xref: /openbmc/linux/drivers/nvme/host/core.c (revision da097dcc)
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 static inline 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 	}
942 }
943 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
944 
945 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
946 {
947 	struct nvme_command *cmd = nvme_req(req)->cmd;
948 	blk_status_t ret = BLK_STS_OK;
949 
950 	if (!(req->rq_flags & RQF_DONTPREP))
951 		nvme_clear_nvme_request(req);
952 
953 	switch (req_op(req)) {
954 	case REQ_OP_DRV_IN:
955 	case REQ_OP_DRV_OUT:
956 		/* these are setup prior to execution in nvme_init_request() */
957 		break;
958 	case REQ_OP_FLUSH:
959 		nvme_setup_flush(ns, cmd);
960 		break;
961 	case REQ_OP_ZONE_RESET_ALL:
962 	case REQ_OP_ZONE_RESET:
963 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
964 		break;
965 	case REQ_OP_ZONE_OPEN:
966 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
967 		break;
968 	case REQ_OP_ZONE_CLOSE:
969 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
970 		break;
971 	case REQ_OP_ZONE_FINISH:
972 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
973 		break;
974 	case REQ_OP_WRITE_ZEROES:
975 		ret = nvme_setup_write_zeroes(ns, req, cmd);
976 		break;
977 	case REQ_OP_DISCARD:
978 		ret = nvme_setup_discard(ns, req, cmd);
979 		break;
980 	case REQ_OP_READ:
981 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
982 		break;
983 	case REQ_OP_WRITE:
984 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
985 		break;
986 	case REQ_OP_ZONE_APPEND:
987 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
988 		break;
989 	default:
990 		WARN_ON_ONCE(1);
991 		return BLK_STS_IOERR;
992 	}
993 
994 	cmd->common.command_id = nvme_cid(req);
995 	trace_nvme_setup_cmd(req, cmd);
996 	return ret;
997 }
998 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
999 
1000 /*
1001  * Return values:
1002  * 0:  success
1003  * >0: nvme controller's cqe status response
1004  * <0: kernel error in lieu of controller response
1005  */
1006 int nvme_execute_rq(struct request *rq, bool at_head)
1007 {
1008 	blk_status_t status;
1009 
1010 	status = blk_execute_rq(rq, at_head);
1011 	if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1012 		return -EINTR;
1013 	if (nvme_req(rq)->status)
1014 		return nvme_req(rq)->status;
1015 	return blk_status_to_errno(status);
1016 }
1017 EXPORT_SYMBOL_NS_GPL(nvme_execute_rq, NVME_TARGET_PASSTHRU);
1018 
1019 /*
1020  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1021  * if the result is positive, it's an NVM Express status code
1022  */
1023 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1024 		union nvme_result *result, void *buffer, unsigned bufflen,
1025 		int qid, int at_head, blk_mq_req_flags_t flags)
1026 {
1027 	struct request *req;
1028 	int ret;
1029 
1030 	if (qid == NVME_QID_ANY)
1031 		req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1032 	else
1033 		req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1034 						qid - 1);
1035 
1036 	if (IS_ERR(req))
1037 		return PTR_ERR(req);
1038 	nvme_init_request(req, cmd);
1039 
1040 	if (buffer && bufflen) {
1041 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1042 		if (ret)
1043 			goto out;
1044 	}
1045 
1046 	ret = nvme_execute_rq(req, at_head);
1047 	if (result && ret >= 0)
1048 		*result = nvme_req(req)->result;
1049  out:
1050 	blk_mq_free_request(req);
1051 	return ret;
1052 }
1053 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1054 
1055 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1056 		void *buffer, unsigned bufflen)
1057 {
1058 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1059 			NVME_QID_ANY, 0, 0);
1060 }
1061 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1062 
1063 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1064 {
1065 	u32 effects = 0;
1066 
1067 	if (ns) {
1068 		effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1069 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1070 			dev_warn_once(ctrl->device,
1071 				"IO command:%02x has unusual effects:%08x\n",
1072 				opcode, effects);
1073 
1074 		/*
1075 		 * NVME_CMD_EFFECTS_CSE_MASK causes a freeze all I/O queues,
1076 		 * which would deadlock when done on an I/O command.  Note that
1077 		 * We already warn about an unusual effect above.
1078 		 */
1079 		effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1080 	} else {
1081 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1082 	}
1083 
1084 	return effects;
1085 }
1086 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1087 
1088 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1089 {
1090 	u32 effects = nvme_command_effects(ctrl, ns, opcode);
1091 
1092 	/*
1093 	 * For simplicity, IO to all namespaces is quiesced even if the command
1094 	 * effects say only one namespace is affected.
1095 	 */
1096 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1097 		mutex_lock(&ctrl->scan_lock);
1098 		mutex_lock(&ctrl->subsys->lock);
1099 		nvme_mpath_start_freeze(ctrl->subsys);
1100 		nvme_mpath_wait_freeze(ctrl->subsys);
1101 		nvme_start_freeze(ctrl);
1102 		nvme_wait_freeze(ctrl);
1103 	}
1104 	return effects;
1105 }
1106 EXPORT_SYMBOL_NS_GPL(nvme_passthru_start, NVME_TARGET_PASSTHRU);
1107 
1108 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1109 		       struct nvme_command *cmd, int status)
1110 {
1111 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1112 		nvme_unfreeze(ctrl);
1113 		nvme_mpath_unfreeze(ctrl->subsys);
1114 		mutex_unlock(&ctrl->subsys->lock);
1115 		mutex_unlock(&ctrl->scan_lock);
1116 	}
1117 	if (effects & NVME_CMD_EFFECTS_CCC) {
1118 		if (!test_and_set_bit(NVME_CTRL_DIRTY_CAPABILITY,
1119 				      &ctrl->flags)) {
1120 			dev_info(ctrl->device,
1121 "controller capabilities changed, reset may be required to take effect.\n");
1122 		}
1123 	}
1124 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1125 		nvme_queue_scan(ctrl);
1126 		flush_work(&ctrl->scan_work);
1127 	}
1128 	if (ns)
1129 		return;
1130 
1131 	switch (cmd->common.opcode) {
1132 	case nvme_admin_set_features:
1133 		switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1134 		case NVME_FEAT_KATO:
1135 			/*
1136 			 * Keep alive commands interval on the host should be
1137 			 * updated when KATO is modified by Set Features
1138 			 * commands.
1139 			 */
1140 			if (!status)
1141 				nvme_update_keep_alive(ctrl, cmd);
1142 			break;
1143 		default:
1144 			break;
1145 		}
1146 		break;
1147 	default:
1148 		break;
1149 	}
1150 }
1151 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1152 
1153 /*
1154  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1155  *
1156  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1157  *   accounting for transport roundtrip times [..].
1158  */
1159 static unsigned long nvme_keep_alive_work_period(struct nvme_ctrl *ctrl)
1160 {
1161 	unsigned long delay = ctrl->kato * HZ / 2;
1162 
1163 	/*
1164 	 * When using Traffic Based Keep Alive, we need to run
1165 	 * nvme_keep_alive_work at twice the normal frequency, as one
1166 	 * command completion can postpone sending a keep alive command
1167 	 * by up to twice the delay between runs.
1168 	 */
1169 	if (ctrl->ctratt & NVME_CTRL_ATTR_TBKAS)
1170 		delay /= 2;
1171 	return delay;
1172 }
1173 
1174 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1175 {
1176 	queue_delayed_work(nvme_wq, &ctrl->ka_work,
1177 			   nvme_keep_alive_work_period(ctrl));
1178 }
1179 
1180 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1181 						 blk_status_t status)
1182 {
1183 	struct nvme_ctrl *ctrl = rq->end_io_data;
1184 	unsigned long flags;
1185 	bool startka = false;
1186 	unsigned long rtt = jiffies - (rq->deadline - rq->timeout);
1187 	unsigned long delay = nvme_keep_alive_work_period(ctrl);
1188 
1189 	/*
1190 	 * Subtract off the keepalive RTT so nvme_keep_alive_work runs
1191 	 * at the desired frequency.
1192 	 */
1193 	if (rtt <= delay) {
1194 		delay -= rtt;
1195 	} else {
1196 		dev_warn(ctrl->device, "long keepalive RTT (%u ms)\n",
1197 			 jiffies_to_msecs(rtt));
1198 		delay = 0;
1199 	}
1200 
1201 	blk_mq_free_request(rq);
1202 
1203 	if (status) {
1204 		dev_err(ctrl->device,
1205 			"failed nvme_keep_alive_end_io error=%d\n",
1206 				status);
1207 		return RQ_END_IO_NONE;
1208 	}
1209 
1210 	ctrl->ka_last_check_time = jiffies;
1211 	ctrl->comp_seen = false;
1212 	spin_lock_irqsave(&ctrl->lock, flags);
1213 	if (ctrl->state == NVME_CTRL_LIVE ||
1214 	    ctrl->state == NVME_CTRL_CONNECTING)
1215 		startka = true;
1216 	spin_unlock_irqrestore(&ctrl->lock, flags);
1217 	if (startka)
1218 		queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1219 	return RQ_END_IO_NONE;
1220 }
1221 
1222 static void nvme_keep_alive_work(struct work_struct *work)
1223 {
1224 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1225 			struct nvme_ctrl, ka_work);
1226 	bool comp_seen = ctrl->comp_seen;
1227 	struct request *rq;
1228 
1229 	ctrl->ka_last_check_time = jiffies;
1230 
1231 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1232 		dev_dbg(ctrl->device,
1233 			"reschedule traffic based keep-alive timer\n");
1234 		ctrl->comp_seen = false;
1235 		nvme_queue_keep_alive_work(ctrl);
1236 		return;
1237 	}
1238 
1239 	rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1240 				  BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1241 	if (IS_ERR(rq)) {
1242 		/* allocation failure, reset the controller */
1243 		dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1244 		nvme_reset_ctrl(ctrl);
1245 		return;
1246 	}
1247 	nvme_init_request(rq, &ctrl->ka_cmd);
1248 
1249 	rq->timeout = ctrl->kato * HZ;
1250 	rq->end_io = nvme_keep_alive_end_io;
1251 	rq->end_io_data = ctrl;
1252 	blk_execute_rq_nowait(rq, false);
1253 }
1254 
1255 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1256 {
1257 	if (unlikely(ctrl->kato == 0))
1258 		return;
1259 
1260 	nvme_queue_keep_alive_work(ctrl);
1261 }
1262 
1263 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1264 {
1265 	if (unlikely(ctrl->kato == 0))
1266 		return;
1267 
1268 	cancel_delayed_work_sync(&ctrl->ka_work);
1269 }
1270 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1271 
1272 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1273 				   struct nvme_command *cmd)
1274 {
1275 	unsigned int new_kato =
1276 		DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1277 
1278 	dev_info(ctrl->device,
1279 		 "keep alive interval updated from %u ms to %u ms\n",
1280 		 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1281 
1282 	nvme_stop_keep_alive(ctrl);
1283 	ctrl->kato = new_kato;
1284 	nvme_start_keep_alive(ctrl);
1285 }
1286 
1287 /*
1288  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1289  * flag, thus sending any new CNS opcodes has a big chance of not working.
1290  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1291  * (but not for any later version).
1292  */
1293 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1294 {
1295 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1296 		return ctrl->vs < NVME_VS(1, 2, 0);
1297 	return ctrl->vs < NVME_VS(1, 1, 0);
1298 }
1299 
1300 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1301 {
1302 	struct nvme_command c = { };
1303 	int error;
1304 
1305 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1306 	c.identify.opcode = nvme_admin_identify;
1307 	c.identify.cns = NVME_ID_CNS_CTRL;
1308 
1309 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1310 	if (!*id)
1311 		return -ENOMEM;
1312 
1313 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1314 			sizeof(struct nvme_id_ctrl));
1315 	if (error)
1316 		kfree(*id);
1317 	return error;
1318 }
1319 
1320 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1321 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1322 {
1323 	const char *warn_str = "ctrl returned bogus length:";
1324 	void *data = cur;
1325 
1326 	switch (cur->nidt) {
1327 	case NVME_NIDT_EUI64:
1328 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1329 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1330 				 warn_str, cur->nidl);
1331 			return -1;
1332 		}
1333 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1334 			return NVME_NIDT_EUI64_LEN;
1335 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1336 		return NVME_NIDT_EUI64_LEN;
1337 	case NVME_NIDT_NGUID:
1338 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1339 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1340 				 warn_str, cur->nidl);
1341 			return -1;
1342 		}
1343 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1344 			return NVME_NIDT_NGUID_LEN;
1345 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1346 		return NVME_NIDT_NGUID_LEN;
1347 	case NVME_NIDT_UUID:
1348 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1349 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1350 				 warn_str, cur->nidl);
1351 			return -1;
1352 		}
1353 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1354 			return NVME_NIDT_UUID_LEN;
1355 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1356 		return NVME_NIDT_UUID_LEN;
1357 	case NVME_NIDT_CSI:
1358 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1359 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1360 				 warn_str, cur->nidl);
1361 			return -1;
1362 		}
1363 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1364 		*csi_seen = true;
1365 		return NVME_NIDT_CSI_LEN;
1366 	default:
1367 		/* Skip unknown types */
1368 		return cur->nidl;
1369 	}
1370 }
1371 
1372 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1373 		struct nvme_ns_info *info)
1374 {
1375 	struct nvme_command c = { };
1376 	bool csi_seen = false;
1377 	int status, pos, len;
1378 	void *data;
1379 
1380 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1381 		return 0;
1382 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1383 		return 0;
1384 
1385 	c.identify.opcode = nvme_admin_identify;
1386 	c.identify.nsid = cpu_to_le32(info->nsid);
1387 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1388 
1389 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1390 	if (!data)
1391 		return -ENOMEM;
1392 
1393 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1394 				      NVME_IDENTIFY_DATA_SIZE);
1395 	if (status) {
1396 		dev_warn(ctrl->device,
1397 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1398 			info->nsid, status);
1399 		goto free_data;
1400 	}
1401 
1402 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1403 		struct nvme_ns_id_desc *cur = data + pos;
1404 
1405 		if (cur->nidl == 0)
1406 			break;
1407 
1408 		len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1409 		if (len < 0)
1410 			break;
1411 
1412 		len += sizeof(*cur);
1413 	}
1414 
1415 	if (nvme_multi_css(ctrl) && !csi_seen) {
1416 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1417 			 info->nsid);
1418 		status = -EINVAL;
1419 	}
1420 
1421 free_data:
1422 	kfree(data);
1423 	return status;
1424 }
1425 
1426 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1427 			struct nvme_id_ns **id)
1428 {
1429 	struct nvme_command c = { };
1430 	int error;
1431 
1432 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1433 	c.identify.opcode = nvme_admin_identify;
1434 	c.identify.nsid = cpu_to_le32(nsid);
1435 	c.identify.cns = NVME_ID_CNS_NS;
1436 
1437 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1438 	if (!*id)
1439 		return -ENOMEM;
1440 
1441 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1442 	if (error) {
1443 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1444 		kfree(*id);
1445 	}
1446 	return error;
1447 }
1448 
1449 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1450 		struct nvme_ns_info *info)
1451 {
1452 	struct nvme_ns_ids *ids = &info->ids;
1453 	struct nvme_id_ns *id;
1454 	int ret;
1455 
1456 	ret = nvme_identify_ns(ctrl, info->nsid, &id);
1457 	if (ret)
1458 		return ret;
1459 
1460 	if (id->ncap == 0) {
1461 		/* namespace not allocated or attached */
1462 		info->is_removed = true;
1463 		ret = -ENODEV;
1464 		goto error;
1465 	}
1466 
1467 	info->anagrpid = id->anagrpid;
1468 	info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1469 	info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1470 	info->is_ready = true;
1471 	if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1472 		dev_info(ctrl->device,
1473 			 "Ignoring bogus Namespace Identifiers\n");
1474 	} else {
1475 		if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1476 		    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1477 			memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1478 		if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1479 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1480 			memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1481 	}
1482 
1483 error:
1484 	kfree(id);
1485 	return ret;
1486 }
1487 
1488 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1489 		struct nvme_ns_info *info)
1490 {
1491 	struct nvme_id_ns_cs_indep *id;
1492 	struct nvme_command c = {
1493 		.identify.opcode	= nvme_admin_identify,
1494 		.identify.nsid		= cpu_to_le32(info->nsid),
1495 		.identify.cns		= NVME_ID_CNS_NS_CS_INDEP,
1496 	};
1497 	int ret;
1498 
1499 	id = kmalloc(sizeof(*id), GFP_KERNEL);
1500 	if (!id)
1501 		return -ENOMEM;
1502 
1503 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1504 	if (!ret) {
1505 		info->anagrpid = id->anagrpid;
1506 		info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1507 		info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1508 		info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1509 	}
1510 	kfree(id);
1511 	return ret;
1512 }
1513 
1514 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1515 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1516 {
1517 	union nvme_result res = { 0 };
1518 	struct nvme_command c = { };
1519 	int ret;
1520 
1521 	c.features.opcode = op;
1522 	c.features.fid = cpu_to_le32(fid);
1523 	c.features.dword11 = cpu_to_le32(dword11);
1524 
1525 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1526 			buffer, buflen, NVME_QID_ANY, 0, 0);
1527 	if (ret >= 0 && result)
1528 		*result = le32_to_cpu(res.u32);
1529 	return ret;
1530 }
1531 
1532 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1533 		      unsigned int dword11, void *buffer, size_t buflen,
1534 		      u32 *result)
1535 {
1536 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1537 			     buflen, result);
1538 }
1539 EXPORT_SYMBOL_GPL(nvme_set_features);
1540 
1541 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1542 		      unsigned int dword11, void *buffer, size_t buflen,
1543 		      u32 *result)
1544 {
1545 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1546 			     buflen, result);
1547 }
1548 EXPORT_SYMBOL_GPL(nvme_get_features);
1549 
1550 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1551 {
1552 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1553 	u32 result;
1554 	int status, nr_io_queues;
1555 
1556 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1557 			&result);
1558 	if (status < 0)
1559 		return status;
1560 
1561 	/*
1562 	 * Degraded controllers might return an error when setting the queue
1563 	 * count.  We still want to be able to bring them online and offer
1564 	 * access to the admin queue, as that might be only way to fix them up.
1565 	 */
1566 	if (status > 0) {
1567 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1568 		*count = 0;
1569 	} else {
1570 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1571 		*count = min(*count, nr_io_queues);
1572 	}
1573 
1574 	return 0;
1575 }
1576 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1577 
1578 #define NVME_AEN_SUPPORTED \
1579 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1580 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1581 
1582 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1583 {
1584 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1585 	int status;
1586 
1587 	if (!supported_aens)
1588 		return;
1589 
1590 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1591 			NULL, 0, &result);
1592 	if (status)
1593 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1594 			 supported_aens);
1595 
1596 	queue_work(nvme_wq, &ctrl->async_event_work);
1597 }
1598 
1599 static int nvme_ns_open(struct nvme_ns *ns)
1600 {
1601 
1602 	/* should never be called due to GENHD_FL_HIDDEN */
1603 	if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1604 		goto fail;
1605 	if (!nvme_get_ns(ns))
1606 		goto fail;
1607 	if (!try_module_get(ns->ctrl->ops->module))
1608 		goto fail_put_ns;
1609 
1610 	return 0;
1611 
1612 fail_put_ns:
1613 	nvme_put_ns(ns);
1614 fail:
1615 	return -ENXIO;
1616 }
1617 
1618 static void nvme_ns_release(struct nvme_ns *ns)
1619 {
1620 
1621 	module_put(ns->ctrl->ops->module);
1622 	nvme_put_ns(ns);
1623 }
1624 
1625 static int nvme_open(struct gendisk *disk, blk_mode_t mode)
1626 {
1627 	return nvme_ns_open(disk->private_data);
1628 }
1629 
1630 static void nvme_release(struct gendisk *disk)
1631 {
1632 	nvme_ns_release(disk->private_data);
1633 }
1634 
1635 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1636 {
1637 	/* some standard values */
1638 	geo->heads = 1 << 6;
1639 	geo->sectors = 1 << 5;
1640 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1641 	return 0;
1642 }
1643 
1644 #ifdef CONFIG_BLK_DEV_INTEGRITY
1645 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1646 				u32 max_integrity_segments)
1647 {
1648 	struct blk_integrity integrity = { };
1649 
1650 	switch (ns->pi_type) {
1651 	case NVME_NS_DPS_PI_TYPE3:
1652 		switch (ns->guard_type) {
1653 		case NVME_NVM_NS_16B_GUARD:
1654 			integrity.profile = &t10_pi_type3_crc;
1655 			integrity.tag_size = sizeof(u16) + sizeof(u32);
1656 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1657 			break;
1658 		case NVME_NVM_NS_64B_GUARD:
1659 			integrity.profile = &ext_pi_type3_crc64;
1660 			integrity.tag_size = sizeof(u16) + 6;
1661 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1662 			break;
1663 		default:
1664 			integrity.profile = NULL;
1665 			break;
1666 		}
1667 		break;
1668 	case NVME_NS_DPS_PI_TYPE1:
1669 	case NVME_NS_DPS_PI_TYPE2:
1670 		switch (ns->guard_type) {
1671 		case NVME_NVM_NS_16B_GUARD:
1672 			integrity.profile = &t10_pi_type1_crc;
1673 			integrity.tag_size = sizeof(u16);
1674 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1675 			break;
1676 		case NVME_NVM_NS_64B_GUARD:
1677 			integrity.profile = &ext_pi_type1_crc64;
1678 			integrity.tag_size = sizeof(u16);
1679 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1680 			break;
1681 		default:
1682 			integrity.profile = NULL;
1683 			break;
1684 		}
1685 		break;
1686 	default:
1687 		integrity.profile = NULL;
1688 		break;
1689 	}
1690 
1691 	integrity.tuple_size = ns->ms;
1692 	blk_integrity_register(disk, &integrity);
1693 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1694 }
1695 #else
1696 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1697 				u32 max_integrity_segments)
1698 {
1699 }
1700 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1701 
1702 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1703 {
1704 	struct nvme_ctrl *ctrl = ns->ctrl;
1705 	struct request_queue *queue = disk->queue;
1706 	u32 size = queue_logical_block_size(queue);
1707 
1708 	if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1709 		ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1710 
1711 	if (ctrl->max_discard_sectors == 0) {
1712 		blk_queue_max_discard_sectors(queue, 0);
1713 		return;
1714 	}
1715 
1716 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1717 			NVME_DSM_MAX_RANGES);
1718 
1719 	queue->limits.discard_granularity = size;
1720 
1721 	/* If discard is already enabled, don't reset queue limits */
1722 	if (queue->limits.max_discard_sectors)
1723 		return;
1724 
1725 	blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1726 	blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1727 
1728 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1729 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1730 }
1731 
1732 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1733 {
1734 	return uuid_equal(&a->uuid, &b->uuid) &&
1735 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1736 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1737 		a->csi == b->csi;
1738 }
1739 
1740 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1741 {
1742 	bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1743 	unsigned lbaf = nvme_lbaf_index(id->flbas);
1744 	struct nvme_ctrl *ctrl = ns->ctrl;
1745 	struct nvme_command c = { };
1746 	struct nvme_id_ns_nvm *nvm;
1747 	int ret = 0;
1748 	u32 elbaf;
1749 
1750 	ns->pi_size = 0;
1751 	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1752 	if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1753 		ns->pi_size = sizeof(struct t10_pi_tuple);
1754 		ns->guard_type = NVME_NVM_NS_16B_GUARD;
1755 		goto set_pi;
1756 	}
1757 
1758 	nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1759 	if (!nvm)
1760 		return -ENOMEM;
1761 
1762 	c.identify.opcode = nvme_admin_identify;
1763 	c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1764 	c.identify.cns = NVME_ID_CNS_CS_NS;
1765 	c.identify.csi = NVME_CSI_NVM;
1766 
1767 	ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1768 	if (ret)
1769 		goto free_data;
1770 
1771 	elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1772 
1773 	/* no support for storage tag formats right now */
1774 	if (nvme_elbaf_sts(elbaf))
1775 		goto free_data;
1776 
1777 	ns->guard_type = nvme_elbaf_guard_type(elbaf);
1778 	switch (ns->guard_type) {
1779 	case NVME_NVM_NS_64B_GUARD:
1780 		ns->pi_size = sizeof(struct crc64_pi_tuple);
1781 		break;
1782 	case NVME_NVM_NS_16B_GUARD:
1783 		ns->pi_size = sizeof(struct t10_pi_tuple);
1784 		break;
1785 	default:
1786 		break;
1787 	}
1788 
1789 free_data:
1790 	kfree(nvm);
1791 set_pi:
1792 	if (ns->pi_size && (first || ns->ms == ns->pi_size))
1793 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1794 	else
1795 		ns->pi_type = 0;
1796 
1797 	return ret;
1798 }
1799 
1800 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1801 {
1802 	struct nvme_ctrl *ctrl = ns->ctrl;
1803 	int ret;
1804 
1805 	ret = nvme_init_ms(ns, id);
1806 	if (ret)
1807 		return ret;
1808 
1809 	ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1810 	if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1811 		return 0;
1812 
1813 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1814 		/*
1815 		 * The NVMe over Fabrics specification only supports metadata as
1816 		 * part of the extended data LBA.  We rely on HCA/HBA support to
1817 		 * remap the separate metadata buffer from the block layer.
1818 		 */
1819 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1820 			return 0;
1821 
1822 		ns->features |= NVME_NS_EXT_LBAS;
1823 
1824 		/*
1825 		 * The current fabrics transport drivers support namespace
1826 		 * metadata formats only if nvme_ns_has_pi() returns true.
1827 		 * Suppress support for all other formats so the namespace will
1828 		 * have a 0 capacity and not be usable through the block stack.
1829 		 *
1830 		 * Note, this check will need to be modified if any drivers
1831 		 * gain the ability to use other metadata formats.
1832 		 */
1833 		if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1834 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1835 	} else {
1836 		/*
1837 		 * For PCIe controllers, we can't easily remap the separate
1838 		 * metadata buffer from the block layer and thus require a
1839 		 * separate metadata buffer for block layer metadata/PI support.
1840 		 * We allow extended LBAs for the passthrough interface, though.
1841 		 */
1842 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
1843 			ns->features |= NVME_NS_EXT_LBAS;
1844 		else
1845 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1846 	}
1847 	return 0;
1848 }
1849 
1850 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1851 		struct request_queue *q)
1852 {
1853 	bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1854 
1855 	if (ctrl->max_hw_sectors) {
1856 		u32 max_segments =
1857 			(ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1858 
1859 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
1860 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1861 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1862 	}
1863 	blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1864 	blk_queue_dma_alignment(q, 3);
1865 	blk_queue_write_cache(q, vwc, vwc);
1866 }
1867 
1868 static void nvme_update_disk_info(struct gendisk *disk,
1869 		struct nvme_ns *ns, struct nvme_id_ns *id)
1870 {
1871 	sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1872 	u32 bs = 1U << ns->lba_shift;
1873 	u32 atomic_bs, phys_bs, io_opt = 0;
1874 
1875 	/*
1876 	 * The block layer can't support LBA sizes larger than the page size
1877 	 * or smaller than a sector size yet, so catch this early and don't
1878 	 * allow block I/O.
1879 	 */
1880 	if (ns->lba_shift > PAGE_SHIFT || ns->lba_shift < SECTOR_SHIFT) {
1881 		capacity = 0;
1882 		bs = (1 << 9);
1883 	}
1884 
1885 	blk_integrity_unregister(disk);
1886 
1887 	atomic_bs = phys_bs = bs;
1888 	if (id->nabo == 0) {
1889 		/*
1890 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1891 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1892 		 * 0 then AWUPF must be used instead.
1893 		 */
1894 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1895 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1896 		else
1897 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1898 	}
1899 
1900 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1901 		/* NPWG = Namespace Preferred Write Granularity */
1902 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1903 		/* NOWS = Namespace Optimal Write Size */
1904 		io_opt = bs * (1 + le16_to_cpu(id->nows));
1905 	}
1906 
1907 	blk_queue_logical_block_size(disk->queue, bs);
1908 	/*
1909 	 * Linux filesystems assume writing a single physical block is
1910 	 * an atomic operation. Hence limit the physical block size to the
1911 	 * value of the Atomic Write Unit Power Fail parameter.
1912 	 */
1913 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1914 	blk_queue_io_min(disk->queue, phys_bs);
1915 	blk_queue_io_opt(disk->queue, io_opt);
1916 
1917 	/*
1918 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
1919 	 * metadata masquerading as Type 0 if supported, otherwise reject block
1920 	 * I/O to namespaces with metadata except when the namespace supports
1921 	 * PI, as it can strip/insert in that case.
1922 	 */
1923 	if (ns->ms) {
1924 		if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1925 		    (ns->features & NVME_NS_METADATA_SUPPORTED))
1926 			nvme_init_integrity(disk, ns,
1927 					    ns->ctrl->max_integrity_segments);
1928 		else if (!nvme_ns_has_pi(ns))
1929 			capacity = 0;
1930 	}
1931 
1932 	set_capacity_and_notify(disk, capacity);
1933 
1934 	nvme_config_discard(disk, ns);
1935 	blk_queue_max_write_zeroes_sectors(disk->queue,
1936 					   ns->ctrl->max_zeroes_sectors);
1937 }
1938 
1939 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
1940 {
1941 	return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
1942 }
1943 
1944 static inline bool nvme_first_scan(struct gendisk *disk)
1945 {
1946 	/* nvme_alloc_ns() scans the disk prior to adding it */
1947 	return !disk_live(disk);
1948 }
1949 
1950 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1951 {
1952 	struct nvme_ctrl *ctrl = ns->ctrl;
1953 	u32 iob;
1954 
1955 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1956 	    is_power_of_2(ctrl->max_hw_sectors))
1957 		iob = ctrl->max_hw_sectors;
1958 	else
1959 		iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1960 
1961 	if (!iob)
1962 		return;
1963 
1964 	if (!is_power_of_2(iob)) {
1965 		if (nvme_first_scan(ns->disk))
1966 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1967 				ns->disk->disk_name, iob);
1968 		return;
1969 	}
1970 
1971 	if (blk_queue_is_zoned(ns->disk->queue)) {
1972 		if (nvme_first_scan(ns->disk))
1973 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
1974 				ns->disk->disk_name);
1975 		return;
1976 	}
1977 
1978 	blk_queue_chunk_sectors(ns->queue, iob);
1979 }
1980 
1981 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
1982 		struct nvme_ns_info *info)
1983 {
1984 	blk_mq_freeze_queue(ns->disk->queue);
1985 	nvme_set_queue_limits(ns->ctrl, ns->queue);
1986 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
1987 	blk_mq_unfreeze_queue(ns->disk->queue);
1988 
1989 	if (nvme_ns_head_multipath(ns->head)) {
1990 		blk_mq_freeze_queue(ns->head->disk->queue);
1991 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
1992 		nvme_mpath_revalidate_paths(ns);
1993 		blk_stack_limits(&ns->head->disk->queue->limits,
1994 				 &ns->queue->limits, 0);
1995 		ns->head->disk->flags |= GENHD_FL_HIDDEN;
1996 		blk_mq_unfreeze_queue(ns->head->disk->queue);
1997 	}
1998 
1999 	/* Hide the block-interface for these devices */
2000 	ns->disk->flags |= GENHD_FL_HIDDEN;
2001 	set_bit(NVME_NS_READY, &ns->flags);
2002 
2003 	return 0;
2004 }
2005 
2006 static int nvme_update_ns_info_block(struct nvme_ns *ns,
2007 		struct nvme_ns_info *info)
2008 {
2009 	struct nvme_id_ns *id;
2010 	unsigned lbaf;
2011 	int ret;
2012 
2013 	ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
2014 	if (ret)
2015 		return ret;
2016 
2017 	if (id->ncap == 0) {
2018 		/* namespace not allocated or attached */
2019 		info->is_removed = true;
2020 		ret = -ENODEV;
2021 		goto error;
2022 	}
2023 
2024 	blk_mq_freeze_queue(ns->disk->queue);
2025 	lbaf = nvme_lbaf_index(id->flbas);
2026 	ns->lba_shift = id->lbaf[lbaf].ds;
2027 	nvme_set_queue_limits(ns->ctrl, ns->queue);
2028 
2029 	ret = nvme_configure_metadata(ns, id);
2030 	if (ret < 0) {
2031 		blk_mq_unfreeze_queue(ns->disk->queue);
2032 		goto out;
2033 	}
2034 	nvme_set_chunk_sectors(ns, id);
2035 	nvme_update_disk_info(ns->disk, ns, id);
2036 
2037 	if (ns->head->ids.csi == NVME_CSI_ZNS) {
2038 		ret = nvme_update_zone_info(ns, lbaf);
2039 		if (ret) {
2040 			blk_mq_unfreeze_queue(ns->disk->queue);
2041 			goto out;
2042 		}
2043 	}
2044 
2045 	/*
2046 	 * Only set the DEAC bit if the device guarantees that reads from
2047 	 * deallocated data return zeroes.  While the DEAC bit does not
2048 	 * require that, it must be a no-op if reads from deallocated data
2049 	 * do not return zeroes.
2050 	 */
2051 	if ((id->dlfeat & 0x7) == 0x1 && (id->dlfeat & (1 << 3)))
2052 		ns->features |= NVME_NS_DEAC;
2053 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2054 	set_bit(NVME_NS_READY, &ns->flags);
2055 	blk_mq_unfreeze_queue(ns->disk->queue);
2056 
2057 	if (blk_queue_is_zoned(ns->queue)) {
2058 		ret = nvme_revalidate_zones(ns);
2059 		if (ret && !nvme_first_scan(ns->disk))
2060 			goto out;
2061 	}
2062 
2063 	if (nvme_ns_head_multipath(ns->head)) {
2064 		blk_mq_freeze_queue(ns->head->disk->queue);
2065 		nvme_update_disk_info(ns->head->disk, ns, id);
2066 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2067 		nvme_mpath_revalidate_paths(ns);
2068 		blk_stack_limits(&ns->head->disk->queue->limits,
2069 				 &ns->queue->limits, 0);
2070 		disk_update_readahead(ns->head->disk);
2071 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2072 	}
2073 
2074 	ret = 0;
2075 out:
2076 	/*
2077 	 * If probing fails due an unsupported feature, hide the block device,
2078 	 * but still allow other access.
2079 	 */
2080 	if (ret == -ENODEV) {
2081 		ns->disk->flags |= GENHD_FL_HIDDEN;
2082 		set_bit(NVME_NS_READY, &ns->flags);
2083 		ret = 0;
2084 	}
2085 
2086 error:
2087 	kfree(id);
2088 	return ret;
2089 }
2090 
2091 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2092 {
2093 	switch (info->ids.csi) {
2094 	case NVME_CSI_ZNS:
2095 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2096 			dev_info(ns->ctrl->device,
2097 	"block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2098 				info->nsid);
2099 			return nvme_update_ns_info_generic(ns, info);
2100 		}
2101 		return nvme_update_ns_info_block(ns, info);
2102 	case NVME_CSI_NVM:
2103 		return nvme_update_ns_info_block(ns, info);
2104 	default:
2105 		dev_info(ns->ctrl->device,
2106 			"block device for nsid %u not supported (csi %u)\n",
2107 			info->nsid, info->ids.csi);
2108 		return nvme_update_ns_info_generic(ns, info);
2109 	}
2110 }
2111 
2112 #ifdef CONFIG_BLK_SED_OPAL
2113 static int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2114 		bool send)
2115 {
2116 	struct nvme_ctrl *ctrl = data;
2117 	struct nvme_command cmd = { };
2118 
2119 	if (send)
2120 		cmd.common.opcode = nvme_admin_security_send;
2121 	else
2122 		cmd.common.opcode = nvme_admin_security_recv;
2123 	cmd.common.nsid = 0;
2124 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2125 	cmd.common.cdw11 = cpu_to_le32(len);
2126 
2127 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2128 			NVME_QID_ANY, 1, 0);
2129 }
2130 
2131 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2132 {
2133 	if (ctrl->oacs & NVME_CTRL_OACS_SEC_SUPP) {
2134 		if (!ctrl->opal_dev)
2135 			ctrl->opal_dev = init_opal_dev(ctrl, &nvme_sec_submit);
2136 		else if (was_suspended)
2137 			opal_unlock_from_suspend(ctrl->opal_dev);
2138 	} else {
2139 		free_opal_dev(ctrl->opal_dev);
2140 		ctrl->opal_dev = NULL;
2141 	}
2142 }
2143 #else
2144 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2145 {
2146 }
2147 #endif /* CONFIG_BLK_SED_OPAL */
2148 
2149 #ifdef CONFIG_BLK_DEV_ZONED
2150 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2151 		unsigned int nr_zones, report_zones_cb cb, void *data)
2152 {
2153 	return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2154 			data);
2155 }
2156 #else
2157 #define nvme_report_zones	NULL
2158 #endif /* CONFIG_BLK_DEV_ZONED */
2159 
2160 const struct block_device_operations nvme_bdev_ops = {
2161 	.owner		= THIS_MODULE,
2162 	.ioctl		= nvme_ioctl,
2163 	.compat_ioctl	= blkdev_compat_ptr_ioctl,
2164 	.open		= nvme_open,
2165 	.release	= nvme_release,
2166 	.getgeo		= nvme_getgeo,
2167 	.report_zones	= nvme_report_zones,
2168 	.pr_ops		= &nvme_pr_ops,
2169 };
2170 
2171 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 mask, u32 val,
2172 		u32 timeout, const char *op)
2173 {
2174 	unsigned long timeout_jiffies = jiffies + timeout * HZ;
2175 	u32 csts;
2176 	int ret;
2177 
2178 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2179 		if (csts == ~0)
2180 			return -ENODEV;
2181 		if ((csts & mask) == val)
2182 			break;
2183 
2184 		usleep_range(1000, 2000);
2185 		if (fatal_signal_pending(current))
2186 			return -EINTR;
2187 		if (time_after(jiffies, timeout_jiffies)) {
2188 			dev_err(ctrl->device,
2189 				"Device not ready; aborting %s, CSTS=0x%x\n",
2190 				op, csts);
2191 			return -ENODEV;
2192 		}
2193 	}
2194 
2195 	return ret;
2196 }
2197 
2198 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown)
2199 {
2200 	int ret;
2201 
2202 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2203 	if (shutdown)
2204 		ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2205 	else
2206 		ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2207 
2208 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2209 	if (ret)
2210 		return ret;
2211 
2212 	if (shutdown) {
2213 		return nvme_wait_ready(ctrl, NVME_CSTS_SHST_MASK,
2214 				       NVME_CSTS_SHST_CMPLT,
2215 				       ctrl->shutdown_timeout, "shutdown");
2216 	}
2217 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2218 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2219 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, 0,
2220 			       (NVME_CAP_TIMEOUT(ctrl->cap) + 1) / 2, "reset");
2221 }
2222 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2223 
2224 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2225 {
2226 	unsigned dev_page_min;
2227 	u32 timeout;
2228 	int ret;
2229 
2230 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2231 	if (ret) {
2232 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2233 		return ret;
2234 	}
2235 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2236 
2237 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2238 		dev_err(ctrl->device,
2239 			"Minimum device page size %u too large for host (%u)\n",
2240 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2241 		return -ENODEV;
2242 	}
2243 
2244 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2245 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2246 	else
2247 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2248 
2249 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS && ctrl->cap & NVME_CAP_CRMS_CRIMS)
2250 		ctrl->ctrl_config |= NVME_CC_CRIME;
2251 
2252 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2253 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2254 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2255 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2256 	if (ret)
2257 		return ret;
2258 
2259 	/* Flush write to device (required if transport is PCI) */
2260 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2261 	if (ret)
2262 		return ret;
2263 
2264 	/* CAP value may change after initial CC write */
2265 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2266 	if (ret)
2267 		return ret;
2268 
2269 	timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2270 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2271 		u32 crto, ready_timeout;
2272 
2273 		ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2274 		if (ret) {
2275 			dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2276 				ret);
2277 			return ret;
2278 		}
2279 
2280 		/*
2281 		 * CRTO should always be greater or equal to CAP.TO, but some
2282 		 * devices are known to get this wrong. Use the larger of the
2283 		 * two values.
2284 		 */
2285 		if (ctrl->ctrl_config & NVME_CC_CRIME)
2286 			ready_timeout = NVME_CRTO_CRIMT(crto);
2287 		else
2288 			ready_timeout = NVME_CRTO_CRWMT(crto);
2289 
2290 		if (ready_timeout < timeout)
2291 			dev_warn_once(ctrl->device, "bad crto:%x cap:%llx\n",
2292 				      crto, ctrl->cap);
2293 		else
2294 			timeout = ready_timeout;
2295 	}
2296 
2297 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2298 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2299 	if (ret)
2300 		return ret;
2301 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, NVME_CSTS_RDY,
2302 			       (timeout + 1) / 2, "initialisation");
2303 }
2304 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2305 
2306 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2307 {
2308 	__le64 ts;
2309 	int ret;
2310 
2311 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2312 		return 0;
2313 
2314 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2315 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2316 			NULL);
2317 	if (ret)
2318 		dev_warn_once(ctrl->device,
2319 			"could not set timestamp (%d)\n", ret);
2320 	return ret;
2321 }
2322 
2323 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2324 {
2325 	struct nvme_feat_host_behavior *host;
2326 	u8 acre = 0, lbafee = 0;
2327 	int ret;
2328 
2329 	/* Don't bother enabling the feature if retry delay is not reported */
2330 	if (ctrl->crdt[0])
2331 		acre = NVME_ENABLE_ACRE;
2332 	if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2333 		lbafee = NVME_ENABLE_LBAFEE;
2334 
2335 	if (!acre && !lbafee)
2336 		return 0;
2337 
2338 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2339 	if (!host)
2340 		return 0;
2341 
2342 	host->acre = acre;
2343 	host->lbafee = lbafee;
2344 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2345 				host, sizeof(*host), NULL);
2346 	kfree(host);
2347 	return ret;
2348 }
2349 
2350 /*
2351  * The function checks whether the given total (exlat + enlat) latency of
2352  * a power state allows the latter to be used as an APST transition target.
2353  * It does so by comparing the latency to the primary and secondary latency
2354  * tolerances defined by module params. If there's a match, the corresponding
2355  * timeout value is returned and the matching tolerance index (1 or 2) is
2356  * reported.
2357  */
2358 static bool nvme_apst_get_transition_time(u64 total_latency,
2359 		u64 *transition_time, unsigned *last_index)
2360 {
2361 	if (total_latency <= apst_primary_latency_tol_us) {
2362 		if (*last_index == 1)
2363 			return false;
2364 		*last_index = 1;
2365 		*transition_time = apst_primary_timeout_ms;
2366 		return true;
2367 	}
2368 	if (apst_secondary_timeout_ms &&
2369 		total_latency <= apst_secondary_latency_tol_us) {
2370 		if (*last_index <= 2)
2371 			return false;
2372 		*last_index = 2;
2373 		*transition_time = apst_secondary_timeout_ms;
2374 		return true;
2375 	}
2376 	return false;
2377 }
2378 
2379 /*
2380  * APST (Autonomous Power State Transition) lets us program a table of power
2381  * state transitions that the controller will perform automatically.
2382  *
2383  * Depending on module params, one of the two supported techniques will be used:
2384  *
2385  * - If the parameters provide explicit timeouts and tolerances, they will be
2386  *   used to build a table with up to 2 non-operational states to transition to.
2387  *   The default parameter values were selected based on the values used by
2388  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2389  *   regeneration of the APST table in the event of switching between external
2390  *   and battery power, the timeouts and tolerances reflect a compromise
2391  *   between values used by Microsoft for AC and battery scenarios.
2392  * - If not, we'll configure the table with a simple heuristic: we are willing
2393  *   to spend at most 2% of the time transitioning between power states.
2394  *   Therefore, when running in any given state, we will enter the next
2395  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2396  *   microseconds, as long as that state's exit latency is under the requested
2397  *   maximum latency.
2398  *
2399  * We will not autonomously enter any non-operational state for which the total
2400  * latency exceeds ps_max_latency_us.
2401  *
2402  * Users can set ps_max_latency_us to zero to turn off APST.
2403  */
2404 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2405 {
2406 	struct nvme_feat_auto_pst *table;
2407 	unsigned apste = 0;
2408 	u64 max_lat_us = 0;
2409 	__le64 target = 0;
2410 	int max_ps = -1;
2411 	int state;
2412 	int ret;
2413 	unsigned last_lt_index = UINT_MAX;
2414 
2415 	/*
2416 	 * If APST isn't supported or if we haven't been initialized yet,
2417 	 * then don't do anything.
2418 	 */
2419 	if (!ctrl->apsta)
2420 		return 0;
2421 
2422 	if (ctrl->npss > 31) {
2423 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2424 		return 0;
2425 	}
2426 
2427 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2428 	if (!table)
2429 		return 0;
2430 
2431 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2432 		/* Turn off APST. */
2433 		dev_dbg(ctrl->device, "APST disabled\n");
2434 		goto done;
2435 	}
2436 
2437 	/*
2438 	 * Walk through all states from lowest- to highest-power.
2439 	 * According to the spec, lower-numbered states use more power.  NPSS,
2440 	 * despite the name, is the index of the lowest-power state, not the
2441 	 * number of states.
2442 	 */
2443 	for (state = (int)ctrl->npss; state >= 0; state--) {
2444 		u64 total_latency_us, exit_latency_us, transition_ms;
2445 
2446 		if (target)
2447 			table->entries[state] = target;
2448 
2449 		/*
2450 		 * Don't allow transitions to the deepest state if it's quirked
2451 		 * off.
2452 		 */
2453 		if (state == ctrl->npss &&
2454 		    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2455 			continue;
2456 
2457 		/*
2458 		 * Is this state a useful non-operational state for higher-power
2459 		 * states to autonomously transition to?
2460 		 */
2461 		if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2462 			continue;
2463 
2464 		exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2465 		if (exit_latency_us > ctrl->ps_max_latency_us)
2466 			continue;
2467 
2468 		total_latency_us = exit_latency_us +
2469 			le32_to_cpu(ctrl->psd[state].entry_lat);
2470 
2471 		/*
2472 		 * This state is good. It can be used as the APST idle target
2473 		 * for higher power states.
2474 		 */
2475 		if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2476 			if (!nvme_apst_get_transition_time(total_latency_us,
2477 					&transition_ms, &last_lt_index))
2478 				continue;
2479 		} else {
2480 			transition_ms = total_latency_us + 19;
2481 			do_div(transition_ms, 20);
2482 			if (transition_ms > (1 << 24) - 1)
2483 				transition_ms = (1 << 24) - 1;
2484 		}
2485 
2486 		target = cpu_to_le64((state << 3) | (transition_ms << 8));
2487 		if (max_ps == -1)
2488 			max_ps = state;
2489 		if (total_latency_us > max_lat_us)
2490 			max_lat_us = total_latency_us;
2491 	}
2492 
2493 	if (max_ps == -1)
2494 		dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2495 	else
2496 		dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2497 			max_ps, max_lat_us, (int)sizeof(*table), table);
2498 	apste = 1;
2499 
2500 done:
2501 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2502 				table, sizeof(*table), NULL);
2503 	if (ret)
2504 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2505 	kfree(table);
2506 	return ret;
2507 }
2508 
2509 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2510 {
2511 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2512 	u64 latency;
2513 
2514 	switch (val) {
2515 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2516 	case PM_QOS_LATENCY_ANY:
2517 		latency = U64_MAX;
2518 		break;
2519 
2520 	default:
2521 		latency = val;
2522 	}
2523 
2524 	if (ctrl->ps_max_latency_us != latency) {
2525 		ctrl->ps_max_latency_us = latency;
2526 		if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
2527 			nvme_configure_apst(ctrl);
2528 	}
2529 }
2530 
2531 struct nvme_core_quirk_entry {
2532 	/*
2533 	 * NVMe model and firmware strings are padded with spaces.  For
2534 	 * simplicity, strings in the quirk table are padded with NULLs
2535 	 * instead.
2536 	 */
2537 	u16 vid;
2538 	const char *mn;
2539 	const char *fr;
2540 	unsigned long quirks;
2541 };
2542 
2543 static const struct nvme_core_quirk_entry core_quirks[] = {
2544 	{
2545 		/*
2546 		 * This Toshiba device seems to die using any APST states.  See:
2547 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2548 		 */
2549 		.vid = 0x1179,
2550 		.mn = "THNSF5256GPUK TOSHIBA",
2551 		.quirks = NVME_QUIRK_NO_APST,
2552 	},
2553 	{
2554 		/*
2555 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2556 		 * condition associated with actions related to suspend to idle
2557 		 * LiteON has resolved the problem in future firmware
2558 		 */
2559 		.vid = 0x14a4,
2560 		.fr = "22301111",
2561 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2562 	},
2563 	{
2564 		/*
2565 		 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2566 		 * aborts I/O during any load, but more easily reproducible
2567 		 * with discards (fstrim).
2568 		 *
2569 		 * The device is left in a state where it is also not possible
2570 		 * to use "nvme set-feature" to disable APST, but booting with
2571 		 * nvme_core.default_ps_max_latency=0 works.
2572 		 */
2573 		.vid = 0x1e0f,
2574 		.mn = "KCD6XVUL6T40",
2575 		.quirks = NVME_QUIRK_NO_APST,
2576 	},
2577 	{
2578 		/*
2579 		 * The external Samsung X5 SSD fails initialization without a
2580 		 * delay before checking if it is ready and has a whole set of
2581 		 * other problems.  To make this even more interesting, it
2582 		 * shares the PCI ID with internal Samsung 970 Evo Plus that
2583 		 * does not need or want these quirks.
2584 		 */
2585 		.vid = 0x144d,
2586 		.mn = "Samsung Portable SSD X5",
2587 		.quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2588 			  NVME_QUIRK_NO_DEEPEST_PS |
2589 			  NVME_QUIRK_IGNORE_DEV_SUBNQN,
2590 	}
2591 };
2592 
2593 /* match is null-terminated but idstr is space-padded. */
2594 static bool string_matches(const char *idstr, const char *match, size_t len)
2595 {
2596 	size_t matchlen;
2597 
2598 	if (!match)
2599 		return true;
2600 
2601 	matchlen = strlen(match);
2602 	WARN_ON_ONCE(matchlen > len);
2603 
2604 	if (memcmp(idstr, match, matchlen))
2605 		return false;
2606 
2607 	for (; matchlen < len; matchlen++)
2608 		if (idstr[matchlen] != ' ')
2609 			return false;
2610 
2611 	return true;
2612 }
2613 
2614 static bool quirk_matches(const struct nvme_id_ctrl *id,
2615 			  const struct nvme_core_quirk_entry *q)
2616 {
2617 	return q->vid == le16_to_cpu(id->vid) &&
2618 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2619 		string_matches(id->fr, q->fr, sizeof(id->fr));
2620 }
2621 
2622 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2623 		struct nvme_id_ctrl *id)
2624 {
2625 	size_t nqnlen;
2626 	int off;
2627 
2628 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2629 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2630 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2631 			strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2632 			return;
2633 		}
2634 
2635 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2636 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2637 	}
2638 
2639 	/*
2640 	 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2641 	 * Base Specification 2.0.  It is slightly different from the format
2642 	 * specified there due to historic reasons, and we can't change it now.
2643 	 */
2644 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2645 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2646 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2647 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2648 	off += sizeof(id->sn);
2649 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2650 	off += sizeof(id->mn);
2651 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2652 }
2653 
2654 static void nvme_release_subsystem(struct device *dev)
2655 {
2656 	struct nvme_subsystem *subsys =
2657 		container_of(dev, struct nvme_subsystem, dev);
2658 
2659 	if (subsys->instance >= 0)
2660 		ida_free(&nvme_instance_ida, subsys->instance);
2661 	kfree(subsys);
2662 }
2663 
2664 static void nvme_destroy_subsystem(struct kref *ref)
2665 {
2666 	struct nvme_subsystem *subsys =
2667 			container_of(ref, struct nvme_subsystem, ref);
2668 
2669 	mutex_lock(&nvme_subsystems_lock);
2670 	list_del(&subsys->entry);
2671 	mutex_unlock(&nvme_subsystems_lock);
2672 
2673 	ida_destroy(&subsys->ns_ida);
2674 	device_del(&subsys->dev);
2675 	put_device(&subsys->dev);
2676 }
2677 
2678 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2679 {
2680 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2681 }
2682 
2683 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2684 {
2685 	struct nvme_subsystem *subsys;
2686 
2687 	lockdep_assert_held(&nvme_subsystems_lock);
2688 
2689 	/*
2690 	 * Fail matches for discovery subsystems. This results
2691 	 * in each discovery controller bound to a unique subsystem.
2692 	 * This avoids issues with validating controller values
2693 	 * that can only be true when there is a single unique subsystem.
2694 	 * There may be multiple and completely independent entities
2695 	 * that provide discovery controllers.
2696 	 */
2697 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2698 		return NULL;
2699 
2700 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2701 		if (strcmp(subsys->subnqn, subsysnqn))
2702 			continue;
2703 		if (!kref_get_unless_zero(&subsys->ref))
2704 			continue;
2705 		return subsys;
2706 	}
2707 
2708 	return NULL;
2709 }
2710 
2711 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2712 {
2713 	return ctrl->opts && ctrl->opts->discovery_nqn;
2714 }
2715 
2716 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2717 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2718 {
2719 	struct nvme_ctrl *tmp;
2720 
2721 	lockdep_assert_held(&nvme_subsystems_lock);
2722 
2723 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2724 		if (nvme_state_terminal(tmp))
2725 			continue;
2726 
2727 		if (tmp->cntlid == ctrl->cntlid) {
2728 			dev_err(ctrl->device,
2729 				"Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2730 				ctrl->cntlid, dev_name(tmp->device),
2731 				subsys->subnqn);
2732 			return false;
2733 		}
2734 
2735 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2736 		    nvme_discovery_ctrl(ctrl))
2737 			continue;
2738 
2739 		dev_err(ctrl->device,
2740 			"Subsystem does not support multiple controllers\n");
2741 		return false;
2742 	}
2743 
2744 	return true;
2745 }
2746 
2747 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2748 {
2749 	struct nvme_subsystem *subsys, *found;
2750 	int ret;
2751 
2752 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2753 	if (!subsys)
2754 		return -ENOMEM;
2755 
2756 	subsys->instance = -1;
2757 	mutex_init(&subsys->lock);
2758 	kref_init(&subsys->ref);
2759 	INIT_LIST_HEAD(&subsys->ctrls);
2760 	INIT_LIST_HEAD(&subsys->nsheads);
2761 	nvme_init_subnqn(subsys, ctrl, id);
2762 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2763 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2764 	subsys->vendor_id = le16_to_cpu(id->vid);
2765 	subsys->cmic = id->cmic;
2766 
2767 	/* Versions prior to 1.4 don't necessarily report a valid type */
2768 	if (id->cntrltype == NVME_CTRL_DISC ||
2769 	    !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2770 		subsys->subtype = NVME_NQN_DISC;
2771 	else
2772 		subsys->subtype = NVME_NQN_NVME;
2773 
2774 	if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2775 		dev_err(ctrl->device,
2776 			"Subsystem %s is not a discovery controller",
2777 			subsys->subnqn);
2778 		kfree(subsys);
2779 		return -EINVAL;
2780 	}
2781 	subsys->awupf = le16_to_cpu(id->awupf);
2782 	nvme_mpath_default_iopolicy(subsys);
2783 
2784 	subsys->dev.class = nvme_subsys_class;
2785 	subsys->dev.release = nvme_release_subsystem;
2786 	subsys->dev.groups = nvme_subsys_attrs_groups;
2787 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2788 	device_initialize(&subsys->dev);
2789 
2790 	mutex_lock(&nvme_subsystems_lock);
2791 	found = __nvme_find_get_subsystem(subsys->subnqn);
2792 	if (found) {
2793 		put_device(&subsys->dev);
2794 		subsys = found;
2795 
2796 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2797 			ret = -EINVAL;
2798 			goto out_put_subsystem;
2799 		}
2800 	} else {
2801 		ret = device_add(&subsys->dev);
2802 		if (ret) {
2803 			dev_err(ctrl->device,
2804 				"failed to register subsystem device.\n");
2805 			put_device(&subsys->dev);
2806 			goto out_unlock;
2807 		}
2808 		ida_init(&subsys->ns_ida);
2809 		list_add_tail(&subsys->entry, &nvme_subsystems);
2810 	}
2811 
2812 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2813 				dev_name(ctrl->device));
2814 	if (ret) {
2815 		dev_err(ctrl->device,
2816 			"failed to create sysfs link from subsystem.\n");
2817 		goto out_put_subsystem;
2818 	}
2819 
2820 	if (!found)
2821 		subsys->instance = ctrl->instance;
2822 	ctrl->subsys = subsys;
2823 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2824 	mutex_unlock(&nvme_subsystems_lock);
2825 	return 0;
2826 
2827 out_put_subsystem:
2828 	nvme_put_subsystem(subsys);
2829 out_unlock:
2830 	mutex_unlock(&nvme_subsystems_lock);
2831 	return ret;
2832 }
2833 
2834 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2835 		void *log, size_t size, u64 offset)
2836 {
2837 	struct nvme_command c = { };
2838 	u32 dwlen = nvme_bytes_to_numd(size);
2839 
2840 	c.get_log_page.opcode = nvme_admin_get_log_page;
2841 	c.get_log_page.nsid = cpu_to_le32(nsid);
2842 	c.get_log_page.lid = log_page;
2843 	c.get_log_page.lsp = lsp;
2844 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2845 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2846 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2847 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2848 	c.get_log_page.csi = csi;
2849 
2850 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2851 }
2852 
2853 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2854 				struct nvme_effects_log **log)
2855 {
2856 	struct nvme_effects_log	*cel = xa_load(&ctrl->cels, csi);
2857 	int ret;
2858 
2859 	if (cel)
2860 		goto out;
2861 
2862 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2863 	if (!cel)
2864 		return -ENOMEM;
2865 
2866 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2867 			cel, sizeof(*cel), 0);
2868 	if (ret) {
2869 		kfree(cel);
2870 		return ret;
2871 	}
2872 
2873 	xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2874 out:
2875 	*log = cel;
2876 	return 0;
2877 }
2878 
2879 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2880 {
2881 	u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2882 
2883 	if (check_shl_overflow(1U, units + page_shift - 9, &val))
2884 		return UINT_MAX;
2885 	return val;
2886 }
2887 
2888 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2889 {
2890 	struct nvme_command c = { };
2891 	struct nvme_id_ctrl_nvm *id;
2892 	int ret;
2893 
2894 	if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2895 		ctrl->max_discard_sectors = UINT_MAX;
2896 		ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2897 	} else {
2898 		ctrl->max_discard_sectors = 0;
2899 		ctrl->max_discard_segments = 0;
2900 	}
2901 
2902 	/*
2903 	 * Even though NVMe spec explicitly states that MDTS is not applicable
2904 	 * to the write-zeroes, we are cautious and limit the size to the
2905 	 * controllers max_hw_sectors value, which is based on the MDTS field
2906 	 * and possibly other limiting factors.
2907 	 */
2908 	if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2909 	    !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2910 		ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2911 	else
2912 		ctrl->max_zeroes_sectors = 0;
2913 
2914 	if (ctrl->subsys->subtype != NVME_NQN_NVME ||
2915 	    nvme_ctrl_limited_cns(ctrl) ||
2916 	    test_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags))
2917 		return 0;
2918 
2919 	id = kzalloc(sizeof(*id), GFP_KERNEL);
2920 	if (!id)
2921 		return -ENOMEM;
2922 
2923 	c.identify.opcode = nvme_admin_identify;
2924 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
2925 	c.identify.csi = NVME_CSI_NVM;
2926 
2927 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2928 	if (ret)
2929 		goto free_data;
2930 
2931 	if (id->dmrl)
2932 		ctrl->max_discard_segments = id->dmrl;
2933 	ctrl->dmrsl = le32_to_cpu(id->dmrsl);
2934 	if (id->wzsl)
2935 		ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2936 
2937 free_data:
2938 	if (ret > 0)
2939 		set_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags);
2940 	kfree(id);
2941 	return ret;
2942 }
2943 
2944 static void nvme_init_known_nvm_effects(struct nvme_ctrl *ctrl)
2945 {
2946 	struct nvme_effects_log	*log = ctrl->effects;
2947 
2948 	log->acs[nvme_admin_format_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
2949 						NVME_CMD_EFFECTS_NCC |
2950 						NVME_CMD_EFFECTS_CSE_MASK);
2951 	log->acs[nvme_admin_sanitize_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
2952 						NVME_CMD_EFFECTS_CSE_MASK);
2953 
2954 	/*
2955 	 * The spec says the result of a security receive command depends on
2956 	 * the previous security send command. As such, many vendors log this
2957 	 * command as one to submitted only when no other commands to the same
2958 	 * namespace are outstanding. The intention is to tell the host to
2959 	 * prevent mixing security send and receive.
2960 	 *
2961 	 * This driver can only enforce such exclusive access against IO
2962 	 * queues, though. We are not readily able to enforce such a rule for
2963 	 * two commands to the admin queue, which is the only queue that
2964 	 * matters for this command.
2965 	 *
2966 	 * Rather than blindly freezing the IO queues for this effect that
2967 	 * doesn't even apply to IO, mask it off.
2968 	 */
2969 	log->acs[nvme_admin_security_recv] &= cpu_to_le32(~NVME_CMD_EFFECTS_CSE_MASK);
2970 
2971 	log->iocs[nvme_cmd_write] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
2972 	log->iocs[nvme_cmd_write_zeroes] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
2973 	log->iocs[nvme_cmd_write_uncor] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
2974 }
2975 
2976 static int nvme_init_effects(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2977 {
2978 	int ret = 0;
2979 
2980 	if (ctrl->effects)
2981 		return 0;
2982 
2983 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2984 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2985 		if (ret < 0)
2986 			return ret;
2987 	}
2988 
2989 	if (!ctrl->effects) {
2990 		ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2991 		if (!ctrl->effects)
2992 			return -ENOMEM;
2993 		xa_store(&ctrl->cels, NVME_CSI_NVM, ctrl->effects, GFP_KERNEL);
2994 	}
2995 
2996 	nvme_init_known_nvm_effects(ctrl);
2997 	return 0;
2998 }
2999 
3000 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3001 {
3002 	struct nvme_id_ctrl *id;
3003 	u32 max_hw_sectors;
3004 	bool prev_apst_enabled;
3005 	int ret;
3006 
3007 	ret = nvme_identify_ctrl(ctrl, &id);
3008 	if (ret) {
3009 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3010 		return -EIO;
3011 	}
3012 
3013 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
3014 		ctrl->cntlid = le16_to_cpu(id->cntlid);
3015 
3016 	if (!ctrl->identified) {
3017 		unsigned int i;
3018 
3019 		/*
3020 		 * Check for quirks.  Quirk can depend on firmware version,
3021 		 * so, in principle, the set of quirks present can change
3022 		 * across a reset.  As a possible future enhancement, we
3023 		 * could re-scan for quirks every time we reinitialize
3024 		 * the device, but we'd have to make sure that the driver
3025 		 * behaves intelligently if the quirks change.
3026 		 */
3027 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3028 			if (quirk_matches(id, &core_quirks[i]))
3029 				ctrl->quirks |= core_quirks[i].quirks;
3030 		}
3031 
3032 		ret = nvme_init_subsystem(ctrl, id);
3033 		if (ret)
3034 			goto out_free;
3035 
3036 		ret = nvme_init_effects(ctrl, id);
3037 		if (ret)
3038 			goto out_free;
3039 	}
3040 	memcpy(ctrl->subsys->firmware_rev, id->fr,
3041 	       sizeof(ctrl->subsys->firmware_rev));
3042 
3043 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3044 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3045 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3046 	}
3047 
3048 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3049 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3050 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3051 
3052 	ctrl->oacs = le16_to_cpu(id->oacs);
3053 	ctrl->oncs = le16_to_cpu(id->oncs);
3054 	ctrl->mtfa = le16_to_cpu(id->mtfa);
3055 	ctrl->oaes = le32_to_cpu(id->oaes);
3056 	ctrl->wctemp = le16_to_cpu(id->wctemp);
3057 	ctrl->cctemp = le16_to_cpu(id->cctemp);
3058 
3059 	atomic_set(&ctrl->abort_limit, id->acl + 1);
3060 	ctrl->vwc = id->vwc;
3061 	if (id->mdts)
3062 		max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3063 	else
3064 		max_hw_sectors = UINT_MAX;
3065 	ctrl->max_hw_sectors =
3066 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3067 
3068 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
3069 	ctrl->sgls = le32_to_cpu(id->sgls);
3070 	ctrl->kas = le16_to_cpu(id->kas);
3071 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
3072 	ctrl->ctratt = le32_to_cpu(id->ctratt);
3073 
3074 	ctrl->cntrltype = id->cntrltype;
3075 	ctrl->dctype = id->dctype;
3076 
3077 	if (id->rtd3e) {
3078 		/* us -> s */
3079 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3080 
3081 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3082 						 shutdown_timeout, 60);
3083 
3084 		if (ctrl->shutdown_timeout != shutdown_timeout)
3085 			dev_info(ctrl->device,
3086 				 "Shutdown timeout set to %u seconds\n",
3087 				 ctrl->shutdown_timeout);
3088 	} else
3089 		ctrl->shutdown_timeout = shutdown_timeout;
3090 
3091 	ctrl->npss = id->npss;
3092 	ctrl->apsta = id->apsta;
3093 	prev_apst_enabled = ctrl->apst_enabled;
3094 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3095 		if (force_apst && id->apsta) {
3096 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3097 			ctrl->apst_enabled = true;
3098 		} else {
3099 			ctrl->apst_enabled = false;
3100 		}
3101 	} else {
3102 		ctrl->apst_enabled = id->apsta;
3103 	}
3104 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3105 
3106 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3107 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3108 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3109 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3110 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3111 
3112 		/*
3113 		 * In fabrics we need to verify the cntlid matches the
3114 		 * admin connect
3115 		 */
3116 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3117 			dev_err(ctrl->device,
3118 				"Mismatching cntlid: Connect %u vs Identify "
3119 				"%u, rejecting\n",
3120 				ctrl->cntlid, le16_to_cpu(id->cntlid));
3121 			ret = -EINVAL;
3122 			goto out_free;
3123 		}
3124 
3125 		if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3126 			dev_err(ctrl->device,
3127 				"keep-alive support is mandatory for fabrics\n");
3128 			ret = -EINVAL;
3129 			goto out_free;
3130 		}
3131 	} else {
3132 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3133 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3134 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3135 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3136 	}
3137 
3138 	ret = nvme_mpath_init_identify(ctrl, id);
3139 	if (ret < 0)
3140 		goto out_free;
3141 
3142 	if (ctrl->apst_enabled && !prev_apst_enabled)
3143 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3144 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3145 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3146 
3147 out_free:
3148 	kfree(id);
3149 	return ret;
3150 }
3151 
3152 /*
3153  * Initialize the cached copies of the Identify data and various controller
3154  * register in our nvme_ctrl structure.  This should be called as soon as
3155  * the admin queue is fully up and running.
3156  */
3157 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended)
3158 {
3159 	int ret;
3160 
3161 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3162 	if (ret) {
3163 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3164 		return ret;
3165 	}
3166 
3167 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3168 
3169 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3170 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3171 
3172 	ret = nvme_init_identify(ctrl);
3173 	if (ret)
3174 		return ret;
3175 
3176 	ret = nvme_configure_apst(ctrl);
3177 	if (ret < 0)
3178 		return ret;
3179 
3180 	ret = nvme_configure_timestamp(ctrl);
3181 	if (ret < 0)
3182 		return ret;
3183 
3184 	ret = nvme_configure_host_options(ctrl);
3185 	if (ret < 0)
3186 		return ret;
3187 
3188 	nvme_configure_opal(ctrl, was_suspended);
3189 
3190 	if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3191 		/*
3192 		 * Do not return errors unless we are in a controller reset,
3193 		 * the controller works perfectly fine without hwmon.
3194 		 */
3195 		ret = nvme_hwmon_init(ctrl);
3196 		if (ret == -EINTR)
3197 			return ret;
3198 	}
3199 
3200 	clear_bit(NVME_CTRL_DIRTY_CAPABILITY, &ctrl->flags);
3201 	ctrl->identified = true;
3202 
3203 	return 0;
3204 }
3205 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3206 
3207 static int nvme_dev_open(struct inode *inode, struct file *file)
3208 {
3209 	struct nvme_ctrl *ctrl =
3210 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3211 
3212 	switch (nvme_ctrl_state(ctrl)) {
3213 	case NVME_CTRL_LIVE:
3214 		break;
3215 	default:
3216 		return -EWOULDBLOCK;
3217 	}
3218 
3219 	nvme_get_ctrl(ctrl);
3220 	if (!try_module_get(ctrl->ops->module)) {
3221 		nvme_put_ctrl(ctrl);
3222 		return -EINVAL;
3223 	}
3224 
3225 	file->private_data = ctrl;
3226 	return 0;
3227 }
3228 
3229 static int nvme_dev_release(struct inode *inode, struct file *file)
3230 {
3231 	struct nvme_ctrl *ctrl =
3232 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3233 
3234 	module_put(ctrl->ops->module);
3235 	nvme_put_ctrl(ctrl);
3236 	return 0;
3237 }
3238 
3239 static const struct file_operations nvme_dev_fops = {
3240 	.owner		= THIS_MODULE,
3241 	.open		= nvme_dev_open,
3242 	.release	= nvme_dev_release,
3243 	.unlocked_ioctl	= nvme_dev_ioctl,
3244 	.compat_ioctl	= compat_ptr_ioctl,
3245 	.uring_cmd	= nvme_dev_uring_cmd,
3246 };
3247 
3248 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3249 		unsigned nsid)
3250 {
3251 	struct nvme_ns_head *h;
3252 
3253 	lockdep_assert_held(&ctrl->subsys->lock);
3254 
3255 	list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3256 		/*
3257 		 * Private namespaces can share NSIDs under some conditions.
3258 		 * In that case we can't use the same ns_head for namespaces
3259 		 * with the same NSID.
3260 		 */
3261 		if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3262 			continue;
3263 		if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3264 			return h;
3265 	}
3266 
3267 	return NULL;
3268 }
3269 
3270 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3271 		struct nvme_ns_ids *ids)
3272 {
3273 	bool has_uuid = !uuid_is_null(&ids->uuid);
3274 	bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3275 	bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3276 	struct nvme_ns_head *h;
3277 
3278 	lockdep_assert_held(&subsys->lock);
3279 
3280 	list_for_each_entry(h, &subsys->nsheads, entry) {
3281 		if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3282 			return -EINVAL;
3283 		if (has_nguid &&
3284 		    memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3285 			return -EINVAL;
3286 		if (has_eui64 &&
3287 		    memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3288 			return -EINVAL;
3289 	}
3290 
3291 	return 0;
3292 }
3293 
3294 static void nvme_cdev_rel(struct device *dev)
3295 {
3296 	ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3297 }
3298 
3299 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3300 {
3301 	cdev_device_del(cdev, cdev_device);
3302 	put_device(cdev_device);
3303 }
3304 
3305 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3306 		const struct file_operations *fops, struct module *owner)
3307 {
3308 	int minor, ret;
3309 
3310 	minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3311 	if (minor < 0)
3312 		return minor;
3313 	cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3314 	cdev_device->class = nvme_ns_chr_class;
3315 	cdev_device->release = nvme_cdev_rel;
3316 	device_initialize(cdev_device);
3317 	cdev_init(cdev, fops);
3318 	cdev->owner = owner;
3319 	ret = cdev_device_add(cdev, cdev_device);
3320 	if (ret)
3321 		put_device(cdev_device);
3322 
3323 	return ret;
3324 }
3325 
3326 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3327 {
3328 	return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3329 }
3330 
3331 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3332 {
3333 	nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3334 	return 0;
3335 }
3336 
3337 static const struct file_operations nvme_ns_chr_fops = {
3338 	.owner		= THIS_MODULE,
3339 	.open		= nvme_ns_chr_open,
3340 	.release	= nvme_ns_chr_release,
3341 	.unlocked_ioctl	= nvme_ns_chr_ioctl,
3342 	.compat_ioctl	= compat_ptr_ioctl,
3343 	.uring_cmd	= nvme_ns_chr_uring_cmd,
3344 	.uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
3345 };
3346 
3347 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3348 {
3349 	int ret;
3350 
3351 	ns->cdev_device.parent = ns->ctrl->device;
3352 	ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3353 			   ns->ctrl->instance, ns->head->instance);
3354 	if (ret)
3355 		return ret;
3356 
3357 	return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3358 			     ns->ctrl->ops->module);
3359 }
3360 
3361 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3362 		struct nvme_ns_info *info)
3363 {
3364 	struct nvme_ns_head *head;
3365 	size_t size = sizeof(*head);
3366 	int ret = -ENOMEM;
3367 
3368 #ifdef CONFIG_NVME_MULTIPATH
3369 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3370 #endif
3371 
3372 	head = kzalloc(size, GFP_KERNEL);
3373 	if (!head)
3374 		goto out;
3375 	ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3376 	if (ret < 0)
3377 		goto out_free_head;
3378 	head->instance = ret;
3379 	INIT_LIST_HEAD(&head->list);
3380 	ret = init_srcu_struct(&head->srcu);
3381 	if (ret)
3382 		goto out_ida_remove;
3383 	head->subsys = ctrl->subsys;
3384 	head->ns_id = info->nsid;
3385 	head->ids = info->ids;
3386 	head->shared = info->is_shared;
3387 	kref_init(&head->ref);
3388 
3389 	if (head->ids.csi) {
3390 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3391 		if (ret)
3392 			goto out_cleanup_srcu;
3393 	} else
3394 		head->effects = ctrl->effects;
3395 
3396 	ret = nvme_mpath_alloc_disk(ctrl, head);
3397 	if (ret)
3398 		goto out_cleanup_srcu;
3399 
3400 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3401 
3402 	kref_get(&ctrl->subsys->ref);
3403 
3404 	return head;
3405 out_cleanup_srcu:
3406 	cleanup_srcu_struct(&head->srcu);
3407 out_ida_remove:
3408 	ida_free(&ctrl->subsys->ns_ida, head->instance);
3409 out_free_head:
3410 	kfree(head);
3411 out:
3412 	if (ret > 0)
3413 		ret = blk_status_to_errno(nvme_error_status(ret));
3414 	return ERR_PTR(ret);
3415 }
3416 
3417 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3418 		struct nvme_ns_ids *ids)
3419 {
3420 	struct nvme_subsystem *s;
3421 	int ret = 0;
3422 
3423 	/*
3424 	 * Note that this check is racy as we try to avoid holding the global
3425 	 * lock over the whole ns_head creation.  But it is only intended as
3426 	 * a sanity check anyway.
3427 	 */
3428 	mutex_lock(&nvme_subsystems_lock);
3429 	list_for_each_entry(s, &nvme_subsystems, entry) {
3430 		if (s == this)
3431 			continue;
3432 		mutex_lock(&s->lock);
3433 		ret = nvme_subsys_check_duplicate_ids(s, ids);
3434 		mutex_unlock(&s->lock);
3435 		if (ret)
3436 			break;
3437 	}
3438 	mutex_unlock(&nvme_subsystems_lock);
3439 
3440 	return ret;
3441 }
3442 
3443 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
3444 {
3445 	struct nvme_ctrl *ctrl = ns->ctrl;
3446 	struct nvme_ns_head *head = NULL;
3447 	int ret;
3448 
3449 	ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
3450 	if (ret) {
3451 		/*
3452 		 * We've found two different namespaces on two different
3453 		 * subsystems that report the same ID.  This is pretty nasty
3454 		 * for anything that actually requires unique device
3455 		 * identification.  In the kernel we need this for multipathing,
3456 		 * and in user space the /dev/disk/by-id/ links rely on it.
3457 		 *
3458 		 * If the device also claims to be multi-path capable back off
3459 		 * here now and refuse the probe the second device as this is a
3460 		 * recipe for data corruption.  If not this is probably a
3461 		 * cheap consumer device if on the PCIe bus, so let the user
3462 		 * proceed and use the shiny toy, but warn that with changing
3463 		 * probing order (which due to our async probing could just be
3464 		 * device taking longer to startup) the other device could show
3465 		 * up at any time.
3466 		 */
3467 		nvme_print_device_info(ctrl);
3468 		if ((ns->ctrl->ops->flags & NVME_F_FABRICS) || /* !PCIe */
3469 		    ((ns->ctrl->subsys->cmic & NVME_CTRL_CMIC_MULTI_CTRL) &&
3470 		     info->is_shared)) {
3471 			dev_err(ctrl->device,
3472 				"ignoring nsid %d because of duplicate IDs\n",
3473 				info->nsid);
3474 			return ret;
3475 		}
3476 
3477 		dev_err(ctrl->device,
3478 			"clearing duplicate IDs for nsid %d\n", info->nsid);
3479 		dev_err(ctrl->device,
3480 			"use of /dev/disk/by-id/ may cause data corruption\n");
3481 		memset(&info->ids.nguid, 0, sizeof(info->ids.nguid));
3482 		memset(&info->ids.uuid, 0, sizeof(info->ids.uuid));
3483 		memset(&info->ids.eui64, 0, sizeof(info->ids.eui64));
3484 		ctrl->quirks |= NVME_QUIRK_BOGUS_NID;
3485 	}
3486 
3487 	mutex_lock(&ctrl->subsys->lock);
3488 	head = nvme_find_ns_head(ctrl, info->nsid);
3489 	if (!head) {
3490 		ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
3491 		if (ret) {
3492 			dev_err(ctrl->device,
3493 				"duplicate IDs in subsystem for nsid %d\n",
3494 				info->nsid);
3495 			goto out_unlock;
3496 		}
3497 		head = nvme_alloc_ns_head(ctrl, info);
3498 		if (IS_ERR(head)) {
3499 			ret = PTR_ERR(head);
3500 			goto out_unlock;
3501 		}
3502 	} else {
3503 		ret = -EINVAL;
3504 		if (!info->is_shared || !head->shared) {
3505 			dev_err(ctrl->device,
3506 				"Duplicate unshared namespace %d\n",
3507 				info->nsid);
3508 			goto out_put_ns_head;
3509 		}
3510 		if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
3511 			dev_err(ctrl->device,
3512 				"IDs don't match for shared namespace %d\n",
3513 					info->nsid);
3514 			goto out_put_ns_head;
3515 		}
3516 
3517 		if (!multipath) {
3518 			dev_warn(ctrl->device,
3519 				"Found shared namespace %d, but multipathing not supported.\n",
3520 				info->nsid);
3521 			dev_warn_once(ctrl->device,
3522 				"Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0.\n");
3523 		}
3524 	}
3525 
3526 	list_add_tail_rcu(&ns->siblings, &head->list);
3527 	ns->head = head;
3528 	mutex_unlock(&ctrl->subsys->lock);
3529 	return 0;
3530 
3531 out_put_ns_head:
3532 	nvme_put_ns_head(head);
3533 out_unlock:
3534 	mutex_unlock(&ctrl->subsys->lock);
3535 	return ret;
3536 }
3537 
3538 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3539 {
3540 	struct nvme_ns *ns, *ret = NULL;
3541 
3542 	down_read(&ctrl->namespaces_rwsem);
3543 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3544 		if (ns->head->ns_id == nsid) {
3545 			if (!nvme_get_ns(ns))
3546 				continue;
3547 			ret = ns;
3548 			break;
3549 		}
3550 		if (ns->head->ns_id > nsid)
3551 			break;
3552 	}
3553 	up_read(&ctrl->namespaces_rwsem);
3554 	return ret;
3555 }
3556 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3557 
3558 /*
3559  * Add the namespace to the controller list while keeping the list ordered.
3560  */
3561 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3562 {
3563 	struct nvme_ns *tmp;
3564 
3565 	list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3566 		if (tmp->head->ns_id < ns->head->ns_id) {
3567 			list_add(&ns->list, &tmp->list);
3568 			return;
3569 		}
3570 	}
3571 	list_add(&ns->list, &ns->ctrl->namespaces);
3572 }
3573 
3574 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
3575 {
3576 	struct nvme_ns *ns;
3577 	struct gendisk *disk;
3578 	int node = ctrl->numa_node;
3579 
3580 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3581 	if (!ns)
3582 		return;
3583 
3584 	disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3585 	if (IS_ERR(disk))
3586 		goto out_free_ns;
3587 	disk->fops = &nvme_bdev_ops;
3588 	disk->private_data = ns;
3589 
3590 	ns->disk = disk;
3591 	ns->queue = disk->queue;
3592 
3593 	if (ctrl->opts && ctrl->opts->data_digest)
3594 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3595 
3596 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3597 	if (ctrl->ops->supports_pci_p2pdma &&
3598 	    ctrl->ops->supports_pci_p2pdma(ctrl))
3599 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3600 
3601 	ns->ctrl = ctrl;
3602 	kref_init(&ns->kref);
3603 
3604 	if (nvme_init_ns_head(ns, info))
3605 		goto out_cleanup_disk;
3606 
3607 	/*
3608 	 * If multipathing is enabled, the device name for all disks and not
3609 	 * just those that represent shared namespaces needs to be based on the
3610 	 * subsystem instance.  Using the controller instance for private
3611 	 * namespaces could lead to naming collisions between shared and private
3612 	 * namespaces if they don't use a common numbering scheme.
3613 	 *
3614 	 * If multipathing is not enabled, disk names must use the controller
3615 	 * instance as shared namespaces will show up as multiple block
3616 	 * devices.
3617 	 */
3618 	if (nvme_ns_head_multipath(ns->head)) {
3619 		sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
3620 			ctrl->instance, ns->head->instance);
3621 		disk->flags |= GENHD_FL_HIDDEN;
3622 	} else if (multipath) {
3623 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
3624 			ns->head->instance);
3625 	} else {
3626 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3627 			ns->head->instance);
3628 	}
3629 
3630 	if (nvme_update_ns_info(ns, info))
3631 		goto out_unlink_ns;
3632 
3633 	down_write(&ctrl->namespaces_rwsem);
3634 	/*
3635 	 * Ensure that no namespaces are added to the ctrl list after the queues
3636 	 * are frozen, thereby avoiding a deadlock between scan and reset.
3637 	 */
3638 	if (test_bit(NVME_CTRL_FROZEN, &ctrl->flags)) {
3639 		up_write(&ctrl->namespaces_rwsem);
3640 		goto out_unlink_ns;
3641 	}
3642 	nvme_ns_add_to_ctrl_list(ns);
3643 	up_write(&ctrl->namespaces_rwsem);
3644 	nvme_get_ctrl(ctrl);
3645 
3646 	if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
3647 		goto out_cleanup_ns_from_list;
3648 
3649 	if (!nvme_ns_head_multipath(ns->head))
3650 		nvme_add_ns_cdev(ns);
3651 
3652 	nvme_mpath_add_disk(ns, info->anagrpid);
3653 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3654 
3655 	return;
3656 
3657  out_cleanup_ns_from_list:
3658 	nvme_put_ctrl(ctrl);
3659 	down_write(&ctrl->namespaces_rwsem);
3660 	list_del_init(&ns->list);
3661 	up_write(&ctrl->namespaces_rwsem);
3662  out_unlink_ns:
3663 	mutex_lock(&ctrl->subsys->lock);
3664 	list_del_rcu(&ns->siblings);
3665 	if (list_empty(&ns->head->list))
3666 		list_del_init(&ns->head->entry);
3667 	mutex_unlock(&ctrl->subsys->lock);
3668 	nvme_put_ns_head(ns->head);
3669  out_cleanup_disk:
3670 	put_disk(disk);
3671  out_free_ns:
3672 	kfree(ns);
3673 }
3674 
3675 static void nvme_ns_remove(struct nvme_ns *ns)
3676 {
3677 	bool last_path = false;
3678 
3679 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3680 		return;
3681 
3682 	clear_bit(NVME_NS_READY, &ns->flags);
3683 	set_capacity(ns->disk, 0);
3684 	nvme_fault_inject_fini(&ns->fault_inject);
3685 
3686 	/*
3687 	 * Ensure that !NVME_NS_READY is seen by other threads to prevent
3688 	 * this ns going back into current_path.
3689 	 */
3690 	synchronize_srcu(&ns->head->srcu);
3691 
3692 	/* wait for concurrent submissions */
3693 	if (nvme_mpath_clear_current_path(ns))
3694 		synchronize_srcu(&ns->head->srcu);
3695 
3696 	mutex_lock(&ns->ctrl->subsys->lock);
3697 	list_del_rcu(&ns->siblings);
3698 	if (list_empty(&ns->head->list)) {
3699 		list_del_init(&ns->head->entry);
3700 		last_path = true;
3701 	}
3702 	mutex_unlock(&ns->ctrl->subsys->lock);
3703 
3704 	/* guarantee not available in head->list */
3705 	synchronize_srcu(&ns->head->srcu);
3706 
3707 	if (!nvme_ns_head_multipath(ns->head))
3708 		nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3709 	del_gendisk(ns->disk);
3710 
3711 	down_write(&ns->ctrl->namespaces_rwsem);
3712 	list_del_init(&ns->list);
3713 	up_write(&ns->ctrl->namespaces_rwsem);
3714 
3715 	if (last_path)
3716 		nvme_mpath_shutdown_disk(ns->head);
3717 	nvme_put_ns(ns);
3718 }
3719 
3720 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3721 {
3722 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3723 
3724 	if (ns) {
3725 		nvme_ns_remove(ns);
3726 		nvme_put_ns(ns);
3727 	}
3728 }
3729 
3730 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
3731 {
3732 	int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3733 
3734 	if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
3735 		dev_err(ns->ctrl->device,
3736 			"identifiers changed for nsid %d\n", ns->head->ns_id);
3737 		goto out;
3738 	}
3739 
3740 	ret = nvme_update_ns_info(ns, info);
3741 out:
3742 	/*
3743 	 * Only remove the namespace if we got a fatal error back from the
3744 	 * device, otherwise ignore the error and just move on.
3745 	 *
3746 	 * TODO: we should probably schedule a delayed retry here.
3747 	 */
3748 	if (ret > 0 && (ret & NVME_SC_DNR))
3749 		nvme_ns_remove(ns);
3750 }
3751 
3752 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3753 {
3754 	struct nvme_ns_info info = { .nsid = nsid };
3755 	struct nvme_ns *ns;
3756 	int ret;
3757 
3758 	if (nvme_identify_ns_descs(ctrl, &info))
3759 		return;
3760 
3761 	if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
3762 		dev_warn(ctrl->device,
3763 			"command set not reported for nsid: %d\n", nsid);
3764 		return;
3765 	}
3766 
3767 	/*
3768 	 * If available try to use the Command Set Idependent Identify Namespace
3769 	 * data structure to find all the generic information that is needed to
3770 	 * set up a namespace.  If not fall back to the legacy version.
3771 	 */
3772 	if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
3773 	    (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS))
3774 		ret = nvme_ns_info_from_id_cs_indep(ctrl, &info);
3775 	else
3776 		ret = nvme_ns_info_from_identify(ctrl, &info);
3777 
3778 	if (info.is_removed)
3779 		nvme_ns_remove_by_nsid(ctrl, nsid);
3780 
3781 	/*
3782 	 * Ignore the namespace if it is not ready. We will get an AEN once it
3783 	 * becomes ready and restart the scan.
3784 	 */
3785 	if (ret || !info.is_ready)
3786 		return;
3787 
3788 	ns = nvme_find_get_ns(ctrl, nsid);
3789 	if (ns) {
3790 		nvme_validate_ns(ns, &info);
3791 		nvme_put_ns(ns);
3792 	} else {
3793 		nvme_alloc_ns(ctrl, &info);
3794 	}
3795 }
3796 
3797 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3798 					unsigned nsid)
3799 {
3800 	struct nvme_ns *ns, *next;
3801 	LIST_HEAD(rm_list);
3802 
3803 	down_write(&ctrl->namespaces_rwsem);
3804 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3805 		if (ns->head->ns_id > nsid)
3806 			list_move_tail(&ns->list, &rm_list);
3807 	}
3808 	up_write(&ctrl->namespaces_rwsem);
3809 
3810 	list_for_each_entry_safe(ns, next, &rm_list, list)
3811 		nvme_ns_remove(ns);
3812 
3813 }
3814 
3815 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3816 {
3817 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3818 	__le32 *ns_list;
3819 	u32 prev = 0;
3820 	int ret = 0, i;
3821 
3822 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3823 	if (!ns_list)
3824 		return -ENOMEM;
3825 
3826 	for (;;) {
3827 		struct nvme_command cmd = {
3828 			.identify.opcode	= nvme_admin_identify,
3829 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
3830 			.identify.nsid		= cpu_to_le32(prev),
3831 		};
3832 
3833 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
3834 					    NVME_IDENTIFY_DATA_SIZE);
3835 		if (ret) {
3836 			dev_warn(ctrl->device,
3837 				"Identify NS List failed (status=0x%x)\n", ret);
3838 			goto free;
3839 		}
3840 
3841 		for (i = 0; i < nr_entries; i++) {
3842 			u32 nsid = le32_to_cpu(ns_list[i]);
3843 
3844 			if (!nsid)	/* end of the list? */
3845 				goto out;
3846 			nvme_scan_ns(ctrl, nsid);
3847 			while (++prev < nsid)
3848 				nvme_ns_remove_by_nsid(ctrl, prev);
3849 		}
3850 	}
3851  out:
3852 	nvme_remove_invalid_namespaces(ctrl, prev);
3853  free:
3854 	kfree(ns_list);
3855 	return ret;
3856 }
3857 
3858 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3859 {
3860 	struct nvme_id_ctrl *id;
3861 	u32 nn, i;
3862 
3863 	if (nvme_identify_ctrl(ctrl, &id))
3864 		return;
3865 	nn = le32_to_cpu(id->nn);
3866 	kfree(id);
3867 
3868 	for (i = 1; i <= nn; i++)
3869 		nvme_scan_ns(ctrl, i);
3870 
3871 	nvme_remove_invalid_namespaces(ctrl, nn);
3872 }
3873 
3874 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3875 {
3876 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3877 	__le32 *log;
3878 	int error;
3879 
3880 	log = kzalloc(log_size, GFP_KERNEL);
3881 	if (!log)
3882 		return;
3883 
3884 	/*
3885 	 * We need to read the log to clear the AEN, but we don't want to rely
3886 	 * on it for the changed namespace information as userspace could have
3887 	 * raced with us in reading the log page, which could cause us to miss
3888 	 * updates.
3889 	 */
3890 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
3891 			NVME_CSI_NVM, log, log_size, 0);
3892 	if (error)
3893 		dev_warn(ctrl->device,
3894 			"reading changed ns log failed: %d\n", error);
3895 
3896 	kfree(log);
3897 }
3898 
3899 static void nvme_scan_work(struct work_struct *work)
3900 {
3901 	struct nvme_ctrl *ctrl =
3902 		container_of(work, struct nvme_ctrl, scan_work);
3903 	int ret;
3904 
3905 	/* No tagset on a live ctrl means IO queues could not created */
3906 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE || !ctrl->tagset)
3907 		return;
3908 
3909 	/*
3910 	 * Identify controller limits can change at controller reset due to
3911 	 * new firmware download, even though it is not common we cannot ignore
3912 	 * such scenario. Controller's non-mdts limits are reported in the unit
3913 	 * of logical blocks that is dependent on the format of attached
3914 	 * namespace. Hence re-read the limits at the time of ns allocation.
3915 	 */
3916 	ret = nvme_init_non_mdts_limits(ctrl);
3917 	if (ret < 0) {
3918 		dev_warn(ctrl->device,
3919 			"reading non-mdts-limits failed: %d\n", ret);
3920 		return;
3921 	}
3922 
3923 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3924 		dev_info(ctrl->device, "rescanning namespaces.\n");
3925 		nvme_clear_changed_ns_log(ctrl);
3926 	}
3927 
3928 	mutex_lock(&ctrl->scan_lock);
3929 	if (nvme_ctrl_limited_cns(ctrl)) {
3930 		nvme_scan_ns_sequential(ctrl);
3931 	} else {
3932 		/*
3933 		 * Fall back to sequential scan if DNR is set to handle broken
3934 		 * devices which should support Identify NS List (as per the VS
3935 		 * they report) but don't actually support it.
3936 		 */
3937 		ret = nvme_scan_ns_list(ctrl);
3938 		if (ret > 0 && ret & NVME_SC_DNR)
3939 			nvme_scan_ns_sequential(ctrl);
3940 	}
3941 	mutex_unlock(&ctrl->scan_lock);
3942 }
3943 
3944 /*
3945  * This function iterates the namespace list unlocked to allow recovery from
3946  * controller failure. It is up to the caller to ensure the namespace list is
3947  * not modified by scan work while this function is executing.
3948  */
3949 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3950 {
3951 	struct nvme_ns *ns, *next;
3952 	LIST_HEAD(ns_list);
3953 
3954 	/*
3955 	 * make sure to requeue I/O to all namespaces as these
3956 	 * might result from the scan itself and must complete
3957 	 * for the scan_work to make progress
3958 	 */
3959 	nvme_mpath_clear_ctrl_paths(ctrl);
3960 
3961 	/*
3962 	 * Unquiesce io queues so any pending IO won't hang, especially
3963 	 * those submitted from scan work
3964 	 */
3965 	nvme_unquiesce_io_queues(ctrl);
3966 
3967 	/* prevent racing with ns scanning */
3968 	flush_work(&ctrl->scan_work);
3969 
3970 	/*
3971 	 * The dead states indicates the controller was not gracefully
3972 	 * disconnected. In that case, we won't be able to flush any data while
3973 	 * removing the namespaces' disks; fail all the queues now to avoid
3974 	 * potentially having to clean up the failed sync later.
3975 	 */
3976 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_DEAD)
3977 		nvme_mark_namespaces_dead(ctrl);
3978 
3979 	/* this is a no-op when called from the controller reset handler */
3980 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
3981 
3982 	down_write(&ctrl->namespaces_rwsem);
3983 	list_splice_init(&ctrl->namespaces, &ns_list);
3984 	up_write(&ctrl->namespaces_rwsem);
3985 
3986 	list_for_each_entry_safe(ns, next, &ns_list, list)
3987 		nvme_ns_remove(ns);
3988 }
3989 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3990 
3991 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env)
3992 {
3993 	const struct nvme_ctrl *ctrl =
3994 		container_of(dev, struct nvme_ctrl, ctrl_device);
3995 	struct nvmf_ctrl_options *opts = ctrl->opts;
3996 	int ret;
3997 
3998 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3999 	if (ret)
4000 		return ret;
4001 
4002 	if (opts) {
4003 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4004 		if (ret)
4005 			return ret;
4006 
4007 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4008 				opts->trsvcid ?: "none");
4009 		if (ret)
4010 			return ret;
4011 
4012 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4013 				opts->host_traddr ?: "none");
4014 		if (ret)
4015 			return ret;
4016 
4017 		ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4018 				opts->host_iface ?: "none");
4019 	}
4020 	return ret;
4021 }
4022 
4023 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4024 {
4025 	char *envp[2] = { envdata, NULL };
4026 
4027 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4028 }
4029 
4030 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4031 {
4032 	char *envp[2] = { NULL, NULL };
4033 	u32 aen_result = ctrl->aen_result;
4034 
4035 	ctrl->aen_result = 0;
4036 	if (!aen_result)
4037 		return;
4038 
4039 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4040 	if (!envp[0])
4041 		return;
4042 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4043 	kfree(envp[0]);
4044 }
4045 
4046 static void nvme_async_event_work(struct work_struct *work)
4047 {
4048 	struct nvme_ctrl *ctrl =
4049 		container_of(work, struct nvme_ctrl, async_event_work);
4050 
4051 	nvme_aen_uevent(ctrl);
4052 
4053 	/*
4054 	 * The transport drivers must guarantee AER submission here is safe by
4055 	 * flushing ctrl async_event_work after changing the controller state
4056 	 * from LIVE and before freeing the admin queue.
4057 	*/
4058 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
4059 		ctrl->ops->submit_async_event(ctrl);
4060 }
4061 
4062 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4063 {
4064 
4065 	u32 csts;
4066 
4067 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4068 		return false;
4069 
4070 	if (csts == ~0)
4071 		return false;
4072 
4073 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4074 }
4075 
4076 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4077 {
4078 	struct nvme_fw_slot_info_log *log;
4079 
4080 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4081 	if (!log)
4082 		return;
4083 
4084 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4085 			log, sizeof(*log), 0))
4086 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4087 	kfree(log);
4088 }
4089 
4090 static void nvme_fw_act_work(struct work_struct *work)
4091 {
4092 	struct nvme_ctrl *ctrl = container_of(work,
4093 				struct nvme_ctrl, fw_act_work);
4094 	unsigned long fw_act_timeout;
4095 
4096 	nvme_auth_stop(ctrl);
4097 
4098 	if (ctrl->mtfa)
4099 		fw_act_timeout = jiffies +
4100 				msecs_to_jiffies(ctrl->mtfa * 100);
4101 	else
4102 		fw_act_timeout = jiffies +
4103 				msecs_to_jiffies(admin_timeout * 1000);
4104 
4105 	nvme_quiesce_io_queues(ctrl);
4106 	while (nvme_ctrl_pp_status(ctrl)) {
4107 		if (time_after(jiffies, fw_act_timeout)) {
4108 			dev_warn(ctrl->device,
4109 				"Fw activation timeout, reset controller\n");
4110 			nvme_try_sched_reset(ctrl);
4111 			return;
4112 		}
4113 		msleep(100);
4114 	}
4115 
4116 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4117 		return;
4118 
4119 	nvme_unquiesce_io_queues(ctrl);
4120 	/* read FW slot information to clear the AER */
4121 	nvme_get_fw_slot_info(ctrl);
4122 
4123 	queue_work(nvme_wq, &ctrl->async_event_work);
4124 }
4125 
4126 static u32 nvme_aer_type(u32 result)
4127 {
4128 	return result & 0x7;
4129 }
4130 
4131 static u32 nvme_aer_subtype(u32 result)
4132 {
4133 	return (result & 0xff00) >> 8;
4134 }
4135 
4136 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4137 {
4138 	u32 aer_notice_type = nvme_aer_subtype(result);
4139 	bool requeue = true;
4140 
4141 	switch (aer_notice_type) {
4142 	case NVME_AER_NOTICE_NS_CHANGED:
4143 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4144 		nvme_queue_scan(ctrl);
4145 		break;
4146 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4147 		/*
4148 		 * We are (ab)using the RESETTING state to prevent subsequent
4149 		 * recovery actions from interfering with the controller's
4150 		 * firmware activation.
4151 		 */
4152 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4153 			requeue = false;
4154 			queue_work(nvme_wq, &ctrl->fw_act_work);
4155 		}
4156 		break;
4157 #ifdef CONFIG_NVME_MULTIPATH
4158 	case NVME_AER_NOTICE_ANA:
4159 		if (!ctrl->ana_log_buf)
4160 			break;
4161 		queue_work(nvme_wq, &ctrl->ana_work);
4162 		break;
4163 #endif
4164 	case NVME_AER_NOTICE_DISC_CHANGED:
4165 		ctrl->aen_result = result;
4166 		break;
4167 	default:
4168 		dev_warn(ctrl->device, "async event result %08x\n", result);
4169 	}
4170 	return requeue;
4171 }
4172 
4173 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4174 {
4175 	dev_warn(ctrl->device, "resetting controller due to AER\n");
4176 	nvme_reset_ctrl(ctrl);
4177 }
4178 
4179 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4180 		volatile union nvme_result *res)
4181 {
4182 	u32 result = le32_to_cpu(res->u32);
4183 	u32 aer_type = nvme_aer_type(result);
4184 	u32 aer_subtype = nvme_aer_subtype(result);
4185 	bool requeue = true;
4186 
4187 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4188 		return;
4189 
4190 	trace_nvme_async_event(ctrl, result);
4191 	switch (aer_type) {
4192 	case NVME_AER_NOTICE:
4193 		requeue = nvme_handle_aen_notice(ctrl, result);
4194 		break;
4195 	case NVME_AER_ERROR:
4196 		/*
4197 		 * For a persistent internal error, don't run async_event_work
4198 		 * to submit a new AER. The controller reset will do it.
4199 		 */
4200 		if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4201 			nvme_handle_aer_persistent_error(ctrl);
4202 			return;
4203 		}
4204 		fallthrough;
4205 	case NVME_AER_SMART:
4206 	case NVME_AER_CSS:
4207 	case NVME_AER_VS:
4208 		ctrl->aen_result = result;
4209 		break;
4210 	default:
4211 		break;
4212 	}
4213 
4214 	if (requeue)
4215 		queue_work(nvme_wq, &ctrl->async_event_work);
4216 }
4217 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4218 
4219 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4220 		const struct blk_mq_ops *ops, unsigned int cmd_size)
4221 {
4222 	int ret;
4223 
4224 	memset(set, 0, sizeof(*set));
4225 	set->ops = ops;
4226 	set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4227 	if (ctrl->ops->flags & NVME_F_FABRICS)
4228 		/* Reserved for fabric connect and keep alive */
4229 		set->reserved_tags = 2;
4230 	set->numa_node = ctrl->numa_node;
4231 	set->flags = BLK_MQ_F_NO_SCHED;
4232 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4233 		set->flags |= BLK_MQ_F_BLOCKING;
4234 	set->cmd_size = cmd_size;
4235 	set->driver_data = ctrl;
4236 	set->nr_hw_queues = 1;
4237 	set->timeout = NVME_ADMIN_TIMEOUT;
4238 	ret = blk_mq_alloc_tag_set(set);
4239 	if (ret)
4240 		return ret;
4241 
4242 	ctrl->admin_q = blk_mq_init_queue(set);
4243 	if (IS_ERR(ctrl->admin_q)) {
4244 		ret = PTR_ERR(ctrl->admin_q);
4245 		goto out_free_tagset;
4246 	}
4247 
4248 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4249 		ctrl->fabrics_q = blk_mq_init_queue(set);
4250 		if (IS_ERR(ctrl->fabrics_q)) {
4251 			ret = PTR_ERR(ctrl->fabrics_q);
4252 			goto out_cleanup_admin_q;
4253 		}
4254 	}
4255 
4256 	ctrl->admin_tagset = set;
4257 	return 0;
4258 
4259 out_cleanup_admin_q:
4260 	blk_mq_destroy_queue(ctrl->admin_q);
4261 	blk_put_queue(ctrl->admin_q);
4262 out_free_tagset:
4263 	blk_mq_free_tag_set(set);
4264 	ctrl->admin_q = NULL;
4265 	ctrl->fabrics_q = NULL;
4266 	return ret;
4267 }
4268 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4269 
4270 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4271 {
4272 	blk_mq_destroy_queue(ctrl->admin_q);
4273 	blk_put_queue(ctrl->admin_q);
4274 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4275 		blk_mq_destroy_queue(ctrl->fabrics_q);
4276 		blk_put_queue(ctrl->fabrics_q);
4277 	}
4278 	blk_mq_free_tag_set(ctrl->admin_tagset);
4279 }
4280 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4281 
4282 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4283 		const struct blk_mq_ops *ops, unsigned int nr_maps,
4284 		unsigned int cmd_size)
4285 {
4286 	int ret;
4287 
4288 	memset(set, 0, sizeof(*set));
4289 	set->ops = ops;
4290 	set->queue_depth = min_t(unsigned, ctrl->sqsize, BLK_MQ_MAX_DEPTH - 1);
4291 	/*
4292 	 * Some Apple controllers requires tags to be unique across admin and
4293 	 * the (only) I/O queue, so reserve the first 32 tags of the I/O queue.
4294 	 */
4295 	if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
4296 		set->reserved_tags = NVME_AQ_DEPTH;
4297 	else if (ctrl->ops->flags & NVME_F_FABRICS)
4298 		/* Reserved for fabric connect */
4299 		set->reserved_tags = 1;
4300 	set->numa_node = ctrl->numa_node;
4301 	set->flags = BLK_MQ_F_SHOULD_MERGE;
4302 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4303 		set->flags |= BLK_MQ_F_BLOCKING;
4304 	set->cmd_size = cmd_size,
4305 	set->driver_data = ctrl;
4306 	set->nr_hw_queues = ctrl->queue_count - 1;
4307 	set->timeout = NVME_IO_TIMEOUT;
4308 	set->nr_maps = nr_maps;
4309 	ret = blk_mq_alloc_tag_set(set);
4310 	if (ret)
4311 		return ret;
4312 
4313 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4314 		ctrl->connect_q = blk_mq_init_queue(set);
4315         	if (IS_ERR(ctrl->connect_q)) {
4316 			ret = PTR_ERR(ctrl->connect_q);
4317 			goto out_free_tag_set;
4318 		}
4319 		blk_queue_flag_set(QUEUE_FLAG_SKIP_TAGSET_QUIESCE,
4320 				   ctrl->connect_q);
4321 	}
4322 
4323 	ctrl->tagset = set;
4324 	return 0;
4325 
4326 out_free_tag_set:
4327 	blk_mq_free_tag_set(set);
4328 	ctrl->connect_q = NULL;
4329 	return ret;
4330 }
4331 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
4332 
4333 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
4334 {
4335 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4336 		blk_mq_destroy_queue(ctrl->connect_q);
4337 		blk_put_queue(ctrl->connect_q);
4338 	}
4339 	blk_mq_free_tag_set(ctrl->tagset);
4340 }
4341 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
4342 
4343 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4344 {
4345 	nvme_mpath_stop(ctrl);
4346 	nvme_auth_stop(ctrl);
4347 	nvme_stop_keep_alive(ctrl);
4348 	nvme_stop_failfast_work(ctrl);
4349 	flush_work(&ctrl->async_event_work);
4350 	cancel_work_sync(&ctrl->fw_act_work);
4351 	if (ctrl->ops->stop_ctrl)
4352 		ctrl->ops->stop_ctrl(ctrl);
4353 }
4354 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4355 
4356 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4357 {
4358 	nvme_start_keep_alive(ctrl);
4359 
4360 	nvme_enable_aen(ctrl);
4361 
4362 	/*
4363 	 * persistent discovery controllers need to send indication to userspace
4364 	 * to re-read the discovery log page to learn about possible changes
4365 	 * that were missed. We identify persistent discovery controllers by
4366 	 * checking that they started once before, hence are reconnecting back.
4367 	 */
4368 	if (test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
4369 	    nvme_discovery_ctrl(ctrl))
4370 		nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
4371 
4372 	if (ctrl->queue_count > 1) {
4373 		nvme_queue_scan(ctrl);
4374 		nvme_unquiesce_io_queues(ctrl);
4375 		nvme_mpath_update(ctrl);
4376 	}
4377 
4378 	nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4379 	set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags);
4380 }
4381 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4382 
4383 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4384 {
4385 	nvme_hwmon_exit(ctrl);
4386 	nvme_fault_inject_fini(&ctrl->fault_inject);
4387 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4388 	cdev_device_del(&ctrl->cdev, ctrl->device);
4389 	nvme_put_ctrl(ctrl);
4390 }
4391 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4392 
4393 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4394 {
4395 	struct nvme_effects_log	*cel;
4396 	unsigned long i;
4397 
4398 	xa_for_each(&ctrl->cels, i, cel) {
4399 		xa_erase(&ctrl->cels, i);
4400 		kfree(cel);
4401 	}
4402 
4403 	xa_destroy(&ctrl->cels);
4404 }
4405 
4406 static void nvme_free_ctrl(struct device *dev)
4407 {
4408 	struct nvme_ctrl *ctrl =
4409 		container_of(dev, struct nvme_ctrl, ctrl_device);
4410 	struct nvme_subsystem *subsys = ctrl->subsys;
4411 
4412 	if (!subsys || ctrl->instance != subsys->instance)
4413 		ida_free(&nvme_instance_ida, ctrl->instance);
4414 
4415 	nvme_free_cels(ctrl);
4416 	nvme_mpath_uninit(ctrl);
4417 	nvme_auth_stop(ctrl);
4418 	nvme_auth_free(ctrl);
4419 	__free_page(ctrl->discard_page);
4420 	free_opal_dev(ctrl->opal_dev);
4421 
4422 	if (subsys) {
4423 		mutex_lock(&nvme_subsystems_lock);
4424 		list_del(&ctrl->subsys_entry);
4425 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4426 		mutex_unlock(&nvme_subsystems_lock);
4427 	}
4428 
4429 	ctrl->ops->free_ctrl(ctrl);
4430 
4431 	if (subsys)
4432 		nvme_put_subsystem(subsys);
4433 }
4434 
4435 /*
4436  * Initialize a NVMe controller structures.  This needs to be called during
4437  * earliest initialization so that we have the initialized structured around
4438  * during probing.
4439  */
4440 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4441 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
4442 {
4443 	int ret;
4444 
4445 	WRITE_ONCE(ctrl->state, NVME_CTRL_NEW);
4446 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4447 	spin_lock_init(&ctrl->lock);
4448 	mutex_init(&ctrl->scan_lock);
4449 	INIT_LIST_HEAD(&ctrl->namespaces);
4450 	xa_init(&ctrl->cels);
4451 	init_rwsem(&ctrl->namespaces_rwsem);
4452 	ctrl->dev = dev;
4453 	ctrl->ops = ops;
4454 	ctrl->quirks = quirks;
4455 	ctrl->numa_node = NUMA_NO_NODE;
4456 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4457 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4458 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4459 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4460 	init_waitqueue_head(&ctrl->state_wq);
4461 
4462 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4463 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4464 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4465 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4466 
4467 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4468 			PAGE_SIZE);
4469 	ctrl->discard_page = alloc_page(GFP_KERNEL);
4470 	if (!ctrl->discard_page) {
4471 		ret = -ENOMEM;
4472 		goto out;
4473 	}
4474 
4475 	ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4476 	if (ret < 0)
4477 		goto out;
4478 	ctrl->instance = ret;
4479 
4480 	device_initialize(&ctrl->ctrl_device);
4481 	ctrl->device = &ctrl->ctrl_device;
4482 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4483 			ctrl->instance);
4484 	ctrl->device->class = nvme_class;
4485 	ctrl->device->parent = ctrl->dev;
4486 	if (ops->dev_attr_groups)
4487 		ctrl->device->groups = ops->dev_attr_groups;
4488 	else
4489 		ctrl->device->groups = nvme_dev_attr_groups;
4490 	ctrl->device->release = nvme_free_ctrl;
4491 	dev_set_drvdata(ctrl->device, ctrl);
4492 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4493 	if (ret)
4494 		goto out_release_instance;
4495 
4496 	nvme_get_ctrl(ctrl);
4497 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
4498 	ctrl->cdev.owner = ops->module;
4499 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4500 	if (ret)
4501 		goto out_free_name;
4502 
4503 	/*
4504 	 * Initialize latency tolerance controls.  The sysfs files won't
4505 	 * be visible to userspace unless the device actually supports APST.
4506 	 */
4507 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4508 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4509 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4510 
4511 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4512 	nvme_mpath_init_ctrl(ctrl);
4513 	ret = nvme_auth_init_ctrl(ctrl);
4514 	if (ret)
4515 		goto out_free_cdev;
4516 
4517 	return 0;
4518 out_free_cdev:
4519 	nvme_fault_inject_fini(&ctrl->fault_inject);
4520 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4521 	cdev_device_del(&ctrl->cdev, ctrl->device);
4522 out_free_name:
4523 	nvme_put_ctrl(ctrl);
4524 	kfree_const(ctrl->device->kobj.name);
4525 out_release_instance:
4526 	ida_free(&nvme_instance_ida, ctrl->instance);
4527 out:
4528 	if (ctrl->discard_page)
4529 		__free_page(ctrl->discard_page);
4530 	return ret;
4531 }
4532 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4533 
4534 /* let I/O to all namespaces fail in preparation for surprise removal */
4535 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
4536 {
4537 	struct nvme_ns *ns;
4538 
4539 	down_read(&ctrl->namespaces_rwsem);
4540 	list_for_each_entry(ns, &ctrl->namespaces, list)
4541 		blk_mark_disk_dead(ns->disk);
4542 	up_read(&ctrl->namespaces_rwsem);
4543 }
4544 EXPORT_SYMBOL_GPL(nvme_mark_namespaces_dead);
4545 
4546 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4547 {
4548 	struct nvme_ns *ns;
4549 
4550 	down_read(&ctrl->namespaces_rwsem);
4551 	list_for_each_entry(ns, &ctrl->namespaces, list)
4552 		blk_mq_unfreeze_queue(ns->queue);
4553 	up_read(&ctrl->namespaces_rwsem);
4554 	clear_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4555 }
4556 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4557 
4558 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4559 {
4560 	struct nvme_ns *ns;
4561 
4562 	down_read(&ctrl->namespaces_rwsem);
4563 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4564 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4565 		if (timeout <= 0)
4566 			break;
4567 	}
4568 	up_read(&ctrl->namespaces_rwsem);
4569 	return timeout;
4570 }
4571 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4572 
4573 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4574 {
4575 	struct nvme_ns *ns;
4576 
4577 	down_read(&ctrl->namespaces_rwsem);
4578 	list_for_each_entry(ns, &ctrl->namespaces, list)
4579 		blk_mq_freeze_queue_wait(ns->queue);
4580 	up_read(&ctrl->namespaces_rwsem);
4581 }
4582 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4583 
4584 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4585 {
4586 	struct nvme_ns *ns;
4587 
4588 	set_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4589 	down_read(&ctrl->namespaces_rwsem);
4590 	list_for_each_entry(ns, &ctrl->namespaces, list)
4591 		blk_freeze_queue_start(ns->queue);
4592 	up_read(&ctrl->namespaces_rwsem);
4593 }
4594 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4595 
4596 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl)
4597 {
4598 	if (!ctrl->tagset)
4599 		return;
4600 	if (!test_and_set_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4601 		blk_mq_quiesce_tagset(ctrl->tagset);
4602 	else
4603 		blk_mq_wait_quiesce_done(ctrl->tagset);
4604 }
4605 EXPORT_SYMBOL_GPL(nvme_quiesce_io_queues);
4606 
4607 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl)
4608 {
4609 	if (!ctrl->tagset)
4610 		return;
4611 	if (test_and_clear_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4612 		blk_mq_unquiesce_tagset(ctrl->tagset);
4613 }
4614 EXPORT_SYMBOL_GPL(nvme_unquiesce_io_queues);
4615 
4616 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl)
4617 {
4618 	if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4619 		blk_mq_quiesce_queue(ctrl->admin_q);
4620 	else
4621 		blk_mq_wait_quiesce_done(ctrl->admin_q->tag_set);
4622 }
4623 EXPORT_SYMBOL_GPL(nvme_quiesce_admin_queue);
4624 
4625 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl)
4626 {
4627 	if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4628 		blk_mq_unquiesce_queue(ctrl->admin_q);
4629 }
4630 EXPORT_SYMBOL_GPL(nvme_unquiesce_admin_queue);
4631 
4632 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4633 {
4634 	struct nvme_ns *ns;
4635 
4636 	down_read(&ctrl->namespaces_rwsem);
4637 	list_for_each_entry(ns, &ctrl->namespaces, list)
4638 		blk_sync_queue(ns->queue);
4639 	up_read(&ctrl->namespaces_rwsem);
4640 }
4641 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4642 
4643 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4644 {
4645 	nvme_sync_io_queues(ctrl);
4646 	if (ctrl->admin_q)
4647 		blk_sync_queue(ctrl->admin_q);
4648 }
4649 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4650 
4651 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4652 {
4653 	if (file->f_op != &nvme_dev_fops)
4654 		return NULL;
4655 	return file->private_data;
4656 }
4657 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4658 
4659 /*
4660  * Check we didn't inadvertently grow the command structure sizes:
4661  */
4662 static inline void _nvme_check_size(void)
4663 {
4664 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4665 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4666 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4667 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4668 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4669 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4670 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4671 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4672 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4673 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4674 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4675 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4676 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4677 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
4678 			NVME_IDENTIFY_DATA_SIZE);
4679 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4680 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
4681 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4682 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4683 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4684 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4685 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4686 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4687 	BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
4688 }
4689 
4690 
4691 static int __init nvme_core_init(void)
4692 {
4693 	int result = -ENOMEM;
4694 
4695 	_nvme_check_size();
4696 
4697 	nvme_wq = alloc_workqueue("nvme-wq",
4698 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4699 	if (!nvme_wq)
4700 		goto out;
4701 
4702 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4703 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4704 	if (!nvme_reset_wq)
4705 		goto destroy_wq;
4706 
4707 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4708 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4709 	if (!nvme_delete_wq)
4710 		goto destroy_reset_wq;
4711 
4712 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4713 			NVME_MINORS, "nvme");
4714 	if (result < 0)
4715 		goto destroy_delete_wq;
4716 
4717 	nvme_class = class_create("nvme");
4718 	if (IS_ERR(nvme_class)) {
4719 		result = PTR_ERR(nvme_class);
4720 		goto unregister_chrdev;
4721 	}
4722 	nvme_class->dev_uevent = nvme_class_uevent;
4723 
4724 	nvme_subsys_class = class_create("nvme-subsystem");
4725 	if (IS_ERR(nvme_subsys_class)) {
4726 		result = PTR_ERR(nvme_subsys_class);
4727 		goto destroy_class;
4728 	}
4729 
4730 	result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4731 				     "nvme-generic");
4732 	if (result < 0)
4733 		goto destroy_subsys_class;
4734 
4735 	nvme_ns_chr_class = class_create("nvme-generic");
4736 	if (IS_ERR(nvme_ns_chr_class)) {
4737 		result = PTR_ERR(nvme_ns_chr_class);
4738 		goto unregister_generic_ns;
4739 	}
4740 
4741 	result = nvme_init_auth();
4742 	if (result)
4743 		goto destroy_ns_chr;
4744 	return 0;
4745 
4746 destroy_ns_chr:
4747 	class_destroy(nvme_ns_chr_class);
4748 unregister_generic_ns:
4749 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4750 destroy_subsys_class:
4751 	class_destroy(nvme_subsys_class);
4752 destroy_class:
4753 	class_destroy(nvme_class);
4754 unregister_chrdev:
4755 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4756 destroy_delete_wq:
4757 	destroy_workqueue(nvme_delete_wq);
4758 destroy_reset_wq:
4759 	destroy_workqueue(nvme_reset_wq);
4760 destroy_wq:
4761 	destroy_workqueue(nvme_wq);
4762 out:
4763 	return result;
4764 }
4765 
4766 static void __exit nvme_core_exit(void)
4767 {
4768 	nvme_exit_auth();
4769 	class_destroy(nvme_ns_chr_class);
4770 	class_destroy(nvme_subsys_class);
4771 	class_destroy(nvme_class);
4772 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4773 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4774 	destroy_workqueue(nvme_delete_wq);
4775 	destroy_workqueue(nvme_reset_wq);
4776 	destroy_workqueue(nvme_wq);
4777 	ida_destroy(&nvme_ns_chr_minor_ida);
4778 	ida_destroy(&nvme_instance_ida);
4779 }
4780 
4781 MODULE_LICENSE("GPL");
4782 MODULE_VERSION("1.0");
4783 module_init(nvme_core_init);
4784 module_exit(nvme_core_exit);
4785