081 	ctrl->vwc = id->vwc;
3082 	if (id->mdts)
3083 		max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3084 	else
3085 		max_hw_sectors = UINT_MAX;
3086 	ctrl->max_hw_sectors =
3087 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3088 
3089 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
3090 	ctrl->sgls = le32_to_cpu(id->sgls);
3091 	ctrl->kas = le16_to_cpu(id->kas);
3092 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
3093 	ctrl->ctratt = le32_to_cpu(id->ctratt);
3094 
3095 	ctrl->cntrltype = id->cntrltype;
3096 	ctrl->dctype = id->dctype;
3097 
3098 	if (id->rtd3e) {
3099 		/* us -> s */
3100 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3101 
3102 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3103 						 shutdown_timeout, 60);
3104 
3105 		if (ctrl->shutdown_timeout != shutdown_timeout)
3106 			dev_info(ctrl->device,
3107 				 "Shutdown timeout set to %u seconds\n",
3108 				 ctrl->shutdown_timeout);
3109 	} else
3110 		ctrl->shutdown_timeout = shutdown_timeout;
3111 
3112 	ctrl->npss = id->npss;
3113 	ctrl->apsta = id->apsta;
3114 	prev_apst_enabled = ctrl->apst_enabled;
3115 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3116 		if (force_apst && id->apsta) {
3117 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3118 			ctrl->apst_enabled = true;
3119 		} else {
3120 			ctrl->apst_enabled = false;
3121 		}
3122 	} else {
3123 		ctrl->apst_enabled = id->apsta;
3124 	}
3125 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3126 
3127 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3128 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3129 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3130 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3131 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3132 
3133 		/*
3134 		 * In fabrics we need to verify the cntlid matches the
3135 		 * admin connect
3136 		 */
3137 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3138 			dev_err(ctrl->device,
3139 				"Mismatching cntlid: Connect %u vs Identify "
3140 				"%u, rejecting\n",
3141 				ctrl->cntlid, le16_to_cpu(id->cntlid));
3142 			ret = -EINVAL;
3143 			goto out_free;
3144 		}
3145 
3146 		if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3147 			dev_err(ctrl->device,
3148 				"keep-alive support is mandatory for fabrics\n");
3149 			ret = -EINVAL;
3150 			goto out_free;
3151 		}
3152 	} else {
3153 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3154 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3155 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3156 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3157 	}
3158 
3159 	ret = nvme_mpath_init_identify(ctrl, id);
3160 	if (ret < 0)
3161 		goto out_free;
3162 
3163 	if (ctrl->apst_enabled && !prev_apst_enabled)
3164 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3165 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3166 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3167 
3168 out_free:
3169 	kfree(id);
3170 	return ret;
3171 }
3172 
3173 /*
3174  * Initialize the cached copies of the Identify data and various controller
3175  * register in our nvme_ctrl structure.  This should be called as soon as
3176  * the admin queue is fully up and running.
3177  */
3178 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended)
3179 {
3180 	int ret;
3181 
3182 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3183 	if (ret) {
3184 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3185 		return ret;
3186 	}
3187 
3188 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3189 
3190 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3191 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3192 
3193 	ret = nvme_init_identify(ctrl);
3194 	if (ret)
3195 		return ret;
3196 
3197 	ret = nvme_configure_apst(ctrl);
3198 	if (ret < 0)
3199 		return ret;
3200 
3201 	ret = nvme_configure_timestamp(ctrl);
3202 	if (ret < 0)
3203 		return ret;
3204 
3205 	ret = nvme_configure_host_options(ctrl);
3206 	if (ret < 0)
3207 		return ret;
3208 
3209 	nvme_configure_opal(ctrl, was_suspended);
3210 
3211 	if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3212 		/*
3213 		 * Do not return errors unless we are in a controller reset,
3214 		 * the controller works perfectly fine without hwmon.
3215 		 */
3216 		ret = nvme_hwmon_init(ctrl);
3217 		if (ret == -EINTR)
3218 			return ret;
3219 	}
3220 
3221 	clear_bit(NVME_CTRL_DIRTY_CAPABILITY, &ctrl->flags);
3222 	ctrl->identified = true;
3223 
3224 	return 0;
3225 }
3226 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3227 
3228 static int nvme_dev_open(struct inode *inode, struct file *file)
3229 {
3230 	struct nvme_ctrl *ctrl =
3231 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3232 
3233 	switch (nvme_ctrl_state(ctrl)) {
3234 	case NVME_CTRL_LIVE:
3235 		break;
3236 	default:
3237 		return -EWOULDBLOCK;
3238 	}
3239 
3240 	nvme_get_ctrl(ctrl);
3241 	if (!try_module_get(ctrl->ops->module)) {
3242 		nvme_put_ctrl(ctrl);
3243 		return -EINVAL;
3244 	}
3245 
3246 	file->private_data = ctrl;
3247 	return 0;
3248 }
3249 
3250 static int nvme_dev_release(struct inode *inode, struct file *file)
3251 {
3252 	struct nvme_ctrl *ctrl =
3253 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3254 
3255 	module_put(ctrl->ops->module);
3256 	nvme_put_ctrl(ctrl);
3257 	return 0;
3258 }
3259 
3260 static const struct file_operations nvme_dev_fops = {
3261 	.owner		= THIS_MODULE,
3262 	.open		= nvme_dev_open,
3263 	.release	= nvme_dev_release,
3264 	.unlocked_ioctl	= nvme_dev_ioctl,
3265 	.compat_ioctl	= compat_ptr_ioctl,
3266 	.uring_cmd	= nvme_dev_uring_cmd,
3267 };
3268 
3269 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3270 		unsigned nsid)
3271 {
3272 	struct nvme_ns_head *h;
3273 
3274 	lockdep_assert_held(&ctrl->subsys->lock);
3275 
3276 	list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3277 		/*
3278 		 * Private namespaces can share NSIDs under some conditions.
3279 		 * In that case we can't use the same ns_head for namespaces
3280 		 * with the same NSID.
3281 		 */
3282 		if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3283 			continue;
3284 		if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3285 			return h;
3286 	}
3287 
3288 	return NULL;
3289 }
3290 
3291 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3292 		struct nvme_ns_ids *ids)
3293 {
3294 	bool has_uuid = !uuid_is_null(&ids->uuid);
3295 	bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3296 	bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3297 	struct nvme_ns_head *h;
3298 
3299 	lockdep_assert_held(&subsys->lock);
3300 
3301 	list_for_each_entry(h, &subsys->nsheads, entry) {
3302 		if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3303 			return -EINVAL;
3304 		if (has_nguid &&
3305 		    memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3306 			return -EINVAL;
3307 		if (has_eui64 &&
3308 		    memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3309 			return -EINVAL;
3310 	}
3311 
3312 	return 0;
3313 }
3314 
3315 static void nvme_cdev_rel(struct device *dev)
3316 {
3317 	ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3318 }
3319 
3320 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3321 {
3322 	cdev_device_del(cdev, cdev_device);
3323 	put_device(cdev_device);
3324 }
3325 
3326 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3327 		const struct file_operations *fops, struct module *owner)
3328 {
3329 	int minor, ret;
3330 
3331 	minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3332 	if (minor < 0)
3333 		return minor;
3334 	cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3335 	cdev_device->class = nvme_ns_chr_class;
3336 	cdev_device->release = nvme_cdev_rel;
3337 	device_initialize(cdev_device);
3338 	cdev_init(cdev, fops);
3339 	cdev->owner = owner;
3340 	ret = cdev_device_add(cdev, cdev_device);
3341 	if (ret)
3342 		put_device(cdev_device);
3343 
3344 	return ret;
3345 }
3346 
3347 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3348 {
3349 	return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3350 }
3351 
3352 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3353 {
3354 	nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3355 	return 0;
3356 }
3357 
3358 static const struct file_operations nvme_ns_chr_fops = {
3359 	.owner		= THIS_MODULE,
3360 	.open		= nvme_ns_chr_open,
3361 	.release	= nvme_ns_chr_release,
3362 	.unlocked_ioctl	= nvme_ns_chr_ioctl,
3363 	.compat_ioctl	= compat_ptr_ioctl,
3364 	.uring_cmd	= nvme_ns_chr_uring_cmd,
3365 	.uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
3366 };
3367 
3368 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3369 {
3370 	int ret;
3371 
3372 	ns->cdev_device.parent = ns->ctrl->device;
3373 	ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3374 			   ns->ctrl->instance, ns->head->instance);
3375 	if (ret)
3376 		return ret;
3377 
3378 	return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3379 			     ns->ctrl->ops->module);
3380 }
3381 
3382 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3383 		struct nvme_ns_info *info)
3384 {
3385 	struct nvme_ns_head *head;
3386 	size_t size = sizeof(*head);
3387 	int ret = -ENOMEM;
3388 
3389 #ifdef CONFIG_NVME_MULTIPATH
3390 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3391 #endif
3392 
3393 	head = kzalloc(size, GFP_KERNEL);
3394 	if (!head)
3395 		goto out;
3396 	ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3397 	if (ret < 0)
3398 		goto out_free_head;
3399 	head->instance = ret;
3400 	INIT_LIST_HEAD(&head->list);
3401 	ret = init_srcu_struct(&head->srcu);
3402 	if (ret)
3403 		goto out_ida_remove;
3404 	head->subsys = ctrl->subsys;
3405 	head->ns_id = info->nsid;
3406 	head->ids = info->ids;
3407 	head->shared = info->is_shared;
3408 	kref_init(&head->ref);
3409 
3410 	if (head->ids.csi) {
3411 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3412 		if (ret)
3413 			goto out_cleanup_srcu;
3414 	} else
3415 		head->effects = ctrl->effects;
3416 
3417 	ret = nvme_mpath_alloc_disk(ctrl, head);
3418 	if (ret)
3419 		goto out_cleanup_srcu;
3420 
3421 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3422 
3423 	kref_get(&ctrl->subsys->ref);
3424 
3425 	return head;
3426 out_cleanup_srcu:
3427 	cleanup_srcu_struct(&head->srcu);
3428 out_ida_remove:
3429 	ida_free(&ctrl->subsys->ns_ida, head->instance);
3430 out_free_head:
3431 	kfree(head);
3432 out:
3433 	if (ret > 0)
3434 		ret = blk_status_to_errno(nvme_error_status(ret));
3435 	return ERR_PTR(ret);
3436 }
3437 
3438 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3439 		struct nvme_ns_ids *ids)
3440 {
3441 	struct nvme_subsystem *s;
3442 	int ret = 0;
3443 
3444 	/*
3445 	 * Note that this check is racy as we try to avoid holding the global
3446 	 * lock over the whole ns_head creation.  But it is only intended as
3447 	 * a sanity check anyway.
3448 	 */
3449 	mutex_lock(&nvme_subsystems_lock);
3450 	list_for_each_entry(s, &nvme_subsystems, entry) {
3451 		if (s == this)
3452 			continue;
3453 		mutex_lock(&s->lock);
3454 		ret = nvme_subsys_check_duplicate_ids(s, ids);
3455 		mutex_unlock(&s->lock);
3456 		if (ret)
3457 			break;
3458 	}
3459 	mutex_unlock(&nvme_subsystems_lock);
3460 
3461 	return ret;
3462 }
3463 
3464 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
3465 {
3466 	struct nvme_ctrl *ctrl = ns->ctrl;
3467 	struct nvme_ns_head *head = NULL;
3468 	int ret;
3469 
3470 	ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
3471 	if (ret) {
3472 		/*
3473 		 * We've found two different namespaces on two different
3474 		 * subsystems that report the same ID.  This is pretty nasty
3475 		 * for anything that actually requires unique device
3476 		 * identification.  In the kernel we need this for multipathing,
3477 		 * and in user space the /dev/disk/by-id/ links rely on it.
3478 		 *
3479 		 * If the device also claims to be multi-path capable back off
3480 		 * here now and refuse the probe the second device as this is a
3481 		 * recipe for data corruption.  If not this is probably a
3482 		 * cheap consumer device if on the PCIe bus, so let the user
3483 		 * proceed and use the shiny toy, but warn that with changing
3484 		 * probing order (which due to our async probing could just be
3485 		 * device taking longer to startup) the other device could show
3486 		 * up at any time.
3487 		 */
3488 		nvme_print_device_info(ctrl);
3489 		if ((ns->ctrl->ops->flags & NVME_F_FABRICS) || /* !PCIe */
3490 		    ((ns->ctrl->subsys->cmic & NVME_CTRL_CMIC_MULTI_CTRL) &&
3491 		     info->is_shared)) {
3492 			dev_err(ctrl->device,
3493 				"ignoring nsid %d because of duplicate IDs\n",
3494 				info->nsid);
3495 			return ret;
3496 		}
3497 
3498 		dev_err(ctrl->device,
3499 			"clearing duplicate IDs for nsid %d\n", info->nsid);
3500 		dev_err(ctrl->device,
3501 			"use of /dev/disk/by-id/ may cause data corruption\n");
3502 		memset(&info->ids.nguid, 0, sizeof(info->ids.nguid));
3503 		memset(&info->ids.uuid, 0, sizeof(info->ids.uuid));
3504 		memset(&info->ids.eui64, 0, sizeof(info->ids.eui64));
3505 		ctrl->quirks |= NVME_QUIRK_BOGUS_NID;
3506 	}
3507 
3508 	mutex_lock(&ctrl->subsys->lock);
3509 	head = nvme_find_ns_head(ctrl, info->nsid);
3510 	if (!head) {
3511 		ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
3512 		if (ret) {
3513 			dev_err(ctrl->device,
3514 				"duplicate IDs in subsystem for nsid %d\n",
3515 				info->nsid);
3516 			goto out_unlock;
3517 		}
3518 		head = nvme_alloc_ns_head(ctrl, info);
3519 		if (IS_ERR(head)) {
3520 			ret = PTR_ERR(head);
3521 			goto out_unlock;
3522 		}
3523 	} else {
3524 		ret = -EINVAL;
3525 		if (!info->is_shared || !head->shared) {
3526 			dev_err(ctrl->device,
3527 				"Duplicate unshared namespace %d\n",
3528 				info->nsid);
3529 			goto out_put_ns_head;
3530 		}
3531 		if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
3532 			dev_err(ctrl->device,
3533 				"IDs don't match for shared namespace %d\n",
3534 					info->nsid);
3535 			goto out_put_ns_head;
3536 		}
3537 
3538 		if (!multipath) {
3539 			dev_warn(ctrl->device,
3540 				"Found shared namespace %d, but multipathing not supported.\n",
3541 				info->nsid);
3542 			dev_warn_once(ctrl->device,
3543 				"Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0.\n");
3544 		}
3545 	}
3546 
3547 	list_add_tail_rcu(&ns->siblings, &head->list);
3548 	ns->head = head;
3549 	mutex_unlock(&ctrl->subsys->lock);
3550 	return 0;
3551 
3552 out_put_ns_head:
3553 	nvme_put_ns_head(head);
3554 out_unlock:
3555 	mutex_unlock(&ctrl->subsys->lock);
3556 	return ret;
3557 }
3558 
3559 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3560 {
3561 	struct nvme_ns *ns, *ret = NULL;
3562 
3563 	down_read(&ctrl->namespaces_rwsem);
3564 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3565 		if (ns->head->ns_id == nsid) {
3566 			if (!nvme_get_ns(ns))
3567 				continue;
3568 			ret = ns;
3569 			break;
3570 		}
3571 		if (ns->head->ns_id > nsid)
3572 			break;
3573 	}
3574 	up_read(&ctrl->namespaces_rwsem);
3575 	return ret;
3576 }
3577 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3578 
3579 /*
3580  * Add the namespace to the controller list while keeping the list ordered.
3581  */
3582 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3583 {
3584 	struct nvme_ns *tmp;
3585 
3586 	list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3587 		if (tmp->head->ns_id < ns->head->ns_id) {
3588 			list_add(&ns->list, &tmp->list);
3589 			return;
3590 		}
3591 	}
3592 	list_add(&ns->list, &ns->ctrl->namespaces);
3593 }
3594 
3595 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
3596 {
3597 	struct nvme_ns *ns;
3598 	struct gendisk *disk;
3599 	int node = ctrl->numa_node;
3600 
3601 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3602 	if (!ns)
3603 		return;
3604 
3605 	disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3606 	if (IS_ERR(disk))
3607 		goto out_free_ns;
3608 	disk->fops = &nvme_bdev_ops;
3609 	disk->private_data = ns;
3610 
3611 	ns->disk = disk;
3612 	ns->queue = disk->queue;
3613 
3614 	if (ctrl->opts && ctrl->opts->data_digest)
3615 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3616 
3617 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3618 	if (ctrl->ops->supports_pci_p2pdma &&
3619 	    ctrl->ops->supports_pci_p2pdma(ctrl))
3620 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3621 
3622 	ns->ctrl = ctrl;
3623 	kref_init(&ns->kref);
3624 
3625 	if (nvme_init_ns_head(ns, info))
3626 		goto out_cleanup_disk;
3627 
3628 	/*
3629 	 * If multipathing is enabled, the device name for all disks and not
3630 	 * just those that represent shared namespaces needs to be based on the
3631 	 * subsystem instance.  Using the controller instance for private
3632 	 * namespaces could lead to naming collisions between shared and private
3633 	 * namespaces if they don't use a common numbering scheme.
3634 	 *
3635 	 * If multipathing is not enabled, disk names must use the controller
3636 	 * instance as shared namespaces will show up as multiple block
3637 	 * devices.
3638 	 */
3639 	if (nvme_ns_head_multipath(ns->head)) {
3640 		sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
3641 			ctrl->instance, ns->head->instance);
3642 		disk->flags |= GENHD_FL_HIDDEN;
3643 	} else if (multipath) {
3644 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
3645 			ns->head->instance);
3646 	} else {
3647 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3648 			ns->head->instance);
3649 	}
3650 
3651 	if (nvme_update_ns_info(ns, info))
3652 		goto out_unlink_ns;
3653 
3654 	down_write(&ctrl->namespaces_rwsem);
3655 	/*
3656 	 * Ensure that no namespaces are added to the ctrl list after the queues
3657 	 * are frozen, thereby avoiding a deadlock between scan and reset.
3658 	 */
3659 	if (test_bit(NVME_CTRL_FROZEN, &ctrl->flags)) {
3660 		up_write(&ctrl->namespaces_rwsem);
3661 		goto out_unlink_ns;
3662 	}
3663 	nvme_ns_add_to_ctrl_list(ns);
3664 	up_write(&ctrl->namespaces_rwsem);
3665 	nvme_get_ctrl(ctrl);
3666 
3667 	if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
3668 		goto out_cleanup_ns_from_list;
3669 
3670 	if (!nvme_ns_head_multipath(ns->head))
3671 		nvme_add_ns_cdev(ns);
3672 
3673 	nvme_mpath_add_disk(ns, info->anagrpid);
3674 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3675 
3676 	return;
3677 
3678  out_cleanup_ns_from_list:
3679 	nvme_put_ctrl(ctrl);
3680 	down_write(&ctrl->namespaces_rwsem);
3681 	list_del_init(&ns->list);
3682 	up_write(&ctrl->namespaces_rwsem);
3683  out_unlink_ns:
3684 	mutex_lock(&ctrl->subsys->lock);
3685 	list_del_rcu(&ns->siblings);
3686 	if (list_empty(&ns->head->list))
3687 		list_del_init(&ns->head->entry);
3688 	mutex_unlock(&ctrl->subsys->lock);
3689 	nvme_put_ns_head(ns->head);
3690  out_cleanup_disk:
3691 	put_disk(disk);
3692  out_free_ns:
3693 	kfree(ns);
3694 }
3695 
3696 static void nvme_ns_remove(struct nvme_ns *ns)
3697 {
3698 	bool last_path = false;
3699 
3700 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3701 		return;
3702 
3703 	clear_bit(NVME_NS_READY, &ns->flags);
3704 	set_capacity(ns->disk, 0);
3705 	nvme_fault_inject_fini(&ns->fault_inject);
3706 
3707 	/*
3708 	 * Ensure that !NVME_NS_READY is seen by other threads to prevent
3709 	 * this ns going back into current_path.
3710 	 */
3711 	synchronize_srcu(&ns->head->srcu);
3712 
3713 	/* wait for concurrent submissions */
3714 	if (nvme_mpath_clear_current_path(ns))
3715 		synchronize_srcu(&ns->head->srcu);
3716 
3717 	mutex_lock(&ns->ctrl->subsys->lock);
3718 	list_del_rcu(&ns->siblings);
3719 	if (list_empty(&ns->head->list)) {
3720 		list_del_init(&ns->head->entry);
3721 		last_path = true;
3722 	}
3723 	mutex_unlock(&ns->ctrl->subsys->lock);
3724 
3725 	/* guarantee not available in head->list */
3726 	synchronize_srcu(&ns->head->srcu);
3727 
3728 	if (!nvme_ns_head_multipath(ns->head))
3729 		nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3730 	del_gendisk(ns->disk);
3731 
3732 	down_write(&ns->ctrl->namespaces_rwsem);
3733 	list_del_init(&ns->list);
3734 	up_write(&ns->ctrl->namespaces_rwsem);
3735 
3736 	if (last_path)
3737 		nvme_mpath_shutdown_disk(ns->head);
3738 	nvme_put_ns(ns);
3739 }
3740 
3741 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3742 {
3743 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3744 
3745 	if (ns) {
3746 		nvme_ns_remove(ns);
3747 		nvme_put_ns(ns);
3748 	}
3749 }
3750 
3751 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
3752 {
3753 	int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3754 
3755 	if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
3756 		dev_err(ns->ctrl->device,
3757 			"identifiers changed for nsid %d\n", ns->head->ns_id);
3758 		goto out;
3759 	}
3760 
3761 	ret = nvme_update_ns_info(ns, info);
3762 out:
3763 	/*
3764 	 * Only remove the namespace if we got a fatal error back from the
3765 	 * device, otherwise ignore the error and just move on.
3766 	 *
3767 	 * TODO: we should probably schedule a delayed retry here.
3768 	 */
3769 	if (ret > 0 && (ret & NVME_SC_DNR))
3770 		nvme_ns_remove(ns);
3771 }
3772 
3773 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3774 {
3775 	struct nvme_ns_info info = { .nsid = nsid };
3776 	struct nvme_ns *ns;
3777 	int ret;
3778 
3779 	if (nvme_identify_ns_descs(ctrl, &info))
3780 		return;
3781 
3782 	if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
3783 		dev_warn(ctrl->device,
3784 			"command set not reported for nsid: %d\n", nsid);
3785 		return;
3786 	}
3787 
3788 	/*
3789 	 * If available try to use the Command Set Idependent Identify Namespace
3790 	 * data structure to find all the generic information that is needed to
3791 	 * set up a namespace.  If not fall back to the legacy version.
3792 	 */
3793 	if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
3794 	    (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS))
3795 		ret = nvme_ns_info_from_id_cs_indep(ctrl, &info);
3796 	else
3797 		ret = nvme_ns_info_from_identify(ctrl, &info);
3798 
3799 	if (info.is_removed)
3800 		nvme_ns_remove_by_nsid(ctrl, nsid);
3801 
3802 	/*
3803 	 * Ignore the namespace if it is not ready. We will get an AEN once it
3804 	 * becomes ready and restart the scan.
3805 	 */
3806 	if (ret || !info.is_ready)
3807 		return;
3808 
3809 	ns = nvme_find_get_ns(ctrl, nsid);
3810 	if (ns) {
3811 		nvme_validate_ns(ns, &info);
3812 		nvme_put_ns(ns);
3813 	} else {
3814 		nvme_alloc_ns(ctrl, &info);
3815 	}
3816 }
3817 
3818 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3819 					unsigned nsid)
3820 {
3821 	struct nvme_ns *ns, *next;
3822 	LIST_HEAD(rm_list);
3823 
3824 	down_write(&ctrl->namespaces_rwsem);
3825 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3826 		if (ns->head->ns_id > nsid)
3827 			list_move_tail(&ns->list, &rm_list);
3828 	}
3829 	up_write(&ctrl->namespaces_rwsem);
3830 
3831 	list_for_each_entry_safe(ns, next, &rm_list, list)
3832 		nvme_ns_remove(ns);
3833 
3834 }
3835 
3836 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3837 {
3838 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3839 	__le32 *ns_list;
3840 	u32 prev = 0;
3841 	int ret = 0, i;
3842 
3843 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3844 	if (!ns_list)
3845 		return -ENOMEM;
3846 
3847 	for (;;) {
3848 		struct nvme_command cmd = {
3849 			.identify.opcode	= nvme_admin_identify,
3850 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
3851 			.identify.nsid		= cpu_to_le32(prev),
3852 		};
3853 
3854 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
3855 					    NVME_IDENTIFY_DATA_SIZE);
3856 		if (ret) {
3857 			dev_warn(ctrl->device,
3858 				"Identify NS List failed (status=0x%x)\n", ret);
3859 			goto free;
3860 		}
3861 
3862 		for (i = 0; i < nr_entries; i++) {
3863 			u32 nsid = le32_to_cpu(ns_list[i]);
3864 
3865 			if (!nsid)	/* end of the list? */
3866 				goto out;
3867 			nvme_scan_ns(ctrl, nsid);
3868 			while (++prev < nsid)
3869 				nvme_ns_remove_by_nsid(ctrl, prev);
3870 		}
3871 	}
3872  out:
3873 	nvme_remove_invalid_namespaces(ctrl, prev);
3874  free:
3875 	kfree(ns_list);
3876 	return ret;
3877 }
3878 
3879 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3880 {
3881 	struct nvme_id_ctrl *id;
3882 	u32 nn, i;
3883 
3884 	if (nvme_identify_ctrl(ctrl, &id))
3885 		return;
3886 	nn = le32_to_cpu(id->nn);
3887 	kfree(id);
3888 
3889 	for (i = 1; i <= nn; i++)
3890 		nvme_scan_ns(ctrl, i);
3891 
3892 	nvme_remove_invalid_namespaces(ctrl, nn);
3893 }
3894 
3895 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3896 {
3897 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3898 	__le32 *log;
3899 	int error;
3900 
3901 	log = kzalloc(log_size, GFP_KERNEL);
3902 	if (!log)
3903 		return;
3904 
3905 	/*
3906 	 * We need to read the log to clear the AEN, but we don't want to rely
3907 	 * on it for the changed namespace information as userspace could have
3908 	 * raced with us in reading the log page, which could cause us to miss
3909 	 * updates.
3910 	 */
3911 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
3912 			NVME_CSI_NVM, log, log_size, 0);
3913 	if (error)
3914 		dev_warn(ctrl->device,
3915 			"reading changed ns log failed: %d\n", error);
3916 
3917 	kfree(log);
3918 }
3919 
3920 static void nvme_scan_work(struct work_struct *work)
3921 {
3922 	struct nvme_ctrl *ctrl =
3923 		container_of(work, struct nvme_ctrl, scan_work);
3924 	int ret;
3925 
3926 	/* No tagset on a live ctrl means IO queues could not created */
3927 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE || !ctrl->tagset)
3928 		return;
3929 
3930 	/*
3931 	 * Identify controller limits can change at controller reset due to
3932 	 * new firmware download, even though it is not common we cannot ignore
3933 	 * such scenario. Controller's non-mdts limits are reported in the unit
3934 	 * of logical blocks that is dependent on the format of attached
3935 	 * namespace. Hence re-read the limits at the time of ns allocation.
3936 	 */
3937 	ret = nvme_init_non_mdts_limits(ctrl);
3938 	if (ret < 0) {
3939 		dev_warn(ctrl->device,
3940 			"reading non-mdts-limits failed: %d\n", ret);
3941 		return;
3942 	}
3943 
3944 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3945 		dev_info(ctrl->device, "rescanning namespaces.\n");
3946 		nvme_clear_changed_ns_log(ctrl);
3947 	}
3948 
3949 	mutex_lock(&ctrl->scan_lock);
3950 	if (nvme_ctrl_limited_cns(ctrl)) {
3951 		nvme_scan_ns_sequential(ctrl);
3952 	} else {
3953 		/*
3954 		 * Fall back to sequential scan if DNR is set to handle broken
3955 		 * devices which should support Identify NS List (as per the VS
3956 		 * they report) but don't actually support it.
3957 		 */
3958 		ret = nvme_scan_ns_list(ctrl);
3959 		if (ret > 0 && ret & NVME_SC_DNR)
3960 			nvme_scan_ns_sequential(ctrl);
3961 	}
3962 	mutex_unlock(&ctrl->scan_lock);
3963 }
3964 
3965 /*
3966  * This function iterates the namespace list unlocked to allow recovery from
3967  * controller failure. It is up to the caller to ensure the namespace list is
3968  * not modified by scan work while this function is executing.
3969  */
3970 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3971 {
3972 	struct nvme_ns *ns, *next;
3973 	LIST_HEAD(ns_list);
3974 
3975 	/*
3976 	 * make sure to requeue I/O to all namespaces as these
3977 	 * might result from the scan itself and must complete
3978 	 * for the scan_work to make progress
3979 	 */
3980 	nvme_mpath_clear_ctrl_paths(ctrl);
3981 
3982 	/*
3983 	 * Unquiesce io queues so any pending IO won't hang, especially
3984 	 * those submitted from scan work
3985 	 */
3986 	nvme_unquiesce_io_queues(ctrl);
3987 
3988 	/* prevent racing with ns scanning */
3989 	flush_work(&ctrl->scan_work);
3990 
3991 	/*
3992 	 * The dead states indicates the controller was not gracefully
3993 	 * disconnected. In that case, we won't be able to flush any data while
3994 	 * removing the namespaces' disks; fail all the queues now to avoid
3995 	 * potentially having to clean up the failed sync later.
3996 	 */
3997 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_DEAD)
3998 		nvme_mark_namespaces_dead(ctrl);
3999 
4000 	/* this is a no-op when called from the controller reset handler */
4001 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4002 
4003 	down_write(&ctrl->namespaces_rwsem);
4004 	list_splice_init(&ctrl->namespaces, &ns_list);
4005 	up_write(&ctrl->namespaces_rwsem);
4006 
4007 	list_for_each_entry_safe(ns, next, &ns_list, list)
4008 		nvme_ns_remove(ns);
4009 }
4010 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4011 
4012 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env)
4013 {
4014 	const struct nvme_ctrl *ctrl =
4015 		container_of(dev, struct nvme_ctrl, ctrl_device);
4016 	struct nvmf_ctrl_options *opts = ctrl->opts;
4017 	int ret;
4018 
4019 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4020 	if (ret)
4021 		return ret;
4022 
4023 	if (opts) {
4024 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4025 		if (ret)
4026 			return ret;
4027 
4028 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4029 				opts->trsvcid ?: "none");
4030 		if (ret)
4031 			return ret;
4032 
4033 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4034 				opts->host_traddr ?: "none");
4035 		if (ret)
4036 			return ret;
4037 
4038 		ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4039 				opts->host_iface ?: "none");
4040 	}
4041 	return ret;
4042 }
4043 
4044 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4045 {
4046 	char *envp[2] = { envdata, NULL };
4047 
4048 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4049 }
4050 
4051 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4052 {
4053 	char *envp[2] = { NULL, NULL };
4054 	u32 aen_result = ctrl->aen_result;
4055 
4056 	ctrl->aen_result = 0;
4057 	if (!aen_result)
4058 		return;
4059 
4060 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4061 	if (!envp[0])
4062 		return;
4063 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4064 	kfree(envp[0]);
4065 }
4066 
4067 static void nvme_async_event_work(struct work_struct *work)
4068 {
4069 	struct nvme_ctrl *ctrl =
4070 		container_of(work, struct nvme_ctrl, async_event_work);
4071 
4072 	nvme_aen_uevent(ctrl);
4073 
4074 	/*
4075 	 * The transport drivers must guarantee AER submission here is safe by
4076 	 * flushing ctrl async_event_work after changing the controller state
4077 	 * from LIVE and before freeing the admin queue.
4078 	*/
4079 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
4080 		ctrl->ops->submit_async_event(ctrl);
4081 }
4082 
4083 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4084 {
4085 
4086 	u32 csts;
4087 
4088 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4089 		return false;
4090 
4091 	if (csts == ~0)
4092 		return false;
4093 
4094 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4095 }
4096 
4097 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4098 {
4099 	struct nvme_fw_slot_info_log *log;
4100 
4101 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4102 	if (!log)
4103 		return;
4104 
4105 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4106 			log, sizeof(*log), 0))
4107 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4108 	kfree(log);
4109 }
4110 
4111 static void nvme_fw_act_work(struct work_struct *work)
4112 {
4113 	struct nvme_ctrl *ctrl = container_of(work,
4114 				struct nvme_ctrl, fw_act_work);
4115 	unsigned long fw_act_timeout;
4116 
4117 	nvme_auth_stop(ctrl);
4118 
4119 	if (ctrl->mtfa)
4120 		fw_act_timeout = jiffies +
4121 				msecs_to_jiffies(ctrl->mtfa * 100);
4122 	else
4123 		fw_act_timeout = jiffies +
4124 				msecs_to_jiffies(admin_timeout * 1000);
4125 
4126 	nvme_quiesce_io_queues(ctrl);
4127 	while (nvme_ctrl_pp_status(ctrl)) {
4128 		if (time_after(jiffies, fw_act_timeout)) {
4129 			dev_warn(ctrl->device,
4130 				"Fw activation timeout, reset controller\n");
4131 			nvme_try_sched_reset(ctrl);
4132 			return;
4133 		}
4134 		msleep(100);
4135 	}
4136 
4137 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4138 		return;
4139 
4140 	nvme_unquiesce_io_queues(ctrl);
4141 	/* read FW slot information to clear the AER */
4142 	nvme_get_fw_slot_info(ctrl);
4143 
4144 	queue_work(nvme_wq, &ctrl->async_event_work);
4145 }
4146 
4147 static u32 nvme_aer_type(u32 result)
4148 {
4149 	return result & 0x7;
4150 }
4151 
4152 static u32 nvme_aer_subtype(u32 result)
4153 {
4154 	return (result & 0xff00) >> 8;
4155 }
4156 
4157 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4158 {
4159 	u32 aer_notice_type = nvme_aer_subtype(result);
4160 	bool requeue = true;
4161 
4162 	switch (aer_notice_type) {
4163 	case NVME_AER_NOTICE_NS_CHANGED:
4164 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4165 		nvme_queue_scan(ctrl);
4166 		break;
4167 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4168 		/*
4169 		 * We are (ab)using the RESETTING state to prevent subsequent
4170 		 * recovery actions from interfering with the controller's
4171 		 * firmware activation.
4172 		 */
4173 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4174 			requeue = false;
4175 			queue_work(nvme_wq, &ctrl->fw_act_work);
4176 		}
4177 		break;
4178 #ifdef CONFIG_NVME_MULTIPATH
4179 	case NVME_AER_NOTICE_ANA:
4180 		if (!ctrl->ana_log_buf)
4181 			break;
4182 		queue_work(nvme_wq, &ctrl->ana_work);
4183 		break;
4184 #endif
4185 	case NVME_AER_NOTICE_DISC_CHANGED:
4186 		ctrl->aen_result = result;
4187 		break;
4188 	default:
4189 		dev_warn(ctrl->device, "async event result %08x\n", result);
4190 	}
4191 	return requeue;
4192 }
4193 
4194 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4195 {
4196 	dev_warn(ctrl->device, "resetting controller due to AER\n");
4197 	nvme_reset_ctrl(ctrl);
4198 }
4199 
4200 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4201 		volatile union nvme_result *res)
4202 {
4203 	u32 result = le32_to_cpu(res->u32);
4204 	u32 aer_type = nvme_aer_type(result);
4205 	u32 aer_subtype = nvme_aer_subtype(result);
4206 	bool requeue = true;
4207 
4208 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4209 		return;
4210 
4211 	trace_nvme_async_event(ctrl, result);
4212 	switch (aer_type) {
4213 	case NVME_AER_NOTICE:
4214 		requeue = nvme_handle_aen_notice(ctrl, result);
4215 		break;
4216 	case NVME_AER_ERROR:
4217 		/*
4218 		 * For a persistent internal error, don't run async_event_work
4219 		 * to submit a new AER. The controller reset will do it.
4220 		 */
4221 		if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4222 			nvme_handle_aer_persistent_error(ctrl);
4223 			return;
4224 		}
4225 		fallthrough;
4226 	case NVME_AER_SMART:
4227 	case NVME_AER_CSS:
4228 	case NVME_AER_VS:
4229 		ctrl->aen_result = result;
4230 		break;
4231 	default:
4232 		break;
4233 	}
4234 
4235 	if (requeue)
4236 		queue_work(nvme_wq, &ctrl->async_event_work);
4237 }
4238 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4239 
4240 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4241 		const struct blk_mq_ops *ops, unsigned int cmd_size)
4242 {
4243 	int ret;
4244 
4245 	memset(set, 0, sizeof(*set));
4246 	set->ops = ops;
4247 	set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4248 	if (ctrl->ops->flags & NVME_F_FABRICS)
4249 		/* Reserved for fabric connect and keep alive */
4250 		set->reserved_tags = 2;
4251 	set->numa_node = ctrl->numa_node;
4252 	set->flags = BLK_MQ_F_NO_SCHED;
4253 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4254 		set->flags |= BLK_MQ_F_BLOCKING;
4255 	set->cmd_size = cmd_size;
4256 	set->driver_data = ctrl;
4257 	set->nr_hw_queues = 1;
4258 	set->timeout = NVME_ADMIN_TIMEOUT;
4259 	ret = blk_mq_alloc_tag_set(set);
4260 	if (ret)
4261 		return ret;
4262 
4263 	ctrl->admin_q = blk_mq_init_queue(set);
4264 	if (IS_ERR(ctrl->admin_q)) {
4265 		ret = PTR_ERR(ctrl->admin_q);
4266 		goto out_free_tagset;
4267 	}
4268 
4269 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4270 		ctrl->fabrics_q = blk_mq_init_queue(set);
4271 		if (IS_ERR(ctrl->fabrics_q)) {
4272 			ret = PTR_ERR(ctrl->fabrics_q);
4273 			goto out_cleanup_admin_q;
4274 		}
4275 	}
4276 
4277 	ctrl->admin_tagset = set;
4278 	return 0;
4279 
4280 out_cleanup_admin_q:
4281 	blk_mq_destroy_queue(ctrl->admin_q);
4282 	blk_put_queue(ctrl->admin_q);
4283 out_free_tagset:
4284 	blk_mq_free_tag_set(set);
4285 	ctrl->admin_q = NULL;
4286 	ctrl->fabrics_q = NULL;
4287 	return ret;
4288 }
4289 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4290 
4291 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4292 {
4293 	blk_mq_destroy_queue(ctrl->admin_q);
4294 	blk_put_queue(ctrl->admin_q);
4295 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4296 		blk_mq_destroy_queue(ctrl->fabrics_q);
4297 		blk_put_queue(ctrl->fabrics_q);
4298 	}
4299 	blk_mq_free_tag_set(ctrl->admin_tagset);
4300 }
4301 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4302 
4303 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4304 		const struct blk_mq_ops *ops, unsigned int nr_maps,
4305 		unsigned int cmd_size)
4306 {
4307 	int ret;
4308 
4309 	memset(set, 0, sizeof(*set));
4310 	set->ops = ops;
4311 	set->queue_depth = min_t(unsigned, ctrl->sqsize, BLK_MQ_MAX_DEPTH - 1);
4312 	/*
4313 	 * Some Apple controllers requires tags to be unique across admin and
4314 	 * the (only) I/O queue, so reserve the first 32 tags of the I/O queue.
4315 	 */
4316 	if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
4317 		set->reserved_tags = NVME_AQ_DEPTH;
4318 	else if (ctrl->ops->flags & NVME_F_FABRICS)
4319 		/* Reserved for fabric connect */
4320 		set->reserved_tags = 1;
4321 	set->numa_node = ctrl->numa_node;
4322 	set->flags = BLK_MQ_F_SHOULD_MERGE;
4323 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4324 		set->flags |= BLK_MQ_F_BLOCKING;
4325 	set->cmd_size = cmd_size,
4326 	set->driver_data = ctrl;
4327 	set->nr_hw_queues = ctrl->queue_count - 1;
4328 	set->timeout = NVME_IO_TIMEOUT;
4329 	set->nr_maps = nr_maps;
4330 	ret = blk_mq_alloc_tag_set(set);
4331 	if (ret)
4332 		return ret;
4333 
4334 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4335 		ctrl->connect_q = blk_mq_init_queue(set);
4336         	if (IS_ERR(ctrl->connect_q)) {
4337 			ret = PTR_ERR(ctrl->connect_q);
4338 			goto out_free_tag_set;
4339 		}
4340 		blk_queue_flag_set(QUEUE_FLAG_SKIP_TAGSET_QUIESCE,
4341 				   ctrl->connect_q);
4342 	}
4343 
4344 	ctrl->tagset = set;
4345 	return 0;
4346 
4347 out_free_tag_set:
4348 	blk_mq_free_tag_set(set);
4349 	ctrl->connect_q = NULL;
4350 	return ret;
4351 }
4352 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
4353 
4354 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
4355 {
4356 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4357 		blk_mq_destroy_queue(ctrl->connect_q);
4358 		blk_put_queue(ctrl->connect_q);
4359 	}
4360 	blk_mq_free_tag_set(ctrl->tagset);
4361 }
4362 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
4363 
4364 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4365 {
4366 	nvme_mpath_stop(ctrl);
4367 	nvme_auth_stop(ctrl);
4368 	nvme_stop_keep_alive(ctrl);
4369 	nvme_stop_failfast_work(ctrl);
4370 	flush_work(&ctrl->async_event_work);
4371 	cancel_work_sync(&ctrl->fw_act_work);
4372 	if (ctrl->ops->stop_ctrl)
4373 		ctrl->ops->stop_ctrl(ctrl);
4374 }
4375 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4376 
4377 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4378 {
4379 	nvme_start_keep_alive(ctrl);
4380 
4381 	nvme_enable_aen(ctrl);
4382 
4383 	/*
4384 	 * persistent discovery controllers need to send indication to userspace
4385 	 * to re-read the discovery log page to learn about possible changes
4386 	 * that were missed. We identify persistent discovery controllers by
4387 	 * checking that they started once before, hence are reconnecting back.
4388 	 */
4389 	if (test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
4390 	    nvme_discovery_ctrl(ctrl))
4391 		nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
4392 
4393 	if (ctrl->queue_count > 1) {
4394 		nvme_queue_scan(ctrl);
4395 		nvme_unquiesce_io_queues(ctrl);
4396 		nvme_mpath_update(ctrl);
4397 	}
4398 
4399 	nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4400 	set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags);
4401 }
4402 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4403 
4404 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4405 {
4406 	nvme_hwmon_exit(ctrl);
4407 	nvme_fault_inject_fini(&ctrl->fault_inject);
4408 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4409 	cdev_device_del(&ctrl->cdev, ctrl->device);
4410 	nvme_put_ctrl(ctrl);
4411 }
4412 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4413 
4414 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4415 {
4416 	struct nvme_effects_log	*cel;
4417 	unsigned long i;
4418 
4419 	xa_for_each(&ctrl->cels, i, cel) {
4420 		xa_erase(&ctrl->cels, i);
4421 		kfree(cel);
4422 	}
4423 
4424 	xa_destroy(&ctrl->cels);
4425 }
4426 
4427 static void nvme_free_ctrl(struct device *dev)
4428 {
4429 	struct nvme_ctrl *ctrl =
4430 		container_of(dev, struct nvme_ctrl, ctrl_device);
4431 	struct nvme_subsystem *subsys = ctrl->subsys;
4432 
4433 	if (!subsys || ctrl->instance != subsys->instance)
4434 		ida_free(&nvme_instance_ida, ctrl->instance);
4435 
4436 	nvme_free_cels(ctrl);
4437 	nvme_mpath_uninit(ctrl);
4438 	nvme_auth_stop(ctrl);
4439 	nvme_auth_free(ctrl);
4440 	__free_page(ctrl->discard_page);
4441 	free_opal_dev(ctrl->opal_dev);
4442 
4443 	if (subsys) {
4444 		mutex_lock(&nvme_subsystems_lock);
4445 		list_del(&ctrl->subsys_entry);
4446 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4447 		mutex_unlock(&nvme_subsystems_lock);
4448 	}
4449 
4450 	ctrl->ops->free_ctrl(ctrl);
4451 
4452 	if (subsys)
4453 		nvme_put_subsystem(subsys);
4454 }
4455 
4456 /*
4457  * Initialize a NVMe controller structures.  This needs to be called during
4458  * earliest initialization so that we have the initialized structured around
4459  * during probing.
4460  */
4461 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4462 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
4463 {
4464 	int ret;
4465 
4466 	WRITE_ONCE(ctrl->state, NVME_CTRL_NEW);
4467 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4468 	spin_lock_init(&ctrl->lock);
4469 	mutex_init(&ctrl->scan_lock);
4470 	INIT_LIST_HEAD(&ctrl->namespaces);
4471 	xa_init(&ctrl->cels);
4472 	init_rwsem(&ctrl->namespaces_rwsem);
4473 	ctrl->dev = dev;
4474 	ctrl->ops = ops;
4475 	ctrl->quirks = quirks;
4476 	ctrl->numa_node = NUMA_NO_NODE;
4477 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4478 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4479 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4480 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4481 	init_waitqueue_head(&ctrl->state_wq);
4482 
4483 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4484 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4485 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4486 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4487 
4488 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4489 			PAGE_SIZE);
4490 	ctrl->discard_page = alloc_page(GFP_KERNEL);
4491 	if (!ctrl->discard_page) {
4492 		ret = -ENOMEM;
4493 		goto out;
4494 	}
4495 
4496 	ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4497 	if (ret < 0)
4498 		goto out;
4499 	ctrl->instance = ret;
4500 
4501 	device_initialize(&ctrl->ctrl_device);
4502 	ctrl->device = &ctrl->ctrl_device;
4503 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4504 			ctrl->instance);
4505 	ctrl->device->class = nvme_class;
4506 	ctrl->device->parent = ctrl->dev;
4507 	if (ops->dev_attr_groups)
4508 		ctrl->device->groups = ops->dev_attr_groups;
4509 	else
4510 		ctrl->device->groups = nvme_dev_attr_groups;
4511 	ctrl->device->release = nvme_free_ctrl;
4512 	dev_set_drvdata(ctrl->device, ctrl);
4513 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4514 	if (ret)
4515 		goto out_release_instance;
4516 
4517 	nvme_get_ctrl(ctrl);
4518 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
4519 	ctrl->cdev.owner = ops->module;
4520 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4521 	if (ret)
4522 		goto out_free_name;
4523 
4524 	/*
4525 	 * Initialize latency tolerance controls.  The sysfs files won't
4526 	 * be visible to userspace unless the device actually supports APST.
4527 	 */
4528 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4529 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4530 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4531 
4532 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4533 	nvme_mpath_init_ctrl(ctrl);
4534 	ret = nvme_auth_init_ctrl(ctrl);
4535 	if (ret)
4536 		goto out_free_cdev;
4537 
4538 	return 0;
4539 out_free_cdev:
4540 	nvme_fault_inject_fini(&ctrl->fault_inject);
4541 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4542 	cdev_device_del(&ctrl->cdev, ctrl->device);
4543 out_free_name:
4544 	nvme_put_ctrl(ctrl);
4545 	kfree_const(ctrl->device->kobj.name);
4546 out_release_instance:
4547 	ida_free(&nvme_instance_ida, ctrl->instance);
4548 out:
4549 	if (ctrl->discard_page)
4550 		__free_page(ctrl->discard_page);
4551 	return ret;
4552 }
4553 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4554 
4555 /* let I/O to all namespaces fail in preparation for surprise removal */
4556 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
4557 {
4558 	struct nvme_ns *ns;
4559 
4560 	down_read(&ctrl->namespaces_rwsem);
4561 	list_for_each_entry(ns, &ctrl->namespaces, list)
4562 		blk_mark_disk_dead(ns->disk);
4563 	up_read(&ctrl->namespaces_rwsem);
4564 }
4565 EXPORT_SYMBOL_GPL(nvme_mark_namespaces_dead);
4566 
4567 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4568 {
4569 	struct nvme_ns *ns;
4570 
4571 	down_read(&ctrl->namespaces_rwsem);
4572 	list_for_each_entry(ns, &ctrl->namespaces, list)
4573 		blk_mq_unfreeze_queue(ns->queue);
4574 	up_read(&ctrl->namespaces_rwsem);
4575 	clear_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4576 }
4577 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4578 
4579 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4580 {
4581 	struct nvme_ns *ns;
4582 
4583 	down_read(&ctrl->namespaces_rwsem);
4584 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4585 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4586 		if (timeout <= 0)
4587 			break;
4588 	}
4589 	up_read(&ctrl->namespaces_rwsem);
4590 	return timeout;
4591 }
4592 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4593 
4594 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4595 {
4596 	struct nvme_ns *ns;
4597 
4598 	down_read(&ctrl->namespaces_rwsem);
4599 	list_for_each_entry(ns, &ctrl->namespaces, list)
4600 		blk_mq_freeze_queue_wait(ns->queue);
4601 	up_read(&ctrl->namespaces_rwsem);
4602 }
4603 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4604 
4605 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4606 {
4607 	struct nvme_ns *ns;
4608 
4609 	set_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4610 	down_read(&ctrl->namespaces_rwsem);
4611 	list_for_each_entry(ns, &ctrl->namespaces, list)
4612 		blk_freeze_queue_start(ns->queue);
4613 	up_read(&ctrl->namespaces_rwsem);
4614 }
4615 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4616 
4617 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl)
4618 {
4619 	if (!ctrl->tagset)
4620 		return;
4621 	if (!test_and_set_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4622 		blk_mq_quiesce_tagset(ctrl->tagset);
4623 	else
4624 		blk_mq_wait_quiesce_done(ctrl->tagset);
4625 }
4626 EXPORT_SYMBOL_GPL(nvme_quiesce_io_queues);
4627 
4628 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl)
4629 {
4630 	if (!ctrl->tagset)
4631 		return;
4632 	if (test_and_clear_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4633 		blk_mq_unquiesce_tagset(ctrl->tagset);
4634 }
4635 EXPORT_SYMBOL_GPL(nvme_unquiesce_io_queues);
4636 
4637 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl)
4638 {
4639 	if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4640 		blk_mq_quiesce_queue(ctrl->admin_q);
4641 	else
4642 		blk_mq_wait_quiesce_done(ctrl->admin_q->tag_set);
4643 }
4644 EXPORT_SYMBOL_GPL(nvme_quiesce_admin_queue);
4645 
4646 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl)
4647 {
4648 	if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4649 		blk_mq_unquiesce_queue(ctrl->admin_q);
4650 }
4651 EXPORT_SYMBOL_GPL(nvme_unquiesce_admin_queue);
4652 
4653 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4654 {
4655 	struct nvme_ns *ns;
4656 
4657 	down_read(&ctrl->namespaces_rwsem);
4658 	list_for_each_entry(ns, &ctrl->namespaces, list)
4659 		blk_sync_queue(ns->queue);
4660 	up_read(&ctrl->namespaces_rwsem);
4661 }
4662 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4663 
4664 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4665 {
4666 	nvme_sync_io_queues(ctrl);
4667 	if (ctrl->admin_q)
4668 		blk_sync_queue(ctrl->admin_q);
4669 }
4670 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4671 
4672 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4673 {
4674 	if (file->f_op != &nvme_dev_fops)
4675 		return NULL;
4676 	return file->private_data;
4677 }
4678 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4679 
4680 /*
4681  * Check we didn't inadvertently grow the command structure sizes:
4682  */
4683 static inline void _nvme_check_size(void)
4684 {
4685 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4686 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4687 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4688 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4689 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4690 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4691 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4692 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4693 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4694 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4695 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4696 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4697 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4698 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
4699 			NVME_IDENTIFY_DATA_SIZE);
4700 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4701 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
4702 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4703 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4704 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4705 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4706 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4707 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4708 	BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
4709 }
4710 
4711 
4712 static int __init nvme_core_init(void)
4713 {
4714 	int result = -ENOMEM;
4715 
4716 	_nvme_check_size();
4717 
4718 	nvme_wq = alloc_workqueue("nvme-wq",
4719 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4720 	if (!nvme_wq)
4721 		goto out;
4722 
4723 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4724 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4725 	if (!nvme_reset_wq)
4726 		goto destroy_wq;
4727 
4728 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4729 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4730 	if (!nvme_delete_wq)
4731 		goto destroy_reset_wq;
4732 
4733 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4734 			NVME_MINORS, "nvme");
4735 	if (result < 0)
4736 		goto destroy_delete_wq;
4737 
4738 	nvme_class = class_create("nvme");
4739 	if (IS_ERR(nvme_class)) {
4740 		result = PTR_ERR(nvme_class);
4741 		goto unregister_chrdev;
4742 	}
4743 	nvme_class->dev_uevent = nvme_class_uevent;
4744 
4745 	nvme_subsys_class = class_create("nvme-subsystem");
4746 	if (IS_ERR(nvme_subsys_class)) {
4747 		result = PTR_ERR(nvme_subsys_class);
4748 		goto destroy_class;
4749 	}
4750 
4751 	result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4752 				     "nvme-generic");
4753 	if (result < 0)
4754 		goto destroy_subsys_class;
4755 
4756 	nvme_ns_chr_class = class_create("nvme-generic");
4757 	if (IS_ERR(nvme_ns_chr_class)) {
4758 		result = PTR_ERR(nvme_ns_chr_class);
4759 		goto unregister_generic_ns;
4760 	}
4761 
4762 	result = nvme_init_auth();
4763 	if (result)
4764 		goto destroy_ns_chr;
4765 	return 0;
4766 
4767 destroy_ns_chr:
4768 	class_destroy(nvme_ns_chr_class);
4769 unregister_generic_ns:
4770 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4771 destroy_subsys_class:
4772 	class_destroy(nvme_subsys_class);
4773 destroy_class:
4774 	class_destroy(nvme_class);
4775 unregister_chrdev:
4776 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4777 destroy_delete_wq:
4778 	destroy_workqueue(nvme_delete_wq);
4779 destroy_reset_wq:
4780 	destroy_workqueue(nvme_reset_wq);
4781 destroy_wq:
4782 	destroy_workqueue(nvme_wq);
4783 out:
4784 	return result;
4785 }
4786 
4787 static void __exit nvme_core_exit(void)
4788 {
4789 	nvme_exit_auth();
4790 	class_destroy(nvme_ns_chr_class);
4791 	class_destroy(nvme_subsys_class);
4792 	class_destroy(nvme_class);
4793 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4794 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4795 	destroy_workqueue(nvme_delete_wq);
4796 	destroy_workqueue(nvme_reset_wq);
4797 	destroy_workqueue(nvme_wq);
4798 	ida_destroy(&nvme_ns_chr_minor_ida);
4799 	ida_destroy(&nvme_instance_ida);
4800 }
4801 
4802 MODULE_LICENSE("GPL");
4803 MODULE_VERSION("1.0");
4804 module_init(nvme_core_init);
4805 module_exit(nvme_core_exit);
4806