xref: /linux/drivers/nvme/host/core.c (revision 8e3ed5440b0c305dcd1d5fa7419bd8066d22ef42)
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 <linux/ratelimit.h>
24 #include <asm/unaligned.h>
25 
26 #include "nvme.h"
27 #include "fabrics.h"
28 #include <linux/nvme-auth.h>
29 
30 #define CREATE_TRACE_POINTS
31 #include "trace.h"
32 
33 #define NVME_MINORS		(1U << MINORBITS)
34 
35 struct nvme_ns_info {
36 	struct nvme_ns_ids ids;
37 	u32 nsid;
38 	__le32 anagrpid;
39 	bool is_shared;
40 	bool is_readonly;
41 	bool is_ready;
42 	bool is_removed;
43 };
44 
45 unsigned int admin_timeout = 60;
46 module_param(admin_timeout, uint, 0644);
47 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
48 EXPORT_SYMBOL_GPL(admin_timeout);
49 
50 unsigned int nvme_io_timeout = 30;
51 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
52 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
53 EXPORT_SYMBOL_GPL(nvme_io_timeout);
54 
55 static unsigned char shutdown_timeout = 5;
56 module_param(shutdown_timeout, byte, 0644);
57 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
58 
59 static u8 nvme_max_retries = 5;
60 module_param_named(max_retries, nvme_max_retries, byte, 0644);
61 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
62 
63 static unsigned long default_ps_max_latency_us = 100000;
64 module_param(default_ps_max_latency_us, ulong, 0644);
65 MODULE_PARM_DESC(default_ps_max_latency_us,
66 		 "max power saving latency for new devices; use PM QOS to change per device");
67 
68 static bool force_apst;
69 module_param(force_apst, bool, 0644);
70 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
71 
72 static unsigned long apst_primary_timeout_ms = 100;
73 module_param(apst_primary_timeout_ms, ulong, 0644);
74 MODULE_PARM_DESC(apst_primary_timeout_ms,
75 	"primary APST timeout in ms");
76 
77 static unsigned long apst_secondary_timeout_ms = 2000;
78 module_param(apst_secondary_timeout_ms, ulong, 0644);
79 MODULE_PARM_DESC(apst_secondary_timeout_ms,
80 	"secondary APST timeout in ms");
81 
82 static unsigned long apst_primary_latency_tol_us = 15000;
83 module_param(apst_primary_latency_tol_us, ulong, 0644);
84 MODULE_PARM_DESC(apst_primary_latency_tol_us,
85 	"primary APST latency tolerance in us");
86 
87 static unsigned long apst_secondary_latency_tol_us = 100000;
88 module_param(apst_secondary_latency_tol_us, ulong, 0644);
89 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
90 	"secondary APST latency tolerance in us");
91 
92 /*
93  * nvme_wq - hosts nvme related works that are not reset or delete
94  * nvme_reset_wq - hosts nvme reset works
95  * nvme_delete_wq - hosts nvme delete works
96  *
97  * nvme_wq will host works such as scan, aen handling, fw activation,
98  * keep-alive, periodic reconnects etc. nvme_reset_wq
99  * runs reset works which also flush works hosted on nvme_wq for
100  * serialization purposes. nvme_delete_wq host controller deletion
101  * works which flush reset works for serialization.
102  */
103 struct workqueue_struct *nvme_wq;
104 EXPORT_SYMBOL_GPL(nvme_wq);
105 
106 struct workqueue_struct *nvme_reset_wq;
107 EXPORT_SYMBOL_GPL(nvme_reset_wq);
108 
109 struct workqueue_struct *nvme_delete_wq;
110 EXPORT_SYMBOL_GPL(nvme_delete_wq);
111 
112 static LIST_HEAD(nvme_subsystems);
113 static DEFINE_MUTEX(nvme_subsystems_lock);
114 
115 static DEFINE_IDA(nvme_instance_ida);
116 static dev_t nvme_ctrl_base_chr_devt;
117 static struct class *nvme_class;
118 static struct class *nvme_subsys_class;
119 
120 static DEFINE_IDA(nvme_ns_chr_minor_ida);
121 static dev_t nvme_ns_chr_devt;
122 static struct class *nvme_ns_chr_class;
123 
124 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
125 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
126 					   unsigned nsid);
127 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
128 				   struct nvme_command *cmd);
129 
130 void nvme_queue_scan(struct nvme_ctrl *ctrl)
131 {
132 	/*
133 	 * Only new queue scan work when admin and IO queues are both alive
134 	 */
135 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE && ctrl->tagset)
136 		queue_work(nvme_wq, &ctrl->scan_work);
137 }
138 
139 /*
140  * Use this function to proceed with scheduling reset_work for a controller
141  * that had previously been set to the resetting state. This is intended for
142  * code paths that can't be interrupted by other reset attempts. A hot removal
143  * may prevent this from succeeding.
144  */
145 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
146 {
147 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_RESETTING)
148 		return -EBUSY;
149 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
150 		return -EBUSY;
151 	return 0;
152 }
153 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
154 
155 static void nvme_failfast_work(struct work_struct *work)
156 {
157 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
158 			struct nvme_ctrl, failfast_work);
159 
160 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_CONNECTING)
161 		return;
162 
163 	set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
164 	dev_info(ctrl->device, "failfast expired\n");
165 	nvme_kick_requeue_lists(ctrl);
166 }
167 
168 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
169 {
170 	if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
171 		return;
172 
173 	schedule_delayed_work(&ctrl->failfast_work,
174 			      ctrl->opts->fast_io_fail_tmo * HZ);
175 }
176 
177 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
178 {
179 	if (!ctrl->opts)
180 		return;
181 
182 	cancel_delayed_work_sync(&ctrl->failfast_work);
183 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
184 }
185 
186 
187 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
188 {
189 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
190 		return -EBUSY;
191 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
192 		return -EBUSY;
193 	return 0;
194 }
195 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
196 
197 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
198 {
199 	int ret;
200 
201 	ret = nvme_reset_ctrl(ctrl);
202 	if (!ret) {
203 		flush_work(&ctrl->reset_work);
204 		if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE)
205 			ret = -ENETRESET;
206 	}
207 
208 	return ret;
209 }
210 
211 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
212 {
213 	dev_info(ctrl->device,
214 		 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
215 
216 	flush_work(&ctrl->reset_work);
217 	nvme_stop_ctrl(ctrl);
218 	nvme_remove_namespaces(ctrl);
219 	ctrl->ops->delete_ctrl(ctrl);
220 	nvme_uninit_ctrl(ctrl);
221 }
222 
223 static void nvme_delete_ctrl_work(struct work_struct *work)
224 {
225 	struct nvme_ctrl *ctrl =
226 		container_of(work, struct nvme_ctrl, delete_work);
227 
228 	nvme_do_delete_ctrl(ctrl);
229 }
230 
231 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
232 {
233 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
234 		return -EBUSY;
235 	if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
236 		return -EBUSY;
237 	return 0;
238 }
239 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
240 
241 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
242 {
243 	/*
244 	 * Keep a reference until nvme_do_delete_ctrl() complete,
245 	 * since ->delete_ctrl can free the controller.
246 	 */
247 	nvme_get_ctrl(ctrl);
248 	if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
249 		nvme_do_delete_ctrl(ctrl);
250 	nvme_put_ctrl(ctrl);
251 }
252 
253 static blk_status_t nvme_error_status(u16 status)
254 {
255 	switch (status & 0x7ff) {
256 	case NVME_SC_SUCCESS:
257 		return BLK_STS_OK;
258 	case NVME_SC_CAP_EXCEEDED:
259 		return BLK_STS_NOSPC;
260 	case NVME_SC_LBA_RANGE:
261 	case NVME_SC_CMD_INTERRUPTED:
262 	case NVME_SC_NS_NOT_READY:
263 		return BLK_STS_TARGET;
264 	case NVME_SC_BAD_ATTRIBUTES:
265 	case NVME_SC_ONCS_NOT_SUPPORTED:
266 	case NVME_SC_INVALID_OPCODE:
267 	case NVME_SC_INVALID_FIELD:
268 	case NVME_SC_INVALID_NS:
269 		return BLK_STS_NOTSUPP;
270 	case NVME_SC_WRITE_FAULT:
271 	case NVME_SC_READ_ERROR:
272 	case NVME_SC_UNWRITTEN_BLOCK:
273 	case NVME_SC_ACCESS_DENIED:
274 	case NVME_SC_READ_ONLY:
275 	case NVME_SC_COMPARE_FAILED:
276 		return BLK_STS_MEDIUM;
277 	case NVME_SC_GUARD_CHECK:
278 	case NVME_SC_APPTAG_CHECK:
279 	case NVME_SC_REFTAG_CHECK:
280 	case NVME_SC_INVALID_PI:
281 		return BLK_STS_PROTECTION;
282 	case NVME_SC_RESERVATION_CONFLICT:
283 		return BLK_STS_RESV_CONFLICT;
284 	case NVME_SC_HOST_PATH_ERROR:
285 		return BLK_STS_TRANSPORT;
286 	case NVME_SC_ZONE_TOO_MANY_ACTIVE:
287 		return BLK_STS_ZONE_ACTIVE_RESOURCE;
288 	case NVME_SC_ZONE_TOO_MANY_OPEN:
289 		return BLK_STS_ZONE_OPEN_RESOURCE;
290 	default:
291 		return BLK_STS_IOERR;
292 	}
293 }
294 
295 static void nvme_retry_req(struct request *req)
296 {
297 	unsigned long delay = 0;
298 	u16 crd;
299 
300 	/* The mask and shift result must be <= 3 */
301 	crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
302 	if (crd)
303 		delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
304 
305 	nvme_req(req)->retries++;
306 	blk_mq_requeue_request(req, false);
307 	blk_mq_delay_kick_requeue_list(req->q, delay);
308 }
309 
310 static void nvme_log_error(struct request *req)
311 {
312 	struct nvme_ns *ns = req->q->queuedata;
313 	struct nvme_request *nr = nvme_req(req);
314 
315 	if (ns) {
316 		pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %u blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
317 		       ns->disk ? ns->disk->disk_name : "?",
318 		       nvme_get_opcode_str(nr->cmd->common.opcode),
319 		       nr->cmd->common.opcode,
320 		       nvme_sect_to_lba(ns->head, blk_rq_pos(req)),
321 		       blk_rq_bytes(req) >> ns->head->lba_shift,
322 		       nvme_get_error_status_str(nr->status),
323 		       nr->status >> 8 & 7,	/* Status Code Type */
324 		       nr->status & 0xff,	/* Status Code */
325 		       nr->status & NVME_SC_MORE ? "MORE " : "",
326 		       nr->status & NVME_SC_DNR  ? "DNR "  : "");
327 		return;
328 	}
329 
330 	pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
331 			   dev_name(nr->ctrl->device),
332 			   nvme_get_admin_opcode_str(nr->cmd->common.opcode),
333 			   nr->cmd->common.opcode,
334 			   nvme_get_error_status_str(nr->status),
335 			   nr->status >> 8 & 7,	/* Status Code Type */
336 			   nr->status & 0xff,	/* Status Code */
337 			   nr->status & NVME_SC_MORE ? "MORE " : "",
338 			   nr->status & NVME_SC_DNR  ? "DNR "  : "");
339 }
340 
341 static void nvme_log_err_passthru(struct request *req)
342 {
343 	struct nvme_ns *ns = req->q->queuedata;
344 	struct nvme_request *nr = nvme_req(req);
345 
346 	pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s"
347 		"cdw10=0x%x cdw11=0x%x cdw12=0x%x cdw13=0x%x cdw14=0x%x cdw15=0x%x\n",
348 		ns ? ns->disk->disk_name : dev_name(nr->ctrl->device),
349 		ns ? nvme_get_opcode_str(nr->cmd->common.opcode) :
350 		     nvme_get_admin_opcode_str(nr->cmd->common.opcode),
351 		nr->cmd->common.opcode,
352 		nvme_get_error_status_str(nr->status),
353 		nr->status >> 8 & 7,	/* Status Code Type */
354 		nr->status & 0xff,	/* Status Code */
355 		nr->status & NVME_SC_MORE ? "MORE " : "",
356 		nr->status & NVME_SC_DNR  ? "DNR "  : "",
357 		nr->cmd->common.cdw10,
358 		nr->cmd->common.cdw11,
359 		nr->cmd->common.cdw12,
360 		nr->cmd->common.cdw13,
361 		nr->cmd->common.cdw14,
362 		nr->cmd->common.cdw14);
363 }
364 
365 enum nvme_disposition {
366 	COMPLETE,
367 	RETRY,
368 	FAILOVER,
369 	AUTHENTICATE,
370 };
371 
372 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
373 {
374 	if (likely(nvme_req(req)->status == 0))
375 		return COMPLETE;
376 
377 	if ((nvme_req(req)->status & 0x7ff) == NVME_SC_AUTH_REQUIRED)
378 		return AUTHENTICATE;
379 
380 	if (blk_noretry_request(req) ||
381 	    (nvme_req(req)->status & NVME_SC_DNR) ||
382 	    nvme_req(req)->retries >= nvme_max_retries)
383 		return COMPLETE;
384 
385 	if (req->cmd_flags & REQ_NVME_MPATH) {
386 		if (nvme_is_path_error(nvme_req(req)->status) ||
387 		    blk_queue_dying(req->q))
388 			return FAILOVER;
389 	} else {
390 		if (blk_queue_dying(req->q))
391 			return COMPLETE;
392 	}
393 
394 	return RETRY;
395 }
396 
397 static inline void nvme_end_req_zoned(struct request *req)
398 {
399 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
400 	    req_op(req) == REQ_OP_ZONE_APPEND) {
401 		struct nvme_ns *ns = req->q->queuedata;
402 
403 		req->__sector = nvme_lba_to_sect(ns->head,
404 			le64_to_cpu(nvme_req(req)->result.u64));
405 	}
406 }
407 
408 static inline void nvme_end_req(struct request *req)
409 {
410 	blk_status_t status = nvme_error_status(nvme_req(req)->status);
411 
412 	if (unlikely(nvme_req(req)->status && !(req->rq_flags & RQF_QUIET))) {
413 		if (blk_rq_is_passthrough(req))
414 			nvme_log_err_passthru(req);
415 		else
416 			nvme_log_error(req);
417 	}
418 	nvme_end_req_zoned(req);
419 	nvme_trace_bio_complete(req);
420 	if (req->cmd_flags & REQ_NVME_MPATH)
421 		nvme_mpath_end_request(req);
422 	blk_mq_end_request(req, status);
423 }
424 
425 void nvme_complete_rq(struct request *req)
426 {
427 	struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
428 
429 	trace_nvme_complete_rq(req);
430 	nvme_cleanup_cmd(req);
431 
432 	/*
433 	 * Completions of long-running commands should not be able to
434 	 * defer sending of periodic keep alives, since the controller
435 	 * may have completed processing such commands a long time ago
436 	 * (arbitrarily close to command submission time).
437 	 * req->deadline - req->timeout is the command submission time
438 	 * in jiffies.
439 	 */
440 	if (ctrl->kas &&
441 	    req->deadline - req->timeout >= ctrl->ka_last_check_time)
442 		ctrl->comp_seen = true;
443 
444 	switch (nvme_decide_disposition(req)) {
445 	case COMPLETE:
446 		nvme_end_req(req);
447 		return;
448 	case RETRY:
449 		nvme_retry_req(req);
450 		return;
451 	case FAILOVER:
452 		nvme_failover_req(req);
453 		return;
454 	case AUTHENTICATE:
455 #ifdef CONFIG_NVME_HOST_AUTH
456 		queue_work(nvme_wq, &ctrl->dhchap_auth_work);
457 		nvme_retry_req(req);
458 #else
459 		nvme_end_req(req);
460 #endif
461 		return;
462 	}
463 }
464 EXPORT_SYMBOL_GPL(nvme_complete_rq);
465 
466 void nvme_complete_batch_req(struct request *req)
467 {
468 	trace_nvme_complete_rq(req);
469 	nvme_cleanup_cmd(req);
470 	nvme_end_req_zoned(req);
471 }
472 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
473 
474 /*
475  * Called to unwind from ->queue_rq on a failed command submission so that the
476  * multipathing code gets called to potentially failover to another path.
477  * The caller needs to unwind all transport specific resource allocations and
478  * must return propagate the return value.
479  */
480 blk_status_t nvme_host_path_error(struct request *req)
481 {
482 	nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
483 	blk_mq_set_request_complete(req);
484 	nvme_complete_rq(req);
485 	return BLK_STS_OK;
486 }
487 EXPORT_SYMBOL_GPL(nvme_host_path_error);
488 
489 bool nvme_cancel_request(struct request *req, void *data)
490 {
491 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
492 				"Cancelling I/O %d", req->tag);
493 
494 	/* don't abort one completed or idle request */
495 	if (blk_mq_rq_state(req) != MQ_RQ_IN_FLIGHT)
496 		return true;
497 
498 	nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
499 	nvme_req(req)->flags |= NVME_REQ_CANCELLED;
500 	blk_mq_complete_request(req);
501 	return true;
502 }
503 EXPORT_SYMBOL_GPL(nvme_cancel_request);
504 
505 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
506 {
507 	if (ctrl->tagset) {
508 		blk_mq_tagset_busy_iter(ctrl->tagset,
509 				nvme_cancel_request, ctrl);
510 		blk_mq_tagset_wait_completed_request(ctrl->tagset);
511 	}
512 }
513 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
514 
515 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
516 {
517 	if (ctrl->admin_tagset) {
518 		blk_mq_tagset_busy_iter(ctrl->admin_tagset,
519 				nvme_cancel_request, ctrl);
520 		blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
521 	}
522 }
523 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
524 
525 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
526 		enum nvme_ctrl_state new_state)
527 {
528 	enum nvme_ctrl_state old_state;
529 	unsigned long flags;
530 	bool changed = false;
531 
532 	spin_lock_irqsave(&ctrl->lock, flags);
533 
534 	old_state = nvme_ctrl_state(ctrl);
535 	switch (new_state) {
536 	case NVME_CTRL_LIVE:
537 		switch (old_state) {
538 		case NVME_CTRL_NEW:
539 		case NVME_CTRL_RESETTING:
540 		case NVME_CTRL_CONNECTING:
541 			changed = true;
542 			fallthrough;
543 		default:
544 			break;
545 		}
546 		break;
547 	case NVME_CTRL_RESETTING:
548 		switch (old_state) {
549 		case NVME_CTRL_NEW:
550 		case NVME_CTRL_LIVE:
551 			changed = true;
552 			fallthrough;
553 		default:
554 			break;
555 		}
556 		break;
557 	case NVME_CTRL_CONNECTING:
558 		switch (old_state) {
559 		case NVME_CTRL_NEW:
560 		case NVME_CTRL_RESETTING:
561 			changed = true;
562 			fallthrough;
563 		default:
564 			break;
565 		}
566 		break;
567 	case NVME_CTRL_DELETING:
568 		switch (old_state) {
569 		case NVME_CTRL_LIVE:
570 		case NVME_CTRL_RESETTING:
571 		case NVME_CTRL_CONNECTING:
572 			changed = true;
573 			fallthrough;
574 		default:
575 			break;
576 		}
577 		break;
578 	case NVME_CTRL_DELETING_NOIO:
579 		switch (old_state) {
580 		case NVME_CTRL_DELETING:
581 		case NVME_CTRL_DEAD:
582 			changed = true;
583 			fallthrough;
584 		default:
585 			break;
586 		}
587 		break;
588 	case NVME_CTRL_DEAD:
589 		switch (old_state) {
590 		case NVME_CTRL_DELETING:
591 			changed = true;
592 			fallthrough;
593 		default:
594 			break;
595 		}
596 		break;
597 	default:
598 		break;
599 	}
600 
601 	if (changed) {
602 		WRITE_ONCE(ctrl->state, new_state);
603 		wake_up_all(&ctrl->state_wq);
604 	}
605 
606 	spin_unlock_irqrestore(&ctrl->lock, flags);
607 	if (!changed)
608 		return false;
609 
610 	if (new_state == NVME_CTRL_LIVE) {
611 		if (old_state == NVME_CTRL_CONNECTING)
612 			nvme_stop_failfast_work(ctrl);
613 		nvme_kick_requeue_lists(ctrl);
614 	} else if (new_state == NVME_CTRL_CONNECTING &&
615 		old_state == NVME_CTRL_RESETTING) {
616 		nvme_start_failfast_work(ctrl);
617 	}
618 	return changed;
619 }
620 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
621 
622 /*
623  * Returns true for sink states that can't ever transition back to live.
624  */
625 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
626 {
627 	switch (nvme_ctrl_state(ctrl)) {
628 	case NVME_CTRL_NEW:
629 	case NVME_CTRL_LIVE:
630 	case NVME_CTRL_RESETTING:
631 	case NVME_CTRL_CONNECTING:
632 		return false;
633 	case NVME_CTRL_DELETING:
634 	case NVME_CTRL_DELETING_NOIO:
635 	case NVME_CTRL_DEAD:
636 		return true;
637 	default:
638 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
639 		return true;
640 	}
641 }
642 
643 /*
644  * Waits for the controller state to be resetting, or returns false if it is
645  * not possible to ever transition to that state.
646  */
647 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
648 {
649 	wait_event(ctrl->state_wq,
650 		   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
651 		   nvme_state_terminal(ctrl));
652 	return nvme_ctrl_state(ctrl) == NVME_CTRL_RESETTING;
653 }
654 EXPORT_SYMBOL_GPL(nvme_wait_reset);
655 
656 static void nvme_free_ns_head(struct kref *ref)
657 {
658 	struct nvme_ns_head *head =
659 		container_of(ref, struct nvme_ns_head, ref);
660 
661 	nvme_mpath_remove_disk(head);
662 	ida_free(&head->subsys->ns_ida, head->instance);
663 	cleanup_srcu_struct(&head->srcu);
664 	nvme_put_subsystem(head->subsys);
665 	kfree(head);
666 }
667 
668 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
669 {
670 	return kref_get_unless_zero(&head->ref);
671 }
672 
673 void nvme_put_ns_head(struct nvme_ns_head *head)
674 {
675 	kref_put(&head->ref, nvme_free_ns_head);
676 }
677 
678 static void nvme_free_ns(struct kref *kref)
679 {
680 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
681 
682 	put_disk(ns->disk);
683 	nvme_put_ns_head(ns->head);
684 	nvme_put_ctrl(ns->ctrl);
685 	kfree(ns);
686 }
687 
688 static inline bool nvme_get_ns(struct nvme_ns *ns)
689 {
690 	return kref_get_unless_zero(&ns->kref);
691 }
692 
693 void nvme_put_ns(struct nvme_ns *ns)
694 {
695 	kref_put(&ns->kref, nvme_free_ns);
696 }
697 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
698 
699 static inline void nvme_clear_nvme_request(struct request *req)
700 {
701 	nvme_req(req)->status = 0;
702 	nvme_req(req)->retries = 0;
703 	nvme_req(req)->flags = 0;
704 	req->rq_flags |= RQF_DONTPREP;
705 }
706 
707 /* initialize a passthrough request */
708 void nvme_init_request(struct request *req, struct nvme_command *cmd)
709 {
710 	struct nvme_request *nr = nvme_req(req);
711 	bool logging_enabled;
712 
713 	if (req->q->queuedata) {
714 		struct nvme_ns *ns = req->q->disk->private_data;
715 
716 		logging_enabled = ns->head->passthru_err_log_enabled;
717 		req->timeout = NVME_IO_TIMEOUT;
718 	} else { /* no queuedata implies admin queue */
719 		logging_enabled = nr->ctrl->passthru_err_log_enabled;
720 		req->timeout = NVME_ADMIN_TIMEOUT;
721 	}
722 
723 	if (!logging_enabled)
724 		req->rq_flags |= RQF_QUIET;
725 
726 	/* passthru commands should let the driver set the SGL flags */
727 	cmd->common.flags &= ~NVME_CMD_SGL_ALL;
728 
729 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
730 	if (req->mq_hctx->type == HCTX_TYPE_POLL)
731 		req->cmd_flags |= REQ_POLLED;
732 	nvme_clear_nvme_request(req);
733 	memcpy(nr->cmd, cmd, sizeof(*cmd));
734 }
735 EXPORT_SYMBOL_GPL(nvme_init_request);
736 
737 /*
738  * For something we're not in a state to send to the device the default action
739  * is to busy it and retry it after the controller state is recovered.  However,
740  * if the controller is deleting or if anything is marked for failfast or
741  * nvme multipath it is immediately failed.
742  *
743  * Note: commands used to initialize the controller will be marked for failfast.
744  * Note: nvme cli/ioctl commands are marked for failfast.
745  */
746 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
747 		struct request *rq)
748 {
749 	enum nvme_ctrl_state state = nvme_ctrl_state(ctrl);
750 
751 	if (state != NVME_CTRL_DELETING_NOIO &&
752 	    state != NVME_CTRL_DELETING &&
753 	    state != NVME_CTRL_DEAD &&
754 	    !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
755 	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
756 		return BLK_STS_RESOURCE;
757 	return nvme_host_path_error(rq);
758 }
759 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
760 
761 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
762 		bool queue_live, enum nvme_ctrl_state state)
763 {
764 	struct nvme_request *req = nvme_req(rq);
765 
766 	/*
767 	 * currently we have a problem sending passthru commands
768 	 * on the admin_q if the controller is not LIVE because we can't
769 	 * make sure that they are going out after the admin connect,
770 	 * controller enable and/or other commands in the initialization
771 	 * sequence. until the controller will be LIVE, fail with
772 	 * BLK_STS_RESOURCE so that they will be rescheduled.
773 	 */
774 	if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
775 		return false;
776 
777 	if (ctrl->ops->flags & NVME_F_FABRICS) {
778 		/*
779 		 * Only allow commands on a live queue, except for the connect
780 		 * command, which is require to set the queue live in the
781 		 * appropinquate states.
782 		 */
783 		switch (state) {
784 		case NVME_CTRL_CONNECTING:
785 			if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
786 			    (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
787 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
788 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
789 				return true;
790 			break;
791 		default:
792 			break;
793 		case NVME_CTRL_DEAD:
794 			return false;
795 		}
796 	}
797 
798 	return queue_live;
799 }
800 EXPORT_SYMBOL_GPL(__nvme_check_ready);
801 
802 static inline void nvme_setup_flush(struct nvme_ns *ns,
803 		struct nvme_command *cmnd)
804 {
805 	memset(cmnd, 0, sizeof(*cmnd));
806 	cmnd->common.opcode = nvme_cmd_flush;
807 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
808 }
809 
810 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
811 		struct nvme_command *cmnd)
812 {
813 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
814 	struct nvme_dsm_range *range;
815 	struct bio *bio;
816 
817 	/*
818 	 * Some devices do not consider the DSM 'Number of Ranges' field when
819 	 * determining how much data to DMA. Always allocate memory for maximum
820 	 * number of segments to prevent device reading beyond end of buffer.
821 	 */
822 	static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
823 
824 	range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
825 	if (!range) {
826 		/*
827 		 * If we fail allocation our range, fallback to the controller
828 		 * discard page. If that's also busy, it's safe to return
829 		 * busy, as we know we can make progress once that's freed.
830 		 */
831 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
832 			return BLK_STS_RESOURCE;
833 
834 		range = page_address(ns->ctrl->discard_page);
835 	}
836 
837 	if (queue_max_discard_segments(req->q) == 1) {
838 		u64 slba = nvme_sect_to_lba(ns->head, blk_rq_pos(req));
839 		u32 nlb = blk_rq_sectors(req) >> (ns->head->lba_shift - 9);
840 
841 		range[0].cattr = cpu_to_le32(0);
842 		range[0].nlb = cpu_to_le32(nlb);
843 		range[0].slba = cpu_to_le64(slba);
844 		n = 1;
845 	} else {
846 		__rq_for_each_bio(bio, req) {
847 			u64 slba = nvme_sect_to_lba(ns->head,
848 						    bio->bi_iter.bi_sector);
849 			u32 nlb = bio->bi_iter.bi_size >> ns->head->lba_shift;
850 
851 			if (n < segments) {
852 				range[n].cattr = cpu_to_le32(0);
853 				range[n].nlb = cpu_to_le32(nlb);
854 				range[n].slba = cpu_to_le64(slba);
855 			}
856 			n++;
857 		}
858 	}
859 
860 	if (WARN_ON_ONCE(n != segments)) {
861 		if (virt_to_page(range) == ns->ctrl->discard_page)
862 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
863 		else
864 			kfree(range);
865 		return BLK_STS_IOERR;
866 	}
867 
868 	memset(cmnd, 0, sizeof(*cmnd));
869 	cmnd->dsm.opcode = nvme_cmd_dsm;
870 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
871 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
872 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
873 
874 	bvec_set_virt(&req->special_vec, range, alloc_size);
875 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
876 
877 	return BLK_STS_OK;
878 }
879 
880 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
881 			      struct request *req)
882 {
883 	u32 upper, lower;
884 	u64 ref48;
885 
886 	/* both rw and write zeroes share the same reftag format */
887 	switch (ns->head->guard_type) {
888 	case NVME_NVM_NS_16B_GUARD:
889 		cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
890 		break;
891 	case NVME_NVM_NS_64B_GUARD:
892 		ref48 = ext_pi_ref_tag(req);
893 		lower = lower_32_bits(ref48);
894 		upper = upper_32_bits(ref48);
895 
896 		cmnd->rw.reftag = cpu_to_le32(lower);
897 		cmnd->rw.cdw3 = cpu_to_le32(upper);
898 		break;
899 	default:
900 		break;
901 	}
902 }
903 
904 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
905 		struct request *req, struct nvme_command *cmnd)
906 {
907 	memset(cmnd, 0, sizeof(*cmnd));
908 
909 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
910 		return nvme_setup_discard(ns, req, cmnd);
911 
912 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
913 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
914 	cmnd->write_zeroes.slba =
915 		cpu_to_le64(nvme_sect_to_lba(ns->head, blk_rq_pos(req)));
916 	cmnd->write_zeroes.length =
917 		cpu_to_le16((blk_rq_bytes(req) >> ns->head->lba_shift) - 1);
918 
919 	if (!(req->cmd_flags & REQ_NOUNMAP) &&
920 	    (ns->head->features & NVME_NS_DEAC))
921 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_WZ_DEAC);
922 
923 	if (nvme_ns_has_pi(ns->head)) {
924 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_RW_PRINFO_PRACT);
925 
926 		switch (ns->head->pi_type) {
927 		case NVME_NS_DPS_PI_TYPE1:
928 		case NVME_NS_DPS_PI_TYPE2:
929 			nvme_set_ref_tag(ns, cmnd, req);
930 			break;
931 		}
932 	}
933 
934 	return BLK_STS_OK;
935 }
936 
937 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
938 		struct request *req, struct nvme_command *cmnd,
939 		enum nvme_opcode op)
940 {
941 	u16 control = 0;
942 	u32 dsmgmt = 0;
943 
944 	if (req->cmd_flags & REQ_FUA)
945 		control |= NVME_RW_FUA;
946 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
947 		control |= NVME_RW_LR;
948 
949 	if (req->cmd_flags & REQ_RAHEAD)
950 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
951 
952 	cmnd->rw.opcode = op;
953 	cmnd->rw.flags = 0;
954 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
955 	cmnd->rw.cdw2 = 0;
956 	cmnd->rw.cdw3 = 0;
957 	cmnd->rw.metadata = 0;
958 	cmnd->rw.slba =
959 		cpu_to_le64(nvme_sect_to_lba(ns->head, blk_rq_pos(req)));
960 	cmnd->rw.length =
961 		cpu_to_le16((blk_rq_bytes(req) >> ns->head->lba_shift) - 1);
962 	cmnd->rw.reftag = 0;
963 	cmnd->rw.apptag = 0;
964 	cmnd->rw.appmask = 0;
965 
966 	if (ns->head->ms) {
967 		/*
968 		 * If formated with metadata, the block layer always provides a
969 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
970 		 * we enable the PRACT bit for protection information or set the
971 		 * namespace capacity to zero to prevent any I/O.
972 		 */
973 		if (!blk_integrity_rq(req)) {
974 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns->head)))
975 				return BLK_STS_NOTSUPP;
976 			control |= NVME_RW_PRINFO_PRACT;
977 		}
978 
979 		switch (ns->head->pi_type) {
980 		case NVME_NS_DPS_PI_TYPE3:
981 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
982 			break;
983 		case NVME_NS_DPS_PI_TYPE1:
984 		case NVME_NS_DPS_PI_TYPE2:
985 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
986 					NVME_RW_PRINFO_PRCHK_REF;
987 			if (op == nvme_cmd_zone_append)
988 				control |= NVME_RW_APPEND_PIREMAP;
989 			nvme_set_ref_tag(ns, cmnd, req);
990 			break;
991 		}
992 	}
993 
994 	cmnd->rw.control = cpu_to_le16(control);
995 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
996 	return 0;
997 }
998 
999 void nvme_cleanup_cmd(struct request *req)
1000 {
1001 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
1002 		struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
1003 
1004 		if (req->special_vec.bv_page == ctrl->discard_page)
1005 			clear_bit_unlock(0, &ctrl->discard_page_busy);
1006 		else
1007 			kfree(bvec_virt(&req->special_vec));
1008 	}
1009 }
1010 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
1011 
1012 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
1013 {
1014 	struct nvme_command *cmd = nvme_req(req)->cmd;
1015 	blk_status_t ret = BLK_STS_OK;
1016 
1017 	if (!(req->rq_flags & RQF_DONTPREP))
1018 		nvme_clear_nvme_request(req);
1019 
1020 	switch (req_op(req)) {
1021 	case REQ_OP_DRV_IN:
1022 	case REQ_OP_DRV_OUT:
1023 		/* these are setup prior to execution in nvme_init_request() */
1024 		break;
1025 	case REQ_OP_FLUSH:
1026 		nvme_setup_flush(ns, cmd);
1027 		break;
1028 	case REQ_OP_ZONE_RESET_ALL:
1029 	case REQ_OP_ZONE_RESET:
1030 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
1031 		break;
1032 	case REQ_OP_ZONE_OPEN:
1033 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
1034 		break;
1035 	case REQ_OP_ZONE_CLOSE:
1036 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
1037 		break;
1038 	case REQ_OP_ZONE_FINISH:
1039 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
1040 		break;
1041 	case REQ_OP_WRITE_ZEROES:
1042 		ret = nvme_setup_write_zeroes(ns, req, cmd);
1043 		break;
1044 	case REQ_OP_DISCARD:
1045 		ret = nvme_setup_discard(ns, req, cmd);
1046 		break;
1047 	case REQ_OP_READ:
1048 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1049 		break;
1050 	case REQ_OP_WRITE:
1051 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1052 		break;
1053 	case REQ_OP_ZONE_APPEND:
1054 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1055 		break;
1056 	default:
1057 		WARN_ON_ONCE(1);
1058 		return BLK_STS_IOERR;
1059 	}
1060 
1061 	cmd->common.command_id = nvme_cid(req);
1062 	trace_nvme_setup_cmd(req, cmd);
1063 	return ret;
1064 }
1065 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1066 
1067 /*
1068  * Return values:
1069  * 0:  success
1070  * >0: nvme controller's cqe status response
1071  * <0: kernel error in lieu of controller response
1072  */
1073 int nvme_execute_rq(struct request *rq, bool at_head)
1074 {
1075 	blk_status_t status;
1076 
1077 	status = blk_execute_rq(rq, at_head);
1078 	if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1079 		return -EINTR;
1080 	if (nvme_req(rq)->status)
1081 		return nvme_req(rq)->status;
1082 	return blk_status_to_errno(status);
1083 }
1084 EXPORT_SYMBOL_NS_GPL(nvme_execute_rq, NVME_TARGET_PASSTHRU);
1085 
1086 /*
1087  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1088  * if the result is positive, it's an NVM Express status code
1089  */
1090 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1091 		union nvme_result *result, void *buffer, unsigned bufflen,
1092 		int qid, nvme_submit_flags_t flags)
1093 {
1094 	struct request *req;
1095 	int ret;
1096 	blk_mq_req_flags_t blk_flags = 0;
1097 
1098 	if (flags & NVME_SUBMIT_NOWAIT)
1099 		blk_flags |= BLK_MQ_REQ_NOWAIT;
1100 	if (flags & NVME_SUBMIT_RESERVED)
1101 		blk_flags |= BLK_MQ_REQ_RESERVED;
1102 	if (qid == NVME_QID_ANY)
1103 		req = blk_mq_alloc_request(q, nvme_req_op(cmd), blk_flags);
1104 	else
1105 		req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), blk_flags,
1106 						qid - 1);
1107 
1108 	if (IS_ERR(req))
1109 		return PTR_ERR(req);
1110 	nvme_init_request(req, cmd);
1111 	if (flags & NVME_SUBMIT_RETRY)
1112 		req->cmd_flags &= ~REQ_FAILFAST_DRIVER;
1113 
1114 	if (buffer && bufflen) {
1115 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1116 		if (ret)
1117 			goto out;
1118 	}
1119 
1120 	ret = nvme_execute_rq(req, flags & NVME_SUBMIT_AT_HEAD);
1121 	if (result && ret >= 0)
1122 		*result = nvme_req(req)->result;
1123  out:
1124 	blk_mq_free_request(req);
1125 	return ret;
1126 }
1127 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1128 
1129 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1130 		void *buffer, unsigned bufflen)
1131 {
1132 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1133 			NVME_QID_ANY, 0);
1134 }
1135 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1136 
1137 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1138 {
1139 	u32 effects = 0;
1140 
1141 	if (ns) {
1142 		effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1143 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1144 			dev_warn_once(ctrl->device,
1145 				"IO command:%02x has unusual effects:%08x\n",
1146 				opcode, effects);
1147 
1148 		/*
1149 		 * NVME_CMD_EFFECTS_CSE_MASK causes a freeze all I/O queues,
1150 		 * which would deadlock when done on an I/O command.  Note that
1151 		 * We already warn about an unusual effect above.
1152 		 */
1153 		effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1154 	} else {
1155 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1156 
1157 		/* Ignore execution restrictions if any relaxation bits are set */
1158 		if (effects & NVME_CMD_EFFECTS_CSER_MASK)
1159 			effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1160 	}
1161 
1162 	return effects;
1163 }
1164 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1165 
1166 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1167 {
1168 	u32 effects = nvme_command_effects(ctrl, ns, opcode);
1169 
1170 	/*
1171 	 * For simplicity, IO to all namespaces is quiesced even if the command
1172 	 * effects say only one namespace is affected.
1173 	 */
1174 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1175 		mutex_lock(&ctrl->scan_lock);
1176 		mutex_lock(&ctrl->subsys->lock);
1177 		nvme_mpath_start_freeze(ctrl->subsys);
1178 		nvme_mpath_wait_freeze(ctrl->subsys);
1179 		nvme_start_freeze(ctrl);
1180 		nvme_wait_freeze(ctrl);
1181 	}
1182 	return effects;
1183 }
1184 EXPORT_SYMBOL_NS_GPL(nvme_passthru_start, NVME_TARGET_PASSTHRU);
1185 
1186 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1187 		       struct nvme_command *cmd, int status)
1188 {
1189 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1190 		nvme_unfreeze(ctrl);
1191 		nvme_mpath_unfreeze(ctrl->subsys);
1192 		mutex_unlock(&ctrl->subsys->lock);
1193 		mutex_unlock(&ctrl->scan_lock);
1194 	}
1195 	if (effects & NVME_CMD_EFFECTS_CCC) {
1196 		if (!test_and_set_bit(NVME_CTRL_DIRTY_CAPABILITY,
1197 				      &ctrl->flags)) {
1198 			dev_info(ctrl->device,
1199 "controller capabilities changed, reset may be required to take effect.\n");
1200 		}
1201 	}
1202 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1203 		nvme_queue_scan(ctrl);
1204 		flush_work(&ctrl->scan_work);
1205 	}
1206 	if (ns)
1207 		return;
1208 
1209 	switch (cmd->common.opcode) {
1210 	case nvme_admin_set_features:
1211 		switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1212 		case NVME_FEAT_KATO:
1213 			/*
1214 			 * Keep alive commands interval on the host should be
1215 			 * updated when KATO is modified by Set Features
1216 			 * commands.
1217 			 */
1218 			if (!status)
1219 				nvme_update_keep_alive(ctrl, cmd);
1220 			break;
1221 		default:
1222 			break;
1223 		}
1224 		break;
1225 	default:
1226 		break;
1227 	}
1228 }
1229 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1230 
1231 /*
1232  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1233  *
1234  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1235  *   accounting for transport roundtrip times [..].
1236  */
1237 static unsigned long nvme_keep_alive_work_period(struct nvme_ctrl *ctrl)
1238 {
1239 	unsigned long delay = ctrl->kato * HZ / 2;
1240 
1241 	/*
1242 	 * When using Traffic Based Keep Alive, we need to run
1243 	 * nvme_keep_alive_work at twice the normal frequency, as one
1244 	 * command completion can postpone sending a keep alive command
1245 	 * by up to twice the delay between runs.
1246 	 */
1247 	if (ctrl->ctratt & NVME_CTRL_ATTR_TBKAS)
1248 		delay /= 2;
1249 	return delay;
1250 }
1251 
1252 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1253 {
1254 	unsigned long now = jiffies;
1255 	unsigned long delay = nvme_keep_alive_work_period(ctrl);
1256 	unsigned long ka_next_check_tm = ctrl->ka_last_check_time + delay;
1257 
1258 	if (time_after(now, ka_next_check_tm))
1259 		delay = 0;
1260 	else
1261 		delay = ka_next_check_tm - now;
1262 
1263 	queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1264 }
1265 
1266 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1267 						 blk_status_t status)
1268 {
1269 	struct nvme_ctrl *ctrl = rq->end_io_data;
1270 	unsigned long flags;
1271 	bool startka = false;
1272 	unsigned long rtt = jiffies - (rq->deadline - rq->timeout);
1273 	unsigned long delay = nvme_keep_alive_work_period(ctrl);
1274 
1275 	/*
1276 	 * Subtract off the keepalive RTT so nvme_keep_alive_work runs
1277 	 * at the desired frequency.
1278 	 */
1279 	if (rtt <= delay) {
1280 		delay -= rtt;
1281 	} else {
1282 		dev_warn(ctrl->device, "long keepalive RTT (%u ms)\n",
1283 			 jiffies_to_msecs(rtt));
1284 		delay = 0;
1285 	}
1286 
1287 	blk_mq_free_request(rq);
1288 
1289 	if (status) {
1290 		dev_err(ctrl->device,
1291 			"failed nvme_keep_alive_end_io error=%d\n",
1292 				status);
1293 		return RQ_END_IO_NONE;
1294 	}
1295 
1296 	ctrl->ka_last_check_time = jiffies;
1297 	ctrl->comp_seen = false;
1298 	spin_lock_irqsave(&ctrl->lock, flags);
1299 	if (ctrl->state == NVME_CTRL_LIVE ||
1300 	    ctrl->state == NVME_CTRL_CONNECTING)
1301 		startka = true;
1302 	spin_unlock_irqrestore(&ctrl->lock, flags);
1303 	if (startka)
1304 		queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1305 	return RQ_END_IO_NONE;
1306 }
1307 
1308 static void nvme_keep_alive_work(struct work_struct *work)
1309 {
1310 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1311 			struct nvme_ctrl, ka_work);
1312 	bool comp_seen = ctrl->comp_seen;
1313 	struct request *rq;
1314 
1315 	ctrl->ka_last_check_time = jiffies;
1316 
1317 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1318 		dev_dbg(ctrl->device,
1319 			"reschedule traffic based keep-alive timer\n");
1320 		ctrl->comp_seen = false;
1321 		nvme_queue_keep_alive_work(ctrl);
1322 		return;
1323 	}
1324 
1325 	rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1326 				  BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1327 	if (IS_ERR(rq)) {
1328 		/* allocation failure, reset the controller */
1329 		dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1330 		nvme_reset_ctrl(ctrl);
1331 		return;
1332 	}
1333 	nvme_init_request(rq, &ctrl->ka_cmd);
1334 
1335 	rq->timeout = ctrl->kato * HZ;
1336 	rq->end_io = nvme_keep_alive_end_io;
1337 	rq->end_io_data = ctrl;
1338 	blk_execute_rq_nowait(rq, false);
1339 }
1340 
1341 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1342 {
1343 	if (unlikely(ctrl->kato == 0))
1344 		return;
1345 
1346 	nvme_queue_keep_alive_work(ctrl);
1347 }
1348 
1349 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1350 {
1351 	if (unlikely(ctrl->kato == 0))
1352 		return;
1353 
1354 	cancel_delayed_work_sync(&ctrl->ka_work);
1355 }
1356 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1357 
1358 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1359 				   struct nvme_command *cmd)
1360 {
1361 	unsigned int new_kato =
1362 		DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1363 
1364 	dev_info(ctrl->device,
1365 		 "keep alive interval updated from %u ms to %u ms\n",
1366 		 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1367 
1368 	nvme_stop_keep_alive(ctrl);
1369 	ctrl->kato = new_kato;
1370 	nvme_start_keep_alive(ctrl);
1371 }
1372 
1373 /*
1374  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1375  * flag, thus sending any new CNS opcodes has a big chance of not working.
1376  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1377  * (but not for any later version).
1378  */
1379 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1380 {
1381 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1382 		return ctrl->vs < NVME_VS(1, 2, 0);
1383 	return ctrl->vs < NVME_VS(1, 1, 0);
1384 }
1385 
1386 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1387 {
1388 	struct nvme_command c = { };
1389 	int error;
1390 
1391 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1392 	c.identify.opcode = nvme_admin_identify;
1393 	c.identify.cns = NVME_ID_CNS_CTRL;
1394 
1395 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1396 	if (!*id)
1397 		return -ENOMEM;
1398 
1399 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1400 			sizeof(struct nvme_id_ctrl));
1401 	if (error)
1402 		kfree(*id);
1403 	return error;
1404 }
1405 
1406 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1407 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1408 {
1409 	const char *warn_str = "ctrl returned bogus length:";
1410 	void *data = cur;
1411 
1412 	switch (cur->nidt) {
1413 	case NVME_NIDT_EUI64:
1414 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1415 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1416 				 warn_str, cur->nidl);
1417 			return -1;
1418 		}
1419 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1420 			return NVME_NIDT_EUI64_LEN;
1421 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1422 		return NVME_NIDT_EUI64_LEN;
1423 	case NVME_NIDT_NGUID:
1424 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1425 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1426 				 warn_str, cur->nidl);
1427 			return -1;
1428 		}
1429 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1430 			return NVME_NIDT_NGUID_LEN;
1431 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1432 		return NVME_NIDT_NGUID_LEN;
1433 	case NVME_NIDT_UUID:
1434 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1435 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1436 				 warn_str, cur->nidl);
1437 			return -1;
1438 		}
1439 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1440 			return NVME_NIDT_UUID_LEN;
1441 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1442 		return NVME_NIDT_UUID_LEN;
1443 	case NVME_NIDT_CSI:
1444 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1445 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1446 				 warn_str, cur->nidl);
1447 			return -1;
1448 		}
1449 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1450 		*csi_seen = true;
1451 		return NVME_NIDT_CSI_LEN;
1452 	default:
1453 		/* Skip unknown types */
1454 		return cur->nidl;
1455 	}
1456 }
1457 
1458 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1459 		struct nvme_ns_info *info)
1460 {
1461 	struct nvme_command c = { };
1462 	bool csi_seen = false;
1463 	int status, pos, len;
1464 	void *data;
1465 
1466 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1467 		return 0;
1468 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1469 		return 0;
1470 
1471 	c.identify.opcode = nvme_admin_identify;
1472 	c.identify.nsid = cpu_to_le32(info->nsid);
1473 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1474 
1475 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1476 	if (!data)
1477 		return -ENOMEM;
1478 
1479 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1480 				      NVME_IDENTIFY_DATA_SIZE);
1481 	if (status) {
1482 		dev_warn(ctrl->device,
1483 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1484 			info->nsid, status);
1485 		goto free_data;
1486 	}
1487 
1488 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1489 		struct nvme_ns_id_desc *cur = data + pos;
1490 
1491 		if (cur->nidl == 0)
1492 			break;
1493 
1494 		len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1495 		if (len < 0)
1496 			break;
1497 
1498 		len += sizeof(*cur);
1499 	}
1500 
1501 	if (nvme_multi_css(ctrl) && !csi_seen) {
1502 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1503 			 info->nsid);
1504 		status = -EINVAL;
1505 	}
1506 
1507 free_data:
1508 	kfree(data);
1509 	return status;
1510 }
1511 
1512 int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1513 			struct nvme_id_ns **id)
1514 {
1515 	struct nvme_command c = { };
1516 	int error;
1517 
1518 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1519 	c.identify.opcode = nvme_admin_identify;
1520 	c.identify.nsid = cpu_to_le32(nsid);
1521 	c.identify.cns = NVME_ID_CNS_NS;
1522 
1523 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1524 	if (!*id)
1525 		return -ENOMEM;
1526 
1527 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1528 	if (error) {
1529 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1530 		kfree(*id);
1531 	}
1532 	return error;
1533 }
1534 
1535 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1536 		struct nvme_ns_info *info)
1537 {
1538 	struct nvme_ns_ids *ids = &info->ids;
1539 	struct nvme_id_ns *id;
1540 	int ret;
1541 
1542 	ret = nvme_identify_ns(ctrl, info->nsid, &id);
1543 	if (ret)
1544 		return ret;
1545 
1546 	if (id->ncap == 0) {
1547 		/* namespace not allocated or attached */
1548 		info->is_removed = true;
1549 		ret = -ENODEV;
1550 		goto error;
1551 	}
1552 
1553 	info->anagrpid = id->anagrpid;
1554 	info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1555 	info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1556 	info->is_ready = true;
1557 	if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1558 		dev_info(ctrl->device,
1559 			 "Ignoring bogus Namespace Identifiers\n");
1560 	} else {
1561 		if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1562 		    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1563 			memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1564 		if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1565 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1566 			memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1567 	}
1568 
1569 error:
1570 	kfree(id);
1571 	return ret;
1572 }
1573 
1574 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1575 		struct nvme_ns_info *info)
1576 {
1577 	struct nvme_id_ns_cs_indep *id;
1578 	struct nvme_command c = {
1579 		.identify.opcode	= nvme_admin_identify,
1580 		.identify.nsid		= cpu_to_le32(info->nsid),
1581 		.identify.cns		= NVME_ID_CNS_NS_CS_INDEP,
1582 	};
1583 	int ret;
1584 
1585 	id = kmalloc(sizeof(*id), GFP_KERNEL);
1586 	if (!id)
1587 		return -ENOMEM;
1588 
1589 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1590 	if (!ret) {
1591 		info->anagrpid = id->anagrpid;
1592 		info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1593 		info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1594 		info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1595 	}
1596 	kfree(id);
1597 	return ret;
1598 }
1599 
1600 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1601 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1602 {
1603 	union nvme_result res = { 0 };
1604 	struct nvme_command c = { };
1605 	int ret;
1606 
1607 	c.features.opcode = op;
1608 	c.features.fid = cpu_to_le32(fid);
1609 	c.features.dword11 = cpu_to_le32(dword11);
1610 
1611 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1612 			buffer, buflen, NVME_QID_ANY, 0);
1613 	if (ret >= 0 && result)
1614 		*result = le32_to_cpu(res.u32);
1615 	return ret;
1616 }
1617 
1618 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1619 		      unsigned int dword11, void *buffer, size_t buflen,
1620 		      u32 *result)
1621 {
1622 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1623 			     buflen, result);
1624 }
1625 EXPORT_SYMBOL_GPL(nvme_set_features);
1626 
1627 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1628 		      unsigned int dword11, void *buffer, size_t buflen,
1629 		      u32 *result)
1630 {
1631 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1632 			     buflen, result);
1633 }
1634 EXPORT_SYMBOL_GPL(nvme_get_features);
1635 
1636 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1637 {
1638 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1639 	u32 result;
1640 	int status, nr_io_queues;
1641 
1642 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1643 			&result);
1644 	if (status < 0)
1645 		return status;
1646 
1647 	/*
1648 	 * Degraded controllers might return an error when setting the queue
1649 	 * count.  We still want to be able to bring them online and offer
1650 	 * access to the admin queue, as that might be only way to fix them up.
1651 	 */
1652 	if (status > 0) {
1653 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1654 		*count = 0;
1655 	} else {
1656 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1657 		*count = min(*count, nr_io_queues);
1658 	}
1659 
1660 	return 0;
1661 }
1662 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1663 
1664 #define NVME_AEN_SUPPORTED \
1665 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1666 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1667 
1668 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1669 {
1670 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1671 	int status;
1672 
1673 	if (!supported_aens)
1674 		return;
1675 
1676 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1677 			NULL, 0, &result);
1678 	if (status)
1679 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1680 			 supported_aens);
1681 
1682 	queue_work(nvme_wq, &ctrl->async_event_work);
1683 }
1684 
1685 static int nvme_ns_open(struct nvme_ns *ns)
1686 {
1687 
1688 	/* should never be called due to GENHD_FL_HIDDEN */
1689 	if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1690 		goto fail;
1691 	if (!nvme_get_ns(ns))
1692 		goto fail;
1693 	if (!try_module_get(ns->ctrl->ops->module))
1694 		goto fail_put_ns;
1695 
1696 	return 0;
1697 
1698 fail_put_ns:
1699 	nvme_put_ns(ns);
1700 fail:
1701 	return -ENXIO;
1702 }
1703 
1704 static void nvme_ns_release(struct nvme_ns *ns)
1705 {
1706 
1707 	module_put(ns->ctrl->ops->module);
1708 	nvme_put_ns(ns);
1709 }
1710 
1711 static int nvme_open(struct gendisk *disk, blk_mode_t mode)
1712 {
1713 	return nvme_ns_open(disk->private_data);
1714 }
1715 
1716 static void nvme_release(struct gendisk *disk)
1717 {
1718 	nvme_ns_release(disk->private_data);
1719 }
1720 
1721 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1722 {
1723 	/* some standard values */
1724 	geo->heads = 1 << 6;
1725 	geo->sectors = 1 << 5;
1726 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1727 	return 0;
1728 }
1729 
1730 #ifdef CONFIG_BLK_DEV_INTEGRITY
1731 static void nvme_init_integrity(struct gendisk *disk,
1732 		struct nvme_ns_head *head, u32 max_integrity_segments)
1733 {
1734 	struct blk_integrity integrity = { };
1735 
1736 	switch (head->pi_type) {
1737 	case NVME_NS_DPS_PI_TYPE3:
1738 		switch (head->guard_type) {
1739 		case NVME_NVM_NS_16B_GUARD:
1740 			integrity.profile = &t10_pi_type3_crc;
1741 			integrity.tag_size = sizeof(u16) + sizeof(u32);
1742 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1743 			break;
1744 		case NVME_NVM_NS_64B_GUARD:
1745 			integrity.profile = &ext_pi_type3_crc64;
1746 			integrity.tag_size = sizeof(u16) + 6;
1747 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1748 			break;
1749 		default:
1750 			integrity.profile = NULL;
1751 			break;
1752 		}
1753 		break;
1754 	case NVME_NS_DPS_PI_TYPE1:
1755 	case NVME_NS_DPS_PI_TYPE2:
1756 		switch (head->guard_type) {
1757 		case NVME_NVM_NS_16B_GUARD:
1758 			integrity.profile = &t10_pi_type1_crc;
1759 			integrity.tag_size = sizeof(u16);
1760 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1761 			break;
1762 		case NVME_NVM_NS_64B_GUARD:
1763 			integrity.profile = &ext_pi_type1_crc64;
1764 			integrity.tag_size = sizeof(u16);
1765 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1766 			break;
1767 		default:
1768 			integrity.profile = NULL;
1769 			break;
1770 		}
1771 		break;
1772 	default:
1773 		integrity.profile = NULL;
1774 		break;
1775 	}
1776 
1777 	integrity.tuple_size = head->ms;
1778 	blk_integrity_register(disk, &integrity);
1779 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1780 }
1781 #else
1782 static void nvme_init_integrity(struct gendisk *disk,
1783 		struct nvme_ns_head *head, u32 max_integrity_segments)
1784 {
1785 }
1786 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1787 
1788 static void nvme_config_discard(struct nvme_ctrl *ctrl, struct gendisk *disk,
1789 		struct nvme_ns_head *head)
1790 {
1791 	struct request_queue *queue = disk->queue;
1792 	u32 max_discard_sectors;
1793 
1794 	if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(head, UINT_MAX)) {
1795 		max_discard_sectors = nvme_lba_to_sect(head, ctrl->dmrsl);
1796 	} else if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
1797 		max_discard_sectors = UINT_MAX;
1798 	} else {
1799 		blk_queue_max_discard_sectors(queue, 0);
1800 		return;
1801 	}
1802 
1803 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1804 			NVME_DSM_MAX_RANGES);
1805 
1806 	/*
1807 	 * If discard is already enabled, don't reset queue limits.
1808 	 *
1809 	 * This works around the fact that the block layer can't cope well with
1810 	 * updating the hardware limits when overridden through sysfs.  This is
1811 	 * harmless because discard limits in NVMe are purely advisory.
1812 	 */
1813 	if (queue->limits.max_discard_sectors)
1814 		return;
1815 
1816 	blk_queue_max_discard_sectors(queue, max_discard_sectors);
1817 	if (ctrl->dmrl)
1818 		blk_queue_max_discard_segments(queue, ctrl->dmrl);
1819 	else
1820 		blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1821 	queue->limits.discard_granularity = queue_logical_block_size(queue);
1822 
1823 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1824 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1825 }
1826 
1827 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1828 {
1829 	return uuid_equal(&a->uuid, &b->uuid) &&
1830 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1831 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1832 		a->csi == b->csi;
1833 }
1834 
1835 static int nvme_init_ms(struct nvme_ctrl *ctrl, struct nvme_ns_head *head,
1836 		struct nvme_id_ns *id)
1837 {
1838 	bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1839 	unsigned lbaf = nvme_lbaf_index(id->flbas);
1840 	struct nvme_command c = { };
1841 	struct nvme_id_ns_nvm *nvm;
1842 	int ret = 0;
1843 	u32 elbaf;
1844 
1845 	head->pi_size = 0;
1846 	head->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1847 	if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1848 		head->pi_size = sizeof(struct t10_pi_tuple);
1849 		head->guard_type = NVME_NVM_NS_16B_GUARD;
1850 		goto set_pi;
1851 	}
1852 
1853 	nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1854 	if (!nvm)
1855 		return -ENOMEM;
1856 
1857 	c.identify.opcode = nvme_admin_identify;
1858 	c.identify.nsid = cpu_to_le32(head->ns_id);
1859 	c.identify.cns = NVME_ID_CNS_CS_NS;
1860 	c.identify.csi = NVME_CSI_NVM;
1861 
1862 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, nvm, sizeof(*nvm));
1863 	if (ret)
1864 		goto free_data;
1865 
1866 	elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1867 
1868 	/* no support for storage tag formats right now */
1869 	if (nvme_elbaf_sts(elbaf))
1870 		goto free_data;
1871 
1872 	head->guard_type = nvme_elbaf_guard_type(elbaf);
1873 	switch (head->guard_type) {
1874 	case NVME_NVM_NS_64B_GUARD:
1875 		head->pi_size = sizeof(struct crc64_pi_tuple);
1876 		break;
1877 	case NVME_NVM_NS_16B_GUARD:
1878 		head->pi_size = sizeof(struct t10_pi_tuple);
1879 		break;
1880 	default:
1881 		break;
1882 	}
1883 
1884 free_data:
1885 	kfree(nvm);
1886 set_pi:
1887 	if (head->pi_size && (first || head->ms == head->pi_size))
1888 		head->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1889 	else
1890 		head->pi_type = 0;
1891 
1892 	return ret;
1893 }
1894 
1895 static int nvme_configure_metadata(struct nvme_ctrl *ctrl,
1896 		struct nvme_ns_head *head, struct nvme_id_ns *id)
1897 {
1898 	int ret;
1899 
1900 	ret = nvme_init_ms(ctrl, head, id);
1901 	if (ret)
1902 		return ret;
1903 
1904 	head->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1905 	if (!head->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1906 		return 0;
1907 
1908 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1909 		/*
1910 		 * The NVMe over Fabrics specification only supports metadata as
1911 		 * part of the extended data LBA.  We rely on HCA/HBA support to
1912 		 * remap the separate metadata buffer from the block layer.
1913 		 */
1914 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1915 			return 0;
1916 
1917 		head->features |= NVME_NS_EXT_LBAS;
1918 
1919 		/*
1920 		 * The current fabrics transport drivers support namespace
1921 		 * metadata formats only if nvme_ns_has_pi() returns true.
1922 		 * Suppress support for all other formats so the namespace will
1923 		 * have a 0 capacity and not be usable through the block stack.
1924 		 *
1925 		 * Note, this check will need to be modified if any drivers
1926 		 * gain the ability to use other metadata formats.
1927 		 */
1928 		if (ctrl->max_integrity_segments && nvme_ns_has_pi(head))
1929 			head->features |= NVME_NS_METADATA_SUPPORTED;
1930 	} else {
1931 		/*
1932 		 * For PCIe controllers, we can't easily remap the separate
1933 		 * metadata buffer from the block layer and thus require a
1934 		 * separate metadata buffer for block layer metadata/PI support.
1935 		 * We allow extended LBAs for the passthrough interface, though.
1936 		 */
1937 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
1938 			head->features |= NVME_NS_EXT_LBAS;
1939 		else
1940 			head->features |= NVME_NS_METADATA_SUPPORTED;
1941 	}
1942 	return 0;
1943 }
1944 
1945 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1946 		struct request_queue *q)
1947 {
1948 	bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1949 
1950 	if (ctrl->max_hw_sectors) {
1951 		u32 max_segments =
1952 			(ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1953 
1954 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
1955 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1956 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1957 	}
1958 	blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1959 	blk_queue_dma_alignment(q, 3);
1960 	blk_queue_write_cache(q, vwc, vwc);
1961 }
1962 
1963 static void nvme_update_disk_info(struct nvme_ctrl *ctrl, struct gendisk *disk,
1964 		struct nvme_ns_head *head, struct nvme_id_ns *id)
1965 {
1966 	sector_t capacity = nvme_lba_to_sect(head, le64_to_cpu(id->nsze));
1967 	u32 bs = 1U << head->lba_shift;
1968 	u32 atomic_bs, phys_bs, io_opt = 0;
1969 
1970 	/*
1971 	 * The block layer can't support LBA sizes larger than the page size
1972 	 * or smaller than a sector size yet, so catch this early and don't
1973 	 * allow block I/O.
1974 	 */
1975 	if (head->lba_shift > PAGE_SHIFT || head->lba_shift < SECTOR_SHIFT) {
1976 		capacity = 0;
1977 		bs = (1 << 9);
1978 	}
1979 
1980 	blk_integrity_unregister(disk);
1981 
1982 	atomic_bs = phys_bs = bs;
1983 	if (id->nabo == 0) {
1984 		/*
1985 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1986 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1987 		 * 0 then AWUPF must be used instead.
1988 		 */
1989 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1990 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1991 		else
1992 			atomic_bs = (1 + ctrl->subsys->awupf) * bs;
1993 	}
1994 
1995 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1996 		/* NPWG = Namespace Preferred Write Granularity */
1997 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1998 		/* NOWS = Namespace Optimal Write Size */
1999 		io_opt = bs * (1 + le16_to_cpu(id->nows));
2000 	}
2001 
2002 	blk_queue_logical_block_size(disk->queue, bs);
2003 	/*
2004 	 * Linux filesystems assume writing a single physical block is
2005 	 * an atomic operation. Hence limit the physical block size to the
2006 	 * value of the Atomic Write Unit Power Fail parameter.
2007 	 */
2008 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2009 	blk_queue_io_min(disk->queue, phys_bs);
2010 	blk_queue_io_opt(disk->queue, io_opt);
2011 
2012 	/*
2013 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
2014 	 * metadata masquerading as Type 0 if supported, otherwise reject block
2015 	 * I/O to namespaces with metadata except when the namespace supports
2016 	 * PI, as it can strip/insert in that case.
2017 	 */
2018 	if (head->ms) {
2019 		if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2020 		    (head->features & NVME_NS_METADATA_SUPPORTED))
2021 			nvme_init_integrity(disk, head,
2022 					    ctrl->max_integrity_segments);
2023 		else if (!nvme_ns_has_pi(head))
2024 			capacity = 0;
2025 	}
2026 
2027 	set_capacity_and_notify(disk, capacity);
2028 
2029 	nvme_config_discard(ctrl, disk, head);
2030 	blk_queue_max_write_zeroes_sectors(disk->queue,
2031 					   ctrl->max_zeroes_sectors);
2032 }
2033 
2034 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
2035 {
2036 	return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
2037 }
2038 
2039 static inline bool nvme_first_scan(struct gendisk *disk)
2040 {
2041 	/* nvme_alloc_ns() scans the disk prior to adding it */
2042 	return !disk_live(disk);
2043 }
2044 
2045 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2046 {
2047 	struct nvme_ctrl *ctrl = ns->ctrl;
2048 	u32 iob;
2049 
2050 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2051 	    is_power_of_2(ctrl->max_hw_sectors))
2052 		iob = ctrl->max_hw_sectors;
2053 	else
2054 		iob = nvme_lba_to_sect(ns->head, le16_to_cpu(id->noiob));
2055 
2056 	if (!iob)
2057 		return;
2058 
2059 	if (!is_power_of_2(iob)) {
2060 		if (nvme_first_scan(ns->disk))
2061 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2062 				ns->disk->disk_name, iob);
2063 		return;
2064 	}
2065 
2066 	if (blk_queue_is_zoned(ns->disk->queue)) {
2067 		if (nvme_first_scan(ns->disk))
2068 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
2069 				ns->disk->disk_name);
2070 		return;
2071 	}
2072 
2073 	blk_queue_chunk_sectors(ns->queue, iob);
2074 }
2075 
2076 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
2077 		struct nvme_ns_info *info)
2078 {
2079 	blk_mq_freeze_queue(ns->disk->queue);
2080 	nvme_set_queue_limits(ns->ctrl, ns->queue);
2081 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2082 	blk_mq_unfreeze_queue(ns->disk->queue);
2083 
2084 	if (nvme_ns_head_multipath(ns->head)) {
2085 		blk_mq_freeze_queue(ns->head->disk->queue);
2086 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2087 		nvme_mpath_revalidate_paths(ns);
2088 		blk_stack_limits(&ns->head->disk->queue->limits,
2089 				 &ns->queue->limits, 0);
2090 		ns->head->disk->flags |= GENHD_FL_HIDDEN;
2091 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2092 	}
2093 
2094 	/* Hide the block-interface for these devices */
2095 	ns->disk->flags |= GENHD_FL_HIDDEN;
2096 	set_bit(NVME_NS_READY, &ns->flags);
2097 
2098 	return 0;
2099 }
2100 
2101 static int nvme_update_ns_info_block(struct nvme_ns *ns,
2102 		struct nvme_ns_info *info)
2103 {
2104 	struct nvme_id_ns *id;
2105 	unsigned lbaf;
2106 	int ret;
2107 
2108 	ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
2109 	if (ret)
2110 		return ret;
2111 
2112 	if (id->ncap == 0) {
2113 		/* namespace not allocated or attached */
2114 		info->is_removed = true;
2115 		ret = -ENODEV;
2116 		goto error;
2117 	}
2118 
2119 	blk_mq_freeze_queue(ns->disk->queue);
2120 	lbaf = nvme_lbaf_index(id->flbas);
2121 	ns->head->lba_shift = id->lbaf[lbaf].ds;
2122 	ns->head->nuse = le64_to_cpu(id->nuse);
2123 	nvme_set_queue_limits(ns->ctrl, ns->queue);
2124 
2125 	ret = nvme_configure_metadata(ns->ctrl, ns->head, id);
2126 	if (ret < 0) {
2127 		blk_mq_unfreeze_queue(ns->disk->queue);
2128 		goto out;
2129 	}
2130 	nvme_set_chunk_sectors(ns, id);
2131 	nvme_update_disk_info(ns->ctrl, ns->disk, ns->head, id);
2132 
2133 	if (ns->head->ids.csi == NVME_CSI_ZNS) {
2134 		ret = nvme_update_zone_info(ns, lbaf);
2135 		if (ret) {
2136 			blk_mq_unfreeze_queue(ns->disk->queue);
2137 			goto out;
2138 		}
2139 	}
2140 
2141 	/*
2142 	 * Only set the DEAC bit if the device guarantees that reads from
2143 	 * deallocated data return zeroes.  While the DEAC bit does not
2144 	 * require that, it must be a no-op if reads from deallocated data
2145 	 * do not return zeroes.
2146 	 */
2147 	if ((id->dlfeat & 0x7) == 0x1 && (id->dlfeat & (1 << 3)))
2148 		ns->head->features |= NVME_NS_DEAC;
2149 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2150 	set_bit(NVME_NS_READY, &ns->flags);
2151 	blk_mq_unfreeze_queue(ns->disk->queue);
2152 
2153 	if (blk_queue_is_zoned(ns->queue)) {
2154 		ret = nvme_revalidate_zones(ns);
2155 		if (ret && !nvme_first_scan(ns->disk))
2156 			goto out;
2157 	}
2158 
2159 	if (nvme_ns_head_multipath(ns->head)) {
2160 		blk_mq_freeze_queue(ns->head->disk->queue);
2161 		nvme_update_disk_info(ns->ctrl, ns->head->disk, ns->head, id);
2162 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2163 		nvme_mpath_revalidate_paths(ns);
2164 		blk_stack_limits(&ns->head->disk->queue->limits,
2165 				 &ns->queue->limits, 0);
2166 		disk_update_readahead(ns->head->disk);
2167 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2168 	}
2169 
2170 	ret = 0;
2171 out:
2172 	/*
2173 	 * If probing fails due an unsupported feature, hide the block device,
2174 	 * but still allow other access.
2175 	 */
2176 	if (ret == -ENODEV) {
2177 		ns->disk->flags |= GENHD_FL_HIDDEN;
2178 		set_bit(NVME_NS_READY, &ns->flags);
2179 		ret = 0;
2180 	}
2181 
2182 error:
2183 	kfree(id);
2184 	return ret;
2185 }
2186 
2187 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2188 {
2189 	switch (info->ids.csi) {
2190 	case NVME_CSI_ZNS:
2191 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2192 			dev_info(ns->ctrl->device,
2193 	"block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2194 				info->nsid);
2195 			return nvme_update_ns_info_generic(ns, info);
2196 		}
2197 		return nvme_update_ns_info_block(ns, info);
2198 	case NVME_CSI_NVM:
2199 		return nvme_update_ns_info_block(ns, info);
2200 	default:
2201 		dev_info(ns->ctrl->device,
2202 			"block device for nsid %u not supported (csi %u)\n",
2203 			info->nsid, info->ids.csi);
2204 		return nvme_update_ns_info_generic(ns, info);
2205 	}
2206 }
2207 
2208 #ifdef CONFIG_BLK_SED_OPAL
2209 static int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2210 		bool send)
2211 {
2212 	struct nvme_ctrl *ctrl = data;
2213 	struct nvme_command cmd = { };
2214 
2215 	if (send)
2216 		cmd.common.opcode = nvme_admin_security_send;
2217 	else
2218 		cmd.common.opcode = nvme_admin_security_recv;
2219 	cmd.common.nsid = 0;
2220 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2221 	cmd.common.cdw11 = cpu_to_le32(len);
2222 
2223 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2224 			NVME_QID_ANY, NVME_SUBMIT_AT_HEAD);
2225 }
2226 
2227 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2228 {
2229 	if (ctrl->oacs & NVME_CTRL_OACS_SEC_SUPP) {
2230 		if (!ctrl->opal_dev)
2231 			ctrl->opal_dev = init_opal_dev(ctrl, &nvme_sec_submit);
2232 		else if (was_suspended)
2233 			opal_unlock_from_suspend(ctrl->opal_dev);
2234 	} else {
2235 		free_opal_dev(ctrl->opal_dev);
2236 		ctrl->opal_dev = NULL;
2237 	}
2238 }
2239 #else
2240 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2241 {
2242 }
2243 #endif /* CONFIG_BLK_SED_OPAL */
2244 
2245 #ifdef CONFIG_BLK_DEV_ZONED
2246 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2247 		unsigned int nr_zones, report_zones_cb cb, void *data)
2248 {
2249 	return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2250 			data);
2251 }
2252 #else
2253 #define nvme_report_zones	NULL
2254 #endif /* CONFIG_BLK_DEV_ZONED */
2255 
2256 const struct block_device_operations nvme_bdev_ops = {
2257 	.owner		= THIS_MODULE,
2258 	.ioctl		= nvme_ioctl,
2259 	.compat_ioctl	= blkdev_compat_ptr_ioctl,
2260 	.open		= nvme_open,
2261 	.release	= nvme_release,
2262 	.getgeo		= nvme_getgeo,
2263 	.report_zones	= nvme_report_zones,
2264 	.pr_ops		= &nvme_pr_ops,
2265 };
2266 
2267 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 mask, u32 val,
2268 		u32 timeout, const char *op)
2269 {
2270 	unsigned long timeout_jiffies = jiffies + timeout * HZ;
2271 	u32 csts;
2272 	int ret;
2273 
2274 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2275 		if (csts == ~0)
2276 			return -ENODEV;
2277 		if ((csts & mask) == val)
2278 			break;
2279 
2280 		usleep_range(1000, 2000);
2281 		if (fatal_signal_pending(current))
2282 			return -EINTR;
2283 		if (time_after(jiffies, timeout_jiffies)) {
2284 			dev_err(ctrl->device,
2285 				"Device not ready; aborting %s, CSTS=0x%x\n",
2286 				op, csts);
2287 			return -ENODEV;
2288 		}
2289 	}
2290 
2291 	return ret;
2292 }
2293 
2294 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown)
2295 {
2296 	int ret;
2297 
2298 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2299 	if (shutdown)
2300 		ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2301 	else
2302 		ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2303 
2304 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2305 	if (ret)
2306 		return ret;
2307 
2308 	if (shutdown) {
2309 		return nvme_wait_ready(ctrl, NVME_CSTS_SHST_MASK,
2310 				       NVME_CSTS_SHST_CMPLT,
2311 				       ctrl->shutdown_timeout, "shutdown");
2312 	}
2313 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2314 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2315 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, 0,
2316 			       (NVME_CAP_TIMEOUT(ctrl->cap) + 1) / 2, "reset");
2317 }
2318 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2319 
2320 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2321 {
2322 	unsigned dev_page_min;
2323 	u32 timeout;
2324 	int ret;
2325 
2326 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2327 	if (ret) {
2328 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2329 		return ret;
2330 	}
2331 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2332 
2333 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2334 		dev_err(ctrl->device,
2335 			"Minimum device page size %u too large for host (%u)\n",
2336 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2337 		return -ENODEV;
2338 	}
2339 
2340 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2341 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2342 	else
2343 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2344 
2345 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS && ctrl->cap & NVME_CAP_CRMS_CRIMS)
2346 		ctrl->ctrl_config |= NVME_CC_CRIME;
2347 
2348 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2349 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2350 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2351 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2352 	if (ret)
2353 		return ret;
2354 
2355 	/* Flush write to device (required if transport is PCI) */
2356 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2357 	if (ret)
2358 		return ret;
2359 
2360 	/* CAP value may change after initial CC write */
2361 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2362 	if (ret)
2363 		return ret;
2364 
2365 	timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2366 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2367 		u32 crto, ready_timeout;
2368 
2369 		ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2370 		if (ret) {
2371 			dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2372 				ret);
2373 			return ret;
2374 		}
2375 
2376 		/*
2377 		 * CRTO should always be greater or equal to CAP.TO, but some
2378 		 * devices are known to get this wrong. Use the larger of the
2379 		 * two values.
2380 		 */
2381 		if (ctrl->ctrl_config & NVME_CC_CRIME)
2382 			ready_timeout = NVME_CRTO_CRIMT(crto);
2383 		else
2384 			ready_timeout = NVME_CRTO_CRWMT(crto);
2385 
2386 		if (ready_timeout < timeout)
2387 			dev_warn_once(ctrl->device, "bad crto:%x cap:%llx\n",
2388 				      crto, ctrl->cap);
2389 		else
2390 			timeout = ready_timeout;
2391 	}
2392 
2393 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2394 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2395 	if (ret)
2396 		return ret;
2397 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, NVME_CSTS_RDY,
2398 			       (timeout + 1) / 2, "initialisation");
2399 }
2400 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2401 
2402 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2403 {
2404 	__le64 ts;
2405 	int ret;
2406 
2407 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2408 		return 0;
2409 
2410 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2411 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2412 			NULL);
2413 	if (ret)
2414 		dev_warn_once(ctrl->device,
2415 			"could not set timestamp (%d)\n", ret);
2416 	return ret;
2417 }
2418 
2419 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2420 {
2421 	struct nvme_feat_host_behavior *host;
2422 	u8 acre = 0, lbafee = 0;
2423 	int ret;
2424 
2425 	/* Don't bother enabling the feature if retry delay is not reported */
2426 	if (ctrl->crdt[0])
2427 		acre = NVME_ENABLE_ACRE;
2428 	if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2429 		lbafee = NVME_ENABLE_LBAFEE;
2430 
2431 	if (!acre && !lbafee)
2432 		return 0;
2433 
2434 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2435 	if (!host)
2436 		return 0;
2437 
2438 	host->acre = acre;
2439 	host->lbafee = lbafee;
2440 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2441 				host, sizeof(*host), NULL);
2442 	kfree(host);
2443 	return ret;
2444 }
2445 
2446 /*
2447  * The function checks whether the given total (exlat + enlat) latency of
2448  * a power state allows the latter to be used as an APST transition target.
2449  * It does so by comparing the latency to the primary and secondary latency
2450  * tolerances defined by module params. If there's a match, the corresponding
2451  * timeout value is returned and the matching tolerance index (1 or 2) is
2452  * reported.
2453  */
2454 static bool nvme_apst_get_transition_time(u64 total_latency,
2455 		u64 *transition_time, unsigned *last_index)
2456 {
2457 	if (total_latency <= apst_primary_latency_tol_us) {
2458 		if (*last_index == 1)
2459 			return false;
2460 		*last_index = 1;
2461 		*transition_time = apst_primary_timeout_ms;
2462 		return true;
2463 	}
2464 	if (apst_secondary_timeout_ms &&
2465 		total_latency <= apst_secondary_latency_tol_us) {
2466 		if (*last_index <= 2)
2467 			return false;
2468 		*last_index = 2;
2469 		*transition_time = apst_secondary_timeout_ms;
2470 		return true;
2471 	}
2472 	return false;
2473 }
2474 
2475 /*
2476  * APST (Autonomous Power State Transition) lets us program a table of power
2477  * state transitions that the controller will perform automatically.
2478  *
2479  * Depending on module params, one of the two supported techniques will be used:
2480  *
2481  * - If the parameters provide explicit timeouts and tolerances, they will be
2482  *   used to build a table with up to 2 non-operational states to transition to.
2483  *   The default parameter values were selected based on the values used by
2484  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2485  *   regeneration of the APST table in the event of switching between external
2486  *   and battery power, the timeouts and tolerances reflect a compromise
2487  *   between values used by Microsoft for AC and battery scenarios.
2488  * - If not, we'll configure the table with a simple heuristic: we are willing
2489  *   to spend at most 2% of the time transitioning between power states.
2490  *   Therefore, when running in any given state, we will enter the next
2491  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2492  *   microseconds, as long as that state's exit latency is under the requested
2493  *   maximum latency.
2494  *
2495  * We will not autonomously enter any non-operational state for which the total
2496  * latency exceeds ps_max_latency_us.
2497  *
2498  * Users can set ps_max_latency_us to zero to turn off APST.
2499  */
2500 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2501 {
2502 	struct nvme_feat_auto_pst *table;
2503 	unsigned apste = 0;
2504 	u64 max_lat_us = 0;
2505 	__le64 target = 0;
2506 	int max_ps = -1;
2507 	int state;
2508 	int ret;
2509 	unsigned last_lt_index = UINT_MAX;
2510 
2511 	/*
2512 	 * If APST isn't supported or if we haven't been initialized yet,
2513 	 * then don't do anything.
2514 	 */
2515 	if (!ctrl->apsta)
2516 		return 0;
2517 
2518 	if (ctrl->npss > 31) {
2519 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2520 		return 0;
2521 	}
2522 
2523 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2524 	if (!table)
2525 		return 0;
2526 
2527 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2528 		/* Turn off APST. */
2529 		dev_dbg(ctrl->device, "APST disabled\n");
2530 		goto done;
2531 	}
2532 
2533 	/*
2534 	 * Walk through all states from lowest- to highest-power.
2535 	 * According to the spec, lower-numbered states use more power.  NPSS,
2536 	 * despite the name, is the index of the lowest-power state, not the
2537 	 * number of states.
2538 	 */
2539 	for (state = (int)ctrl->npss; state >= 0; state--) {
2540 		u64 total_latency_us, exit_latency_us, transition_ms;
2541 
2542 		if (target)
2543 			table->entries[state] = target;
2544 
2545 		/*
2546 		 * Don't allow transitions to the deepest state if it's quirked
2547 		 * off.
2548 		 */
2549 		if (state == ctrl->npss &&
2550 		    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2551 			continue;
2552 
2553 		/*
2554 		 * Is this state a useful non-operational state for higher-power
2555 		 * states to autonomously transition to?
2556 		 */
2557 		if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2558 			continue;
2559 
2560 		exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2561 		if (exit_latency_us > ctrl->ps_max_latency_us)
2562 			continue;
2563 
2564 		total_latency_us = exit_latency_us +
2565 			le32_to_cpu(ctrl->psd[state].entry_lat);
2566 
2567 		/*
2568 		 * This state is good. It can be used as the APST idle target
2569 		 * for higher power states.
2570 		 */
2571 		if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2572 			if (!nvme_apst_get_transition_time(total_latency_us,
2573 					&transition_ms, &last_lt_index))
2574 				continue;
2575 		} else {
2576 			transition_ms = total_latency_us + 19;
2577 			do_div(transition_ms, 20);
2578 			if (transition_ms > (1 << 24) - 1)
2579 				transition_ms = (1 << 24) - 1;
2580 		}
2581 
2582 		target = cpu_to_le64((state << 3) | (transition_ms << 8));
2583 		if (max_ps == -1)
2584 			max_ps = state;
2585 		if (total_latency_us > max_lat_us)
2586 			max_lat_us = total_latency_us;
2587 	}
2588 
2589 	if (max_ps == -1)
2590 		dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2591 	else
2592 		dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2593 			max_ps, max_lat_us, (int)sizeof(*table), table);
2594 	apste = 1;
2595 
2596 done:
2597 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2598 				table, sizeof(*table), NULL);
2599 	if (ret)
2600 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2601 	kfree(table);
2602 	return ret;
2603 }
2604 
2605 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2606 {
2607 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2608 	u64 latency;
2609 
2610 	switch (val) {
2611 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2612 	case PM_QOS_LATENCY_ANY:
2613 		latency = U64_MAX;
2614 		break;
2615 
2616 	default:
2617 		latency = val;
2618 	}
2619 
2620 	if (ctrl->ps_max_latency_us != latency) {
2621 		ctrl->ps_max_latency_us = latency;
2622 		if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
2623 			nvme_configure_apst(ctrl);
2624 	}
2625 }
2626 
2627 struct nvme_core_quirk_entry {
2628 	/*
2629 	 * NVMe model and firmware strings are padded with spaces.  For
2630 	 * simplicity, strings in the quirk table are padded with NULLs
2631 	 * instead.
2632 	 */
2633 	u16 vid;
2634 	const char *mn;
2635 	const char *fr;
2636 	unsigned long quirks;
2637 };
2638 
2639 static const struct nvme_core_quirk_entry core_quirks[] = {
2640 	{
2641 		/*
2642 		 * This Toshiba device seems to die using any APST states.  See:
2643 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2644 		 */
2645 		.vid = 0x1179,
2646 		.mn = "THNSF5256GPUK TOSHIBA",
2647 		.quirks = NVME_QUIRK_NO_APST,
2648 	},
2649 	{
2650 		/*
2651 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2652 		 * condition associated with actions related to suspend to idle
2653 		 * LiteON has resolved the problem in future firmware
2654 		 */
2655 		.vid = 0x14a4,
2656 		.fr = "22301111",
2657 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2658 	},
2659 	{
2660 		/*
2661 		 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2662 		 * aborts I/O during any load, but more easily reproducible
2663 		 * with discards (fstrim).
2664 		 *
2665 		 * The device is left in a state where it is also not possible
2666 		 * to use "nvme set-feature" to disable APST, but booting with
2667 		 * nvme_core.default_ps_max_latency=0 works.
2668 		 */
2669 		.vid = 0x1e0f,
2670 		.mn = "KCD6XVUL6T40",
2671 		.quirks = NVME_QUIRK_NO_APST,
2672 	},
2673 	{
2674 		/*
2675 		 * The external Samsung X5 SSD fails initialization without a
2676 		 * delay before checking if it is ready and has a whole set of
2677 		 * other problems.  To make this even more interesting, it
2678 		 * shares the PCI ID with internal Samsung 970 Evo Plus that
2679 		 * does not need or want these quirks.
2680 		 */
2681 		.vid = 0x144d,
2682 		.mn = "Samsung Portable SSD X5",
2683 		.quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2684 			  NVME_QUIRK_NO_DEEPEST_PS |
2685 			  NVME_QUIRK_IGNORE_DEV_SUBNQN,
2686 	}
2687 };
2688 
2689 /* match is null-terminated but idstr is space-padded. */
2690 static bool string_matches(const char *idstr, const char *match, size_t len)
2691 {
2692 	size_t matchlen;
2693 
2694 	if (!match)
2695 		return true;
2696 
2697 	matchlen = strlen(match);
2698 	WARN_ON_ONCE(matchlen > len);
2699 
2700 	if (memcmp(idstr, match, matchlen))
2701 		return false;
2702 
2703 	for (; matchlen < len; matchlen++)
2704 		if (idstr[matchlen] != ' ')
2705 			return false;
2706 
2707 	return true;
2708 }
2709 
2710 static bool quirk_matches(const struct nvme_id_ctrl *id,
2711 			  const struct nvme_core_quirk_entry *q)
2712 {
2713 	return q->vid == le16_to_cpu(id->vid) &&
2714 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2715 		string_matches(id->fr, q->fr, sizeof(id->fr));
2716 }
2717 
2718 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2719 		struct nvme_id_ctrl *id)
2720 {
2721 	size_t nqnlen;
2722 	int off;
2723 
2724 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2725 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2726 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2727 			strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2728 			return;
2729 		}
2730 
2731 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2732 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2733 	}
2734 
2735 	/*
2736 	 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2737 	 * Base Specification 2.0.  It is slightly different from the format
2738 	 * specified there due to historic reasons, and we can't change it now.
2739 	 */
2740 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2741 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2742 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2743 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2744 	off += sizeof(id->sn);
2745 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2746 	off += sizeof(id->mn);
2747 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2748 }
2749 
2750 static void nvme_release_subsystem(struct device *dev)
2751 {
2752 	struct nvme_subsystem *subsys =
2753 		container_of(dev, struct nvme_subsystem, dev);
2754 
2755 	if (subsys->instance >= 0)
2756 		ida_free(&nvme_instance_ida, subsys->instance);
2757 	kfree(subsys);
2758 }
2759 
2760 static void nvme_destroy_subsystem(struct kref *ref)
2761 {
2762 	struct nvme_subsystem *subsys =
2763 			container_of(ref, struct nvme_subsystem, ref);
2764 
2765 	mutex_lock(&nvme_subsystems_lock);
2766 	list_del(&subsys->entry);
2767 	mutex_unlock(&nvme_subsystems_lock);
2768 
2769 	ida_destroy(&subsys->ns_ida);
2770 	device_del(&subsys->dev);
2771 	put_device(&subsys->dev);
2772 }
2773 
2774 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2775 {
2776 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2777 }
2778 
2779 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2780 {
2781 	struct nvme_subsystem *subsys;
2782 
2783 	lockdep_assert_held(&nvme_subsystems_lock);
2784 
2785 	/*
2786 	 * Fail matches for discovery subsystems. This results
2787 	 * in each discovery controller bound to a unique subsystem.
2788 	 * This avoids issues with validating controller values
2789 	 * that can only be true when there is a single unique subsystem.
2790 	 * There may be multiple and completely independent entities
2791 	 * that provide discovery controllers.
2792 	 */
2793 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2794 		return NULL;
2795 
2796 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2797 		if (strcmp(subsys->subnqn, subsysnqn))
2798 			continue;
2799 		if (!kref_get_unless_zero(&subsys->ref))
2800 			continue;
2801 		return subsys;
2802 	}
2803 
2804 	return NULL;
2805 }
2806 
2807 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2808 {
2809 	return ctrl->opts && ctrl->opts->discovery_nqn;
2810 }
2811 
2812 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2813 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2814 {
2815 	struct nvme_ctrl *tmp;
2816 
2817 	lockdep_assert_held(&nvme_subsystems_lock);
2818 
2819 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2820 		if (nvme_state_terminal(tmp))
2821 			continue;
2822 
2823 		if (tmp->cntlid == ctrl->cntlid) {
2824 			dev_err(ctrl->device,
2825 				"Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2826 				ctrl->cntlid, dev_name(tmp->device),
2827 				subsys->subnqn);
2828 			return false;
2829 		}
2830 
2831 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2832 		    nvme_discovery_ctrl(ctrl))
2833 			continue;
2834 
2835 		dev_err(ctrl->device,
2836 			"Subsystem does not support multiple controllers\n");
2837 		return false;
2838 	}
2839 
2840 	return true;
2841 }
2842 
2843 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2844 {
2845 	struct nvme_subsystem *subsys, *found;
2846 	int ret;
2847 
2848 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2849 	if (!subsys)
2850 		return -ENOMEM;
2851 
2852 	subsys->instance = -1;
2853 	mutex_init(&subsys->lock);
2854 	kref_init(&subsys->ref);
2855 	INIT_LIST_HEAD(&subsys->ctrls);
2856 	INIT_LIST_HEAD(&subsys->nsheads);
2857 	nvme_init_subnqn(subsys, ctrl, id);
2858 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2859 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2860 	subsys->vendor_id = le16_to_cpu(id->vid);
2861 	subsys->cmic = id->cmic;
2862 
2863 	/* Versions prior to 1.4 don't necessarily report a valid type */
2864 	if (id->cntrltype == NVME_CTRL_DISC ||
2865 	    !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2866 		subsys->subtype = NVME_NQN_DISC;
2867 	else
2868 		subsys->subtype = NVME_NQN_NVME;
2869 
2870 	if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2871 		dev_err(ctrl->device,
2872 			"Subsystem %s is not a discovery controller",
2873 			subsys->subnqn);
2874 		kfree(subsys);
2875 		return -EINVAL;
2876 	}
2877 	subsys->awupf = le16_to_cpu(id->awupf);
2878 	nvme_mpath_default_iopolicy(subsys);
2879 
2880 	subsys->dev.class = nvme_subsys_class;
2881 	subsys->dev.release = nvme_release_subsystem;
2882 	subsys->dev.groups = nvme_subsys_attrs_groups;
2883 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2884 	device_initialize(&subsys->dev);
2885 
2886 	mutex_lock(&nvme_subsystems_lock);
2887 	found = __nvme_find_get_subsystem(subsys->subnqn);
2888 	if (found) {
2889 		put_device(&subsys->dev);
2890 		subsys = found;
2891 
2892 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2893 			ret = -EINVAL;
2894 			goto out_put_subsystem;
2895 		}
2896 	} else {
2897 		ret = device_add(&subsys->dev);
2898 		if (ret) {
2899 			dev_err(ctrl->device,
2900 				"failed to register subsystem device.\n");
2901 			put_device(&subsys->dev);
2902 			goto out_unlock;
2903 		}
2904 		ida_init(&subsys->ns_ida);
2905 		list_add_tail(&subsys->entry, &nvme_subsystems);
2906 	}
2907 
2908 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2909 				dev_name(ctrl->device));
2910 	if (ret) {
2911 		dev_err(ctrl->device,
2912 			"failed to create sysfs link from subsystem.\n");
2913 		goto out_put_subsystem;
2914 	}
2915 
2916 	if (!found)
2917 		subsys->instance = ctrl->instance;
2918 	ctrl->subsys = subsys;
2919 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2920 	mutex_unlock(&nvme_subsystems_lock);
2921 	return 0;
2922 
2923 out_put_subsystem:
2924 	nvme_put_subsystem(subsys);
2925 out_unlock:
2926 	mutex_unlock(&nvme_subsystems_lock);
2927 	return ret;
2928 }
2929 
2930 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2931 		void *log, size_t size, u64 offset)
2932 {
2933 	struct nvme_command c = { };
2934 	u32 dwlen = nvme_bytes_to_numd(size);
2935 
2936 	c.get_log_page.opcode = nvme_admin_get_log_page;
2937 	c.get_log_page.nsid = cpu_to_le32(nsid);
2938 	c.get_log_page.lid = log_page;
2939 	c.get_log_page.lsp = lsp;
2940 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2941 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2942 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2943 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2944 	c.get_log_page.csi = csi;
2945 
2946 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2947 }
2948 
2949 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2950 				struct nvme_effects_log **log)
2951 {
2952 	struct nvme_effects_log	*cel = xa_load(&ctrl->cels, csi);
2953 	int ret;
2954 
2955 	if (cel)
2956 		goto out;
2957 
2958 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2959 	if (!cel)
2960 		return -ENOMEM;
2961 
2962 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2963 			cel, sizeof(*cel), 0);
2964 	if (ret) {
2965 		kfree(cel);
2966 		return ret;
2967 	}
2968 
2969 	xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2970 out:
2971 	*log = cel;
2972 	return 0;
2973 }
2974 
2975 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2976 {
2977 	u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2978 
2979 	if (check_shl_overflow(1U, units + page_shift - 9, &val))
2980 		return UINT_MAX;
2981 	return val;
2982 }
2983 
2984 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2985 {
2986 	struct nvme_command c = { };
2987 	struct nvme_id_ctrl_nvm *id;
2988 	int ret;
2989 
2990 	/*
2991 	 * Even though NVMe spec explicitly states that MDTS is not applicable
2992 	 * to the write-zeroes, we are cautious and limit the size to the
2993 	 * controllers max_hw_sectors value, which is based on the MDTS field
2994 	 * and possibly other limiting factors.
2995 	 */
2996 	if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2997 	    !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2998 		ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2999 	else
3000 		ctrl->max_zeroes_sectors = 0;
3001 
3002 	if (ctrl->subsys->subtype != NVME_NQN_NVME ||
3003 	    nvme_ctrl_limited_cns(ctrl) ||
3004 	    test_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags))
3005 		return 0;
3006 
3007 	id = kzalloc(sizeof(*id), GFP_KERNEL);
3008 	if (!id)
3009 		return -ENOMEM;
3010 
3011 	c.identify.opcode = nvme_admin_identify;
3012 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
3013 	c.identify.csi = NVME_CSI_NVM;
3014 
3015 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
3016 	if (ret)
3017 		goto free_data;
3018 
3019 	ctrl->dmrl = id->dmrl;
3020 	ctrl->dmrsl = le32_to_cpu(id->dmrsl);
3021 	if (id->wzsl)
3022 		ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
3023 
3024 free_data:
3025 	if (ret > 0)
3026 		set_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags);
3027 	kfree(id);
3028 	return ret;
3029 }
3030 
3031 static void nvme_init_known_nvm_effects(struct nvme_ctrl *ctrl)
3032 {
3033 	struct nvme_effects_log	*log = ctrl->effects;
3034 
3035 	log->acs[nvme_admin_format_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
3036 						NVME_CMD_EFFECTS_NCC |
3037 						NVME_CMD_EFFECTS_CSE_MASK);
3038 	log->acs[nvme_admin_sanitize_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
3039 						NVME_CMD_EFFECTS_CSE_MASK);
3040 
3041 	/*
3042 	 * The spec says the result of a security receive command depends on
3043 	 * the previous security send command. As such, many vendors log this
3044 	 * command as one to submitted only when no other commands to the same
3045 	 * namespace are outstanding. The intention is to tell the host to
3046 	 * prevent mixing security send and receive.
3047 	 *
3048 	 * This driver can only enforce such exclusive access against IO
3049 	 * queues, though. We are not readily able to enforce such a rule for
3050 	 * two commands to the admin queue, which is the only queue that
3051 	 * matters for this command.
3052 	 *
3053 	 * Rather than blindly freezing the IO queues for this effect that
3054 	 * doesn't even apply to IO, mask it off.
3055 	 */
3056 	log->acs[nvme_admin_security_recv] &= cpu_to_le32(~NVME_CMD_EFFECTS_CSE_MASK);
3057 
3058 	log->iocs[nvme_cmd_write] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3059 	log->iocs[nvme_cmd_write_zeroes] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3060 	log->iocs[nvme_cmd_write_uncor] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3061 }
3062 
3063 static int nvme_init_effects(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
3064 {
3065 	int ret = 0;
3066 
3067 	if (ctrl->effects)
3068 		return 0;
3069 
3070 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3071 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3072 		if (ret < 0)
3073 			return ret;
3074 	}
3075 
3076 	if (!ctrl->effects) {
3077 		ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
3078 		if (!ctrl->effects)
3079 			return -ENOMEM;
3080 		xa_store(&ctrl->cels, NVME_CSI_NVM, ctrl->effects, GFP_KERNEL);
3081 	}
3082 
3083 	nvme_init_known_nvm_effects(ctrl);
3084 	return 0;
3085 }
3086 
3087 static int nvme_check_ctrl_fabric_info(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
3088 {
3089 	/*
3090 	 * In fabrics we need to verify the cntlid matches the
3091 	 * admin connect
3092 	 */
3093 	if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3094 		dev_err(ctrl->device,
3095 			"Mismatching cntlid: Connect %u vs Identify %u, rejecting\n",
3096 			ctrl->cntlid, le16_to_cpu(id->cntlid));
3097 		return -EINVAL;
3098 	}
3099 
3100 	if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3101 		dev_err(ctrl->device,
3102 			"keep-alive support is mandatory for fabrics\n");
3103 		return -EINVAL;
3104 	}
3105 
3106 	if (!nvme_discovery_ctrl(ctrl) && ctrl->ioccsz < 4) {
3107 		dev_err(ctrl->device,
3108 			"I/O queue command capsule supported size %d < 4\n",
3109 			ctrl->ioccsz);
3110 		return -EINVAL;
3111 	}
3112 
3113 	if (!nvme_discovery_ctrl(ctrl) && ctrl->iorcsz < 1) {
3114 		dev_err(ctrl->device,
3115 			"I/O queue response capsule supported size %d < 1\n",
3116 			ctrl->iorcsz);
3117 		return -EINVAL;
3118 	}
3119 
3120 	return 0;
3121 }
3122 
3123 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3124 {
3125 	struct nvme_id_ctrl *id;
3126 	u32 max_hw_sectors;
3127 	bool prev_apst_enabled;
3128 	int ret;
3129 
3130 	ret = nvme_identify_ctrl(ctrl, &id);
3131 	if (ret) {
3132 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3133 		return -EIO;
3134 	}
3135 
3136 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
3137 		ctrl->cntlid = le16_to_cpu(id->cntlid);
3138 
3139 	if (!ctrl->identified) {
3140 		unsigned int i;
3141 
3142 		/*
3143 		 * Check for quirks.  Quirk can depend on firmware version,
3144 		 * so, in principle, the set of quirks present can change
3145 		 * across a reset.  As a possible future enhancement, we
3146 		 * could re-scan for quirks every time we reinitialize
3147 		 * the device, but we'd have to make sure that the driver
3148 		 * behaves intelligently if the quirks change.
3149 		 */
3150 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3151 			if (quirk_matches(id, &core_quirks[i]))
3152 				ctrl->quirks |= core_quirks[i].quirks;
3153 		}
3154 
3155 		ret = nvme_init_subsystem(ctrl, id);
3156 		if (ret)
3157 			goto out_free;
3158 
3159 		ret = nvme_init_effects(ctrl, id);
3160 		if (ret)
3161 			goto out_free;
3162 	}
3163 	memcpy(ctrl->subsys->firmware_rev, id->fr,
3164 	       sizeof(ctrl->subsys->firmware_rev));
3165 
3166 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3167 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3168 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3169 	}
3170 
3171 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3172 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3173 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3174 
3175 	ctrl->oacs = le16_to_cpu(id->oacs);
3176 	ctrl->oncs = le16_to_cpu(id->oncs);
3177 	ctrl->mtfa = le16_to_cpu(id->mtfa);
3178 	ctrl->oaes = le32_to_cpu(id->oaes);
3179 	ctrl->wctemp = le16_to_cpu(id->wctemp);
3180 	ctrl->cctemp = le16_to_cpu(id->cctemp);
3181 
3182 	atomic_set(&ctrl->abort_limit, id->acl + 1);
3183 	ctrl->vwc = id->vwc;
3184 	if (id->mdts)
3185 		max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3186 	else
3187 		max_hw_sectors = UINT_MAX;
3188 	ctrl->max_hw_sectors =
3189 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3190 
3191 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
3192 	ctrl->sgls = le32_to_cpu(id->sgls);
3193 	ctrl->kas = le16_to_cpu(id->kas);
3194 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
3195 	ctrl->ctratt = le32_to_cpu(id->ctratt);
3196 
3197 	ctrl->cntrltype = id->cntrltype;
3198 	ctrl->dctype = id->dctype;
3199 
3200 	if (id->rtd3e) {
3201 		/* us -> s */
3202 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3203 
3204 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3205 						 shutdown_timeout, 60);
3206 
3207 		if (ctrl->shutdown_timeout != shutdown_timeout)
3208 			dev_info(ctrl->device,
3209 				 "Shutdown timeout set to %u seconds\n",
3210 				 ctrl->shutdown_timeout);
3211 	} else
3212 		ctrl->shutdown_timeout = shutdown_timeout;
3213 
3214 	ctrl->npss = id->npss;
3215 	ctrl->apsta = id->apsta;
3216 	prev_apst_enabled = ctrl->apst_enabled;
3217 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3218 		if (force_apst && id->apsta) {
3219 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3220 			ctrl->apst_enabled = true;
3221 		} else {
3222 			ctrl->apst_enabled = false;
3223 		}
3224 	} else {
3225 		ctrl->apst_enabled = id->apsta;
3226 	}
3227 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3228 
3229 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3230 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3231 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3232 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3233 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3234 
3235 		ret = nvme_check_ctrl_fabric_info(ctrl, id);
3236 		if (ret)
3237 			goto out_free;
3238 	} else {
3239 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3240 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3241 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3242 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3243 	}
3244 
3245 	ret = nvme_mpath_init_identify(ctrl, id);
3246 	if (ret < 0)
3247 		goto out_free;
3248 
3249 	if (ctrl->apst_enabled && !prev_apst_enabled)
3250 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3251 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3252 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3253 
3254 out_free:
3255 	kfree(id);
3256 	return ret;
3257 }
3258 
3259 /*
3260  * Initialize the cached copies of the Identify data and various controller
3261  * register in our nvme_ctrl structure.  This should be called as soon as
3262  * the admin queue is fully up and running.
3263  */
3264 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended)
3265 {
3266 	int ret;
3267 
3268 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3269 	if (ret) {
3270 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3271 		return ret;
3272 	}
3273 
3274 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3275 
3276 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3277 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3278 
3279 	ret = nvme_init_identify(ctrl);
3280 	if (ret)
3281 		return ret;
3282 
3283 	ret = nvme_configure_apst(ctrl);
3284 	if (ret < 0)
3285 		return ret;
3286 
3287 	ret = nvme_configure_timestamp(ctrl);
3288 	if (ret < 0)
3289 		return ret;
3290 
3291 	ret = nvme_configure_host_options(ctrl);
3292 	if (ret < 0)
3293 		return ret;
3294 
3295 	nvme_configure_opal(ctrl, was_suspended);
3296 
3297 	if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3298 		/*
3299 		 * Do not return errors unless we are in a controller reset,
3300 		 * the controller works perfectly fine without hwmon.
3301 		 */
3302 		ret = nvme_hwmon_init(ctrl);
3303 		if (ret == -EINTR)
3304 			return ret;
3305 	}
3306 
3307 	clear_bit(NVME_CTRL_DIRTY_CAPABILITY, &ctrl->flags);
3308 	ctrl->identified = true;
3309 
3310 	nvme_start_keep_alive(ctrl);
3311 
3312 	return 0;
3313 }
3314 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3315 
3316 static int nvme_dev_open(struct inode *inode, struct file *file)
3317 {
3318 	struct nvme_ctrl *ctrl =
3319 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3320 
3321 	switch (nvme_ctrl_state(ctrl)) {
3322 	case NVME_CTRL_LIVE:
3323 		break;
3324 	default:
3325 		return -EWOULDBLOCK;
3326 	}
3327 
3328 	nvme_get_ctrl(ctrl);
3329 	if (!try_module_get(ctrl->ops->module)) {
3330 		nvme_put_ctrl(ctrl);
3331 		return -EINVAL;
3332 	}
3333 
3334 	file->private_data = ctrl;
3335 	return 0;
3336 }
3337 
3338 static int nvme_dev_release(struct inode *inode, struct file *file)
3339 {
3340 	struct nvme_ctrl *ctrl =
3341 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3342 
3343 	module_put(ctrl->ops->module);
3344 	nvme_put_ctrl(ctrl);
3345 	return 0;
3346 }
3347 
3348 static const struct file_operations nvme_dev_fops = {
3349 	.owner		= THIS_MODULE,
3350 	.open		= nvme_dev_open,
3351 	.release	= nvme_dev_release,
3352 	.unlocked_ioctl	= nvme_dev_ioctl,
3353 	.compat_ioctl	= compat_ptr_ioctl,
3354 	.uring_cmd	= nvme_dev_uring_cmd,
3355 };
3356 
3357 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3358 		unsigned nsid)
3359 {
3360 	struct nvme_ns_head *h;
3361 
3362 	lockdep_assert_held(&ctrl->subsys->lock);
3363 
3364 	list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3365 		/*
3366 		 * Private namespaces can share NSIDs under some conditions.
3367 		 * In that case we can't use the same ns_head for namespaces
3368 		 * with the same NSID.
3369 		 */
3370 		if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3371 			continue;
3372 		if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3373 			return h;
3374 	}
3375 
3376 	return NULL;
3377 }
3378 
3379 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3380 		struct nvme_ns_ids *ids)
3381 {
3382 	bool has_uuid = !uuid_is_null(&ids->uuid);
3383 	bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3384 	bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3385 	struct nvme_ns_head *h;
3386 
3387 	lockdep_assert_held(&subsys->lock);
3388 
3389 	list_for_each_entry(h, &subsys->nsheads, entry) {
3390 		if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3391 			return -EINVAL;
3392 		if (has_nguid &&
3393 		    memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3394 			return -EINVAL;
3395 		if (has_eui64 &&
3396 		    memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3397 			return -EINVAL;
3398 	}
3399 
3400 	return 0;
3401 }
3402 
3403 static void nvme_cdev_rel(struct device *dev)
3404 {
3405 	ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3406 }
3407 
3408 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3409 {
3410 	cdev_device_del(cdev, cdev_device);
3411 	put_device(cdev_device);
3412 }
3413 
3414 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3415 		const struct file_operations *fops, struct module *owner)
3416 {
3417 	int minor, ret;
3418 
3419 	minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3420 	if (minor < 0)
3421 		return minor;
3422 	cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3423 	cdev_device->class = nvme_ns_chr_class;
3424 	cdev_device->release = nvme_cdev_rel;
3425 	device_initialize(cdev_device);
3426 	cdev_init(cdev, fops);
3427 	cdev->owner = owner;
3428 	ret = cdev_device_add(cdev, cdev_device);
3429 	if (ret)
3430 		put_device(cdev_device);
3431 
3432 	return ret;
3433 }
3434 
3435 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3436 {
3437 	return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3438 }
3439 
3440 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3441 {
3442 	nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3443 	return 0;
3444 }
3445 
3446 static const struct file_operations nvme_ns_chr_fops = {
3447 	.owner		= THIS_MODULE,
3448 	.open		= nvme_ns_chr_open,
3449 	.release	= nvme_ns_chr_release,
3450 	.unlocked_ioctl	= nvme_ns_chr_ioctl,
3451 	.compat_ioctl	= compat_ptr_ioctl,
3452 	.uring_cmd	= nvme_ns_chr_uring_cmd,
3453 	.uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
3454 };
3455 
3456 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3457 {
3458 	int ret;
3459 
3460 	ns->cdev_device.parent = ns->ctrl->device;
3461 	ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3462 			   ns->ctrl->instance, ns->head->instance);
3463 	if (ret)
3464 		return ret;
3465 
3466 	return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3467 			     ns->ctrl->ops->module);
3468 }
3469 
3470 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3471 		struct nvme_ns_info *info)
3472 {
3473 	struct nvme_ns_head *head;
3474 	size_t size = sizeof(*head);
3475 	int ret = -ENOMEM;
3476 
3477 #ifdef CONFIG_NVME_MULTIPATH
3478 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3479 #endif
3480 
3481 	head = kzalloc(size, GFP_KERNEL);
3482 	if (!head)
3483 		goto out;
3484 	ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3485 	if (ret < 0)
3486 		goto out_free_head;
3487 	head->instance = ret;
3488 	INIT_LIST_HEAD(&head->list);
3489 	ret = init_srcu_struct(&head->srcu);
3490 	if (ret)
3491 		goto out_ida_remove;
3492 	head->subsys = ctrl->subsys;
3493 	head->ns_id = info->nsid;
3494 	head->ids = info->ids;
3495 	head->shared = info->is_shared;
3496 	ratelimit_state_init(&head->rs_nuse, 5 * HZ, 1);
3497 	ratelimit_set_flags(&head->rs_nuse, RATELIMIT_MSG_ON_RELEASE);
3498 	kref_init(&head->ref);
3499 
3500 	if (head->ids.csi) {
3501 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3502 		if (ret)
3503 			goto out_cleanup_srcu;
3504 	} else
3505 		head->effects = ctrl->effects;
3506 
3507 	ret = nvme_mpath_alloc_disk(ctrl, head);
3508 	if (ret)
3509 		goto out_cleanup_srcu;
3510 
3511 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3512 
3513 	kref_get(&ctrl->subsys->ref);
3514 
3515 	return head;
3516 out_cleanup_srcu:
3517 	cleanup_srcu_struct(&head->srcu);
3518 out_ida_remove:
3519 	ida_free(&ctrl->subsys->ns_ida, head->instance);
3520 out_free_head:
3521 	kfree(head);
3522 out:
3523 	if (ret > 0)
3524 		ret = blk_status_to_errno(nvme_error_status(ret));
3525 	return ERR_PTR(ret);
3526 }
3527 
3528 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3529 		struct nvme_ns_ids *ids)
3530 {
3531 	struct nvme_subsystem *s;
3532 	int ret = 0;
3533 
3534 	/*
3535 	 * Note that this check is racy as we try to avoid holding the global
3536 	 * lock over the whole ns_head creation.  But it is only intended as
3537 	 * a sanity check anyway.
3538 	 */
3539 	mutex_lock(&nvme_subsystems_lock);
3540 	list_for_each_entry(s, &nvme_subsystems, entry) {
3541 		if (s == this)
3542 			continue;
3543 		mutex_lock(&s->lock);
3544 		ret = nvme_subsys_check_duplicate_ids(s, ids);
3545 		mutex_unlock(&s->lock);
3546 		if (ret)
3547 			break;
3548 	}
3549 	mutex_unlock(&nvme_subsystems_lock);
3550 
3551 	return ret;
3552 }
3553 
3554 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
3555 {
3556 	struct nvme_ctrl *ctrl = ns->ctrl;
3557 	struct nvme_ns_head *head = NULL;
3558 	int ret;
3559 
3560 	ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
3561 	if (ret) {
3562 		/*
3563 		 * We've found two different namespaces on two different
3564 		 * subsystems that report the same ID.  This is pretty nasty
3565 		 * for anything that actually requires unique device
3566 		 * identification.  In the kernel we need this for multipathing,
3567 		 * and in user space the /dev/disk/by-id/ links rely on it.
3568 		 *
3569 		 * If the device also claims to be multi-path capable back off
3570 		 * here now and refuse the probe the second device as this is a
3571 		 * recipe for data corruption.  If not this is probably a
3572 		 * cheap consumer device if on the PCIe bus, so let the user
3573 		 * proceed and use the shiny toy, but warn that with changing
3574 		 * probing order (which due to our async probing could just be
3575 		 * device taking longer to startup) the other device could show
3576 		 * up at any time.
3577 		 */
3578 		nvme_print_device_info(ctrl);
3579 		if ((ns->ctrl->ops->flags & NVME_F_FABRICS) || /* !PCIe */
3580 		    ((ns->ctrl->subsys->cmic & NVME_CTRL_CMIC_MULTI_CTRL) &&
3581 		     info->is_shared)) {
3582 			dev_err(ctrl->device,
3583 				"ignoring nsid %d because of duplicate IDs\n",
3584 				info->nsid);
3585 			return ret;
3586 		}
3587 
3588 		dev_err(ctrl->device,
3589 			"clearing duplicate IDs for nsid %d\n", info->nsid);
3590 		dev_err(ctrl->device,
3591 			"use of /dev/disk/by-id/ may cause data corruption\n");
3592 		memset(&info->ids.nguid, 0, sizeof(info->ids.nguid));
3593 		memset(&info->ids.uuid, 0, sizeof(info->ids.uuid));
3594 		memset(&info->ids.eui64, 0, sizeof(info->ids.eui64));
3595 		ctrl->quirks |= NVME_QUIRK_BOGUS_NID;
3596 	}
3597 
3598 	mutex_lock(&ctrl->subsys->lock);
3599 	head = nvme_find_ns_head(ctrl, info->nsid);
3600 	if (!head) {
3601 		ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
3602 		if (ret) {
3603 			dev_err(ctrl->device,
3604 				"duplicate IDs in subsystem for nsid %d\n",
3605 				info->nsid);
3606 			goto out_unlock;
3607 		}
3608 		head = nvme_alloc_ns_head(ctrl, info);
3609 		if (IS_ERR(head)) {
3610 			ret = PTR_ERR(head);
3611 			goto out_unlock;
3612 		}
3613 	} else {
3614 		ret = -EINVAL;
3615 		if (!info->is_shared || !head->shared) {
3616 			dev_err(ctrl->device,
3617 				"Duplicate unshared namespace %d\n",
3618 				info->nsid);
3619 			goto out_put_ns_head;
3620 		}
3621 		if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
3622 			dev_err(ctrl->device,
3623 				"IDs don't match for shared namespace %d\n",
3624 					info->nsid);
3625 			goto out_put_ns_head;
3626 		}
3627 
3628 		if (!multipath) {
3629 			dev_warn(ctrl->device,
3630 				"Found shared namespace %d, but multipathing not supported.\n",
3631 				info->nsid);
3632 			dev_warn_once(ctrl->device,
3633 				"Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
3634 		}
3635 	}
3636 
3637 	list_add_tail_rcu(&ns->siblings, &head->list);
3638 	ns->head = head;
3639 	mutex_unlock(&ctrl->subsys->lock);
3640 	return 0;
3641 
3642 out_put_ns_head:
3643 	nvme_put_ns_head(head);
3644 out_unlock:
3645 	mutex_unlock(&ctrl->subsys->lock);
3646 	return ret;
3647 }
3648 
3649 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3650 {
3651 	struct nvme_ns *ns, *ret = NULL;
3652 
3653 	down_read(&ctrl->namespaces_rwsem);
3654 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3655 		if (ns->head->ns_id == nsid) {
3656 			if (!nvme_get_ns(ns))
3657 				continue;
3658 			ret = ns;
3659 			break;
3660 		}
3661 		if (ns->head->ns_id > nsid)
3662 			break;
3663 	}
3664 	up_read(&ctrl->namespaces_rwsem);
3665 	return ret;
3666 }
3667 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3668 
3669 /*
3670  * Add the namespace to the controller list while keeping the list ordered.
3671  */
3672 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3673 {
3674 	struct nvme_ns *tmp;
3675 
3676 	list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3677 		if (tmp->head->ns_id < ns->head->ns_id) {
3678 			list_add(&ns->list, &tmp->list);
3679 			return;
3680 		}
3681 	}
3682 	list_add(&ns->list, &ns->ctrl->namespaces);
3683 }
3684 
3685 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
3686 {
3687 	struct nvme_ns *ns;
3688 	struct gendisk *disk;
3689 	int node = ctrl->numa_node;
3690 
3691 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3692 	if (!ns)
3693 		return;
3694 
3695 	disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3696 	if (IS_ERR(disk))
3697 		goto out_free_ns;
3698 	disk->fops = &nvme_bdev_ops;
3699 	disk->private_data = ns;
3700 
3701 	ns->disk = disk;
3702 	ns->queue = disk->queue;
3703 
3704 	if (ctrl->opts && ctrl->opts->data_digest)
3705 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3706 
3707 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3708 	if (ctrl->ops->supports_pci_p2pdma &&
3709 	    ctrl->ops->supports_pci_p2pdma(ctrl))
3710 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3711 
3712 	ns->ctrl = ctrl;
3713 	kref_init(&ns->kref);
3714 
3715 	if (nvme_init_ns_head(ns, info))
3716 		goto out_cleanup_disk;
3717 
3718 	/*
3719 	 * If multipathing is enabled, the device name for all disks and not
3720 	 * just those that represent shared namespaces needs to be based on the
3721 	 * subsystem instance.  Using the controller instance for private
3722 	 * namespaces could lead to naming collisions between shared and private
3723 	 * namespaces if they don't use a common numbering scheme.
3724 	 *
3725 	 * If multipathing is not enabled, disk names must use the controller
3726 	 * instance as shared namespaces will show up as multiple block
3727 	 * devices.
3728 	 */
3729 	if (nvme_ns_head_multipath(ns->head)) {
3730 		sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
3731 			ctrl->instance, ns->head->instance);
3732 		disk->flags |= GENHD_FL_HIDDEN;
3733 	} else if (multipath) {
3734 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
3735 			ns->head->instance);
3736 	} else {
3737 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3738 			ns->head->instance);
3739 	}
3740 
3741 	if (nvme_update_ns_info(ns, info))
3742 		goto out_unlink_ns;
3743 
3744 	down_write(&ctrl->namespaces_rwsem);
3745 	/*
3746 	 * Ensure that no namespaces are added to the ctrl list after the queues
3747 	 * are frozen, thereby avoiding a deadlock between scan and reset.
3748 	 */
3749 	if (test_bit(NVME_CTRL_FROZEN, &ctrl->flags)) {
3750 		up_write(&ctrl->namespaces_rwsem);
3751 		goto out_unlink_ns;
3752 	}
3753 	nvme_ns_add_to_ctrl_list(ns);
3754 	up_write(&ctrl->namespaces_rwsem);
3755 	nvme_get_ctrl(ctrl);
3756 
3757 	if (device_add_disk(ctrl->device, ns->disk, nvme_ns_attr_groups))
3758 		goto out_cleanup_ns_from_list;
3759 
3760 	if (!nvme_ns_head_multipath(ns->head))
3761 		nvme_add_ns_cdev(ns);
3762 
3763 	nvme_mpath_add_disk(ns, info->anagrpid);
3764 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3765 
3766 	/*
3767 	 * Set ns->disk->device->driver_data to ns so we can access
3768 	 * ns->head->passthru_err_log_enabled in
3769 	 * nvme_io_passthru_err_log_enabled_[store | show]().
3770 	 */
3771 	dev_set_drvdata(disk_to_dev(ns->disk), ns);
3772 
3773 	return;
3774 
3775  out_cleanup_ns_from_list:
3776 	nvme_put_ctrl(ctrl);
3777 	down_write(&ctrl->namespaces_rwsem);
3778 	list_del_init(&ns->list);
3779 	up_write(&ctrl->namespaces_rwsem);
3780  out_unlink_ns:
3781 	mutex_lock(&ctrl->subsys->lock);
3782 	list_del_rcu(&ns->siblings);
3783 	if (list_empty(&ns->head->list))
3784 		list_del_init(&ns->head->entry);
3785 	mutex_unlock(&ctrl->subsys->lock);
3786 	nvme_put_ns_head(ns->head);
3787  out_cleanup_disk:
3788 	put_disk(disk);
3789  out_free_ns:
3790 	kfree(ns);
3791 }
3792 
3793 static void nvme_ns_remove(struct nvme_ns *ns)
3794 {
3795 	bool last_path = false;
3796 
3797 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3798 		return;
3799 
3800 	clear_bit(NVME_NS_READY, &ns->flags);
3801 	set_capacity(ns->disk, 0);
3802 	nvme_fault_inject_fini(&ns->fault_inject);
3803 
3804 	/*
3805 	 * Ensure that !NVME_NS_READY is seen by other threads to prevent
3806 	 * this ns going back into current_path.
3807 	 */
3808 	synchronize_srcu(&ns->head->srcu);
3809 
3810 	/* wait for concurrent submissions */
3811 	if (nvme_mpath_clear_current_path(ns))
3812 		synchronize_srcu(&ns->head->srcu);
3813 
3814 	mutex_lock(&ns->ctrl->subsys->lock);
3815 	list_del_rcu(&ns->siblings);
3816 	if (list_empty(&ns->head->list)) {
3817 		list_del_init(&ns->head->entry);
3818 		last_path = true;
3819 	}
3820 	mutex_unlock(&ns->ctrl->subsys->lock);
3821 
3822 	/* guarantee not available in head->list */
3823 	synchronize_srcu(&ns->head->srcu);
3824 
3825 	if (!nvme_ns_head_multipath(ns->head))
3826 		nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3827 	del_gendisk(ns->disk);
3828 
3829 	down_write(&ns->ctrl->namespaces_rwsem);
3830 	list_del_init(&ns->list);
3831 	up_write(&ns->ctrl->namespaces_rwsem);
3832 
3833 	if (last_path)
3834 		nvme_mpath_shutdown_disk(ns->head);
3835 	nvme_put_ns(ns);
3836 }
3837 
3838 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3839 {
3840 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3841 
3842 	if (ns) {
3843 		nvme_ns_remove(ns);
3844 		nvme_put_ns(ns);
3845 	}
3846 }
3847 
3848 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
3849 {
3850 	int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3851 
3852 	if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
3853 		dev_err(ns->ctrl->device,
3854 			"identifiers changed for nsid %d\n", ns->head->ns_id);
3855 		goto out;
3856 	}
3857 
3858 	ret = nvme_update_ns_info(ns, info);
3859 out:
3860 	/*
3861 	 * Only remove the namespace if we got a fatal error back from the
3862 	 * device, otherwise ignore the error and just move on.
3863 	 *
3864 	 * TODO: we should probably schedule a delayed retry here.
3865 	 */
3866 	if (ret > 0 && (ret & NVME_SC_DNR))
3867 		nvme_ns_remove(ns);
3868 }
3869 
3870 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3871 {
3872 	struct nvme_ns_info info = { .nsid = nsid };
3873 	struct nvme_ns *ns;
3874 	int ret;
3875 
3876 	if (nvme_identify_ns_descs(ctrl, &info))
3877 		return;
3878 
3879 	if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
3880 		dev_warn(ctrl->device,
3881 			"command set not reported for nsid: %d\n", nsid);
3882 		return;
3883 	}
3884 
3885 	/*
3886 	 * If available try to use the Command Set Idependent Identify Namespace
3887 	 * data structure to find all the generic information that is needed to
3888 	 * set up a namespace.  If not fall back to the legacy version.
3889 	 */
3890 	if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
3891 	    (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS))
3892 		ret = nvme_ns_info_from_id_cs_indep(ctrl, &info);
3893 	else
3894 		ret = nvme_ns_info_from_identify(ctrl, &info);
3895 
3896 	if (info.is_removed)
3897 		nvme_ns_remove_by_nsid(ctrl, nsid);
3898 
3899 	/*
3900 	 * Ignore the namespace if it is not ready. We will get an AEN once it
3901 	 * becomes ready and restart the scan.
3902 	 */
3903 	if (ret || !info.is_ready)
3904 		return;
3905 
3906 	ns = nvme_find_get_ns(ctrl, nsid);
3907 	if (ns) {
3908 		nvme_validate_ns(ns, &info);
3909 		nvme_put_ns(ns);
3910 	} else {
3911 		nvme_alloc_ns(ctrl, &info);
3912 	}
3913 }
3914 
3915 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3916 					unsigned nsid)
3917 {
3918 	struct nvme_ns *ns, *next;
3919 	LIST_HEAD(rm_list);
3920 
3921 	down_write(&ctrl->namespaces_rwsem);
3922 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3923 		if (ns->head->ns_id > nsid)
3924 			list_move_tail(&ns->list, &rm_list);
3925 	}
3926 	up_write(&ctrl->namespaces_rwsem);
3927 
3928 	list_for_each_entry_safe(ns, next, &rm_list, list)
3929 		nvme_ns_remove(ns);
3930 
3931 }
3932 
3933 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3934 {
3935 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3936 	__le32 *ns_list;
3937 	u32 prev = 0;
3938 	int ret = 0, i;
3939 
3940 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3941 	if (!ns_list)
3942 		return -ENOMEM;
3943 
3944 	for (;;) {
3945 		struct nvme_command cmd = {
3946 			.identify.opcode	= nvme_admin_identify,
3947 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
3948 			.identify.nsid		= cpu_to_le32(prev),
3949 		};
3950 
3951 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
3952 					    NVME_IDENTIFY_DATA_SIZE);
3953 		if (ret) {
3954 			dev_warn(ctrl->device,
3955 				"Identify NS List failed (status=0x%x)\n", ret);
3956 			goto free;
3957 		}
3958 
3959 		for (i = 0; i < nr_entries; i++) {
3960 			u32 nsid = le32_to_cpu(ns_list[i]);
3961 
3962 			if (!nsid)	/* end of the list? */
3963 				goto out;
3964 			nvme_scan_ns(ctrl, nsid);
3965 			while (++prev < nsid)
3966 				nvme_ns_remove_by_nsid(ctrl, prev);
3967 		}
3968 	}
3969  out:
3970 	nvme_remove_invalid_namespaces(ctrl, prev);
3971  free:
3972 	kfree(ns_list);
3973 	return ret;
3974 }
3975 
3976 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3977 {
3978 	struct nvme_id_ctrl *id;
3979 	u32 nn, i;
3980 
3981 	if (nvme_identify_ctrl(ctrl, &id))
3982 		return;
3983 	nn = le32_to_cpu(id->nn);
3984 	kfree(id);
3985 
3986 	for (i = 1; i <= nn; i++)
3987 		nvme_scan_ns(ctrl, i);
3988 
3989 	nvme_remove_invalid_namespaces(ctrl, nn);
3990 }
3991 
3992 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3993 {
3994 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3995 	__le32 *log;
3996 	int error;
3997 
3998 	log = kzalloc(log_size, GFP_KERNEL);
3999 	if (!log)
4000 		return;
4001 
4002 	/*
4003 	 * We need to read the log to clear the AEN, but we don't want to rely
4004 	 * on it for the changed namespace information as userspace could have
4005 	 * raced with us in reading the log page, which could cause us to miss
4006 	 * updates.
4007 	 */
4008 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4009 			NVME_CSI_NVM, log, log_size, 0);
4010 	if (error)
4011 		dev_warn(ctrl->device,
4012 			"reading changed ns log failed: %d\n", error);
4013 
4014 	kfree(log);
4015 }
4016 
4017 static void nvme_scan_work(struct work_struct *work)
4018 {
4019 	struct nvme_ctrl *ctrl =
4020 		container_of(work, struct nvme_ctrl, scan_work);
4021 	int ret;
4022 
4023 	/* No tagset on a live ctrl means IO queues could not created */
4024 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE || !ctrl->tagset)
4025 		return;
4026 
4027 	/*
4028 	 * Identify controller limits can change at controller reset due to
4029 	 * new firmware download, even though it is not common we cannot ignore
4030 	 * such scenario. Controller's non-mdts limits are reported in the unit
4031 	 * of logical blocks that is dependent on the format of attached
4032 	 * namespace. Hence re-read the limits at the time of ns allocation.
4033 	 */
4034 	ret = nvme_init_non_mdts_limits(ctrl);
4035 	if (ret < 0) {
4036 		dev_warn(ctrl->device,
4037 			"reading non-mdts-limits failed: %d\n", ret);
4038 		return;
4039 	}
4040 
4041 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4042 		dev_info(ctrl->device, "rescanning namespaces.\n");
4043 		nvme_clear_changed_ns_log(ctrl);
4044 	}
4045 
4046 	mutex_lock(&ctrl->scan_lock);
4047 	if (nvme_ctrl_limited_cns(ctrl)) {
4048 		nvme_scan_ns_sequential(ctrl);
4049 	} else {
4050 		/*
4051 		 * Fall back to sequential scan if DNR is set to handle broken
4052 		 * devices which should support Identify NS List (as per the VS
4053 		 * they report) but don't actually support it.
4054 		 */
4055 		ret = nvme_scan_ns_list(ctrl);
4056 		if (ret > 0 && ret & NVME_SC_DNR)
4057 			nvme_scan_ns_sequential(ctrl);
4058 	}
4059 	mutex_unlock(&ctrl->scan_lock);
4060 }
4061 
4062 /*
4063  * This function iterates the namespace list unlocked to allow recovery from
4064  * controller failure. It is up to the caller to ensure the namespace list is
4065  * not modified by scan work while this function is executing.
4066  */
4067 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4068 {
4069 	struct nvme_ns *ns, *next;
4070 	LIST_HEAD(ns_list);
4071 
4072 	/*
4073 	 * make sure to requeue I/O to all namespaces as these
4074 	 * might result from the scan itself and must complete
4075 	 * for the scan_work to make progress
4076 	 */
4077 	nvme_mpath_clear_ctrl_paths(ctrl);
4078 
4079 	/*
4080 	 * Unquiesce io queues so any pending IO won't hang, especially
4081 	 * those submitted from scan work
4082 	 */
4083 	nvme_unquiesce_io_queues(ctrl);
4084 
4085 	/* prevent racing with ns scanning */
4086 	flush_work(&ctrl->scan_work);
4087 
4088 	/*
4089 	 * The dead states indicates the controller was not gracefully
4090 	 * disconnected. In that case, we won't be able to flush any data while
4091 	 * removing the namespaces' disks; fail all the queues now to avoid
4092 	 * potentially having to clean up the failed sync later.
4093 	 */
4094 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_DEAD)
4095 		nvme_mark_namespaces_dead(ctrl);
4096 
4097 	/* this is a no-op when called from the controller reset handler */
4098 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4099 
4100 	down_write(&ctrl->namespaces_rwsem);
4101 	list_splice_init(&ctrl->namespaces, &ns_list);
4102 	up_write(&ctrl->namespaces_rwsem);
4103 
4104 	list_for_each_entry_safe(ns, next, &ns_list, list)
4105 		nvme_ns_remove(ns);
4106 }
4107 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4108 
4109 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env)
4110 {
4111 	const struct nvme_ctrl *ctrl =
4112 		container_of(dev, struct nvme_ctrl, ctrl_device);
4113 	struct nvmf_ctrl_options *opts = ctrl->opts;
4114 	int ret;
4115 
4116 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4117 	if (ret)
4118 		return ret;
4119 
4120 	if (opts) {
4121 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4122 		if (ret)
4123 			return ret;
4124 
4125 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4126 				opts->trsvcid ?: "none");
4127 		if (ret)
4128 			return ret;
4129 
4130 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4131 				opts->host_traddr ?: "none");
4132 		if (ret)
4133 			return ret;
4134 
4135 		ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4136 				opts->host_iface ?: "none");
4137 	}
4138 	return ret;
4139 }
4140 
4141 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4142 {
4143 	char *envp[2] = { envdata, NULL };
4144 
4145 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4146 }
4147 
4148 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4149 {
4150 	char *envp[2] = { NULL, NULL };
4151 	u32 aen_result = ctrl->aen_result;
4152 
4153 	ctrl->aen_result = 0;
4154 	if (!aen_result)
4155 		return;
4156 
4157 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4158 	if (!envp[0])
4159 		return;
4160 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4161 	kfree(envp[0]);
4162 }
4163 
4164 static void nvme_async_event_work(struct work_struct *work)
4165 {
4166 	struct nvme_ctrl *ctrl =
4167 		container_of(work, struct nvme_ctrl, async_event_work);
4168 
4169 	nvme_aen_uevent(ctrl);
4170 
4171 	/*
4172 	 * The transport drivers must guarantee AER submission here is safe by
4173 	 * flushing ctrl async_event_work after changing the controller state
4174 	 * from LIVE and before freeing the admin queue.
4175 	*/
4176 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
4177 		ctrl->ops->submit_async_event(ctrl);
4178 }
4179 
4180 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4181 {
4182 
4183 	u32 csts;
4184 
4185 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4186 		return false;
4187 
4188 	if (csts == ~0)
4189 		return false;
4190 
4191 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4192 }
4193 
4194 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4195 {
4196 	struct nvme_fw_slot_info_log *log;
4197 	u8 next_fw_slot, cur_fw_slot;
4198 
4199 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4200 	if (!log)
4201 		return;
4202 
4203 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4204 			 log, sizeof(*log), 0)) {
4205 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4206 		goto out_free_log;
4207 	}
4208 
4209 	cur_fw_slot = log->afi & 0x7;
4210 	next_fw_slot = (log->afi & 0x70) >> 4;
4211 	if (!cur_fw_slot || (next_fw_slot && (cur_fw_slot != next_fw_slot))) {
4212 		dev_info(ctrl->device,
4213 			 "Firmware is activated after next Controller Level Reset\n");
4214 		goto out_free_log;
4215 	}
4216 
4217 	memcpy(ctrl->subsys->firmware_rev, &log->frs[cur_fw_slot - 1],
4218 		sizeof(ctrl->subsys->firmware_rev));
4219 
4220 out_free_log:
4221 	kfree(log);
4222 }
4223 
4224 static void nvme_fw_act_work(struct work_struct *work)
4225 {
4226 	struct nvme_ctrl *ctrl = container_of(work,
4227 				struct nvme_ctrl, fw_act_work);
4228 	unsigned long fw_act_timeout;
4229 
4230 	nvme_auth_stop(ctrl);
4231 
4232 	if (ctrl->mtfa)
4233 		fw_act_timeout = jiffies +
4234 				msecs_to_jiffies(ctrl->mtfa * 100);
4235 	else
4236 		fw_act_timeout = jiffies +
4237 				msecs_to_jiffies(admin_timeout * 1000);
4238 
4239 	nvme_quiesce_io_queues(ctrl);
4240 	while (nvme_ctrl_pp_status(ctrl)) {
4241 		if (time_after(jiffies, fw_act_timeout)) {
4242 			dev_warn(ctrl->device,
4243 				"Fw activation timeout, reset controller\n");
4244 			nvme_try_sched_reset(ctrl);
4245 			return;
4246 		}
4247 		msleep(100);
4248 	}
4249 
4250 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4251 		return;
4252 
4253 	nvme_unquiesce_io_queues(ctrl);
4254 	/* read FW slot information to clear the AER */
4255 	nvme_get_fw_slot_info(ctrl);
4256 
4257 	queue_work(nvme_wq, &ctrl->async_event_work);
4258 }
4259 
4260 static u32 nvme_aer_type(u32 result)
4261 {
4262 	return result & 0x7;
4263 }
4264 
4265 static u32 nvme_aer_subtype(u32 result)
4266 {
4267 	return (result & 0xff00) >> 8;
4268 }
4269 
4270 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4271 {
4272 	u32 aer_notice_type = nvme_aer_subtype(result);
4273 	bool requeue = true;
4274 
4275 	switch (aer_notice_type) {
4276 	case NVME_AER_NOTICE_NS_CHANGED:
4277 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4278 		nvme_queue_scan(ctrl);
4279 		break;
4280 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4281 		/*
4282 		 * We are (ab)using the RESETTING state to prevent subsequent
4283 		 * recovery actions from interfering with the controller's
4284 		 * firmware activation.
4285 		 */
4286 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4287 			requeue = false;
4288 			queue_work(nvme_wq, &ctrl->fw_act_work);
4289 		}
4290 		break;
4291 #ifdef CONFIG_NVME_MULTIPATH
4292 	case NVME_AER_NOTICE_ANA:
4293 		if (!ctrl->ana_log_buf)
4294 			break;
4295 		queue_work(nvme_wq, &ctrl->ana_work);
4296 		break;
4297 #endif
4298 	case NVME_AER_NOTICE_DISC_CHANGED:
4299 		ctrl->aen_result = result;
4300 		break;
4301 	default:
4302 		dev_warn(ctrl->device, "async event result %08x\n", result);
4303 	}
4304 	return requeue;
4305 }
4306 
4307 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4308 {
4309 	dev_warn(ctrl->device, "resetting controller due to AER\n");
4310 	nvme_reset_ctrl(ctrl);
4311 }
4312 
4313 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4314 		volatile union nvme_result *res)
4315 {
4316 	u32 result = le32_to_cpu(res->u32);
4317 	u32 aer_type = nvme_aer_type(result);
4318 	u32 aer_subtype = nvme_aer_subtype(result);
4319 	bool requeue = true;
4320 
4321 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4322 		return;
4323 
4324 	trace_nvme_async_event(ctrl, result);
4325 	switch (aer_type) {
4326 	case NVME_AER_NOTICE:
4327 		requeue = nvme_handle_aen_notice(ctrl, result);
4328 		break;
4329 	case NVME_AER_ERROR:
4330 		/*
4331 		 * For a persistent internal error, don't run async_event_work
4332 		 * to submit a new AER. The controller reset will do it.
4333 		 */
4334 		if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4335 			nvme_handle_aer_persistent_error(ctrl);
4336 			return;
4337 		}
4338 		fallthrough;
4339 	case NVME_AER_SMART:
4340 	case NVME_AER_CSS:
4341 	case NVME_AER_VS:
4342 		ctrl->aen_result = result;
4343 		break;
4344 	default:
4345 		break;
4346 	}
4347 
4348 	if (requeue)
4349 		queue_work(nvme_wq, &ctrl->async_event_work);
4350 }
4351 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4352 
4353 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4354 		const struct blk_mq_ops *ops, unsigned int cmd_size)
4355 {
4356 	int ret;
4357 
4358 	memset(set, 0, sizeof(*set));
4359 	set->ops = ops;
4360 	set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4361 	if (ctrl->ops->flags & NVME_F_FABRICS)
4362 		set->reserved_tags = NVMF_RESERVED_TAGS;
4363 	set->numa_node = ctrl->numa_node;
4364 	set->flags = BLK_MQ_F_NO_SCHED;
4365 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4366 		set->flags |= BLK_MQ_F_BLOCKING;
4367 	set->cmd_size = cmd_size;
4368 	set->driver_data = ctrl;
4369 	set->nr_hw_queues = 1;
4370 	set->timeout = NVME_ADMIN_TIMEOUT;
4371 	ret = blk_mq_alloc_tag_set(set);
4372 	if (ret)
4373 		return ret;
4374 
4375 	ctrl->admin_q = blk_mq_init_queue(set);
4376 	if (IS_ERR(ctrl->admin_q)) {
4377 		ret = PTR_ERR(ctrl->admin_q);
4378 		goto out_free_tagset;
4379 	}
4380 
4381 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4382 		ctrl->fabrics_q = blk_mq_init_queue(set);
4383 		if (IS_ERR(ctrl->fabrics_q)) {
4384 			ret = PTR_ERR(ctrl->fabrics_q);
4385 			goto out_cleanup_admin_q;
4386 		}
4387 	}
4388 
4389 	ctrl->admin_tagset = set;
4390 	return 0;
4391 
4392 out_cleanup_admin_q:
4393 	blk_mq_destroy_queue(ctrl->admin_q);
4394 	blk_put_queue(ctrl->admin_q);
4395 out_free_tagset:
4396 	blk_mq_free_tag_set(set);
4397 	ctrl->admin_q = NULL;
4398 	ctrl->fabrics_q = NULL;
4399 	return ret;
4400 }
4401 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4402 
4403 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4404 {
4405 	blk_mq_destroy_queue(ctrl->admin_q);
4406 	blk_put_queue(ctrl->admin_q);
4407 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4408 		blk_mq_destroy_queue(ctrl->fabrics_q);
4409 		blk_put_queue(ctrl->fabrics_q);
4410 	}
4411 	blk_mq_free_tag_set(ctrl->admin_tagset);
4412 }
4413 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4414 
4415 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4416 		const struct blk_mq_ops *ops, unsigned int nr_maps,
4417 		unsigned int cmd_size)
4418 {
4419 	int ret;
4420 
4421 	memset(set, 0, sizeof(*set));
4422 	set->ops = ops;
4423 	set->queue_depth = min_t(unsigned, ctrl->sqsize, BLK_MQ_MAX_DEPTH - 1);
4424 	/*
4425 	 * Some Apple controllers requires tags to be unique across admin and
4426 	 * the (only) I/O queue, so reserve the first 32 tags of the I/O queue.
4427 	 */
4428 	if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
4429 		set->reserved_tags = NVME_AQ_DEPTH;
4430 	else if (ctrl->ops->flags & NVME_F_FABRICS)
4431 		set->reserved_tags = NVMF_RESERVED_TAGS;
4432 	set->numa_node = ctrl->numa_node;
4433 	set->flags = BLK_MQ_F_SHOULD_MERGE;
4434 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4435 		set->flags |= BLK_MQ_F_BLOCKING;
4436 	set->cmd_size = cmd_size,
4437 	set->driver_data = ctrl;
4438 	set->nr_hw_queues = ctrl->queue_count - 1;
4439 	set->timeout = NVME_IO_TIMEOUT;
4440 	set->nr_maps = nr_maps;
4441 	ret = blk_mq_alloc_tag_set(set);
4442 	if (ret)
4443 		return ret;
4444 
4445 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4446 		ctrl->connect_q = blk_mq_init_queue(set);
4447         	if (IS_ERR(ctrl->connect_q)) {
4448 			ret = PTR_ERR(ctrl->connect_q);
4449 			goto out_free_tag_set;
4450 		}
4451 		blk_queue_flag_set(QUEUE_FLAG_SKIP_TAGSET_QUIESCE,
4452 				   ctrl->connect_q);
4453 	}
4454 
4455 	ctrl->tagset = set;
4456 	return 0;
4457 
4458 out_free_tag_set:
4459 	blk_mq_free_tag_set(set);
4460 	ctrl->connect_q = NULL;
4461 	return ret;
4462 }
4463 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
4464 
4465 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
4466 {
4467 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4468 		blk_mq_destroy_queue(ctrl->connect_q);
4469 		blk_put_queue(ctrl->connect_q);
4470 	}
4471 	blk_mq_free_tag_set(ctrl->tagset);
4472 }
4473 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
4474 
4475 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4476 {
4477 	nvme_mpath_stop(ctrl);
4478 	nvme_auth_stop(ctrl);
4479 	nvme_stop_keep_alive(ctrl);
4480 	nvme_stop_failfast_work(ctrl);
4481 	flush_work(&ctrl->async_event_work);
4482 	cancel_work_sync(&ctrl->fw_act_work);
4483 	if (ctrl->ops->stop_ctrl)
4484 		ctrl->ops->stop_ctrl(ctrl);
4485 }
4486 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4487 
4488 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4489 {
4490 	nvme_enable_aen(ctrl);
4491 
4492 	/*
4493 	 * persistent discovery controllers need to send indication to userspace
4494 	 * to re-read the discovery log page to learn about possible changes
4495 	 * that were missed. We identify persistent discovery controllers by
4496 	 * checking that they started once before, hence are reconnecting back.
4497 	 */
4498 	if (test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
4499 	    nvme_discovery_ctrl(ctrl))
4500 		nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
4501 
4502 	if (ctrl->queue_count > 1) {
4503 		nvme_queue_scan(ctrl);
4504 		nvme_unquiesce_io_queues(ctrl);
4505 		nvme_mpath_update(ctrl);
4506 	}
4507 
4508 	nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4509 	set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags);
4510 }
4511 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4512 
4513 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4514 {
4515 	nvme_hwmon_exit(ctrl);
4516 	nvme_fault_inject_fini(&ctrl->fault_inject);
4517 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4518 	cdev_device_del(&ctrl->cdev, ctrl->device);
4519 	nvme_put_ctrl(ctrl);
4520 }
4521 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4522 
4523 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4524 {
4525 	struct nvme_effects_log	*cel;
4526 	unsigned long i;
4527 
4528 	xa_for_each(&ctrl->cels, i, cel) {
4529 		xa_erase(&ctrl->cels, i);
4530 		kfree(cel);
4531 	}
4532 
4533 	xa_destroy(&ctrl->cels);
4534 }
4535 
4536 static void nvme_free_ctrl(struct device *dev)
4537 {
4538 	struct nvme_ctrl *ctrl =
4539 		container_of(dev, struct nvme_ctrl, ctrl_device);
4540 	struct nvme_subsystem *subsys = ctrl->subsys;
4541 
4542 	if (!subsys || ctrl->instance != subsys->instance)
4543 		ida_free(&nvme_instance_ida, ctrl->instance);
4544 	key_put(ctrl->tls_key);
4545 	nvme_free_cels(ctrl);
4546 	nvme_mpath_uninit(ctrl);
4547 	nvme_auth_stop(ctrl);
4548 	nvme_auth_free(ctrl);
4549 	__free_page(ctrl->discard_page);
4550 	free_opal_dev(ctrl->opal_dev);
4551 
4552 	if (subsys) {
4553 		mutex_lock(&nvme_subsystems_lock);
4554 		list_del(&ctrl->subsys_entry);
4555 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4556 		mutex_unlock(&nvme_subsystems_lock);
4557 	}
4558 
4559 	ctrl->ops->free_ctrl(ctrl);
4560 
4561 	if (subsys)
4562 		nvme_put_subsystem(subsys);
4563 }
4564 
4565 /*
4566  * Initialize a NVMe controller structures.  This needs to be called during
4567  * earliest initialization so that we have the initialized structured around
4568  * during probing.
4569  */
4570 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4571 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
4572 {
4573 	int ret;
4574 
4575 	WRITE_ONCE(ctrl->state, NVME_CTRL_NEW);
4576 	ctrl->passthru_err_log_enabled = false;
4577 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4578 	spin_lock_init(&ctrl->lock);
4579 	mutex_init(&ctrl->scan_lock);
4580 	INIT_LIST_HEAD(&ctrl->namespaces);
4581 	xa_init(&ctrl->cels);
4582 	init_rwsem(&ctrl->namespaces_rwsem);
4583 	ctrl->dev = dev;
4584 	ctrl->ops = ops;
4585 	ctrl->quirks = quirks;
4586 	ctrl->numa_node = NUMA_NO_NODE;
4587 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4588 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4589 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4590 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4591 	init_waitqueue_head(&ctrl->state_wq);
4592 
4593 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4594 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4595 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4596 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4597 	ctrl->ka_last_check_time = jiffies;
4598 
4599 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4600 			PAGE_SIZE);
4601 	ctrl->discard_page = alloc_page(GFP_KERNEL);
4602 	if (!ctrl->discard_page) {
4603 		ret = -ENOMEM;
4604 		goto out;
4605 	}
4606 
4607 	ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4608 	if (ret < 0)
4609 		goto out;
4610 	ctrl->instance = ret;
4611 
4612 	device_initialize(&ctrl->ctrl_device);
4613 	ctrl->device = &ctrl->ctrl_device;
4614 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4615 			ctrl->instance);
4616 	ctrl->device->class = nvme_class;
4617 	ctrl->device->parent = ctrl->dev;
4618 	if (ops->dev_attr_groups)
4619 		ctrl->device->groups = ops->dev_attr_groups;
4620 	else
4621 		ctrl->device->groups = nvme_dev_attr_groups;
4622 	ctrl->device->release = nvme_free_ctrl;
4623 	dev_set_drvdata(ctrl->device, ctrl);
4624 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4625 	if (ret)
4626 		goto out_release_instance;
4627 
4628 	nvme_get_ctrl(ctrl);
4629 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
4630 	ctrl->cdev.owner = ops->module;
4631 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4632 	if (ret)
4633 		goto out_free_name;
4634 
4635 	/*
4636 	 * Initialize latency tolerance controls.  The sysfs files won't
4637 	 * be visible to userspace unless the device actually supports APST.
4638 	 */
4639 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4640 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4641 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4642 
4643 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4644 	nvme_mpath_init_ctrl(ctrl);
4645 	ret = nvme_auth_init_ctrl(ctrl);
4646 	if (ret)
4647 		goto out_free_cdev;
4648 
4649 	return 0;
4650 out_free_cdev:
4651 	nvme_fault_inject_fini(&ctrl->fault_inject);
4652 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4653 	cdev_device_del(&ctrl->cdev, ctrl->device);
4654 out_free_name:
4655 	nvme_put_ctrl(ctrl);
4656 	kfree_const(ctrl->device->kobj.name);
4657 out_release_instance:
4658 	ida_free(&nvme_instance_ida, ctrl->instance);
4659 out:
4660 	if (ctrl->discard_page)
4661 		__free_page(ctrl->discard_page);
4662 	return ret;
4663 }
4664 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4665 
4666 /* let I/O to all namespaces fail in preparation for surprise removal */
4667 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
4668 {
4669 	struct nvme_ns *ns;
4670 
4671 	down_read(&ctrl->namespaces_rwsem);
4672 	list_for_each_entry(ns, &ctrl->namespaces, list)
4673 		blk_mark_disk_dead(ns->disk);
4674 	up_read(&ctrl->namespaces_rwsem);
4675 }
4676 EXPORT_SYMBOL_GPL(nvme_mark_namespaces_dead);
4677 
4678 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4679 {
4680 	struct nvme_ns *ns;
4681 
4682 	down_read(&ctrl->namespaces_rwsem);
4683 	list_for_each_entry(ns, &ctrl->namespaces, list)
4684 		blk_mq_unfreeze_queue(ns->queue);
4685 	up_read(&ctrl->namespaces_rwsem);
4686 	clear_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4687 }
4688 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4689 
4690 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4691 {
4692 	struct nvme_ns *ns;
4693 
4694 	down_read(&ctrl->namespaces_rwsem);
4695 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4696 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4697 		if (timeout <= 0)
4698 			break;
4699 	}
4700 	up_read(&ctrl->namespaces_rwsem);
4701 	return timeout;
4702 }
4703 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4704 
4705 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4706 {
4707 	struct nvme_ns *ns;
4708 
4709 	down_read(&ctrl->namespaces_rwsem);
4710 	list_for_each_entry(ns, &ctrl->namespaces, list)
4711 		blk_mq_freeze_queue_wait(ns->queue);
4712 	up_read(&ctrl->namespaces_rwsem);
4713 }
4714 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4715 
4716 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4717 {
4718 	struct nvme_ns *ns;
4719 
4720 	set_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4721 	down_read(&ctrl->namespaces_rwsem);
4722 	list_for_each_entry(ns, &ctrl->namespaces, list)
4723 		blk_freeze_queue_start(ns->queue);
4724 	up_read(&ctrl->namespaces_rwsem);
4725 }
4726 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4727 
4728 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl)
4729 {
4730 	if (!ctrl->tagset)
4731 		return;
4732 	if (!test_and_set_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4733 		blk_mq_quiesce_tagset(ctrl->tagset);
4734 	else
4735 		blk_mq_wait_quiesce_done(ctrl->tagset);
4736 }
4737 EXPORT_SYMBOL_GPL(nvme_quiesce_io_queues);
4738 
4739 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl)
4740 {
4741 	if (!ctrl->tagset)
4742 		return;
4743 	if (test_and_clear_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4744 		blk_mq_unquiesce_tagset(ctrl->tagset);
4745 }
4746 EXPORT_SYMBOL_GPL(nvme_unquiesce_io_queues);
4747 
4748 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl)
4749 {
4750 	if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4751 		blk_mq_quiesce_queue(ctrl->admin_q);
4752 	else
4753 		blk_mq_wait_quiesce_done(ctrl->admin_q->tag_set);
4754 }
4755 EXPORT_SYMBOL_GPL(nvme_quiesce_admin_queue);
4756 
4757 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl)
4758 {
4759 	if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4760 		blk_mq_unquiesce_queue(ctrl->admin_q);
4761 }
4762 EXPORT_SYMBOL_GPL(nvme_unquiesce_admin_queue);
4763 
4764 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4765 {
4766 	struct nvme_ns *ns;
4767 
4768 	down_read(&ctrl->namespaces_rwsem);
4769 	list_for_each_entry(ns, &ctrl->namespaces, list)
4770 		blk_sync_queue(ns->queue);
4771 	up_read(&ctrl->namespaces_rwsem);
4772 }
4773 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4774 
4775 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4776 {
4777 	nvme_sync_io_queues(ctrl);
4778 	if (ctrl->admin_q)
4779 		blk_sync_queue(ctrl->admin_q);
4780 }
4781 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4782 
4783 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4784 {
4785 	if (file->f_op != &nvme_dev_fops)
4786 		return NULL;
4787 	return file->private_data;
4788 }
4789 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4790 
4791 /*
4792  * Check we didn't inadvertently grow the command structure sizes:
4793  */
4794 static inline void _nvme_check_size(void)
4795 {
4796 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4797 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4798 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4799 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4800 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4801 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4802 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4803 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4804 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4805 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4806 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4807 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4808 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4809 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
4810 			NVME_IDENTIFY_DATA_SIZE);
4811 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4812 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
4813 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4814 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4815 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4816 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4817 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4818 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4819 	BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
4820 }
4821 
4822 
4823 static int __init nvme_core_init(void)
4824 {
4825 	int result = -ENOMEM;
4826 
4827 	_nvme_check_size();
4828 
4829 	nvme_wq = alloc_workqueue("nvme-wq",
4830 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4831 	if (!nvme_wq)
4832 		goto out;
4833 
4834 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4835 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4836 	if (!nvme_reset_wq)
4837 		goto destroy_wq;
4838 
4839 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4840 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4841 	if (!nvme_delete_wq)
4842 		goto destroy_reset_wq;
4843 
4844 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4845 			NVME_MINORS, "nvme");
4846 	if (result < 0)
4847 		goto destroy_delete_wq;
4848 
4849 	nvme_class = class_create("nvme");
4850 	if (IS_ERR(nvme_class)) {
4851 		result = PTR_ERR(nvme_class);
4852 		goto unregister_chrdev;
4853 	}
4854 	nvme_class->dev_uevent = nvme_class_uevent;
4855 
4856 	nvme_subsys_class = class_create("nvme-subsystem");
4857 	if (IS_ERR(nvme_subsys_class)) {
4858 		result = PTR_ERR(nvme_subsys_class);
4859 		goto destroy_class;
4860 	}
4861 
4862 	result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4863 				     "nvme-generic");
4864 	if (result < 0)
4865 		goto destroy_subsys_class;
4866 
4867 	nvme_ns_chr_class = class_create("nvme-generic");
4868 	if (IS_ERR(nvme_ns_chr_class)) {
4869 		result = PTR_ERR(nvme_ns_chr_class);
4870 		goto unregister_generic_ns;
4871 	}
4872 	result = nvme_init_auth();
4873 	if (result)
4874 		goto destroy_ns_chr;
4875 	return 0;
4876 
4877 destroy_ns_chr:
4878 	class_destroy(nvme_ns_chr_class);
4879 unregister_generic_ns:
4880 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4881 destroy_subsys_class:
4882 	class_destroy(nvme_subsys_class);
4883 destroy_class:
4884 	class_destroy(nvme_class);
4885 unregister_chrdev:
4886 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4887 destroy_delete_wq:
4888 	destroy_workqueue(nvme_delete_wq);
4889 destroy_reset_wq:
4890 	destroy_workqueue(nvme_reset_wq);
4891 destroy_wq:
4892 	destroy_workqueue(nvme_wq);
4893 out:
4894 	return result;
4895 }
4896 
4897 static void __exit nvme_core_exit(void)
4898 {
4899 	nvme_exit_auth();
4900 	class_destroy(nvme_ns_chr_class);
4901 	class_destroy(nvme_subsys_class);
4902 	class_destroy(nvme_class);
4903 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4904 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4905 	destroy_workqueue(nvme_delete_wq);
4906 	destroy_workqueue(nvme_reset_wq);
4907 	destroy_workqueue(nvme_wq);
4908 	ida_destroy(&nvme_ns_chr_minor_ida);
4909 	ida_destroy(&nvme_instance_ida);
4910 }
4911 
4912 MODULE_LICENSE("GPL");
4913 MODULE_VERSION("1.0");
4914 MODULE_DESCRIPTION("NVMe host core framework");
4915 module_init(nvme_core_init);
4916 module_exit(nvme_core_exit);
4917