xref: /openbmc/linux/drivers/nvme/host/core.c (revision 8d1af5c6)
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 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  * Waits for the controller state to be resetting, or returns false if it is
592  * not possible to ever transition to that state.
593  */
594 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
595 {
596 	wait_event(ctrl->state_wq,
597 		   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
598 		   nvme_state_terminal(ctrl));
599 	return nvme_ctrl_state(ctrl) == NVME_CTRL_RESETTING;
600 }
601 EXPORT_SYMBOL_GPL(nvme_wait_reset);
602 
603 static void nvme_free_ns_head(struct kref *ref)
604 {
605 	struct nvme_ns_head *head =
606 		container_of(ref, struct nvme_ns_head, ref);
607 
608 	nvme_mpath_remove_disk(head);
609 	ida_free(&head->subsys->ns_ida, head->instance);
610 	cleanup_srcu_struct(&head->srcu);
611 	nvme_put_subsystem(head->subsys);
612 	kfree(head);
613 }
614 
615 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
616 {
617 	return kref_get_unless_zero(&head->ref);
618 }
619 
620 void nvme_put_ns_head(struct nvme_ns_head *head)
621 {
622 	kref_put(&head->ref, nvme_free_ns_head);
623 }
624 
625 static void nvme_free_ns(struct kref *kref)
626 {
627 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
628 
629 	put_disk(ns->disk);
630 	nvme_put_ns_head(ns->head);
631 	nvme_put_ctrl(ns->ctrl);
632 	kfree(ns);
633 }
634 
635 bool nvme_get_ns(struct nvme_ns *ns)
636 {
637 	return kref_get_unless_zero(&ns->kref);
638 }
639 
640 void nvme_put_ns(struct nvme_ns *ns)
641 {
642 	kref_put(&ns->kref, nvme_free_ns);
643 }
644 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
645 
646 static inline void nvme_clear_nvme_request(struct request *req)
647 {
648 	nvme_req(req)->status = 0;
649 	nvme_req(req)->retries = 0;
650 	nvme_req(req)->flags = 0;
651 	req->rq_flags |= RQF_DONTPREP;
652 }
653 
654 /* initialize a passthrough request */
655 void nvme_init_request(struct request *req, struct nvme_command *cmd)
656 {
657 	if (req->q->queuedata)
658 		req->timeout = NVME_IO_TIMEOUT;
659 	else /* no queuedata implies admin queue */
660 		req->timeout = NVME_ADMIN_TIMEOUT;
661 
662 	/* passthru commands should let the driver set the SGL flags */
663 	cmd->common.flags &= ~NVME_CMD_SGL_ALL;
664 
665 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
666 	if (req->mq_hctx->type == HCTX_TYPE_POLL)
667 		req->cmd_flags |= REQ_POLLED;
668 	nvme_clear_nvme_request(req);
669 	req->rq_flags |= RQF_QUIET;
670 	memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
671 }
672 EXPORT_SYMBOL_GPL(nvme_init_request);
673 
674 /*
675  * For something we're not in a state to send to the device the default action
676  * is to busy it and retry it after the controller state is recovered.  However,
677  * if the controller is deleting or if anything is marked for failfast or
678  * nvme multipath it is immediately failed.
679  *
680  * Note: commands used to initialize the controller will be marked for failfast.
681  * Note: nvme cli/ioctl commands are marked for failfast.
682  */
683 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
684 		struct request *rq)
685 {
686 	enum nvme_ctrl_state state = nvme_ctrl_state(ctrl);
687 
688 	if (state != NVME_CTRL_DELETING_NOIO &&
689 	    state != NVME_CTRL_DELETING &&
690 	    state != NVME_CTRL_DEAD &&
691 	    !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
692 	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
693 		return BLK_STS_RESOURCE;
694 	return nvme_host_path_error(rq);
695 }
696 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
697 
698 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
699 		bool queue_live)
700 {
701 	struct nvme_request *req = nvme_req(rq);
702 
703 	/*
704 	 * currently we have a problem sending passthru commands
705 	 * on the admin_q if the controller is not LIVE because we can't
706 	 * make sure that they are going out after the admin connect,
707 	 * controller enable and/or other commands in the initialization
708 	 * sequence. until the controller will be LIVE, fail with
709 	 * BLK_STS_RESOURCE so that they will be rescheduled.
710 	 */
711 	if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
712 		return false;
713 
714 	if (ctrl->ops->flags & NVME_F_FABRICS) {
715 		/*
716 		 * Only allow commands on a live queue, except for the connect
717 		 * command, which is require to set the queue live in the
718 		 * appropinquate states.
719 		 */
720 		switch (nvme_ctrl_state(ctrl)) {
721 		case NVME_CTRL_CONNECTING:
722 			if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
723 			    (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
724 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
725 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
726 				return true;
727 			break;
728 		default:
729 			break;
730 		case NVME_CTRL_DEAD:
731 			return false;
732 		}
733 	}
734 
735 	return queue_live;
736 }
737 EXPORT_SYMBOL_GPL(__nvme_check_ready);
738 
739 static inline void nvme_setup_flush(struct nvme_ns *ns,
740 		struct nvme_command *cmnd)
741 {
742 	memset(cmnd, 0, sizeof(*cmnd));
743 	cmnd->common.opcode = nvme_cmd_flush;
744 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
745 }
746 
747 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
748 		struct nvme_command *cmnd)
749 {
750 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
751 	struct nvme_dsm_range *range;
752 	struct bio *bio;
753 
754 	/*
755 	 * Some devices do not consider the DSM 'Number of Ranges' field when
756 	 * determining how much data to DMA. Always allocate memory for maximum
757 	 * number of segments to prevent device reading beyond end of buffer.
758 	 */
759 	static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
760 
761 	range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
762 	if (!range) {
763 		/*
764 		 * If we fail allocation our range, fallback to the controller
765 		 * discard page. If that's also busy, it's safe to return
766 		 * busy, as we know we can make progress once that's freed.
767 		 */
768 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
769 			return BLK_STS_RESOURCE;
770 
771 		range = page_address(ns->ctrl->discard_page);
772 	}
773 
774 	if (queue_max_discard_segments(req->q) == 1) {
775 		u64 slba = nvme_sect_to_lba(ns, blk_rq_pos(req));
776 		u32 nlb = blk_rq_sectors(req) >> (ns->lba_shift - 9);
777 
778 		range[0].cattr = cpu_to_le32(0);
779 		range[0].nlb = cpu_to_le32(nlb);
780 		range[0].slba = cpu_to_le64(slba);
781 		n = 1;
782 	} else {
783 		__rq_for_each_bio(bio, req) {
784 			u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
785 			u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
786 
787 			if (n < segments) {
788 				range[n].cattr = cpu_to_le32(0);
789 				range[n].nlb = cpu_to_le32(nlb);
790 				range[n].slba = cpu_to_le64(slba);
791 			}
792 			n++;
793 		}
794 	}
795 
796 	if (WARN_ON_ONCE(n != segments)) {
797 		if (virt_to_page(range) == ns->ctrl->discard_page)
798 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
799 		else
800 			kfree(range);
801 		return BLK_STS_IOERR;
802 	}
803 
804 	memset(cmnd, 0, sizeof(*cmnd));
805 	cmnd->dsm.opcode = nvme_cmd_dsm;
806 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
807 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
808 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
809 
810 	bvec_set_virt(&req->special_vec, range, alloc_size);
811 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
812 
813 	return BLK_STS_OK;
814 }
815 
816 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
817 			      struct request *req)
818 {
819 	u32 upper, lower;
820 	u64 ref48;
821 
822 	/* both rw and write zeroes share the same reftag format */
823 	switch (ns->guard_type) {
824 	case NVME_NVM_NS_16B_GUARD:
825 		cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
826 		break;
827 	case NVME_NVM_NS_64B_GUARD:
828 		ref48 = ext_pi_ref_tag(req);
829 		lower = lower_32_bits(ref48);
830 		upper = upper_32_bits(ref48);
831 
832 		cmnd->rw.reftag = cpu_to_le32(lower);
833 		cmnd->rw.cdw3 = cpu_to_le32(upper);
834 		break;
835 	default:
836 		break;
837 	}
838 }
839 
840 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
841 		struct request *req, struct nvme_command *cmnd)
842 {
843 	memset(cmnd, 0, sizeof(*cmnd));
844 
845 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
846 		return nvme_setup_discard(ns, req, cmnd);
847 
848 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
849 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
850 	cmnd->write_zeroes.slba =
851 		cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
852 	cmnd->write_zeroes.length =
853 		cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
854 
855 	if (!(req->cmd_flags & REQ_NOUNMAP) && (ns->features & NVME_NS_DEAC))
856 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_WZ_DEAC);
857 
858 	if (nvme_ns_has_pi(ns)) {
859 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_RW_PRINFO_PRACT);
860 
861 		switch (ns->pi_type) {
862 		case NVME_NS_DPS_PI_TYPE1:
863 		case NVME_NS_DPS_PI_TYPE2:
864 			nvme_set_ref_tag(ns, cmnd, req);
865 			break;
866 		}
867 	}
868 
869 	return BLK_STS_OK;
870 }
871 
872 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
873 		struct request *req, struct nvme_command *cmnd,
874 		enum nvme_opcode op)
875 {
876 	u16 control = 0;
877 	u32 dsmgmt = 0;
878 
879 	if (req->cmd_flags & REQ_FUA)
880 		control |= NVME_RW_FUA;
881 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
882 		control |= NVME_RW_LR;
883 
884 	if (req->cmd_flags & REQ_RAHEAD)
885 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
886 
887 	cmnd->rw.opcode = op;
888 	cmnd->rw.flags = 0;
889 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
890 	cmnd->rw.cdw2 = 0;
891 	cmnd->rw.cdw3 = 0;
892 	cmnd->rw.metadata = 0;
893 	cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
894 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
895 	cmnd->rw.reftag = 0;
896 	cmnd->rw.apptag = 0;
897 	cmnd->rw.appmask = 0;
898 
899 	if (ns->ms) {
900 		/*
901 		 * If formated with metadata, the block layer always provides a
902 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
903 		 * we enable the PRACT bit for protection information or set the
904 		 * namespace capacity to zero to prevent any I/O.
905 		 */
906 		if (!blk_integrity_rq(req)) {
907 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
908 				return BLK_STS_NOTSUPP;
909 			control |= NVME_RW_PRINFO_PRACT;
910 		}
911 
912 		switch (ns->pi_type) {
913 		case NVME_NS_DPS_PI_TYPE3:
914 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
915 			break;
916 		case NVME_NS_DPS_PI_TYPE1:
917 		case NVME_NS_DPS_PI_TYPE2:
918 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
919 					NVME_RW_PRINFO_PRCHK_REF;
920 			if (op == nvme_cmd_zone_append)
921 				control |= NVME_RW_APPEND_PIREMAP;
922 			nvme_set_ref_tag(ns, cmnd, req);
923 			break;
924 		}
925 	}
926 
927 	cmnd->rw.control = cpu_to_le16(control);
928 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
929 	return 0;
930 }
931 
932 void nvme_cleanup_cmd(struct request *req)
933 {
934 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
935 		struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
936 
937 		if (req->special_vec.bv_page == ctrl->discard_page)
938 			clear_bit_unlock(0, &ctrl->discard_page_busy);
939 		else
940 			kfree(bvec_virt(&req->special_vec));
941 		req->rq_flags &= ~RQF_SPECIAL_PAYLOAD;
942 	}
943 }
944 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
945 
946 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
947 {
948 	struct nvme_command *cmd = nvme_req(req)->cmd;
949 	blk_status_t ret = BLK_STS_OK;
950 
951 	if (!(req->rq_flags & RQF_DONTPREP))
952 		nvme_clear_nvme_request(req);
953 
954 	switch (req_op(req)) {
955 	case REQ_OP_DRV_IN:
956 	case REQ_OP_DRV_OUT:
957 		/* these are setup prior to execution in nvme_init_request() */
958 		break;
959 	case REQ_OP_FLUSH:
960 		nvme_setup_flush(ns, cmd);
961 		break;
962 	case REQ_OP_ZONE_RESET_ALL:
963 	case REQ_OP_ZONE_RESET:
964 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
965 		break;
966 	case REQ_OP_ZONE_OPEN:
967 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
968 		break;
969 	case REQ_OP_ZONE_CLOSE:
970 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
971 		break;
972 	case REQ_OP_ZONE_FINISH:
973 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
974 		break;
975 	case REQ_OP_WRITE_ZEROES:
976 		ret = nvme_setup_write_zeroes(ns, req, cmd);
977 		break;
978 	case REQ_OP_DISCARD:
979 		ret = nvme_setup_discard(ns, req, cmd);
980 		break;
981 	case REQ_OP_READ:
982 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
983 		break;
984 	case REQ_OP_WRITE:
985 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
986 		break;
987 	case REQ_OP_ZONE_APPEND:
988 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
989 		break;
990 	default:
991 		WARN_ON_ONCE(1);
992 		return BLK_STS_IOERR;
993 	}
994 
995 	cmd->common.command_id = nvme_cid(req);
996 	trace_nvme_setup_cmd(req, cmd);
997 	return ret;
998 }
999 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1000 
1001 /*
1002  * Return values:
1003  * 0:  success
1004  * >0: nvme controller's cqe status response
1005  * <0: kernel error in lieu of controller response
1006  */
1007 int nvme_execute_rq(struct request *rq, bool at_head)
1008 {
1009 	blk_status_t status;
1010 
1011 	status = blk_execute_rq(rq, at_head);
1012 	if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1013 		return -EINTR;
1014 	if (nvme_req(rq)->status)
1015 		return nvme_req(rq)->status;
1016 	return blk_status_to_errno(status);
1017 }
1018 EXPORT_SYMBOL_NS_GPL(nvme_execute_rq, NVME_TARGET_PASSTHRU);
1019 
1020 /*
1021  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1022  * if the result is positive, it's an NVM Express status code
1023  */
1024 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1025 		union nvme_result *result, void *buffer, unsigned bufflen,
1026 		int qid, int at_head, blk_mq_req_flags_t flags)
1027 {
1028 	struct request *req;
1029 	int ret;
1030 
1031 	if (qid == NVME_QID_ANY)
1032 		req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1033 	else
1034 		req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1035 						qid - 1);
1036 
1037 	if (IS_ERR(req))
1038 		return PTR_ERR(req);
1039 	nvme_init_request(req, cmd);
1040 
1041 	if (buffer && bufflen) {
1042 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1043 		if (ret)
1044 			goto out;
1045 	}
1046 
1047 	ret = nvme_execute_rq(req, at_head);
1048 	if (result && ret >= 0)
1049 		*result = nvme_req(req)->result;
1050  out:
1051 	blk_mq_free_request(req);
1052 	return ret;
1053 }
1054 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1055 
1056 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1057 		void *buffer, unsigned bufflen)
1058 {
1059 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1060 			NVME_QID_ANY, 0, 0);
1061 }
1062 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1063 
1064 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1065 {
1066 	u32 effects = 0;
1067 
1068 	if (ns) {
1069 		effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1070 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1071 			dev_warn_once(ctrl->device,
1072 				"IO command:%02x has unusual effects:%08x\n",
1073 				opcode, effects);
1074 
1075 		/*
1076 		 * NVME_CMD_EFFECTS_CSE_MASK causes a freeze all I/O queues,
1077 		 * which would deadlock when done on an I/O command.  Note that
1078 		 * We already warn about an unusual effect above.
1079 		 */
1080 		effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1081 	} else {
1082 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1083 	}
1084 
1085 	return effects;
1086 }
1087 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1088 
1089 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1090 {
1091 	u32 effects = nvme_command_effects(ctrl, ns, opcode);
1092 
1093 	/*
1094 	 * For simplicity, IO to all namespaces is quiesced even if the command
1095 	 * effects say only one namespace is affected.
1096 	 */
1097 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1098 		mutex_lock(&ctrl->scan_lock);
1099 		mutex_lock(&ctrl->subsys->lock);
1100 		nvme_mpath_start_freeze(ctrl->subsys);
1101 		nvme_mpath_wait_freeze(ctrl->subsys);
1102 		nvme_start_freeze(ctrl);
1103 		nvme_wait_freeze(ctrl);
1104 	}
1105 	return effects;
1106 }
1107 EXPORT_SYMBOL_NS_GPL(nvme_passthru_start, NVME_TARGET_PASSTHRU);
1108 
1109 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1110 		       struct nvme_command *cmd, int status)
1111 {
1112 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1113 		nvme_unfreeze(ctrl);
1114 		nvme_mpath_unfreeze(ctrl->subsys);
1115 		mutex_unlock(&ctrl->subsys->lock);
1116 		mutex_unlock(&ctrl->scan_lock);
1117 	}
1118 	if (effects & NVME_CMD_EFFECTS_CCC) {
1119 		if (!test_and_set_bit(NVME_CTRL_DIRTY_CAPABILITY,
1120 				      &ctrl->flags)) {
1121 			dev_info(ctrl->device,
1122 "controller capabilities changed, reset may be required to take effect.\n");
1123 		}
1124 	}
1125 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1126 		nvme_queue_scan(ctrl);
1127 		flush_work(&ctrl->scan_work);
1128 	}
1129 	if (ns)
1130 		return;
1131 
1132 	switch (cmd->common.opcode) {
1133 	case nvme_admin_set_features:
1134 		switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1135 		case NVME_FEAT_KATO:
1136 			/*
1137 			 * Keep alive commands interval on the host should be
1138 			 * updated when KATO is modified by Set Features
1139 			 * commands.
1140 			 */
1141 			if (!status)
1142 				nvme_update_keep_alive(ctrl, cmd);
1143 			break;
1144 		default:
1145 			break;
1146 		}
1147 		break;
1148 	default:
1149 		break;
1150 	}
1151 }
1152 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1153 
1154 /*
1155  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1156  *
1157  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1158  *   accounting for transport roundtrip times [..].
1159  */
1160 static unsigned long nvme_keep_alive_work_period(struct nvme_ctrl *ctrl)
1161 {
1162 	unsigned long delay = ctrl->kato * HZ / 2;
1163 
1164 	/*
1165 	 * When using Traffic Based Keep Alive, we need to run
1166 	 * nvme_keep_alive_work at twice the normal frequency, as one
1167 	 * command completion can postpone sending a keep alive command
1168 	 * by up to twice the delay between runs.
1169 	 */
1170 	if (ctrl->ctratt & NVME_CTRL_ATTR_TBKAS)
1171 		delay /= 2;
1172 	return delay;
1173 }
1174 
1175 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1176 {
1177 	queue_delayed_work(nvme_wq, &ctrl->ka_work,
1178 			   nvme_keep_alive_work_period(ctrl));
1179 }
1180 
1181 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1182 						 blk_status_t status)
1183 {
1184 	struct nvme_ctrl *ctrl = rq->end_io_data;
1185 	unsigned long flags;
1186 	bool startka = false;
1187 	unsigned long rtt = jiffies - (rq->deadline - rq->timeout);
1188 	unsigned long delay = nvme_keep_alive_work_period(ctrl);
1189 
1190 	/*
1191 	 * Subtract off the keepalive RTT so nvme_keep_alive_work runs
1192 	 * at the desired frequency.
1193 	 */
1194 	if (rtt <= delay) {
1195 		delay -= rtt;
1196 	} else {
1197 		dev_warn(ctrl->device, "long keepalive RTT (%u ms)\n",
1198 			 jiffies_to_msecs(rtt));
1199 		delay = 0;
1200 	}
1201 
1202 	blk_mq_free_request(rq);
1203 
1204 	if (status) {
1205 		dev_err(ctrl->device,
1206 			"failed nvme_keep_alive_end_io error=%d\n",
1207 				status);
1208 		return RQ_END_IO_NONE;
1209 	}
1210 
1211 	ctrl->ka_last_check_time = jiffies;
1212 	ctrl->comp_seen = false;
1213 	spin_lock_irqsave(&ctrl->lock, flags);
1214 	if (ctrl->state == NVME_CTRL_LIVE ||
1215 	    ctrl->state == NVME_CTRL_CONNECTING)
1216 		startka = true;
1217 	spin_unlock_irqrestore(&ctrl->lock, flags);
1218 	if (startka)
1219 		queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1220 	return RQ_END_IO_NONE;
1221 }
1222 
1223 static void nvme_keep_alive_work(struct work_struct *work)
1224 {
1225 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1226 			struct nvme_ctrl, ka_work);
1227 	bool comp_seen = ctrl->comp_seen;
1228 	struct request *rq;
1229 
1230 	ctrl->ka_last_check_time = jiffies;
1231 
1232 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1233 		dev_dbg(ctrl->device,
1234 			"reschedule traffic based keep-alive timer\n");
1235 		ctrl->comp_seen = false;
1236 		nvme_queue_keep_alive_work(ctrl);
1237 		return;
1238 	}
1239 
1240 	rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1241 				  BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1242 	if (IS_ERR(rq)) {
1243 		/* allocation failure, reset the controller */
1244 		dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1245 		nvme_reset_ctrl(ctrl);
1246 		return;
1247 	}
1248 	nvme_init_request(rq, &ctrl->ka_cmd);
1249 
1250 	rq->timeout = ctrl->kato * HZ;
1251 	rq->end_io = nvme_keep_alive_end_io;
1252 	rq->end_io_data = ctrl;
1253 	blk_execute_rq_nowait(rq, false);
1254 }
1255 
1256 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1257 {
1258 	if (unlikely(ctrl->kato == 0))
1259 		return;
1260 
1261 	nvme_queue_keep_alive_work(ctrl);
1262 }
1263 
1264 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1265 {
1266 	if (unlikely(ctrl->kato == 0))
1267 		return;
1268 
1269 	cancel_delayed_work_sync(&ctrl->ka_work);
1270 }
1271 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1272 
1273 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1274 				   struct nvme_command *cmd)
1275 {
1276 	unsigned int new_kato =
1277 		DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1278 
1279 	dev_info(ctrl->device,
1280 		 "keep alive interval updated from %u ms to %u ms\n",
1281 		 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1282 
1283 	nvme_stop_keep_alive(ctrl);
1284 	ctrl->kato = new_kato;
1285 	nvme_start_keep_alive(ctrl);
1286 }
1287 
1288 /*
1289  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1290  * flag, thus sending any new CNS opcodes has a big chance of not working.
1291  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1292  * (but not for any later version).
1293  */
1294 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1295 {
1296 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1297 		return ctrl->vs < NVME_VS(1, 2, 0);
1298 	return ctrl->vs < NVME_VS(1, 1, 0);
1299 }
1300 
1301 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1302 {
1303 	struct nvme_command c = { };
1304 	int error;
1305 
1306 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1307 	c.identify.opcode = nvme_admin_identify;
1308 	c.identify.cns = NVME_ID_CNS_CTRL;
1309 
1310 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1311 	if (!*id)
1312 		return -ENOMEM;
1313 
1314 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1315 			sizeof(struct nvme_id_ctrl));
1316 	if (error) {
1317 		kfree(*id);
1318 		*id = NULL;
1319 	}
1320 	return error;
1321 }
1322 
1323 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1324 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1325 {
1326 	const char *warn_str = "ctrl returned bogus length:";
1327 	void *data = cur;
1328 
1329 	switch (cur->nidt) {
1330 	case NVME_NIDT_EUI64:
1331 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1332 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1333 				 warn_str, cur->nidl);
1334 			return -1;
1335 		}
1336 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1337 			return NVME_NIDT_EUI64_LEN;
1338 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1339 		return NVME_NIDT_EUI64_LEN;
1340 	case NVME_NIDT_NGUID:
1341 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1342 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1343 				 warn_str, cur->nidl);
1344 			return -1;
1345 		}
1346 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1347 			return NVME_NIDT_NGUID_LEN;
1348 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1349 		return NVME_NIDT_NGUID_LEN;
1350 	case NVME_NIDT_UUID:
1351 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1352 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1353 				 warn_str, cur->nidl);
1354 			return -1;
1355 		}
1356 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1357 			return NVME_NIDT_UUID_LEN;
1358 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1359 		return NVME_NIDT_UUID_LEN;
1360 	case NVME_NIDT_CSI:
1361 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1362 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1363 				 warn_str, cur->nidl);
1364 			return -1;
1365 		}
1366 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1367 		*csi_seen = true;
1368 		return NVME_NIDT_CSI_LEN;
1369 	default:
1370 		/* Skip unknown types */
1371 		return cur->nidl;
1372 	}
1373 }
1374 
1375 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1376 		struct nvme_ns_info *info)
1377 {
1378 	struct nvme_command c = { };
1379 	bool csi_seen = false;
1380 	int status, pos, len;
1381 	void *data;
1382 
1383 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1384 		return 0;
1385 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1386 		return 0;
1387 
1388 	c.identify.opcode = nvme_admin_identify;
1389 	c.identify.nsid = cpu_to_le32(info->nsid);
1390 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1391 
1392 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1393 	if (!data)
1394 		return -ENOMEM;
1395 
1396 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1397 				      NVME_IDENTIFY_DATA_SIZE);
1398 	if (status) {
1399 		dev_warn(ctrl->device,
1400 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1401 			info->nsid, status);
1402 		goto free_data;
1403 	}
1404 
1405 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1406 		struct nvme_ns_id_desc *cur = data + pos;
1407 
1408 		if (cur->nidl == 0)
1409 			break;
1410 
1411 		len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1412 		if (len < 0)
1413 			break;
1414 
1415 		len += sizeof(*cur);
1416 	}
1417 
1418 	if (nvme_multi_css(ctrl) && !csi_seen) {
1419 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1420 			 info->nsid);
1421 		status = -EINVAL;
1422 	}
1423 
1424 free_data:
1425 	kfree(data);
1426 	return status;
1427 }
1428 
1429 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1430 			struct nvme_id_ns **id)
1431 {
1432 	struct nvme_command c = { };
1433 	int error;
1434 
1435 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1436 	c.identify.opcode = nvme_admin_identify;
1437 	c.identify.nsid = cpu_to_le32(nsid);
1438 	c.identify.cns = NVME_ID_CNS_NS;
1439 
1440 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1441 	if (!*id)
1442 		return -ENOMEM;
1443 
1444 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1445 	if (error) {
1446 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1447 		kfree(*id);
1448 		*id = NULL;
1449 	}
1450 	return error;
1451 }
1452 
1453 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1454 		struct nvme_ns_info *info)
1455 {
1456 	struct nvme_ns_ids *ids = &info->ids;
1457 	struct nvme_id_ns *id;
1458 	int ret;
1459 
1460 	ret = nvme_identify_ns(ctrl, info->nsid, &id);
1461 	if (ret)
1462 		return ret;
1463 
1464 	if (id->ncap == 0) {
1465 		/* namespace not allocated or attached */
1466 		info->is_removed = true;
1467 		ret = -ENODEV;
1468 		goto error;
1469 	}
1470 
1471 	info->anagrpid = id->anagrpid;
1472 	info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1473 	info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1474 	info->is_ready = true;
1475 	if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1476 		dev_info(ctrl->device,
1477 			 "Ignoring bogus Namespace Identifiers\n");
1478 	} else {
1479 		if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1480 		    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1481 			memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1482 		if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1483 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1484 			memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1485 	}
1486 
1487 error:
1488 	kfree(id);
1489 	return ret;
1490 }
1491 
1492 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1493 		struct nvme_ns_info *info)
1494 {
1495 	struct nvme_id_ns_cs_indep *id;
1496 	struct nvme_command c = {
1497 		.identify.opcode	= nvme_admin_identify,
1498 		.identify.nsid		= cpu_to_le32(info->nsid),
1499 		.identify.cns		= NVME_ID_CNS_NS_CS_INDEP,
1500 	};
1501 	int ret;
1502 
1503 	id = kmalloc(sizeof(*id), GFP_KERNEL);
1504 	if (!id)
1505 		return -ENOMEM;
1506 
1507 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1508 	if (!ret) {
1509 		info->anagrpid = id->anagrpid;
1510 		info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1511 		info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1512 		info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1513 	}
1514 	kfree(id);
1515 	return ret;
1516 }
1517 
1518 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1519 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1520 {
1521 	union nvme_result res = { 0 };
1522 	struct nvme_command c = { };
1523 	int ret;
1524 
1525 	c.features.opcode = op;
1526 	c.features.fid = cpu_to_le32(fid);
1527 	c.features.dword11 = cpu_to_le32(dword11);
1528 
1529 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1530 			buffer, buflen, NVME_QID_ANY, 0, 0);
1531 	if (ret >= 0 && result)
1532 		*result = le32_to_cpu(res.u32);
1533 	return ret;
1534 }
1535 
1536 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1537 		      unsigned int dword11, void *buffer, size_t buflen,
1538 		      u32 *result)
1539 {
1540 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1541 			     buflen, result);
1542 }
1543 EXPORT_SYMBOL_GPL(nvme_set_features);
1544 
1545 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1546 		      unsigned int dword11, void *buffer, size_t buflen,
1547 		      u32 *result)
1548 {
1549 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1550 			     buflen, result);
1551 }
1552 EXPORT_SYMBOL_GPL(nvme_get_features);
1553 
1554 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1555 {
1556 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1557 	u32 result;
1558 	int status, nr_io_queues;
1559 
1560 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1561 			&result);
1562 	if (status < 0)
1563 		return status;
1564 
1565 	/*
1566 	 * Degraded controllers might return an error when setting the queue
1567 	 * count.  We still want to be able to bring them online and offer
1568 	 * access to the admin queue, as that might be only way to fix them up.
1569 	 */
1570 	if (status > 0) {
1571 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1572 		*count = 0;
1573 	} else {
1574 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1575 		*count = min(*count, nr_io_queues);
1576 	}
1577 
1578 	return 0;
1579 }
1580 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1581 
1582 #define NVME_AEN_SUPPORTED \
1583 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1584 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1585 
1586 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1587 {
1588 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1589 	int status;
1590 
1591 	if (!supported_aens)
1592 		return;
1593 
1594 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1595 			NULL, 0, &result);
1596 	if (status)
1597 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1598 			 supported_aens);
1599 
1600 	queue_work(nvme_wq, &ctrl->async_event_work);
1601 }
1602 
1603 static int nvme_ns_open(struct nvme_ns *ns)
1604 {
1605 
1606 	/* should never be called due to GENHD_FL_HIDDEN */
1607 	if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1608 		goto fail;
1609 	if (!nvme_get_ns(ns))
1610 		goto fail;
1611 	if (!try_module_get(ns->ctrl->ops->module))
1612 		goto fail_put_ns;
1613 
1614 	return 0;
1615 
1616 fail_put_ns:
1617 	nvme_put_ns(ns);
1618 fail:
1619 	return -ENXIO;
1620 }
1621 
1622 static void nvme_ns_release(struct nvme_ns *ns)
1623 {
1624 
1625 	module_put(ns->ctrl->ops->module);
1626 	nvme_put_ns(ns);
1627 }
1628 
1629 static int nvme_open(struct gendisk *disk, blk_mode_t mode)
1630 {
1631 	return nvme_ns_open(disk->private_data);
1632 }
1633 
1634 static void nvme_release(struct gendisk *disk)
1635 {
1636 	nvme_ns_release(disk->private_data);
1637 }
1638 
1639 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1640 {
1641 	/* some standard values */
1642 	geo->heads = 1 << 6;
1643 	geo->sectors = 1 << 5;
1644 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1645 	return 0;
1646 }
1647 
1648 #ifdef CONFIG_BLK_DEV_INTEGRITY
1649 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1650 				u32 max_integrity_segments)
1651 {
1652 	struct blk_integrity integrity = { };
1653 
1654 	switch (ns->pi_type) {
1655 	case NVME_NS_DPS_PI_TYPE3:
1656 		switch (ns->guard_type) {
1657 		case NVME_NVM_NS_16B_GUARD:
1658 			integrity.profile = &t10_pi_type3_crc;
1659 			integrity.tag_size = sizeof(u16) + sizeof(u32);
1660 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1661 			break;
1662 		case NVME_NVM_NS_64B_GUARD:
1663 			integrity.profile = &ext_pi_type3_crc64;
1664 			integrity.tag_size = sizeof(u16) + 6;
1665 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1666 			break;
1667 		default:
1668 			integrity.profile = NULL;
1669 			break;
1670 		}
1671 		break;
1672 	case NVME_NS_DPS_PI_TYPE1:
1673 	case NVME_NS_DPS_PI_TYPE2:
1674 		switch (ns->guard_type) {
1675 		case NVME_NVM_NS_16B_GUARD:
1676 			integrity.profile = &t10_pi_type1_crc;
1677 			integrity.tag_size = sizeof(u16);
1678 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1679 			break;
1680 		case NVME_NVM_NS_64B_GUARD:
1681 			integrity.profile = &ext_pi_type1_crc64;
1682 			integrity.tag_size = sizeof(u16);
1683 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1684 			break;
1685 		default:
1686 			integrity.profile = NULL;
1687 			break;
1688 		}
1689 		break;
1690 	default:
1691 		integrity.profile = NULL;
1692 		break;
1693 	}
1694 
1695 	integrity.tuple_size = ns->ms;
1696 	blk_integrity_register(disk, &integrity);
1697 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1698 }
1699 #else
1700 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1701 				u32 max_integrity_segments)
1702 {
1703 }
1704 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1705 
1706 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1707 {
1708 	struct nvme_ctrl *ctrl = ns->ctrl;
1709 	struct request_queue *queue = disk->queue;
1710 	u32 size = queue_logical_block_size(queue);
1711 
1712 	if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1713 		ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1714 
1715 	if (ctrl->max_discard_sectors == 0) {
1716 		blk_queue_max_discard_sectors(queue, 0);
1717 		return;
1718 	}
1719 
1720 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1721 			NVME_DSM_MAX_RANGES);
1722 
1723 	queue->limits.discard_granularity = size;
1724 
1725 	/* If discard is already enabled, don't reset queue limits */
1726 	if (queue->limits.max_discard_sectors)
1727 		return;
1728 
1729 	blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1730 	blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1731 
1732 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1733 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1734 }
1735 
1736 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1737 {
1738 	return uuid_equal(&a->uuid, &b->uuid) &&
1739 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1740 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1741 		a->csi == b->csi;
1742 }
1743 
1744 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1745 {
1746 	bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1747 	unsigned lbaf = nvme_lbaf_index(id->flbas);
1748 	struct nvme_ctrl *ctrl = ns->ctrl;
1749 	struct nvme_command c = { };
1750 	struct nvme_id_ns_nvm *nvm;
1751 	int ret = 0;
1752 	u32 elbaf;
1753 
1754 	ns->pi_size = 0;
1755 	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1756 	if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1757 		ns->pi_size = sizeof(struct t10_pi_tuple);
1758 		ns->guard_type = NVME_NVM_NS_16B_GUARD;
1759 		goto set_pi;
1760 	}
1761 
1762 	nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1763 	if (!nvm)
1764 		return -ENOMEM;
1765 
1766 	c.identify.opcode = nvme_admin_identify;
1767 	c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1768 	c.identify.cns = NVME_ID_CNS_CS_NS;
1769 	c.identify.csi = NVME_CSI_NVM;
1770 
1771 	ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1772 	if (ret)
1773 		goto free_data;
1774 
1775 	elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1776 
1777 	/* no support for storage tag formats right now */
1778 	if (nvme_elbaf_sts(elbaf))
1779 		goto free_data;
1780 
1781 	ns->guard_type = nvme_elbaf_guard_type(elbaf);
1782 	switch (ns->guard_type) {
1783 	case NVME_NVM_NS_64B_GUARD:
1784 		ns->pi_size = sizeof(struct crc64_pi_tuple);
1785 		break;
1786 	case NVME_NVM_NS_16B_GUARD:
1787 		ns->pi_size = sizeof(struct t10_pi_tuple);
1788 		break;
1789 	default:
1790 		break;
1791 	}
1792 
1793 free_data:
1794 	kfree(nvm);
1795 set_pi:
1796 	if (ns->pi_size && (first || ns->ms == ns->pi_size))
1797 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1798 	else
1799 		ns->pi_type = 0;
1800 
1801 	return ret;
1802 }
1803 
1804 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1805 {
1806 	struct nvme_ctrl *ctrl = ns->ctrl;
1807 	int ret;
1808 
1809 	ret = nvme_init_ms(ns, id);
1810 	if (ret)
1811 		return ret;
1812 
1813 	ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1814 	if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1815 		return 0;
1816 
1817 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1818 		/*
1819 		 * The NVMe over Fabrics specification only supports metadata as
1820 		 * part of the extended data LBA.  We rely on HCA/HBA support to
1821 		 * remap the separate metadata buffer from the block layer.
1822 		 */
1823 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1824 			return 0;
1825 
1826 		ns->features |= NVME_NS_EXT_LBAS;
1827 
1828 		/*
1829 		 * The current fabrics transport drivers support namespace
1830 		 * metadata formats only if nvme_ns_has_pi() returns true.
1831 		 * Suppress support for all other formats so the namespace will
1832 		 * have a 0 capacity and not be usable through the block stack.
1833 		 *
1834 		 * Note, this check will need to be modified if any drivers
1835 		 * gain the ability to use other metadata formats.
1836 		 */
1837 		if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1838 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1839 	} else {
1840 		/*
1841 		 * For PCIe controllers, we can't easily remap the separate
1842 		 * metadata buffer from the block layer and thus require a
1843 		 * separate metadata buffer for block layer metadata/PI support.
1844 		 * We allow extended LBAs for the passthrough interface, though.
1845 		 */
1846 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
1847 			ns->features |= NVME_NS_EXT_LBAS;
1848 		else
1849 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1850 	}
1851 	return 0;
1852 }
1853 
1854 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1855 		struct request_queue *q)
1856 {
1857 	bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1858 
1859 	if (ctrl->max_hw_sectors) {
1860 		u32 max_segments =
1861 			(ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1862 
1863 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
1864 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1865 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1866 	}
1867 	blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1868 	blk_queue_dma_alignment(q, 3);
1869 	blk_queue_write_cache(q, vwc, vwc);
1870 }
1871 
1872 static void nvme_update_disk_info(struct gendisk *disk,
1873 		struct nvme_ns *ns, struct nvme_id_ns *id)
1874 {
1875 	sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1876 	u32 bs = 1U << ns->lba_shift;
1877 	u32 atomic_bs, phys_bs, io_opt = 0;
1878 
1879 	/*
1880 	 * The block layer can't support LBA sizes larger than the page size
1881 	 * or smaller than a sector size yet, so catch this early and don't
1882 	 * allow block I/O.
1883 	 */
1884 	if (ns->lba_shift > PAGE_SHIFT || ns->lba_shift < SECTOR_SHIFT) {
1885 		capacity = 0;
1886 		bs = (1 << 9);
1887 	}
1888 
1889 	blk_integrity_unregister(disk);
1890 
1891 	atomic_bs = phys_bs = bs;
1892 	if (id->nabo == 0) {
1893 		/*
1894 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1895 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1896 		 * 0 then AWUPF must be used instead.
1897 		 */
1898 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1899 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1900 		else
1901 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1902 	}
1903 
1904 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1905 		/* NPWG = Namespace Preferred Write Granularity */
1906 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1907 		/* NOWS = Namespace Optimal Write Size */
1908 		io_opt = bs * (1 + le16_to_cpu(id->nows));
1909 	}
1910 
1911 	blk_queue_logical_block_size(disk->queue, bs);
1912 	/*
1913 	 * Linux filesystems assume writing a single physical block is
1914 	 * an atomic operation. Hence limit the physical block size to the
1915 	 * value of the Atomic Write Unit Power Fail parameter.
1916 	 */
1917 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1918 	blk_queue_io_min(disk->queue, phys_bs);
1919 	blk_queue_io_opt(disk->queue, io_opt);
1920 
1921 	/*
1922 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
1923 	 * metadata masquerading as Type 0 if supported, otherwise reject block
1924 	 * I/O to namespaces with metadata except when the namespace supports
1925 	 * PI, as it can strip/insert in that case.
1926 	 */
1927 	if (ns->ms) {
1928 		if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1929 		    (ns->features & NVME_NS_METADATA_SUPPORTED))
1930 			nvme_init_integrity(disk, ns,
1931 					    ns->ctrl->max_integrity_segments);
1932 		else if (!nvme_ns_has_pi(ns))
1933 			capacity = 0;
1934 	}
1935 
1936 	set_capacity_and_notify(disk, capacity);
1937 
1938 	nvme_config_discard(disk, ns);
1939 	blk_queue_max_write_zeroes_sectors(disk->queue,
1940 					   ns->ctrl->max_zeroes_sectors);
1941 }
1942 
1943 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
1944 {
1945 	return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
1946 }
1947 
1948 static inline bool nvme_first_scan(struct gendisk *disk)
1949 {
1950 	/* nvme_alloc_ns() scans the disk prior to adding it */
1951 	return !disk_live(disk);
1952 }
1953 
1954 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1955 {
1956 	struct nvme_ctrl *ctrl = ns->ctrl;
1957 	u32 iob;
1958 
1959 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1960 	    is_power_of_2(ctrl->max_hw_sectors))
1961 		iob = ctrl->max_hw_sectors;
1962 	else
1963 		iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1964 
1965 	if (!iob)
1966 		return;
1967 
1968 	if (!is_power_of_2(iob)) {
1969 		if (nvme_first_scan(ns->disk))
1970 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1971 				ns->disk->disk_name, iob);
1972 		return;
1973 	}
1974 
1975 	if (blk_queue_is_zoned(ns->disk->queue)) {
1976 		if (nvme_first_scan(ns->disk))
1977 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
1978 				ns->disk->disk_name);
1979 		return;
1980 	}
1981 
1982 	blk_queue_chunk_sectors(ns->queue, iob);
1983 }
1984 
1985 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
1986 		struct nvme_ns_info *info)
1987 {
1988 	blk_mq_freeze_queue(ns->disk->queue);
1989 	nvme_set_queue_limits(ns->ctrl, ns->queue);
1990 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
1991 	blk_mq_unfreeze_queue(ns->disk->queue);
1992 
1993 	if (nvme_ns_head_multipath(ns->head)) {
1994 		blk_mq_freeze_queue(ns->head->disk->queue);
1995 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
1996 		nvme_mpath_revalidate_paths(ns);
1997 		blk_stack_limits(&ns->head->disk->queue->limits,
1998 				 &ns->queue->limits, 0);
1999 		ns->head->disk->flags |= GENHD_FL_HIDDEN;
2000 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2001 	}
2002 
2003 	/* Hide the block-interface for these devices */
2004 	ns->disk->flags |= GENHD_FL_HIDDEN;
2005 	set_bit(NVME_NS_READY, &ns->flags);
2006 
2007 	return 0;
2008 }
2009 
2010 static int nvme_update_ns_info_block(struct nvme_ns *ns,
2011 		struct nvme_ns_info *info)
2012 {
2013 	struct nvme_id_ns *id;
2014 	unsigned lbaf;
2015 	int ret;
2016 
2017 	ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
2018 	if (ret)
2019 		return ret;
2020 
2021 	if (id->ncap == 0) {
2022 		/* namespace not allocated or attached */
2023 		info->is_removed = true;
2024 		ret = -ENODEV;
2025 		goto error;
2026 	}
2027 
2028 	blk_mq_freeze_queue(ns->disk->queue);
2029 	lbaf = nvme_lbaf_index(id->flbas);
2030 	ns->lba_shift = id->lbaf[lbaf].ds;
2031 	nvme_set_queue_limits(ns->ctrl, ns->queue);
2032 
2033 	ret = nvme_configure_metadata(ns, id);
2034 	if (ret < 0) {
2035 		blk_mq_unfreeze_queue(ns->disk->queue);
2036 		goto out;
2037 	}
2038 	nvme_set_chunk_sectors(ns, id);
2039 	nvme_update_disk_info(ns->disk, ns, id);
2040 
2041 	if (ns->head->ids.csi == NVME_CSI_ZNS) {
2042 		ret = nvme_update_zone_info(ns, lbaf);
2043 		if (ret) {
2044 			blk_mq_unfreeze_queue(ns->disk->queue);
2045 			goto out;
2046 		}
2047 	}
2048 
2049 	/*
2050 	 * Only set the DEAC bit if the device guarantees that reads from
2051 	 * deallocated data return zeroes.  While the DEAC bit does not
2052 	 * require that, it must be a no-op if reads from deallocated data
2053 	 * do not return zeroes.
2054 	 */
2055 	if ((id->dlfeat & 0x7) == 0x1 && (id->dlfeat & (1 << 3)))
2056 		ns->features |= NVME_NS_DEAC;
2057 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2058 	set_bit(NVME_NS_READY, &ns->flags);
2059 	blk_mq_unfreeze_queue(ns->disk->queue);
2060 
2061 	if (blk_queue_is_zoned(ns->queue)) {
2062 		ret = nvme_revalidate_zones(ns);
2063 		if (ret && !nvme_first_scan(ns->disk))
2064 			goto out;
2065 	}
2066 
2067 	if (nvme_ns_head_multipath(ns->head)) {
2068 		blk_mq_freeze_queue(ns->head->disk->queue);
2069 		nvme_update_disk_info(ns->head->disk, ns, id);
2070 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2071 		nvme_mpath_revalidate_paths(ns);
2072 		blk_stack_limits(&ns->head->disk->queue->limits,
2073 				 &ns->queue->limits, 0);
2074 		disk_update_readahead(ns->head->disk);
2075 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2076 	}
2077 
2078 	ret = 0;
2079 out:
2080 	/*
2081 	 * If probing fails due an unsupported feature, hide the block device,
2082 	 * but still allow other access.
2083 	 */
2084 	if (ret == -ENODEV) {
2085 		ns->disk->flags |= GENHD_FL_HIDDEN;
2086 		set_bit(NVME_NS_READY, &ns->flags);
2087 		ret = 0;
2088 	}
2089 
2090 error:
2091 	kfree(id);
2092 	return ret;
2093 }
2094 
2095 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2096 {
2097 	switch (info->ids.csi) {
2098 	case NVME_CSI_ZNS:
2099 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2100 			dev_info(ns->ctrl->device,
2101 	"block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2102 				info->nsid);
2103 			return nvme_update_ns_info_generic(ns, info);
2104 		}
2105 		return nvme_update_ns_info_block(ns, info);
2106 	case NVME_CSI_NVM:
2107 		return nvme_update_ns_info_block(ns, info);
2108 	default:
2109 		dev_info(ns->ctrl->device,
2110 			"block device for nsid %u not supported (csi %u)\n",
2111 			info->nsid, info->ids.csi);
2112 		return nvme_update_ns_info_generic(ns, info);
2113 	}
2114 }
2115 
2116 #ifdef CONFIG_BLK_SED_OPAL
2117 static int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2118 		bool send)
2119 {
2120 	struct nvme_ctrl *ctrl = data;
2121 	struct nvme_command cmd = { };
2122 
2123 	if (send)
2124 		cmd.common.opcode = nvme_admin_security_send;
2125 	else
2126 		cmd.common.opcode = nvme_admin_security_recv;
2127 	cmd.common.nsid = 0;
2128 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2129 	cmd.common.cdw11 = cpu_to_le32(len);
2130 
2131 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2132 			NVME_QID_ANY, 1, 0);
2133 }
2134 
2135 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2136 {
2137 	if (ctrl->oacs & NVME_CTRL_OACS_SEC_SUPP) {
2138 		if (!ctrl->opal_dev)
2139 			ctrl->opal_dev = init_opal_dev(ctrl, &nvme_sec_submit);
2140 		else if (was_suspended)
2141 			opal_unlock_from_suspend(ctrl->opal_dev);
2142 	} else {
2143 		free_opal_dev(ctrl->opal_dev);
2144 		ctrl->opal_dev = NULL;
2145 	}
2146 }
2147 #else
2148 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2149 {
2150 }
2151 #endif /* CONFIG_BLK_SED_OPAL */
2152 
2153 #ifdef CONFIG_BLK_DEV_ZONED
2154 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2155 		unsigned int nr_zones, report_zones_cb cb, void *data)
2156 {
2157 	return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2158 			data);
2159 }
2160 #else
2161 #define nvme_report_zones	NULL
2162 #endif /* CONFIG_BLK_DEV_ZONED */
2163 
2164 const struct block_device_operations nvme_bdev_ops = {
2165 	.owner		= THIS_MODULE,
2166 	.ioctl		= nvme_ioctl,
2167 	.compat_ioctl	= blkdev_compat_ptr_ioctl,
2168 	.open		= nvme_open,
2169 	.release	= nvme_release,
2170 	.getgeo		= nvme_getgeo,
2171 	.report_zones	= nvme_report_zones,
2172 	.pr_ops		= &nvme_pr_ops,
2173 };
2174 
2175 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 mask, u32 val,
2176 		u32 timeout, const char *op)
2177 {
2178 	unsigned long timeout_jiffies = jiffies + timeout * HZ;
2179 	u32 csts;
2180 	int ret;
2181 
2182 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2183 		if (csts == ~0)
2184 			return -ENODEV;
2185 		if ((csts & mask) == val)
2186 			break;
2187 
2188 		usleep_range(1000, 2000);
2189 		if (fatal_signal_pending(current))
2190 			return -EINTR;
2191 		if (time_after(jiffies, timeout_jiffies)) {
2192 			dev_err(ctrl->device,
2193 				"Device not ready; aborting %s, CSTS=0x%x\n",
2194 				op, csts);
2195 			return -ENODEV;
2196 		}
2197 	}
2198 
2199 	return ret;
2200 }
2201 
2202 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown)
2203 {
2204 	int ret;
2205 
2206 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2207 	if (shutdown)
2208 		ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2209 	else
2210 		ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2211 
2212 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2213 	if (ret)
2214 		return ret;
2215 
2216 	if (shutdown) {
2217 		return nvme_wait_ready(ctrl, NVME_CSTS_SHST_MASK,
2218 				       NVME_CSTS_SHST_CMPLT,
2219 				       ctrl->shutdown_timeout, "shutdown");
2220 	}
2221 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2222 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2223 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, 0,
2224 			       (NVME_CAP_TIMEOUT(ctrl->cap) + 1) / 2, "reset");
2225 }
2226 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2227 
2228 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2229 {
2230 	unsigned dev_page_min;
2231 	u32 timeout;
2232 	int ret;
2233 
2234 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2235 	if (ret) {
2236 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2237 		return ret;
2238 	}
2239 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2240 
2241 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2242 		dev_err(ctrl->device,
2243 			"Minimum device page size %u too large for host (%u)\n",
2244 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2245 		return -ENODEV;
2246 	}
2247 
2248 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2249 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2250 	else
2251 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2252 
2253 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS && ctrl->cap & NVME_CAP_CRMS_CRIMS)
2254 		ctrl->ctrl_config |= NVME_CC_CRIME;
2255 
2256 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2257 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2258 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2259 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2260 	if (ret)
2261 		return ret;
2262 
2263 	/* Flush write to device (required if transport is PCI) */
2264 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2265 	if (ret)
2266 		return ret;
2267 
2268 	/* CAP value may change after initial CC write */
2269 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2270 	if (ret)
2271 		return ret;
2272 
2273 	timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2274 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2275 		u32 crto, ready_timeout;
2276 
2277 		ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2278 		if (ret) {
2279 			dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2280 				ret);
2281 			return ret;
2282 		}
2283 
2284 		/*
2285 		 * CRTO should always be greater or equal to CAP.TO, but some
2286 		 * devices are known to get this wrong. Use the larger of the
2287 		 * two values.
2288 		 */
2289 		if (ctrl->ctrl_config & NVME_CC_CRIME)
2290 			ready_timeout = NVME_CRTO_CRIMT(crto);
2291 		else
2292 			ready_timeout = NVME_CRTO_CRWMT(crto);
2293 
2294 		if (ready_timeout < timeout)
2295 			dev_warn_once(ctrl->device, "bad crto:%x cap:%llx\n",
2296 				      crto, ctrl->cap);
2297 		else
2298 			timeout = ready_timeout;
2299 	}
2300 
2301 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2302 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2303 	if (ret)
2304 		return ret;
2305 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, NVME_CSTS_RDY,
2306 			       (timeout + 1) / 2, "initialisation");
2307 }
2308 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2309 
2310 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2311 {
2312 	__le64 ts;
2313 	int ret;
2314 
2315 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2316 		return 0;
2317 
2318 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2319 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2320 			NULL);
2321 	if (ret)
2322 		dev_warn_once(ctrl->device,
2323 			"could not set timestamp (%d)\n", ret);
2324 	return ret;
2325 }
2326 
2327 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2328 {
2329 	struct nvme_feat_host_behavior *host;
2330 	u8 acre = 0, lbafee = 0;
2331 	int ret;
2332 
2333 	/* Don't bother enabling the feature if retry delay is not reported */
2334 	if (ctrl->crdt[0])
2335 		acre = NVME_ENABLE_ACRE;
2336 	if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2337 		lbafee = NVME_ENABLE_LBAFEE;
2338 
2339 	if (!acre && !lbafee)
2340 		return 0;
2341 
2342 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2343 	if (!host)
2344 		return 0;
2345 
2346 	host->acre = acre;
2347 	host->lbafee = lbafee;
2348 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2349 				host, sizeof(*host), NULL);
2350 	kfree(host);
2351 	return ret;
2352 }
2353 
2354 /*
2355  * The function checks whether the given total (exlat + enlat) latency of
2356  * a power state allows the latter to be used as an APST transition target.
2357  * It does so by comparing the latency to the primary and secondary latency
2358  * tolerances defined by module params. If there's a match, the corresponding
2359  * timeout value is returned and the matching tolerance index (1 or 2) is
2360  * reported.
2361  */
2362 static bool nvme_apst_get_transition_time(u64 total_latency,
2363 		u64 *transition_time, unsigned *last_index)
2364 {
2365 	if (total_latency <= apst_primary_latency_tol_us) {
2366 		if (*last_index == 1)
2367 			return false;
2368 		*last_index = 1;
2369 		*transition_time = apst_primary_timeout_ms;
2370 		return true;
2371 	}
2372 	if (apst_secondary_timeout_ms &&
2373 		total_latency <= apst_secondary_latency_tol_us) {
2374 		if (*last_index <= 2)
2375 			return false;
2376 		*last_index = 2;
2377 		*transition_time = apst_secondary_timeout_ms;
2378 		return true;
2379 	}
2380 	return false;
2381 }
2382 
2383 /*
2384  * APST (Autonomous Power State Transition) lets us program a table of power
2385  * state transitions that the controller will perform automatically.
2386  *
2387  * Depending on module params, one of the two supported techniques will be used:
2388  *
2389  * - If the parameters provide explicit timeouts and tolerances, they will be
2390  *   used to build a table with up to 2 non-operational states to transition to.
2391  *   The default parameter values were selected based on the values used by
2392  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2393  *   regeneration of the APST table in the event of switching between external
2394  *   and battery power, the timeouts and tolerances reflect a compromise
2395  *   between values used by Microsoft for AC and battery scenarios.
2396  * - If not, we'll configure the table with a simple heuristic: we are willing
2397  *   to spend at most 2% of the time transitioning between power states.
2398  *   Therefore, when running in any given state, we will enter the next
2399  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2400  *   microseconds, as long as that state's exit latency is under the requested
2401  *   maximum latency.
2402  *
2403  * We will not autonomously enter any non-operational state for which the total
2404  * latency exceeds ps_max_latency_us.
2405  *
2406  * Users can set ps_max_latency_us to zero to turn off APST.
2407  */
2408 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2409 {
2410 	struct nvme_feat_auto_pst *table;
2411 	unsigned apste = 0;
2412 	u64 max_lat_us = 0;
2413 	__le64 target = 0;
2414 	int max_ps = -1;
2415 	int state;
2416 	int ret;
2417 	unsigned last_lt_index = UINT_MAX;
2418 
2419 	/*
2420 	 * If APST isn't supported or if we haven't been initialized yet,
2421 	 * then don't do anything.
2422 	 */
2423 	if (!ctrl->apsta)
2424 		return 0;
2425 
2426 	if (ctrl->npss > 31) {
2427 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2428 		return 0;
2429 	}
2430 
2431 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2432 	if (!table)
2433 		return 0;
2434 
2435 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2436 		/* Turn off APST. */
2437 		dev_dbg(ctrl->device, "APST disabled\n");
2438 		goto done;
2439 	}
2440 
2441 	/*
2442 	 * Walk through all states from lowest- to highest-power.
2443 	 * According to the spec, lower-numbered states use more power.  NPSS,
2444 	 * despite the name, is the index of the lowest-power state, not the
2445 	 * number of states.
2446 	 */
2447 	for (state = (int)ctrl->npss; state >= 0; state--) {
2448 		u64 total_latency_us, exit_latency_us, transition_ms;
2449 
2450 		if (target)
2451 			table->entries[state] = target;
2452 
2453 		/*
2454 		 * Don't allow transitions to the deepest state if it's quirked
2455 		 * off.
2456 		 */
2457 		if (state == ctrl->npss &&
2458 		    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2459 			continue;
2460 
2461 		/*
2462 		 * Is this state a useful non-operational state for higher-power
2463 		 * states to autonomously transition to?
2464 		 */
2465 		if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2466 			continue;
2467 
2468 		exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2469 		if (exit_latency_us > ctrl->ps_max_latency_us)
2470 			continue;
2471 
2472 		total_latency_us = exit_latency_us +
2473 			le32_to_cpu(ctrl->psd[state].entry_lat);
2474 
2475 		/*
2476 		 * This state is good. It can be used as the APST idle target
2477 		 * for higher power states.
2478 		 */
2479 		if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2480 			if (!nvme_apst_get_transition_time(total_latency_us,
2481 					&transition_ms, &last_lt_index))
2482 				continue;
2483 		} else {
2484 			transition_ms = total_latency_us + 19;
2485 			do_div(transition_ms, 20);
2486 			if (transition_ms > (1 << 24) - 1)
2487 				transition_ms = (1 << 24) - 1;
2488 		}
2489 
2490 		target = cpu_to_le64((state << 3) | (transition_ms << 8));
2491 		if (max_ps == -1)
2492 			max_ps = state;
2493 		if (total_latency_us > max_lat_us)
2494 			max_lat_us = total_latency_us;
2495 	}
2496 
2497 	if (max_ps == -1)
2498 		dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2499 	else
2500 		dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2501 			max_ps, max_lat_us, (int)sizeof(*table), table);
2502 	apste = 1;
2503 
2504 done:
2505 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2506 				table, sizeof(*table), NULL);
2507 	if (ret)
2508 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2509 	kfree(table);
2510 	return ret;
2511 }
2512 
2513 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2514 {
2515 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2516 	u64 latency;
2517 
2518 	switch (val) {
2519 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2520 	case PM_QOS_LATENCY_ANY:
2521 		latency = U64_MAX;
2522 		break;
2523 
2524 	default:
2525 		latency = val;
2526 	}
2527 
2528 	if (ctrl->ps_max_latency_us != latency) {
2529 		ctrl->ps_max_latency_us = latency;
2530 		if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
2531 			nvme_configure_apst(ctrl);
2532 	}
2533 }
2534 
2535 struct nvme_core_quirk_entry {
2536 	/*
2537 	 * NVMe model and firmware strings are padded with spaces.  For
2538 	 * simplicity, strings in the quirk table are padded with NULLs
2539 	 * instead.
2540 	 */
2541 	u16 vid;
2542 	const char *mn;
2543 	const char *fr;
2544 	unsigned long quirks;
2545 };
2546 
2547 static const struct nvme_core_quirk_entry core_quirks[] = {
2548 	{
2549 		/*
2550 		 * This Toshiba device seems to die using any APST states.  See:
2551 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2552 		 */
2553 		.vid = 0x1179,
2554 		.mn = "THNSF5256GPUK TOSHIBA",
2555 		.quirks = NVME_QUIRK_NO_APST,
2556 	},
2557 	{
2558 		/*
2559 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2560 		 * condition associated with actions related to suspend to idle
2561 		 * LiteON has resolved the problem in future firmware
2562 		 */
2563 		.vid = 0x14a4,
2564 		.fr = "22301111",
2565 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2566 	},
2567 	{
2568 		/*
2569 		 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2570 		 * aborts I/O during any load, but more easily reproducible
2571 		 * with discards (fstrim).
2572 		 *
2573 		 * The device is left in a state where it is also not possible
2574 		 * to use "nvme set-feature" to disable APST, but booting with
2575 		 * nvme_core.default_ps_max_latency=0 works.
2576 		 */
2577 		.vid = 0x1e0f,
2578 		.mn = "KCD6XVUL6T40",
2579 		.quirks = NVME_QUIRK_NO_APST,
2580 	},
2581 	{
2582 		/*
2583 		 * The external Samsung X5 SSD fails initialization without a
2584 		 * delay before checking if it is ready and has a whole set of
2585 		 * other problems.  To make this even more interesting, it
2586 		 * shares the PCI ID with internal Samsung 970 Evo Plus that
2587 		 * does not need or want these quirks.
2588 		 */
2589 		.vid = 0x144d,
2590 		.mn = "Samsung Portable SSD X5",
2591 		.quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2592 			  NVME_QUIRK_NO_DEEPEST_PS |
2593 			  NVME_QUIRK_IGNORE_DEV_SUBNQN,
2594 	}
2595 };
2596 
2597 /* match is null-terminated but idstr is space-padded. */
2598 static bool string_matches(const char *idstr, const char *match, size_t len)
2599 {
2600 	size_t matchlen;
2601 
2602 	if (!match)
2603 		return true;
2604 
2605 	matchlen = strlen(match);
2606 	WARN_ON_ONCE(matchlen > len);
2607 
2608 	if (memcmp(idstr, match, matchlen))
2609 		return false;
2610 
2611 	for (; matchlen < len; matchlen++)
2612 		if (idstr[matchlen] != ' ')
2613 			return false;
2614 
2615 	return true;
2616 }
2617 
2618 static bool quirk_matches(const struct nvme_id_ctrl *id,
2619 			  const struct nvme_core_quirk_entry *q)
2620 {
2621 	return q->vid == le16_to_cpu(id->vid) &&
2622 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2623 		string_matches(id->fr, q->fr, sizeof(id->fr));
2624 }
2625 
2626 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2627 		struct nvme_id_ctrl *id)
2628 {
2629 	size_t nqnlen;
2630 	int off;
2631 
2632 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2633 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2634 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2635 			strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2636 			return;
2637 		}
2638 
2639 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2640 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2641 	}
2642 
2643 	/*
2644 	 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2645 	 * Base Specification 2.0.  It is slightly different from the format
2646 	 * specified there due to historic reasons, and we can't change it now.
2647 	 */
2648 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2649 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2650 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2651 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2652 	off += sizeof(id->sn);
2653 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2654 	off += sizeof(id->mn);
2655 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2656 }
2657 
2658 static void nvme_release_subsystem(struct device *dev)
2659 {
2660 	struct nvme_subsystem *subsys =
2661 		container_of(dev, struct nvme_subsystem, dev);
2662 
2663 	if (subsys->instance >= 0)
2664 		ida_free(&nvme_instance_ida, subsys->instance);
2665 	kfree(subsys);
2666 }
2667 
2668 static void nvme_destroy_subsystem(struct kref *ref)
2669 {
2670 	struct nvme_subsystem *subsys =
2671 			container_of(ref, struct nvme_subsystem, ref);
2672 
2673 	mutex_lock(&nvme_subsystems_lock);
2674 	list_del(&subsys->entry);
2675 	mutex_unlock(&nvme_subsystems_lock);
2676 
2677 	ida_destroy(&subsys->ns_ida);
2678 	device_del(&subsys->dev);
2679 	put_device(&subsys->dev);
2680 }
2681 
2682 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2683 {
2684 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2685 }
2686 
2687 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2688 {
2689 	struct nvme_subsystem *subsys;
2690 
2691 	lockdep_assert_held(&nvme_subsystems_lock);
2692 
2693 	/*
2694 	 * Fail matches for discovery subsystems. This results
2695 	 * in each discovery controller bound to a unique subsystem.
2696 	 * This avoids issues with validating controller values
2697 	 * that can only be true when there is a single unique subsystem.
2698 	 * There may be multiple and completely independent entities
2699 	 * that provide discovery controllers.
2700 	 */
2701 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2702 		return NULL;
2703 
2704 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2705 		if (strcmp(subsys->subnqn, subsysnqn))
2706 			continue;
2707 		if (!kref_get_unless_zero(&subsys->ref))
2708 			continue;
2709 		return subsys;
2710 	}
2711 
2712 	return NULL;
2713 }
2714 
2715 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2716 {
2717 	return ctrl->opts && ctrl->opts->discovery_nqn;
2718 }
2719 
2720 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2721 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2722 {
2723 	struct nvme_ctrl *tmp;
2724 
2725 	lockdep_assert_held(&nvme_subsystems_lock);
2726 
2727 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2728 		if (nvme_state_terminal(tmp))
2729 			continue;
2730 
2731 		if (tmp->cntlid == ctrl->cntlid) {
2732 			dev_err(ctrl->device,
2733 				"Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2734 				ctrl->cntlid, dev_name(tmp->device),
2735 				subsys->subnqn);
2736 			return false;
2737 		}
2738 
2739 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2740 		    nvme_discovery_ctrl(ctrl))
2741 			continue;
2742 
2743 		dev_err(ctrl->device,
2744 			"Subsystem does not support multiple controllers\n");
2745 		return false;
2746 	}
2747 
2748 	return true;
2749 }
2750 
2751 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2752 {
2753 	struct nvme_subsystem *subsys, *found;
2754 	int ret;
2755 
2756 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2757 	if (!subsys)
2758 		return -ENOMEM;
2759 
2760 	subsys->instance = -1;
2761 	mutex_init(&subsys->lock);
2762 	kref_init(&subsys->ref);
2763 	INIT_LIST_HEAD(&subsys->ctrls);
2764 	INIT_LIST_HEAD(&subsys->nsheads);
2765 	nvme_init_subnqn(subsys, ctrl, id);
2766 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2767 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2768 	subsys->vendor_id = le16_to_cpu(id->vid);
2769 	subsys->cmic = id->cmic;
2770 
2771 	/* Versions prior to 1.4 don't necessarily report a valid type */
2772 	if (id->cntrltype == NVME_CTRL_DISC ||
2773 	    !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2774 		subsys->subtype = NVME_NQN_DISC;
2775 	else
2776 		subsys->subtype = NVME_NQN_NVME;
2777 
2778 	if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2779 		dev_err(ctrl->device,
2780 			"Subsystem %s is not a discovery controller",
2781 			subsys->subnqn);
2782 		kfree(subsys);
2783 		return -EINVAL;
2784 	}
2785 	subsys->awupf = le16_to_cpu(id->awupf);
2786 	nvme_mpath_default_iopolicy(subsys);
2787 
2788 	subsys->dev.class = nvme_subsys_class;
2789 	subsys->dev.release = nvme_release_subsystem;
2790 	subsys->dev.groups = nvme_subsys_attrs_groups;
2791 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2792 	device_initialize(&subsys->dev);
2793 
2794 	mutex_lock(&nvme_subsystems_lock);
2795 	found = __nvme_find_get_subsystem(subsys->subnqn);
2796 	if (found) {
2797 		put_device(&subsys->dev);
2798 		subsys = found;
2799 
2800 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2801 			ret = -EINVAL;
2802 			goto out_put_subsystem;
2803 		}
2804 	} else {
2805 		ret = device_add(&subsys->dev);
2806 		if (ret) {
2807 			dev_err(ctrl->device,
2808 				"failed to register subsystem device.\n");
2809 			put_device(&subsys->dev);
2810 			goto out_unlock;
2811 		}
2812 		ida_init(&subsys->ns_ida);
2813 		list_add_tail(&subsys->entry, &nvme_subsystems);
2814 	}
2815 
2816 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2817 				dev_name(ctrl->device));
2818 	if (ret) {
2819 		dev_err(ctrl->device,
2820 			"failed to create sysfs link from subsystem.\n");
2821 		goto out_put_subsystem;
2822 	}
2823 
2824 	if (!found)
2825 		subsys->instance = ctrl->instance;
2826 	ctrl->subsys = subsys;
2827 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2828 	mutex_unlock(&nvme_subsystems_lock);
2829 	return 0;
2830 
2831 out_put_subsystem:
2832 	nvme_put_subsystem(subsys);
2833 out_unlock:
2834 	mutex_unlock(&nvme_subsystems_lock);
2835 	return ret;
2836 }
2837 
2838 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2839 		void *log, size_t size, u64 offset)
2840 {
2841 	struct nvme_command c = { };
2842 	u32 dwlen = nvme_bytes_to_numd(size);
2843 
2844 	c.get_log_page.opcode = nvme_admin_get_log_page;
2845 	c.get_log_page.nsid = cpu_to_le32(nsid);
2846 	c.get_log_page.lid = log_page;
2847 	c.get_log_page.lsp = lsp;
2848 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2849 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2850 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2851 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2852 	c.get_log_page.csi = csi;
2853 
2854 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2855 }
2856 
2857 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2858 				struct nvme_effects_log **log)
2859 {
2860 	struct nvme_effects_log	*cel = xa_load(&ctrl->cels, csi);
2861 	int ret;
2862 
2863 	if (cel)
2864 		goto out;
2865 
2866 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2867 	if (!cel)
2868 		return -ENOMEM;
2869 
2870 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2871 			cel, sizeof(*cel), 0);
2872 	if (ret) {
2873 		kfree(cel);
2874 		return ret;
2875 	}
2876 
2877 	xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2878 out:
2879 	*log = cel;
2880 	return 0;
2881 }
2882 
2883 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2884 {
2885 	u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2886 
2887 	if (check_shl_overflow(1U, units + page_shift - 9, &val))
2888 		return UINT_MAX;
2889 	return val;
2890 }
2891 
2892 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2893 {
2894 	struct nvme_command c = { };
2895 	struct nvme_id_ctrl_nvm *id;
2896 	int ret;
2897 
2898 	if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2899 		ctrl->max_discard_sectors = UINT_MAX;
2900 		ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2901 	} else {
2902 		ctrl->max_discard_sectors = 0;
2903 		ctrl->max_discard_segments = 0;
2904 	}
2905 
2906 	/*
2907 	 * Even though NVMe spec explicitly states that MDTS is not applicable
2908 	 * to the write-zeroes, we are cautious and limit the size to the
2909 	 * controllers max_hw_sectors value, which is based on the MDTS field
2910 	 * and possibly other limiting factors.
2911 	 */
2912 	if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2913 	    !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2914 		ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2915 	else
2916 		ctrl->max_zeroes_sectors = 0;
2917 
2918 	if (ctrl->subsys->subtype != NVME_NQN_NVME ||
2919 	    nvme_ctrl_limited_cns(ctrl) ||
2920 	    test_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags))
2921 		return 0;
2922 
2923 	id = kzalloc(sizeof(*id), GFP_KERNEL);
2924 	if (!id)
2925 		return -ENOMEM;
2926 
2927 	c.identify.opcode = nvme_admin_identify;
2928 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
2929 	c.identify.csi = NVME_CSI_NVM;
2930 
2931 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2932 	if (ret)
2933 		goto free_data;
2934 
2935 	if (id->dmrl)
2936 		ctrl->max_discard_segments = id->dmrl;
2937 	ctrl->dmrsl = le32_to_cpu(id->dmrsl);
2938 	if (id->wzsl)
2939 		ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2940 
2941 free_data:
2942 	if (ret > 0)
2943 		set_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags);
2944 	kfree(id);
2945 	return ret;
2946 }
2947 
2948 static void nvme_init_known_nvm_effects(struct nvme_ctrl *ctrl)
2949 {
2950 	struct nvme_effects_log	*log = ctrl->effects;
2951 
2952 	log->acs[nvme_admin_format_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
2953 						NVME_CMD_EFFECTS_NCC |
2954 						NVME_CMD_EFFECTS_CSE_MASK);
2955 	log->acs[nvme_admin_sanitize_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
2956 						NVME_CMD_EFFECTS_CSE_MASK);
2957 
2958 	/*
2959 	 * The spec says the result of a security receive command depends on
2960 	 * the previous security send command. As such, many vendors log this
2961 	 * command as one to submitted only when no other commands to the same
2962 	 * namespace are outstanding. The intention is to tell the host to
2963 	 * prevent mixing security send and receive.
2964 	 *
2965 	 * This driver can only enforce such exclusive access against IO
2966 	 * queues, though. We are not readily able to enforce such a rule for
2967 	 * two commands to the admin queue, which is the only queue that
2968 	 * matters for this command.
2969 	 *
2970 	 * Rather than blindly freezing the IO queues for this effect that
2971 	 * doesn't even apply to IO, mask it off.
2972 	 */
2973 	log->acs[nvme_admin_security_recv] &= cpu_to_le32(~NVME_CMD_EFFECTS_CSE_MASK);
2974 
2975 	log->iocs[nvme_cmd_write] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
2976 	log->iocs[nvme_cmd_write_zeroes] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
2977 	log->iocs[nvme_cmd_write_uncor] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
2978 }
2979 
2980 static int nvme_init_effects(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2981 {
2982 	int ret = 0;
2983 
2984 	if (ctrl->effects)
2985 		return 0;
2986 
2987 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2988 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2989 		if (ret < 0)
2990 			return ret;
2991 	}
2992 
2993 	if (!ctrl->effects) {
2994 		ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2995 		if (!ctrl->effects)
2996 			return -ENOMEM;
2997 		xa_store(&ctrl->cels, NVME_CSI_NVM, ctrl->effects, GFP_KERNEL);
2998 	}
2999 
3000 	nvme_init_known_nvm_effects(ctrl);
3001 	return 0;
3002 }
3003 
3004 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3005 {
3006 	struct nvme_id_ctrl *id;
3007 	u32 max_hw_sectors;
3008 	bool prev_apst_enabled;
3009 	int ret;
3010 
3011 	ret = nvme_identify_ctrl(ctrl, &id);
3012 	if (ret) {
3013 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3014 		return -EIO;
3015 	}
3016 
3017 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
3018 		ctrl->cntlid = le16_to_cpu(id->cntlid);
3019 
3020 	if (!ctrl->identified) {
3021 		unsigned int i;
3022 
3023 		/*
3024 		 * Check for quirks.  Quirk can depend on firmware version,
3025 		 * so, in principle, the set of quirks present can change
3026 		 * across a reset.  As a possible future enhancement, we
3027 		 * could re-scan for quirks every time we reinitialize
3028 		 * the device, but we'd have to make sure that the driver
3029 		 * behaves intelligently if the quirks change.
3030 		 */
3031 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3032 			if (quirk_matches(id, &core_quirks[i]))
3033 				ctrl->quirks |= core_quirks[i].quirks;
3034 		}
3035 
3036 		ret = nvme_init_subsystem(ctrl, id);
3037 		if (ret)
3038 			goto out_free;
3039 
3040 		ret = nvme_init_effects(ctrl, id);
3041 		if (ret)
3042 			goto out_free;
3043 	}
3044 	memcpy(ctrl->subsys->firmware_rev, id->fr,
3045 	       sizeof(ctrl->subsys->firmware_rev));
3046 
3047 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3048 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3049 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3050 	}
3051 
3052 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3053 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3054 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3055 
3056 	ctrl->oacs = le16_to_cpu(id->oacs);
3057 	ctrl->oncs = le16_to_cpu(id->oncs);
3058 	ctrl->mtfa = le16_to_cpu(id->mtfa);
3059 	ctrl->oaes = le32_to_cpu(id->oaes);
3060 	ctrl->wctemp = le16_to_cpu(id->wctemp);
3061 	ctrl->cctemp = le16_to_cpu(id->cctemp);
3062 
3063 	atomic_set(&ctrl->abort_limit, id->acl + 1);
3064 	ctrl->vwc = id->vwc;
3065 	if (id->mdts)
3066 		max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3067 	else
3068 		max_hw_sectors = UINT_MAX;
3069 	ctrl->max_hw_sectors =
3070 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3071 
3072 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
3073 	ctrl->sgls = le32_to_cpu(id->sgls);
3074 	ctrl->kas = le16_to_cpu(id->kas);
3075 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
3076 	ctrl->ctratt = le32_to_cpu(id->ctratt);
3077 
3078 	ctrl->cntrltype = id->cntrltype;
3079 	ctrl->dctype = id->dctype;
3080 
3081 	if (id->rtd3e) {
3082 		/* us -> s */
3083 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3084 
3085 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3086 						 shutdown_timeout, 60);
3087 
3088 		if (ctrl->shutdown_timeout != shutdown_timeout)
3089 			dev_info(ctrl->device,
3090 				 "Shutdown timeout set to %u seconds\n",
3091 				 ctrl->shutdown_timeout);
3092 	} else
3093 		ctrl->shutdown_timeout = shutdown_timeout;
3094 
3095 	ctrl->npss = id->npss;
3096 	ctrl->apsta = id->apsta;
3097 	prev_apst_enabled = ctrl->apst_enabled;
3098 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3099 		if (force_apst && id->apsta) {
3100 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3101 			ctrl->apst_enabled = true;
3102 		} else {
3103 			ctrl->apst_enabled = false;
3104 		}
3105 	} else {
3106 		ctrl->apst_enabled = id->apsta;
3107 	}
3108 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3109 
3110 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3111 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3112 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3113 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3114 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3115 
3116 		/*
3117 		 * In fabrics we need to verify the cntlid matches the
3118 		 * admin connect
3119 		 */
3120 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3121 			dev_err(ctrl->device,
3122 				"Mismatching cntlid: Connect %u vs Identify "
3123 				"%u, rejecting\n",
3124 				ctrl->cntlid, le16_to_cpu(id->cntlid));
3125 			ret = -EINVAL;
3126 			goto out_free;
3127 		}
3128 
3129 		if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3130 			dev_err(ctrl->device,
3131 				"keep-alive support is mandatory for fabrics\n");
3132 			ret = -EINVAL;
3133 			goto out_free;
3134 		}
3135 	} else {
3136 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3137 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3138 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3139 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3140 	}
3141 
3142 	ret = nvme_mpath_init_identify(ctrl, id);
3143 	if (ret < 0)
3144 		goto out_free;
3145 
3146 	if (ctrl->apst_enabled && !prev_apst_enabled)
3147 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3148 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3149 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3150 
3151 out_free:
3152 	kfree(id);
3153 	return ret;
3154 }
3155 
3156 /*
3157  * Initialize the cached copies of the Identify data and various controller
3158  * register in our nvme_ctrl structure.  This should be called as soon as
3159  * the admin queue is fully up and running.
3160  */
3161 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended)
3162 {
3163 	int ret;
3164 
3165 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3166 	if (ret) {
3167 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3168 		return ret;
3169 	}
3170 
3171 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3172 
3173 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3174 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3175 
3176 	ret = nvme_init_identify(ctrl);
3177 	if (ret)
3178 		return ret;
3179 
3180 	ret = nvme_configure_apst(ctrl);
3181 	if (ret < 0)
3182 		return ret;
3183 
3184 	ret = nvme_configure_timestamp(ctrl);
3185 	if (ret < 0)
3186 		return ret;
3187 
3188 	ret = nvme_configure_host_options(ctrl);
3189 	if (ret < 0)
3190 		return ret;
3191 
3192 	nvme_configure_opal(ctrl, was_suspended);
3193 
3194 	if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3195 		/*
3196 		 * Do not return errors unless we are in a controller reset,
3197 		 * the controller works perfectly fine without hwmon.
3198 		 */
3199 		ret = nvme_hwmon_init(ctrl);
3200 		if (ret == -EINTR)
3201 			return ret;
3202 	}
3203 
3204 	clear_bit(NVME_CTRL_DIRTY_CAPABILITY, &ctrl->flags);
3205 	ctrl->identified = true;
3206 
3207 	return 0;
3208 }
3209 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3210 
3211 static int nvme_dev_open(struct inode *inode, struct file *file)
3212 {
3213 	struct nvme_ctrl *ctrl =
3214 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3215 
3216 	switch (nvme_ctrl_state(ctrl)) {
3217 	case NVME_CTRL_LIVE:
3218 		break;
3219 	default:
3220 		return -EWOULDBLOCK;
3221 	}
3222 
3223 	nvme_get_ctrl(ctrl);
3224 	if (!try_module_get(ctrl->ops->module)) {
3225 		nvme_put_ctrl(ctrl);
3226 		return -EINVAL;
3227 	}
3228 
3229 	file->private_data = ctrl;
3230 	return 0;
3231 }
3232 
3233 static int nvme_dev_release(struct inode *inode, struct file *file)
3234 {
3235 	struct nvme_ctrl *ctrl =
3236 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3237 
3238 	module_put(ctrl->ops->module);
3239 	nvme_put_ctrl(ctrl);
3240 	return 0;
3241 }
3242 
3243 static const struct file_operations nvme_dev_fops = {
3244 	.owner		= THIS_MODULE,
3245 	.open		= nvme_dev_open,
3246 	.release	= nvme_dev_release,
3247 	.unlocked_ioctl	= nvme_dev_ioctl,
3248 	.compat_ioctl	= compat_ptr_ioctl,
3249 	.uring_cmd	= nvme_dev_uring_cmd,
3250 };
3251 
3252 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3253 		unsigned nsid)
3254 {
3255 	struct nvme_ns_head *h;
3256 
3257 	lockdep_assert_held(&ctrl->subsys->lock);
3258 
3259 	list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3260 		/*
3261 		 * Private namespaces can share NSIDs under some conditions.
3262 		 * In that case we can't use the same ns_head for namespaces
3263 		 * with the same NSID.
3264 		 */
3265 		if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3266 			continue;
3267 		if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3268 			return h;
3269 	}
3270 
3271 	return NULL;
3272 }
3273 
3274 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3275 		struct nvme_ns_ids *ids)
3276 {
3277 	bool has_uuid = !uuid_is_null(&ids->uuid);
3278 	bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3279 	bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3280 	struct nvme_ns_head *h;
3281 
3282 	lockdep_assert_held(&subsys->lock);
3283 
3284 	list_for_each_entry(h, &subsys->nsheads, entry) {
3285 		if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3286 			return -EINVAL;
3287 		if (has_nguid &&
3288 		    memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3289 			return -EINVAL;
3290 		if (has_eui64 &&
3291 		    memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3292 			return -EINVAL;
3293 	}
3294 
3295 	return 0;
3296 }
3297 
3298 static void nvme_cdev_rel(struct device *dev)
3299 {
3300 	ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3301 }
3302 
3303 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3304 {
3305 	cdev_device_del(cdev, cdev_device);
3306 	put_device(cdev_device);
3307 }
3308 
3309 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3310 		const struct file_operations *fops, struct module *owner)
3311 {
3312 	int minor, ret;
3313 
3314 	minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3315 	if (minor < 0)
3316 		return minor;
3317 	cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3318 	cdev_device->class = nvme_ns_chr_class;
3319 	cdev_device->release = nvme_cdev_rel;
3320 	device_initialize(cdev_device);
3321 	cdev_init(cdev, fops);
3322 	cdev->owner = owner;
3323 	ret = cdev_device_add(cdev, cdev_device);
3324 	if (ret)
3325 		put_device(cdev_device);
3326 
3327 	return ret;
3328 }
3329 
3330 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3331 {
3332 	return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3333 }
3334 
3335 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3336 {
3337 	nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3338 	return 0;
3339 }
3340 
3341 static const struct file_operations nvme_ns_chr_fops = {
3342 	.owner		= THIS_MODULE,
3343 	.open		= nvme_ns_chr_open,
3344 	.release	= nvme_ns_chr_release,
3345 	.unlocked_ioctl	= nvme_ns_chr_ioctl,
3346 	.compat_ioctl	= compat_ptr_ioctl,
3347 	.uring_cmd	= nvme_ns_chr_uring_cmd,
3348 	.uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
3349 };
3350 
3351 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3352 {
3353 	int ret;
3354 
3355 	ns->cdev_device.parent = ns->ctrl->device;
3356 	ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3357 			   ns->ctrl->instance, ns->head->instance);
3358 	if (ret)
3359 		return ret;
3360 
3361 	return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3362 			     ns->ctrl->ops->module);
3363 }
3364 
3365 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3366 		struct nvme_ns_info *info)
3367 {
3368 	struct nvme_ns_head *head;
3369 	size_t size = sizeof(*head);
3370 	int ret = -ENOMEM;
3371 
3372 #ifdef CONFIG_NVME_MULTIPATH
3373 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3374 #endif
3375 
3376 	head = kzalloc(size, GFP_KERNEL);
3377 	if (!head)
3378 		goto out;
3379 	ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3380 	if (ret < 0)
3381 		goto out_free_head;
3382 	head->instance = ret;
3383 	INIT_LIST_HEAD(&head->list);
3384 	ret = init_srcu_struct(&head->srcu);
3385 	if (ret)
3386 		goto out_ida_remove;
3387 	head->subsys = ctrl->subsys;
3388 	head->ns_id = info->nsid;
3389 	head->ids = info->ids;
3390 	head->shared = info->is_shared;
3391 	kref_init(&head->ref);
3392 
3393 	if (head->ids.csi) {
3394 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3395 		if (ret)
3396 			goto out_cleanup_srcu;
3397 	} else
3398 		head->effects = ctrl->effects;
3399 
3400 	ret = nvme_mpath_alloc_disk(ctrl, head);
3401 	if (ret)
3402 		goto out_cleanup_srcu;
3403 
3404 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3405 
3406 	kref_get(&ctrl->subsys->ref);
3407 
3408 	return head;
3409 out_cleanup_srcu:
3410 	cleanup_srcu_struct(&head->srcu);
3411 out_ida_remove:
3412 	ida_free(&ctrl->subsys->ns_ida, head->instance);
3413 out_free_head:
3414 	kfree(head);
3415 out:
3416 	if (ret > 0)
3417 		ret = blk_status_to_errno(nvme_error_status(ret));
3418 	return ERR_PTR(ret);
3419 }
3420 
3421 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3422 		struct nvme_ns_ids *ids)
3423 {
3424 	struct nvme_subsystem *s;
3425 	int ret = 0;
3426 
3427 	/*
3428 	 * Note that this check is racy as we try to avoid holding the global
3429 	 * lock over the whole ns_head creation.  But it is only intended as
3430 	 * a sanity check anyway.
3431 	 */
3432 	mutex_lock(&nvme_subsystems_lock);
3433 	list_for_each_entry(s, &nvme_subsystems, entry) {
3434 		if (s == this)
3435 			continue;
3436 		mutex_lock(&s->lock);
3437 		ret = nvme_subsys_check_duplicate_ids(s, ids);
3438 		mutex_unlock(&s->lock);
3439 		if (ret)
3440 			break;
3441 	}
3442 	mutex_unlock(&nvme_subsystems_lock);
3443 
3444 	return ret;
3445 }
3446 
3447 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
3448 {
3449 	struct nvme_ctrl *ctrl = ns->ctrl;
3450 	struct nvme_ns_head *head = NULL;
3451 	int ret;
3452 
3453 	ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
3454 	if (ret) {
3455 		/*
3456 		 * We've found two different namespaces on two different
3457 		 * subsystems that report the same ID.  This is pretty nasty
3458 		 * for anything that actually requires unique device
3459 		 * identification.  In the kernel we need this for multipathing,
3460 		 * and in user space the /dev/disk/by-id/ links rely on it.
3461 		 *
3462 		 * If the device also claims to be multi-path capable back off
3463 		 * here now and refuse the probe the second device as this is a
3464 		 * recipe for data corruption.  If not this is probably a
3465 		 * cheap consumer device if on the PCIe bus, so let the user
3466 		 * proceed and use the shiny toy, but warn that with changing
3467 		 * probing order (which due to our async probing could just be
3468 		 * device taking longer to startup) the other device could show
3469 		 * up at any time.
3470 		 */
3471 		nvme_print_device_info(ctrl);
3472 		if ((ns->ctrl->ops->flags & NVME_F_FABRICS) || /* !PCIe */
3473 		    ((ns->ctrl->subsys->cmic & NVME_CTRL_CMIC_MULTI_CTRL) &&
3474 		     info->is_shared)) {
3475 			dev_err(ctrl->device,
3476 				"ignoring nsid %d because of duplicate IDs\n",
3477 				info->nsid);
3478 			return ret;
3479 		}
3480 
3481 		dev_err(ctrl->device,
3482 			"clearing duplicate IDs for nsid %d\n", info->nsid);
3483 		dev_err(ctrl->device,
3484 			"use of /dev/disk/by-id/ may cause data corruption\n");
3485 		memset(&info->ids.nguid, 0, sizeof(info->ids.nguid));
3486 		memset(&info->ids.uuid, 0, sizeof(info->ids.uuid));
3487 		memset(&info->ids.eui64, 0, sizeof(info->ids.eui64));
3488 		ctrl->quirks |= NVME_QUIRK_BOGUS_NID;
3489 	}
3490 
3491 	mutex_lock(&ctrl->subsys->lock);
3492 	head = nvme_find_ns_head(ctrl, info->nsid);
3493 	if (!head) {
3494 		ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
3495 		if (ret) {
3496 			dev_err(ctrl->device,
3497 				"duplicate IDs in subsystem for nsid %d\n",
3498 				info->nsid);
3499 			goto out_unlock;
3500 		}
3501 		head = nvme_alloc_ns_head(ctrl, info);
3502 		if (IS_ERR(head)) {
3503 			ret = PTR_ERR(head);
3504 			goto out_unlock;
3505 		}
3506 	} else {
3507 		ret = -EINVAL;
3508 		if (!info->is_shared || !head->shared) {
3509 			dev_err(ctrl->device,
3510 				"Duplicate unshared namespace %d\n",
3511 				info->nsid);
3512 			goto out_put_ns_head;
3513 		}
3514 		if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
3515 			dev_err(ctrl->device,
3516 				"IDs don't match for shared namespace %d\n",
3517 					info->nsid);
3518 			goto out_put_ns_head;
3519 		}
3520 
3521 		if (!multipath) {
3522 			dev_warn(ctrl->device,
3523 				"Found shared namespace %d, but multipathing not supported.\n",
3524 				info->nsid);
3525 			dev_warn_once(ctrl->device,
3526 				"Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0.\n");
3527 		}
3528 	}
3529 
3530 	list_add_tail_rcu(&ns->siblings, &head->list);
3531 	ns->head = head;
3532 	mutex_unlock(&ctrl->subsys->lock);
3533 	return 0;
3534 
3535 out_put_ns_head:
3536 	nvme_put_ns_head(head);
3537 out_unlock:
3538 	mutex_unlock(&ctrl->subsys->lock);
3539 	return ret;
3540 }
3541 
3542 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3543 {
3544 	struct nvme_ns *ns, *ret = NULL;
3545 	int srcu_idx;
3546 
3547 	srcu_idx = srcu_read_lock(&ctrl->srcu);
3548 	list_for_each_entry_rcu(ns, &ctrl->namespaces, list) {
3549 		if (ns->head->ns_id == nsid) {
3550 			if (!nvme_get_ns(ns))
3551 				continue;
3552 			ret = ns;
3553 			break;
3554 		}
3555 		if (ns->head->ns_id > nsid)
3556 			break;
3557 	}
3558 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
3559 	return ret;
3560 }
3561 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3562 
3563 /*
3564  * Add the namespace to the controller list while keeping the list ordered.
3565  */
3566 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3567 {
3568 	struct nvme_ns *tmp;
3569 
3570 	list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3571 		if (tmp->head->ns_id < ns->head->ns_id) {
3572 			list_add_rcu(&ns->list, &tmp->list);
3573 			return;
3574 		}
3575 	}
3576 	list_add(&ns->list, &ns->ctrl->namespaces);
3577 }
3578 
3579 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
3580 {
3581 	struct nvme_ns *ns;
3582 	struct gendisk *disk;
3583 	int node = ctrl->numa_node;
3584 
3585 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3586 	if (!ns)
3587 		return;
3588 
3589 	disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3590 	if (IS_ERR(disk))
3591 		goto out_free_ns;
3592 	disk->fops = &nvme_bdev_ops;
3593 	disk->private_data = ns;
3594 
3595 	ns->disk = disk;
3596 	ns->queue = disk->queue;
3597 
3598 	if (ctrl->opts && ctrl->opts->data_digest)
3599 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3600 
3601 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3602 	if (ctrl->ops->supports_pci_p2pdma &&
3603 	    ctrl->ops->supports_pci_p2pdma(ctrl))
3604 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3605 
3606 	ns->ctrl = ctrl;
3607 	kref_init(&ns->kref);
3608 
3609 	if (nvme_init_ns_head(ns, info))
3610 		goto out_cleanup_disk;
3611 
3612 	/*
3613 	 * If multipathing is enabled, the device name for all disks and not
3614 	 * just those that represent shared namespaces needs to be based on the
3615 	 * subsystem instance.  Using the controller instance for private
3616 	 * namespaces could lead to naming collisions between shared and private
3617 	 * namespaces if they don't use a common numbering scheme.
3618 	 *
3619 	 * If multipathing is not enabled, disk names must use the controller
3620 	 * instance as shared namespaces will show up as multiple block
3621 	 * devices.
3622 	 */
3623 	if (nvme_ns_head_multipath(ns->head)) {
3624 		sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
3625 			ctrl->instance, ns->head->instance);
3626 		disk->flags |= GENHD_FL_HIDDEN;
3627 	} else if (multipath) {
3628 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
3629 			ns->head->instance);
3630 	} else {
3631 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3632 			ns->head->instance);
3633 	}
3634 
3635 	if (nvme_update_ns_info(ns, info))
3636 		goto out_unlink_ns;
3637 
3638 	mutex_lock(&ctrl->namespaces_lock);
3639 	/*
3640 	 * Ensure that no namespaces are added to the ctrl list after the queues
3641 	 * are frozen, thereby avoiding a deadlock between scan and reset.
3642 	 */
3643 	if (test_bit(NVME_CTRL_FROZEN, &ctrl->flags)) {
3644 		mutex_unlock(&ctrl->namespaces_lock);
3645 		goto out_unlink_ns;
3646 	}
3647 	nvme_ns_add_to_ctrl_list(ns);
3648 	mutex_unlock(&ctrl->namespaces_lock);
3649 	synchronize_srcu(&ctrl->srcu);
3650 	nvme_get_ctrl(ctrl);
3651 
3652 	if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
3653 		goto out_cleanup_ns_from_list;
3654 
3655 	if (!nvme_ns_head_multipath(ns->head))
3656 		nvme_add_ns_cdev(ns);
3657 
3658 	nvme_mpath_add_disk(ns, info->anagrpid);
3659 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3660 
3661 	return;
3662 
3663  out_cleanup_ns_from_list:
3664 	nvme_put_ctrl(ctrl);
3665 	mutex_lock(&ctrl->namespaces_lock);
3666 	list_del_rcu(&ns->list);
3667 	mutex_unlock(&ctrl->namespaces_lock);
3668 	synchronize_srcu(&ctrl->srcu);
3669  out_unlink_ns:
3670 	mutex_lock(&ctrl->subsys->lock);
3671 	list_del_rcu(&ns->siblings);
3672 	if (list_empty(&ns->head->list))
3673 		list_del_init(&ns->head->entry);
3674 	mutex_unlock(&ctrl->subsys->lock);
3675 	nvme_put_ns_head(ns->head);
3676  out_cleanup_disk:
3677 	put_disk(disk);
3678  out_free_ns:
3679 	kfree(ns);
3680 }
3681 
3682 static void nvme_ns_remove(struct nvme_ns *ns)
3683 {
3684 	bool last_path = false;
3685 
3686 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3687 		return;
3688 
3689 	clear_bit(NVME_NS_READY, &ns->flags);
3690 	set_capacity(ns->disk, 0);
3691 	nvme_fault_inject_fini(&ns->fault_inject);
3692 
3693 	/*
3694 	 * Ensure that !NVME_NS_READY is seen by other threads to prevent
3695 	 * this ns going back into current_path.
3696 	 */
3697 	synchronize_srcu(&ns->head->srcu);
3698 
3699 	/* wait for concurrent submissions */
3700 	if (nvme_mpath_clear_current_path(ns))
3701 		synchronize_srcu(&ns->head->srcu);
3702 
3703 	mutex_lock(&ns->ctrl->subsys->lock);
3704 	list_del_rcu(&ns->siblings);
3705 	if (list_empty(&ns->head->list)) {
3706 		list_del_init(&ns->head->entry);
3707 		last_path = true;
3708 	}
3709 	mutex_unlock(&ns->ctrl->subsys->lock);
3710 
3711 	/* guarantee not available in head->list */
3712 	synchronize_srcu(&ns->head->srcu);
3713 
3714 	if (!nvme_ns_head_multipath(ns->head))
3715 		nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3716 	del_gendisk(ns->disk);
3717 
3718 	mutex_lock(&ns->ctrl->namespaces_lock);
3719 	list_del_rcu(&ns->list);
3720 	mutex_unlock(&ns->ctrl->namespaces_lock);
3721 	synchronize_srcu(&ns->ctrl->srcu);
3722 
3723 	if (last_path)
3724 		nvme_mpath_shutdown_disk(ns->head);
3725 	nvme_put_ns(ns);
3726 }
3727 
3728 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3729 {
3730 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3731 
3732 	if (ns) {
3733 		nvme_ns_remove(ns);
3734 		nvme_put_ns(ns);
3735 	}
3736 }
3737 
3738 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
3739 {
3740 	int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3741 
3742 	if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
3743 		dev_err(ns->ctrl->device,
3744 			"identifiers changed for nsid %d\n", ns->head->ns_id);
3745 		goto out;
3746 	}
3747 
3748 	ret = nvme_update_ns_info(ns, info);
3749 out:
3750 	/*
3751 	 * Only remove the namespace if we got a fatal error back from the
3752 	 * device, otherwise ignore the error and just move on.
3753 	 *
3754 	 * TODO: we should probably schedule a delayed retry here.
3755 	 */
3756 	if (ret > 0 && (ret & NVME_SC_DNR))
3757 		nvme_ns_remove(ns);
3758 }
3759 
3760 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3761 {
3762 	struct nvme_ns_info info = { .nsid = nsid };
3763 	struct nvme_ns *ns;
3764 	int ret;
3765 
3766 	if (nvme_identify_ns_descs(ctrl, &info))
3767 		return;
3768 
3769 	if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
3770 		dev_warn(ctrl->device,
3771 			"command set not reported for nsid: %d\n", nsid);
3772 		return;
3773 	}
3774 
3775 	/*
3776 	 * If available try to use the Command Set Idependent Identify Namespace
3777 	 * data structure to find all the generic information that is needed to
3778 	 * set up a namespace.  If not fall back to the legacy version.
3779 	 */
3780 	if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
3781 	    (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS))
3782 		ret = nvme_ns_info_from_id_cs_indep(ctrl, &info);
3783 	else
3784 		ret = nvme_ns_info_from_identify(ctrl, &info);
3785 
3786 	if (info.is_removed)
3787 		nvme_ns_remove_by_nsid(ctrl, nsid);
3788 
3789 	/*
3790 	 * Ignore the namespace if it is not ready. We will get an AEN once it
3791 	 * becomes ready and restart the scan.
3792 	 */
3793 	if (ret || !info.is_ready)
3794 		return;
3795 
3796 	ns = nvme_find_get_ns(ctrl, nsid);
3797 	if (ns) {
3798 		nvme_validate_ns(ns, &info);
3799 		nvme_put_ns(ns);
3800 	} else {
3801 		nvme_alloc_ns(ctrl, &info);
3802 	}
3803 }
3804 
3805 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3806 					unsigned nsid)
3807 {
3808 	struct nvme_ns *ns, *next;
3809 	LIST_HEAD(rm_list);
3810 
3811 	mutex_lock(&ctrl->namespaces_lock);
3812 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3813 		if (ns->head->ns_id > nsid) {
3814 			list_del_rcu(&ns->list);
3815 			synchronize_srcu(&ctrl->srcu);
3816 			list_add_tail_rcu(&ns->list, &rm_list);
3817 		}
3818 	}
3819 	mutex_unlock(&ctrl->namespaces_lock);
3820 
3821 	list_for_each_entry_safe(ns, next, &rm_list, list)
3822 		nvme_ns_remove(ns);
3823 }
3824 
3825 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3826 {
3827 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3828 	__le32 *ns_list;
3829 	u32 prev = 0;
3830 	int ret = 0, i;
3831 
3832 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3833 	if (!ns_list)
3834 		return -ENOMEM;
3835 
3836 	for (;;) {
3837 		struct nvme_command cmd = {
3838 			.identify.opcode	= nvme_admin_identify,
3839 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
3840 			.identify.nsid		= cpu_to_le32(prev),
3841 		};
3842 
3843 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
3844 					    NVME_IDENTIFY_DATA_SIZE);
3845 		if (ret) {
3846 			dev_warn(ctrl->device,
3847 				"Identify NS List failed (status=0x%x)\n", ret);
3848 			goto free;
3849 		}
3850 
3851 		for (i = 0; i < nr_entries; i++) {
3852 			u32 nsid = le32_to_cpu(ns_list[i]);
3853 
3854 			if (!nsid)	/* end of the list? */
3855 				goto out;
3856 			nvme_scan_ns(ctrl, nsid);
3857 			while (++prev < nsid)
3858 				nvme_ns_remove_by_nsid(ctrl, prev);
3859 		}
3860 	}
3861  out:
3862 	nvme_remove_invalid_namespaces(ctrl, prev);
3863  free:
3864 	kfree(ns_list);
3865 	return ret;
3866 }
3867 
3868 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3869 {
3870 	struct nvme_id_ctrl *id;
3871 	u32 nn, i;
3872 
3873 	if (nvme_identify_ctrl(ctrl, &id))
3874 		return;
3875 	nn = le32_to_cpu(id->nn);
3876 	kfree(id);
3877 
3878 	for (i = 1; i <= nn; i++)
3879 		nvme_scan_ns(ctrl, i);
3880 
3881 	nvme_remove_invalid_namespaces(ctrl, nn);
3882 }
3883 
3884 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3885 {
3886 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3887 	__le32 *log;
3888 	int error;
3889 
3890 	log = kzalloc(log_size, GFP_KERNEL);
3891 	if (!log)
3892 		return;
3893 
3894 	/*
3895 	 * We need to read the log to clear the AEN, but we don't want to rely
3896 	 * on it for the changed namespace information as userspace could have
3897 	 * raced with us in reading the log page, which could cause us to miss
3898 	 * updates.
3899 	 */
3900 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
3901 			NVME_CSI_NVM, log, log_size, 0);
3902 	if (error)
3903 		dev_warn(ctrl->device,
3904 			"reading changed ns log failed: %d\n", error);
3905 
3906 	kfree(log);
3907 }
3908 
3909 static void nvme_scan_work(struct work_struct *work)
3910 {
3911 	struct nvme_ctrl *ctrl =
3912 		container_of(work, struct nvme_ctrl, scan_work);
3913 	int ret;
3914 
3915 	/* No tagset on a live ctrl means IO queues could not created */
3916 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE || !ctrl->tagset)
3917 		return;
3918 
3919 	/*
3920 	 * Identify controller limits can change at controller reset due to
3921 	 * new firmware download, even though it is not common we cannot ignore
3922 	 * such scenario. Controller's non-mdts limits are reported in the unit
3923 	 * of logical blocks that is dependent on the format of attached
3924 	 * namespace. Hence re-read the limits at the time of ns allocation.
3925 	 */
3926 	ret = nvme_init_non_mdts_limits(ctrl);
3927 	if (ret < 0) {
3928 		dev_warn(ctrl->device,
3929 			"reading non-mdts-limits failed: %d\n", ret);
3930 		return;
3931 	}
3932 
3933 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3934 		dev_info(ctrl->device, "rescanning namespaces.\n");
3935 		nvme_clear_changed_ns_log(ctrl);
3936 	}
3937 
3938 	mutex_lock(&ctrl->scan_lock);
3939 	if (nvme_ctrl_limited_cns(ctrl)) {
3940 		nvme_scan_ns_sequential(ctrl);
3941 	} else {
3942 		/*
3943 		 * Fall back to sequential scan if DNR is set to handle broken
3944 		 * devices which should support Identify NS List (as per the VS
3945 		 * they report) but don't actually support it.
3946 		 */
3947 		ret = nvme_scan_ns_list(ctrl);
3948 		if (ret > 0 && ret & NVME_SC_DNR)
3949 			nvme_scan_ns_sequential(ctrl);
3950 	}
3951 	mutex_unlock(&ctrl->scan_lock);
3952 }
3953 
3954 /*
3955  * This function iterates the namespace list unlocked to allow recovery from
3956  * controller failure. It is up to the caller to ensure the namespace list is
3957  * not modified by scan work while this function is executing.
3958  */
3959 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3960 {
3961 	struct nvme_ns *ns, *next;
3962 	LIST_HEAD(ns_list);
3963 
3964 	/*
3965 	 * make sure to requeue I/O to all namespaces as these
3966 	 * might result from the scan itself and must complete
3967 	 * for the scan_work to make progress
3968 	 */
3969 	nvme_mpath_clear_ctrl_paths(ctrl);
3970 
3971 	/*
3972 	 * Unquiesce io queues so any pending IO won't hang, especially
3973 	 * those submitted from scan work
3974 	 */
3975 	nvme_unquiesce_io_queues(ctrl);
3976 
3977 	/* prevent racing with ns scanning */
3978 	flush_work(&ctrl->scan_work);
3979 
3980 	/*
3981 	 * The dead states indicates the controller was not gracefully
3982 	 * disconnected. In that case, we won't be able to flush any data while
3983 	 * removing the namespaces' disks; fail all the queues now to avoid
3984 	 * potentially having to clean up the failed sync later.
3985 	 */
3986 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_DEAD)
3987 		nvme_mark_namespaces_dead(ctrl);
3988 
3989 	/* this is a no-op when called from the controller reset handler */
3990 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
3991 
3992 	mutex_lock(&ctrl->namespaces_lock);
3993 	list_splice_init_rcu(&ctrl->namespaces, &ns_list, synchronize_rcu);
3994 	mutex_unlock(&ctrl->namespaces_lock);
3995 	synchronize_srcu(&ctrl->srcu);
3996 
3997 	list_for_each_entry_safe(ns, next, &ns_list, list)
3998 		nvme_ns_remove(ns);
3999 }
4000 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4001 
4002 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env)
4003 {
4004 	const struct nvme_ctrl *ctrl =
4005 		container_of(dev, struct nvme_ctrl, ctrl_device);
4006 	struct nvmf_ctrl_options *opts = ctrl->opts;
4007 	int ret;
4008 
4009 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4010 	if (ret)
4011 		return ret;
4012 
4013 	if (opts) {
4014 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4015 		if (ret)
4016 			return ret;
4017 
4018 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4019 				opts->trsvcid ?: "none");
4020 		if (ret)
4021 			return ret;
4022 
4023 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4024 				opts->host_traddr ?: "none");
4025 		if (ret)
4026 			return ret;
4027 
4028 		ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4029 				opts->host_iface ?: "none");
4030 	}
4031 	return ret;
4032 }
4033 
4034 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4035 {
4036 	char *envp[2] = { envdata, NULL };
4037 
4038 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4039 }
4040 
4041 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4042 {
4043 	char *envp[2] = { NULL, NULL };
4044 	u32 aen_result = ctrl->aen_result;
4045 
4046 	ctrl->aen_result = 0;
4047 	if (!aen_result)
4048 		return;
4049 
4050 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4051 	if (!envp[0])
4052 		return;
4053 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4054 	kfree(envp[0]);
4055 }
4056 
4057 static void nvme_async_event_work(struct work_struct *work)
4058 {
4059 	struct nvme_ctrl *ctrl =
4060 		container_of(work, struct nvme_ctrl, async_event_work);
4061 
4062 	nvme_aen_uevent(ctrl);
4063 
4064 	/*
4065 	 * The transport drivers must guarantee AER submission here is safe by
4066 	 * flushing ctrl async_event_work after changing the controller state
4067 	 * from LIVE and before freeing the admin queue.
4068 	*/
4069 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
4070 		ctrl->ops->submit_async_event(ctrl);
4071 }
4072 
4073 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4074 {
4075 
4076 	u32 csts;
4077 
4078 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4079 		return false;
4080 
4081 	if (csts == ~0)
4082 		return false;
4083 
4084 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4085 }
4086 
4087 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4088 {
4089 	struct nvme_fw_slot_info_log *log;
4090 
4091 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4092 	if (!log)
4093 		return;
4094 
4095 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4096 			log, sizeof(*log), 0))
4097 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4098 	kfree(log);
4099 }
4100 
4101 static void nvme_fw_act_work(struct work_struct *work)
4102 {
4103 	struct nvme_ctrl *ctrl = container_of(work,
4104 				struct nvme_ctrl, fw_act_work);
4105 	unsigned long fw_act_timeout;
4106 
4107 	nvme_auth_stop(ctrl);
4108 
4109 	if (ctrl->mtfa)
4110 		fw_act_timeout = jiffies +
4111 				msecs_to_jiffies(ctrl->mtfa * 100);
4112 	else
4113 		fw_act_timeout = jiffies +
4114 				msecs_to_jiffies(admin_timeout * 1000);
4115 
4116 	nvme_quiesce_io_queues(ctrl);
4117 	while (nvme_ctrl_pp_status(ctrl)) {
4118 		if (time_after(jiffies, fw_act_timeout)) {
4119 			dev_warn(ctrl->device,
4120 				"Fw activation timeout, reset controller\n");
4121 			nvme_try_sched_reset(ctrl);
4122 			return;
4123 		}
4124 		msleep(100);
4125 	}
4126 
4127 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4128 		return;
4129 
4130 	nvme_unquiesce_io_queues(ctrl);
4131 	/* read FW slot information to clear the AER */
4132 	nvme_get_fw_slot_info(ctrl);
4133 
4134 	queue_work(nvme_wq, &ctrl->async_event_work);
4135 }
4136 
4137 static u32 nvme_aer_type(u32 result)
4138 {
4139 	return result & 0x7;
4140 }
4141 
4142 static u32 nvme_aer_subtype(u32 result)
4143 {
4144 	return (result & 0xff00) >> 8;
4145 }
4146 
4147 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4148 {
4149 	u32 aer_notice_type = nvme_aer_subtype(result);
4150 	bool requeue = true;
4151 
4152 	switch (aer_notice_type) {
4153 	case NVME_AER_NOTICE_NS_CHANGED:
4154 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4155 		nvme_queue_scan(ctrl);
4156 		break;
4157 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4158 		/*
4159 		 * We are (ab)using the RESETTING state to prevent subsequent
4160 		 * recovery actions from interfering with the controller's
4161 		 * firmware activation.
4162 		 */
4163 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4164 			requeue = false;
4165 			queue_work(nvme_wq, &ctrl->fw_act_work);
4166 		}
4167 		break;
4168 #ifdef CONFIG_NVME_MULTIPATH
4169 	case NVME_AER_NOTICE_ANA:
4170 		if (!ctrl->ana_log_buf)
4171 			break;
4172 		queue_work(nvme_wq, &ctrl->ana_work);
4173 		break;
4174 #endif
4175 	case NVME_AER_NOTICE_DISC_CHANGED:
4176 		ctrl->aen_result = result;
4177 		break;
4178 	default:
4179 		dev_warn(ctrl->device, "async event result %08x\n", result);
4180 	}
4181 	return requeue;
4182 }
4183 
4184 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4185 {
4186 	dev_warn(ctrl->device, "resetting controller due to AER\n");
4187 	nvme_reset_ctrl(ctrl);
4188 }
4189 
4190 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4191 		volatile union nvme_result *res)
4192 {
4193 	u32 result = le32_to_cpu(res->u32);
4194 	u32 aer_type = nvme_aer_type(result);
4195 	u32 aer_subtype = nvme_aer_subtype(result);
4196 	bool requeue = true;
4197 
4198 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4199 		return;
4200 
4201 	trace_nvme_async_event(ctrl, result);
4202 	switch (aer_type) {
4203 	case NVME_AER_NOTICE:
4204 		requeue = nvme_handle_aen_notice(ctrl, result);
4205 		break;
4206 	case NVME_AER_ERROR:
4207 		/*
4208 		 * For a persistent internal error, don't run async_event_work
4209 		 * to submit a new AER. The controller reset will do it.
4210 		 */
4211 		if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4212 			nvme_handle_aer_persistent_error(ctrl);
4213 			return;
4214 		}
4215 		fallthrough;
4216 	case NVME_AER_SMART:
4217 	case NVME_AER_CSS:
4218 	case NVME_AER_VS:
4219 		ctrl->aen_result = result;
4220 		break;
4221 	default:
4222 		break;
4223 	}
4224 
4225 	if (requeue)
4226 		queue_work(nvme_wq, &ctrl->async_event_work);
4227 }
4228 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4229 
4230 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4231 		const struct blk_mq_ops *ops, unsigned int cmd_size)
4232 {
4233 	int ret;
4234 
4235 	memset(set, 0, sizeof(*set));
4236 	set->ops = ops;
4237 	set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4238 	if (ctrl->ops->flags & NVME_F_FABRICS)
4239 		/* Reserved for fabric connect and keep alive */
4240 		set->reserved_tags = 2;
4241 	set->numa_node = ctrl->numa_node;
4242 	set->flags = BLK_MQ_F_NO_SCHED;
4243 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4244 		set->flags |= BLK_MQ_F_BLOCKING;
4245 	set->cmd_size = cmd_size;
4246 	set->driver_data = ctrl;
4247 	set->nr_hw_queues = 1;
4248 	set->timeout = NVME_ADMIN_TIMEOUT;
4249 	ret = blk_mq_alloc_tag_set(set);
4250 	if (ret)
4251 		return ret;
4252 
4253 	ctrl->admin_q = blk_mq_init_queue(set);
4254 	if (IS_ERR(ctrl->admin_q)) {
4255 		ret = PTR_ERR(ctrl->admin_q);
4256 		goto out_free_tagset;
4257 	}
4258 
4259 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4260 		ctrl->fabrics_q = blk_mq_init_queue(set);
4261 		if (IS_ERR(ctrl->fabrics_q)) {
4262 			ret = PTR_ERR(ctrl->fabrics_q);
4263 			goto out_cleanup_admin_q;
4264 		}
4265 	}
4266 
4267 	ctrl->admin_tagset = set;
4268 	return 0;
4269 
4270 out_cleanup_admin_q:
4271 	blk_mq_destroy_queue(ctrl->admin_q);
4272 	blk_put_queue(ctrl->admin_q);
4273 out_free_tagset:
4274 	blk_mq_free_tag_set(set);
4275 	ctrl->admin_q = NULL;
4276 	ctrl->fabrics_q = NULL;
4277 	return ret;
4278 }
4279 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4280 
4281 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4282 {
4283 	blk_mq_destroy_queue(ctrl->admin_q);
4284 	blk_put_queue(ctrl->admin_q);
4285 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4286 		blk_mq_destroy_queue(ctrl->fabrics_q);
4287 		blk_put_queue(ctrl->fabrics_q);
4288 	}
4289 	blk_mq_free_tag_set(ctrl->admin_tagset);
4290 }
4291 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4292 
4293 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4294 		const struct blk_mq_ops *ops, unsigned int nr_maps,
4295 		unsigned int cmd_size)
4296 {
4297 	int ret;
4298 
4299 	memset(set, 0, sizeof(*set));
4300 	set->ops = ops;
4301 	set->queue_depth = min_t(unsigned, ctrl->sqsize, BLK_MQ_MAX_DEPTH - 1);
4302 	/*
4303 	 * Some Apple controllers requires tags to be unique across admin and
4304 	 * the (only) I/O queue, so reserve the first 32 tags of the I/O queue.
4305 	 */
4306 	if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
4307 		set->reserved_tags = NVME_AQ_DEPTH;
4308 	else if (ctrl->ops->flags & NVME_F_FABRICS)
4309 		/* Reserved for fabric connect */
4310 		set->reserved_tags = 1;
4311 	set->numa_node = ctrl->numa_node;
4312 	set->flags = BLK_MQ_F_SHOULD_MERGE;
4313 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4314 		set->flags |= BLK_MQ_F_BLOCKING;
4315 	set->cmd_size = cmd_size,
4316 	set->driver_data = ctrl;
4317 	set->nr_hw_queues = ctrl->queue_count - 1;
4318 	set->timeout = NVME_IO_TIMEOUT;
4319 	set->nr_maps = nr_maps;
4320 	ret = blk_mq_alloc_tag_set(set);
4321 	if (ret)
4322 		return ret;
4323 
4324 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4325 		ctrl->connect_q = blk_mq_init_queue(set);
4326         	if (IS_ERR(ctrl->connect_q)) {
4327 			ret = PTR_ERR(ctrl->connect_q);
4328 			goto out_free_tag_set;
4329 		}
4330 		blk_queue_flag_set(QUEUE_FLAG_SKIP_TAGSET_QUIESCE,
4331 				   ctrl->connect_q);
4332 	}
4333 
4334 	ctrl->tagset = set;
4335 	return 0;
4336 
4337 out_free_tag_set:
4338 	blk_mq_free_tag_set(set);
4339 	ctrl->connect_q = NULL;
4340 	return ret;
4341 }
4342 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
4343 
4344 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
4345 {
4346 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4347 		blk_mq_destroy_queue(ctrl->connect_q);
4348 		blk_put_queue(ctrl->connect_q);
4349 	}
4350 	blk_mq_free_tag_set(ctrl->tagset);
4351 }
4352 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
4353 
4354 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4355 {
4356 	nvme_mpath_stop(ctrl);
4357 	nvme_auth_stop(ctrl);
4358 	nvme_stop_keep_alive(ctrl);
4359 	nvme_stop_failfast_work(ctrl);
4360 	flush_work(&ctrl->async_event_work);
4361 	cancel_work_sync(&ctrl->fw_act_work);
4362 	if (ctrl->ops->stop_ctrl)
4363 		ctrl->ops->stop_ctrl(ctrl);
4364 }
4365 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4366 
4367 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4368 {
4369 	nvme_start_keep_alive(ctrl);
4370 
4371 	nvme_enable_aen(ctrl);
4372 
4373 	/*
4374 	 * persistent discovery controllers need to send indication to userspace
4375 	 * to re-read the discovery log page to learn about possible changes
4376 	 * that were missed. We identify persistent discovery controllers by
4377 	 * checking that they started once before, hence are reconnecting back.
4378 	 */
4379 	if (test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
4380 	    nvme_discovery_ctrl(ctrl))
4381 		nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
4382 
4383 	if (ctrl->queue_count > 1) {
4384 		nvme_queue_scan(ctrl);
4385 		nvme_unquiesce_io_queues(ctrl);
4386 		nvme_mpath_update(ctrl);
4387 	}
4388 
4389 	nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4390 	set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags);
4391 }
4392 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4393 
4394 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4395 {
4396 	nvme_hwmon_exit(ctrl);
4397 	nvme_fault_inject_fini(&ctrl->fault_inject);
4398 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4399 	cdev_device_del(&ctrl->cdev, ctrl->device);
4400 	nvme_put_ctrl(ctrl);
4401 }
4402 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4403 
4404 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4405 {
4406 	struct nvme_effects_log	*cel;
4407 	unsigned long i;
4408 
4409 	xa_for_each(&ctrl->cels, i, cel) {
4410 		xa_erase(&ctrl->cels, i);
4411 		kfree(cel);
4412 	}
4413 
4414 	xa_destroy(&ctrl->cels);
4415 }
4416 
4417 static void nvme_free_ctrl(struct device *dev)
4418 {
4419 	struct nvme_ctrl *ctrl =
4420 		container_of(dev, struct nvme_ctrl, ctrl_device);
4421 	struct nvme_subsystem *subsys = ctrl->subsys;
4422 
4423 	if (!subsys || ctrl->instance != subsys->instance)
4424 		ida_free(&nvme_instance_ida, ctrl->instance);
4425 
4426 	nvme_free_cels(ctrl);
4427 	nvme_mpath_uninit(ctrl);
4428 	cleanup_srcu_struct(&ctrl->srcu);
4429 	nvme_auth_stop(ctrl);
4430 	nvme_auth_free(ctrl);
4431 	__free_page(ctrl->discard_page);
4432 	free_opal_dev(ctrl->opal_dev);
4433 
4434 	if (subsys) {
4435 		mutex_lock(&nvme_subsystems_lock);
4436 		list_del(&ctrl->subsys_entry);
4437 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4438 		mutex_unlock(&nvme_subsystems_lock);
4439 	}
4440 
4441 	ctrl->ops->free_ctrl(ctrl);
4442 
4443 	if (subsys)
4444 		nvme_put_subsystem(subsys);
4445 }
4446 
4447 /*
4448  * Initialize a NVMe controller structures.  This needs to be called during
4449  * earliest initialization so that we have the initialized structured around
4450  * during probing.
4451  */
4452 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4453 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
4454 {
4455 	int ret;
4456 
4457 	WRITE_ONCE(ctrl->state, NVME_CTRL_NEW);
4458 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4459 	spin_lock_init(&ctrl->lock);
4460 	mutex_init(&ctrl->namespaces_lock);
4461 
4462 	ret = init_srcu_struct(&ctrl->srcu);
4463 	if (ret)
4464 		return ret;
4465 
4466 	mutex_init(&ctrl->scan_lock);
4467 	INIT_LIST_HEAD(&ctrl->namespaces);
4468 	xa_init(&ctrl->cels);
4469 	ctrl->dev = dev;
4470 	ctrl->ops = ops;
4471 	ctrl->quirks = quirks;
4472 	ctrl->numa_node = NUMA_NO_NODE;
4473 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4474 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4475 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4476 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4477 	init_waitqueue_head(&ctrl->state_wq);
4478 
4479 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4480 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4481 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4482 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4483 
4484 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4485 			PAGE_SIZE);
4486 	ctrl->discard_page = alloc_page(GFP_KERNEL);
4487 	if (!ctrl->discard_page) {
4488 		ret = -ENOMEM;
4489 		goto out;
4490 	}
4491 
4492 	ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4493 	if (ret < 0)
4494 		goto out;
4495 	ctrl->instance = ret;
4496 
4497 	device_initialize(&ctrl->ctrl_device);
4498 	ctrl->device = &ctrl->ctrl_device;
4499 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4500 			ctrl->instance);
4501 	ctrl->device->class = nvme_class;
4502 	ctrl->device->parent = ctrl->dev;
4503 	if (ops->dev_attr_groups)
4504 		ctrl->device->groups = ops->dev_attr_groups;
4505 	else
4506 		ctrl->device->groups = nvme_dev_attr_groups;
4507 	ctrl->device->release = nvme_free_ctrl;
4508 	dev_set_drvdata(ctrl->device, ctrl);
4509 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4510 	if (ret)
4511 		goto out_release_instance;
4512 
4513 	nvme_get_ctrl(ctrl);
4514 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
4515 	ctrl->cdev.owner = ops->module;
4516 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4517 	if (ret)
4518 		goto out_free_name;
4519 
4520 	/*
4521 	 * Initialize latency tolerance controls.  The sysfs files won't
4522 	 * be visible to userspace unless the device actually supports APST.
4523 	 */
4524 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4525 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4526 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4527 
4528 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4529 	nvme_mpath_init_ctrl(ctrl);
4530 	ret = nvme_auth_init_ctrl(ctrl);
4531 	if (ret)
4532 		goto out_free_cdev;
4533 
4534 	return 0;
4535 out_free_cdev:
4536 	nvme_fault_inject_fini(&ctrl->fault_inject);
4537 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4538 	cdev_device_del(&ctrl->cdev, ctrl->device);
4539 out_free_name:
4540 	nvme_put_ctrl(ctrl);
4541 	kfree_const(ctrl->device->kobj.name);
4542 out_release_instance:
4543 	ida_free(&nvme_instance_ida, ctrl->instance);
4544 out:
4545 	if (ctrl->discard_page)
4546 		__free_page(ctrl->discard_page);
4547 	cleanup_srcu_struct(&ctrl->srcu);
4548 	return ret;
4549 }
4550 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4551 
4552 /* let I/O to all namespaces fail in preparation for surprise removal */
4553 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
4554 {
4555 	struct nvme_ns *ns;
4556 	int srcu_idx;
4557 
4558 	srcu_idx = srcu_read_lock(&ctrl->srcu);
4559 	list_for_each_entry_rcu(ns, &ctrl->namespaces, list)
4560 		blk_mark_disk_dead(ns->disk);
4561 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
4562 }
4563 EXPORT_SYMBOL_GPL(nvme_mark_namespaces_dead);
4564 
4565 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4566 {
4567 	struct nvme_ns *ns;
4568 	int srcu_idx;
4569 
4570 	srcu_idx = srcu_read_lock(&ctrl->srcu);
4571 	list_for_each_entry_rcu(ns, &ctrl->namespaces, list)
4572 		blk_mq_unfreeze_queue(ns->queue);
4573 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
4574 	clear_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4575 }
4576 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4577 
4578 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4579 {
4580 	struct nvme_ns *ns;
4581 	int srcu_idx;
4582 
4583 	srcu_idx = srcu_read_lock(&ctrl->srcu);
4584 	list_for_each_entry_rcu(ns, &ctrl->namespaces, list) {
4585 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4586 		if (timeout <= 0)
4587 			break;
4588 	}
4589 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
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 	int srcu_idx;
4598 
4599 	srcu_idx = srcu_read_lock(&ctrl->srcu);
4600 	list_for_each_entry_rcu(ns, &ctrl->namespaces, list)
4601 		blk_mq_freeze_queue_wait(ns->queue);
4602 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
4603 }
4604 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4605 
4606 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4607 {
4608 	struct nvme_ns *ns;
4609 	int srcu_idx;
4610 
4611 	set_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4612 	srcu_idx = srcu_read_lock(&ctrl->srcu);
4613 	list_for_each_entry_rcu(ns, &ctrl->namespaces, list)
4614 		blk_freeze_queue_start(ns->queue);
4615 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
4616 }
4617 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4618 
4619 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl)
4620 {
4621 	if (!ctrl->tagset)
4622 		return;
4623 	if (!test_and_set_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4624 		blk_mq_quiesce_tagset(ctrl->tagset);
4625 	else
4626 		blk_mq_wait_quiesce_done(ctrl->tagset);
4627 }
4628 EXPORT_SYMBOL_GPL(nvme_quiesce_io_queues);
4629 
4630 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl)
4631 {
4632 	if (!ctrl->tagset)
4633 		return;
4634 	if (test_and_clear_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4635 		blk_mq_unquiesce_tagset(ctrl->tagset);
4636 }
4637 EXPORT_SYMBOL_GPL(nvme_unquiesce_io_queues);
4638 
4639 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl)
4640 {
4641 	if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4642 		blk_mq_quiesce_queue(ctrl->admin_q);
4643 	else
4644 		blk_mq_wait_quiesce_done(ctrl->admin_q->tag_set);
4645 }
4646 EXPORT_SYMBOL_GPL(nvme_quiesce_admin_queue);
4647 
4648 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl)
4649 {
4650 	if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4651 		blk_mq_unquiesce_queue(ctrl->admin_q);
4652 }
4653 EXPORT_SYMBOL_GPL(nvme_unquiesce_admin_queue);
4654 
4655 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4656 {
4657 	struct nvme_ns *ns;
4658 	int srcu_idx;
4659 
4660 	srcu_idx = srcu_read_lock(&ctrl->srcu);
4661 	list_for_each_entry_rcu(ns, &ctrl->namespaces, list)
4662 		blk_sync_queue(ns->queue);
4663 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
4664 }
4665 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4666 
4667 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4668 {
4669 	nvme_sync_io_queues(ctrl);
4670 	if (ctrl->admin_q)
4671 		blk_sync_queue(ctrl->admin_q);
4672 }
4673 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4674 
4675 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4676 {
4677 	if (file->f_op != &nvme_dev_fops)
4678 		return NULL;
4679 	return file->private_data;
4680 }
4681 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4682 
4683 /*
4684  * Check we didn't inadvertently grow the command structure sizes:
4685  */
4686 static inline void _nvme_check_size(void)
4687 {
4688 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4689 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4690 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4691 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4692 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4693 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4694 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4695 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4696 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4697 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4698 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4699 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4700 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4701 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
4702 			NVME_IDENTIFY_DATA_SIZE);
4703 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4704 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
4705 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4706 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4707 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4708 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4709 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4710 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4711 	BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
4712 }
4713 
4714 
4715 static int __init nvme_core_init(void)
4716 {
4717 	int result = -ENOMEM;
4718 
4719 	_nvme_check_size();
4720 
4721 	nvme_wq = alloc_workqueue("nvme-wq",
4722 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4723 	if (!nvme_wq)
4724 		goto out;
4725 
4726 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4727 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4728 	if (!nvme_reset_wq)
4729 		goto destroy_wq;
4730 
4731 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4732 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4733 	if (!nvme_delete_wq)
4734 		goto destroy_reset_wq;
4735 
4736 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4737 			NVME_MINORS, "nvme");
4738 	if (result < 0)
4739 		goto destroy_delete_wq;
4740 
4741 	nvme_class = class_create("nvme");
4742 	if (IS_ERR(nvme_class)) {
4743 		result = PTR_ERR(nvme_class);
4744 		goto unregister_chrdev;
4745 	}
4746 	nvme_class->dev_uevent = nvme_class_uevent;
4747 
4748 	nvme_subsys_class = class_create("nvme-subsystem");
4749 	if (IS_ERR(nvme_subsys_class)) {
4750 		result = PTR_ERR(nvme_subsys_class);
4751 		goto destroy_class;
4752 	}
4753 
4754 	result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4755 				     "nvme-generic");
4756 	if (result < 0)
4757 		goto destroy_subsys_class;
4758 
4759 	nvme_ns_chr_class = class_create("nvme-generic");
4760 	if (IS_ERR(nvme_ns_chr_class)) {
4761 		result = PTR_ERR(nvme_ns_chr_class);
4762 		goto unregister_generic_ns;
4763 	}
4764 
4765 	result = nvme_init_auth();
4766 	if (result)
4767 		goto destroy_ns_chr;
4768 	return 0;
4769 
4770 destroy_ns_chr:
4771 	class_destroy(nvme_ns_chr_class);
4772 unregister_generic_ns:
4773 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4774 destroy_subsys_class:
4775 	class_destroy(nvme_subsys_class);
4776 destroy_class:
4777 	class_destroy(nvme_class);
4778 unregister_chrdev:
4779 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4780 destroy_delete_wq:
4781 	destroy_workqueue(nvme_delete_wq);
4782 destroy_reset_wq:
4783 	destroy_workqueue(nvme_reset_wq);
4784 destroy_wq:
4785 	destroy_workqueue(nvme_wq);
4786 out:
4787 	return result;
4788 }
4789 
4790 static void __exit nvme_core_exit(void)
4791 {
4792 	nvme_exit_auth();
4793 	class_destroy(nvme_ns_chr_class);
4794 	class_destroy(nvme_subsys_class);
4795 	class_destroy(nvme_class);
4796 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4797 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4798 	destroy_workqueue(nvme_delete_wq);
4799 	destroy_workqueue(nvme_reset_wq);
4800 	destroy_workqueue(nvme_wq);
4801 	ida_destroy(&nvme_ns_chr_minor_ida);
4802 	ida_destroy(&nvme_instance_ida);
4803 }
4804 
4805 MODULE_LICENSE("GPL");
4806 MODULE_VERSION("1.0");
4807 module_init(nvme_core_init);
4808 module_exit(nvme_core_exit);
4809