xref: /linux/drivers/nvme/host/core.c (revision eb01fe7abbe2d0b38824d2a93fdb4cc3eaf2ccc1)
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 int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env);
118 static const struct class nvme_class = {
119 	.name = "nvme",
120 	.dev_uevent = nvme_class_uevent,
121 };
122 
123 static const struct class nvme_subsys_class = {
124 	.name = "nvme-subsystem",
125 };
126 
127 static DEFINE_IDA(nvme_ns_chr_minor_ida);
128 static dev_t nvme_ns_chr_devt;
129 static const struct class nvme_ns_chr_class = {
130 	.name = "nvme-generic",
131 };
132 
133 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
134 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
135 					   unsigned nsid);
136 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
137 				   struct nvme_command *cmd);
138 
139 void nvme_queue_scan(struct nvme_ctrl *ctrl)
140 {
141 	/*
142 	 * Only new queue scan work when admin and IO queues are both alive
143 	 */
144 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE && ctrl->tagset)
145 		queue_work(nvme_wq, &ctrl->scan_work);
146 }
147 
148 /*
149  * Use this function to proceed with scheduling reset_work for a controller
150  * that had previously been set to the resetting state. This is intended for
151  * code paths that can't be interrupted by other reset attempts. A hot removal
152  * may prevent this from succeeding.
153  */
154 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
155 {
156 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_RESETTING)
157 		return -EBUSY;
158 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
159 		return -EBUSY;
160 	return 0;
161 }
162 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
163 
164 static void nvme_failfast_work(struct work_struct *work)
165 {
166 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
167 			struct nvme_ctrl, failfast_work);
168 
169 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_CONNECTING)
170 		return;
171 
172 	set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
173 	dev_info(ctrl->device, "failfast expired\n");
174 	nvme_kick_requeue_lists(ctrl);
175 }
176 
177 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
178 {
179 	if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
180 		return;
181 
182 	schedule_delayed_work(&ctrl->failfast_work,
183 			      ctrl->opts->fast_io_fail_tmo * HZ);
184 }
185 
186 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
187 {
188 	if (!ctrl->opts)
189 		return;
190 
191 	cancel_delayed_work_sync(&ctrl->failfast_work);
192 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
193 }
194 
195 
196 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
197 {
198 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
199 		return -EBUSY;
200 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
201 		return -EBUSY;
202 	return 0;
203 }
204 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
205 
206 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
207 {
208 	int ret;
209 
210 	ret = nvme_reset_ctrl(ctrl);
211 	if (!ret) {
212 		flush_work(&ctrl->reset_work);
213 		if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE)
214 			ret = -ENETRESET;
215 	}
216 
217 	return ret;
218 }
219 
220 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
221 {
222 	dev_info(ctrl->device,
223 		 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
224 
225 	flush_work(&ctrl->reset_work);
226 	nvme_stop_ctrl(ctrl);
227 	nvme_remove_namespaces(ctrl);
228 	ctrl->ops->delete_ctrl(ctrl);
229 	nvme_uninit_ctrl(ctrl);
230 }
231 
232 static void nvme_delete_ctrl_work(struct work_struct *work)
233 {
234 	struct nvme_ctrl *ctrl =
235 		container_of(work, struct nvme_ctrl, delete_work);
236 
237 	nvme_do_delete_ctrl(ctrl);
238 }
239 
240 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
241 {
242 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
243 		return -EBUSY;
244 	if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
245 		return -EBUSY;
246 	return 0;
247 }
248 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
249 
250 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
251 {
252 	/*
253 	 * Keep a reference until nvme_do_delete_ctrl() complete,
254 	 * since ->delete_ctrl can free the controller.
255 	 */
256 	nvme_get_ctrl(ctrl);
257 	if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
258 		nvme_do_delete_ctrl(ctrl);
259 	nvme_put_ctrl(ctrl);
260 }
261 
262 static blk_status_t nvme_error_status(u16 status)
263 {
264 	switch (status & 0x7ff) {
265 	case NVME_SC_SUCCESS:
266 		return BLK_STS_OK;
267 	case NVME_SC_CAP_EXCEEDED:
268 		return BLK_STS_NOSPC;
269 	case NVME_SC_LBA_RANGE:
270 	case NVME_SC_CMD_INTERRUPTED:
271 	case NVME_SC_NS_NOT_READY:
272 		return BLK_STS_TARGET;
273 	case NVME_SC_BAD_ATTRIBUTES:
274 	case NVME_SC_ONCS_NOT_SUPPORTED:
275 	case NVME_SC_INVALID_OPCODE:
276 	case NVME_SC_INVALID_FIELD:
277 	case NVME_SC_INVALID_NS:
278 		return BLK_STS_NOTSUPP;
279 	case NVME_SC_WRITE_FAULT:
280 	case NVME_SC_READ_ERROR:
281 	case NVME_SC_UNWRITTEN_BLOCK:
282 	case NVME_SC_ACCESS_DENIED:
283 	case NVME_SC_READ_ONLY:
284 	case NVME_SC_COMPARE_FAILED:
285 		return BLK_STS_MEDIUM;
286 	case NVME_SC_GUARD_CHECK:
287 	case NVME_SC_APPTAG_CHECK:
288 	case NVME_SC_REFTAG_CHECK:
289 	case NVME_SC_INVALID_PI:
290 		return BLK_STS_PROTECTION;
291 	case NVME_SC_RESERVATION_CONFLICT:
292 		return BLK_STS_RESV_CONFLICT;
293 	case NVME_SC_HOST_PATH_ERROR:
294 		return BLK_STS_TRANSPORT;
295 	case NVME_SC_ZONE_TOO_MANY_ACTIVE:
296 		return BLK_STS_ZONE_ACTIVE_RESOURCE;
297 	case NVME_SC_ZONE_TOO_MANY_OPEN:
298 		return BLK_STS_ZONE_OPEN_RESOURCE;
299 	default:
300 		return BLK_STS_IOERR;
301 	}
302 }
303 
304 static void nvme_retry_req(struct request *req)
305 {
306 	unsigned long delay = 0;
307 	u16 crd;
308 
309 	/* The mask and shift result must be <= 3 */
310 	crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
311 	if (crd)
312 		delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
313 
314 	nvme_req(req)->retries++;
315 	blk_mq_requeue_request(req, false);
316 	blk_mq_delay_kick_requeue_list(req->q, delay);
317 }
318 
319 static void nvme_log_error(struct request *req)
320 {
321 	struct nvme_ns *ns = req->q->queuedata;
322 	struct nvme_request *nr = nvme_req(req);
323 
324 	if (ns) {
325 		pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %u blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
326 		       ns->disk ? ns->disk->disk_name : "?",
327 		       nvme_get_opcode_str(nr->cmd->common.opcode),
328 		       nr->cmd->common.opcode,
329 		       nvme_sect_to_lba(ns->head, blk_rq_pos(req)),
330 		       blk_rq_bytes(req) >> ns->head->lba_shift,
331 		       nvme_get_error_status_str(nr->status),
332 		       nr->status >> 8 & 7,	/* Status Code Type */
333 		       nr->status & 0xff,	/* Status Code */
334 		       nr->status & NVME_SC_MORE ? "MORE " : "",
335 		       nr->status & NVME_SC_DNR  ? "DNR "  : "");
336 		return;
337 	}
338 
339 	pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
340 			   dev_name(nr->ctrl->device),
341 			   nvme_get_admin_opcode_str(nr->cmd->common.opcode),
342 			   nr->cmd->common.opcode,
343 			   nvme_get_error_status_str(nr->status),
344 			   nr->status >> 8 & 7,	/* Status Code Type */
345 			   nr->status & 0xff,	/* Status Code */
346 			   nr->status & NVME_SC_MORE ? "MORE " : "",
347 			   nr->status & NVME_SC_DNR  ? "DNR "  : "");
348 }
349 
350 static void nvme_log_err_passthru(struct request *req)
351 {
352 	struct nvme_ns *ns = req->q->queuedata;
353 	struct nvme_request *nr = nvme_req(req);
354 
355 	pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s"
356 		"cdw10=0x%x cdw11=0x%x cdw12=0x%x cdw13=0x%x cdw14=0x%x cdw15=0x%x\n",
357 		ns ? ns->disk->disk_name : dev_name(nr->ctrl->device),
358 		ns ? nvme_get_opcode_str(nr->cmd->common.opcode) :
359 		     nvme_get_admin_opcode_str(nr->cmd->common.opcode),
360 		nr->cmd->common.opcode,
361 		nvme_get_error_status_str(nr->status),
362 		nr->status >> 8 & 7,	/* Status Code Type */
363 		nr->status & 0xff,	/* Status Code */
364 		nr->status & NVME_SC_MORE ? "MORE " : "",
365 		nr->status & NVME_SC_DNR  ? "DNR "  : "",
366 		nr->cmd->common.cdw10,
367 		nr->cmd->common.cdw11,
368 		nr->cmd->common.cdw12,
369 		nr->cmd->common.cdw13,
370 		nr->cmd->common.cdw14,
371 		nr->cmd->common.cdw14);
372 }
373 
374 enum nvme_disposition {
375 	COMPLETE,
376 	RETRY,
377 	FAILOVER,
378 	AUTHENTICATE,
379 };
380 
381 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
382 {
383 	if (likely(nvme_req(req)->status == 0))
384 		return COMPLETE;
385 
386 	if ((nvme_req(req)->status & 0x7ff) == NVME_SC_AUTH_REQUIRED)
387 		return AUTHENTICATE;
388 
389 	if (blk_noretry_request(req) ||
390 	    (nvme_req(req)->status & NVME_SC_DNR) ||
391 	    nvme_req(req)->retries >= nvme_max_retries)
392 		return COMPLETE;
393 
394 	if (req->cmd_flags & REQ_NVME_MPATH) {
395 		if (nvme_is_path_error(nvme_req(req)->status) ||
396 		    blk_queue_dying(req->q))
397 			return FAILOVER;
398 	} else {
399 		if (blk_queue_dying(req->q))
400 			return COMPLETE;
401 	}
402 
403 	return RETRY;
404 }
405 
406 static inline void nvme_end_req_zoned(struct request *req)
407 {
408 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
409 	    req_op(req) == REQ_OP_ZONE_APPEND) {
410 		struct nvme_ns *ns = req->q->queuedata;
411 
412 		req->__sector = nvme_lba_to_sect(ns->head,
413 			le64_to_cpu(nvme_req(req)->result.u64));
414 	}
415 }
416 
417 static inline void nvme_end_req(struct request *req)
418 {
419 	blk_status_t status = nvme_error_status(nvme_req(req)->status);
420 
421 	if (unlikely(nvme_req(req)->status && !(req->rq_flags & RQF_QUIET))) {
422 		if (blk_rq_is_passthrough(req))
423 			nvme_log_err_passthru(req);
424 		else
425 			nvme_log_error(req);
426 	}
427 	nvme_end_req_zoned(req);
428 	nvme_trace_bio_complete(req);
429 	if (req->cmd_flags & REQ_NVME_MPATH)
430 		nvme_mpath_end_request(req);
431 	blk_mq_end_request(req, status);
432 }
433 
434 void nvme_complete_rq(struct request *req)
435 {
436 	struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
437 
438 	trace_nvme_complete_rq(req);
439 	nvme_cleanup_cmd(req);
440 
441 	/*
442 	 * Completions of long-running commands should not be able to
443 	 * defer sending of periodic keep alives, since the controller
444 	 * may have completed processing such commands a long time ago
445 	 * (arbitrarily close to command submission time).
446 	 * req->deadline - req->timeout is the command submission time
447 	 * in jiffies.
448 	 */
449 	if (ctrl->kas &&
450 	    req->deadline - req->timeout >= ctrl->ka_last_check_time)
451 		ctrl->comp_seen = true;
452 
453 	switch (nvme_decide_disposition(req)) {
454 	case COMPLETE:
455 		nvme_end_req(req);
456 		return;
457 	case RETRY:
458 		nvme_retry_req(req);
459 		return;
460 	case FAILOVER:
461 		nvme_failover_req(req);
462 		return;
463 	case AUTHENTICATE:
464 #ifdef CONFIG_NVME_HOST_AUTH
465 		queue_work(nvme_wq, &ctrl->dhchap_auth_work);
466 		nvme_retry_req(req);
467 #else
468 		nvme_end_req(req);
469 #endif
470 		return;
471 	}
472 }
473 EXPORT_SYMBOL_GPL(nvme_complete_rq);
474 
475 void nvme_complete_batch_req(struct request *req)
476 {
477 	trace_nvme_complete_rq(req);
478 	nvme_cleanup_cmd(req);
479 	nvme_end_req_zoned(req);
480 }
481 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
482 
483 /*
484  * Called to unwind from ->queue_rq on a failed command submission so that the
485  * multipathing code gets called to potentially failover to another path.
486  * The caller needs to unwind all transport specific resource allocations and
487  * must return propagate the return value.
488  */
489 blk_status_t nvme_host_path_error(struct request *req)
490 {
491 	nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
492 	blk_mq_set_request_complete(req);
493 	nvme_complete_rq(req);
494 	return BLK_STS_OK;
495 }
496 EXPORT_SYMBOL_GPL(nvme_host_path_error);
497 
498 bool nvme_cancel_request(struct request *req, void *data)
499 {
500 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
501 				"Cancelling I/O %d", req->tag);
502 
503 	/* don't abort one completed or idle request */
504 	if (blk_mq_rq_state(req) != MQ_RQ_IN_FLIGHT)
505 		return true;
506 
507 	nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
508 	nvme_req(req)->flags |= NVME_REQ_CANCELLED;
509 	blk_mq_complete_request(req);
510 	return true;
511 }
512 EXPORT_SYMBOL_GPL(nvme_cancel_request);
513 
514 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
515 {
516 	if (ctrl->tagset) {
517 		blk_mq_tagset_busy_iter(ctrl->tagset,
518 				nvme_cancel_request, ctrl);
519 		blk_mq_tagset_wait_completed_request(ctrl->tagset);
520 	}
521 }
522 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
523 
524 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
525 {
526 	if (ctrl->admin_tagset) {
527 		blk_mq_tagset_busy_iter(ctrl->admin_tagset,
528 				nvme_cancel_request, ctrl);
529 		blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
530 	}
531 }
532 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
533 
534 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
535 		enum nvme_ctrl_state new_state)
536 {
537 	enum nvme_ctrl_state old_state;
538 	unsigned long flags;
539 	bool changed = false;
540 
541 	spin_lock_irqsave(&ctrl->lock, flags);
542 
543 	old_state = nvme_ctrl_state(ctrl);
544 	switch (new_state) {
545 	case NVME_CTRL_LIVE:
546 		switch (old_state) {
547 		case NVME_CTRL_NEW:
548 		case NVME_CTRL_RESETTING:
549 		case NVME_CTRL_CONNECTING:
550 			changed = true;
551 			fallthrough;
552 		default:
553 			break;
554 		}
555 		break;
556 	case NVME_CTRL_RESETTING:
557 		switch (old_state) {
558 		case NVME_CTRL_NEW:
559 		case NVME_CTRL_LIVE:
560 			changed = true;
561 			fallthrough;
562 		default:
563 			break;
564 		}
565 		break;
566 	case NVME_CTRL_CONNECTING:
567 		switch (old_state) {
568 		case NVME_CTRL_NEW:
569 		case NVME_CTRL_RESETTING:
570 			changed = true;
571 			fallthrough;
572 		default:
573 			break;
574 		}
575 		break;
576 	case NVME_CTRL_DELETING:
577 		switch (old_state) {
578 		case NVME_CTRL_LIVE:
579 		case NVME_CTRL_RESETTING:
580 		case NVME_CTRL_CONNECTING:
581 			changed = true;
582 			fallthrough;
583 		default:
584 			break;
585 		}
586 		break;
587 	case NVME_CTRL_DELETING_NOIO:
588 		switch (old_state) {
589 		case NVME_CTRL_DELETING:
590 		case NVME_CTRL_DEAD:
591 			changed = true;
592 			fallthrough;
593 		default:
594 			break;
595 		}
596 		break;
597 	case NVME_CTRL_DEAD:
598 		switch (old_state) {
599 		case NVME_CTRL_DELETING:
600 			changed = true;
601 			fallthrough;
602 		default:
603 			break;
604 		}
605 		break;
606 	default:
607 		break;
608 	}
609 
610 	if (changed) {
611 		WRITE_ONCE(ctrl->state, new_state);
612 		wake_up_all(&ctrl->state_wq);
613 	}
614 
615 	spin_unlock_irqrestore(&ctrl->lock, flags);
616 	if (!changed)
617 		return false;
618 
619 	if (new_state == NVME_CTRL_LIVE) {
620 		if (old_state == NVME_CTRL_CONNECTING)
621 			nvme_stop_failfast_work(ctrl);
622 		nvme_kick_requeue_lists(ctrl);
623 	} else if (new_state == NVME_CTRL_CONNECTING &&
624 		old_state == NVME_CTRL_RESETTING) {
625 		nvme_start_failfast_work(ctrl);
626 	}
627 	return changed;
628 }
629 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
630 
631 /*
632  * Returns true for sink states that can't ever transition back to live.
633  */
634 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
635 {
636 	switch (nvme_ctrl_state(ctrl)) {
637 	case NVME_CTRL_NEW:
638 	case NVME_CTRL_LIVE:
639 	case NVME_CTRL_RESETTING:
640 	case NVME_CTRL_CONNECTING:
641 		return false;
642 	case NVME_CTRL_DELETING:
643 	case NVME_CTRL_DELETING_NOIO:
644 	case NVME_CTRL_DEAD:
645 		return true;
646 	default:
647 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
648 		return true;
649 	}
650 }
651 
652 /*
653  * Waits for the controller state to be resetting, or returns false if it is
654  * not possible to ever transition to that state.
655  */
656 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
657 {
658 	wait_event(ctrl->state_wq,
659 		   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
660 		   nvme_state_terminal(ctrl));
661 	return nvme_ctrl_state(ctrl) == NVME_CTRL_RESETTING;
662 }
663 EXPORT_SYMBOL_GPL(nvme_wait_reset);
664 
665 static void nvme_free_ns_head(struct kref *ref)
666 {
667 	struct nvme_ns_head *head =
668 		container_of(ref, struct nvme_ns_head, ref);
669 
670 	nvme_mpath_remove_disk(head);
671 	ida_free(&head->subsys->ns_ida, head->instance);
672 	cleanup_srcu_struct(&head->srcu);
673 	nvme_put_subsystem(head->subsys);
674 	kfree(head);
675 }
676 
677 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
678 {
679 	return kref_get_unless_zero(&head->ref);
680 }
681 
682 void nvme_put_ns_head(struct nvme_ns_head *head)
683 {
684 	kref_put(&head->ref, nvme_free_ns_head);
685 }
686 
687 static void nvme_free_ns(struct kref *kref)
688 {
689 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
690 
691 	put_disk(ns->disk);
692 	nvme_put_ns_head(ns->head);
693 	nvme_put_ctrl(ns->ctrl);
694 	kfree(ns);
695 }
696 
697 static inline bool nvme_get_ns(struct nvme_ns *ns)
698 {
699 	return kref_get_unless_zero(&ns->kref);
700 }
701 
702 void nvme_put_ns(struct nvme_ns *ns)
703 {
704 	kref_put(&ns->kref, nvme_free_ns);
705 }
706 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
707 
708 static inline void nvme_clear_nvme_request(struct request *req)
709 {
710 	nvme_req(req)->status = 0;
711 	nvme_req(req)->retries = 0;
712 	nvme_req(req)->flags = 0;
713 	req->rq_flags |= RQF_DONTPREP;
714 }
715 
716 /* initialize a passthrough request */
717 void nvme_init_request(struct request *req, struct nvme_command *cmd)
718 {
719 	struct nvme_request *nr = nvme_req(req);
720 	bool logging_enabled;
721 
722 	if (req->q->queuedata) {
723 		struct nvme_ns *ns = req->q->disk->private_data;
724 
725 		logging_enabled = ns->head->passthru_err_log_enabled;
726 		req->timeout = NVME_IO_TIMEOUT;
727 	} else { /* no queuedata implies admin queue */
728 		logging_enabled = nr->ctrl->passthru_err_log_enabled;
729 		req->timeout = NVME_ADMIN_TIMEOUT;
730 	}
731 
732 	if (!logging_enabled)
733 		req->rq_flags |= RQF_QUIET;
734 
735 	/* passthru commands should let the driver set the SGL flags */
736 	cmd->common.flags &= ~NVME_CMD_SGL_ALL;
737 
738 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
739 	if (req->mq_hctx->type == HCTX_TYPE_POLL)
740 		req->cmd_flags |= REQ_POLLED;
741 	nvme_clear_nvme_request(req);
742 	memcpy(nr->cmd, cmd, sizeof(*cmd));
743 }
744 EXPORT_SYMBOL_GPL(nvme_init_request);
745 
746 /*
747  * For something we're not in a state to send to the device the default action
748  * is to busy it and retry it after the controller state is recovered.  However,
749  * if the controller is deleting or if anything is marked for failfast or
750  * nvme multipath it is immediately failed.
751  *
752  * Note: commands used to initialize the controller will be marked for failfast.
753  * Note: nvme cli/ioctl commands are marked for failfast.
754  */
755 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
756 		struct request *rq)
757 {
758 	enum nvme_ctrl_state state = nvme_ctrl_state(ctrl);
759 
760 	if (state != NVME_CTRL_DELETING_NOIO &&
761 	    state != NVME_CTRL_DELETING &&
762 	    state != NVME_CTRL_DEAD &&
763 	    !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
764 	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
765 		return BLK_STS_RESOURCE;
766 	return nvme_host_path_error(rq);
767 }
768 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
769 
770 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
771 		bool queue_live, enum nvme_ctrl_state state)
772 {
773 	struct nvme_request *req = nvme_req(rq);
774 
775 	/*
776 	 * currently we have a problem sending passthru commands
777 	 * on the admin_q if the controller is not LIVE because we can't
778 	 * make sure that they are going out after the admin connect,
779 	 * controller enable and/or other commands in the initialization
780 	 * sequence. until the controller will be LIVE, fail with
781 	 * BLK_STS_RESOURCE so that they will be rescheduled.
782 	 */
783 	if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
784 		return false;
785 
786 	if (ctrl->ops->flags & NVME_F_FABRICS) {
787 		/*
788 		 * Only allow commands on a live queue, except for the connect
789 		 * command, which is require to set the queue live in the
790 		 * appropinquate states.
791 		 */
792 		switch (state) {
793 		case NVME_CTRL_CONNECTING:
794 			if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
795 			    (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
796 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
797 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
798 				return true;
799 			break;
800 		default:
801 			break;
802 		case NVME_CTRL_DEAD:
803 			return false;
804 		}
805 	}
806 
807 	return queue_live;
808 }
809 EXPORT_SYMBOL_GPL(__nvme_check_ready);
810 
811 static inline void nvme_setup_flush(struct nvme_ns *ns,
812 		struct nvme_command *cmnd)
813 {
814 	memset(cmnd, 0, sizeof(*cmnd));
815 	cmnd->common.opcode = nvme_cmd_flush;
816 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
817 }
818 
819 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
820 		struct nvme_command *cmnd)
821 {
822 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
823 	struct nvme_dsm_range *range;
824 	struct bio *bio;
825 
826 	/*
827 	 * Some devices do not consider the DSM 'Number of Ranges' field when
828 	 * determining how much data to DMA. Always allocate memory for maximum
829 	 * number of segments to prevent device reading beyond end of buffer.
830 	 */
831 	static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
832 
833 	range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
834 	if (!range) {
835 		/*
836 		 * If we fail allocation our range, fallback to the controller
837 		 * discard page. If that's also busy, it's safe to return
838 		 * busy, as we know we can make progress once that's freed.
839 		 */
840 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
841 			return BLK_STS_RESOURCE;
842 
843 		range = page_address(ns->ctrl->discard_page);
844 	}
845 
846 	if (queue_max_discard_segments(req->q) == 1) {
847 		u64 slba = nvme_sect_to_lba(ns->head, blk_rq_pos(req));
848 		u32 nlb = blk_rq_sectors(req) >> (ns->head->lba_shift - 9);
849 
850 		range[0].cattr = cpu_to_le32(0);
851 		range[0].nlb = cpu_to_le32(nlb);
852 		range[0].slba = cpu_to_le64(slba);
853 		n = 1;
854 	} else {
855 		__rq_for_each_bio(bio, req) {
856 			u64 slba = nvme_sect_to_lba(ns->head,
857 						    bio->bi_iter.bi_sector);
858 			u32 nlb = bio->bi_iter.bi_size >> ns->head->lba_shift;
859 
860 			if (n < segments) {
861 				range[n].cattr = cpu_to_le32(0);
862 				range[n].nlb = cpu_to_le32(nlb);
863 				range[n].slba = cpu_to_le64(slba);
864 			}
865 			n++;
866 		}
867 	}
868 
869 	if (WARN_ON_ONCE(n != segments)) {
870 		if (virt_to_page(range) == ns->ctrl->discard_page)
871 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
872 		else
873 			kfree(range);
874 		return BLK_STS_IOERR;
875 	}
876 
877 	memset(cmnd, 0, sizeof(*cmnd));
878 	cmnd->dsm.opcode = nvme_cmd_dsm;
879 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
880 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
881 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
882 
883 	bvec_set_virt(&req->special_vec, range, alloc_size);
884 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
885 
886 	return BLK_STS_OK;
887 }
888 
889 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
890 			      struct request *req)
891 {
892 	u32 upper, lower;
893 	u64 ref48;
894 
895 	/* both rw and write zeroes share the same reftag format */
896 	switch (ns->head->guard_type) {
897 	case NVME_NVM_NS_16B_GUARD:
898 		cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
899 		break;
900 	case NVME_NVM_NS_64B_GUARD:
901 		ref48 = ext_pi_ref_tag(req);
902 		lower = lower_32_bits(ref48);
903 		upper = upper_32_bits(ref48);
904 
905 		cmnd->rw.reftag = cpu_to_le32(lower);
906 		cmnd->rw.cdw3 = cpu_to_le32(upper);
907 		break;
908 	default:
909 		break;
910 	}
911 }
912 
913 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
914 		struct request *req, struct nvme_command *cmnd)
915 {
916 	memset(cmnd, 0, sizeof(*cmnd));
917 
918 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
919 		return nvme_setup_discard(ns, req, cmnd);
920 
921 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
922 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
923 	cmnd->write_zeroes.slba =
924 		cpu_to_le64(nvme_sect_to_lba(ns->head, blk_rq_pos(req)));
925 	cmnd->write_zeroes.length =
926 		cpu_to_le16((blk_rq_bytes(req) >> ns->head->lba_shift) - 1);
927 
928 	if (!(req->cmd_flags & REQ_NOUNMAP) &&
929 	    (ns->head->features & NVME_NS_DEAC))
930 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_WZ_DEAC);
931 
932 	if (nvme_ns_has_pi(ns->head)) {
933 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_RW_PRINFO_PRACT);
934 
935 		switch (ns->head->pi_type) {
936 		case NVME_NS_DPS_PI_TYPE1:
937 		case NVME_NS_DPS_PI_TYPE2:
938 			nvme_set_ref_tag(ns, cmnd, req);
939 			break;
940 		}
941 	}
942 
943 	return BLK_STS_OK;
944 }
945 
946 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
947 		struct request *req, struct nvme_command *cmnd,
948 		enum nvme_opcode op)
949 {
950 	u16 control = 0;
951 	u32 dsmgmt = 0;
952 
953 	if (req->cmd_flags & REQ_FUA)
954 		control |= NVME_RW_FUA;
955 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
956 		control |= NVME_RW_LR;
957 
958 	if (req->cmd_flags & REQ_RAHEAD)
959 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
960 
961 	cmnd->rw.opcode = op;
962 	cmnd->rw.flags = 0;
963 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
964 	cmnd->rw.cdw2 = 0;
965 	cmnd->rw.cdw3 = 0;
966 	cmnd->rw.metadata = 0;
967 	cmnd->rw.slba =
968 		cpu_to_le64(nvme_sect_to_lba(ns->head, blk_rq_pos(req)));
969 	cmnd->rw.length =
970 		cpu_to_le16((blk_rq_bytes(req) >> ns->head->lba_shift) - 1);
971 	cmnd->rw.reftag = 0;
972 	cmnd->rw.apptag = 0;
973 	cmnd->rw.appmask = 0;
974 
975 	if (ns->head->ms) {
976 		/*
977 		 * If formated with metadata, the block layer always provides a
978 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
979 		 * we enable the PRACT bit for protection information or set the
980 		 * namespace capacity to zero to prevent any I/O.
981 		 */
982 		if (!blk_integrity_rq(req)) {
983 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns->head)))
984 				return BLK_STS_NOTSUPP;
985 			control |= NVME_RW_PRINFO_PRACT;
986 		}
987 
988 		switch (ns->head->pi_type) {
989 		case NVME_NS_DPS_PI_TYPE3:
990 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
991 			break;
992 		case NVME_NS_DPS_PI_TYPE1:
993 		case NVME_NS_DPS_PI_TYPE2:
994 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
995 					NVME_RW_PRINFO_PRCHK_REF;
996 			if (op == nvme_cmd_zone_append)
997 				control |= NVME_RW_APPEND_PIREMAP;
998 			nvme_set_ref_tag(ns, cmnd, req);
999 			break;
1000 		}
1001 	}
1002 
1003 	cmnd->rw.control = cpu_to_le16(control);
1004 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
1005 	return 0;
1006 }
1007 
1008 void nvme_cleanup_cmd(struct request *req)
1009 {
1010 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
1011 		struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
1012 
1013 		if (req->special_vec.bv_page == ctrl->discard_page)
1014 			clear_bit_unlock(0, &ctrl->discard_page_busy);
1015 		else
1016 			kfree(bvec_virt(&req->special_vec));
1017 	}
1018 }
1019 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
1020 
1021 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
1022 {
1023 	struct nvme_command *cmd = nvme_req(req)->cmd;
1024 	blk_status_t ret = BLK_STS_OK;
1025 
1026 	if (!(req->rq_flags & RQF_DONTPREP))
1027 		nvme_clear_nvme_request(req);
1028 
1029 	switch (req_op(req)) {
1030 	case REQ_OP_DRV_IN:
1031 	case REQ_OP_DRV_OUT:
1032 		/* these are setup prior to execution in nvme_init_request() */
1033 		break;
1034 	case REQ_OP_FLUSH:
1035 		nvme_setup_flush(ns, cmd);
1036 		break;
1037 	case REQ_OP_ZONE_RESET_ALL:
1038 	case REQ_OP_ZONE_RESET:
1039 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
1040 		break;
1041 	case REQ_OP_ZONE_OPEN:
1042 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
1043 		break;
1044 	case REQ_OP_ZONE_CLOSE:
1045 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
1046 		break;
1047 	case REQ_OP_ZONE_FINISH:
1048 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
1049 		break;
1050 	case REQ_OP_WRITE_ZEROES:
1051 		ret = nvme_setup_write_zeroes(ns, req, cmd);
1052 		break;
1053 	case REQ_OP_DISCARD:
1054 		ret = nvme_setup_discard(ns, req, cmd);
1055 		break;
1056 	case REQ_OP_READ:
1057 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1058 		break;
1059 	case REQ_OP_WRITE:
1060 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1061 		break;
1062 	case REQ_OP_ZONE_APPEND:
1063 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1064 		break;
1065 	default:
1066 		WARN_ON_ONCE(1);
1067 		return BLK_STS_IOERR;
1068 	}
1069 
1070 	cmd->common.command_id = nvme_cid(req);
1071 	trace_nvme_setup_cmd(req, cmd);
1072 	return ret;
1073 }
1074 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1075 
1076 /*
1077  * Return values:
1078  * 0:  success
1079  * >0: nvme controller's cqe status response
1080  * <0: kernel error in lieu of controller response
1081  */
1082 int nvme_execute_rq(struct request *rq, bool at_head)
1083 {
1084 	blk_status_t status;
1085 
1086 	status = blk_execute_rq(rq, at_head);
1087 	if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1088 		return -EINTR;
1089 	if (nvme_req(rq)->status)
1090 		return nvme_req(rq)->status;
1091 	return blk_status_to_errno(status);
1092 }
1093 EXPORT_SYMBOL_NS_GPL(nvme_execute_rq, NVME_TARGET_PASSTHRU);
1094 
1095 /*
1096  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1097  * if the result is positive, it's an NVM Express status code
1098  */
1099 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1100 		union nvme_result *result, void *buffer, unsigned bufflen,
1101 		int qid, nvme_submit_flags_t flags)
1102 {
1103 	struct request *req;
1104 	int ret;
1105 	blk_mq_req_flags_t blk_flags = 0;
1106 
1107 	if (flags & NVME_SUBMIT_NOWAIT)
1108 		blk_flags |= BLK_MQ_REQ_NOWAIT;
1109 	if (flags & NVME_SUBMIT_RESERVED)
1110 		blk_flags |= BLK_MQ_REQ_RESERVED;
1111 	if (qid == NVME_QID_ANY)
1112 		req = blk_mq_alloc_request(q, nvme_req_op(cmd), blk_flags);
1113 	else
1114 		req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), blk_flags,
1115 						qid - 1);
1116 
1117 	if (IS_ERR(req))
1118 		return PTR_ERR(req);
1119 	nvme_init_request(req, cmd);
1120 	if (flags & NVME_SUBMIT_RETRY)
1121 		req->cmd_flags &= ~REQ_FAILFAST_DRIVER;
1122 
1123 	if (buffer && bufflen) {
1124 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1125 		if (ret)
1126 			goto out;
1127 	}
1128 
1129 	ret = nvme_execute_rq(req, flags & NVME_SUBMIT_AT_HEAD);
1130 	if (result && ret >= 0)
1131 		*result = nvme_req(req)->result;
1132  out:
1133 	blk_mq_free_request(req);
1134 	return ret;
1135 }
1136 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1137 
1138 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1139 		void *buffer, unsigned bufflen)
1140 {
1141 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1142 			NVME_QID_ANY, 0);
1143 }
1144 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1145 
1146 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1147 {
1148 	u32 effects = 0;
1149 
1150 	if (ns) {
1151 		effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1152 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1153 			dev_warn_once(ctrl->device,
1154 				"IO command:%02x has unusual effects:%08x\n",
1155 				opcode, effects);
1156 
1157 		/*
1158 		 * NVME_CMD_EFFECTS_CSE_MASK causes a freeze all I/O queues,
1159 		 * which would deadlock when done on an I/O command.  Note that
1160 		 * We already warn about an unusual effect above.
1161 		 */
1162 		effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1163 	} else {
1164 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1165 
1166 		/* Ignore execution restrictions if any relaxation bits are set */
1167 		if (effects & NVME_CMD_EFFECTS_CSER_MASK)
1168 			effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1169 	}
1170 
1171 	return effects;
1172 }
1173 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1174 
1175 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1176 {
1177 	u32 effects = nvme_command_effects(ctrl, ns, opcode);
1178 
1179 	/*
1180 	 * For simplicity, IO to all namespaces is quiesced even if the command
1181 	 * effects say only one namespace is affected.
1182 	 */
1183 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1184 		mutex_lock(&ctrl->scan_lock);
1185 		mutex_lock(&ctrl->subsys->lock);
1186 		nvme_mpath_start_freeze(ctrl->subsys);
1187 		nvme_mpath_wait_freeze(ctrl->subsys);
1188 		nvme_start_freeze(ctrl);
1189 		nvme_wait_freeze(ctrl);
1190 	}
1191 	return effects;
1192 }
1193 EXPORT_SYMBOL_NS_GPL(nvme_passthru_start, NVME_TARGET_PASSTHRU);
1194 
1195 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1196 		       struct nvme_command *cmd, int status)
1197 {
1198 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1199 		nvme_unfreeze(ctrl);
1200 		nvme_mpath_unfreeze(ctrl->subsys);
1201 		mutex_unlock(&ctrl->subsys->lock);
1202 		mutex_unlock(&ctrl->scan_lock);
1203 	}
1204 	if (effects & NVME_CMD_EFFECTS_CCC) {
1205 		if (!test_and_set_bit(NVME_CTRL_DIRTY_CAPABILITY,
1206 				      &ctrl->flags)) {
1207 			dev_info(ctrl->device,
1208 "controller capabilities changed, reset may be required to take effect.\n");
1209 		}
1210 	}
1211 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1212 		nvme_queue_scan(ctrl);
1213 		flush_work(&ctrl->scan_work);
1214 	}
1215 	if (ns)
1216 		return;
1217 
1218 	switch (cmd->common.opcode) {
1219 	case nvme_admin_set_features:
1220 		switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1221 		case NVME_FEAT_KATO:
1222 			/*
1223 			 * Keep alive commands interval on the host should be
1224 			 * updated when KATO is modified by Set Features
1225 			 * commands.
1226 			 */
1227 			if (!status)
1228 				nvme_update_keep_alive(ctrl, cmd);
1229 			break;
1230 		default:
1231 			break;
1232 		}
1233 		break;
1234 	default:
1235 		break;
1236 	}
1237 }
1238 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1239 
1240 /*
1241  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1242  *
1243  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1244  *   accounting for transport roundtrip times [..].
1245  */
1246 static unsigned long nvme_keep_alive_work_period(struct nvme_ctrl *ctrl)
1247 {
1248 	unsigned long delay = ctrl->kato * HZ / 2;
1249 
1250 	/*
1251 	 * When using Traffic Based Keep Alive, we need to run
1252 	 * nvme_keep_alive_work at twice the normal frequency, as one
1253 	 * command completion can postpone sending a keep alive command
1254 	 * by up to twice the delay between runs.
1255 	 */
1256 	if (ctrl->ctratt & NVME_CTRL_ATTR_TBKAS)
1257 		delay /= 2;
1258 	return delay;
1259 }
1260 
1261 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1262 {
1263 	unsigned long now = jiffies;
1264 	unsigned long delay = nvme_keep_alive_work_period(ctrl);
1265 	unsigned long ka_next_check_tm = ctrl->ka_last_check_time + delay;
1266 
1267 	if (time_after(now, ka_next_check_tm))
1268 		delay = 0;
1269 	else
1270 		delay = ka_next_check_tm - now;
1271 
1272 	queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1273 }
1274 
1275 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1276 						 blk_status_t status)
1277 {
1278 	struct nvme_ctrl *ctrl = rq->end_io_data;
1279 	unsigned long flags;
1280 	bool startka = false;
1281 	unsigned long rtt = jiffies - (rq->deadline - rq->timeout);
1282 	unsigned long delay = nvme_keep_alive_work_period(ctrl);
1283 
1284 	/*
1285 	 * Subtract off the keepalive RTT so nvme_keep_alive_work runs
1286 	 * at the desired frequency.
1287 	 */
1288 	if (rtt <= delay) {
1289 		delay -= rtt;
1290 	} else {
1291 		dev_warn(ctrl->device, "long keepalive RTT (%u ms)\n",
1292 			 jiffies_to_msecs(rtt));
1293 		delay = 0;
1294 	}
1295 
1296 	blk_mq_free_request(rq);
1297 
1298 	if (status) {
1299 		dev_err(ctrl->device,
1300 			"failed nvme_keep_alive_end_io error=%d\n",
1301 				status);
1302 		return RQ_END_IO_NONE;
1303 	}
1304 
1305 	ctrl->ka_last_check_time = jiffies;
1306 	ctrl->comp_seen = false;
1307 	spin_lock_irqsave(&ctrl->lock, flags);
1308 	if (ctrl->state == NVME_CTRL_LIVE ||
1309 	    ctrl->state == NVME_CTRL_CONNECTING)
1310 		startka = true;
1311 	spin_unlock_irqrestore(&ctrl->lock, flags);
1312 	if (startka)
1313 		queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1314 	return RQ_END_IO_NONE;
1315 }
1316 
1317 static void nvme_keep_alive_work(struct work_struct *work)
1318 {
1319 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1320 			struct nvme_ctrl, ka_work);
1321 	bool comp_seen = ctrl->comp_seen;
1322 	struct request *rq;
1323 
1324 	ctrl->ka_last_check_time = jiffies;
1325 
1326 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1327 		dev_dbg(ctrl->device,
1328 			"reschedule traffic based keep-alive timer\n");
1329 		ctrl->comp_seen = false;
1330 		nvme_queue_keep_alive_work(ctrl);
1331 		return;
1332 	}
1333 
1334 	rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1335 				  BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1336 	if (IS_ERR(rq)) {
1337 		/* allocation failure, reset the controller */
1338 		dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1339 		nvme_reset_ctrl(ctrl);
1340 		return;
1341 	}
1342 	nvme_init_request(rq, &ctrl->ka_cmd);
1343 
1344 	rq->timeout = ctrl->kato * HZ;
1345 	rq->end_io = nvme_keep_alive_end_io;
1346 	rq->end_io_data = ctrl;
1347 	blk_execute_rq_nowait(rq, false);
1348 }
1349 
1350 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1351 {
1352 	if (unlikely(ctrl->kato == 0))
1353 		return;
1354 
1355 	nvme_queue_keep_alive_work(ctrl);
1356 }
1357 
1358 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1359 {
1360 	if (unlikely(ctrl->kato == 0))
1361 		return;
1362 
1363 	cancel_delayed_work_sync(&ctrl->ka_work);
1364 }
1365 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1366 
1367 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1368 				   struct nvme_command *cmd)
1369 {
1370 	unsigned int new_kato =
1371 		DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1372 
1373 	dev_info(ctrl->device,
1374 		 "keep alive interval updated from %u ms to %u ms\n",
1375 		 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1376 
1377 	nvme_stop_keep_alive(ctrl);
1378 	ctrl->kato = new_kato;
1379 	nvme_start_keep_alive(ctrl);
1380 }
1381 
1382 /*
1383  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1384  * flag, thus sending any new CNS opcodes has a big chance of not working.
1385  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1386  * (but not for any later version).
1387  */
1388 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1389 {
1390 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1391 		return ctrl->vs < NVME_VS(1, 2, 0);
1392 	return ctrl->vs < NVME_VS(1, 1, 0);
1393 }
1394 
1395 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1396 {
1397 	struct nvme_command c = { };
1398 	int error;
1399 
1400 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1401 	c.identify.opcode = nvme_admin_identify;
1402 	c.identify.cns = NVME_ID_CNS_CTRL;
1403 
1404 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1405 	if (!*id)
1406 		return -ENOMEM;
1407 
1408 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1409 			sizeof(struct nvme_id_ctrl));
1410 	if (error) {
1411 		kfree(*id);
1412 		*id = NULL;
1413 	}
1414 	return error;
1415 }
1416 
1417 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1418 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1419 {
1420 	const char *warn_str = "ctrl returned bogus length:";
1421 	void *data = cur;
1422 
1423 	switch (cur->nidt) {
1424 	case NVME_NIDT_EUI64:
1425 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1426 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1427 				 warn_str, cur->nidl);
1428 			return -1;
1429 		}
1430 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1431 			return NVME_NIDT_EUI64_LEN;
1432 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1433 		return NVME_NIDT_EUI64_LEN;
1434 	case NVME_NIDT_NGUID:
1435 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1436 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1437 				 warn_str, cur->nidl);
1438 			return -1;
1439 		}
1440 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1441 			return NVME_NIDT_NGUID_LEN;
1442 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1443 		return NVME_NIDT_NGUID_LEN;
1444 	case NVME_NIDT_UUID:
1445 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1446 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1447 				 warn_str, cur->nidl);
1448 			return -1;
1449 		}
1450 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1451 			return NVME_NIDT_UUID_LEN;
1452 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1453 		return NVME_NIDT_UUID_LEN;
1454 	case NVME_NIDT_CSI:
1455 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1456 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1457 				 warn_str, cur->nidl);
1458 			return -1;
1459 		}
1460 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1461 		*csi_seen = true;
1462 		return NVME_NIDT_CSI_LEN;
1463 	default:
1464 		/* Skip unknown types */
1465 		return cur->nidl;
1466 	}
1467 }
1468 
1469 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1470 		struct nvme_ns_info *info)
1471 {
1472 	struct nvme_command c = { };
1473 	bool csi_seen = false;
1474 	int status, pos, len;
1475 	void *data;
1476 
1477 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1478 		return 0;
1479 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1480 		return 0;
1481 
1482 	c.identify.opcode = nvme_admin_identify;
1483 	c.identify.nsid = cpu_to_le32(info->nsid);
1484 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1485 
1486 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1487 	if (!data)
1488 		return -ENOMEM;
1489 
1490 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1491 				      NVME_IDENTIFY_DATA_SIZE);
1492 	if (status) {
1493 		dev_warn(ctrl->device,
1494 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1495 			info->nsid, status);
1496 		goto free_data;
1497 	}
1498 
1499 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1500 		struct nvme_ns_id_desc *cur = data + pos;
1501 
1502 		if (cur->nidl == 0)
1503 			break;
1504 
1505 		len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1506 		if (len < 0)
1507 			break;
1508 
1509 		len += sizeof(*cur);
1510 	}
1511 
1512 	if (nvme_multi_css(ctrl) && !csi_seen) {
1513 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1514 			 info->nsid);
1515 		status = -EINVAL;
1516 	}
1517 
1518 free_data:
1519 	kfree(data);
1520 	return status;
1521 }
1522 
1523 int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1524 			struct nvme_id_ns **id)
1525 {
1526 	struct nvme_command c = { };
1527 	int error;
1528 
1529 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1530 	c.identify.opcode = nvme_admin_identify;
1531 	c.identify.nsid = cpu_to_le32(nsid);
1532 	c.identify.cns = NVME_ID_CNS_NS;
1533 
1534 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1535 	if (!*id)
1536 		return -ENOMEM;
1537 
1538 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1539 	if (error) {
1540 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1541 		kfree(*id);
1542 		*id = NULL;
1543 	}
1544 	return error;
1545 }
1546 
1547 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1548 		struct nvme_ns_info *info)
1549 {
1550 	struct nvme_ns_ids *ids = &info->ids;
1551 	struct nvme_id_ns *id;
1552 	int ret;
1553 
1554 	ret = nvme_identify_ns(ctrl, info->nsid, &id);
1555 	if (ret)
1556 		return ret;
1557 
1558 	if (id->ncap == 0) {
1559 		/* namespace not allocated or attached */
1560 		info->is_removed = true;
1561 		ret = -ENODEV;
1562 		goto error;
1563 	}
1564 
1565 	info->anagrpid = id->anagrpid;
1566 	info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1567 	info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1568 	info->is_ready = true;
1569 	if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1570 		dev_info(ctrl->device,
1571 			 "Ignoring bogus Namespace Identifiers\n");
1572 	} else {
1573 		if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1574 		    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1575 			memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1576 		if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1577 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1578 			memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1579 	}
1580 
1581 error:
1582 	kfree(id);
1583 	return ret;
1584 }
1585 
1586 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1587 		struct nvme_ns_info *info)
1588 {
1589 	struct nvme_id_ns_cs_indep *id;
1590 	struct nvme_command c = {
1591 		.identify.opcode	= nvme_admin_identify,
1592 		.identify.nsid		= cpu_to_le32(info->nsid),
1593 		.identify.cns		= NVME_ID_CNS_NS_CS_INDEP,
1594 	};
1595 	int ret;
1596 
1597 	id = kmalloc(sizeof(*id), GFP_KERNEL);
1598 	if (!id)
1599 		return -ENOMEM;
1600 
1601 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1602 	if (!ret) {
1603 		info->anagrpid = id->anagrpid;
1604 		info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1605 		info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1606 		info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1607 	}
1608 	kfree(id);
1609 	return ret;
1610 }
1611 
1612 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1613 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1614 {
1615 	union nvme_result res = { 0 };
1616 	struct nvme_command c = { };
1617 	int ret;
1618 
1619 	c.features.opcode = op;
1620 	c.features.fid = cpu_to_le32(fid);
1621 	c.features.dword11 = cpu_to_le32(dword11);
1622 
1623 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1624 			buffer, buflen, NVME_QID_ANY, 0);
1625 	if (ret >= 0 && result)
1626 		*result = le32_to_cpu(res.u32);
1627 	return ret;
1628 }
1629 
1630 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1631 		      unsigned int dword11, void *buffer, size_t buflen,
1632 		      u32 *result)
1633 {
1634 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1635 			     buflen, result);
1636 }
1637 EXPORT_SYMBOL_GPL(nvme_set_features);
1638 
1639 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1640 		      unsigned int dword11, void *buffer, size_t buflen,
1641 		      u32 *result)
1642 {
1643 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1644 			     buflen, result);
1645 }
1646 EXPORT_SYMBOL_GPL(nvme_get_features);
1647 
1648 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1649 {
1650 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1651 	u32 result;
1652 	int status, nr_io_queues;
1653 
1654 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1655 			&result);
1656 	if (status < 0)
1657 		return status;
1658 
1659 	/*
1660 	 * Degraded controllers might return an error when setting the queue
1661 	 * count.  We still want to be able to bring them online and offer
1662 	 * access to the admin queue, as that might be only way to fix them up.
1663 	 */
1664 	if (status > 0) {
1665 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1666 		*count = 0;
1667 	} else {
1668 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1669 		*count = min(*count, nr_io_queues);
1670 	}
1671 
1672 	return 0;
1673 }
1674 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1675 
1676 #define NVME_AEN_SUPPORTED \
1677 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1678 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1679 
1680 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1681 {
1682 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1683 	int status;
1684 
1685 	if (!supported_aens)
1686 		return;
1687 
1688 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1689 			NULL, 0, &result);
1690 	if (status)
1691 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1692 			 supported_aens);
1693 
1694 	queue_work(nvme_wq, &ctrl->async_event_work);
1695 }
1696 
1697 static int nvme_ns_open(struct nvme_ns *ns)
1698 {
1699 
1700 	/* should never be called due to GENHD_FL_HIDDEN */
1701 	if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1702 		goto fail;
1703 	if (!nvme_get_ns(ns))
1704 		goto fail;
1705 	if (!try_module_get(ns->ctrl->ops->module))
1706 		goto fail_put_ns;
1707 
1708 	return 0;
1709 
1710 fail_put_ns:
1711 	nvme_put_ns(ns);
1712 fail:
1713 	return -ENXIO;
1714 }
1715 
1716 static void nvme_ns_release(struct nvme_ns *ns)
1717 {
1718 
1719 	module_put(ns->ctrl->ops->module);
1720 	nvme_put_ns(ns);
1721 }
1722 
1723 static int nvme_open(struct gendisk *disk, blk_mode_t mode)
1724 {
1725 	return nvme_ns_open(disk->private_data);
1726 }
1727 
1728 static void nvme_release(struct gendisk *disk)
1729 {
1730 	nvme_ns_release(disk->private_data);
1731 }
1732 
1733 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1734 {
1735 	/* some standard values */
1736 	geo->heads = 1 << 6;
1737 	geo->sectors = 1 << 5;
1738 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1739 	return 0;
1740 }
1741 
1742 static bool nvme_init_integrity(struct gendisk *disk, struct nvme_ns_head *head)
1743 {
1744 	struct blk_integrity integrity = { };
1745 
1746 	blk_integrity_unregister(disk);
1747 
1748 	if (!head->ms)
1749 		return true;
1750 
1751 	/*
1752 	 * PI can always be supported as we can ask the controller to simply
1753 	 * insert/strip it, which is not possible for other kinds of metadata.
1754 	 */
1755 	if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) ||
1756 	    !(head->features & NVME_NS_METADATA_SUPPORTED))
1757 		return nvme_ns_has_pi(head);
1758 
1759 	switch (head->pi_type) {
1760 	case NVME_NS_DPS_PI_TYPE3:
1761 		switch (head->guard_type) {
1762 		case NVME_NVM_NS_16B_GUARD:
1763 			integrity.profile = &t10_pi_type3_crc;
1764 			integrity.tag_size = sizeof(u16) + sizeof(u32);
1765 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1766 			break;
1767 		case NVME_NVM_NS_64B_GUARD:
1768 			integrity.profile = &ext_pi_type3_crc64;
1769 			integrity.tag_size = sizeof(u16) + 6;
1770 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1771 			break;
1772 		default:
1773 			integrity.profile = NULL;
1774 			break;
1775 		}
1776 		break;
1777 	case NVME_NS_DPS_PI_TYPE1:
1778 	case NVME_NS_DPS_PI_TYPE2:
1779 		switch (head->guard_type) {
1780 		case NVME_NVM_NS_16B_GUARD:
1781 			integrity.profile = &t10_pi_type1_crc;
1782 			integrity.tag_size = sizeof(u16);
1783 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1784 			break;
1785 		case NVME_NVM_NS_64B_GUARD:
1786 			integrity.profile = &ext_pi_type1_crc64;
1787 			integrity.tag_size = sizeof(u16);
1788 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1789 			break;
1790 		default:
1791 			integrity.profile = NULL;
1792 			break;
1793 		}
1794 		break;
1795 	default:
1796 		integrity.profile = NULL;
1797 		break;
1798 	}
1799 
1800 	integrity.tuple_size = head->ms;
1801 	integrity.pi_offset = head->pi_offset;
1802 	blk_integrity_register(disk, &integrity);
1803 	return true;
1804 }
1805 
1806 static void nvme_config_discard(struct nvme_ns *ns, struct queue_limits *lim)
1807 {
1808 	struct nvme_ctrl *ctrl = ns->ctrl;
1809 
1810 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1811 			NVME_DSM_MAX_RANGES);
1812 
1813 	if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns->head, UINT_MAX))
1814 		lim->max_hw_discard_sectors =
1815 			nvme_lba_to_sect(ns->head, ctrl->dmrsl);
1816 	else if (ctrl->oncs & NVME_CTRL_ONCS_DSM)
1817 		lim->max_hw_discard_sectors = UINT_MAX;
1818 	else
1819 		lim->max_hw_discard_sectors = 0;
1820 
1821 	lim->discard_granularity = lim->logical_block_size;
1822 
1823 	if (ctrl->dmrl)
1824 		lim->max_discard_segments = ctrl->dmrl;
1825 	else
1826 		lim->max_discard_segments = NVME_DSM_MAX_RANGES;
1827 }
1828 
1829 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1830 {
1831 	return uuid_equal(&a->uuid, &b->uuid) &&
1832 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1833 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1834 		a->csi == b->csi;
1835 }
1836 
1837 static int nvme_identify_ns_nvm(struct nvme_ctrl *ctrl, unsigned int nsid,
1838 		struct nvme_id_ns_nvm **nvmp)
1839 {
1840 	struct nvme_command c = {
1841 		.identify.opcode	= nvme_admin_identify,
1842 		.identify.nsid		= cpu_to_le32(nsid),
1843 		.identify.cns		= NVME_ID_CNS_CS_NS,
1844 		.identify.csi		= NVME_CSI_NVM,
1845 	};
1846 	struct nvme_id_ns_nvm *nvm;
1847 	int ret;
1848 
1849 	nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1850 	if (!nvm)
1851 		return -ENOMEM;
1852 
1853 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, nvm, sizeof(*nvm));
1854 	if (ret)
1855 		kfree(nvm);
1856 	else
1857 		*nvmp = nvm;
1858 	return ret;
1859 }
1860 
1861 static void nvme_configure_pi_elbas(struct nvme_ns_head *head,
1862 		struct nvme_id_ns *id, struct nvme_id_ns_nvm *nvm)
1863 {
1864 	u32 elbaf = le32_to_cpu(nvm->elbaf[nvme_lbaf_index(id->flbas)]);
1865 
1866 	/* no support for storage tag formats right now */
1867 	if (nvme_elbaf_sts(elbaf))
1868 		return;
1869 
1870 	head->guard_type = nvme_elbaf_guard_type(elbaf);
1871 	switch (head->guard_type) {
1872 	case NVME_NVM_NS_64B_GUARD:
1873 		head->pi_size = sizeof(struct crc64_pi_tuple);
1874 		break;
1875 	case NVME_NVM_NS_16B_GUARD:
1876 		head->pi_size = sizeof(struct t10_pi_tuple);
1877 		break;
1878 	default:
1879 		break;
1880 	}
1881 }
1882 
1883 static void nvme_configure_metadata(struct nvme_ctrl *ctrl,
1884 		struct nvme_ns_head *head, struct nvme_id_ns *id,
1885 		struct nvme_id_ns_nvm *nvm)
1886 {
1887 	head->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1888 	head->pi_type = 0;
1889 	head->pi_size = 0;
1890 	head->pi_offset = 0;
1891 	head->ms = le16_to_cpu(id->lbaf[nvme_lbaf_index(id->flbas)].ms);
1892 	if (!head->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1893 		return;
1894 
1895 	if (nvm && (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1896 		nvme_configure_pi_elbas(head, id, nvm);
1897 	} else {
1898 		head->pi_size = sizeof(struct t10_pi_tuple);
1899 		head->guard_type = NVME_NVM_NS_16B_GUARD;
1900 	}
1901 
1902 	if (head->pi_size && head->ms >= head->pi_size)
1903 		head->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1904 	if (!(id->dps & NVME_NS_DPS_PI_FIRST))
1905 		head->pi_offset = head->ms - head->pi_size;
1906 
1907 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1908 		/*
1909 		 * The NVMe over Fabrics specification only supports metadata as
1910 		 * part of the extended data LBA.  We rely on HCA/HBA support to
1911 		 * remap the separate metadata buffer from the block layer.
1912 		 */
1913 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1914 			return;
1915 
1916 		head->features |= NVME_NS_EXT_LBAS;
1917 
1918 		/*
1919 		 * The current fabrics transport drivers support namespace
1920 		 * metadata formats only if nvme_ns_has_pi() returns true.
1921 		 * Suppress support for all other formats so the namespace will
1922 		 * have a 0 capacity and not be usable through the block stack.
1923 		 *
1924 		 * Note, this check will need to be modified if any drivers
1925 		 * gain the ability to use other metadata formats.
1926 		 */
1927 		if (ctrl->max_integrity_segments && nvme_ns_has_pi(head))
1928 			head->features |= NVME_NS_METADATA_SUPPORTED;
1929 	} else {
1930 		/*
1931 		 * For PCIe controllers, we can't easily remap the separate
1932 		 * metadata buffer from the block layer and thus require a
1933 		 * separate metadata buffer for block layer metadata/PI support.
1934 		 * We allow extended LBAs for the passthrough interface, though.
1935 		 */
1936 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
1937 			head->features |= NVME_NS_EXT_LBAS;
1938 		else
1939 			head->features |= NVME_NS_METADATA_SUPPORTED;
1940 	}
1941 }
1942 
1943 static u32 nvme_max_drv_segments(struct nvme_ctrl *ctrl)
1944 {
1945 	return ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> SECTOR_SHIFT) + 1;
1946 }
1947 
1948 static void nvme_set_ctrl_limits(struct nvme_ctrl *ctrl,
1949 		struct queue_limits *lim)
1950 {
1951 	lim->max_hw_sectors = ctrl->max_hw_sectors;
1952 	lim->max_segments = min_t(u32, USHRT_MAX,
1953 		min_not_zero(nvme_max_drv_segments(ctrl), ctrl->max_segments));
1954 	lim->max_integrity_segments = ctrl->max_integrity_segments;
1955 	lim->virt_boundary_mask = NVME_CTRL_PAGE_SIZE - 1;
1956 	lim->max_segment_size = UINT_MAX;
1957 	lim->dma_alignment = 3;
1958 }
1959 
1960 static bool nvme_update_disk_info(struct nvme_ns *ns, struct nvme_id_ns *id,
1961 		struct queue_limits *lim)
1962 {
1963 	struct nvme_ns_head *head = ns->head;
1964 	u32 bs = 1U << head->lba_shift;
1965 	u32 atomic_bs, phys_bs, io_opt = 0;
1966 	bool valid = true;
1967 
1968 	/*
1969 	 * The block layer can't support LBA sizes larger than the page size
1970 	 * or smaller than a sector size yet, so catch this early and don't
1971 	 * allow block I/O.
1972 	 */
1973 	if (head->lba_shift > PAGE_SHIFT || head->lba_shift < SECTOR_SHIFT) {
1974 		bs = (1 << 9);
1975 		valid = false;
1976 	}
1977 
1978 	atomic_bs = phys_bs = bs;
1979 	if (id->nabo == 0) {
1980 		/*
1981 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1982 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1983 		 * 0 then AWUPF must be used instead.
1984 		 */
1985 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1986 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1987 		else
1988 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1989 	}
1990 
1991 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1992 		/* NPWG = Namespace Preferred Write Granularity */
1993 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1994 		/* NOWS = Namespace Optimal Write Size */
1995 		io_opt = bs * (1 + le16_to_cpu(id->nows));
1996 	}
1997 
1998 	/*
1999 	 * Linux filesystems assume writing a single physical block is
2000 	 * an atomic operation. Hence limit the physical block size to the
2001 	 * value of the Atomic Write Unit Power Fail parameter.
2002 	 */
2003 	lim->logical_block_size = bs;
2004 	lim->physical_block_size = min(phys_bs, atomic_bs);
2005 	lim->io_min = phys_bs;
2006 	lim->io_opt = io_opt;
2007 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
2008 		lim->max_write_zeroes_sectors = UINT_MAX;
2009 	else
2010 		lim->max_write_zeroes_sectors = ns->ctrl->max_zeroes_sectors;
2011 	return valid;
2012 }
2013 
2014 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
2015 {
2016 	return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
2017 }
2018 
2019 static inline bool nvme_first_scan(struct gendisk *disk)
2020 {
2021 	/* nvme_alloc_ns() scans the disk prior to adding it */
2022 	return !disk_live(disk);
2023 }
2024 
2025 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id,
2026 		struct queue_limits *lim)
2027 {
2028 	struct nvme_ctrl *ctrl = ns->ctrl;
2029 	u32 iob;
2030 
2031 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2032 	    is_power_of_2(ctrl->max_hw_sectors))
2033 		iob = ctrl->max_hw_sectors;
2034 	else
2035 		iob = nvme_lba_to_sect(ns->head, le16_to_cpu(id->noiob));
2036 
2037 	if (!iob)
2038 		return;
2039 
2040 	if (!is_power_of_2(iob)) {
2041 		if (nvme_first_scan(ns->disk))
2042 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2043 				ns->disk->disk_name, iob);
2044 		return;
2045 	}
2046 
2047 	if (blk_queue_is_zoned(ns->disk->queue)) {
2048 		if (nvme_first_scan(ns->disk))
2049 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
2050 				ns->disk->disk_name);
2051 		return;
2052 	}
2053 
2054 	lim->chunk_sectors = iob;
2055 }
2056 
2057 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
2058 		struct nvme_ns_info *info)
2059 {
2060 	struct queue_limits lim;
2061 	int ret;
2062 
2063 	blk_mq_freeze_queue(ns->disk->queue);
2064 	lim = queue_limits_start_update(ns->disk->queue);
2065 	nvme_set_ctrl_limits(ns->ctrl, &lim);
2066 	ret = queue_limits_commit_update(ns->disk->queue, &lim);
2067 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2068 	blk_mq_unfreeze_queue(ns->disk->queue);
2069 
2070 	/* Hide the block-interface for these devices */
2071 	if (!ret)
2072 		ret = -ENODEV;
2073 	return ret;
2074 }
2075 
2076 static int nvme_update_ns_info_block(struct nvme_ns *ns,
2077 		struct nvme_ns_info *info)
2078 {
2079 	bool vwc = ns->ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2080 	struct queue_limits lim;
2081 	struct nvme_id_ns_nvm *nvm = NULL;
2082 	struct nvme_id_ns *id;
2083 	sector_t capacity;
2084 	unsigned lbaf;
2085 	int ret;
2086 
2087 	ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
2088 	if (ret)
2089 		return ret;
2090 
2091 	if (id->ncap == 0) {
2092 		/* namespace not allocated or attached */
2093 		info->is_removed = true;
2094 		ret = -ENODEV;
2095 		goto out;
2096 	}
2097 
2098 	if (ns->ctrl->ctratt & NVME_CTRL_ATTR_ELBAS) {
2099 		ret = nvme_identify_ns_nvm(ns->ctrl, info->nsid, &nvm);
2100 		if (ret < 0)
2101 			goto out;
2102 	}
2103 
2104 	blk_mq_freeze_queue(ns->disk->queue);
2105 	lbaf = nvme_lbaf_index(id->flbas);
2106 	ns->head->lba_shift = id->lbaf[lbaf].ds;
2107 	ns->head->nuse = le64_to_cpu(id->nuse);
2108 	capacity = nvme_lba_to_sect(ns->head, le64_to_cpu(id->nsze));
2109 
2110 	lim = queue_limits_start_update(ns->disk->queue);
2111 	nvme_set_ctrl_limits(ns->ctrl, &lim);
2112 	nvme_configure_metadata(ns->ctrl, ns->head, id, nvm);
2113 	nvme_set_chunk_sectors(ns, id, &lim);
2114 	if (!nvme_update_disk_info(ns, id, &lim))
2115 		capacity = 0;
2116 	nvme_config_discard(ns, &lim);
2117 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
2118 	    ns->head->ids.csi == NVME_CSI_ZNS) {
2119 		ret = nvme_update_zone_info(ns, lbaf, &lim);
2120 		if (ret) {
2121 			blk_mq_unfreeze_queue(ns->disk->queue);
2122 			goto out;
2123 		}
2124 	}
2125 	ret = queue_limits_commit_update(ns->disk->queue, &lim);
2126 	if (ret) {
2127 		blk_mq_unfreeze_queue(ns->disk->queue);
2128 		goto out;
2129 	}
2130 
2131 	/*
2132 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
2133 	 * metadata masquerading as Type 0 if supported, otherwise reject block
2134 	 * I/O to namespaces with metadata except when the namespace supports
2135 	 * PI, as it can strip/insert in that case.
2136 	 */
2137 	if (!nvme_init_integrity(ns->disk, ns->head))
2138 		capacity = 0;
2139 
2140 	set_capacity_and_notify(ns->disk, capacity);
2141 
2142 	/*
2143 	 * Only set the DEAC bit if the device guarantees that reads from
2144 	 * deallocated data return zeroes.  While the DEAC bit does not
2145 	 * require that, it must be a no-op if reads from deallocated data
2146 	 * do not return zeroes.
2147 	 */
2148 	if ((id->dlfeat & 0x7) == 0x1 && (id->dlfeat & (1 << 3)))
2149 		ns->head->features |= NVME_NS_DEAC;
2150 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2151 	blk_queue_write_cache(ns->disk->queue, vwc, vwc);
2152 	set_bit(NVME_NS_READY, &ns->flags);
2153 	blk_mq_unfreeze_queue(ns->disk->queue);
2154 
2155 	if (blk_queue_is_zoned(ns->queue)) {
2156 		ret = blk_revalidate_disk_zones(ns->disk, NULL);
2157 		if (ret && !nvme_first_scan(ns->disk))
2158 			goto out;
2159 	}
2160 
2161 	ret = 0;
2162 out:
2163 	kfree(nvm);
2164 	kfree(id);
2165 	return ret;
2166 }
2167 
2168 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2169 {
2170 	bool unsupported = false;
2171 	int ret;
2172 
2173 	switch (info->ids.csi) {
2174 	case NVME_CSI_ZNS:
2175 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2176 			dev_info(ns->ctrl->device,
2177 	"block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2178 				info->nsid);
2179 			ret = nvme_update_ns_info_generic(ns, info);
2180 			break;
2181 		}
2182 		ret = nvme_update_ns_info_block(ns, info);
2183 		break;
2184 	case NVME_CSI_NVM:
2185 		ret = nvme_update_ns_info_block(ns, info);
2186 		break;
2187 	default:
2188 		dev_info(ns->ctrl->device,
2189 			"block device for nsid %u not supported (csi %u)\n",
2190 			info->nsid, info->ids.csi);
2191 		ret = nvme_update_ns_info_generic(ns, info);
2192 		break;
2193 	}
2194 
2195 	/*
2196 	 * If probing fails due an unsupported feature, hide the block device,
2197 	 * but still allow other access.
2198 	 */
2199 	if (ret == -ENODEV) {
2200 		ns->disk->flags |= GENHD_FL_HIDDEN;
2201 		set_bit(NVME_NS_READY, &ns->flags);
2202 		unsupported = true;
2203 		ret = 0;
2204 	}
2205 
2206 	if (!ret && nvme_ns_head_multipath(ns->head)) {
2207 		struct queue_limits lim;
2208 
2209 		blk_mq_freeze_queue(ns->head->disk->queue);
2210 		if (unsupported)
2211 			ns->head->disk->flags |= GENHD_FL_HIDDEN;
2212 		else
2213 			nvme_init_integrity(ns->head->disk, ns->head);
2214 		set_capacity_and_notify(ns->head->disk, get_capacity(ns->disk));
2215 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2216 		nvme_mpath_revalidate_paths(ns);
2217 
2218 		lim = queue_limits_start_update(ns->head->disk->queue);
2219 		queue_limits_stack_bdev(&lim, ns->disk->part0, 0,
2220 					ns->head->disk->disk_name);
2221 		ret = queue_limits_commit_update(ns->head->disk->queue, &lim);
2222 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2223 	}
2224 
2225 	return ret;
2226 }
2227 
2228 #ifdef CONFIG_BLK_SED_OPAL
2229 static int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2230 		bool send)
2231 {
2232 	struct nvme_ctrl *ctrl = data;
2233 	struct nvme_command cmd = { };
2234 
2235 	if (send)
2236 		cmd.common.opcode = nvme_admin_security_send;
2237 	else
2238 		cmd.common.opcode = nvme_admin_security_recv;
2239 	cmd.common.nsid = 0;
2240 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2241 	cmd.common.cdw11 = cpu_to_le32(len);
2242 
2243 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2244 			NVME_QID_ANY, NVME_SUBMIT_AT_HEAD);
2245 }
2246 
2247 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2248 {
2249 	if (ctrl->oacs & NVME_CTRL_OACS_SEC_SUPP) {
2250 		if (!ctrl->opal_dev)
2251 			ctrl->opal_dev = init_opal_dev(ctrl, &nvme_sec_submit);
2252 		else if (was_suspended)
2253 			opal_unlock_from_suspend(ctrl->opal_dev);
2254 	} else {
2255 		free_opal_dev(ctrl->opal_dev);
2256 		ctrl->opal_dev = NULL;
2257 	}
2258 }
2259 #else
2260 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2261 {
2262 }
2263 #endif /* CONFIG_BLK_SED_OPAL */
2264 
2265 #ifdef CONFIG_BLK_DEV_ZONED
2266 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2267 		unsigned int nr_zones, report_zones_cb cb, void *data)
2268 {
2269 	return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2270 			data);
2271 }
2272 #else
2273 #define nvme_report_zones	NULL
2274 #endif /* CONFIG_BLK_DEV_ZONED */
2275 
2276 const struct block_device_operations nvme_bdev_ops = {
2277 	.owner		= THIS_MODULE,
2278 	.ioctl		= nvme_ioctl,
2279 	.compat_ioctl	= blkdev_compat_ptr_ioctl,
2280 	.open		= nvme_open,
2281 	.release	= nvme_release,
2282 	.getgeo		= nvme_getgeo,
2283 	.report_zones	= nvme_report_zones,
2284 	.pr_ops		= &nvme_pr_ops,
2285 };
2286 
2287 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 mask, u32 val,
2288 		u32 timeout, const char *op)
2289 {
2290 	unsigned long timeout_jiffies = jiffies + timeout * HZ;
2291 	u32 csts;
2292 	int ret;
2293 
2294 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2295 		if (csts == ~0)
2296 			return -ENODEV;
2297 		if ((csts & mask) == val)
2298 			break;
2299 
2300 		usleep_range(1000, 2000);
2301 		if (fatal_signal_pending(current))
2302 			return -EINTR;
2303 		if (time_after(jiffies, timeout_jiffies)) {
2304 			dev_err(ctrl->device,
2305 				"Device not ready; aborting %s, CSTS=0x%x\n",
2306 				op, csts);
2307 			return -ENODEV;
2308 		}
2309 	}
2310 
2311 	return ret;
2312 }
2313 
2314 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown)
2315 {
2316 	int ret;
2317 
2318 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2319 	if (shutdown)
2320 		ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2321 	else
2322 		ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2323 
2324 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2325 	if (ret)
2326 		return ret;
2327 
2328 	if (shutdown) {
2329 		return nvme_wait_ready(ctrl, NVME_CSTS_SHST_MASK,
2330 				       NVME_CSTS_SHST_CMPLT,
2331 				       ctrl->shutdown_timeout, "shutdown");
2332 	}
2333 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2334 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2335 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, 0,
2336 			       (NVME_CAP_TIMEOUT(ctrl->cap) + 1) / 2, "reset");
2337 }
2338 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2339 
2340 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2341 {
2342 	unsigned dev_page_min;
2343 	u32 timeout;
2344 	int ret;
2345 
2346 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2347 	if (ret) {
2348 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2349 		return ret;
2350 	}
2351 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2352 
2353 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2354 		dev_err(ctrl->device,
2355 			"Minimum device page size %u too large for host (%u)\n",
2356 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2357 		return -ENODEV;
2358 	}
2359 
2360 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2361 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2362 	else
2363 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2364 
2365 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS && ctrl->cap & NVME_CAP_CRMS_CRIMS)
2366 		ctrl->ctrl_config |= NVME_CC_CRIME;
2367 
2368 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2369 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2370 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2371 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2372 	if (ret)
2373 		return ret;
2374 
2375 	/* Flush write to device (required if transport is PCI) */
2376 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2377 	if (ret)
2378 		return ret;
2379 
2380 	/* CAP value may change after initial CC write */
2381 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2382 	if (ret)
2383 		return ret;
2384 
2385 	timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2386 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2387 		u32 crto, ready_timeout;
2388 
2389 		ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2390 		if (ret) {
2391 			dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2392 				ret);
2393 			return ret;
2394 		}
2395 
2396 		/*
2397 		 * CRTO should always be greater or equal to CAP.TO, but some
2398 		 * devices are known to get this wrong. Use the larger of the
2399 		 * two values.
2400 		 */
2401 		if (ctrl->ctrl_config & NVME_CC_CRIME)
2402 			ready_timeout = NVME_CRTO_CRIMT(crto);
2403 		else
2404 			ready_timeout = NVME_CRTO_CRWMT(crto);
2405 
2406 		if (ready_timeout < timeout)
2407 			dev_warn_once(ctrl->device, "bad crto:%x cap:%llx\n",
2408 				      crto, ctrl->cap);
2409 		else
2410 			timeout = ready_timeout;
2411 	}
2412 
2413 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2414 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2415 	if (ret)
2416 		return ret;
2417 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, NVME_CSTS_RDY,
2418 			       (timeout + 1) / 2, "initialisation");
2419 }
2420 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2421 
2422 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2423 {
2424 	__le64 ts;
2425 	int ret;
2426 
2427 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2428 		return 0;
2429 
2430 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2431 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2432 			NULL);
2433 	if (ret)
2434 		dev_warn_once(ctrl->device,
2435 			"could not set timestamp (%d)\n", ret);
2436 	return ret;
2437 }
2438 
2439 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2440 {
2441 	struct nvme_feat_host_behavior *host;
2442 	u8 acre = 0, lbafee = 0;
2443 	int ret;
2444 
2445 	/* Don't bother enabling the feature if retry delay is not reported */
2446 	if (ctrl->crdt[0])
2447 		acre = NVME_ENABLE_ACRE;
2448 	if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2449 		lbafee = NVME_ENABLE_LBAFEE;
2450 
2451 	if (!acre && !lbafee)
2452 		return 0;
2453 
2454 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2455 	if (!host)
2456 		return 0;
2457 
2458 	host->acre = acre;
2459 	host->lbafee = lbafee;
2460 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2461 				host, sizeof(*host), NULL);
2462 	kfree(host);
2463 	return ret;
2464 }
2465 
2466 /*
2467  * The function checks whether the given total (exlat + enlat) latency of
2468  * a power state allows the latter to be used as an APST transition target.
2469  * It does so by comparing the latency to the primary and secondary latency
2470  * tolerances defined by module params. If there's a match, the corresponding
2471  * timeout value is returned and the matching tolerance index (1 or 2) is
2472  * reported.
2473  */
2474 static bool nvme_apst_get_transition_time(u64 total_latency,
2475 		u64 *transition_time, unsigned *last_index)
2476 {
2477 	if (total_latency <= apst_primary_latency_tol_us) {
2478 		if (*last_index == 1)
2479 			return false;
2480 		*last_index = 1;
2481 		*transition_time = apst_primary_timeout_ms;
2482 		return true;
2483 	}
2484 	if (apst_secondary_timeout_ms &&
2485 		total_latency <= apst_secondary_latency_tol_us) {
2486 		if (*last_index <= 2)
2487 			return false;
2488 		*last_index = 2;
2489 		*transition_time = apst_secondary_timeout_ms;
2490 		return true;
2491 	}
2492 	return false;
2493 }
2494 
2495 /*
2496  * APST (Autonomous Power State Transition) lets us program a table of power
2497  * state transitions that the controller will perform automatically.
2498  *
2499  * Depending on module params, one of the two supported techniques will be used:
2500  *
2501  * - If the parameters provide explicit timeouts and tolerances, they will be
2502  *   used to build a table with up to 2 non-operational states to transition to.
2503  *   The default parameter values were selected based on the values used by
2504  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2505  *   regeneration of the APST table in the event of switching between external
2506  *   and battery power, the timeouts and tolerances reflect a compromise
2507  *   between values used by Microsoft for AC and battery scenarios.
2508  * - If not, we'll configure the table with a simple heuristic: we are willing
2509  *   to spend at most 2% of the time transitioning between power states.
2510  *   Therefore, when running in any given state, we will enter the next
2511  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2512  *   microseconds, as long as that state's exit latency is under the requested
2513  *   maximum latency.
2514  *
2515  * We will not autonomously enter any non-operational state for which the total
2516  * latency exceeds ps_max_latency_us.
2517  *
2518  * Users can set ps_max_latency_us to zero to turn off APST.
2519  */
2520 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2521 {
2522 	struct nvme_feat_auto_pst *table;
2523 	unsigned apste = 0;
2524 	u64 max_lat_us = 0;
2525 	__le64 target = 0;
2526 	int max_ps = -1;
2527 	int state;
2528 	int ret;
2529 	unsigned last_lt_index = UINT_MAX;
2530 
2531 	/*
2532 	 * If APST isn't supported or if we haven't been initialized yet,
2533 	 * then don't do anything.
2534 	 */
2535 	if (!ctrl->apsta)
2536 		return 0;
2537 
2538 	if (ctrl->npss > 31) {
2539 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2540 		return 0;
2541 	}
2542 
2543 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2544 	if (!table)
2545 		return 0;
2546 
2547 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2548 		/* Turn off APST. */
2549 		dev_dbg(ctrl->device, "APST disabled\n");
2550 		goto done;
2551 	}
2552 
2553 	/*
2554 	 * Walk through all states from lowest- to highest-power.
2555 	 * According to the spec, lower-numbered states use more power.  NPSS,
2556 	 * despite the name, is the index of the lowest-power state, not the
2557 	 * number of states.
2558 	 */
2559 	for (state = (int)ctrl->npss; state >= 0; state--) {
2560 		u64 total_latency_us, exit_latency_us, transition_ms;
2561 
2562 		if (target)
2563 			table->entries[state] = target;
2564 
2565 		/*
2566 		 * Don't allow transitions to the deepest state if it's quirked
2567 		 * off.
2568 		 */
2569 		if (state == ctrl->npss &&
2570 		    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2571 			continue;
2572 
2573 		/*
2574 		 * Is this state a useful non-operational state for higher-power
2575 		 * states to autonomously transition to?
2576 		 */
2577 		if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2578 			continue;
2579 
2580 		exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2581 		if (exit_latency_us > ctrl->ps_max_latency_us)
2582 			continue;
2583 
2584 		total_latency_us = exit_latency_us +
2585 			le32_to_cpu(ctrl->psd[state].entry_lat);
2586 
2587 		/*
2588 		 * This state is good. It can be used as the APST idle target
2589 		 * for higher power states.
2590 		 */
2591 		if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2592 			if (!nvme_apst_get_transition_time(total_latency_us,
2593 					&transition_ms, &last_lt_index))
2594 				continue;
2595 		} else {
2596 			transition_ms = total_latency_us + 19;
2597 			do_div(transition_ms, 20);
2598 			if (transition_ms > (1 << 24) - 1)
2599 				transition_ms = (1 << 24) - 1;
2600 		}
2601 
2602 		target = cpu_to_le64((state << 3) | (transition_ms << 8));
2603 		if (max_ps == -1)
2604 			max_ps = state;
2605 		if (total_latency_us > max_lat_us)
2606 			max_lat_us = total_latency_us;
2607 	}
2608 
2609 	if (max_ps == -1)
2610 		dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2611 	else
2612 		dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2613 			max_ps, max_lat_us, (int)sizeof(*table), table);
2614 	apste = 1;
2615 
2616 done:
2617 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2618 				table, sizeof(*table), NULL);
2619 	if (ret)
2620 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2621 	kfree(table);
2622 	return ret;
2623 }
2624 
2625 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2626 {
2627 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2628 	u64 latency;
2629 
2630 	switch (val) {
2631 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2632 	case PM_QOS_LATENCY_ANY:
2633 		latency = U64_MAX;
2634 		break;
2635 
2636 	default:
2637 		latency = val;
2638 	}
2639 
2640 	if (ctrl->ps_max_latency_us != latency) {
2641 		ctrl->ps_max_latency_us = latency;
2642 		if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
2643 			nvme_configure_apst(ctrl);
2644 	}
2645 }
2646 
2647 struct nvme_core_quirk_entry {
2648 	/*
2649 	 * NVMe model and firmware strings are padded with spaces.  For
2650 	 * simplicity, strings in the quirk table are padded with NULLs
2651 	 * instead.
2652 	 */
2653 	u16 vid;
2654 	const char *mn;
2655 	const char *fr;
2656 	unsigned long quirks;
2657 };
2658 
2659 static const struct nvme_core_quirk_entry core_quirks[] = {
2660 	{
2661 		/*
2662 		 * This Toshiba device seems to die using any APST states.  See:
2663 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2664 		 */
2665 		.vid = 0x1179,
2666 		.mn = "THNSF5256GPUK TOSHIBA",
2667 		.quirks = NVME_QUIRK_NO_APST,
2668 	},
2669 	{
2670 		/*
2671 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2672 		 * condition associated with actions related to suspend to idle
2673 		 * LiteON has resolved the problem in future firmware
2674 		 */
2675 		.vid = 0x14a4,
2676 		.fr = "22301111",
2677 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2678 	},
2679 	{
2680 		/*
2681 		 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2682 		 * aborts I/O during any load, but more easily reproducible
2683 		 * with discards (fstrim).
2684 		 *
2685 		 * The device is left in a state where it is also not possible
2686 		 * to use "nvme set-feature" to disable APST, but booting with
2687 		 * nvme_core.default_ps_max_latency=0 works.
2688 		 */
2689 		.vid = 0x1e0f,
2690 		.mn = "KCD6XVUL6T40",
2691 		.quirks = NVME_QUIRK_NO_APST,
2692 	},
2693 	{
2694 		/*
2695 		 * The external Samsung X5 SSD fails initialization without a
2696 		 * delay before checking if it is ready and has a whole set of
2697 		 * other problems.  To make this even more interesting, it
2698 		 * shares the PCI ID with internal Samsung 970 Evo Plus that
2699 		 * does not need or want these quirks.
2700 		 */
2701 		.vid = 0x144d,
2702 		.mn = "Samsung Portable SSD X5",
2703 		.quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2704 			  NVME_QUIRK_NO_DEEPEST_PS |
2705 			  NVME_QUIRK_IGNORE_DEV_SUBNQN,
2706 	}
2707 };
2708 
2709 /* match is null-terminated but idstr is space-padded. */
2710 static bool string_matches(const char *idstr, const char *match, size_t len)
2711 {
2712 	size_t matchlen;
2713 
2714 	if (!match)
2715 		return true;
2716 
2717 	matchlen = strlen(match);
2718 	WARN_ON_ONCE(matchlen > len);
2719 
2720 	if (memcmp(idstr, match, matchlen))
2721 		return false;
2722 
2723 	for (; matchlen < len; matchlen++)
2724 		if (idstr[matchlen] != ' ')
2725 			return false;
2726 
2727 	return true;
2728 }
2729 
2730 static bool quirk_matches(const struct nvme_id_ctrl *id,
2731 			  const struct nvme_core_quirk_entry *q)
2732 {
2733 	return q->vid == le16_to_cpu(id->vid) &&
2734 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2735 		string_matches(id->fr, q->fr, sizeof(id->fr));
2736 }
2737 
2738 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2739 		struct nvme_id_ctrl *id)
2740 {
2741 	size_t nqnlen;
2742 	int off;
2743 
2744 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2745 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2746 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2747 			strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2748 			return;
2749 		}
2750 
2751 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2752 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2753 	}
2754 
2755 	/*
2756 	 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2757 	 * Base Specification 2.0.  It is slightly different from the format
2758 	 * specified there due to historic reasons, and we can't change it now.
2759 	 */
2760 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2761 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2762 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2763 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2764 	off += sizeof(id->sn);
2765 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2766 	off += sizeof(id->mn);
2767 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2768 }
2769 
2770 static void nvme_release_subsystem(struct device *dev)
2771 {
2772 	struct nvme_subsystem *subsys =
2773 		container_of(dev, struct nvme_subsystem, dev);
2774 
2775 	if (subsys->instance >= 0)
2776 		ida_free(&nvme_instance_ida, subsys->instance);
2777 	kfree(subsys);
2778 }
2779 
2780 static void nvme_destroy_subsystem(struct kref *ref)
2781 {
2782 	struct nvme_subsystem *subsys =
2783 			container_of(ref, struct nvme_subsystem, ref);
2784 
2785 	mutex_lock(&nvme_subsystems_lock);
2786 	list_del(&subsys->entry);
2787 	mutex_unlock(&nvme_subsystems_lock);
2788 
2789 	ida_destroy(&subsys->ns_ida);
2790 	device_del(&subsys->dev);
2791 	put_device(&subsys->dev);
2792 }
2793 
2794 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2795 {
2796 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2797 }
2798 
2799 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2800 {
2801 	struct nvme_subsystem *subsys;
2802 
2803 	lockdep_assert_held(&nvme_subsystems_lock);
2804 
2805 	/*
2806 	 * Fail matches for discovery subsystems. This results
2807 	 * in each discovery controller bound to a unique subsystem.
2808 	 * This avoids issues with validating controller values
2809 	 * that can only be true when there is a single unique subsystem.
2810 	 * There may be multiple and completely independent entities
2811 	 * that provide discovery controllers.
2812 	 */
2813 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2814 		return NULL;
2815 
2816 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2817 		if (strcmp(subsys->subnqn, subsysnqn))
2818 			continue;
2819 		if (!kref_get_unless_zero(&subsys->ref))
2820 			continue;
2821 		return subsys;
2822 	}
2823 
2824 	return NULL;
2825 }
2826 
2827 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2828 {
2829 	return ctrl->opts && ctrl->opts->discovery_nqn;
2830 }
2831 
2832 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2833 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2834 {
2835 	struct nvme_ctrl *tmp;
2836 
2837 	lockdep_assert_held(&nvme_subsystems_lock);
2838 
2839 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2840 		if (nvme_state_terminal(tmp))
2841 			continue;
2842 
2843 		if (tmp->cntlid == ctrl->cntlid) {
2844 			dev_err(ctrl->device,
2845 				"Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2846 				ctrl->cntlid, dev_name(tmp->device),
2847 				subsys->subnqn);
2848 			return false;
2849 		}
2850 
2851 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2852 		    nvme_discovery_ctrl(ctrl))
2853 			continue;
2854 
2855 		dev_err(ctrl->device,
2856 			"Subsystem does not support multiple controllers\n");
2857 		return false;
2858 	}
2859 
2860 	return true;
2861 }
2862 
2863 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2864 {
2865 	struct nvme_subsystem *subsys, *found;
2866 	int ret;
2867 
2868 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2869 	if (!subsys)
2870 		return -ENOMEM;
2871 
2872 	subsys->instance = -1;
2873 	mutex_init(&subsys->lock);
2874 	kref_init(&subsys->ref);
2875 	INIT_LIST_HEAD(&subsys->ctrls);
2876 	INIT_LIST_HEAD(&subsys->nsheads);
2877 	nvme_init_subnqn(subsys, ctrl, id);
2878 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2879 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2880 	subsys->vendor_id = le16_to_cpu(id->vid);
2881 	subsys->cmic = id->cmic;
2882 
2883 	/* Versions prior to 1.4 don't necessarily report a valid type */
2884 	if (id->cntrltype == NVME_CTRL_DISC ||
2885 	    !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2886 		subsys->subtype = NVME_NQN_DISC;
2887 	else
2888 		subsys->subtype = NVME_NQN_NVME;
2889 
2890 	if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2891 		dev_err(ctrl->device,
2892 			"Subsystem %s is not a discovery controller",
2893 			subsys->subnqn);
2894 		kfree(subsys);
2895 		return -EINVAL;
2896 	}
2897 	subsys->awupf = le16_to_cpu(id->awupf);
2898 	nvme_mpath_default_iopolicy(subsys);
2899 
2900 	subsys->dev.class = &nvme_subsys_class;
2901 	subsys->dev.release = nvme_release_subsystem;
2902 	subsys->dev.groups = nvme_subsys_attrs_groups;
2903 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2904 	device_initialize(&subsys->dev);
2905 
2906 	mutex_lock(&nvme_subsystems_lock);
2907 	found = __nvme_find_get_subsystem(subsys->subnqn);
2908 	if (found) {
2909 		put_device(&subsys->dev);
2910 		subsys = found;
2911 
2912 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2913 			ret = -EINVAL;
2914 			goto out_put_subsystem;
2915 		}
2916 	} else {
2917 		ret = device_add(&subsys->dev);
2918 		if (ret) {
2919 			dev_err(ctrl->device,
2920 				"failed to register subsystem device.\n");
2921 			put_device(&subsys->dev);
2922 			goto out_unlock;
2923 		}
2924 		ida_init(&subsys->ns_ida);
2925 		list_add_tail(&subsys->entry, &nvme_subsystems);
2926 	}
2927 
2928 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2929 				dev_name(ctrl->device));
2930 	if (ret) {
2931 		dev_err(ctrl->device,
2932 			"failed to create sysfs link from subsystem.\n");
2933 		goto out_put_subsystem;
2934 	}
2935 
2936 	if (!found)
2937 		subsys->instance = ctrl->instance;
2938 	ctrl->subsys = subsys;
2939 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2940 	mutex_unlock(&nvme_subsystems_lock);
2941 	return 0;
2942 
2943 out_put_subsystem:
2944 	nvme_put_subsystem(subsys);
2945 out_unlock:
2946 	mutex_unlock(&nvme_subsystems_lock);
2947 	return ret;
2948 }
2949 
2950 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2951 		void *log, size_t size, u64 offset)
2952 {
2953 	struct nvme_command c = { };
2954 	u32 dwlen = nvme_bytes_to_numd(size);
2955 
2956 	c.get_log_page.opcode = nvme_admin_get_log_page;
2957 	c.get_log_page.nsid = cpu_to_le32(nsid);
2958 	c.get_log_page.lid = log_page;
2959 	c.get_log_page.lsp = lsp;
2960 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2961 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2962 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2963 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2964 	c.get_log_page.csi = csi;
2965 
2966 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2967 }
2968 
2969 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2970 				struct nvme_effects_log **log)
2971 {
2972 	struct nvme_effects_log	*cel = xa_load(&ctrl->cels, csi);
2973 	int ret;
2974 
2975 	if (cel)
2976 		goto out;
2977 
2978 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2979 	if (!cel)
2980 		return -ENOMEM;
2981 
2982 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2983 			cel, sizeof(*cel), 0);
2984 	if (ret) {
2985 		kfree(cel);
2986 		return ret;
2987 	}
2988 
2989 	xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2990 out:
2991 	*log = cel;
2992 	return 0;
2993 }
2994 
2995 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2996 {
2997 	u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2998 
2999 	if (check_shl_overflow(1U, units + page_shift - 9, &val))
3000 		return UINT_MAX;
3001 	return val;
3002 }
3003 
3004 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
3005 {
3006 	struct nvme_command c = { };
3007 	struct nvme_id_ctrl_nvm *id;
3008 	int ret;
3009 
3010 	/*
3011 	 * Even though NVMe spec explicitly states that MDTS is not applicable
3012 	 * to the write-zeroes, we are cautious and limit the size to the
3013 	 * controllers max_hw_sectors value, which is based on the MDTS field
3014 	 * and possibly other limiting factors.
3015 	 */
3016 	if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
3017 	    !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
3018 		ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
3019 	else
3020 		ctrl->max_zeroes_sectors = 0;
3021 
3022 	if (ctrl->subsys->subtype != NVME_NQN_NVME ||
3023 	    nvme_ctrl_limited_cns(ctrl) ||
3024 	    test_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags))
3025 		return 0;
3026 
3027 	id = kzalloc(sizeof(*id), GFP_KERNEL);
3028 	if (!id)
3029 		return -ENOMEM;
3030 
3031 	c.identify.opcode = nvme_admin_identify;
3032 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
3033 	c.identify.csi = NVME_CSI_NVM;
3034 
3035 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
3036 	if (ret)
3037 		goto free_data;
3038 
3039 	ctrl->dmrl = id->dmrl;
3040 	ctrl->dmrsl = le32_to_cpu(id->dmrsl);
3041 	if (id->wzsl)
3042 		ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
3043 
3044 free_data:
3045 	if (ret > 0)
3046 		set_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags);
3047 	kfree(id);
3048 	return ret;
3049 }
3050 
3051 static void nvme_init_known_nvm_effects(struct nvme_ctrl *ctrl)
3052 {
3053 	struct nvme_effects_log	*log = ctrl->effects;
3054 
3055 	log->acs[nvme_admin_format_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
3056 						NVME_CMD_EFFECTS_NCC |
3057 						NVME_CMD_EFFECTS_CSE_MASK);
3058 	log->acs[nvme_admin_sanitize_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
3059 						NVME_CMD_EFFECTS_CSE_MASK);
3060 
3061 	/*
3062 	 * The spec says the result of a security receive command depends on
3063 	 * the previous security send command. As such, many vendors log this
3064 	 * command as one to submitted only when no other commands to the same
3065 	 * namespace are outstanding. The intention is to tell the host to
3066 	 * prevent mixing security send and receive.
3067 	 *
3068 	 * This driver can only enforce such exclusive access against IO
3069 	 * queues, though. We are not readily able to enforce such a rule for
3070 	 * two commands to the admin queue, which is the only queue that
3071 	 * matters for this command.
3072 	 *
3073 	 * Rather than blindly freezing the IO queues for this effect that
3074 	 * doesn't even apply to IO, mask it off.
3075 	 */
3076 	log->acs[nvme_admin_security_recv] &= cpu_to_le32(~NVME_CMD_EFFECTS_CSE_MASK);
3077 
3078 	log->iocs[nvme_cmd_write] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3079 	log->iocs[nvme_cmd_write_zeroes] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3080 	log->iocs[nvme_cmd_write_uncor] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3081 }
3082 
3083 static int nvme_init_effects(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
3084 {
3085 	int ret = 0;
3086 
3087 	if (ctrl->effects)
3088 		return 0;
3089 
3090 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3091 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3092 		if (ret < 0)
3093 			return ret;
3094 	}
3095 
3096 	if (!ctrl->effects) {
3097 		ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
3098 		if (!ctrl->effects)
3099 			return -ENOMEM;
3100 		xa_store(&ctrl->cels, NVME_CSI_NVM, ctrl->effects, GFP_KERNEL);
3101 	}
3102 
3103 	nvme_init_known_nvm_effects(ctrl);
3104 	return 0;
3105 }
3106 
3107 static int nvme_check_ctrl_fabric_info(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
3108 {
3109 	/*
3110 	 * In fabrics we need to verify the cntlid matches the
3111 	 * admin connect
3112 	 */
3113 	if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3114 		dev_err(ctrl->device,
3115 			"Mismatching cntlid: Connect %u vs Identify %u, rejecting\n",
3116 			ctrl->cntlid, le16_to_cpu(id->cntlid));
3117 		return -EINVAL;
3118 	}
3119 
3120 	if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3121 		dev_err(ctrl->device,
3122 			"keep-alive support is mandatory for fabrics\n");
3123 		return -EINVAL;
3124 	}
3125 
3126 	if (!nvme_discovery_ctrl(ctrl) && ctrl->ioccsz < 4) {
3127 		dev_err(ctrl->device,
3128 			"I/O queue command capsule supported size %d < 4\n",
3129 			ctrl->ioccsz);
3130 		return -EINVAL;
3131 	}
3132 
3133 	if (!nvme_discovery_ctrl(ctrl) && ctrl->iorcsz < 1) {
3134 		dev_err(ctrl->device,
3135 			"I/O queue response capsule supported size %d < 1\n",
3136 			ctrl->iorcsz);
3137 		return -EINVAL;
3138 	}
3139 
3140 	if (!ctrl->maxcmd) {
3141 		dev_err(ctrl->device, "Maximum outstanding commands is 0\n");
3142 		return -EINVAL;
3143 	}
3144 
3145 	return 0;
3146 }
3147 
3148 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3149 {
3150 	struct queue_limits lim;
3151 	struct nvme_id_ctrl *id;
3152 	u32 max_hw_sectors;
3153 	bool prev_apst_enabled;
3154 	int ret;
3155 
3156 	ret = nvme_identify_ctrl(ctrl, &id);
3157 	if (ret) {
3158 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3159 		return -EIO;
3160 	}
3161 
3162 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
3163 		ctrl->cntlid = le16_to_cpu(id->cntlid);
3164 
3165 	if (!ctrl->identified) {
3166 		unsigned int i;
3167 
3168 		/*
3169 		 * Check for quirks.  Quirk can depend on firmware version,
3170 		 * so, in principle, the set of quirks present can change
3171 		 * across a reset.  As a possible future enhancement, we
3172 		 * could re-scan for quirks every time we reinitialize
3173 		 * the device, but we'd have to make sure that the driver
3174 		 * behaves intelligently if the quirks change.
3175 		 */
3176 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3177 			if (quirk_matches(id, &core_quirks[i]))
3178 				ctrl->quirks |= core_quirks[i].quirks;
3179 		}
3180 
3181 		ret = nvme_init_subsystem(ctrl, id);
3182 		if (ret)
3183 			goto out_free;
3184 
3185 		ret = nvme_init_effects(ctrl, id);
3186 		if (ret)
3187 			goto out_free;
3188 	}
3189 	memcpy(ctrl->subsys->firmware_rev, id->fr,
3190 	       sizeof(ctrl->subsys->firmware_rev));
3191 
3192 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3193 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3194 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3195 	}
3196 
3197 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3198 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3199 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3200 
3201 	ctrl->oacs = le16_to_cpu(id->oacs);
3202 	ctrl->oncs = le16_to_cpu(id->oncs);
3203 	ctrl->mtfa = le16_to_cpu(id->mtfa);
3204 	ctrl->oaes = le32_to_cpu(id->oaes);
3205 	ctrl->wctemp = le16_to_cpu(id->wctemp);
3206 	ctrl->cctemp = le16_to_cpu(id->cctemp);
3207 
3208 	atomic_set(&ctrl->abort_limit, id->acl + 1);
3209 	ctrl->vwc = id->vwc;
3210 	if (id->mdts)
3211 		max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3212 	else
3213 		max_hw_sectors = UINT_MAX;
3214 	ctrl->max_hw_sectors =
3215 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3216 
3217 	lim = queue_limits_start_update(ctrl->admin_q);
3218 	nvme_set_ctrl_limits(ctrl, &lim);
3219 	ret = queue_limits_commit_update(ctrl->admin_q, &lim);
3220 	if (ret)
3221 		goto out_free;
3222 
3223 	ctrl->sgls = le32_to_cpu(id->sgls);
3224 	ctrl->kas = le16_to_cpu(id->kas);
3225 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
3226 	ctrl->ctratt = le32_to_cpu(id->ctratt);
3227 
3228 	ctrl->cntrltype = id->cntrltype;
3229 	ctrl->dctype = id->dctype;
3230 
3231 	if (id->rtd3e) {
3232 		/* us -> s */
3233 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3234 
3235 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3236 						 shutdown_timeout, 60);
3237 
3238 		if (ctrl->shutdown_timeout != shutdown_timeout)
3239 			dev_info(ctrl->device,
3240 				 "Shutdown timeout set to %u seconds\n",
3241 				 ctrl->shutdown_timeout);
3242 	} else
3243 		ctrl->shutdown_timeout = shutdown_timeout;
3244 
3245 	ctrl->npss = id->npss;
3246 	ctrl->apsta = id->apsta;
3247 	prev_apst_enabled = ctrl->apst_enabled;
3248 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3249 		if (force_apst && id->apsta) {
3250 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3251 			ctrl->apst_enabled = true;
3252 		} else {
3253 			ctrl->apst_enabled = false;
3254 		}
3255 	} else {
3256 		ctrl->apst_enabled = id->apsta;
3257 	}
3258 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3259 
3260 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3261 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3262 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3263 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3264 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3265 
3266 		ret = nvme_check_ctrl_fabric_info(ctrl, id);
3267 		if (ret)
3268 			goto out_free;
3269 	} else {
3270 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3271 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3272 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3273 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3274 	}
3275 
3276 	ret = nvme_mpath_init_identify(ctrl, id);
3277 	if (ret < 0)
3278 		goto out_free;
3279 
3280 	if (ctrl->apst_enabled && !prev_apst_enabled)
3281 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3282 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3283 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3284 
3285 out_free:
3286 	kfree(id);
3287 	return ret;
3288 }
3289 
3290 /*
3291  * Initialize the cached copies of the Identify data and various controller
3292  * register in our nvme_ctrl structure.  This should be called as soon as
3293  * the admin queue is fully up and running.
3294  */
3295 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended)
3296 {
3297 	int ret;
3298 
3299 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3300 	if (ret) {
3301 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3302 		return ret;
3303 	}
3304 
3305 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3306 
3307 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3308 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3309 
3310 	ret = nvme_init_identify(ctrl);
3311 	if (ret)
3312 		return ret;
3313 
3314 	ret = nvme_configure_apst(ctrl);
3315 	if (ret < 0)
3316 		return ret;
3317 
3318 	ret = nvme_configure_timestamp(ctrl);
3319 	if (ret < 0)
3320 		return ret;
3321 
3322 	ret = nvme_configure_host_options(ctrl);
3323 	if (ret < 0)
3324 		return ret;
3325 
3326 	nvme_configure_opal(ctrl, was_suspended);
3327 
3328 	if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3329 		/*
3330 		 * Do not return errors unless we are in a controller reset,
3331 		 * the controller works perfectly fine without hwmon.
3332 		 */
3333 		ret = nvme_hwmon_init(ctrl);
3334 		if (ret == -EINTR)
3335 			return ret;
3336 	}
3337 
3338 	clear_bit(NVME_CTRL_DIRTY_CAPABILITY, &ctrl->flags);
3339 	ctrl->identified = true;
3340 
3341 	nvme_start_keep_alive(ctrl);
3342 
3343 	return 0;
3344 }
3345 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3346 
3347 static int nvme_dev_open(struct inode *inode, struct file *file)
3348 {
3349 	struct nvme_ctrl *ctrl =
3350 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3351 
3352 	switch (nvme_ctrl_state(ctrl)) {
3353 	case NVME_CTRL_LIVE:
3354 		break;
3355 	default:
3356 		return -EWOULDBLOCK;
3357 	}
3358 
3359 	nvme_get_ctrl(ctrl);
3360 	if (!try_module_get(ctrl->ops->module)) {
3361 		nvme_put_ctrl(ctrl);
3362 		return -EINVAL;
3363 	}
3364 
3365 	file->private_data = ctrl;
3366 	return 0;
3367 }
3368 
3369 static int nvme_dev_release(struct inode *inode, struct file *file)
3370 {
3371 	struct nvme_ctrl *ctrl =
3372 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3373 
3374 	module_put(ctrl->ops->module);
3375 	nvme_put_ctrl(ctrl);
3376 	return 0;
3377 }
3378 
3379 static const struct file_operations nvme_dev_fops = {
3380 	.owner		= THIS_MODULE,
3381 	.open		= nvme_dev_open,
3382 	.release	= nvme_dev_release,
3383 	.unlocked_ioctl	= nvme_dev_ioctl,
3384 	.compat_ioctl	= compat_ptr_ioctl,
3385 	.uring_cmd	= nvme_dev_uring_cmd,
3386 };
3387 
3388 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3389 		unsigned nsid)
3390 {
3391 	struct nvme_ns_head *h;
3392 
3393 	lockdep_assert_held(&ctrl->subsys->lock);
3394 
3395 	list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3396 		/*
3397 		 * Private namespaces can share NSIDs under some conditions.
3398 		 * In that case we can't use the same ns_head for namespaces
3399 		 * with the same NSID.
3400 		 */
3401 		if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3402 			continue;
3403 		if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3404 			return h;
3405 	}
3406 
3407 	return NULL;
3408 }
3409 
3410 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3411 		struct nvme_ns_ids *ids)
3412 {
3413 	bool has_uuid = !uuid_is_null(&ids->uuid);
3414 	bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3415 	bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3416 	struct nvme_ns_head *h;
3417 
3418 	lockdep_assert_held(&subsys->lock);
3419 
3420 	list_for_each_entry(h, &subsys->nsheads, entry) {
3421 		if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3422 			return -EINVAL;
3423 		if (has_nguid &&
3424 		    memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3425 			return -EINVAL;
3426 		if (has_eui64 &&
3427 		    memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3428 			return -EINVAL;
3429 	}
3430 
3431 	return 0;
3432 }
3433 
3434 static void nvme_cdev_rel(struct device *dev)
3435 {
3436 	ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3437 }
3438 
3439 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3440 {
3441 	cdev_device_del(cdev, cdev_device);
3442 	put_device(cdev_device);
3443 }
3444 
3445 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3446 		const struct file_operations *fops, struct module *owner)
3447 {
3448 	int minor, ret;
3449 
3450 	minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3451 	if (minor < 0)
3452 		return minor;
3453 	cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3454 	cdev_device->class = &nvme_ns_chr_class;
3455 	cdev_device->release = nvme_cdev_rel;
3456 	device_initialize(cdev_device);
3457 	cdev_init(cdev, fops);
3458 	cdev->owner = owner;
3459 	ret = cdev_device_add(cdev, cdev_device);
3460 	if (ret)
3461 		put_device(cdev_device);
3462 
3463 	return ret;
3464 }
3465 
3466 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3467 {
3468 	return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3469 }
3470 
3471 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3472 {
3473 	nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3474 	return 0;
3475 }
3476 
3477 static const struct file_operations nvme_ns_chr_fops = {
3478 	.owner		= THIS_MODULE,
3479 	.open		= nvme_ns_chr_open,
3480 	.release	= nvme_ns_chr_release,
3481 	.unlocked_ioctl	= nvme_ns_chr_ioctl,
3482 	.compat_ioctl	= compat_ptr_ioctl,
3483 	.uring_cmd	= nvme_ns_chr_uring_cmd,
3484 	.uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
3485 };
3486 
3487 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3488 {
3489 	int ret;
3490 
3491 	ns->cdev_device.parent = ns->ctrl->device;
3492 	ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3493 			   ns->ctrl->instance, ns->head->instance);
3494 	if (ret)
3495 		return ret;
3496 
3497 	return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3498 			     ns->ctrl->ops->module);
3499 }
3500 
3501 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3502 		struct nvme_ns_info *info)
3503 {
3504 	struct nvme_ns_head *head;
3505 	size_t size = sizeof(*head);
3506 	int ret = -ENOMEM;
3507 
3508 #ifdef CONFIG_NVME_MULTIPATH
3509 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3510 #endif
3511 
3512 	head = kzalloc(size, GFP_KERNEL);
3513 	if (!head)
3514 		goto out;
3515 	ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3516 	if (ret < 0)
3517 		goto out_free_head;
3518 	head->instance = ret;
3519 	INIT_LIST_HEAD(&head->list);
3520 	ret = init_srcu_struct(&head->srcu);
3521 	if (ret)
3522 		goto out_ida_remove;
3523 	head->subsys = ctrl->subsys;
3524 	head->ns_id = info->nsid;
3525 	head->ids = info->ids;
3526 	head->shared = info->is_shared;
3527 	ratelimit_state_init(&head->rs_nuse, 5 * HZ, 1);
3528 	ratelimit_set_flags(&head->rs_nuse, RATELIMIT_MSG_ON_RELEASE);
3529 	kref_init(&head->ref);
3530 
3531 	if (head->ids.csi) {
3532 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3533 		if (ret)
3534 			goto out_cleanup_srcu;
3535 	} else
3536 		head->effects = ctrl->effects;
3537 
3538 	ret = nvme_mpath_alloc_disk(ctrl, head);
3539 	if (ret)
3540 		goto out_cleanup_srcu;
3541 
3542 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3543 
3544 	kref_get(&ctrl->subsys->ref);
3545 
3546 	return head;
3547 out_cleanup_srcu:
3548 	cleanup_srcu_struct(&head->srcu);
3549 out_ida_remove:
3550 	ida_free(&ctrl->subsys->ns_ida, head->instance);
3551 out_free_head:
3552 	kfree(head);
3553 out:
3554 	if (ret > 0)
3555 		ret = blk_status_to_errno(nvme_error_status(ret));
3556 	return ERR_PTR(ret);
3557 }
3558 
3559 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3560 		struct nvme_ns_ids *ids)
3561 {
3562 	struct nvme_subsystem *s;
3563 	int ret = 0;
3564 
3565 	/*
3566 	 * Note that this check is racy as we try to avoid holding the global
3567 	 * lock over the whole ns_head creation.  But it is only intended as
3568 	 * a sanity check anyway.
3569 	 */
3570 	mutex_lock(&nvme_subsystems_lock);
3571 	list_for_each_entry(s, &nvme_subsystems, entry) {
3572 		if (s == this)
3573 			continue;
3574 		mutex_lock(&s->lock);
3575 		ret = nvme_subsys_check_duplicate_ids(s, ids);
3576 		mutex_unlock(&s->lock);
3577 		if (ret)
3578 			break;
3579 	}
3580 	mutex_unlock(&nvme_subsystems_lock);
3581 
3582 	return ret;
3583 }
3584 
3585 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
3586 {
3587 	struct nvme_ctrl *ctrl = ns->ctrl;
3588 	struct nvme_ns_head *head = NULL;
3589 	int ret;
3590 
3591 	ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
3592 	if (ret) {
3593 		/*
3594 		 * We've found two different namespaces on two different
3595 		 * subsystems that report the same ID.  This is pretty nasty
3596 		 * for anything that actually requires unique device
3597 		 * identification.  In the kernel we need this for multipathing,
3598 		 * and in user space the /dev/disk/by-id/ links rely on it.
3599 		 *
3600 		 * If the device also claims to be multi-path capable back off
3601 		 * here now and refuse the probe the second device as this is a
3602 		 * recipe for data corruption.  If not this is probably a
3603 		 * cheap consumer device if on the PCIe bus, so let the user
3604 		 * proceed and use the shiny toy, but warn that with changing
3605 		 * probing order (which due to our async probing could just be
3606 		 * device taking longer to startup) the other device could show
3607 		 * up at any time.
3608 		 */
3609 		nvme_print_device_info(ctrl);
3610 		if ((ns->ctrl->ops->flags & NVME_F_FABRICS) || /* !PCIe */
3611 		    ((ns->ctrl->subsys->cmic & NVME_CTRL_CMIC_MULTI_CTRL) &&
3612 		     info->is_shared)) {
3613 			dev_err(ctrl->device,
3614 				"ignoring nsid %d because of duplicate IDs\n",
3615 				info->nsid);
3616 			return ret;
3617 		}
3618 
3619 		dev_err(ctrl->device,
3620 			"clearing duplicate IDs for nsid %d\n", info->nsid);
3621 		dev_err(ctrl->device,
3622 			"use of /dev/disk/by-id/ may cause data corruption\n");
3623 		memset(&info->ids.nguid, 0, sizeof(info->ids.nguid));
3624 		memset(&info->ids.uuid, 0, sizeof(info->ids.uuid));
3625 		memset(&info->ids.eui64, 0, sizeof(info->ids.eui64));
3626 		ctrl->quirks |= NVME_QUIRK_BOGUS_NID;
3627 	}
3628 
3629 	mutex_lock(&ctrl->subsys->lock);
3630 	head = nvme_find_ns_head(ctrl, info->nsid);
3631 	if (!head) {
3632 		ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
3633 		if (ret) {
3634 			dev_err(ctrl->device,
3635 				"duplicate IDs in subsystem for nsid %d\n",
3636 				info->nsid);
3637 			goto out_unlock;
3638 		}
3639 		head = nvme_alloc_ns_head(ctrl, info);
3640 		if (IS_ERR(head)) {
3641 			ret = PTR_ERR(head);
3642 			goto out_unlock;
3643 		}
3644 	} else {
3645 		ret = -EINVAL;
3646 		if (!info->is_shared || !head->shared) {
3647 			dev_err(ctrl->device,
3648 				"Duplicate unshared namespace %d\n",
3649 				info->nsid);
3650 			goto out_put_ns_head;
3651 		}
3652 		if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
3653 			dev_err(ctrl->device,
3654 				"IDs don't match for shared namespace %d\n",
3655 					info->nsid);
3656 			goto out_put_ns_head;
3657 		}
3658 
3659 		if (!multipath) {
3660 			dev_warn(ctrl->device,
3661 				"Found shared namespace %d, but multipathing not supported.\n",
3662 				info->nsid);
3663 			dev_warn_once(ctrl->device,
3664 				"Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
3665 		}
3666 	}
3667 
3668 	list_add_tail_rcu(&ns->siblings, &head->list);
3669 	ns->head = head;
3670 	mutex_unlock(&ctrl->subsys->lock);
3671 	return 0;
3672 
3673 out_put_ns_head:
3674 	nvme_put_ns_head(head);
3675 out_unlock:
3676 	mutex_unlock(&ctrl->subsys->lock);
3677 	return ret;
3678 }
3679 
3680 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3681 {
3682 	struct nvme_ns *ns, *ret = NULL;
3683 
3684 	down_read(&ctrl->namespaces_rwsem);
3685 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3686 		if (ns->head->ns_id == nsid) {
3687 			if (!nvme_get_ns(ns))
3688 				continue;
3689 			ret = ns;
3690 			break;
3691 		}
3692 		if (ns->head->ns_id > nsid)
3693 			break;
3694 	}
3695 	up_read(&ctrl->namespaces_rwsem);
3696 	return ret;
3697 }
3698 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3699 
3700 /*
3701  * Add the namespace to the controller list while keeping the list ordered.
3702  */
3703 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3704 {
3705 	struct nvme_ns *tmp;
3706 
3707 	list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3708 		if (tmp->head->ns_id < ns->head->ns_id) {
3709 			list_add(&ns->list, &tmp->list);
3710 			return;
3711 		}
3712 	}
3713 	list_add(&ns->list, &ns->ctrl->namespaces);
3714 }
3715 
3716 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
3717 {
3718 	struct nvme_ns *ns;
3719 	struct gendisk *disk;
3720 	int node = ctrl->numa_node;
3721 
3722 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3723 	if (!ns)
3724 		return;
3725 
3726 	disk = blk_mq_alloc_disk(ctrl->tagset, NULL, ns);
3727 	if (IS_ERR(disk))
3728 		goto out_free_ns;
3729 	disk->fops = &nvme_bdev_ops;
3730 	disk->private_data = ns;
3731 
3732 	ns->disk = disk;
3733 	ns->queue = disk->queue;
3734 
3735 	if (ctrl->opts && ctrl->opts->data_digest)
3736 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3737 
3738 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3739 	if (ctrl->ops->supports_pci_p2pdma &&
3740 	    ctrl->ops->supports_pci_p2pdma(ctrl))
3741 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3742 
3743 	ns->ctrl = ctrl;
3744 	kref_init(&ns->kref);
3745 
3746 	if (nvme_init_ns_head(ns, info))
3747 		goto out_cleanup_disk;
3748 
3749 	/*
3750 	 * If multipathing is enabled, the device name for all disks and not
3751 	 * just those that represent shared namespaces needs to be based on the
3752 	 * subsystem instance.  Using the controller instance for private
3753 	 * namespaces could lead to naming collisions between shared and private
3754 	 * namespaces if they don't use a common numbering scheme.
3755 	 *
3756 	 * If multipathing is not enabled, disk names must use the controller
3757 	 * instance as shared namespaces will show up as multiple block
3758 	 * devices.
3759 	 */
3760 	if (nvme_ns_head_multipath(ns->head)) {
3761 		sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
3762 			ctrl->instance, ns->head->instance);
3763 		disk->flags |= GENHD_FL_HIDDEN;
3764 	} else if (multipath) {
3765 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
3766 			ns->head->instance);
3767 	} else {
3768 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3769 			ns->head->instance);
3770 	}
3771 
3772 	if (nvme_update_ns_info(ns, info))
3773 		goto out_unlink_ns;
3774 
3775 	down_write(&ctrl->namespaces_rwsem);
3776 	/*
3777 	 * Ensure that no namespaces are added to the ctrl list after the queues
3778 	 * are frozen, thereby avoiding a deadlock between scan and reset.
3779 	 */
3780 	if (test_bit(NVME_CTRL_FROZEN, &ctrl->flags)) {
3781 		up_write(&ctrl->namespaces_rwsem);
3782 		goto out_unlink_ns;
3783 	}
3784 	nvme_ns_add_to_ctrl_list(ns);
3785 	up_write(&ctrl->namespaces_rwsem);
3786 	nvme_get_ctrl(ctrl);
3787 
3788 	if (device_add_disk(ctrl->device, ns->disk, nvme_ns_attr_groups))
3789 		goto out_cleanup_ns_from_list;
3790 
3791 	if (!nvme_ns_head_multipath(ns->head))
3792 		nvme_add_ns_cdev(ns);
3793 
3794 	nvme_mpath_add_disk(ns, info->anagrpid);
3795 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3796 
3797 	/*
3798 	 * Set ns->disk->device->driver_data to ns so we can access
3799 	 * ns->head->passthru_err_log_enabled in
3800 	 * nvme_io_passthru_err_log_enabled_[store | show]().
3801 	 */
3802 	dev_set_drvdata(disk_to_dev(ns->disk), ns);
3803 
3804 	return;
3805 
3806  out_cleanup_ns_from_list:
3807 	nvme_put_ctrl(ctrl);
3808 	down_write(&ctrl->namespaces_rwsem);
3809 	list_del_init(&ns->list);
3810 	up_write(&ctrl->namespaces_rwsem);
3811  out_unlink_ns:
3812 	mutex_lock(&ctrl->subsys->lock);
3813 	list_del_rcu(&ns->siblings);
3814 	if (list_empty(&ns->head->list))
3815 		list_del_init(&ns->head->entry);
3816 	mutex_unlock(&ctrl->subsys->lock);
3817 	nvme_put_ns_head(ns->head);
3818  out_cleanup_disk:
3819 	put_disk(disk);
3820  out_free_ns:
3821 	kfree(ns);
3822 }
3823 
3824 static void nvme_ns_remove(struct nvme_ns *ns)
3825 {
3826 	bool last_path = false;
3827 
3828 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3829 		return;
3830 
3831 	clear_bit(NVME_NS_READY, &ns->flags);
3832 	set_capacity(ns->disk, 0);
3833 	nvme_fault_inject_fini(&ns->fault_inject);
3834 
3835 	/*
3836 	 * Ensure that !NVME_NS_READY is seen by other threads to prevent
3837 	 * this ns going back into current_path.
3838 	 */
3839 	synchronize_srcu(&ns->head->srcu);
3840 
3841 	/* wait for concurrent submissions */
3842 	if (nvme_mpath_clear_current_path(ns))
3843 		synchronize_srcu(&ns->head->srcu);
3844 
3845 	mutex_lock(&ns->ctrl->subsys->lock);
3846 	list_del_rcu(&ns->siblings);
3847 	if (list_empty(&ns->head->list)) {
3848 		list_del_init(&ns->head->entry);
3849 		last_path = true;
3850 	}
3851 	mutex_unlock(&ns->ctrl->subsys->lock);
3852 
3853 	/* guarantee not available in head->list */
3854 	synchronize_srcu(&ns->head->srcu);
3855 
3856 	if (!nvme_ns_head_multipath(ns->head))
3857 		nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3858 	del_gendisk(ns->disk);
3859 
3860 	down_write(&ns->ctrl->namespaces_rwsem);
3861 	list_del_init(&ns->list);
3862 	up_write(&ns->ctrl->namespaces_rwsem);
3863 
3864 	if (last_path)
3865 		nvme_mpath_shutdown_disk(ns->head);
3866 	nvme_put_ns(ns);
3867 }
3868 
3869 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3870 {
3871 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3872 
3873 	if (ns) {
3874 		nvme_ns_remove(ns);
3875 		nvme_put_ns(ns);
3876 	}
3877 }
3878 
3879 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
3880 {
3881 	int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3882 
3883 	if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
3884 		dev_err(ns->ctrl->device,
3885 			"identifiers changed for nsid %d\n", ns->head->ns_id);
3886 		goto out;
3887 	}
3888 
3889 	ret = nvme_update_ns_info(ns, info);
3890 out:
3891 	/*
3892 	 * Only remove the namespace if we got a fatal error back from the
3893 	 * device, otherwise ignore the error and just move on.
3894 	 *
3895 	 * TODO: we should probably schedule a delayed retry here.
3896 	 */
3897 	if (ret > 0 && (ret & NVME_SC_DNR))
3898 		nvme_ns_remove(ns);
3899 }
3900 
3901 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3902 {
3903 	struct nvme_ns_info info = { .nsid = nsid };
3904 	struct nvme_ns *ns;
3905 	int ret;
3906 
3907 	if (nvme_identify_ns_descs(ctrl, &info))
3908 		return;
3909 
3910 	if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
3911 		dev_warn(ctrl->device,
3912 			"command set not reported for nsid: %d\n", nsid);
3913 		return;
3914 	}
3915 
3916 	/*
3917 	 * If available try to use the Command Set Idependent Identify Namespace
3918 	 * data structure to find all the generic information that is needed to
3919 	 * set up a namespace.  If not fall back to the legacy version.
3920 	 */
3921 	if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
3922 	    (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS))
3923 		ret = nvme_ns_info_from_id_cs_indep(ctrl, &info);
3924 	else
3925 		ret = nvme_ns_info_from_identify(ctrl, &info);
3926 
3927 	if (info.is_removed)
3928 		nvme_ns_remove_by_nsid(ctrl, nsid);
3929 
3930 	/*
3931 	 * Ignore the namespace if it is not ready. We will get an AEN once it
3932 	 * becomes ready and restart the scan.
3933 	 */
3934 	if (ret || !info.is_ready)
3935 		return;
3936 
3937 	ns = nvme_find_get_ns(ctrl, nsid);
3938 	if (ns) {
3939 		nvme_validate_ns(ns, &info);
3940 		nvme_put_ns(ns);
3941 	} else {
3942 		nvme_alloc_ns(ctrl, &info);
3943 	}
3944 }
3945 
3946 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3947 					unsigned nsid)
3948 {
3949 	struct nvme_ns *ns, *next;
3950 	LIST_HEAD(rm_list);
3951 
3952 	down_write(&ctrl->namespaces_rwsem);
3953 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3954 		if (ns->head->ns_id > nsid)
3955 			list_move_tail(&ns->list, &rm_list);
3956 	}
3957 	up_write(&ctrl->namespaces_rwsem);
3958 
3959 	list_for_each_entry_safe(ns, next, &rm_list, list)
3960 		nvme_ns_remove(ns);
3961 
3962 }
3963 
3964 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3965 {
3966 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3967 	__le32 *ns_list;
3968 	u32 prev = 0;
3969 	int ret = 0, i;
3970 
3971 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3972 	if (!ns_list)
3973 		return -ENOMEM;
3974 
3975 	for (;;) {
3976 		struct nvme_command cmd = {
3977 			.identify.opcode	= nvme_admin_identify,
3978 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
3979 			.identify.nsid		= cpu_to_le32(prev),
3980 		};
3981 
3982 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
3983 					    NVME_IDENTIFY_DATA_SIZE);
3984 		if (ret) {
3985 			dev_warn(ctrl->device,
3986 				"Identify NS List failed (status=0x%x)\n", ret);
3987 			goto free;
3988 		}
3989 
3990 		for (i = 0; i < nr_entries; i++) {
3991 			u32 nsid = le32_to_cpu(ns_list[i]);
3992 
3993 			if (!nsid)	/* end of the list? */
3994 				goto out;
3995 			nvme_scan_ns(ctrl, nsid);
3996 			while (++prev < nsid)
3997 				nvme_ns_remove_by_nsid(ctrl, prev);
3998 		}
3999 	}
4000  out:
4001 	nvme_remove_invalid_namespaces(ctrl, prev);
4002  free:
4003 	kfree(ns_list);
4004 	return ret;
4005 }
4006 
4007 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4008 {
4009 	struct nvme_id_ctrl *id;
4010 	u32 nn, i;
4011 
4012 	if (nvme_identify_ctrl(ctrl, &id))
4013 		return;
4014 	nn = le32_to_cpu(id->nn);
4015 	kfree(id);
4016 
4017 	for (i = 1; i <= nn; i++)
4018 		nvme_scan_ns(ctrl, i);
4019 
4020 	nvme_remove_invalid_namespaces(ctrl, nn);
4021 }
4022 
4023 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4024 {
4025 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4026 	__le32 *log;
4027 	int error;
4028 
4029 	log = kzalloc(log_size, GFP_KERNEL);
4030 	if (!log)
4031 		return;
4032 
4033 	/*
4034 	 * We need to read the log to clear the AEN, but we don't want to rely
4035 	 * on it for the changed namespace information as userspace could have
4036 	 * raced with us in reading the log page, which could cause us to miss
4037 	 * updates.
4038 	 */
4039 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4040 			NVME_CSI_NVM, log, log_size, 0);
4041 	if (error)
4042 		dev_warn(ctrl->device,
4043 			"reading changed ns log failed: %d\n", error);
4044 
4045 	kfree(log);
4046 }
4047 
4048 static void nvme_scan_work(struct work_struct *work)
4049 {
4050 	struct nvme_ctrl *ctrl =
4051 		container_of(work, struct nvme_ctrl, scan_work);
4052 	int ret;
4053 
4054 	/* No tagset on a live ctrl means IO queues could not created */
4055 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE || !ctrl->tagset)
4056 		return;
4057 
4058 	/*
4059 	 * Identify controller limits can change at controller reset due to
4060 	 * new firmware download, even though it is not common we cannot ignore
4061 	 * such scenario. Controller's non-mdts limits are reported in the unit
4062 	 * of logical blocks that is dependent on the format of attached
4063 	 * namespace. Hence re-read the limits at the time of ns allocation.
4064 	 */
4065 	ret = nvme_init_non_mdts_limits(ctrl);
4066 	if (ret < 0) {
4067 		dev_warn(ctrl->device,
4068 			"reading non-mdts-limits failed: %d\n", ret);
4069 		return;
4070 	}
4071 
4072 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4073 		dev_info(ctrl->device, "rescanning namespaces.\n");
4074 		nvme_clear_changed_ns_log(ctrl);
4075 	}
4076 
4077 	mutex_lock(&ctrl->scan_lock);
4078 	if (nvme_ctrl_limited_cns(ctrl)) {
4079 		nvme_scan_ns_sequential(ctrl);
4080 	} else {
4081 		/*
4082 		 * Fall back to sequential scan if DNR is set to handle broken
4083 		 * devices which should support Identify NS List (as per the VS
4084 		 * they report) but don't actually support it.
4085 		 */
4086 		ret = nvme_scan_ns_list(ctrl);
4087 		if (ret > 0 && ret & NVME_SC_DNR)
4088 			nvme_scan_ns_sequential(ctrl);
4089 	}
4090 	mutex_unlock(&ctrl->scan_lock);
4091 }
4092 
4093 /*
4094  * This function iterates the namespace list unlocked to allow recovery from
4095  * controller failure. It is up to the caller to ensure the namespace list is
4096  * not modified by scan work while this function is executing.
4097  */
4098 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4099 {
4100 	struct nvme_ns *ns, *next;
4101 	LIST_HEAD(ns_list);
4102 
4103 	/*
4104 	 * make sure to requeue I/O to all namespaces as these
4105 	 * might result from the scan itself and must complete
4106 	 * for the scan_work to make progress
4107 	 */
4108 	nvme_mpath_clear_ctrl_paths(ctrl);
4109 
4110 	/*
4111 	 * Unquiesce io queues so any pending IO won't hang, especially
4112 	 * those submitted from scan work
4113 	 */
4114 	nvme_unquiesce_io_queues(ctrl);
4115 
4116 	/* prevent racing with ns scanning */
4117 	flush_work(&ctrl->scan_work);
4118 
4119 	/*
4120 	 * The dead states indicates the controller was not gracefully
4121 	 * disconnected. In that case, we won't be able to flush any data while
4122 	 * removing the namespaces' disks; fail all the queues now to avoid
4123 	 * potentially having to clean up the failed sync later.
4124 	 */
4125 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_DEAD)
4126 		nvme_mark_namespaces_dead(ctrl);
4127 
4128 	/* this is a no-op when called from the controller reset handler */
4129 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4130 
4131 	down_write(&ctrl->namespaces_rwsem);
4132 	list_splice_init(&ctrl->namespaces, &ns_list);
4133 	up_write(&ctrl->namespaces_rwsem);
4134 
4135 	list_for_each_entry_safe(ns, next, &ns_list, list)
4136 		nvme_ns_remove(ns);
4137 }
4138 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4139 
4140 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env)
4141 {
4142 	const struct nvme_ctrl *ctrl =
4143 		container_of(dev, struct nvme_ctrl, ctrl_device);
4144 	struct nvmf_ctrl_options *opts = ctrl->opts;
4145 	int ret;
4146 
4147 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4148 	if (ret)
4149 		return ret;
4150 
4151 	if (opts) {
4152 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4153 		if (ret)
4154 			return ret;
4155 
4156 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4157 				opts->trsvcid ?: "none");
4158 		if (ret)
4159 			return ret;
4160 
4161 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4162 				opts->host_traddr ?: "none");
4163 		if (ret)
4164 			return ret;
4165 
4166 		ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4167 				opts->host_iface ?: "none");
4168 	}
4169 	return ret;
4170 }
4171 
4172 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4173 {
4174 	char *envp[2] = { envdata, NULL };
4175 
4176 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4177 }
4178 
4179 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4180 {
4181 	char *envp[2] = { NULL, NULL };
4182 	u32 aen_result = ctrl->aen_result;
4183 
4184 	ctrl->aen_result = 0;
4185 	if (!aen_result)
4186 		return;
4187 
4188 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4189 	if (!envp[0])
4190 		return;
4191 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4192 	kfree(envp[0]);
4193 }
4194 
4195 static void nvme_async_event_work(struct work_struct *work)
4196 {
4197 	struct nvme_ctrl *ctrl =
4198 		container_of(work, struct nvme_ctrl, async_event_work);
4199 
4200 	nvme_aen_uevent(ctrl);
4201 
4202 	/*
4203 	 * The transport drivers must guarantee AER submission here is safe by
4204 	 * flushing ctrl async_event_work after changing the controller state
4205 	 * from LIVE and before freeing the admin queue.
4206 	*/
4207 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
4208 		ctrl->ops->submit_async_event(ctrl);
4209 }
4210 
4211 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4212 {
4213 
4214 	u32 csts;
4215 
4216 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4217 		return false;
4218 
4219 	if (csts == ~0)
4220 		return false;
4221 
4222 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4223 }
4224 
4225 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4226 {
4227 	struct nvme_fw_slot_info_log *log;
4228 	u8 next_fw_slot, cur_fw_slot;
4229 
4230 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4231 	if (!log)
4232 		return;
4233 
4234 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4235 			 log, sizeof(*log), 0)) {
4236 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4237 		goto out_free_log;
4238 	}
4239 
4240 	cur_fw_slot = log->afi & 0x7;
4241 	next_fw_slot = (log->afi & 0x70) >> 4;
4242 	if (!cur_fw_slot || (next_fw_slot && (cur_fw_slot != next_fw_slot))) {
4243 		dev_info(ctrl->device,
4244 			 "Firmware is activated after next Controller Level Reset\n");
4245 		goto out_free_log;
4246 	}
4247 
4248 	memcpy(ctrl->subsys->firmware_rev, &log->frs[cur_fw_slot - 1],
4249 		sizeof(ctrl->subsys->firmware_rev));
4250 
4251 out_free_log:
4252 	kfree(log);
4253 }
4254 
4255 static void nvme_fw_act_work(struct work_struct *work)
4256 {
4257 	struct nvme_ctrl *ctrl = container_of(work,
4258 				struct nvme_ctrl, fw_act_work);
4259 	unsigned long fw_act_timeout;
4260 
4261 	nvme_auth_stop(ctrl);
4262 
4263 	if (ctrl->mtfa)
4264 		fw_act_timeout = jiffies +
4265 				msecs_to_jiffies(ctrl->mtfa * 100);
4266 	else
4267 		fw_act_timeout = jiffies +
4268 				msecs_to_jiffies(admin_timeout * 1000);
4269 
4270 	nvme_quiesce_io_queues(ctrl);
4271 	while (nvme_ctrl_pp_status(ctrl)) {
4272 		if (time_after(jiffies, fw_act_timeout)) {
4273 			dev_warn(ctrl->device,
4274 				"Fw activation timeout, reset controller\n");
4275 			nvme_try_sched_reset(ctrl);
4276 			return;
4277 		}
4278 		msleep(100);
4279 	}
4280 
4281 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4282 		return;
4283 
4284 	nvme_unquiesce_io_queues(ctrl);
4285 	/* read FW slot information to clear the AER */
4286 	nvme_get_fw_slot_info(ctrl);
4287 
4288 	queue_work(nvme_wq, &ctrl->async_event_work);
4289 }
4290 
4291 static u32 nvme_aer_type(u32 result)
4292 {
4293 	return result & 0x7;
4294 }
4295 
4296 static u32 nvme_aer_subtype(u32 result)
4297 {
4298 	return (result & 0xff00) >> 8;
4299 }
4300 
4301 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4302 {
4303 	u32 aer_notice_type = nvme_aer_subtype(result);
4304 	bool requeue = true;
4305 
4306 	switch (aer_notice_type) {
4307 	case NVME_AER_NOTICE_NS_CHANGED:
4308 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4309 		nvme_queue_scan(ctrl);
4310 		break;
4311 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4312 		/*
4313 		 * We are (ab)using the RESETTING state to prevent subsequent
4314 		 * recovery actions from interfering with the controller's
4315 		 * firmware activation.
4316 		 */
4317 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4318 			requeue = false;
4319 			queue_work(nvme_wq, &ctrl->fw_act_work);
4320 		}
4321 		break;
4322 #ifdef CONFIG_NVME_MULTIPATH
4323 	case NVME_AER_NOTICE_ANA:
4324 		if (!ctrl->ana_log_buf)
4325 			break;
4326 		queue_work(nvme_wq, &ctrl->ana_work);
4327 		break;
4328 #endif
4329 	case NVME_AER_NOTICE_DISC_CHANGED:
4330 		ctrl->aen_result = result;
4331 		break;
4332 	default:
4333 		dev_warn(ctrl->device, "async event result %08x\n", result);
4334 	}
4335 	return requeue;
4336 }
4337 
4338 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4339 {
4340 	dev_warn(ctrl->device, "resetting controller due to AER\n");
4341 	nvme_reset_ctrl(ctrl);
4342 }
4343 
4344 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4345 		volatile union nvme_result *res)
4346 {
4347 	u32 result = le32_to_cpu(res->u32);
4348 	u32 aer_type = nvme_aer_type(result);
4349 	u32 aer_subtype = nvme_aer_subtype(result);
4350 	bool requeue = true;
4351 
4352 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4353 		return;
4354 
4355 	trace_nvme_async_event(ctrl, result);
4356 	switch (aer_type) {
4357 	case NVME_AER_NOTICE:
4358 		requeue = nvme_handle_aen_notice(ctrl, result);
4359 		break;
4360 	case NVME_AER_ERROR:
4361 		/*
4362 		 * For a persistent internal error, don't run async_event_work
4363 		 * to submit a new AER. The controller reset will do it.
4364 		 */
4365 		if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4366 			nvme_handle_aer_persistent_error(ctrl);
4367 			return;
4368 		}
4369 		fallthrough;
4370 	case NVME_AER_SMART:
4371 	case NVME_AER_CSS:
4372 	case NVME_AER_VS:
4373 		ctrl->aen_result = result;
4374 		break;
4375 	default:
4376 		break;
4377 	}
4378 
4379 	if (requeue)
4380 		queue_work(nvme_wq, &ctrl->async_event_work);
4381 }
4382 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4383 
4384 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4385 		const struct blk_mq_ops *ops, unsigned int cmd_size)
4386 {
4387 	struct queue_limits lim = {};
4388 	int ret;
4389 
4390 	memset(set, 0, sizeof(*set));
4391 	set->ops = ops;
4392 	set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4393 	if (ctrl->ops->flags & NVME_F_FABRICS)
4394 		set->reserved_tags = NVMF_RESERVED_TAGS;
4395 	set->numa_node = ctrl->numa_node;
4396 	set->flags = BLK_MQ_F_NO_SCHED;
4397 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4398 		set->flags |= BLK_MQ_F_BLOCKING;
4399 	set->cmd_size = cmd_size;
4400 	set->driver_data = ctrl;
4401 	set->nr_hw_queues = 1;
4402 	set->timeout = NVME_ADMIN_TIMEOUT;
4403 	ret = blk_mq_alloc_tag_set(set);
4404 	if (ret)
4405 		return ret;
4406 
4407 	ctrl->admin_q = blk_mq_alloc_queue(set, &lim, NULL);
4408 	if (IS_ERR(ctrl->admin_q)) {
4409 		ret = PTR_ERR(ctrl->admin_q);
4410 		goto out_free_tagset;
4411 	}
4412 
4413 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4414 		ctrl->fabrics_q = blk_mq_alloc_queue(set, NULL, NULL);
4415 		if (IS_ERR(ctrl->fabrics_q)) {
4416 			ret = PTR_ERR(ctrl->fabrics_q);
4417 			goto out_cleanup_admin_q;
4418 		}
4419 	}
4420 
4421 	ctrl->admin_tagset = set;
4422 	return 0;
4423 
4424 out_cleanup_admin_q:
4425 	blk_mq_destroy_queue(ctrl->admin_q);
4426 	blk_put_queue(ctrl->admin_q);
4427 out_free_tagset:
4428 	blk_mq_free_tag_set(set);
4429 	ctrl->admin_q = NULL;
4430 	ctrl->fabrics_q = NULL;
4431 	return ret;
4432 }
4433 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4434 
4435 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4436 {
4437 	blk_mq_destroy_queue(ctrl->admin_q);
4438 	blk_put_queue(ctrl->admin_q);
4439 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4440 		blk_mq_destroy_queue(ctrl->fabrics_q);
4441 		blk_put_queue(ctrl->fabrics_q);
4442 	}
4443 	blk_mq_free_tag_set(ctrl->admin_tagset);
4444 }
4445 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4446 
4447 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4448 		const struct blk_mq_ops *ops, unsigned int nr_maps,
4449 		unsigned int cmd_size)
4450 {
4451 	int ret;
4452 
4453 	memset(set, 0, sizeof(*set));
4454 	set->ops = ops;
4455 	set->queue_depth = min_t(unsigned, ctrl->sqsize, BLK_MQ_MAX_DEPTH - 1);
4456 	/*
4457 	 * Some Apple controllers requires tags to be unique across admin and
4458 	 * the (only) I/O queue, so reserve the first 32 tags of the I/O queue.
4459 	 */
4460 	if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
4461 		set->reserved_tags = NVME_AQ_DEPTH;
4462 	else if (ctrl->ops->flags & NVME_F_FABRICS)
4463 		set->reserved_tags = NVMF_RESERVED_TAGS;
4464 	set->numa_node = ctrl->numa_node;
4465 	set->flags = BLK_MQ_F_SHOULD_MERGE;
4466 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4467 		set->flags |= BLK_MQ_F_BLOCKING;
4468 	set->cmd_size = cmd_size,
4469 	set->driver_data = ctrl;
4470 	set->nr_hw_queues = ctrl->queue_count - 1;
4471 	set->timeout = NVME_IO_TIMEOUT;
4472 	set->nr_maps = nr_maps;
4473 	ret = blk_mq_alloc_tag_set(set);
4474 	if (ret)
4475 		return ret;
4476 
4477 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4478 		ctrl->connect_q = blk_mq_alloc_queue(set, NULL, NULL);
4479         	if (IS_ERR(ctrl->connect_q)) {
4480 			ret = PTR_ERR(ctrl->connect_q);
4481 			goto out_free_tag_set;
4482 		}
4483 		blk_queue_flag_set(QUEUE_FLAG_SKIP_TAGSET_QUIESCE,
4484 				   ctrl->connect_q);
4485 	}
4486 
4487 	ctrl->tagset = set;
4488 	return 0;
4489 
4490 out_free_tag_set:
4491 	blk_mq_free_tag_set(set);
4492 	ctrl->connect_q = NULL;
4493 	return ret;
4494 }
4495 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
4496 
4497 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
4498 {
4499 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4500 		blk_mq_destroy_queue(ctrl->connect_q);
4501 		blk_put_queue(ctrl->connect_q);
4502 	}
4503 	blk_mq_free_tag_set(ctrl->tagset);
4504 }
4505 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
4506 
4507 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4508 {
4509 	nvme_mpath_stop(ctrl);
4510 	nvme_auth_stop(ctrl);
4511 	nvme_stop_keep_alive(ctrl);
4512 	nvme_stop_failfast_work(ctrl);
4513 	flush_work(&ctrl->async_event_work);
4514 	cancel_work_sync(&ctrl->fw_act_work);
4515 	if (ctrl->ops->stop_ctrl)
4516 		ctrl->ops->stop_ctrl(ctrl);
4517 }
4518 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4519 
4520 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4521 {
4522 	nvme_enable_aen(ctrl);
4523 
4524 	/*
4525 	 * persistent discovery controllers need to send indication to userspace
4526 	 * to re-read the discovery log page to learn about possible changes
4527 	 * that were missed. We identify persistent discovery controllers by
4528 	 * checking that they started once before, hence are reconnecting back.
4529 	 */
4530 	if (test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
4531 	    nvme_discovery_ctrl(ctrl))
4532 		nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
4533 
4534 	if (ctrl->queue_count > 1) {
4535 		nvme_queue_scan(ctrl);
4536 		nvme_unquiesce_io_queues(ctrl);
4537 		nvme_mpath_update(ctrl);
4538 	}
4539 
4540 	nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4541 	set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags);
4542 }
4543 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4544 
4545 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4546 {
4547 	nvme_hwmon_exit(ctrl);
4548 	nvme_fault_inject_fini(&ctrl->fault_inject);
4549 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4550 	cdev_device_del(&ctrl->cdev, ctrl->device);
4551 	nvme_put_ctrl(ctrl);
4552 }
4553 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4554 
4555 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4556 {
4557 	struct nvme_effects_log	*cel;
4558 	unsigned long i;
4559 
4560 	xa_for_each(&ctrl->cels, i, cel) {
4561 		xa_erase(&ctrl->cels, i);
4562 		kfree(cel);
4563 	}
4564 
4565 	xa_destroy(&ctrl->cels);
4566 }
4567 
4568 static void nvme_free_ctrl(struct device *dev)
4569 {
4570 	struct nvme_ctrl *ctrl =
4571 		container_of(dev, struct nvme_ctrl, ctrl_device);
4572 	struct nvme_subsystem *subsys = ctrl->subsys;
4573 
4574 	if (!subsys || ctrl->instance != subsys->instance)
4575 		ida_free(&nvme_instance_ida, ctrl->instance);
4576 	key_put(ctrl->tls_key);
4577 	nvme_free_cels(ctrl);
4578 	nvme_mpath_uninit(ctrl);
4579 	nvme_auth_stop(ctrl);
4580 	nvme_auth_free(ctrl);
4581 	__free_page(ctrl->discard_page);
4582 	free_opal_dev(ctrl->opal_dev);
4583 
4584 	if (subsys) {
4585 		mutex_lock(&nvme_subsystems_lock);
4586 		list_del(&ctrl->subsys_entry);
4587 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4588 		mutex_unlock(&nvme_subsystems_lock);
4589 	}
4590 
4591 	ctrl->ops->free_ctrl(ctrl);
4592 
4593 	if (subsys)
4594 		nvme_put_subsystem(subsys);
4595 }
4596 
4597 /*
4598  * Initialize a NVMe controller structures.  This needs to be called during
4599  * earliest initialization so that we have the initialized structured around
4600  * during probing.
4601  */
4602 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4603 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
4604 {
4605 	int ret;
4606 
4607 	WRITE_ONCE(ctrl->state, NVME_CTRL_NEW);
4608 	ctrl->passthru_err_log_enabled = false;
4609 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4610 	spin_lock_init(&ctrl->lock);
4611 	mutex_init(&ctrl->scan_lock);
4612 	INIT_LIST_HEAD(&ctrl->namespaces);
4613 	xa_init(&ctrl->cels);
4614 	init_rwsem(&ctrl->namespaces_rwsem);
4615 	ctrl->dev = dev;
4616 	ctrl->ops = ops;
4617 	ctrl->quirks = quirks;
4618 	ctrl->numa_node = NUMA_NO_NODE;
4619 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4620 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4621 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4622 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4623 	init_waitqueue_head(&ctrl->state_wq);
4624 
4625 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4626 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4627 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4628 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4629 	ctrl->ka_last_check_time = jiffies;
4630 
4631 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4632 			PAGE_SIZE);
4633 	ctrl->discard_page = alloc_page(GFP_KERNEL);
4634 	if (!ctrl->discard_page) {
4635 		ret = -ENOMEM;
4636 		goto out;
4637 	}
4638 
4639 	ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4640 	if (ret < 0)
4641 		goto out;
4642 	ctrl->instance = ret;
4643 
4644 	device_initialize(&ctrl->ctrl_device);
4645 	ctrl->device = &ctrl->ctrl_device;
4646 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4647 			ctrl->instance);
4648 	ctrl->device->class = &nvme_class;
4649 	ctrl->device->parent = ctrl->dev;
4650 	if (ops->dev_attr_groups)
4651 		ctrl->device->groups = ops->dev_attr_groups;
4652 	else
4653 		ctrl->device->groups = nvme_dev_attr_groups;
4654 	ctrl->device->release = nvme_free_ctrl;
4655 	dev_set_drvdata(ctrl->device, ctrl);
4656 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4657 	if (ret)
4658 		goto out_release_instance;
4659 
4660 	nvme_get_ctrl(ctrl);
4661 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
4662 	ctrl->cdev.owner = ops->module;
4663 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4664 	if (ret)
4665 		goto out_free_name;
4666 
4667 	/*
4668 	 * Initialize latency tolerance controls.  The sysfs files won't
4669 	 * be visible to userspace unless the device actually supports APST.
4670 	 */
4671 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4672 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4673 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4674 
4675 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4676 	nvme_mpath_init_ctrl(ctrl);
4677 	ret = nvme_auth_init_ctrl(ctrl);
4678 	if (ret)
4679 		goto out_free_cdev;
4680 
4681 	return 0;
4682 out_free_cdev:
4683 	nvme_fault_inject_fini(&ctrl->fault_inject);
4684 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4685 	cdev_device_del(&ctrl->cdev, ctrl->device);
4686 out_free_name:
4687 	nvme_put_ctrl(ctrl);
4688 	kfree_const(ctrl->device->kobj.name);
4689 out_release_instance:
4690 	ida_free(&nvme_instance_ida, ctrl->instance);
4691 out:
4692 	if (ctrl->discard_page)
4693 		__free_page(ctrl->discard_page);
4694 	return ret;
4695 }
4696 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4697 
4698 /* let I/O to all namespaces fail in preparation for surprise removal */
4699 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
4700 {
4701 	struct nvme_ns *ns;
4702 
4703 	down_read(&ctrl->namespaces_rwsem);
4704 	list_for_each_entry(ns, &ctrl->namespaces, list)
4705 		blk_mark_disk_dead(ns->disk);
4706 	up_read(&ctrl->namespaces_rwsem);
4707 }
4708 EXPORT_SYMBOL_GPL(nvme_mark_namespaces_dead);
4709 
4710 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4711 {
4712 	struct nvme_ns *ns;
4713 
4714 	down_read(&ctrl->namespaces_rwsem);
4715 	list_for_each_entry(ns, &ctrl->namespaces, list)
4716 		blk_mq_unfreeze_queue(ns->queue);
4717 	up_read(&ctrl->namespaces_rwsem);
4718 	clear_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4719 }
4720 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4721 
4722 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4723 {
4724 	struct nvme_ns *ns;
4725 
4726 	down_read(&ctrl->namespaces_rwsem);
4727 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4728 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4729 		if (timeout <= 0)
4730 			break;
4731 	}
4732 	up_read(&ctrl->namespaces_rwsem);
4733 	return timeout;
4734 }
4735 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4736 
4737 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4738 {
4739 	struct nvme_ns *ns;
4740 
4741 	down_read(&ctrl->namespaces_rwsem);
4742 	list_for_each_entry(ns, &ctrl->namespaces, list)
4743 		blk_mq_freeze_queue_wait(ns->queue);
4744 	up_read(&ctrl->namespaces_rwsem);
4745 }
4746 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4747 
4748 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4749 {
4750 	struct nvme_ns *ns;
4751 
4752 	set_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4753 	down_read(&ctrl->namespaces_rwsem);
4754 	list_for_each_entry(ns, &ctrl->namespaces, list)
4755 		blk_freeze_queue_start(ns->queue);
4756 	up_read(&ctrl->namespaces_rwsem);
4757 }
4758 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4759 
4760 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl)
4761 {
4762 	if (!ctrl->tagset)
4763 		return;
4764 	if (!test_and_set_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4765 		blk_mq_quiesce_tagset(ctrl->tagset);
4766 	else
4767 		blk_mq_wait_quiesce_done(ctrl->tagset);
4768 }
4769 EXPORT_SYMBOL_GPL(nvme_quiesce_io_queues);
4770 
4771 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl)
4772 {
4773 	if (!ctrl->tagset)
4774 		return;
4775 	if (test_and_clear_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4776 		blk_mq_unquiesce_tagset(ctrl->tagset);
4777 }
4778 EXPORT_SYMBOL_GPL(nvme_unquiesce_io_queues);
4779 
4780 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl)
4781 {
4782 	if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4783 		blk_mq_quiesce_queue(ctrl->admin_q);
4784 	else
4785 		blk_mq_wait_quiesce_done(ctrl->admin_q->tag_set);
4786 }
4787 EXPORT_SYMBOL_GPL(nvme_quiesce_admin_queue);
4788 
4789 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl)
4790 {
4791 	if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4792 		blk_mq_unquiesce_queue(ctrl->admin_q);
4793 }
4794 EXPORT_SYMBOL_GPL(nvme_unquiesce_admin_queue);
4795 
4796 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4797 {
4798 	struct nvme_ns *ns;
4799 
4800 	down_read(&ctrl->namespaces_rwsem);
4801 	list_for_each_entry(ns, &ctrl->namespaces, list)
4802 		blk_sync_queue(ns->queue);
4803 	up_read(&ctrl->namespaces_rwsem);
4804 }
4805 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4806 
4807 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4808 {
4809 	nvme_sync_io_queues(ctrl);
4810 	if (ctrl->admin_q)
4811 		blk_sync_queue(ctrl->admin_q);
4812 }
4813 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4814 
4815 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4816 {
4817 	if (file->f_op != &nvme_dev_fops)
4818 		return NULL;
4819 	return file->private_data;
4820 }
4821 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4822 
4823 /*
4824  * Check we didn't inadvertently grow the command structure sizes:
4825  */
4826 static inline void _nvme_check_size(void)
4827 {
4828 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4829 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4830 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4831 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4832 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4833 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4834 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4835 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4836 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4837 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4838 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4839 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4840 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4841 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
4842 			NVME_IDENTIFY_DATA_SIZE);
4843 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4844 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
4845 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4846 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4847 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4848 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4849 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4850 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4851 	BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
4852 }
4853 
4854 
4855 static int __init nvme_core_init(void)
4856 {
4857 	int result = -ENOMEM;
4858 
4859 	_nvme_check_size();
4860 
4861 	nvme_wq = alloc_workqueue("nvme-wq",
4862 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4863 	if (!nvme_wq)
4864 		goto out;
4865 
4866 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4867 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4868 	if (!nvme_reset_wq)
4869 		goto destroy_wq;
4870 
4871 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4872 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4873 	if (!nvme_delete_wq)
4874 		goto destroy_reset_wq;
4875 
4876 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4877 			NVME_MINORS, "nvme");
4878 	if (result < 0)
4879 		goto destroy_delete_wq;
4880 
4881 	result = class_register(&nvme_class);
4882 	if (result)
4883 		goto unregister_chrdev;
4884 
4885 	result = class_register(&nvme_subsys_class);
4886 	if (result)
4887 		goto destroy_class;
4888 
4889 	result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4890 				     "nvme-generic");
4891 	if (result < 0)
4892 		goto destroy_subsys_class;
4893 
4894 	result = class_register(&nvme_ns_chr_class);
4895 	if (result)
4896 		goto unregister_generic_ns;
4897 
4898 	result = nvme_init_auth();
4899 	if (result)
4900 		goto destroy_ns_chr;
4901 	return 0;
4902 
4903 destroy_ns_chr:
4904 	class_unregister(&nvme_ns_chr_class);
4905 unregister_generic_ns:
4906 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4907 destroy_subsys_class:
4908 	class_unregister(&nvme_subsys_class);
4909 destroy_class:
4910 	class_unregister(&nvme_class);
4911 unregister_chrdev:
4912 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4913 destroy_delete_wq:
4914 	destroy_workqueue(nvme_delete_wq);
4915 destroy_reset_wq:
4916 	destroy_workqueue(nvme_reset_wq);
4917 destroy_wq:
4918 	destroy_workqueue(nvme_wq);
4919 out:
4920 	return result;
4921 }
4922 
4923 static void __exit nvme_core_exit(void)
4924 {
4925 	nvme_exit_auth();
4926 	class_unregister(&nvme_ns_chr_class);
4927 	class_unregister(&nvme_subsys_class);
4928 	class_unregister(&nvme_class);
4929 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4930 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4931 	destroy_workqueue(nvme_delete_wq);
4932 	destroy_workqueue(nvme_reset_wq);
4933 	destroy_workqueue(nvme_wq);
4934 	ida_destroy(&nvme_ns_chr_minor_ida);
4935 	ida_destroy(&nvme_instance_ida);
4936 }
4937 
4938 MODULE_LICENSE("GPL");
4939 MODULE_VERSION("1.0");
4940 MODULE_DESCRIPTION("NVMe host core framework");
4941 module_init(nvme_core_init);
4942 module_exit(nvme_core_exit);
4943