xref: /linux/drivers/nvme/host/core.c (revision 268531be211f18c55f7ff5a1641d32c0fd0571cd)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * NVM Express device driver
4  * Copyright (c) 2011-2014, Intel Corporation.
5  */
6 
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/blk-integrity.h>
10 #include <linux/compat.h>
11 #include <linux/delay.h>
12 #include <linux/errno.h>
13 #include <linux/hdreg.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
19 #include <linux/pr.h>
20 #include <linux/ptrace.h>
21 #include <linux/nvme_ioctl.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
24 
25 #include "nvme.h"
26 #include "fabrics.h"
27 #include <linux/nvme-auth.h>
28 
29 #define CREATE_TRACE_POINTS
30 #include "trace.h"
31 
32 #define NVME_MINORS		(1U << MINORBITS)
33 
34 struct nvme_ns_info {
35 	struct nvme_ns_ids ids;
36 	u32 nsid;
37 	__le32 anagrpid;
38 	bool is_shared;
39 	bool is_readonly;
40 	bool is_ready;
41 	bool is_removed;
42 };
43 
44 unsigned int admin_timeout = 60;
45 module_param(admin_timeout, uint, 0644);
46 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
47 EXPORT_SYMBOL_GPL(admin_timeout);
48 
49 unsigned int nvme_io_timeout = 30;
50 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
51 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
52 EXPORT_SYMBOL_GPL(nvme_io_timeout);
53 
54 static unsigned char shutdown_timeout = 5;
55 module_param(shutdown_timeout, byte, 0644);
56 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
57 
58 static u8 nvme_max_retries = 5;
59 module_param_named(max_retries, nvme_max_retries, byte, 0644);
60 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
61 
62 static unsigned long default_ps_max_latency_us = 100000;
63 module_param(default_ps_max_latency_us, ulong, 0644);
64 MODULE_PARM_DESC(default_ps_max_latency_us,
65 		 "max power saving latency for new devices; use PM QOS to change per device");
66 
67 static bool force_apst;
68 module_param(force_apst, bool, 0644);
69 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
70 
71 static unsigned long apst_primary_timeout_ms = 100;
72 module_param(apst_primary_timeout_ms, ulong, 0644);
73 MODULE_PARM_DESC(apst_primary_timeout_ms,
74 	"primary APST timeout in ms");
75 
76 static unsigned long apst_secondary_timeout_ms = 2000;
77 module_param(apst_secondary_timeout_ms, ulong, 0644);
78 MODULE_PARM_DESC(apst_secondary_timeout_ms,
79 	"secondary APST timeout in ms");
80 
81 static unsigned long apst_primary_latency_tol_us = 15000;
82 module_param(apst_primary_latency_tol_us, ulong, 0644);
83 MODULE_PARM_DESC(apst_primary_latency_tol_us,
84 	"primary APST latency tolerance in us");
85 
86 static unsigned long apst_secondary_latency_tol_us = 100000;
87 module_param(apst_secondary_latency_tol_us, ulong, 0644);
88 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
89 	"secondary APST latency tolerance in us");
90 
91 /*
92  * nvme_wq - hosts nvme related works that are not reset or delete
93  * nvme_reset_wq - hosts nvme reset works
94  * nvme_delete_wq - hosts nvme delete works
95  *
96  * nvme_wq will host works such as scan, aen handling, fw activation,
97  * keep-alive, periodic reconnects etc. nvme_reset_wq
98  * runs reset works which also flush works hosted on nvme_wq for
99  * serialization purposes. nvme_delete_wq host controller deletion
100  * works which flush reset works for serialization.
101  */
102 struct workqueue_struct *nvme_wq;
103 EXPORT_SYMBOL_GPL(nvme_wq);
104 
105 struct workqueue_struct *nvme_reset_wq;
106 EXPORT_SYMBOL_GPL(nvme_reset_wq);
107 
108 struct workqueue_struct *nvme_delete_wq;
109 EXPORT_SYMBOL_GPL(nvme_delete_wq);
110 
111 static LIST_HEAD(nvme_subsystems);
112 static DEFINE_MUTEX(nvme_subsystems_lock);
113 
114 static DEFINE_IDA(nvme_instance_ida);
115 static dev_t nvme_ctrl_base_chr_devt;
116 static struct class *nvme_class;
117 static struct class *nvme_subsys_class;
118 
119 static DEFINE_IDA(nvme_ns_chr_minor_ida);
120 static dev_t nvme_ns_chr_devt;
121 static struct class *nvme_ns_chr_class;
122 
123 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
124 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
125 					   unsigned nsid);
126 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
127 				   struct nvme_command *cmd);
128 
129 void nvme_queue_scan(struct nvme_ctrl *ctrl)
130 {
131 	/*
132 	 * Only new queue scan work when admin and IO queues are both alive
133 	 */
134 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE && ctrl->tagset)
135 		queue_work(nvme_wq, &ctrl->scan_work);
136 }
137 
138 /*
139  * Use this function to proceed with scheduling reset_work for a controller
140  * that had previously been set to the resetting state. This is intended for
141  * code paths that can't be interrupted by other reset attempts. A hot removal
142  * may prevent this from succeeding.
143  */
144 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
145 {
146 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_RESETTING)
147 		return -EBUSY;
148 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
149 		return -EBUSY;
150 	return 0;
151 }
152 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
153 
154 static void nvme_failfast_work(struct work_struct *work)
155 {
156 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
157 			struct nvme_ctrl, failfast_work);
158 
159 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_CONNECTING)
160 		return;
161 
162 	set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
163 	dev_info(ctrl->device, "failfast expired\n");
164 	nvme_kick_requeue_lists(ctrl);
165 }
166 
167 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
168 {
169 	if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
170 		return;
171 
172 	schedule_delayed_work(&ctrl->failfast_work,
173 			      ctrl->opts->fast_io_fail_tmo * HZ);
174 }
175 
176 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
177 {
178 	if (!ctrl->opts)
179 		return;
180 
181 	cancel_delayed_work_sync(&ctrl->failfast_work);
182 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
183 }
184 
185 
186 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
187 {
188 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
189 		return -EBUSY;
190 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
191 		return -EBUSY;
192 	return 0;
193 }
194 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
195 
196 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
197 {
198 	int ret;
199 
200 	ret = nvme_reset_ctrl(ctrl);
201 	if (!ret) {
202 		flush_work(&ctrl->reset_work);
203 		if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE)
204 			ret = -ENETRESET;
205 	}
206 
207 	return ret;
208 }
209 
210 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
211 {
212 	dev_info(ctrl->device,
213 		 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
214 
215 	flush_work(&ctrl->reset_work);
216 	nvme_stop_ctrl(ctrl);
217 	nvme_remove_namespaces(ctrl);
218 	ctrl->ops->delete_ctrl(ctrl);
219 	nvme_uninit_ctrl(ctrl);
220 }
221 
222 static void nvme_delete_ctrl_work(struct work_struct *work)
223 {
224 	struct nvme_ctrl *ctrl =
225 		container_of(work, struct nvme_ctrl, delete_work);
226 
227 	nvme_do_delete_ctrl(ctrl);
228 }
229 
230 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
231 {
232 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
233 		return -EBUSY;
234 	if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
235 		return -EBUSY;
236 	return 0;
237 }
238 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
239 
240 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
241 {
242 	/*
243 	 * Keep a reference until nvme_do_delete_ctrl() complete,
244 	 * since ->delete_ctrl can free the controller.
245 	 */
246 	nvme_get_ctrl(ctrl);
247 	if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
248 		nvme_do_delete_ctrl(ctrl);
249 	nvme_put_ctrl(ctrl);
250 }
251 
252 static blk_status_t nvme_error_status(u16 status)
253 {
254 	switch (status & 0x7ff) {
255 	case NVME_SC_SUCCESS:
256 		return BLK_STS_OK;
257 	case NVME_SC_CAP_EXCEEDED:
258 		return BLK_STS_NOSPC;
259 	case NVME_SC_LBA_RANGE:
260 	case NVME_SC_CMD_INTERRUPTED:
261 	case NVME_SC_NS_NOT_READY:
262 		return BLK_STS_TARGET;
263 	case NVME_SC_BAD_ATTRIBUTES:
264 	case NVME_SC_ONCS_NOT_SUPPORTED:
265 	case NVME_SC_INVALID_OPCODE:
266 	case NVME_SC_INVALID_FIELD:
267 	case NVME_SC_INVALID_NS:
268 		return BLK_STS_NOTSUPP;
269 	case NVME_SC_WRITE_FAULT:
270 	case NVME_SC_READ_ERROR:
271 	case NVME_SC_UNWRITTEN_BLOCK:
272 	case NVME_SC_ACCESS_DENIED:
273 	case NVME_SC_READ_ONLY:
274 	case NVME_SC_COMPARE_FAILED:
275 		return BLK_STS_MEDIUM;
276 	case NVME_SC_GUARD_CHECK:
277 	case NVME_SC_APPTAG_CHECK:
278 	case NVME_SC_REFTAG_CHECK:
279 	case NVME_SC_INVALID_PI:
280 		return BLK_STS_PROTECTION;
281 	case NVME_SC_RESERVATION_CONFLICT:
282 		return BLK_STS_RESV_CONFLICT;
283 	case NVME_SC_HOST_PATH_ERROR:
284 		return BLK_STS_TRANSPORT;
285 	case NVME_SC_ZONE_TOO_MANY_ACTIVE:
286 		return BLK_STS_ZONE_ACTIVE_RESOURCE;
287 	case NVME_SC_ZONE_TOO_MANY_OPEN:
288 		return BLK_STS_ZONE_OPEN_RESOURCE;
289 	default:
290 		return BLK_STS_IOERR;
291 	}
292 }
293 
294 static void nvme_retry_req(struct request *req)
295 {
296 	unsigned long delay = 0;
297 	u16 crd;
298 
299 	/* The mask and shift result must be <= 3 */
300 	crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
301 	if (crd)
302 		delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
303 
304 	nvme_req(req)->retries++;
305 	blk_mq_requeue_request(req, false);
306 	blk_mq_delay_kick_requeue_list(req->q, delay);
307 }
308 
309 static void nvme_log_error(struct request *req)
310 {
311 	struct nvme_ns *ns = req->q->queuedata;
312 	struct nvme_request *nr = nvme_req(req);
313 
314 	if (ns) {
315 		pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %llu blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
316 		       ns->disk ? ns->disk->disk_name : "?",
317 		       nvme_get_opcode_str(nr->cmd->common.opcode),
318 		       nr->cmd->common.opcode,
319 		       (unsigned long long)nvme_sect_to_lba(ns, blk_rq_pos(req)),
320 		       (unsigned long long)blk_rq_bytes(req) >> ns->lba_shift,
321 		       nvme_get_error_status_str(nr->status),
322 		       nr->status >> 8 & 7,	/* Status Code Type */
323 		       nr->status & 0xff,	/* Status Code */
324 		       nr->status & NVME_SC_MORE ? "MORE " : "",
325 		       nr->status & NVME_SC_DNR  ? "DNR "  : "");
326 		return;
327 	}
328 
329 	pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
330 			   dev_name(nr->ctrl->device),
331 			   nvme_get_admin_opcode_str(nr->cmd->common.opcode),
332 			   nr->cmd->common.opcode,
333 			   nvme_get_error_status_str(nr->status),
334 			   nr->status >> 8 & 7,	/* Status Code Type */
335 			   nr->status & 0xff,	/* Status Code */
336 			   nr->status & NVME_SC_MORE ? "MORE " : "",
337 			   nr->status & NVME_SC_DNR  ? "DNR "  : "");
338 }
339 
340 enum nvme_disposition {
341 	COMPLETE,
342 	RETRY,
343 	FAILOVER,
344 	AUTHENTICATE,
345 };
346 
347 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
348 {
349 	if (likely(nvme_req(req)->status == 0))
350 		return COMPLETE;
351 
352 	if ((nvme_req(req)->status & 0x7ff) == NVME_SC_AUTH_REQUIRED)
353 		return AUTHENTICATE;
354 
355 	if (blk_noretry_request(req) ||
356 	    (nvme_req(req)->status & NVME_SC_DNR) ||
357 	    nvme_req(req)->retries >= nvme_max_retries)
358 		return COMPLETE;
359 
360 	if (req->cmd_flags & REQ_NVME_MPATH) {
361 		if (nvme_is_path_error(nvme_req(req)->status) ||
362 		    blk_queue_dying(req->q))
363 			return FAILOVER;
364 	} else {
365 		if (blk_queue_dying(req->q))
366 			return COMPLETE;
367 	}
368 
369 	return RETRY;
370 }
371 
372 static inline void nvme_end_req_zoned(struct request *req)
373 {
374 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
375 	    req_op(req) == REQ_OP_ZONE_APPEND)
376 		req->__sector = nvme_lba_to_sect(req->q->queuedata,
377 			le64_to_cpu(nvme_req(req)->result.u64));
378 }
379 
380 static inline void nvme_end_req(struct request *req)
381 {
382 	blk_status_t status = nvme_error_status(nvme_req(req)->status);
383 
384 	if (unlikely(nvme_req(req)->status && !(req->rq_flags & RQF_QUIET)))
385 		nvme_log_error(req);
386 	nvme_end_req_zoned(req);
387 	nvme_trace_bio_complete(req);
388 	if (req->cmd_flags & REQ_NVME_MPATH)
389 		nvme_mpath_end_request(req);
390 	blk_mq_end_request(req, status);
391 }
392 
393 void nvme_complete_rq(struct request *req)
394 {
395 	struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
396 
397 	trace_nvme_complete_rq(req);
398 	nvme_cleanup_cmd(req);
399 
400 	/*
401 	 * Completions of long-running commands should not be able to
402 	 * defer sending of periodic keep alives, since the controller
403 	 * may have completed processing such commands a long time ago
404 	 * (arbitrarily close to command submission time).
405 	 * req->deadline - req->timeout is the command submission time
406 	 * in jiffies.
407 	 */
408 	if (ctrl->kas &&
409 	    req->deadline - req->timeout >= ctrl->ka_last_check_time)
410 		ctrl->comp_seen = true;
411 
412 	switch (nvme_decide_disposition(req)) {
413 	case COMPLETE:
414 		nvme_end_req(req);
415 		return;
416 	case RETRY:
417 		nvme_retry_req(req);
418 		return;
419 	case FAILOVER:
420 		nvme_failover_req(req);
421 		return;
422 	case AUTHENTICATE:
423 #ifdef CONFIG_NVME_HOST_AUTH
424 		queue_work(nvme_wq, &ctrl->dhchap_auth_work);
425 		nvme_retry_req(req);
426 #else
427 		nvme_end_req(req);
428 #endif
429 		return;
430 	}
431 }
432 EXPORT_SYMBOL_GPL(nvme_complete_rq);
433 
434 void nvme_complete_batch_req(struct request *req)
435 {
436 	trace_nvme_complete_rq(req);
437 	nvme_cleanup_cmd(req);
438 	nvme_end_req_zoned(req);
439 }
440 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
441 
442 /*
443  * Called to unwind from ->queue_rq on a failed command submission so that the
444  * multipathing code gets called to potentially failover to another path.
445  * The caller needs to unwind all transport specific resource allocations and
446  * must return propagate the return value.
447  */
448 blk_status_t nvme_host_path_error(struct request *req)
449 {
450 	nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
451 	blk_mq_set_request_complete(req);
452 	nvme_complete_rq(req);
453 	return BLK_STS_OK;
454 }
455 EXPORT_SYMBOL_GPL(nvme_host_path_error);
456 
457 bool nvme_cancel_request(struct request *req, void *data)
458 {
459 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
460 				"Cancelling I/O %d", req->tag);
461 
462 	/* don't abort one completed or idle request */
463 	if (blk_mq_rq_state(req) != MQ_RQ_IN_FLIGHT)
464 		return true;
465 
466 	nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
467 	nvme_req(req)->flags |= NVME_REQ_CANCELLED;
468 	blk_mq_complete_request(req);
469 	return true;
470 }
471 EXPORT_SYMBOL_GPL(nvme_cancel_request);
472 
473 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
474 {
475 	if (ctrl->tagset) {
476 		blk_mq_tagset_busy_iter(ctrl->tagset,
477 				nvme_cancel_request, ctrl);
478 		blk_mq_tagset_wait_completed_request(ctrl->tagset);
479 	}
480 }
481 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
482 
483 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
484 {
485 	if (ctrl->admin_tagset) {
486 		blk_mq_tagset_busy_iter(ctrl->admin_tagset,
487 				nvme_cancel_request, ctrl);
488 		blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
489 	}
490 }
491 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
492 
493 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
494 		enum nvme_ctrl_state new_state)
495 {
496 	enum nvme_ctrl_state old_state;
497 	unsigned long flags;
498 	bool changed = false;
499 
500 	spin_lock_irqsave(&ctrl->lock, flags);
501 
502 	old_state = nvme_ctrl_state(ctrl);
503 	switch (new_state) {
504 	case NVME_CTRL_LIVE:
505 		switch (old_state) {
506 		case NVME_CTRL_NEW:
507 		case NVME_CTRL_RESETTING:
508 		case NVME_CTRL_CONNECTING:
509 			changed = true;
510 			fallthrough;
511 		default:
512 			break;
513 		}
514 		break;
515 	case NVME_CTRL_RESETTING:
516 		switch (old_state) {
517 		case NVME_CTRL_NEW:
518 		case NVME_CTRL_LIVE:
519 			changed = true;
520 			fallthrough;
521 		default:
522 			break;
523 		}
524 		break;
525 	case NVME_CTRL_CONNECTING:
526 		switch (old_state) {
527 		case NVME_CTRL_NEW:
528 		case NVME_CTRL_RESETTING:
529 			changed = true;
530 			fallthrough;
531 		default:
532 			break;
533 		}
534 		break;
535 	case NVME_CTRL_DELETING:
536 		switch (old_state) {
537 		case NVME_CTRL_LIVE:
538 		case NVME_CTRL_RESETTING:
539 		case NVME_CTRL_CONNECTING:
540 			changed = true;
541 			fallthrough;
542 		default:
543 			break;
544 		}
545 		break;
546 	case NVME_CTRL_DELETING_NOIO:
547 		switch (old_state) {
548 		case NVME_CTRL_DELETING:
549 		case NVME_CTRL_DEAD:
550 			changed = true;
551 			fallthrough;
552 		default:
553 			break;
554 		}
555 		break;
556 	case NVME_CTRL_DEAD:
557 		switch (old_state) {
558 		case NVME_CTRL_DELETING:
559 			changed = true;
560 			fallthrough;
561 		default:
562 			break;
563 		}
564 		break;
565 	default:
566 		break;
567 	}
568 
569 	if (changed) {
570 		WRITE_ONCE(ctrl->state, new_state);
571 		wake_up_all(&ctrl->state_wq);
572 	}
573 
574 	spin_unlock_irqrestore(&ctrl->lock, flags);
575 	if (!changed)
576 		return false;
577 
578 	if (new_state == NVME_CTRL_LIVE) {
579 		if (old_state == NVME_CTRL_CONNECTING)
580 			nvme_stop_failfast_work(ctrl);
581 		nvme_kick_requeue_lists(ctrl);
582 	} else if (new_state == NVME_CTRL_CONNECTING &&
583 		old_state == NVME_CTRL_RESETTING) {
584 		nvme_start_failfast_work(ctrl);
585 	}
586 	return changed;
587 }
588 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
589 
590 /*
591  * Returns true for sink states that can't ever transition back to live.
592  */
593 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
594 {
595 	switch (nvme_ctrl_state(ctrl)) {
596 	case NVME_CTRL_NEW:
597 	case NVME_CTRL_LIVE:
598 	case NVME_CTRL_RESETTING:
599 	case NVME_CTRL_CONNECTING:
600 		return false;
601 	case NVME_CTRL_DELETING:
602 	case NVME_CTRL_DELETING_NOIO:
603 	case NVME_CTRL_DEAD:
604 		return true;
605 	default:
606 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
607 		return true;
608 	}
609 }
610 
611 /*
612  * Waits for the controller state to be resetting, or returns false if it is
613  * not possible to ever transition to that state.
614  */
615 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
616 {
617 	wait_event(ctrl->state_wq,
618 		   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
619 		   nvme_state_terminal(ctrl));
620 	return nvme_ctrl_state(ctrl) == NVME_CTRL_RESETTING;
621 }
622 EXPORT_SYMBOL_GPL(nvme_wait_reset);
623 
624 static void nvme_free_ns_head(struct kref *ref)
625 {
626 	struct nvme_ns_head *head =
627 		container_of(ref, struct nvme_ns_head, ref);
628 
629 	nvme_mpath_remove_disk(head);
630 	ida_free(&head->subsys->ns_ida, head->instance);
631 	cleanup_srcu_struct(&head->srcu);
632 	nvme_put_subsystem(head->subsys);
633 	kfree(head);
634 }
635 
636 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
637 {
638 	return kref_get_unless_zero(&head->ref);
639 }
640 
641 void nvme_put_ns_head(struct nvme_ns_head *head)
642 {
643 	kref_put(&head->ref, nvme_free_ns_head);
644 }
645 
646 static void nvme_free_ns(struct kref *kref)
647 {
648 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
649 
650 	put_disk(ns->disk);
651 	nvme_put_ns_head(ns->head);
652 	nvme_put_ctrl(ns->ctrl);
653 	kfree(ns);
654 }
655 
656 static inline bool nvme_get_ns(struct nvme_ns *ns)
657 {
658 	return kref_get_unless_zero(&ns->kref);
659 }
660 
661 void nvme_put_ns(struct nvme_ns *ns)
662 {
663 	kref_put(&ns->kref, nvme_free_ns);
664 }
665 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
666 
667 static inline void nvme_clear_nvme_request(struct request *req)
668 {
669 	nvme_req(req)->status = 0;
670 	nvme_req(req)->retries = 0;
671 	nvme_req(req)->flags = 0;
672 	req->rq_flags |= RQF_DONTPREP;
673 }
674 
675 /* initialize a passthrough request */
676 void nvme_init_request(struct request *req, struct nvme_command *cmd)
677 {
678 	if (req->q->queuedata)
679 		req->timeout = NVME_IO_TIMEOUT;
680 	else /* no queuedata implies admin queue */
681 		req->timeout = NVME_ADMIN_TIMEOUT;
682 
683 	/* passthru commands should let the driver set the SGL flags */
684 	cmd->common.flags &= ~NVME_CMD_SGL_ALL;
685 
686 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
687 	if (req->mq_hctx->type == HCTX_TYPE_POLL)
688 		req->cmd_flags |= REQ_POLLED;
689 	nvme_clear_nvme_request(req);
690 	req->rq_flags |= RQF_QUIET;
691 	memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
692 }
693 EXPORT_SYMBOL_GPL(nvme_init_request);
694 
695 /*
696  * For something we're not in a state to send to the device the default action
697  * is to busy it and retry it after the controller state is recovered.  However,
698  * if the controller is deleting or if anything is marked for failfast or
699  * nvme multipath it is immediately failed.
700  *
701  * Note: commands used to initialize the controller will be marked for failfast.
702  * Note: nvme cli/ioctl commands are marked for failfast.
703  */
704 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
705 		struct request *rq)
706 {
707 	enum nvme_ctrl_state state = nvme_ctrl_state(ctrl);
708 
709 	if (state != NVME_CTRL_DELETING_NOIO &&
710 	    state != NVME_CTRL_DELETING &&
711 	    state != NVME_CTRL_DEAD &&
712 	    !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
713 	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
714 		return BLK_STS_RESOURCE;
715 	return nvme_host_path_error(rq);
716 }
717 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
718 
719 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
720 		bool queue_live)
721 {
722 	struct nvme_request *req = nvme_req(rq);
723 
724 	/*
725 	 * currently we have a problem sending passthru commands
726 	 * on the admin_q if the controller is not LIVE because we can't
727 	 * make sure that they are going out after the admin connect,
728 	 * controller enable and/or other commands in the initialization
729 	 * sequence. until the controller will be LIVE, fail with
730 	 * BLK_STS_RESOURCE so that they will be rescheduled.
731 	 */
732 	if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
733 		return false;
734 
735 	if (ctrl->ops->flags & NVME_F_FABRICS) {
736 		/*
737 		 * Only allow commands on a live queue, except for the connect
738 		 * command, which is require to set the queue live in the
739 		 * appropinquate states.
740 		 */
741 		switch (nvme_ctrl_state(ctrl)) {
742 		case NVME_CTRL_CONNECTING:
743 			if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
744 			    (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
745 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
746 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
747 				return true;
748 			break;
749 		default:
750 			break;
751 		case NVME_CTRL_DEAD:
752 			return false;
753 		}
754 	}
755 
756 	return queue_live;
757 }
758 EXPORT_SYMBOL_GPL(__nvme_check_ready);
759 
760 static inline void nvme_setup_flush(struct nvme_ns *ns,
761 		struct nvme_command *cmnd)
762 {
763 	memset(cmnd, 0, sizeof(*cmnd));
764 	cmnd->common.opcode = nvme_cmd_flush;
765 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
766 }
767 
768 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
769 		struct nvme_command *cmnd)
770 {
771 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
772 	struct nvme_dsm_range *range;
773 	struct bio *bio;
774 
775 	/*
776 	 * Some devices do not consider the DSM 'Number of Ranges' field when
777 	 * determining how much data to DMA. Always allocate memory for maximum
778 	 * number of segments to prevent device reading beyond end of buffer.
779 	 */
780 	static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
781 
782 	range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
783 	if (!range) {
784 		/*
785 		 * If we fail allocation our range, fallback to the controller
786 		 * discard page. If that's also busy, it's safe to return
787 		 * busy, as we know we can make progress once that's freed.
788 		 */
789 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
790 			return BLK_STS_RESOURCE;
791 
792 		range = page_address(ns->ctrl->discard_page);
793 	}
794 
795 	if (queue_max_discard_segments(req->q) == 1) {
796 		u64 slba = nvme_sect_to_lba(ns, blk_rq_pos(req));
797 		u32 nlb = blk_rq_sectors(req) >> (ns->lba_shift - 9);
798 
799 		range[0].cattr = cpu_to_le32(0);
800 		range[0].nlb = cpu_to_le32(nlb);
801 		range[0].slba = cpu_to_le64(slba);
802 		n = 1;
803 	} else {
804 		__rq_for_each_bio(bio, req) {
805 			u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
806 			u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
807 
808 			if (n < segments) {
809 				range[n].cattr = cpu_to_le32(0);
810 				range[n].nlb = cpu_to_le32(nlb);
811 				range[n].slba = cpu_to_le64(slba);
812 			}
813 			n++;
814 		}
815 	}
816 
817 	if (WARN_ON_ONCE(n != segments)) {
818 		if (virt_to_page(range) == ns->ctrl->discard_page)
819 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
820 		else
821 			kfree(range);
822 		return BLK_STS_IOERR;
823 	}
824 
825 	memset(cmnd, 0, sizeof(*cmnd));
826 	cmnd->dsm.opcode = nvme_cmd_dsm;
827 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
828 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
829 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
830 
831 	bvec_set_virt(&req->special_vec, range, alloc_size);
832 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
833 
834 	return BLK_STS_OK;
835 }
836 
837 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
838 			      struct request *req)
839 {
840 	u32 upper, lower;
841 	u64 ref48;
842 
843 	/* both rw and write zeroes share the same reftag format */
844 	switch (ns->guard_type) {
845 	case NVME_NVM_NS_16B_GUARD:
846 		cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
847 		break;
848 	case NVME_NVM_NS_64B_GUARD:
849 		ref48 = ext_pi_ref_tag(req);
850 		lower = lower_32_bits(ref48);
851 		upper = upper_32_bits(ref48);
852 
853 		cmnd->rw.reftag = cpu_to_le32(lower);
854 		cmnd->rw.cdw3 = cpu_to_le32(upper);
855 		break;
856 	default:
857 		break;
858 	}
859 }
860 
861 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
862 		struct request *req, struct nvme_command *cmnd)
863 {
864 	memset(cmnd, 0, sizeof(*cmnd));
865 
866 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
867 		return nvme_setup_discard(ns, req, cmnd);
868 
869 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
870 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
871 	cmnd->write_zeroes.slba =
872 		cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
873 	cmnd->write_zeroes.length =
874 		cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
875 
876 	if (!(req->cmd_flags & REQ_NOUNMAP) && (ns->features & NVME_NS_DEAC))
877 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_WZ_DEAC);
878 
879 	if (nvme_ns_has_pi(ns)) {
880 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_RW_PRINFO_PRACT);
881 
882 		switch (ns->pi_type) {
883 		case NVME_NS_DPS_PI_TYPE1:
884 		case NVME_NS_DPS_PI_TYPE2:
885 			nvme_set_ref_tag(ns, cmnd, req);
886 			break;
887 		}
888 	}
889 
890 	return BLK_STS_OK;
891 }
892 
893 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
894 		struct request *req, struct nvme_command *cmnd,
895 		enum nvme_opcode op)
896 {
897 	u16 control = 0;
898 	u32 dsmgmt = 0;
899 
900 	if (req->cmd_flags & REQ_FUA)
901 		control |= NVME_RW_FUA;
902 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
903 		control |= NVME_RW_LR;
904 
905 	if (req->cmd_flags & REQ_RAHEAD)
906 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
907 
908 	cmnd->rw.opcode = op;
909 	cmnd->rw.flags = 0;
910 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
911 	cmnd->rw.cdw2 = 0;
912 	cmnd->rw.cdw3 = 0;
913 	cmnd->rw.metadata = 0;
914 	cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
915 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
916 	cmnd->rw.reftag = 0;
917 	cmnd->rw.apptag = 0;
918 	cmnd->rw.appmask = 0;
919 
920 	if (ns->ms) {
921 		/*
922 		 * If formated with metadata, the block layer always provides a
923 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
924 		 * we enable the PRACT bit for protection information or set the
925 		 * namespace capacity to zero to prevent any I/O.
926 		 */
927 		if (!blk_integrity_rq(req)) {
928 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
929 				return BLK_STS_NOTSUPP;
930 			control |= NVME_RW_PRINFO_PRACT;
931 		}
932 
933 		switch (ns->pi_type) {
934 		case NVME_NS_DPS_PI_TYPE3:
935 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
936 			break;
937 		case NVME_NS_DPS_PI_TYPE1:
938 		case NVME_NS_DPS_PI_TYPE2:
939 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
940 					NVME_RW_PRINFO_PRCHK_REF;
941 			if (op == nvme_cmd_zone_append)
942 				control |= NVME_RW_APPEND_PIREMAP;
943 			nvme_set_ref_tag(ns, cmnd, req);
944 			break;
945 		}
946 	}
947 
948 	cmnd->rw.control = cpu_to_le16(control);
949 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
950 	return 0;
951 }
952 
953 void nvme_cleanup_cmd(struct request *req)
954 {
955 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
956 		struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
957 
958 		if (req->special_vec.bv_page == ctrl->discard_page)
959 			clear_bit_unlock(0, &ctrl->discard_page_busy);
960 		else
961 			kfree(bvec_virt(&req->special_vec));
962 	}
963 }
964 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
965 
966 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
967 {
968 	struct nvme_command *cmd = nvme_req(req)->cmd;
969 	blk_status_t ret = BLK_STS_OK;
970 
971 	if (!(req->rq_flags & RQF_DONTPREP))
972 		nvme_clear_nvme_request(req);
973 
974 	switch (req_op(req)) {
975 	case REQ_OP_DRV_IN:
976 	case REQ_OP_DRV_OUT:
977 		/* these are setup prior to execution in nvme_init_request() */
978 		break;
979 	case REQ_OP_FLUSH:
980 		nvme_setup_flush(ns, cmd);
981 		break;
982 	case REQ_OP_ZONE_RESET_ALL:
983 	case REQ_OP_ZONE_RESET:
984 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
985 		break;
986 	case REQ_OP_ZONE_OPEN:
987 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
988 		break;
989 	case REQ_OP_ZONE_CLOSE:
990 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
991 		break;
992 	case REQ_OP_ZONE_FINISH:
993 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
994 		break;
995 	case REQ_OP_WRITE_ZEROES:
996 		ret = nvme_setup_write_zeroes(ns, req, cmd);
997 		break;
998 	case REQ_OP_DISCARD:
999 		ret = nvme_setup_discard(ns, req, cmd);
1000 		break;
1001 	case REQ_OP_READ:
1002 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1003 		break;
1004 	case REQ_OP_WRITE:
1005 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1006 		break;
1007 	case REQ_OP_ZONE_APPEND:
1008 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1009 		break;
1010 	default:
1011 		WARN_ON_ONCE(1);
1012 		return BLK_STS_IOERR;
1013 	}
1014 
1015 	cmd->common.command_id = nvme_cid(req);
1016 	trace_nvme_setup_cmd(req, cmd);
1017 	return ret;
1018 }
1019 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1020 
1021 /*
1022  * Return values:
1023  * 0:  success
1024  * >0: nvme controller's cqe status response
1025  * <0: kernel error in lieu of controller response
1026  */
1027 int nvme_execute_rq(struct request *rq, bool at_head)
1028 {
1029 	blk_status_t status;
1030 
1031 	status = blk_execute_rq(rq, at_head);
1032 	if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1033 		return -EINTR;
1034 	if (nvme_req(rq)->status)
1035 		return nvme_req(rq)->status;
1036 	return blk_status_to_errno(status);
1037 }
1038 EXPORT_SYMBOL_NS_GPL(nvme_execute_rq, NVME_TARGET_PASSTHRU);
1039 
1040 /*
1041  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1042  * if the result is positive, it's an NVM Express status code
1043  */
1044 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1045 		union nvme_result *result, void *buffer, unsigned bufflen,
1046 		int qid, int at_head, blk_mq_req_flags_t flags)
1047 {
1048 	struct request *req;
1049 	int ret;
1050 
1051 	if (qid == NVME_QID_ANY)
1052 		req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1053 	else
1054 		req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1055 						qid - 1);
1056 
1057 	if (IS_ERR(req))
1058 		return PTR_ERR(req);
1059 	nvme_init_request(req, cmd);
1060 
1061 	if (buffer && bufflen) {
1062 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1063 		if (ret)
1064 			goto out;
1065 	}
1066 
1067 	ret = nvme_execute_rq(req, at_head);
1068 	if (result && ret >= 0)
1069 		*result = nvme_req(req)->result;
1070  out:
1071 	blk_mq_free_request(req);
1072 	return ret;
1073 }
1074 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1075 
1076 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1077 		void *buffer, unsigned bufflen)
1078 {
1079 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1080 			NVME_QID_ANY, 0, 0);
1081 }
1082 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1083 
1084 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1085 {
1086 	u32 effects = 0;
1087 
1088 	if (ns) {
1089 		effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1090 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1091 			dev_warn_once(ctrl->device,
1092 				"IO command:%02x has unusual effects:%08x\n",
1093 				opcode, effects);
1094 
1095 		/*
1096 		 * NVME_CMD_EFFECTS_CSE_MASK causes a freeze all I/O queues,
1097 		 * which would deadlock when done on an I/O command.  Note that
1098 		 * We already warn about an unusual effect above.
1099 		 */
1100 		effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1101 	} else {
1102 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1103 	}
1104 
1105 	return effects;
1106 }
1107 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1108 
1109 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1110 {
1111 	u32 effects = nvme_command_effects(ctrl, ns, opcode);
1112 
1113 	/*
1114 	 * For simplicity, IO to all namespaces is quiesced even if the command
1115 	 * effects say only one namespace is affected.
1116 	 */
1117 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1118 		mutex_lock(&ctrl->scan_lock);
1119 		mutex_lock(&ctrl->subsys->lock);
1120 		nvme_mpath_start_freeze(ctrl->subsys);
1121 		nvme_mpath_wait_freeze(ctrl->subsys);
1122 		nvme_start_freeze(ctrl);
1123 		nvme_wait_freeze(ctrl);
1124 	}
1125 	return effects;
1126 }
1127 EXPORT_SYMBOL_NS_GPL(nvme_passthru_start, NVME_TARGET_PASSTHRU);
1128 
1129 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1130 		       struct nvme_command *cmd, int status)
1131 {
1132 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1133 		nvme_unfreeze(ctrl);
1134 		nvme_mpath_unfreeze(ctrl->subsys);
1135 		mutex_unlock(&ctrl->subsys->lock);
1136 		mutex_unlock(&ctrl->scan_lock);
1137 	}
1138 	if (effects & NVME_CMD_EFFECTS_CCC) {
1139 		if (!test_and_set_bit(NVME_CTRL_DIRTY_CAPABILITY,
1140 				      &ctrl->flags)) {
1141 			dev_info(ctrl->device,
1142 "controller capabilities changed, reset may be required to take effect.\n");
1143 		}
1144 	}
1145 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1146 		nvme_queue_scan(ctrl);
1147 		flush_work(&ctrl->scan_work);
1148 	}
1149 	if (ns)
1150 		return;
1151 
1152 	switch (cmd->common.opcode) {
1153 	case nvme_admin_set_features:
1154 		switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1155 		case NVME_FEAT_KATO:
1156 			/*
1157 			 * Keep alive commands interval on the host should be
1158 			 * updated when KATO is modified by Set Features
1159 			 * commands.
1160 			 */
1161 			if (!status)
1162 				nvme_update_keep_alive(ctrl, cmd);
1163 			break;
1164 		default:
1165 			break;
1166 		}
1167 		break;
1168 	default:
1169 		break;
1170 	}
1171 }
1172 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1173 
1174 /*
1175  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1176  *
1177  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1178  *   accounting for transport roundtrip times [..].
1179  */
1180 static unsigned long nvme_keep_alive_work_period(struct nvme_ctrl *ctrl)
1181 {
1182 	unsigned long delay = ctrl->kato * HZ / 2;
1183 
1184 	/*
1185 	 * When using Traffic Based Keep Alive, we need to run
1186 	 * nvme_keep_alive_work at twice the normal frequency, as one
1187 	 * command completion can postpone sending a keep alive command
1188 	 * by up to twice the delay between runs.
1189 	 */
1190 	if (ctrl->ctratt & NVME_CTRL_ATTR_TBKAS)
1191 		delay /= 2;
1192 	return delay;
1193 }
1194 
1195 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1196 {
1197 	unsigned long now = jiffies;
1198 	unsigned long delay = nvme_keep_alive_work_period(ctrl);
1199 	unsigned long ka_next_check_tm = ctrl->ka_last_check_time + delay;
1200 
1201 	if (time_after(now, ka_next_check_tm))
1202 		delay = 0;
1203 	else
1204 		delay = ka_next_check_tm - now;
1205 
1206 	queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1207 }
1208 
1209 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1210 						 blk_status_t status)
1211 {
1212 	struct nvme_ctrl *ctrl = rq->end_io_data;
1213 	unsigned long flags;
1214 	bool startka = false;
1215 	unsigned long rtt = jiffies - (rq->deadline - rq->timeout);
1216 	unsigned long delay = nvme_keep_alive_work_period(ctrl);
1217 
1218 	/*
1219 	 * Subtract off the keepalive RTT so nvme_keep_alive_work runs
1220 	 * at the desired frequency.
1221 	 */
1222 	if (rtt <= delay) {
1223 		delay -= rtt;
1224 	} else {
1225 		dev_warn(ctrl->device, "long keepalive RTT (%u ms)\n",
1226 			 jiffies_to_msecs(rtt));
1227 		delay = 0;
1228 	}
1229 
1230 	blk_mq_free_request(rq);
1231 
1232 	if (status) {
1233 		dev_err(ctrl->device,
1234 			"failed nvme_keep_alive_end_io error=%d\n",
1235 				status);
1236 		return RQ_END_IO_NONE;
1237 	}
1238 
1239 	ctrl->ka_last_check_time = jiffies;
1240 	ctrl->comp_seen = false;
1241 	spin_lock_irqsave(&ctrl->lock, flags);
1242 	if (ctrl->state == NVME_CTRL_LIVE ||
1243 	    ctrl->state == NVME_CTRL_CONNECTING)
1244 		startka = true;
1245 	spin_unlock_irqrestore(&ctrl->lock, flags);
1246 	if (startka)
1247 		queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1248 	return RQ_END_IO_NONE;
1249 }
1250 
1251 static void nvme_keep_alive_work(struct work_struct *work)
1252 {
1253 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1254 			struct nvme_ctrl, ka_work);
1255 	bool comp_seen = ctrl->comp_seen;
1256 	struct request *rq;
1257 
1258 	ctrl->ka_last_check_time = jiffies;
1259 
1260 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1261 		dev_dbg(ctrl->device,
1262 			"reschedule traffic based keep-alive timer\n");
1263 		ctrl->comp_seen = false;
1264 		nvme_queue_keep_alive_work(ctrl);
1265 		return;
1266 	}
1267 
1268 	rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1269 				  BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1270 	if (IS_ERR(rq)) {
1271 		/* allocation failure, reset the controller */
1272 		dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1273 		nvme_reset_ctrl(ctrl);
1274 		return;
1275 	}
1276 	nvme_init_request(rq, &ctrl->ka_cmd);
1277 
1278 	rq->timeout = ctrl->kato * HZ;
1279 	rq->end_io = nvme_keep_alive_end_io;
1280 	rq->end_io_data = ctrl;
1281 	blk_execute_rq_nowait(rq, false);
1282 }
1283 
1284 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1285 {
1286 	if (unlikely(ctrl->kato == 0))
1287 		return;
1288 
1289 	nvme_queue_keep_alive_work(ctrl);
1290 }
1291 
1292 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1293 {
1294 	if (unlikely(ctrl->kato == 0))
1295 		return;
1296 
1297 	cancel_delayed_work_sync(&ctrl->ka_work);
1298 }
1299 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1300 
1301 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1302 				   struct nvme_command *cmd)
1303 {
1304 	unsigned int new_kato =
1305 		DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1306 
1307 	dev_info(ctrl->device,
1308 		 "keep alive interval updated from %u ms to %u ms\n",
1309 		 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1310 
1311 	nvme_stop_keep_alive(ctrl);
1312 	ctrl->kato = new_kato;
1313 	nvme_start_keep_alive(ctrl);
1314 }
1315 
1316 /*
1317  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1318  * flag, thus sending any new CNS opcodes has a big chance of not working.
1319  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1320  * (but not for any later version).
1321  */
1322 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1323 {
1324 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1325 		return ctrl->vs < NVME_VS(1, 2, 0);
1326 	return ctrl->vs < NVME_VS(1, 1, 0);
1327 }
1328 
1329 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1330 {
1331 	struct nvme_command c = { };
1332 	int error;
1333 
1334 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1335 	c.identify.opcode = nvme_admin_identify;
1336 	c.identify.cns = NVME_ID_CNS_CTRL;
1337 
1338 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1339 	if (!*id)
1340 		return -ENOMEM;
1341 
1342 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1343 			sizeof(struct nvme_id_ctrl));
1344 	if (error)
1345 		kfree(*id);
1346 	return error;
1347 }
1348 
1349 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1350 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1351 {
1352 	const char *warn_str = "ctrl returned bogus length:";
1353 	void *data = cur;
1354 
1355 	switch (cur->nidt) {
1356 	case NVME_NIDT_EUI64:
1357 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1358 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1359 				 warn_str, cur->nidl);
1360 			return -1;
1361 		}
1362 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1363 			return NVME_NIDT_EUI64_LEN;
1364 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1365 		return NVME_NIDT_EUI64_LEN;
1366 	case NVME_NIDT_NGUID:
1367 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1368 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1369 				 warn_str, cur->nidl);
1370 			return -1;
1371 		}
1372 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1373 			return NVME_NIDT_NGUID_LEN;
1374 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1375 		return NVME_NIDT_NGUID_LEN;
1376 	case NVME_NIDT_UUID:
1377 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1378 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1379 				 warn_str, cur->nidl);
1380 			return -1;
1381 		}
1382 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1383 			return NVME_NIDT_UUID_LEN;
1384 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1385 		return NVME_NIDT_UUID_LEN;
1386 	case NVME_NIDT_CSI:
1387 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1388 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1389 				 warn_str, cur->nidl);
1390 			return -1;
1391 		}
1392 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1393 		*csi_seen = true;
1394 		return NVME_NIDT_CSI_LEN;
1395 	default:
1396 		/* Skip unknown types */
1397 		return cur->nidl;
1398 	}
1399 }
1400 
1401 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1402 		struct nvme_ns_info *info)
1403 {
1404 	struct nvme_command c = { };
1405 	bool csi_seen = false;
1406 	int status, pos, len;
1407 	void *data;
1408 
1409 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1410 		return 0;
1411 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1412 		return 0;
1413 
1414 	c.identify.opcode = nvme_admin_identify;
1415 	c.identify.nsid = cpu_to_le32(info->nsid);
1416 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1417 
1418 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1419 	if (!data)
1420 		return -ENOMEM;
1421 
1422 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1423 				      NVME_IDENTIFY_DATA_SIZE);
1424 	if (status) {
1425 		dev_warn(ctrl->device,
1426 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1427 			info->nsid, status);
1428 		goto free_data;
1429 	}
1430 
1431 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1432 		struct nvme_ns_id_desc *cur = data + pos;
1433 
1434 		if (cur->nidl == 0)
1435 			break;
1436 
1437 		len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1438 		if (len < 0)
1439 			break;
1440 
1441 		len += sizeof(*cur);
1442 	}
1443 
1444 	if (nvme_multi_css(ctrl) && !csi_seen) {
1445 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1446 			 info->nsid);
1447 		status = -EINVAL;
1448 	}
1449 
1450 free_data:
1451 	kfree(data);
1452 	return status;
1453 }
1454 
1455 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1456 			struct nvme_id_ns **id)
1457 {
1458 	struct nvme_command c = { };
1459 	int error;
1460 
1461 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1462 	c.identify.opcode = nvme_admin_identify;
1463 	c.identify.nsid = cpu_to_le32(nsid);
1464 	c.identify.cns = NVME_ID_CNS_NS;
1465 
1466 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1467 	if (!*id)
1468 		return -ENOMEM;
1469 
1470 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1471 	if (error) {
1472 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1473 		kfree(*id);
1474 	}
1475 	return error;
1476 }
1477 
1478 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1479 		struct nvme_ns_info *info)
1480 {
1481 	struct nvme_ns_ids *ids = &info->ids;
1482 	struct nvme_id_ns *id;
1483 	int ret;
1484 
1485 	ret = nvme_identify_ns(ctrl, info->nsid, &id);
1486 	if (ret)
1487 		return ret;
1488 
1489 	if (id->ncap == 0) {
1490 		/* namespace not allocated or attached */
1491 		info->is_removed = true;
1492 		ret = -ENODEV;
1493 		goto error;
1494 	}
1495 
1496 	info->anagrpid = id->anagrpid;
1497 	info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1498 	info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1499 	info->is_ready = true;
1500 	if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1501 		dev_info(ctrl->device,
1502 			 "Ignoring bogus Namespace Identifiers\n");
1503 	} else {
1504 		if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1505 		    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1506 			memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1507 		if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1508 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1509 			memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1510 	}
1511 
1512 error:
1513 	kfree(id);
1514 	return ret;
1515 }
1516 
1517 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1518 		struct nvme_ns_info *info)
1519 {
1520 	struct nvme_id_ns_cs_indep *id;
1521 	struct nvme_command c = {
1522 		.identify.opcode	= nvme_admin_identify,
1523 		.identify.nsid		= cpu_to_le32(info->nsid),
1524 		.identify.cns		= NVME_ID_CNS_NS_CS_INDEP,
1525 	};
1526 	int ret;
1527 
1528 	id = kmalloc(sizeof(*id), GFP_KERNEL);
1529 	if (!id)
1530 		return -ENOMEM;
1531 
1532 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1533 	if (!ret) {
1534 		info->anagrpid = id->anagrpid;
1535 		info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1536 		info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1537 		info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1538 	}
1539 	kfree(id);
1540 	return ret;
1541 }
1542 
1543 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1544 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1545 {
1546 	union nvme_result res = { 0 };
1547 	struct nvme_command c = { };
1548 	int ret;
1549 
1550 	c.features.opcode = op;
1551 	c.features.fid = cpu_to_le32(fid);
1552 	c.features.dword11 = cpu_to_le32(dword11);
1553 
1554 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1555 			buffer, buflen, NVME_QID_ANY, 0, 0);
1556 	if (ret >= 0 && result)
1557 		*result = le32_to_cpu(res.u32);
1558 	return ret;
1559 }
1560 
1561 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1562 		      unsigned int dword11, void *buffer, size_t buflen,
1563 		      u32 *result)
1564 {
1565 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1566 			     buflen, result);
1567 }
1568 EXPORT_SYMBOL_GPL(nvme_set_features);
1569 
1570 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1571 		      unsigned int dword11, void *buffer, size_t buflen,
1572 		      u32 *result)
1573 {
1574 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1575 			     buflen, result);
1576 }
1577 EXPORT_SYMBOL_GPL(nvme_get_features);
1578 
1579 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1580 {
1581 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1582 	u32 result;
1583 	int status, nr_io_queues;
1584 
1585 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1586 			&result);
1587 	if (status < 0)
1588 		return status;
1589 
1590 	/*
1591 	 * Degraded controllers might return an error when setting the queue
1592 	 * count.  We still want to be able to bring them online and offer
1593 	 * access to the admin queue, as that might be only way to fix them up.
1594 	 */
1595 	if (status > 0) {
1596 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1597 		*count = 0;
1598 	} else {
1599 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1600 		*count = min(*count, nr_io_queues);
1601 	}
1602 
1603 	return 0;
1604 }
1605 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1606 
1607 #define NVME_AEN_SUPPORTED \
1608 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1609 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1610 
1611 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1612 {
1613 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1614 	int status;
1615 
1616 	if (!supported_aens)
1617 		return;
1618 
1619 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1620 			NULL, 0, &result);
1621 	if (status)
1622 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1623 			 supported_aens);
1624 
1625 	queue_work(nvme_wq, &ctrl->async_event_work);
1626 }
1627 
1628 static int nvme_ns_open(struct nvme_ns *ns)
1629 {
1630 
1631 	/* should never be called due to GENHD_FL_HIDDEN */
1632 	if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1633 		goto fail;
1634 	if (!nvme_get_ns(ns))
1635 		goto fail;
1636 	if (!try_module_get(ns->ctrl->ops->module))
1637 		goto fail_put_ns;
1638 
1639 	return 0;
1640 
1641 fail_put_ns:
1642 	nvme_put_ns(ns);
1643 fail:
1644 	return -ENXIO;
1645 }
1646 
1647 static void nvme_ns_release(struct nvme_ns *ns)
1648 {
1649 
1650 	module_put(ns->ctrl->ops->module);
1651 	nvme_put_ns(ns);
1652 }
1653 
1654 static int nvme_open(struct gendisk *disk, blk_mode_t mode)
1655 {
1656 	return nvme_ns_open(disk->private_data);
1657 }
1658 
1659 static void nvme_release(struct gendisk *disk)
1660 {
1661 	nvme_ns_release(disk->private_data);
1662 }
1663 
1664 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1665 {
1666 	/* some standard values */
1667 	geo->heads = 1 << 6;
1668 	geo->sectors = 1 << 5;
1669 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1670 	return 0;
1671 }
1672 
1673 #ifdef CONFIG_BLK_DEV_INTEGRITY
1674 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1675 				u32 max_integrity_segments)
1676 {
1677 	struct blk_integrity integrity = { };
1678 
1679 	switch (ns->pi_type) {
1680 	case NVME_NS_DPS_PI_TYPE3:
1681 		switch (ns->guard_type) {
1682 		case NVME_NVM_NS_16B_GUARD:
1683 			integrity.profile = &t10_pi_type3_crc;
1684 			integrity.tag_size = sizeof(u16) + sizeof(u32);
1685 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1686 			break;
1687 		case NVME_NVM_NS_64B_GUARD:
1688 			integrity.profile = &ext_pi_type3_crc64;
1689 			integrity.tag_size = sizeof(u16) + 6;
1690 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1691 			break;
1692 		default:
1693 			integrity.profile = NULL;
1694 			break;
1695 		}
1696 		break;
1697 	case NVME_NS_DPS_PI_TYPE1:
1698 	case NVME_NS_DPS_PI_TYPE2:
1699 		switch (ns->guard_type) {
1700 		case NVME_NVM_NS_16B_GUARD:
1701 			integrity.profile = &t10_pi_type1_crc;
1702 			integrity.tag_size = sizeof(u16);
1703 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1704 			break;
1705 		case NVME_NVM_NS_64B_GUARD:
1706 			integrity.profile = &ext_pi_type1_crc64;
1707 			integrity.tag_size = sizeof(u16);
1708 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1709 			break;
1710 		default:
1711 			integrity.profile = NULL;
1712 			break;
1713 		}
1714 		break;
1715 	default:
1716 		integrity.profile = NULL;
1717 		break;
1718 	}
1719 
1720 	integrity.tuple_size = ns->ms;
1721 	blk_integrity_register(disk, &integrity);
1722 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1723 }
1724 #else
1725 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1726 				u32 max_integrity_segments)
1727 {
1728 }
1729 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1730 
1731 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1732 {
1733 	struct nvme_ctrl *ctrl = ns->ctrl;
1734 	struct request_queue *queue = disk->queue;
1735 	u32 size = queue_logical_block_size(queue);
1736 
1737 	if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1738 		ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1739 
1740 	if (ctrl->max_discard_sectors == 0) {
1741 		blk_queue_max_discard_sectors(queue, 0);
1742 		return;
1743 	}
1744 
1745 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1746 			NVME_DSM_MAX_RANGES);
1747 
1748 	queue->limits.discard_granularity = size;
1749 
1750 	/* If discard is already enabled, don't reset queue limits */
1751 	if (queue->limits.max_discard_sectors)
1752 		return;
1753 
1754 	blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1755 	blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1756 
1757 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1758 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1759 }
1760 
1761 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1762 {
1763 	return uuid_equal(&a->uuid, &b->uuid) &&
1764 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1765 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1766 		a->csi == b->csi;
1767 }
1768 
1769 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1770 {
1771 	bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1772 	unsigned lbaf = nvme_lbaf_index(id->flbas);
1773 	struct nvme_ctrl *ctrl = ns->ctrl;
1774 	struct nvme_command c = { };
1775 	struct nvme_id_ns_nvm *nvm;
1776 	int ret = 0;
1777 	u32 elbaf;
1778 
1779 	ns->pi_size = 0;
1780 	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1781 	if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1782 		ns->pi_size = sizeof(struct t10_pi_tuple);
1783 		ns->guard_type = NVME_NVM_NS_16B_GUARD;
1784 		goto set_pi;
1785 	}
1786 
1787 	nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1788 	if (!nvm)
1789 		return -ENOMEM;
1790 
1791 	c.identify.opcode = nvme_admin_identify;
1792 	c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1793 	c.identify.cns = NVME_ID_CNS_CS_NS;
1794 	c.identify.csi = NVME_CSI_NVM;
1795 
1796 	ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1797 	if (ret)
1798 		goto free_data;
1799 
1800 	elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1801 
1802 	/* no support for storage tag formats right now */
1803 	if (nvme_elbaf_sts(elbaf))
1804 		goto free_data;
1805 
1806 	ns->guard_type = nvme_elbaf_guard_type(elbaf);
1807 	switch (ns->guard_type) {
1808 	case NVME_NVM_NS_64B_GUARD:
1809 		ns->pi_size = sizeof(struct crc64_pi_tuple);
1810 		break;
1811 	case NVME_NVM_NS_16B_GUARD:
1812 		ns->pi_size = sizeof(struct t10_pi_tuple);
1813 		break;
1814 	default:
1815 		break;
1816 	}
1817 
1818 free_data:
1819 	kfree(nvm);
1820 set_pi:
1821 	if (ns->pi_size && (first || ns->ms == ns->pi_size))
1822 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1823 	else
1824 		ns->pi_type = 0;
1825 
1826 	return ret;
1827 }
1828 
1829 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1830 {
1831 	struct nvme_ctrl *ctrl = ns->ctrl;
1832 	int ret;
1833 
1834 	ret = nvme_init_ms(ns, id);
1835 	if (ret)
1836 		return ret;
1837 
1838 	ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1839 	if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1840 		return 0;
1841 
1842 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1843 		/*
1844 		 * The NVMe over Fabrics specification only supports metadata as
1845 		 * part of the extended data LBA.  We rely on HCA/HBA support to
1846 		 * remap the separate metadata buffer from the block layer.
1847 		 */
1848 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1849 			return 0;
1850 
1851 		ns->features |= NVME_NS_EXT_LBAS;
1852 
1853 		/*
1854 		 * The current fabrics transport drivers support namespace
1855 		 * metadata formats only if nvme_ns_has_pi() returns true.
1856 		 * Suppress support for all other formats so the namespace will
1857 		 * have a 0 capacity and not be usable through the block stack.
1858 		 *
1859 		 * Note, this check will need to be modified if any drivers
1860 		 * gain the ability to use other metadata formats.
1861 		 */
1862 		if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1863 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1864 	} else {
1865 		/*
1866 		 * For PCIe controllers, we can't easily remap the separate
1867 		 * metadata buffer from the block layer and thus require a
1868 		 * separate metadata buffer for block layer metadata/PI support.
1869 		 * We allow extended LBAs for the passthrough interface, though.
1870 		 */
1871 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
1872 			ns->features |= NVME_NS_EXT_LBAS;
1873 		else
1874 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1875 	}
1876 	return 0;
1877 }
1878 
1879 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1880 		struct request_queue *q)
1881 {
1882 	bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1883 
1884 	if (ctrl->max_hw_sectors) {
1885 		u32 max_segments =
1886 			(ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1887 
1888 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
1889 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1890 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1891 	}
1892 	blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1893 	blk_queue_dma_alignment(q, 3);
1894 	blk_queue_write_cache(q, vwc, vwc);
1895 }
1896 
1897 static void nvme_update_disk_info(struct gendisk *disk,
1898 		struct nvme_ns *ns, struct nvme_id_ns *id)
1899 {
1900 	sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1901 	u32 bs = 1U << ns->lba_shift;
1902 	u32 atomic_bs, phys_bs, io_opt = 0;
1903 
1904 	/*
1905 	 * The block layer can't support LBA sizes larger than the page size
1906 	 * or smaller than a sector size yet, so catch this early and don't
1907 	 * allow block I/O.
1908 	 */
1909 	if (ns->lba_shift > PAGE_SHIFT || ns->lba_shift < SECTOR_SHIFT) {
1910 		capacity = 0;
1911 		bs = (1 << 9);
1912 	}
1913 
1914 	blk_integrity_unregister(disk);
1915 
1916 	atomic_bs = phys_bs = bs;
1917 	if (id->nabo == 0) {
1918 		/*
1919 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1920 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1921 		 * 0 then AWUPF must be used instead.
1922 		 */
1923 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1924 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1925 		else
1926 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1927 	}
1928 
1929 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1930 		/* NPWG = Namespace Preferred Write Granularity */
1931 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1932 		/* NOWS = Namespace Optimal Write Size */
1933 		io_opt = bs * (1 + le16_to_cpu(id->nows));
1934 	}
1935 
1936 	blk_queue_logical_block_size(disk->queue, bs);
1937 	/*
1938 	 * Linux filesystems assume writing a single physical block is
1939 	 * an atomic operation. Hence limit the physical block size to the
1940 	 * value of the Atomic Write Unit Power Fail parameter.
1941 	 */
1942 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1943 	blk_queue_io_min(disk->queue, phys_bs);
1944 	blk_queue_io_opt(disk->queue, io_opt);
1945 
1946 	/*
1947 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
1948 	 * metadata masquerading as Type 0 if supported, otherwise reject block
1949 	 * I/O to namespaces with metadata except when the namespace supports
1950 	 * PI, as it can strip/insert in that case.
1951 	 */
1952 	if (ns->ms) {
1953 		if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1954 		    (ns->features & NVME_NS_METADATA_SUPPORTED))
1955 			nvme_init_integrity(disk, ns,
1956 					    ns->ctrl->max_integrity_segments);
1957 		else if (!nvme_ns_has_pi(ns))
1958 			capacity = 0;
1959 	}
1960 
1961 	set_capacity_and_notify(disk, capacity);
1962 
1963 	nvme_config_discard(disk, ns);
1964 	blk_queue_max_write_zeroes_sectors(disk->queue,
1965 					   ns->ctrl->max_zeroes_sectors);
1966 }
1967 
1968 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
1969 {
1970 	return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
1971 }
1972 
1973 static inline bool nvme_first_scan(struct gendisk *disk)
1974 {
1975 	/* nvme_alloc_ns() scans the disk prior to adding it */
1976 	return !disk_live(disk);
1977 }
1978 
1979 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1980 {
1981 	struct nvme_ctrl *ctrl = ns->ctrl;
1982 	u32 iob;
1983 
1984 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1985 	    is_power_of_2(ctrl->max_hw_sectors))
1986 		iob = ctrl->max_hw_sectors;
1987 	else
1988 		iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1989 
1990 	if (!iob)
1991 		return;
1992 
1993 	if (!is_power_of_2(iob)) {
1994 		if (nvme_first_scan(ns->disk))
1995 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1996 				ns->disk->disk_name, iob);
1997 		return;
1998 	}
1999 
2000 	if (blk_queue_is_zoned(ns->disk->queue)) {
2001 		if (nvme_first_scan(ns->disk))
2002 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
2003 				ns->disk->disk_name);
2004 		return;
2005 	}
2006 
2007 	blk_queue_chunk_sectors(ns->queue, iob);
2008 }
2009 
2010 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
2011 		struct nvme_ns_info *info)
2012 {
2013 	blk_mq_freeze_queue(ns->disk->queue);
2014 	nvme_set_queue_limits(ns->ctrl, ns->queue);
2015 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2016 	blk_mq_unfreeze_queue(ns->disk->queue);
2017 
2018 	if (nvme_ns_head_multipath(ns->head)) {
2019 		blk_mq_freeze_queue(ns->head->disk->queue);
2020 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2021 		nvme_mpath_revalidate_paths(ns);
2022 		blk_stack_limits(&ns->head->disk->queue->limits,
2023 				 &ns->queue->limits, 0);
2024 		ns->head->disk->flags |= GENHD_FL_HIDDEN;
2025 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2026 	}
2027 
2028 	/* Hide the block-interface for these devices */
2029 	ns->disk->flags |= GENHD_FL_HIDDEN;
2030 	set_bit(NVME_NS_READY, &ns->flags);
2031 
2032 	return 0;
2033 }
2034 
2035 static int nvme_update_ns_info_block(struct nvme_ns *ns,
2036 		struct nvme_ns_info *info)
2037 {
2038 	struct nvme_id_ns *id;
2039 	unsigned lbaf;
2040 	int ret;
2041 
2042 	ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
2043 	if (ret)
2044 		return ret;
2045 
2046 	if (id->ncap == 0) {
2047 		/* namespace not allocated or attached */
2048 		info->is_removed = true;
2049 		ret = -ENODEV;
2050 		goto error;
2051 	}
2052 
2053 	blk_mq_freeze_queue(ns->disk->queue);
2054 	lbaf = nvme_lbaf_index(id->flbas);
2055 	ns->lba_shift = id->lbaf[lbaf].ds;
2056 	nvme_set_queue_limits(ns->ctrl, ns->queue);
2057 
2058 	ret = nvme_configure_metadata(ns, id);
2059 	if (ret < 0) {
2060 		blk_mq_unfreeze_queue(ns->disk->queue);
2061 		goto out;
2062 	}
2063 	nvme_set_chunk_sectors(ns, id);
2064 	nvme_update_disk_info(ns->disk, ns, id);
2065 
2066 	if (ns->head->ids.csi == NVME_CSI_ZNS) {
2067 		ret = nvme_update_zone_info(ns, lbaf);
2068 		if (ret) {
2069 			blk_mq_unfreeze_queue(ns->disk->queue);
2070 			goto out;
2071 		}
2072 	}
2073 
2074 	/*
2075 	 * Only set the DEAC bit if the device guarantees that reads from
2076 	 * deallocated data return zeroes.  While the DEAC bit does not
2077 	 * require that, it must be a no-op if reads from deallocated data
2078 	 * do not return zeroes.
2079 	 */
2080 	if ((id->dlfeat & 0x7) == 0x1 && (id->dlfeat & (1 << 3)))
2081 		ns->features |= NVME_NS_DEAC;
2082 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2083 	set_bit(NVME_NS_READY, &ns->flags);
2084 	blk_mq_unfreeze_queue(ns->disk->queue);
2085 
2086 	if (blk_queue_is_zoned(ns->queue)) {
2087 		ret = nvme_revalidate_zones(ns);
2088 		if (ret && !nvme_first_scan(ns->disk))
2089 			goto out;
2090 	}
2091 
2092 	if (nvme_ns_head_multipath(ns->head)) {
2093 		blk_mq_freeze_queue(ns->head->disk->queue);
2094 		nvme_update_disk_info(ns->head->disk, ns, id);
2095 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2096 		nvme_mpath_revalidate_paths(ns);
2097 		blk_stack_limits(&ns->head->disk->queue->limits,
2098 				 &ns->queue->limits, 0);
2099 		disk_update_readahead(ns->head->disk);
2100 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2101 	}
2102 
2103 	ret = 0;
2104 out:
2105 	/*
2106 	 * If probing fails due an unsupported feature, hide the block device,
2107 	 * but still allow other access.
2108 	 */
2109 	if (ret == -ENODEV) {
2110 		ns->disk->flags |= GENHD_FL_HIDDEN;
2111 		set_bit(NVME_NS_READY, &ns->flags);
2112 		ret = 0;
2113 	}
2114 
2115 error:
2116 	kfree(id);
2117 	return ret;
2118 }
2119 
2120 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2121 {
2122 	switch (info->ids.csi) {
2123 	case NVME_CSI_ZNS:
2124 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2125 			dev_info(ns->ctrl->device,
2126 	"block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2127 				info->nsid);
2128 			return nvme_update_ns_info_generic(ns, info);
2129 		}
2130 		return nvme_update_ns_info_block(ns, info);
2131 	case NVME_CSI_NVM:
2132 		return nvme_update_ns_info_block(ns, info);
2133 	default:
2134 		dev_info(ns->ctrl->device,
2135 			"block device for nsid %u not supported (csi %u)\n",
2136 			info->nsid, info->ids.csi);
2137 		return nvme_update_ns_info_generic(ns, info);
2138 	}
2139 }
2140 
2141 #ifdef CONFIG_BLK_SED_OPAL
2142 static int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2143 		bool send)
2144 {
2145 	struct nvme_ctrl *ctrl = data;
2146 	struct nvme_command cmd = { };
2147 
2148 	if (send)
2149 		cmd.common.opcode = nvme_admin_security_send;
2150 	else
2151 		cmd.common.opcode = nvme_admin_security_recv;
2152 	cmd.common.nsid = 0;
2153 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2154 	cmd.common.cdw11 = cpu_to_le32(len);
2155 
2156 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2157 			NVME_QID_ANY, 1, 0);
2158 }
2159 
2160 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2161 {
2162 	if (ctrl->oacs & NVME_CTRL_OACS_SEC_SUPP) {
2163 		if (!ctrl->opal_dev)
2164 			ctrl->opal_dev = init_opal_dev(ctrl, &nvme_sec_submit);
2165 		else if (was_suspended)
2166 			opal_unlock_from_suspend(ctrl->opal_dev);
2167 	} else {
2168 		free_opal_dev(ctrl->opal_dev);
2169 		ctrl->opal_dev = NULL;
2170 	}
2171 }
2172 #else
2173 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2174 {
2175 }
2176 #endif /* CONFIG_BLK_SED_OPAL */
2177 
2178 #ifdef CONFIG_BLK_DEV_ZONED
2179 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2180 		unsigned int nr_zones, report_zones_cb cb, void *data)
2181 {
2182 	return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2183 			data);
2184 }
2185 #else
2186 #define nvme_report_zones	NULL
2187 #endif /* CONFIG_BLK_DEV_ZONED */
2188 
2189 const struct block_device_operations nvme_bdev_ops = {
2190 	.owner		= THIS_MODULE,
2191 	.ioctl		= nvme_ioctl,
2192 	.compat_ioctl	= blkdev_compat_ptr_ioctl,
2193 	.open		= nvme_open,
2194 	.release	= nvme_release,
2195 	.getgeo		= nvme_getgeo,
2196 	.report_zones	= nvme_report_zones,
2197 	.pr_ops		= &nvme_pr_ops,
2198 };
2199 
2200 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 mask, u32 val,
2201 		u32 timeout, const char *op)
2202 {
2203 	unsigned long timeout_jiffies = jiffies + timeout * HZ;
2204 	u32 csts;
2205 	int ret;
2206 
2207 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2208 		if (csts == ~0)
2209 			return -ENODEV;
2210 		if ((csts & mask) == val)
2211 			break;
2212 
2213 		usleep_range(1000, 2000);
2214 		if (fatal_signal_pending(current))
2215 			return -EINTR;
2216 		if (time_after(jiffies, timeout_jiffies)) {
2217 			dev_err(ctrl->device,
2218 				"Device not ready; aborting %s, CSTS=0x%x\n",
2219 				op, csts);
2220 			return -ENODEV;
2221 		}
2222 	}
2223 
2224 	return ret;
2225 }
2226 
2227 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown)
2228 {
2229 	int ret;
2230 
2231 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2232 	if (shutdown)
2233 		ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2234 	else
2235 		ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2236 
2237 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2238 	if (ret)
2239 		return ret;
2240 
2241 	if (shutdown) {
2242 		return nvme_wait_ready(ctrl, NVME_CSTS_SHST_MASK,
2243 				       NVME_CSTS_SHST_CMPLT,
2244 				       ctrl->shutdown_timeout, "shutdown");
2245 	}
2246 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2247 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2248 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, 0,
2249 			       (NVME_CAP_TIMEOUT(ctrl->cap) + 1) / 2, "reset");
2250 }
2251 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2252 
2253 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2254 {
2255 	unsigned dev_page_min;
2256 	u32 timeout;
2257 	int ret;
2258 
2259 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2260 	if (ret) {
2261 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2262 		return ret;
2263 	}
2264 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2265 
2266 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2267 		dev_err(ctrl->device,
2268 			"Minimum device page size %u too large for host (%u)\n",
2269 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2270 		return -ENODEV;
2271 	}
2272 
2273 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2274 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2275 	else
2276 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2277 
2278 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS && ctrl->cap & NVME_CAP_CRMS_CRIMS)
2279 		ctrl->ctrl_config |= NVME_CC_CRIME;
2280 
2281 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2282 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2283 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2284 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2285 	if (ret)
2286 		return ret;
2287 
2288 	/* Flush write to device (required if transport is PCI) */
2289 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2290 	if (ret)
2291 		return ret;
2292 
2293 	/* CAP value may change after initial CC write */
2294 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2295 	if (ret)
2296 		return ret;
2297 
2298 	timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2299 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2300 		u32 crto, ready_timeout;
2301 
2302 		ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2303 		if (ret) {
2304 			dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2305 				ret);
2306 			return ret;
2307 		}
2308 
2309 		/*
2310 		 * CRTO should always be greater or equal to CAP.TO, but some
2311 		 * devices are known to get this wrong. Use the larger of the
2312 		 * two values.
2313 		 */
2314 		if (ctrl->ctrl_config & NVME_CC_CRIME)
2315 			ready_timeout = NVME_CRTO_CRIMT(crto);
2316 		else
2317 			ready_timeout = NVME_CRTO_CRWMT(crto);
2318 
2319 		if (ready_timeout < timeout)
2320 			dev_warn_once(ctrl->device, "bad crto:%x cap:%llx\n",
2321 				      crto, ctrl->cap);
2322 		else
2323 			timeout = ready_timeout;
2324 	}
2325 
2326 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2327 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2328 	if (ret)
2329 		return ret;
2330 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, NVME_CSTS_RDY,
2331 			       (timeout + 1) / 2, "initialisation");
2332 }
2333 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2334 
2335 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2336 {
2337 	__le64 ts;
2338 	int ret;
2339 
2340 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2341 		return 0;
2342 
2343 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2344 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2345 			NULL);
2346 	if (ret)
2347 		dev_warn_once(ctrl->device,
2348 			"could not set timestamp (%d)\n", ret);
2349 	return ret;
2350 }
2351 
2352 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2353 {
2354 	struct nvme_feat_host_behavior *host;
2355 	u8 acre = 0, lbafee = 0;
2356 	int ret;
2357 
2358 	/* Don't bother enabling the feature if retry delay is not reported */
2359 	if (ctrl->crdt[0])
2360 		acre = NVME_ENABLE_ACRE;
2361 	if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2362 		lbafee = NVME_ENABLE_LBAFEE;
2363 
2364 	if (!acre && !lbafee)
2365 		return 0;
2366 
2367 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2368 	if (!host)
2369 		return 0;
2370 
2371 	host->acre = acre;
2372 	host->lbafee = lbafee;
2373 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2374 				host, sizeof(*host), NULL);
2375 	kfree(host);
2376 	return ret;
2377 }
2378 
2379 /*
2380  * The function checks whether the given total (exlat + enlat) latency of
2381  * a power state allows the latter to be used as an APST transition target.
2382  * It does so by comparing the latency to the primary and secondary latency
2383  * tolerances defined by module params. If there's a match, the corresponding
2384  * timeout value is returned and the matching tolerance index (1 or 2) is
2385  * reported.
2386  */
2387 static bool nvme_apst_get_transition_time(u64 total_latency,
2388 		u64 *transition_time, unsigned *last_index)
2389 {
2390 	if (total_latency <= apst_primary_latency_tol_us) {
2391 		if (*last_index == 1)
2392 			return false;
2393 		*last_index = 1;
2394 		*transition_time = apst_primary_timeout_ms;
2395 		return true;
2396 	}
2397 	if (apst_secondary_timeout_ms &&
2398 		total_latency <= apst_secondary_latency_tol_us) {
2399 		if (*last_index <= 2)
2400 			return false;
2401 		*last_index = 2;
2402 		*transition_time = apst_secondary_timeout_ms;
2403 		return true;
2404 	}
2405 	return false;
2406 }
2407 
2408 /*
2409  * APST (Autonomous Power State Transition) lets us program a table of power
2410  * state transitions that the controller will perform automatically.
2411  *
2412  * Depending on module params, one of the two supported techniques will be used:
2413  *
2414  * - If the parameters provide explicit timeouts and tolerances, they will be
2415  *   used to build a table with up to 2 non-operational states to transition to.
2416  *   The default parameter values were selected based on the values used by
2417  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2418  *   regeneration of the APST table in the event of switching between external
2419  *   and battery power, the timeouts and tolerances reflect a compromise
2420  *   between values used by Microsoft for AC and battery scenarios.
2421  * - If not, we'll configure the table with a simple heuristic: we are willing
2422  *   to spend at most 2% of the time transitioning between power states.
2423  *   Therefore, when running in any given state, we will enter the next
2424  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2425  *   microseconds, as long as that state's exit latency is under the requested
2426  *   maximum latency.
2427  *
2428  * We will not autonomously enter any non-operational state for which the total
2429  * latency exceeds ps_max_latency_us.
2430  *
2431  * Users can set ps_max_latency_us to zero to turn off APST.
2432  */
2433 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2434 {
2435 	struct nvme_feat_auto_pst *table;
2436 	unsigned apste = 0;
2437 	u64 max_lat_us = 0;
2438 	__le64 target = 0;
2439 	int max_ps = -1;
2440 	int state;
2441 	int ret;
2442 	unsigned last_lt_index = UINT_MAX;
2443 
2444 	/*
2445 	 * If APST isn't supported or if we haven't been initialized yet,
2446 	 * then don't do anything.
2447 	 */
2448 	if (!ctrl->apsta)
2449 		return 0;
2450 
2451 	if (ctrl->npss > 31) {
2452 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2453 		return 0;
2454 	}
2455 
2456 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2457 	if (!table)
2458 		return 0;
2459 
2460 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2461 		/* Turn off APST. */
2462 		dev_dbg(ctrl->device, "APST disabled\n");
2463 		goto done;
2464 	}
2465 
2466 	/*
2467 	 * Walk through all states from lowest- to highest-power.
2468 	 * According to the spec, lower-numbered states use more power.  NPSS,
2469 	 * despite the name, is the index of the lowest-power state, not the
2470 	 * number of states.
2471 	 */
2472 	for (state = (int)ctrl->npss; state >= 0; state--) {
2473 		u64 total_latency_us, exit_latency_us, transition_ms;
2474 
2475 		if (target)
2476 			table->entries[state] = target;
2477 
2478 		/*
2479 		 * Don't allow transitions to the deepest state if it's quirked
2480 		 * off.
2481 		 */
2482 		if (state == ctrl->npss &&
2483 		    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2484 			continue;
2485 
2486 		/*
2487 		 * Is this state a useful non-operational state for higher-power
2488 		 * states to autonomously transition to?
2489 		 */
2490 		if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2491 			continue;
2492 
2493 		exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2494 		if (exit_latency_us > ctrl->ps_max_latency_us)
2495 			continue;
2496 
2497 		total_latency_us = exit_latency_us +
2498 			le32_to_cpu(ctrl->psd[state].entry_lat);
2499 
2500 		/*
2501 		 * This state is good. It can be used as the APST idle target
2502 		 * for higher power states.
2503 		 */
2504 		if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2505 			if (!nvme_apst_get_transition_time(total_latency_us,
2506 					&transition_ms, &last_lt_index))
2507 				continue;
2508 		} else {
2509 			transition_ms = total_latency_us + 19;
2510 			do_div(transition_ms, 20);
2511 			if (transition_ms > (1 << 24) - 1)
2512 				transition_ms = (1 << 24) - 1;
2513 		}
2514 
2515 		target = cpu_to_le64((state << 3) | (transition_ms << 8));
2516 		if (max_ps == -1)
2517 			max_ps = state;
2518 		if (total_latency_us > max_lat_us)
2519 			max_lat_us = total_latency_us;
2520 	}
2521 
2522 	if (max_ps == -1)
2523 		dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2524 	else
2525 		dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2526 			max_ps, max_lat_us, (int)sizeof(*table), table);
2527 	apste = 1;
2528 
2529 done:
2530 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2531 				table, sizeof(*table), NULL);
2532 	if (ret)
2533 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2534 	kfree(table);
2535 	return ret;
2536 }
2537 
2538 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2539 {
2540 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2541 	u64 latency;
2542 
2543 	switch (val) {
2544 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2545 	case PM_QOS_LATENCY_ANY:
2546 		latency = U64_MAX;
2547 		break;
2548 
2549 	default:
2550 		latency = val;
2551 	}
2552 
2553 	if (ctrl->ps_max_latency_us != latency) {
2554 		ctrl->ps_max_latency_us = latency;
2555 		if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
2556 			nvme_configure_apst(ctrl);
2557 	}
2558 }
2559 
2560 struct nvme_core_quirk_entry {
2561 	/*
2562 	 * NVMe model and firmware strings are padded with spaces.  For
2563 	 * simplicity, strings in the quirk table are padded with NULLs
2564 	 * instead.
2565 	 */
2566 	u16 vid;
2567 	const char *mn;
2568 	const char *fr;
2569 	unsigned long quirks;
2570 };
2571 
2572 static const struct nvme_core_quirk_entry core_quirks[] = {
2573 	{
2574 		/*
2575 		 * This Toshiba device seems to die using any APST states.  See:
2576 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2577 		 */
2578 		.vid = 0x1179,
2579 		.mn = "THNSF5256GPUK TOSHIBA",
2580 		.quirks = NVME_QUIRK_NO_APST,
2581 	},
2582 	{
2583 		/*
2584 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2585 		 * condition associated with actions related to suspend to idle
2586 		 * LiteON has resolved the problem in future firmware
2587 		 */
2588 		.vid = 0x14a4,
2589 		.fr = "22301111",
2590 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2591 	},
2592 	{
2593 		/*
2594 		 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2595 		 * aborts I/O during any load, but more easily reproducible
2596 		 * with discards (fstrim).
2597 		 *
2598 		 * The device is left in a state where it is also not possible
2599 		 * to use "nvme set-feature" to disable APST, but booting with
2600 		 * nvme_core.default_ps_max_latency=0 works.
2601 		 */
2602 		.vid = 0x1e0f,
2603 		.mn = "KCD6XVUL6T40",
2604 		.quirks = NVME_QUIRK_NO_APST,
2605 	},
2606 	{
2607 		/*
2608 		 * The external Samsung X5 SSD fails initialization without a
2609 		 * delay before checking if it is ready and has a whole set of
2610 		 * other problems.  To make this even more interesting, it
2611 		 * shares the PCI ID with internal Samsung 970 Evo Plus that
2612 		 * does not need or want these quirks.
2613 		 */
2614 		.vid = 0x144d,
2615 		.mn = "Samsung Portable SSD X5",
2616 		.quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2617 			  NVME_QUIRK_NO_DEEPEST_PS |
2618 			  NVME_QUIRK_IGNORE_DEV_SUBNQN,
2619 	}
2620 };
2621 
2622 /* match is null-terminated but idstr is space-padded. */
2623 static bool string_matches(const char *idstr, const char *match, size_t len)
2624 {
2625 	size_t matchlen;
2626 
2627 	if (!match)
2628 		return true;
2629 
2630 	matchlen = strlen(match);
2631 	WARN_ON_ONCE(matchlen > len);
2632 
2633 	if (memcmp(idstr, match, matchlen))
2634 		return false;
2635 
2636 	for (; matchlen < len; matchlen++)
2637 		if (idstr[matchlen] != ' ')
2638 			return false;
2639 
2640 	return true;
2641 }
2642 
2643 static bool quirk_matches(const struct nvme_id_ctrl *id,
2644 			  const struct nvme_core_quirk_entry *q)
2645 {
2646 	return q->vid == le16_to_cpu(id->vid) &&
2647 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2648 		string_matches(id->fr, q->fr, sizeof(id->fr));
2649 }
2650 
2651 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2652 		struct nvme_id_ctrl *id)
2653 {
2654 	size_t nqnlen;
2655 	int off;
2656 
2657 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2658 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2659 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2660 			strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2661 			return;
2662 		}
2663 
2664 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2665 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2666 	}
2667 
2668 	/*
2669 	 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2670 	 * Base Specification 2.0.  It is slightly different from the format
2671 	 * specified there due to historic reasons, and we can't change it now.
2672 	 */
2673 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2674 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2675 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2676 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2677 	off += sizeof(id->sn);
2678 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2679 	off += sizeof(id->mn);
2680 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2681 }
2682 
2683 static void nvme_release_subsystem(struct device *dev)
2684 {
2685 	struct nvme_subsystem *subsys =
2686 		container_of(dev, struct nvme_subsystem, dev);
2687 
2688 	if (subsys->instance >= 0)
2689 		ida_free(&nvme_instance_ida, subsys->instance);
2690 	kfree(subsys);
2691 }
2692 
2693 static void nvme_destroy_subsystem(struct kref *ref)
2694 {
2695 	struct nvme_subsystem *subsys =
2696 			container_of(ref, struct nvme_subsystem, ref);
2697 
2698 	mutex_lock(&nvme_subsystems_lock);
2699 	list_del(&subsys->entry);
2700 	mutex_unlock(&nvme_subsystems_lock);
2701 
2702 	ida_destroy(&subsys->ns_ida);
2703 	device_del(&subsys->dev);
2704 	put_device(&subsys->dev);
2705 }
2706 
2707 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2708 {
2709 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2710 }
2711 
2712 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2713 {
2714 	struct nvme_subsystem *subsys;
2715 
2716 	lockdep_assert_held(&nvme_subsystems_lock);
2717 
2718 	/*
2719 	 * Fail matches for discovery subsystems. This results
2720 	 * in each discovery controller bound to a unique subsystem.
2721 	 * This avoids issues with validating controller values
2722 	 * that can only be true when there is a single unique subsystem.
2723 	 * There may be multiple and completely independent entities
2724 	 * that provide discovery controllers.
2725 	 */
2726 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2727 		return NULL;
2728 
2729 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2730 		if (strcmp(subsys->subnqn, subsysnqn))
2731 			continue;
2732 		if (!kref_get_unless_zero(&subsys->ref))
2733 			continue;
2734 		return subsys;
2735 	}
2736 
2737 	return NULL;
2738 }
2739 
2740 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2741 {
2742 	return ctrl->opts && ctrl->opts->discovery_nqn;
2743 }
2744 
2745 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2746 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2747 {
2748 	struct nvme_ctrl *tmp;
2749 
2750 	lockdep_assert_held(&nvme_subsystems_lock);
2751 
2752 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2753 		if (nvme_state_terminal(tmp))
2754 			continue;
2755 
2756 		if (tmp->cntlid == ctrl->cntlid) {
2757 			dev_err(ctrl->device,
2758 				"Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2759 				ctrl->cntlid, dev_name(tmp->device),
2760 				subsys->subnqn);
2761 			return false;
2762 		}
2763 
2764 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2765 		    nvme_discovery_ctrl(ctrl))
2766 			continue;
2767 
2768 		dev_err(ctrl->device,
2769 			"Subsystem does not support multiple controllers\n");
2770 		return false;
2771 	}
2772 
2773 	return true;
2774 }
2775 
2776 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2777 {
2778 	struct nvme_subsystem *subsys, *found;
2779 	int ret;
2780 
2781 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2782 	if (!subsys)
2783 		return -ENOMEM;
2784 
2785 	subsys->instance = -1;
2786 	mutex_init(&subsys->lock);
2787 	kref_init(&subsys->ref);
2788 	INIT_LIST_HEAD(&subsys->ctrls);
2789 	INIT_LIST_HEAD(&subsys->nsheads);
2790 	nvme_init_subnqn(subsys, ctrl, id);
2791 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2792 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2793 	subsys->vendor_id = le16_to_cpu(id->vid);
2794 	subsys->cmic = id->cmic;
2795 
2796 	/* Versions prior to 1.4 don't necessarily report a valid type */
2797 	if (id->cntrltype == NVME_CTRL_DISC ||
2798 	    !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2799 		subsys->subtype = NVME_NQN_DISC;
2800 	else
2801 		subsys->subtype = NVME_NQN_NVME;
2802 
2803 	if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2804 		dev_err(ctrl->device,
2805 			"Subsystem %s is not a discovery controller",
2806 			subsys->subnqn);
2807 		kfree(subsys);
2808 		return -EINVAL;
2809 	}
2810 	subsys->awupf = le16_to_cpu(id->awupf);
2811 	nvme_mpath_default_iopolicy(subsys);
2812 
2813 	subsys->dev.class = nvme_subsys_class;
2814 	subsys->dev.release = nvme_release_subsystem;
2815 	subsys->dev.groups = nvme_subsys_attrs_groups;
2816 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2817 	device_initialize(&subsys->dev);
2818 
2819 	mutex_lock(&nvme_subsystems_lock);
2820 	found = __nvme_find_get_subsystem(subsys->subnqn);
2821 	if (found) {
2822 		put_device(&subsys->dev);
2823 		subsys = found;
2824 
2825 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2826 			ret = -EINVAL;
2827 			goto out_put_subsystem;
2828 		}
2829 	} else {
2830 		ret = device_add(&subsys->dev);
2831 		if (ret) {
2832 			dev_err(ctrl->device,
2833 				"failed to register subsystem device.\n");
2834 			put_device(&subsys->dev);
2835 			goto out_unlock;
2836 		}
2837 		ida_init(&subsys->ns_ida);
2838 		list_add_tail(&subsys->entry, &nvme_subsystems);
2839 	}
2840 
2841 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2842 				dev_name(ctrl->device));
2843 	if (ret) {
2844 		dev_err(ctrl->device,
2845 			"failed to create sysfs link from subsystem.\n");
2846 		goto out_put_subsystem;
2847 	}
2848 
2849 	if (!found)
2850 		subsys->instance = ctrl->instance;
2851 	ctrl->subsys = subsys;
2852 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2853 	mutex_unlock(&nvme_subsystems_lock);
2854 	return 0;
2855 
2856 out_put_subsystem:
2857 	nvme_put_subsystem(subsys);
2858 out_unlock:
2859 	mutex_unlock(&nvme_subsystems_lock);
2860 	return ret;
2861 }
2862 
2863 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2864 		void *log, size_t size, u64 offset)
2865 {
2866 	struct nvme_command c = { };
2867 	u32 dwlen = nvme_bytes_to_numd(size);
2868 
2869 	c.get_log_page.opcode = nvme_admin_get_log_page;
2870 	c.get_log_page.nsid = cpu_to_le32(nsid);
2871 	c.get_log_page.lid = log_page;
2872 	c.get_log_page.lsp = lsp;
2873 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2874 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2875 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2876 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2877 	c.get_log_page.csi = csi;
2878 
2879 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2880 }
2881 
2882 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2883 				struct nvme_effects_log **log)
2884 {
2885 	struct nvme_effects_log	*cel = xa_load(&ctrl->cels, csi);
2886 	int ret;
2887 
2888 	if (cel)
2889 		goto out;
2890 
2891 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2892 	if (!cel)
2893 		return -ENOMEM;
2894 
2895 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2896 			cel, sizeof(*cel), 0);
2897 	if (ret) {
2898 		kfree(cel);
2899 		return ret;
2900 	}
2901 
2902 	xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2903 out:
2904 	*log = cel;
2905 	return 0;
2906 }
2907 
2908 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2909 {
2910 	u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2911 
2912 	if (check_shl_overflow(1U, units + page_shift - 9, &val))
2913 		return UINT_MAX;
2914 	return val;
2915 }
2916 
2917 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2918 {
2919 	struct nvme_command c = { };
2920 	struct nvme_id_ctrl_nvm *id;
2921 	int ret;
2922 
2923 	if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2924 		ctrl->max_discard_sectors = UINT_MAX;
2925 		ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2926 	} else {
2927 		ctrl->max_discard_sectors = 0;
2928 		ctrl->max_discard_segments = 0;
2929 	}
2930 
2931 	/*
2932 	 * Even though NVMe spec explicitly states that MDTS is not applicable
2933 	 * to the write-zeroes, we are cautious and limit the size to the
2934 	 * controllers max_hw_sectors value, which is based on the MDTS field
2935 	 * and possibly other limiting factors.
2936 	 */
2937 	if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2938 	    !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2939 		ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2940 	else
2941 		ctrl->max_zeroes_sectors = 0;
2942 
2943 	if (ctrl->subsys->subtype != NVME_NQN_NVME ||
2944 	    nvme_ctrl_limited_cns(ctrl) ||
2945 	    test_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags))
2946 		return 0;
2947 
2948 	id = kzalloc(sizeof(*id), GFP_KERNEL);
2949 	if (!id)
2950 		return -ENOMEM;
2951 
2952 	c.identify.opcode = nvme_admin_identify;
2953 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
2954 	c.identify.csi = NVME_CSI_NVM;
2955 
2956 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2957 	if (ret)
2958 		goto free_data;
2959 
2960 	if (id->dmrl)
2961 		ctrl->max_discard_segments = id->dmrl;
2962 	ctrl->dmrsl = le32_to_cpu(id->dmrsl);
2963 	if (id->wzsl)
2964 		ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2965 
2966 free_data:
2967 	if (ret > 0)
2968 		set_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags);
2969 	kfree(id);
2970 	return ret;
2971 }
2972 
2973 static void nvme_init_known_nvm_effects(struct nvme_ctrl *ctrl)
2974 {
2975 	struct nvme_effects_log	*log = ctrl->effects;
2976 
2977 	log->acs[nvme_admin_format_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
2978 						NVME_CMD_EFFECTS_NCC |
2979 						NVME_CMD_EFFECTS_CSE_MASK);
2980 	log->acs[nvme_admin_sanitize_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
2981 						NVME_CMD_EFFECTS_CSE_MASK);
2982 
2983 	/*
2984 	 * The spec says the result of a security receive command depends on
2985 	 * the previous security send command. As such, many vendors log this
2986 	 * command as one to submitted only when no other commands to the same
2987 	 * namespace are outstanding. The intention is to tell the host to
2988 	 * prevent mixing security send and receive.
2989 	 *
2990 	 * This driver can only enforce such exclusive access against IO
2991 	 * queues, though. We are not readily able to enforce such a rule for
2992 	 * two commands to the admin queue, which is the only queue that
2993 	 * matters for this command.
2994 	 *
2995 	 * Rather than blindly freezing the IO queues for this effect that
2996 	 * doesn't even apply to IO, mask it off.
2997 	 */
2998 	log->acs[nvme_admin_security_recv] &= cpu_to_le32(~NVME_CMD_EFFECTS_CSE_MASK);
2999 
3000 	log->iocs[nvme_cmd_write] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3001 	log->iocs[nvme_cmd_write_zeroes] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3002 	log->iocs[nvme_cmd_write_uncor] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3003 }
3004 
3005 static int nvme_init_effects(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
3006 {
3007 	int ret = 0;
3008 
3009 	if (ctrl->effects)
3010 		return 0;
3011 
3012 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3013 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3014 		if (ret < 0)
3015 			return ret;
3016 	}
3017 
3018 	if (!ctrl->effects) {
3019 		ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
3020 		if (!ctrl->effects)
3021 			return -ENOMEM;
3022 		xa_store(&ctrl->cels, NVME_CSI_NVM, ctrl->effects, GFP_KERNEL);
3023 	}
3024 
3025 	nvme_init_known_nvm_effects(ctrl);
3026 	return 0;
3027 }
3028 
3029 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3030 {
3031 	struct nvme_id_ctrl *id;
3032 	u32 max_hw_sectors;
3033 	bool prev_apst_enabled;
3034 	int ret;
3035 
3036 	ret = nvme_identify_ctrl(ctrl, &id);
3037 	if (ret) {
3038 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3039 		return -EIO;
3040 	}
3041 
3042 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
3043 		ctrl->cntlid = le16_to_cpu(id->cntlid);
3044 
3045 	if (!ctrl->identified) {
3046 		unsigned int i;
3047 
3048 		/*
3049 		 * Check for quirks.  Quirk can depend on firmware version,
3050 		 * so, in principle, the set of quirks present can change
3051 		 * across a reset.  As a possible future enhancement, we
3052 		 * could re-scan for quirks every time we reinitialize
3053 		 * the device, but we'd have to make sure that the driver
3054 		 * behaves intelligently if the quirks change.
3055 		 */
3056 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3057 			if (quirk_matches(id, &core_quirks[i]))
3058 				ctrl->quirks |= core_quirks[i].quirks;
3059 		}
3060 
3061 		ret = nvme_init_subsystem(ctrl, id);
3062 		if (ret)
3063 			goto out_free;
3064 
3065 		ret = nvme_init_effects(ctrl, id);
3066 		if (ret)
3067 			goto out_free;
3068 	}
3069 	memcpy(ctrl->subsys->firmware_rev, id->fr,
3070 	       sizeof(ctrl->subsys->firmware_rev));
3071 
3072 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3073 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3074 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3075 	}
3076 
3077 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3078 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3079 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3080 
3081 	ctrl->oacs = le16_to_cpu(id->oacs);
3082 	ctrl->oncs = le16_to_cpu(id->oncs);
3083 	ctrl->mtfa = le16_to_cpu(id->mtfa);
3084 	ctrl->oaes = le32_to_cpu(id->oaes);
3085 	ctrl->wctemp = le16_to_cpu(id->wctemp);
3086 	ctrl->cctemp = le16_to_cpu(id->cctemp);
3087 
3088 	atomic_set(&ctrl->abort_limit, id->acl + 1);
3089 	ctrl->vwc = id->vwc;
3090 	if (id->mdts)
3091 		max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3092 	else
3093 		max_hw_sectors = UINT_MAX;
3094 	ctrl->max_hw_sectors =
3095 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3096 
3097 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
3098 	ctrl->sgls = le32_to_cpu(id->sgls);
3099 	ctrl->kas = le16_to_cpu(id->kas);
3100 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
3101 	ctrl->ctratt = le32_to_cpu(id->ctratt);
3102 
3103 	ctrl->cntrltype = id->cntrltype;
3104 	ctrl->dctype = id->dctype;
3105 
3106 	if (id->rtd3e) {
3107 		/* us -> s */
3108 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3109 
3110 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3111 						 shutdown_timeout, 60);
3112 
3113 		if (ctrl->shutdown_timeout != shutdown_timeout)
3114 			dev_info(ctrl->device,
3115 				 "Shutdown timeout set to %u seconds\n",
3116 				 ctrl->shutdown_timeout);
3117 	} else
3118 		ctrl->shutdown_timeout = shutdown_timeout;
3119 
3120 	ctrl->npss = id->npss;
3121 	ctrl->apsta = id->apsta;
3122 	prev_apst_enabled = ctrl->apst_enabled;
3123 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3124 		if (force_apst && id->apsta) {
3125 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3126 			ctrl->apst_enabled = true;
3127 		} else {
3128 			ctrl->apst_enabled = false;
3129 		}
3130 	} else {
3131 		ctrl->apst_enabled = id->apsta;
3132 	}
3133 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3134 
3135 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3136 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3137 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3138 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3139 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3140 
3141 		/*
3142 		 * In fabrics we need to verify the cntlid matches the
3143 		 * admin connect
3144 		 */
3145 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3146 			dev_err(ctrl->device,
3147 				"Mismatching cntlid: Connect %u vs Identify "
3148 				"%u, rejecting\n",
3149 				ctrl->cntlid, le16_to_cpu(id->cntlid));
3150 			ret = -EINVAL;
3151 			goto out_free;
3152 		}
3153 
3154 		if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3155 			dev_err(ctrl->device,
3156 				"keep-alive support is mandatory for fabrics\n");
3157 			ret = -EINVAL;
3158 			goto out_free;
3159 		}
3160 	} else {
3161 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3162 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3163 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3164 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3165 	}
3166 
3167 	ret = nvme_mpath_init_identify(ctrl, id);
3168 	if (ret < 0)
3169 		goto out_free;
3170 
3171 	if (ctrl->apst_enabled && !prev_apst_enabled)
3172 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3173 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3174 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3175 
3176 out_free:
3177 	kfree(id);
3178 	return ret;
3179 }
3180 
3181 /*
3182  * Initialize the cached copies of the Identify data and various controller
3183  * register in our nvme_ctrl structure.  This should be called as soon as
3184  * the admin queue is fully up and running.
3185  */
3186 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended)
3187 {
3188 	int ret;
3189 
3190 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3191 	if (ret) {
3192 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3193 		return ret;
3194 	}
3195 
3196 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3197 
3198 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3199 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3200 
3201 	ret = nvme_init_identify(ctrl);
3202 	if (ret)
3203 		return ret;
3204 
3205 	ret = nvme_configure_apst(ctrl);
3206 	if (ret < 0)
3207 		return ret;
3208 
3209 	ret = nvme_configure_timestamp(ctrl);
3210 	if (ret < 0)
3211 		return ret;
3212 
3213 	ret = nvme_configure_host_options(ctrl);
3214 	if (ret < 0)
3215 		return ret;
3216 
3217 	nvme_configure_opal(ctrl, was_suspended);
3218 
3219 	if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3220 		/*
3221 		 * Do not return errors unless we are in a controller reset,
3222 		 * the controller works perfectly fine without hwmon.
3223 		 */
3224 		ret = nvme_hwmon_init(ctrl);
3225 		if (ret == -EINTR)
3226 			return ret;
3227 	}
3228 
3229 	clear_bit(NVME_CTRL_DIRTY_CAPABILITY, &ctrl->flags);
3230 	ctrl->identified = true;
3231 
3232 	nvme_start_keep_alive(ctrl);
3233 
3234 	return 0;
3235 }
3236 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3237 
3238 static int nvme_dev_open(struct inode *inode, struct file *file)
3239 {
3240 	struct nvme_ctrl *ctrl =
3241 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3242 
3243 	switch (nvme_ctrl_state(ctrl)) {
3244 	case NVME_CTRL_LIVE:
3245 		break;
3246 	default:
3247 		return -EWOULDBLOCK;
3248 	}
3249 
3250 	nvme_get_ctrl(ctrl);
3251 	if (!try_module_get(ctrl->ops->module)) {
3252 		nvme_put_ctrl(ctrl);
3253 		return -EINVAL;
3254 	}
3255 
3256 	file->private_data = ctrl;
3257 	return 0;
3258 }
3259 
3260 static int nvme_dev_release(struct inode *inode, struct file *file)
3261 {
3262 	struct nvme_ctrl *ctrl =
3263 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3264 
3265 	module_put(ctrl->ops->module);
3266 	nvme_put_ctrl(ctrl);
3267 	return 0;
3268 }
3269 
3270 static const struct file_operations nvme_dev_fops = {
3271 	.owner		= THIS_MODULE,
3272 	.open		= nvme_dev_open,
3273 	.release	= nvme_dev_release,
3274 	.unlocked_ioctl	= nvme_dev_ioctl,
3275 	.compat_ioctl	= compat_ptr_ioctl,
3276 	.uring_cmd	= nvme_dev_uring_cmd,
3277 };
3278 
3279 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3280 		unsigned nsid)
3281 {
3282 	struct nvme_ns_head *h;
3283 
3284 	lockdep_assert_held(&ctrl->subsys->lock);
3285 
3286 	list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3287 		/*
3288 		 * Private namespaces can share NSIDs under some conditions.
3289 		 * In that case we can't use the same ns_head for namespaces
3290 		 * with the same NSID.
3291 		 */
3292 		if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3293 			continue;
3294 		if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3295 			return h;
3296 	}
3297 
3298 	return NULL;
3299 }
3300 
3301 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3302 		struct nvme_ns_ids *ids)
3303 {
3304 	bool has_uuid = !uuid_is_null(&ids->uuid);
3305 	bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3306 	bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3307 	struct nvme_ns_head *h;
3308 
3309 	lockdep_assert_held(&subsys->lock);
3310 
3311 	list_for_each_entry(h, &subsys->nsheads, entry) {
3312 		if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3313 			return -EINVAL;
3314 		if (has_nguid &&
3315 		    memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3316 			return -EINVAL;
3317 		if (has_eui64 &&
3318 		    memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3319 			return -EINVAL;
3320 	}
3321 
3322 	return 0;
3323 }
3324 
3325 static void nvme_cdev_rel(struct device *dev)
3326 {
3327 	ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3328 }
3329 
3330 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3331 {
3332 	cdev_device_del(cdev, cdev_device);
3333 	put_device(cdev_device);
3334 }
3335 
3336 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3337 		const struct file_operations *fops, struct module *owner)
3338 {
3339 	int minor, ret;
3340 
3341 	minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3342 	if (minor < 0)
3343 		return minor;
3344 	cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3345 	cdev_device->class = nvme_ns_chr_class;
3346 	cdev_device->release = nvme_cdev_rel;
3347 	device_initialize(cdev_device);
3348 	cdev_init(cdev, fops);
3349 	cdev->owner = owner;
3350 	ret = cdev_device_add(cdev, cdev_device);
3351 	if (ret)
3352 		put_device(cdev_device);
3353 
3354 	return ret;
3355 }
3356 
3357 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3358 {
3359 	return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3360 }
3361 
3362 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3363 {
3364 	nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3365 	return 0;
3366 }
3367 
3368 static const struct file_operations nvme_ns_chr_fops = {
3369 	.owner		= THIS_MODULE,
3370 	.open		= nvme_ns_chr_open,
3371 	.release	= nvme_ns_chr_release,
3372 	.unlocked_ioctl	= nvme_ns_chr_ioctl,
3373 	.compat_ioctl	= compat_ptr_ioctl,
3374 	.uring_cmd	= nvme_ns_chr_uring_cmd,
3375 	.uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
3376 };
3377 
3378 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3379 {
3380 	int ret;
3381 
3382 	ns->cdev_device.parent = ns->ctrl->device;
3383 	ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3384 			   ns->ctrl->instance, ns->head->instance);
3385 	if (ret)
3386 		return ret;
3387 
3388 	return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3389 			     ns->ctrl->ops->module);
3390 }
3391 
3392 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3393 		struct nvme_ns_info *info)
3394 {
3395 	struct nvme_ns_head *head;
3396 	size_t size = sizeof(*head);
3397 	int ret = -ENOMEM;
3398 
3399 #ifdef CONFIG_NVME_MULTIPATH
3400 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3401 #endif
3402 
3403 	head = kzalloc(size, GFP_KERNEL);
3404 	if (!head)
3405 		goto out;
3406 	ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3407 	if (ret < 0)
3408 		goto out_free_head;
3409 	head->instance = ret;
3410 	INIT_LIST_HEAD(&head->list);
3411 	ret = init_srcu_struct(&head->srcu);
3412 	if (ret)
3413 		goto out_ida_remove;
3414 	head->subsys = ctrl->subsys;
3415 	head->ns_id = info->nsid;
3416 	head->ids = info->ids;
3417 	head->shared = info->is_shared;
3418 	kref_init(&head->ref);
3419 
3420 	if (head->ids.csi) {
3421 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3422 		if (ret)
3423 			goto out_cleanup_srcu;
3424 	} else
3425 		head->effects = ctrl->effects;
3426 
3427 	ret = nvme_mpath_alloc_disk(ctrl, head);
3428 	if (ret)
3429 		goto out_cleanup_srcu;
3430 
3431 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3432 
3433 	kref_get(&ctrl->subsys->ref);
3434 
3435 	return head;
3436 out_cleanup_srcu:
3437 	cleanup_srcu_struct(&head->srcu);
3438 out_ida_remove:
3439 	ida_free(&ctrl->subsys->ns_ida, head->instance);
3440 out_free_head:
3441 	kfree(head);
3442 out:
3443 	if (ret > 0)
3444 		ret = blk_status_to_errno(nvme_error_status(ret));
3445 	return ERR_PTR(ret);
3446 }
3447 
3448 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3449 		struct nvme_ns_ids *ids)
3450 {
3451 	struct nvme_subsystem *s;
3452 	int ret = 0;
3453 
3454 	/*
3455 	 * Note that this check is racy as we try to avoid holding the global
3456 	 * lock over the whole ns_head creation.  But it is only intended as
3457 	 * a sanity check anyway.
3458 	 */
3459 	mutex_lock(&nvme_subsystems_lock);
3460 	list_for_each_entry(s, &nvme_subsystems, entry) {
3461 		if (s == this)
3462 			continue;
3463 		mutex_lock(&s->lock);
3464 		ret = nvme_subsys_check_duplicate_ids(s, ids);
3465 		mutex_unlock(&s->lock);
3466 		if (ret)
3467 			break;
3468 	}
3469 	mutex_unlock(&nvme_subsystems_lock);
3470 
3471 	return ret;
3472 }
3473 
3474 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
3475 {
3476 	struct nvme_ctrl *ctrl = ns->ctrl;
3477 	struct nvme_ns_head *head = NULL;
3478 	int ret;
3479 
3480 	ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
3481 	if (ret) {
3482 		/*
3483 		 * We've found two different namespaces on two different
3484 		 * subsystems that report the same ID.  This is pretty nasty
3485 		 * for anything that actually requires unique device
3486 		 * identification.  In the kernel we need this for multipathing,
3487 		 * and in user space the /dev/disk/by-id/ links rely on it.
3488 		 *
3489 		 * If the device also claims to be multi-path capable back off
3490 		 * here now and refuse the probe the second device as this is a
3491 		 * recipe for data corruption.  If not this is probably a
3492 		 * cheap consumer device if on the PCIe bus, so let the user
3493 		 * proceed and use the shiny toy, but warn that with changing
3494 		 * probing order (which due to our async probing could just be
3495 		 * device taking longer to startup) the other device could show
3496 		 * up at any time.
3497 		 */
3498 		nvme_print_device_info(ctrl);
3499 		if ((ns->ctrl->ops->flags & NVME_F_FABRICS) || /* !PCIe */
3500 		    ((ns->ctrl->subsys->cmic & NVME_CTRL_CMIC_MULTI_CTRL) &&
3501 		     info->is_shared)) {
3502 			dev_err(ctrl->device,
3503 				"ignoring nsid %d because of duplicate IDs\n",
3504 				info->nsid);
3505 			return ret;
3506 		}
3507 
3508 		dev_err(ctrl->device,
3509 			"clearing duplicate IDs for nsid %d\n", info->nsid);
3510 		dev_err(ctrl->device,
3511 			"use of /dev/disk/by-id/ may cause data corruption\n");
3512 		memset(&info->ids.nguid, 0, sizeof(info->ids.nguid));
3513 		memset(&info->ids.uuid, 0, sizeof(info->ids.uuid));
3514 		memset(&info->ids.eui64, 0, sizeof(info->ids.eui64));
3515 		ctrl->quirks |= NVME_QUIRK_BOGUS_NID;
3516 	}
3517 
3518 	mutex_lock(&ctrl->subsys->lock);
3519 	head = nvme_find_ns_head(ctrl, info->nsid);
3520 	if (!head) {
3521 		ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
3522 		if (ret) {
3523 			dev_err(ctrl->device,
3524 				"duplicate IDs in subsystem for nsid %d\n",
3525 				info->nsid);
3526 			goto out_unlock;
3527 		}
3528 		head = nvme_alloc_ns_head(ctrl, info);
3529 		if (IS_ERR(head)) {
3530 			ret = PTR_ERR(head);
3531 			goto out_unlock;
3532 		}
3533 	} else {
3534 		ret = -EINVAL;
3535 		if (!info->is_shared || !head->shared) {
3536 			dev_err(ctrl->device,
3537 				"Duplicate unshared namespace %d\n",
3538 				info->nsid);
3539 			goto out_put_ns_head;
3540 		}
3541 		if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
3542 			dev_err(ctrl->device,
3543 				"IDs don't match for shared namespace %d\n",
3544 					info->nsid);
3545 			goto out_put_ns_head;
3546 		}
3547 
3548 		if (!multipath) {
3549 			dev_warn(ctrl->device,
3550 				"Found shared namespace %d, but multipathing not supported.\n",
3551 				info->nsid);
3552 			dev_warn_once(ctrl->device,
3553 				"Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
3554 		}
3555 	}
3556 
3557 	list_add_tail_rcu(&ns->siblings, &head->list);
3558 	ns->head = head;
3559 	mutex_unlock(&ctrl->subsys->lock);
3560 	return 0;
3561 
3562 out_put_ns_head:
3563 	nvme_put_ns_head(head);
3564 out_unlock:
3565 	mutex_unlock(&ctrl->subsys->lock);
3566 	return ret;
3567 }
3568 
3569 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3570 {
3571 	struct nvme_ns *ns, *ret = NULL;
3572 
3573 	down_read(&ctrl->namespaces_rwsem);
3574 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3575 		if (ns->head->ns_id == nsid) {
3576 			if (!nvme_get_ns(ns))
3577 				continue;
3578 			ret = ns;
3579 			break;
3580 		}
3581 		if (ns->head->ns_id > nsid)
3582 			break;
3583 	}
3584 	up_read(&ctrl->namespaces_rwsem);
3585 	return ret;
3586 }
3587 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3588 
3589 /*
3590  * Add the namespace to the controller list while keeping the list ordered.
3591  */
3592 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3593 {
3594 	struct nvme_ns *tmp;
3595 
3596 	list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3597 		if (tmp->head->ns_id < ns->head->ns_id) {
3598 			list_add(&ns->list, &tmp->list);
3599 			return;
3600 		}
3601 	}
3602 	list_add(&ns->list, &ns->ctrl->namespaces);
3603 }
3604 
3605 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
3606 {
3607 	struct nvme_ns *ns;
3608 	struct gendisk *disk;
3609 	int node = ctrl->numa_node;
3610 
3611 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3612 	if (!ns)
3613 		return;
3614 
3615 	disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3616 	if (IS_ERR(disk))
3617 		goto out_free_ns;
3618 	disk->fops = &nvme_bdev_ops;
3619 	disk->private_data = ns;
3620 
3621 	ns->disk = disk;
3622 	ns->queue = disk->queue;
3623 
3624 	if (ctrl->opts && ctrl->opts->data_digest)
3625 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3626 
3627 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3628 	if (ctrl->ops->supports_pci_p2pdma &&
3629 	    ctrl->ops->supports_pci_p2pdma(ctrl))
3630 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3631 
3632 	ns->ctrl = ctrl;
3633 	kref_init(&ns->kref);
3634 
3635 	if (nvme_init_ns_head(ns, info))
3636 		goto out_cleanup_disk;
3637 
3638 	/*
3639 	 * If multipathing is enabled, the device name for all disks and not
3640 	 * just those that represent shared namespaces needs to be based on the
3641 	 * subsystem instance.  Using the controller instance for private
3642 	 * namespaces could lead to naming collisions between shared and private
3643 	 * namespaces if they don't use a common numbering scheme.
3644 	 *
3645 	 * If multipathing is not enabled, disk names must use the controller
3646 	 * instance as shared namespaces will show up as multiple block
3647 	 * devices.
3648 	 */
3649 	if (nvme_ns_head_multipath(ns->head)) {
3650 		sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
3651 			ctrl->instance, ns->head->instance);
3652 		disk->flags |= GENHD_FL_HIDDEN;
3653 	} else if (multipath) {
3654 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
3655 			ns->head->instance);
3656 	} else {
3657 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3658 			ns->head->instance);
3659 	}
3660 
3661 	if (nvme_update_ns_info(ns, info))
3662 		goto out_unlink_ns;
3663 
3664 	down_write(&ctrl->namespaces_rwsem);
3665 	/*
3666 	 * Ensure that no namespaces are added to the ctrl list after the queues
3667 	 * are frozen, thereby avoiding a deadlock between scan and reset.
3668 	 */
3669 	if (test_bit(NVME_CTRL_FROZEN, &ctrl->flags)) {
3670 		up_write(&ctrl->namespaces_rwsem);
3671 		goto out_unlink_ns;
3672 	}
3673 	nvme_ns_add_to_ctrl_list(ns);
3674 	up_write(&ctrl->namespaces_rwsem);
3675 	nvme_get_ctrl(ctrl);
3676 
3677 	if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
3678 		goto out_cleanup_ns_from_list;
3679 
3680 	if (!nvme_ns_head_multipath(ns->head))
3681 		nvme_add_ns_cdev(ns);
3682 
3683 	nvme_mpath_add_disk(ns, info->anagrpid);
3684 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3685 
3686 	return;
3687 
3688  out_cleanup_ns_from_list:
3689 	nvme_put_ctrl(ctrl);
3690 	down_write(&ctrl->namespaces_rwsem);
3691 	list_del_init(&ns->list);
3692 	up_write(&ctrl->namespaces_rwsem);
3693  out_unlink_ns:
3694 	mutex_lock(&ctrl->subsys->lock);
3695 	list_del_rcu(&ns->siblings);
3696 	if (list_empty(&ns->head->list))
3697 		list_del_init(&ns->head->entry);
3698 	mutex_unlock(&ctrl->subsys->lock);
3699 	nvme_put_ns_head(ns->head);
3700  out_cleanup_disk:
3701 	put_disk(disk);
3702  out_free_ns:
3703 	kfree(ns);
3704 }
3705 
3706 static void nvme_ns_remove(struct nvme_ns *ns)
3707 {
3708 	bool last_path = false;
3709 
3710 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3711 		return;
3712 
3713 	clear_bit(NVME_NS_READY, &ns->flags);
3714 	set_capacity(ns->disk, 0);
3715 	nvme_fault_inject_fini(&ns->fault_inject);
3716 
3717 	/*
3718 	 * Ensure that !NVME_NS_READY is seen by other threads to prevent
3719 	 * this ns going back into current_path.
3720 	 */
3721 	synchronize_srcu(&ns->head->srcu);
3722 
3723 	/* wait for concurrent submissions */
3724 	if (nvme_mpath_clear_current_path(ns))
3725 		synchronize_srcu(&ns->head->srcu);
3726 
3727 	mutex_lock(&ns->ctrl->subsys->lock);
3728 	list_del_rcu(&ns->siblings);
3729 	if (list_empty(&ns->head->list)) {
3730 		list_del_init(&ns->head->entry);
3731 		last_path = true;
3732 	}
3733 	mutex_unlock(&ns->ctrl->subsys->lock);
3734 
3735 	/* guarantee not available in head->list */
3736 	synchronize_srcu(&ns->head->srcu);
3737 
3738 	if (!nvme_ns_head_multipath(ns->head))
3739 		nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3740 	del_gendisk(ns->disk);
3741 
3742 	down_write(&ns->ctrl->namespaces_rwsem);
3743 	list_del_init(&ns->list);
3744 	up_write(&ns->ctrl->namespaces_rwsem);
3745 
3746 	if (last_path)
3747 		nvme_mpath_shutdown_disk(ns->head);
3748 	nvme_put_ns(ns);
3749 }
3750 
3751 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3752 {
3753 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3754 
3755 	if (ns) {
3756 		nvme_ns_remove(ns);
3757 		nvme_put_ns(ns);
3758 	}
3759 }
3760 
3761 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
3762 {
3763 	int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3764 
3765 	if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
3766 		dev_err(ns->ctrl->device,
3767 			"identifiers changed for nsid %d\n", ns->head->ns_id);
3768 		goto out;
3769 	}
3770 
3771 	ret = nvme_update_ns_info(ns, info);
3772 out:
3773 	/*
3774 	 * Only remove the namespace if we got a fatal error back from the
3775 	 * device, otherwise ignore the error and just move on.
3776 	 *
3777 	 * TODO: we should probably schedule a delayed retry here.
3778 	 */
3779 	if (ret > 0 && (ret & NVME_SC_DNR))
3780 		nvme_ns_remove(ns);
3781 }
3782 
3783 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3784 {
3785 	struct nvme_ns_info info = { .nsid = nsid };
3786 	struct nvme_ns *ns;
3787 	int ret;
3788 
3789 	if (nvme_identify_ns_descs(ctrl, &info))
3790 		return;
3791 
3792 	if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
3793 		dev_warn(ctrl->device,
3794 			"command set not reported for nsid: %d\n", nsid);
3795 		return;
3796 	}
3797 
3798 	/*
3799 	 * If available try to use the Command Set Idependent Identify Namespace
3800 	 * data structure to find all the generic information that is needed to
3801 	 * set up a namespace.  If not fall back to the legacy version.
3802 	 */
3803 	if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
3804 	    (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS))
3805 		ret = nvme_ns_info_from_id_cs_indep(ctrl, &info);
3806 	else
3807 		ret = nvme_ns_info_from_identify(ctrl, &info);
3808 
3809 	if (info.is_removed)
3810 		nvme_ns_remove_by_nsid(ctrl, nsid);
3811 
3812 	/*
3813 	 * Ignore the namespace if it is not ready. We will get an AEN once it
3814 	 * becomes ready and restart the scan.
3815 	 */
3816 	if (ret || !info.is_ready)
3817 		return;
3818 
3819 	ns = nvme_find_get_ns(ctrl, nsid);
3820 	if (ns) {
3821 		nvme_validate_ns(ns, &info);
3822 		nvme_put_ns(ns);
3823 	} else {
3824 		nvme_alloc_ns(ctrl, &info);
3825 	}
3826 }
3827 
3828 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3829 					unsigned nsid)
3830 {
3831 	struct nvme_ns *ns, *next;
3832 	LIST_HEAD(rm_list);
3833 
3834 	down_write(&ctrl->namespaces_rwsem);
3835 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3836 		if (ns->head->ns_id > nsid)
3837 			list_move_tail(&ns->list, &rm_list);
3838 	}
3839 	up_write(&ctrl->namespaces_rwsem);
3840 
3841 	list_for_each_entry_safe(ns, next, &rm_list, list)
3842 		nvme_ns_remove(ns);
3843 
3844 }
3845 
3846 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3847 {
3848 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3849 	__le32 *ns_list;
3850 	u32 prev = 0;
3851 	int ret = 0, i;
3852 
3853 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3854 	if (!ns_list)
3855 		return -ENOMEM;
3856 
3857 	for (;;) {
3858 		struct nvme_command cmd = {
3859 			.identify.opcode	= nvme_admin_identify,
3860 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
3861 			.identify.nsid		= cpu_to_le32(prev),
3862 		};
3863 
3864 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
3865 					    NVME_IDENTIFY_DATA_SIZE);
3866 		if (ret) {
3867 			dev_warn(ctrl->device,
3868 				"Identify NS List failed (status=0x%x)\n", ret);
3869 			goto free;
3870 		}
3871 
3872 		for (i = 0; i < nr_entries; i++) {
3873 			u32 nsid = le32_to_cpu(ns_list[i]);
3874 
3875 			if (!nsid)	/* end of the list? */
3876 				goto out;
3877 			nvme_scan_ns(ctrl, nsid);
3878 			while (++prev < nsid)
3879 				nvme_ns_remove_by_nsid(ctrl, prev);
3880 		}
3881 	}
3882  out:
3883 	nvme_remove_invalid_namespaces(ctrl, prev);
3884  free:
3885 	kfree(ns_list);
3886 	return ret;
3887 }
3888 
3889 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3890 {
3891 	struct nvme_id_ctrl *id;
3892 	u32 nn, i;
3893 
3894 	if (nvme_identify_ctrl(ctrl, &id))
3895 		return;
3896 	nn = le32_to_cpu(id->nn);
3897 	kfree(id);
3898 
3899 	for (i = 1; i <= nn; i++)
3900 		nvme_scan_ns(ctrl, i);
3901 
3902 	nvme_remove_invalid_namespaces(ctrl, nn);
3903 }
3904 
3905 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3906 {
3907 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3908 	__le32 *log;
3909 	int error;
3910 
3911 	log = kzalloc(log_size, GFP_KERNEL);
3912 	if (!log)
3913 		return;
3914 
3915 	/*
3916 	 * We need to read the log to clear the AEN, but we don't want to rely
3917 	 * on it for the changed namespace information as userspace could have
3918 	 * raced with us in reading the log page, which could cause us to miss
3919 	 * updates.
3920 	 */
3921 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
3922 			NVME_CSI_NVM, log, log_size, 0);
3923 	if (error)
3924 		dev_warn(ctrl->device,
3925 			"reading changed ns log failed: %d\n", error);
3926 
3927 	kfree(log);
3928 }
3929 
3930 static void nvme_scan_work(struct work_struct *work)
3931 {
3932 	struct nvme_ctrl *ctrl =
3933 		container_of(work, struct nvme_ctrl, scan_work);
3934 	int ret;
3935 
3936 	/* No tagset on a live ctrl means IO queues could not created */
3937 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE || !ctrl->tagset)
3938 		return;
3939 
3940 	/*
3941 	 * Identify controller limits can change at controller reset due to
3942 	 * new firmware download, even though it is not common we cannot ignore
3943 	 * such scenario. Controller's non-mdts limits are reported in the unit
3944 	 * of logical blocks that is dependent on the format of attached
3945 	 * namespace. Hence re-read the limits at the time of ns allocation.
3946 	 */
3947 	ret = nvme_init_non_mdts_limits(ctrl);
3948 	if (ret < 0) {
3949 		dev_warn(ctrl->device,
3950 			"reading non-mdts-limits failed: %d\n", ret);
3951 		return;
3952 	}
3953 
3954 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3955 		dev_info(ctrl->device, "rescanning namespaces.\n");
3956 		nvme_clear_changed_ns_log(ctrl);
3957 	}
3958 
3959 	mutex_lock(&ctrl->scan_lock);
3960 	if (nvme_ctrl_limited_cns(ctrl)) {
3961 		nvme_scan_ns_sequential(ctrl);
3962 	} else {
3963 		/*
3964 		 * Fall back to sequential scan if DNR is set to handle broken
3965 		 * devices which should support Identify NS List (as per the VS
3966 		 * they report) but don't actually support it.
3967 		 */
3968 		ret = nvme_scan_ns_list(ctrl);
3969 		if (ret > 0 && ret & NVME_SC_DNR)
3970 			nvme_scan_ns_sequential(ctrl);
3971 	}
3972 	mutex_unlock(&ctrl->scan_lock);
3973 }
3974 
3975 /*
3976  * This function iterates the namespace list unlocked to allow recovery from
3977  * controller failure. It is up to the caller to ensure the namespace list is
3978  * not modified by scan work while this function is executing.
3979  */
3980 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3981 {
3982 	struct nvme_ns *ns, *next;
3983 	LIST_HEAD(ns_list);
3984 
3985 	/*
3986 	 * make sure to requeue I/O to all namespaces as these
3987 	 * might result from the scan itself and must complete
3988 	 * for the scan_work to make progress
3989 	 */
3990 	nvme_mpath_clear_ctrl_paths(ctrl);
3991 
3992 	/*
3993 	 * Unquiesce io queues so any pending IO won't hang, especially
3994 	 * those submitted from scan work
3995 	 */
3996 	nvme_unquiesce_io_queues(ctrl);
3997 
3998 	/* prevent racing with ns scanning */
3999 	flush_work(&ctrl->scan_work);
4000 
4001 	/*
4002 	 * The dead states indicates the controller was not gracefully
4003 	 * disconnected. In that case, we won't be able to flush any data while
4004 	 * removing the namespaces' disks; fail all the queues now to avoid
4005 	 * potentially having to clean up the failed sync later.
4006 	 */
4007 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_DEAD)
4008 		nvme_mark_namespaces_dead(ctrl);
4009 
4010 	/* this is a no-op when called from the controller reset handler */
4011 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4012 
4013 	down_write(&ctrl->namespaces_rwsem);
4014 	list_splice_init(&ctrl->namespaces, &ns_list);
4015 	up_write(&ctrl->namespaces_rwsem);
4016 
4017 	list_for_each_entry_safe(ns, next, &ns_list, list)
4018 		nvme_ns_remove(ns);
4019 }
4020 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4021 
4022 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env)
4023 {
4024 	const struct nvme_ctrl *ctrl =
4025 		container_of(dev, struct nvme_ctrl, ctrl_device);
4026 	struct nvmf_ctrl_options *opts = ctrl->opts;
4027 	int ret;
4028 
4029 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4030 	if (ret)
4031 		return ret;
4032 
4033 	if (opts) {
4034 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4035 		if (ret)
4036 			return ret;
4037 
4038 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4039 				opts->trsvcid ?: "none");
4040 		if (ret)
4041 			return ret;
4042 
4043 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4044 				opts->host_traddr ?: "none");
4045 		if (ret)
4046 			return ret;
4047 
4048 		ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4049 				opts->host_iface ?: "none");
4050 	}
4051 	return ret;
4052 }
4053 
4054 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4055 {
4056 	char *envp[2] = { envdata, NULL };
4057 
4058 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4059 }
4060 
4061 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4062 {
4063 	char *envp[2] = { NULL, NULL };
4064 	u32 aen_result = ctrl->aen_result;
4065 
4066 	ctrl->aen_result = 0;
4067 	if (!aen_result)
4068 		return;
4069 
4070 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4071 	if (!envp[0])
4072 		return;
4073 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4074 	kfree(envp[0]);
4075 }
4076 
4077 static void nvme_async_event_work(struct work_struct *work)
4078 {
4079 	struct nvme_ctrl *ctrl =
4080 		container_of(work, struct nvme_ctrl, async_event_work);
4081 
4082 	nvme_aen_uevent(ctrl);
4083 
4084 	/*
4085 	 * The transport drivers must guarantee AER submission here is safe by
4086 	 * flushing ctrl async_event_work after changing the controller state
4087 	 * from LIVE and before freeing the admin queue.
4088 	*/
4089 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
4090 		ctrl->ops->submit_async_event(ctrl);
4091 }
4092 
4093 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4094 {
4095 
4096 	u32 csts;
4097 
4098 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4099 		return false;
4100 
4101 	if (csts == ~0)
4102 		return false;
4103 
4104 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4105 }
4106 
4107 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4108 {
4109 	struct nvme_fw_slot_info_log *log;
4110 
4111 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4112 	if (!log)
4113 		return;
4114 
4115 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4116 			 log, sizeof(*log), 0)) {
4117 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4118 		goto out_free_log;
4119 	}
4120 
4121 	if (log->afi & 0x70 || !(log->afi & 0x7)) {
4122 		dev_info(ctrl->device,
4123 			 "Firmware is activated after next Controller Level Reset\n");
4124 		goto out_free_log;
4125 	}
4126 
4127 	memcpy(ctrl->subsys->firmware_rev, &log->frs[(log->afi & 0x7) - 1],
4128 		sizeof(ctrl->subsys->firmware_rev));
4129 
4130 out_free_log:
4131 	kfree(log);
4132 }
4133 
4134 static void nvme_fw_act_work(struct work_struct *work)
4135 {
4136 	struct nvme_ctrl *ctrl = container_of(work,
4137 				struct nvme_ctrl, fw_act_work);
4138 	unsigned long fw_act_timeout;
4139 
4140 	if (ctrl->mtfa)
4141 		fw_act_timeout = jiffies +
4142 				msecs_to_jiffies(ctrl->mtfa * 100);
4143 	else
4144 		fw_act_timeout = jiffies +
4145 				msecs_to_jiffies(admin_timeout * 1000);
4146 
4147 	nvme_quiesce_io_queues(ctrl);
4148 	while (nvme_ctrl_pp_status(ctrl)) {
4149 		if (time_after(jiffies, fw_act_timeout)) {
4150 			dev_warn(ctrl->device,
4151 				"Fw activation timeout, reset controller\n");
4152 			nvme_try_sched_reset(ctrl);
4153 			return;
4154 		}
4155 		msleep(100);
4156 	}
4157 
4158 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4159 		return;
4160 
4161 	nvme_unquiesce_io_queues(ctrl);
4162 	/* read FW slot information to clear the AER */
4163 	nvme_get_fw_slot_info(ctrl);
4164 
4165 	queue_work(nvme_wq, &ctrl->async_event_work);
4166 }
4167 
4168 static u32 nvme_aer_type(u32 result)
4169 {
4170 	return result & 0x7;
4171 }
4172 
4173 static u32 nvme_aer_subtype(u32 result)
4174 {
4175 	return (result & 0xff00) >> 8;
4176 }
4177 
4178 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4179 {
4180 	u32 aer_notice_type = nvme_aer_subtype(result);
4181 	bool requeue = true;
4182 
4183 	switch (aer_notice_type) {
4184 	case NVME_AER_NOTICE_NS_CHANGED:
4185 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4186 		nvme_queue_scan(ctrl);
4187 		break;
4188 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4189 		/*
4190 		 * We are (ab)using the RESETTING state to prevent subsequent
4191 		 * recovery actions from interfering with the controller's
4192 		 * firmware activation.
4193 		 */
4194 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4195 			nvme_auth_stop(ctrl);
4196 			requeue = false;
4197 			queue_work(nvme_wq, &ctrl->fw_act_work);
4198 		}
4199 		break;
4200 #ifdef CONFIG_NVME_MULTIPATH
4201 	case NVME_AER_NOTICE_ANA:
4202 		if (!ctrl->ana_log_buf)
4203 			break;
4204 		queue_work(nvme_wq, &ctrl->ana_work);
4205 		break;
4206 #endif
4207 	case NVME_AER_NOTICE_DISC_CHANGED:
4208 		ctrl->aen_result = result;
4209 		break;
4210 	default:
4211 		dev_warn(ctrl->device, "async event result %08x\n", result);
4212 	}
4213 	return requeue;
4214 }
4215 
4216 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4217 {
4218 	dev_warn(ctrl->device, "resetting controller due to AER\n");
4219 	nvme_reset_ctrl(ctrl);
4220 }
4221 
4222 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4223 		volatile union nvme_result *res)
4224 {
4225 	u32 result = le32_to_cpu(res->u32);
4226 	u32 aer_type = nvme_aer_type(result);
4227 	u32 aer_subtype = nvme_aer_subtype(result);
4228 	bool requeue = true;
4229 
4230 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4231 		return;
4232 
4233 	trace_nvme_async_event(ctrl, result);
4234 	switch (aer_type) {
4235 	case NVME_AER_NOTICE:
4236 		requeue = nvme_handle_aen_notice(ctrl, result);
4237 		break;
4238 	case NVME_AER_ERROR:
4239 		/*
4240 		 * For a persistent internal error, don't run async_event_work
4241 		 * to submit a new AER. The controller reset will do it.
4242 		 */
4243 		if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4244 			nvme_handle_aer_persistent_error(ctrl);
4245 			return;
4246 		}
4247 		fallthrough;
4248 	case NVME_AER_SMART:
4249 	case NVME_AER_CSS:
4250 	case NVME_AER_VS:
4251 		ctrl->aen_result = result;
4252 		break;
4253 	default:
4254 		break;
4255 	}
4256 
4257 	if (requeue)
4258 		queue_work(nvme_wq, &ctrl->async_event_work);
4259 }
4260 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4261 
4262 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4263 		const struct blk_mq_ops *ops, unsigned int cmd_size)
4264 {
4265 	int ret;
4266 
4267 	memset(set, 0, sizeof(*set));
4268 	set->ops = ops;
4269 	set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4270 	if (ctrl->ops->flags & NVME_F_FABRICS)
4271 		set->reserved_tags = NVMF_RESERVED_TAGS;
4272 	set->numa_node = ctrl->numa_node;
4273 	set->flags = BLK_MQ_F_NO_SCHED;
4274 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4275 		set->flags |= BLK_MQ_F_BLOCKING;
4276 	set->cmd_size = cmd_size;
4277 	set->driver_data = ctrl;
4278 	set->nr_hw_queues = 1;
4279 	set->timeout = NVME_ADMIN_TIMEOUT;
4280 	ret = blk_mq_alloc_tag_set(set);
4281 	if (ret)
4282 		return ret;
4283 
4284 	ctrl->admin_q = blk_mq_init_queue(set);
4285 	if (IS_ERR(ctrl->admin_q)) {
4286 		ret = PTR_ERR(ctrl->admin_q);
4287 		goto out_free_tagset;
4288 	}
4289 
4290 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4291 		ctrl->fabrics_q = blk_mq_init_queue(set);
4292 		if (IS_ERR(ctrl->fabrics_q)) {
4293 			ret = PTR_ERR(ctrl->fabrics_q);
4294 			goto out_cleanup_admin_q;
4295 		}
4296 	}
4297 
4298 	ctrl->admin_tagset = set;
4299 	return 0;
4300 
4301 out_cleanup_admin_q:
4302 	blk_mq_destroy_queue(ctrl->admin_q);
4303 	blk_put_queue(ctrl->admin_q);
4304 out_free_tagset:
4305 	blk_mq_free_tag_set(set);
4306 	ctrl->admin_q = NULL;
4307 	ctrl->fabrics_q = NULL;
4308 	return ret;
4309 }
4310 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4311 
4312 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4313 {
4314 	blk_mq_destroy_queue(ctrl->admin_q);
4315 	blk_put_queue(ctrl->admin_q);
4316 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4317 		blk_mq_destroy_queue(ctrl->fabrics_q);
4318 		blk_put_queue(ctrl->fabrics_q);
4319 	}
4320 	blk_mq_free_tag_set(ctrl->admin_tagset);
4321 }
4322 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4323 
4324 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4325 		const struct blk_mq_ops *ops, unsigned int nr_maps,
4326 		unsigned int cmd_size)
4327 {
4328 	int ret;
4329 
4330 	memset(set, 0, sizeof(*set));
4331 	set->ops = ops;
4332 	set->queue_depth = min_t(unsigned, ctrl->sqsize, BLK_MQ_MAX_DEPTH - 1);
4333 	/*
4334 	 * Some Apple controllers requires tags to be unique across admin and
4335 	 * the (only) I/O queue, so reserve the first 32 tags of the I/O queue.
4336 	 */
4337 	if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
4338 		set->reserved_tags = NVME_AQ_DEPTH;
4339 	else if (ctrl->ops->flags & NVME_F_FABRICS)
4340 		set->reserved_tags = NVMF_RESERVED_TAGS;
4341 	set->numa_node = ctrl->numa_node;
4342 	set->flags = BLK_MQ_F_SHOULD_MERGE;
4343 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4344 		set->flags |= BLK_MQ_F_BLOCKING;
4345 	set->cmd_size = cmd_size,
4346 	set->driver_data = ctrl;
4347 	set->nr_hw_queues = ctrl->queue_count - 1;
4348 	set->timeout = NVME_IO_TIMEOUT;
4349 	set->nr_maps = nr_maps;
4350 	ret = blk_mq_alloc_tag_set(set);
4351 	if (ret)
4352 		return ret;
4353 
4354 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4355 		ctrl->connect_q = blk_mq_init_queue(set);
4356         	if (IS_ERR(ctrl->connect_q)) {
4357 			ret = PTR_ERR(ctrl->connect_q);
4358 			goto out_free_tag_set;
4359 		}
4360 		blk_queue_flag_set(QUEUE_FLAG_SKIP_TAGSET_QUIESCE,
4361 				   ctrl->connect_q);
4362 	}
4363 
4364 	ctrl->tagset = set;
4365 	return 0;
4366 
4367 out_free_tag_set:
4368 	blk_mq_free_tag_set(set);
4369 	ctrl->connect_q = NULL;
4370 	return ret;
4371 }
4372 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
4373 
4374 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
4375 {
4376 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4377 		blk_mq_destroy_queue(ctrl->connect_q);
4378 		blk_put_queue(ctrl->connect_q);
4379 	}
4380 	blk_mq_free_tag_set(ctrl->tagset);
4381 }
4382 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
4383 
4384 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4385 {
4386 	nvme_mpath_stop(ctrl);
4387 	nvme_auth_stop(ctrl);
4388 	nvme_stop_keep_alive(ctrl);
4389 	nvme_stop_failfast_work(ctrl);
4390 	flush_work(&ctrl->async_event_work);
4391 	cancel_work_sync(&ctrl->fw_act_work);
4392 	if (ctrl->ops->stop_ctrl)
4393 		ctrl->ops->stop_ctrl(ctrl);
4394 }
4395 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4396 
4397 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4398 {
4399 	nvme_enable_aen(ctrl);
4400 
4401 	/*
4402 	 * persistent discovery controllers need to send indication to userspace
4403 	 * to re-read the discovery log page to learn about possible changes
4404 	 * that were missed. We identify persistent discovery controllers by
4405 	 * checking that they started once before, hence are reconnecting back.
4406 	 */
4407 	if (test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
4408 	    nvme_discovery_ctrl(ctrl))
4409 		nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
4410 
4411 	if (ctrl->queue_count > 1) {
4412 		nvme_queue_scan(ctrl);
4413 		nvme_unquiesce_io_queues(ctrl);
4414 		nvme_mpath_update(ctrl);
4415 	}
4416 
4417 	nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4418 	set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags);
4419 }
4420 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4421 
4422 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4423 {
4424 	nvme_hwmon_exit(ctrl);
4425 	nvme_fault_inject_fini(&ctrl->fault_inject);
4426 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4427 	cdev_device_del(&ctrl->cdev, ctrl->device);
4428 	nvme_put_ctrl(ctrl);
4429 }
4430 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4431 
4432 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4433 {
4434 	struct nvme_effects_log	*cel;
4435 	unsigned long i;
4436 
4437 	xa_for_each(&ctrl->cels, i, cel) {
4438 		xa_erase(&ctrl->cels, i);
4439 		kfree(cel);
4440 	}
4441 
4442 	xa_destroy(&ctrl->cels);
4443 }
4444 
4445 static void nvme_free_ctrl(struct device *dev)
4446 {
4447 	struct nvme_ctrl *ctrl =
4448 		container_of(dev, struct nvme_ctrl, ctrl_device);
4449 	struct nvme_subsystem *subsys = ctrl->subsys;
4450 
4451 	if (!subsys || ctrl->instance != subsys->instance)
4452 		ida_free(&nvme_instance_ida, ctrl->instance);
4453 	key_put(ctrl->tls_key);
4454 	nvme_free_cels(ctrl);
4455 	nvme_mpath_uninit(ctrl);
4456 	nvme_auth_stop(ctrl);
4457 	nvme_auth_free(ctrl);
4458 	__free_page(ctrl->discard_page);
4459 	free_opal_dev(ctrl->opal_dev);
4460 
4461 	if (subsys) {
4462 		mutex_lock(&nvme_subsystems_lock);
4463 		list_del(&ctrl->subsys_entry);
4464 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4465 		mutex_unlock(&nvme_subsystems_lock);
4466 	}
4467 
4468 	ctrl->ops->free_ctrl(ctrl);
4469 
4470 	if (subsys)
4471 		nvme_put_subsystem(subsys);
4472 }
4473 
4474 /*
4475  * Initialize a NVMe controller structures.  This needs to be called during
4476  * earliest initialization so that we have the initialized structured around
4477  * during probing.
4478  */
4479 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4480 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
4481 {
4482 	int ret;
4483 
4484 	WRITE_ONCE(ctrl->state, NVME_CTRL_NEW);
4485 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4486 	spin_lock_init(&ctrl->lock);
4487 	mutex_init(&ctrl->scan_lock);
4488 	INIT_LIST_HEAD(&ctrl->namespaces);
4489 	xa_init(&ctrl->cels);
4490 	init_rwsem(&ctrl->namespaces_rwsem);
4491 	ctrl->dev = dev;
4492 	ctrl->ops = ops;
4493 	ctrl->quirks = quirks;
4494 	ctrl->numa_node = NUMA_NO_NODE;
4495 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4496 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4497 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4498 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4499 	init_waitqueue_head(&ctrl->state_wq);
4500 
4501 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4502 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4503 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4504 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4505 	ctrl->ka_last_check_time = jiffies;
4506 
4507 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4508 			PAGE_SIZE);
4509 	ctrl->discard_page = alloc_page(GFP_KERNEL);
4510 	if (!ctrl->discard_page) {
4511 		ret = -ENOMEM;
4512 		goto out;
4513 	}
4514 
4515 	ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4516 	if (ret < 0)
4517 		goto out;
4518 	ctrl->instance = ret;
4519 
4520 	device_initialize(&ctrl->ctrl_device);
4521 	ctrl->device = &ctrl->ctrl_device;
4522 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4523 			ctrl->instance);
4524 	ctrl->device->class = nvme_class;
4525 	ctrl->device->parent = ctrl->dev;
4526 	if (ops->dev_attr_groups)
4527 		ctrl->device->groups = ops->dev_attr_groups;
4528 	else
4529 		ctrl->device->groups = nvme_dev_attr_groups;
4530 	ctrl->device->release = nvme_free_ctrl;
4531 	dev_set_drvdata(ctrl->device, ctrl);
4532 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4533 	if (ret)
4534 		goto out_release_instance;
4535 
4536 	nvme_get_ctrl(ctrl);
4537 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
4538 	ctrl->cdev.owner = ops->module;
4539 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4540 	if (ret)
4541 		goto out_free_name;
4542 
4543 	/*
4544 	 * Initialize latency tolerance controls.  The sysfs files won't
4545 	 * be visible to userspace unless the device actually supports APST.
4546 	 */
4547 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4548 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4549 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4550 
4551 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4552 	nvme_mpath_init_ctrl(ctrl);
4553 	ret = nvme_auth_init_ctrl(ctrl);
4554 	if (ret)
4555 		goto out_free_cdev;
4556 
4557 	return 0;
4558 out_free_cdev:
4559 	nvme_fault_inject_fini(&ctrl->fault_inject);
4560 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4561 	cdev_device_del(&ctrl->cdev, ctrl->device);
4562 out_free_name:
4563 	nvme_put_ctrl(ctrl);
4564 	kfree_const(ctrl->device->kobj.name);
4565 out_release_instance:
4566 	ida_free(&nvme_instance_ida, ctrl->instance);
4567 out:
4568 	if (ctrl->discard_page)
4569 		__free_page(ctrl->discard_page);
4570 	return ret;
4571 }
4572 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4573 
4574 /* let I/O to all namespaces fail in preparation for surprise removal */
4575 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
4576 {
4577 	struct nvme_ns *ns;
4578 
4579 	down_read(&ctrl->namespaces_rwsem);
4580 	list_for_each_entry(ns, &ctrl->namespaces, list)
4581 		blk_mark_disk_dead(ns->disk);
4582 	up_read(&ctrl->namespaces_rwsem);
4583 }
4584 EXPORT_SYMBOL_GPL(nvme_mark_namespaces_dead);
4585 
4586 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4587 {
4588 	struct nvme_ns *ns;
4589 
4590 	down_read(&ctrl->namespaces_rwsem);
4591 	list_for_each_entry(ns, &ctrl->namespaces, list)
4592 		blk_mq_unfreeze_queue(ns->queue);
4593 	up_read(&ctrl->namespaces_rwsem);
4594 	clear_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4595 }
4596 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4597 
4598 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4599 {
4600 	struct nvme_ns *ns;
4601 
4602 	down_read(&ctrl->namespaces_rwsem);
4603 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4604 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4605 		if (timeout <= 0)
4606 			break;
4607 	}
4608 	up_read(&ctrl->namespaces_rwsem);
4609 	return timeout;
4610 }
4611 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4612 
4613 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4614 {
4615 	struct nvme_ns *ns;
4616 
4617 	down_read(&ctrl->namespaces_rwsem);
4618 	list_for_each_entry(ns, &ctrl->namespaces, list)
4619 		blk_mq_freeze_queue_wait(ns->queue);
4620 	up_read(&ctrl->namespaces_rwsem);
4621 }
4622 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4623 
4624 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4625 {
4626 	struct nvme_ns *ns;
4627 
4628 	set_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4629 	down_read(&ctrl->namespaces_rwsem);
4630 	list_for_each_entry(ns, &ctrl->namespaces, list)
4631 		blk_freeze_queue_start(ns->queue);
4632 	up_read(&ctrl->namespaces_rwsem);
4633 }
4634 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4635 
4636 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl)
4637 {
4638 	if (!ctrl->tagset)
4639 		return;
4640 	if (!test_and_set_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4641 		blk_mq_quiesce_tagset(ctrl->tagset);
4642 	else
4643 		blk_mq_wait_quiesce_done(ctrl->tagset);
4644 }
4645 EXPORT_SYMBOL_GPL(nvme_quiesce_io_queues);
4646 
4647 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl)
4648 {
4649 	if (!ctrl->tagset)
4650 		return;
4651 	if (test_and_clear_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4652 		blk_mq_unquiesce_tagset(ctrl->tagset);
4653 }
4654 EXPORT_SYMBOL_GPL(nvme_unquiesce_io_queues);
4655 
4656 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl)
4657 {
4658 	if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4659 		blk_mq_quiesce_queue(ctrl->admin_q);
4660 	else
4661 		blk_mq_wait_quiesce_done(ctrl->admin_q->tag_set);
4662 }
4663 EXPORT_SYMBOL_GPL(nvme_quiesce_admin_queue);
4664 
4665 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl)
4666 {
4667 	if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4668 		blk_mq_unquiesce_queue(ctrl->admin_q);
4669 }
4670 EXPORT_SYMBOL_GPL(nvme_unquiesce_admin_queue);
4671 
4672 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4673 {
4674 	struct nvme_ns *ns;
4675 
4676 	down_read(&ctrl->namespaces_rwsem);
4677 	list_for_each_entry(ns, &ctrl->namespaces, list)
4678 		blk_sync_queue(ns->queue);
4679 	up_read(&ctrl->namespaces_rwsem);
4680 }
4681 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4682 
4683 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4684 {
4685 	nvme_sync_io_queues(ctrl);
4686 	if (ctrl->admin_q)
4687 		blk_sync_queue(ctrl->admin_q);
4688 }
4689 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4690 
4691 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4692 {
4693 	if (file->f_op != &nvme_dev_fops)
4694 		return NULL;
4695 	return file->private_data;
4696 }
4697 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4698 
4699 /*
4700  * Check we didn't inadvertently grow the command structure sizes:
4701  */
4702 static inline void _nvme_check_size(void)
4703 {
4704 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4705 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4706 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4707 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4708 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4709 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4710 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4711 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4712 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4713 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4714 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4715 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4716 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4717 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
4718 			NVME_IDENTIFY_DATA_SIZE);
4719 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4720 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
4721 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4722 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4723 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4724 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4725 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4726 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4727 	BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
4728 }
4729 
4730 
4731 static int __init nvme_core_init(void)
4732 {
4733 	int result = -ENOMEM;
4734 
4735 	_nvme_check_size();
4736 
4737 	nvme_wq = alloc_workqueue("nvme-wq",
4738 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4739 	if (!nvme_wq)
4740 		goto out;
4741 
4742 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4743 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4744 	if (!nvme_reset_wq)
4745 		goto destroy_wq;
4746 
4747 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4748 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4749 	if (!nvme_delete_wq)
4750 		goto destroy_reset_wq;
4751 
4752 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4753 			NVME_MINORS, "nvme");
4754 	if (result < 0)
4755 		goto destroy_delete_wq;
4756 
4757 	nvme_class = class_create("nvme");
4758 	if (IS_ERR(nvme_class)) {
4759 		result = PTR_ERR(nvme_class);
4760 		goto unregister_chrdev;
4761 	}
4762 	nvme_class->dev_uevent = nvme_class_uevent;
4763 
4764 	nvme_subsys_class = class_create("nvme-subsystem");
4765 	if (IS_ERR(nvme_subsys_class)) {
4766 		result = PTR_ERR(nvme_subsys_class);
4767 		goto destroy_class;
4768 	}
4769 
4770 	result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4771 				     "nvme-generic");
4772 	if (result < 0)
4773 		goto destroy_subsys_class;
4774 
4775 	nvme_ns_chr_class = class_create("nvme-generic");
4776 	if (IS_ERR(nvme_ns_chr_class)) {
4777 		result = PTR_ERR(nvme_ns_chr_class);
4778 		goto unregister_generic_ns;
4779 	}
4780 	result = nvme_init_auth();
4781 	if (result)
4782 		goto destroy_ns_chr;
4783 	return 0;
4784 
4785 destroy_ns_chr:
4786 	class_destroy(nvme_ns_chr_class);
4787 unregister_generic_ns:
4788 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4789 destroy_subsys_class:
4790 	class_destroy(nvme_subsys_class);
4791 destroy_class:
4792 	class_destroy(nvme_class);
4793 unregister_chrdev:
4794 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4795 destroy_delete_wq:
4796 	destroy_workqueue(nvme_delete_wq);
4797 destroy_reset_wq:
4798 	destroy_workqueue(nvme_reset_wq);
4799 destroy_wq:
4800 	destroy_workqueue(nvme_wq);
4801 out:
4802 	return result;
4803 }
4804 
4805 static void __exit nvme_core_exit(void)
4806 {
4807 	nvme_exit_auth();
4808 	class_destroy(nvme_ns_chr_class);
4809 	class_destroy(nvme_subsys_class);
4810 	class_destroy(nvme_class);
4811 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4812 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4813 	destroy_workqueue(nvme_delete_wq);
4814 	destroy_workqueue(nvme_reset_wq);
4815 	destroy_workqueue(nvme_wq);
4816 	ida_destroy(&nvme_ns_chr_minor_ida);
4817 	ida_destroy(&nvme_instance_ida);
4818 }
4819 
4820 MODULE_LICENSE("GPL");
4821 MODULE_VERSION("1.0");
4822 module_init(nvme_core_init);
4823 module_exit(nvme_core_exit);
4824