xref: /openbmc/linux/drivers/nvme/host/core.c (revision 8bdc2a19)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * NVM Express device driver
4  * Copyright (c) 2011-2014, Intel Corporation.
5  */
6 
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/blk-integrity.h>
10 #include <linux/compat.h>
11 #include <linux/delay.h>
12 #include <linux/errno.h>
13 #include <linux/hdreg.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
19 #include <linux/pr.h>
20 #include <linux/ptrace.h>
21 #include <linux/nvme_ioctl.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
24 
25 #include "nvme.h"
26 #include "fabrics.h"
27 
28 #define CREATE_TRACE_POINTS
29 #include "trace.h"
30 
31 #define NVME_MINORS		(1U << MINORBITS)
32 
33 unsigned int admin_timeout = 60;
34 module_param(admin_timeout, uint, 0644);
35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
36 EXPORT_SYMBOL_GPL(admin_timeout);
37 
38 unsigned int nvme_io_timeout = 30;
39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
42 
43 static unsigned char shutdown_timeout = 5;
44 module_param(shutdown_timeout, byte, 0644);
45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
46 
47 static u8 nvme_max_retries = 5;
48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
50 
51 static unsigned long default_ps_max_latency_us = 100000;
52 module_param(default_ps_max_latency_us, ulong, 0644);
53 MODULE_PARM_DESC(default_ps_max_latency_us,
54 		 "max power saving latency for new devices; use PM QOS to change per device");
55 
56 static bool force_apst;
57 module_param(force_apst, bool, 0644);
58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
59 
60 static unsigned long apst_primary_timeout_ms = 100;
61 module_param(apst_primary_timeout_ms, ulong, 0644);
62 MODULE_PARM_DESC(apst_primary_timeout_ms,
63 	"primary APST timeout in ms");
64 
65 static unsigned long apst_secondary_timeout_ms = 2000;
66 module_param(apst_secondary_timeout_ms, ulong, 0644);
67 MODULE_PARM_DESC(apst_secondary_timeout_ms,
68 	"secondary APST timeout in ms");
69 
70 static unsigned long apst_primary_latency_tol_us = 15000;
71 module_param(apst_primary_latency_tol_us, ulong, 0644);
72 MODULE_PARM_DESC(apst_primary_latency_tol_us,
73 	"primary APST latency tolerance in us");
74 
75 static unsigned long apst_secondary_latency_tol_us = 100000;
76 module_param(apst_secondary_latency_tol_us, ulong, 0644);
77 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
78 	"secondary APST latency tolerance in us");
79 
80 /*
81  * nvme_wq - hosts nvme related works that are not reset or delete
82  * nvme_reset_wq - hosts nvme reset works
83  * nvme_delete_wq - hosts nvme delete works
84  *
85  * nvme_wq will host works such as scan, aen handling, fw activation,
86  * keep-alive, periodic reconnects etc. nvme_reset_wq
87  * runs reset works which also flush works hosted on nvme_wq for
88  * serialization purposes. nvme_delete_wq host controller deletion
89  * works which flush reset works for serialization.
90  */
91 struct workqueue_struct *nvme_wq;
92 EXPORT_SYMBOL_GPL(nvme_wq);
93 
94 struct workqueue_struct *nvme_reset_wq;
95 EXPORT_SYMBOL_GPL(nvme_reset_wq);
96 
97 struct workqueue_struct *nvme_delete_wq;
98 EXPORT_SYMBOL_GPL(nvme_delete_wq);
99 
100 static LIST_HEAD(nvme_subsystems);
101 static DEFINE_MUTEX(nvme_subsystems_lock);
102 
103 static DEFINE_IDA(nvme_instance_ida);
104 static dev_t nvme_ctrl_base_chr_devt;
105 static struct class *nvme_class;
106 static struct class *nvme_subsys_class;
107 
108 static DEFINE_IDA(nvme_ns_chr_minor_ida);
109 static dev_t nvme_ns_chr_devt;
110 static struct class *nvme_ns_chr_class;
111 
112 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
113 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
114 					   unsigned nsid);
115 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
116 				   struct nvme_command *cmd);
117 
118 void nvme_queue_scan(struct nvme_ctrl *ctrl)
119 {
120 	/*
121 	 * Only new queue scan work when admin and IO queues are both alive
122 	 */
123 	if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
124 		queue_work(nvme_wq, &ctrl->scan_work);
125 }
126 
127 /*
128  * Use this function to proceed with scheduling reset_work for a controller
129  * that had previously been set to the resetting state. This is intended for
130  * code paths that can't be interrupted by other reset attempts. A hot removal
131  * may prevent this from succeeding.
132  */
133 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
134 {
135 	if (ctrl->state != NVME_CTRL_RESETTING)
136 		return -EBUSY;
137 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
138 		return -EBUSY;
139 	return 0;
140 }
141 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
142 
143 static void nvme_failfast_work(struct work_struct *work)
144 {
145 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
146 			struct nvme_ctrl, failfast_work);
147 
148 	if (ctrl->state != NVME_CTRL_CONNECTING)
149 		return;
150 
151 	set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
152 	dev_info(ctrl->device, "failfast expired\n");
153 	nvme_kick_requeue_lists(ctrl);
154 }
155 
156 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
157 {
158 	if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
159 		return;
160 
161 	schedule_delayed_work(&ctrl->failfast_work,
162 			      ctrl->opts->fast_io_fail_tmo * HZ);
163 }
164 
165 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
166 {
167 	if (!ctrl->opts)
168 		return;
169 
170 	cancel_delayed_work_sync(&ctrl->failfast_work);
171 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
172 }
173 
174 
175 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
176 {
177 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
178 		return -EBUSY;
179 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
180 		return -EBUSY;
181 	return 0;
182 }
183 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
184 
185 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
186 {
187 	int ret;
188 
189 	ret = nvme_reset_ctrl(ctrl);
190 	if (!ret) {
191 		flush_work(&ctrl->reset_work);
192 		if (ctrl->state != NVME_CTRL_LIVE)
193 			ret = -ENETRESET;
194 	}
195 
196 	return ret;
197 }
198 
199 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
200 {
201 	dev_info(ctrl->device,
202 		 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
203 
204 	flush_work(&ctrl->reset_work);
205 	nvme_stop_ctrl(ctrl);
206 	nvme_remove_namespaces(ctrl);
207 	ctrl->ops->delete_ctrl(ctrl);
208 	nvme_uninit_ctrl(ctrl);
209 }
210 
211 static void nvme_delete_ctrl_work(struct work_struct *work)
212 {
213 	struct nvme_ctrl *ctrl =
214 		container_of(work, struct nvme_ctrl, delete_work);
215 
216 	nvme_do_delete_ctrl(ctrl);
217 }
218 
219 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
220 {
221 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
222 		return -EBUSY;
223 	if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
224 		return -EBUSY;
225 	return 0;
226 }
227 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
228 
229 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
230 {
231 	/*
232 	 * Keep a reference until nvme_do_delete_ctrl() complete,
233 	 * since ->delete_ctrl can free the controller.
234 	 */
235 	nvme_get_ctrl(ctrl);
236 	if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
237 		nvme_do_delete_ctrl(ctrl);
238 	nvme_put_ctrl(ctrl);
239 }
240 
241 static blk_status_t nvme_error_status(u16 status)
242 {
243 	switch (status & 0x7ff) {
244 	case NVME_SC_SUCCESS:
245 		return BLK_STS_OK;
246 	case NVME_SC_CAP_EXCEEDED:
247 		return BLK_STS_NOSPC;
248 	case NVME_SC_LBA_RANGE:
249 	case NVME_SC_CMD_INTERRUPTED:
250 	case NVME_SC_NS_NOT_READY:
251 		return BLK_STS_TARGET;
252 	case NVME_SC_BAD_ATTRIBUTES:
253 	case NVME_SC_ONCS_NOT_SUPPORTED:
254 	case NVME_SC_INVALID_OPCODE:
255 	case NVME_SC_INVALID_FIELD:
256 	case NVME_SC_INVALID_NS:
257 		return BLK_STS_NOTSUPP;
258 	case NVME_SC_WRITE_FAULT:
259 	case NVME_SC_READ_ERROR:
260 	case NVME_SC_UNWRITTEN_BLOCK:
261 	case NVME_SC_ACCESS_DENIED:
262 	case NVME_SC_READ_ONLY:
263 	case NVME_SC_COMPARE_FAILED:
264 		return BLK_STS_MEDIUM;
265 	case NVME_SC_GUARD_CHECK:
266 	case NVME_SC_APPTAG_CHECK:
267 	case NVME_SC_REFTAG_CHECK:
268 	case NVME_SC_INVALID_PI:
269 		return BLK_STS_PROTECTION;
270 	case NVME_SC_RESERVATION_CONFLICT:
271 		return BLK_STS_NEXUS;
272 	case NVME_SC_HOST_PATH_ERROR:
273 		return BLK_STS_TRANSPORT;
274 	case NVME_SC_ZONE_TOO_MANY_ACTIVE:
275 		return BLK_STS_ZONE_ACTIVE_RESOURCE;
276 	case NVME_SC_ZONE_TOO_MANY_OPEN:
277 		return BLK_STS_ZONE_OPEN_RESOURCE;
278 	default:
279 		return BLK_STS_IOERR;
280 	}
281 }
282 
283 static void nvme_retry_req(struct request *req)
284 {
285 	unsigned long delay = 0;
286 	u16 crd;
287 
288 	/* The mask and shift result must be <= 3 */
289 	crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
290 	if (crd)
291 		delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
292 
293 	nvme_req(req)->retries++;
294 	blk_mq_requeue_request(req, false);
295 	blk_mq_delay_kick_requeue_list(req->q, delay);
296 }
297 
298 static void nvme_log_error(struct request *req)
299 {
300 	struct nvme_ns *ns = req->q->queuedata;
301 	struct nvme_request *nr = nvme_req(req);
302 
303 	if (ns) {
304 		pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %llu blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
305 		       ns->disk ? ns->disk->disk_name : "?",
306 		       nvme_get_opcode_str(nr->cmd->common.opcode),
307 		       nr->cmd->common.opcode,
308 		       (unsigned long long)nvme_sect_to_lba(ns, blk_rq_pos(req)),
309 		       (unsigned long long)blk_rq_bytes(req) >> ns->lba_shift,
310 		       nvme_get_error_status_str(nr->status),
311 		       nr->status >> 8 & 7,	/* Status Code Type */
312 		       nr->status & 0xff,	/* Status Code */
313 		       nr->status & NVME_SC_MORE ? "MORE " : "",
314 		       nr->status & NVME_SC_DNR  ? "DNR "  : "");
315 		return;
316 	}
317 
318 	pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
319 			   dev_name(nr->ctrl->device),
320 			   nvme_get_admin_opcode_str(nr->cmd->common.opcode),
321 			   nr->cmd->common.opcode,
322 			   nvme_get_error_status_str(nr->status),
323 			   nr->status >> 8 & 7,	/* Status Code Type */
324 			   nr->status & 0xff,	/* Status Code */
325 			   nr->status & NVME_SC_MORE ? "MORE " : "",
326 			   nr->status & NVME_SC_DNR  ? "DNR "  : "");
327 }
328 
329 enum nvme_disposition {
330 	COMPLETE,
331 	RETRY,
332 	FAILOVER,
333 };
334 
335 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
336 {
337 	if (likely(nvme_req(req)->status == 0))
338 		return COMPLETE;
339 
340 	if (blk_noretry_request(req) ||
341 	    (nvme_req(req)->status & NVME_SC_DNR) ||
342 	    nvme_req(req)->retries >= nvme_max_retries)
343 		return COMPLETE;
344 
345 	if (req->cmd_flags & REQ_NVME_MPATH) {
346 		if (nvme_is_path_error(nvme_req(req)->status) ||
347 		    blk_queue_dying(req->q))
348 			return FAILOVER;
349 	} else {
350 		if (blk_queue_dying(req->q))
351 			return COMPLETE;
352 	}
353 
354 	return RETRY;
355 }
356 
357 static inline void nvme_end_req_zoned(struct request *req)
358 {
359 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
360 	    req_op(req) == REQ_OP_ZONE_APPEND)
361 		req->__sector = nvme_lba_to_sect(req->q->queuedata,
362 			le64_to_cpu(nvme_req(req)->result.u64));
363 }
364 
365 static inline void nvme_end_req(struct request *req)
366 {
367 	blk_status_t status = nvme_error_status(nvme_req(req)->status);
368 
369 	if (unlikely(nvme_req(req)->status && !(req->rq_flags & RQF_QUIET)))
370 		nvme_log_error(req);
371 	nvme_end_req_zoned(req);
372 	nvme_trace_bio_complete(req);
373 	blk_mq_end_request(req, status);
374 }
375 
376 void nvme_complete_rq(struct request *req)
377 {
378 	trace_nvme_complete_rq(req);
379 	nvme_cleanup_cmd(req);
380 
381 	if (nvme_req(req)->ctrl->kas)
382 		nvme_req(req)->ctrl->comp_seen = true;
383 
384 	switch (nvme_decide_disposition(req)) {
385 	case COMPLETE:
386 		nvme_end_req(req);
387 		return;
388 	case RETRY:
389 		nvme_retry_req(req);
390 		return;
391 	case FAILOVER:
392 		nvme_failover_req(req);
393 		return;
394 	}
395 }
396 EXPORT_SYMBOL_GPL(nvme_complete_rq);
397 
398 void nvme_complete_batch_req(struct request *req)
399 {
400 	trace_nvme_complete_rq(req);
401 	nvme_cleanup_cmd(req);
402 	nvme_end_req_zoned(req);
403 }
404 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
405 
406 /*
407  * Called to unwind from ->queue_rq on a failed command submission so that the
408  * multipathing code gets called to potentially failover to another path.
409  * The caller needs to unwind all transport specific resource allocations and
410  * must return propagate the return value.
411  */
412 blk_status_t nvme_host_path_error(struct request *req)
413 {
414 	nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
415 	blk_mq_set_request_complete(req);
416 	nvme_complete_rq(req);
417 	return BLK_STS_OK;
418 }
419 EXPORT_SYMBOL_GPL(nvme_host_path_error);
420 
421 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
422 {
423 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
424 				"Cancelling I/O %d", req->tag);
425 
426 	/* don't abort one completed request */
427 	if (blk_mq_request_completed(req))
428 		return true;
429 
430 	nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
431 	nvme_req(req)->flags |= NVME_REQ_CANCELLED;
432 	blk_mq_complete_request(req);
433 	return true;
434 }
435 EXPORT_SYMBOL_GPL(nvme_cancel_request);
436 
437 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
438 {
439 	if (ctrl->tagset) {
440 		blk_mq_tagset_busy_iter(ctrl->tagset,
441 				nvme_cancel_request, ctrl);
442 		blk_mq_tagset_wait_completed_request(ctrl->tagset);
443 	}
444 }
445 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
446 
447 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
448 {
449 	if (ctrl->admin_tagset) {
450 		blk_mq_tagset_busy_iter(ctrl->admin_tagset,
451 				nvme_cancel_request, ctrl);
452 		blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
453 	}
454 }
455 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
456 
457 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
458 		enum nvme_ctrl_state new_state)
459 {
460 	enum nvme_ctrl_state old_state;
461 	unsigned long flags;
462 	bool changed = false;
463 
464 	spin_lock_irqsave(&ctrl->lock, flags);
465 
466 	old_state = ctrl->state;
467 	switch (new_state) {
468 	case NVME_CTRL_LIVE:
469 		switch (old_state) {
470 		case NVME_CTRL_NEW:
471 		case NVME_CTRL_RESETTING:
472 		case NVME_CTRL_CONNECTING:
473 			changed = true;
474 			fallthrough;
475 		default:
476 			break;
477 		}
478 		break;
479 	case NVME_CTRL_RESETTING:
480 		switch (old_state) {
481 		case NVME_CTRL_NEW:
482 		case NVME_CTRL_LIVE:
483 			changed = true;
484 			fallthrough;
485 		default:
486 			break;
487 		}
488 		break;
489 	case NVME_CTRL_CONNECTING:
490 		switch (old_state) {
491 		case NVME_CTRL_NEW:
492 		case NVME_CTRL_RESETTING:
493 			changed = true;
494 			fallthrough;
495 		default:
496 			break;
497 		}
498 		break;
499 	case NVME_CTRL_DELETING:
500 		switch (old_state) {
501 		case NVME_CTRL_LIVE:
502 		case NVME_CTRL_RESETTING:
503 		case NVME_CTRL_CONNECTING:
504 			changed = true;
505 			fallthrough;
506 		default:
507 			break;
508 		}
509 		break;
510 	case NVME_CTRL_DELETING_NOIO:
511 		switch (old_state) {
512 		case NVME_CTRL_DELETING:
513 		case NVME_CTRL_DEAD:
514 			changed = true;
515 			fallthrough;
516 		default:
517 			break;
518 		}
519 		break;
520 	case NVME_CTRL_DEAD:
521 		switch (old_state) {
522 		case NVME_CTRL_DELETING:
523 			changed = true;
524 			fallthrough;
525 		default:
526 			break;
527 		}
528 		break;
529 	default:
530 		break;
531 	}
532 
533 	if (changed) {
534 		ctrl->state = new_state;
535 		wake_up_all(&ctrl->state_wq);
536 	}
537 
538 	spin_unlock_irqrestore(&ctrl->lock, flags);
539 	if (!changed)
540 		return false;
541 
542 	if (ctrl->state == NVME_CTRL_LIVE) {
543 		if (old_state == NVME_CTRL_CONNECTING)
544 			nvme_stop_failfast_work(ctrl);
545 		nvme_kick_requeue_lists(ctrl);
546 	} else if (ctrl->state == NVME_CTRL_CONNECTING &&
547 		old_state == NVME_CTRL_RESETTING) {
548 		nvme_start_failfast_work(ctrl);
549 	}
550 	return changed;
551 }
552 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
553 
554 /*
555  * Returns true for sink states that can't ever transition back to live.
556  */
557 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
558 {
559 	switch (ctrl->state) {
560 	case NVME_CTRL_NEW:
561 	case NVME_CTRL_LIVE:
562 	case NVME_CTRL_RESETTING:
563 	case NVME_CTRL_CONNECTING:
564 		return false;
565 	case NVME_CTRL_DELETING:
566 	case NVME_CTRL_DELETING_NOIO:
567 	case NVME_CTRL_DEAD:
568 		return true;
569 	default:
570 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
571 		return true;
572 	}
573 }
574 
575 /*
576  * Waits for the controller state to be resetting, or returns false if it is
577  * not possible to ever transition to that state.
578  */
579 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
580 {
581 	wait_event(ctrl->state_wq,
582 		   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
583 		   nvme_state_terminal(ctrl));
584 	return ctrl->state == NVME_CTRL_RESETTING;
585 }
586 EXPORT_SYMBOL_GPL(nvme_wait_reset);
587 
588 static void nvme_free_ns_head(struct kref *ref)
589 {
590 	struct nvme_ns_head *head =
591 		container_of(ref, struct nvme_ns_head, ref);
592 
593 	nvme_mpath_remove_disk(head);
594 	ida_free(&head->subsys->ns_ida, head->instance);
595 	cleanup_srcu_struct(&head->srcu);
596 	nvme_put_subsystem(head->subsys);
597 	kfree(head);
598 }
599 
600 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
601 {
602 	return kref_get_unless_zero(&head->ref);
603 }
604 
605 void nvme_put_ns_head(struct nvme_ns_head *head)
606 {
607 	kref_put(&head->ref, nvme_free_ns_head);
608 }
609 
610 static void nvme_free_ns(struct kref *kref)
611 {
612 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
613 
614 	put_disk(ns->disk);
615 	nvme_put_ns_head(ns->head);
616 	nvme_put_ctrl(ns->ctrl);
617 	kfree(ns);
618 }
619 
620 static inline bool nvme_get_ns(struct nvme_ns *ns)
621 {
622 	return kref_get_unless_zero(&ns->kref);
623 }
624 
625 void nvme_put_ns(struct nvme_ns *ns)
626 {
627 	kref_put(&ns->kref, nvme_free_ns);
628 }
629 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
630 
631 static inline void nvme_clear_nvme_request(struct request *req)
632 {
633 	nvme_req(req)->status = 0;
634 	nvme_req(req)->retries = 0;
635 	nvme_req(req)->flags = 0;
636 	req->rq_flags |= RQF_DONTPREP;
637 }
638 
639 /* initialize a passthrough request */
640 void nvme_init_request(struct request *req, struct nvme_command *cmd)
641 {
642 	if (req->q->queuedata)
643 		req->timeout = NVME_IO_TIMEOUT;
644 	else /* no queuedata implies admin queue */
645 		req->timeout = NVME_ADMIN_TIMEOUT;
646 
647 	/* passthru commands should let the driver set the SGL flags */
648 	cmd->common.flags &= ~NVME_CMD_SGL_ALL;
649 
650 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
651 	if (req->mq_hctx->type == HCTX_TYPE_POLL)
652 		req->cmd_flags |= REQ_POLLED;
653 	nvme_clear_nvme_request(req);
654 	memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
655 }
656 EXPORT_SYMBOL_GPL(nvme_init_request);
657 
658 /*
659  * For something we're not in a state to send to the device the default action
660  * is to busy it and retry it after the controller state is recovered.  However,
661  * if the controller is deleting or if anything is marked for failfast or
662  * nvme multipath it is immediately failed.
663  *
664  * Note: commands used to initialize the controller will be marked for failfast.
665  * Note: nvme cli/ioctl commands are marked for failfast.
666  */
667 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
668 		struct request *rq)
669 {
670 	if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
671 	    ctrl->state != NVME_CTRL_DELETING &&
672 	    ctrl->state != NVME_CTRL_DEAD &&
673 	    !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
674 	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
675 		return BLK_STS_RESOURCE;
676 	return nvme_host_path_error(rq);
677 }
678 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
679 
680 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
681 		bool queue_live)
682 {
683 	struct nvme_request *req = nvme_req(rq);
684 
685 	/*
686 	 * currently we have a problem sending passthru commands
687 	 * on the admin_q if the controller is not LIVE because we can't
688 	 * make sure that they are going out after the admin connect,
689 	 * controller enable and/or other commands in the initialization
690 	 * sequence. until the controller will be LIVE, fail with
691 	 * BLK_STS_RESOURCE so that they will be rescheduled.
692 	 */
693 	if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
694 		return false;
695 
696 	if (ctrl->ops->flags & NVME_F_FABRICS) {
697 		/*
698 		 * Only allow commands on a live queue, except for the connect
699 		 * command, which is require to set the queue live in the
700 		 * appropinquate states.
701 		 */
702 		switch (ctrl->state) {
703 		case NVME_CTRL_CONNECTING:
704 			if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
705 			    req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
706 				return true;
707 			break;
708 		default:
709 			break;
710 		case NVME_CTRL_DEAD:
711 			return false;
712 		}
713 	}
714 
715 	return queue_live;
716 }
717 EXPORT_SYMBOL_GPL(__nvme_check_ready);
718 
719 static inline void nvme_setup_flush(struct nvme_ns *ns,
720 		struct nvme_command *cmnd)
721 {
722 	memset(cmnd, 0, sizeof(*cmnd));
723 	cmnd->common.opcode = nvme_cmd_flush;
724 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
725 }
726 
727 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
728 		struct nvme_command *cmnd)
729 {
730 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
731 	struct nvme_dsm_range *range;
732 	struct bio *bio;
733 
734 	/*
735 	 * Some devices do not consider the DSM 'Number of Ranges' field when
736 	 * determining how much data to DMA. Always allocate memory for maximum
737 	 * number of segments to prevent device reading beyond end of buffer.
738 	 */
739 	static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
740 
741 	range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
742 	if (!range) {
743 		/*
744 		 * If we fail allocation our range, fallback to the controller
745 		 * discard page. If that's also busy, it's safe to return
746 		 * busy, as we know we can make progress once that's freed.
747 		 */
748 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
749 			return BLK_STS_RESOURCE;
750 
751 		range = page_address(ns->ctrl->discard_page);
752 	}
753 
754 	__rq_for_each_bio(bio, req) {
755 		u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
756 		u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
757 
758 		if (n < segments) {
759 			range[n].cattr = cpu_to_le32(0);
760 			range[n].nlb = cpu_to_le32(nlb);
761 			range[n].slba = cpu_to_le64(slba);
762 		}
763 		n++;
764 	}
765 
766 	if (WARN_ON_ONCE(n != segments)) {
767 		if (virt_to_page(range) == ns->ctrl->discard_page)
768 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
769 		else
770 			kfree(range);
771 		return BLK_STS_IOERR;
772 	}
773 
774 	memset(cmnd, 0, sizeof(*cmnd));
775 	cmnd->dsm.opcode = nvme_cmd_dsm;
776 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
777 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
778 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
779 
780 	req->special_vec.bv_page = virt_to_page(range);
781 	req->special_vec.bv_offset = offset_in_page(range);
782 	req->special_vec.bv_len = alloc_size;
783 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
784 
785 	return BLK_STS_OK;
786 }
787 
788 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
789 			      struct request *req)
790 {
791 	u32 upper, lower;
792 	u64 ref48;
793 
794 	/* both rw and write zeroes share the same reftag format */
795 	switch (ns->guard_type) {
796 	case NVME_NVM_NS_16B_GUARD:
797 		cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
798 		break;
799 	case NVME_NVM_NS_64B_GUARD:
800 		ref48 = ext_pi_ref_tag(req);
801 		lower = lower_32_bits(ref48);
802 		upper = upper_32_bits(ref48);
803 
804 		cmnd->rw.reftag = cpu_to_le32(lower);
805 		cmnd->rw.cdw3 = cpu_to_le32(upper);
806 		break;
807 	default:
808 		break;
809 	}
810 }
811 
812 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
813 		struct request *req, struct nvme_command *cmnd)
814 {
815 	memset(cmnd, 0, sizeof(*cmnd));
816 
817 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
818 		return nvme_setup_discard(ns, req, cmnd);
819 
820 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
821 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
822 	cmnd->write_zeroes.slba =
823 		cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
824 	cmnd->write_zeroes.length =
825 		cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
826 
827 	if (nvme_ns_has_pi(ns)) {
828 		cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
829 
830 		switch (ns->pi_type) {
831 		case NVME_NS_DPS_PI_TYPE1:
832 		case NVME_NS_DPS_PI_TYPE2:
833 			nvme_set_ref_tag(ns, cmnd, req);
834 			break;
835 		}
836 	}
837 
838 	return BLK_STS_OK;
839 }
840 
841 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
842 		struct request *req, struct nvme_command *cmnd,
843 		enum nvme_opcode op)
844 {
845 	u16 control = 0;
846 	u32 dsmgmt = 0;
847 
848 	if (req->cmd_flags & REQ_FUA)
849 		control |= NVME_RW_FUA;
850 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
851 		control |= NVME_RW_LR;
852 
853 	if (req->cmd_flags & REQ_RAHEAD)
854 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
855 
856 	cmnd->rw.opcode = op;
857 	cmnd->rw.flags = 0;
858 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
859 	cmnd->rw.cdw2 = 0;
860 	cmnd->rw.cdw3 = 0;
861 	cmnd->rw.metadata = 0;
862 	cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
863 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
864 	cmnd->rw.reftag = 0;
865 	cmnd->rw.apptag = 0;
866 	cmnd->rw.appmask = 0;
867 
868 	if (ns->ms) {
869 		/*
870 		 * If formated with metadata, the block layer always provides a
871 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
872 		 * we enable the PRACT bit for protection information or set the
873 		 * namespace capacity to zero to prevent any I/O.
874 		 */
875 		if (!blk_integrity_rq(req)) {
876 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
877 				return BLK_STS_NOTSUPP;
878 			control |= NVME_RW_PRINFO_PRACT;
879 		}
880 
881 		switch (ns->pi_type) {
882 		case NVME_NS_DPS_PI_TYPE3:
883 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
884 			break;
885 		case NVME_NS_DPS_PI_TYPE1:
886 		case NVME_NS_DPS_PI_TYPE2:
887 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
888 					NVME_RW_PRINFO_PRCHK_REF;
889 			if (op == nvme_cmd_zone_append)
890 				control |= NVME_RW_APPEND_PIREMAP;
891 			nvme_set_ref_tag(ns, cmnd, req);
892 			break;
893 		}
894 	}
895 
896 	cmnd->rw.control = cpu_to_le16(control);
897 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
898 	return 0;
899 }
900 
901 void nvme_cleanup_cmd(struct request *req)
902 {
903 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
904 		struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
905 
906 		if (req->special_vec.bv_page == ctrl->discard_page)
907 			clear_bit_unlock(0, &ctrl->discard_page_busy);
908 		else
909 			kfree(bvec_virt(&req->special_vec));
910 	}
911 }
912 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
913 
914 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
915 {
916 	struct nvme_command *cmd = nvme_req(req)->cmd;
917 	blk_status_t ret = BLK_STS_OK;
918 
919 	if (!(req->rq_flags & RQF_DONTPREP))
920 		nvme_clear_nvme_request(req);
921 
922 	switch (req_op(req)) {
923 	case REQ_OP_DRV_IN:
924 	case REQ_OP_DRV_OUT:
925 		/* these are setup prior to execution in nvme_init_request() */
926 		break;
927 	case REQ_OP_FLUSH:
928 		nvme_setup_flush(ns, cmd);
929 		break;
930 	case REQ_OP_ZONE_RESET_ALL:
931 	case REQ_OP_ZONE_RESET:
932 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
933 		break;
934 	case REQ_OP_ZONE_OPEN:
935 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
936 		break;
937 	case REQ_OP_ZONE_CLOSE:
938 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
939 		break;
940 	case REQ_OP_ZONE_FINISH:
941 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
942 		break;
943 	case REQ_OP_WRITE_ZEROES:
944 		ret = nvme_setup_write_zeroes(ns, req, cmd);
945 		break;
946 	case REQ_OP_DISCARD:
947 		ret = nvme_setup_discard(ns, req, cmd);
948 		break;
949 	case REQ_OP_READ:
950 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
951 		break;
952 	case REQ_OP_WRITE:
953 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
954 		break;
955 	case REQ_OP_ZONE_APPEND:
956 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
957 		break;
958 	default:
959 		WARN_ON_ONCE(1);
960 		return BLK_STS_IOERR;
961 	}
962 
963 	cmd->common.command_id = nvme_cid(req);
964 	trace_nvme_setup_cmd(req, cmd);
965 	return ret;
966 }
967 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
968 
969 /*
970  * Return values:
971  * 0:  success
972  * >0: nvme controller's cqe status response
973  * <0: kernel error in lieu of controller response
974  */
975 static int nvme_execute_rq(struct request *rq, bool at_head)
976 {
977 	blk_status_t status;
978 
979 	status = blk_execute_rq(rq, at_head);
980 	if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
981 		return -EINTR;
982 	if (nvme_req(rq)->status)
983 		return nvme_req(rq)->status;
984 	return blk_status_to_errno(status);
985 }
986 
987 /*
988  * Returns 0 on success.  If the result is negative, it's a Linux error code;
989  * if the result is positive, it's an NVM Express status code
990  */
991 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
992 		union nvme_result *result, void *buffer, unsigned bufflen,
993 		unsigned timeout, int qid, int at_head,
994 		blk_mq_req_flags_t flags)
995 {
996 	struct request *req;
997 	int ret;
998 
999 	if (qid == NVME_QID_ANY)
1000 		req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1001 	else
1002 		req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1003 						qid ? qid - 1 : 0);
1004 
1005 	if (IS_ERR(req))
1006 		return PTR_ERR(req);
1007 	nvme_init_request(req, cmd);
1008 
1009 	if (timeout)
1010 		req->timeout = timeout;
1011 
1012 	if (buffer && bufflen) {
1013 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1014 		if (ret)
1015 			goto out;
1016 	}
1017 
1018 	req->rq_flags |= RQF_QUIET;
1019 	ret = nvme_execute_rq(req, at_head);
1020 	if (result && ret >= 0)
1021 		*result = nvme_req(req)->result;
1022  out:
1023 	blk_mq_free_request(req);
1024 	return ret;
1025 }
1026 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1027 
1028 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1029 		void *buffer, unsigned bufflen)
1030 {
1031 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1032 			NVME_QID_ANY, 0, 0);
1033 }
1034 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1035 
1036 static u32 nvme_known_admin_effects(u8 opcode)
1037 {
1038 	switch (opcode) {
1039 	case nvme_admin_format_nvm:
1040 		return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1041 			NVME_CMD_EFFECTS_CSE_MASK;
1042 	case nvme_admin_sanitize_nvm:
1043 		return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1044 	default:
1045 		break;
1046 	}
1047 	return 0;
1048 }
1049 
1050 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1051 {
1052 	u32 effects = 0;
1053 
1054 	if (ns) {
1055 		if (ns->head->effects)
1056 			effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1057 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1058 			dev_warn_once(ctrl->device,
1059 				"IO command:%02x has unhandled effects:%08x\n",
1060 				opcode, effects);
1061 		return 0;
1062 	}
1063 
1064 	if (ctrl->effects)
1065 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1066 	effects |= nvme_known_admin_effects(opcode);
1067 
1068 	return effects;
1069 }
1070 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1071 
1072 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1073 			       u8 opcode)
1074 {
1075 	u32 effects = nvme_command_effects(ctrl, ns, opcode);
1076 
1077 	/*
1078 	 * For simplicity, IO to all namespaces is quiesced even if the command
1079 	 * effects say only one namespace is affected.
1080 	 */
1081 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1082 		mutex_lock(&ctrl->scan_lock);
1083 		mutex_lock(&ctrl->subsys->lock);
1084 		nvme_mpath_start_freeze(ctrl->subsys);
1085 		nvme_mpath_wait_freeze(ctrl->subsys);
1086 		nvme_start_freeze(ctrl);
1087 		nvme_wait_freeze(ctrl);
1088 	}
1089 	return effects;
1090 }
1091 
1092 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1093 			      struct nvme_command *cmd, int status)
1094 {
1095 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1096 		nvme_unfreeze(ctrl);
1097 		nvme_mpath_unfreeze(ctrl->subsys);
1098 		mutex_unlock(&ctrl->subsys->lock);
1099 		nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1100 		mutex_unlock(&ctrl->scan_lock);
1101 	}
1102 	if (effects & NVME_CMD_EFFECTS_CCC)
1103 		nvme_init_ctrl_finish(ctrl);
1104 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1105 		nvme_queue_scan(ctrl);
1106 		flush_work(&ctrl->scan_work);
1107 	}
1108 
1109 	switch (cmd->common.opcode) {
1110 	case nvme_admin_set_features:
1111 		switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1112 		case NVME_FEAT_KATO:
1113 			/*
1114 			 * Keep alive commands interval on the host should be
1115 			 * updated when KATO is modified by Set Features
1116 			 * commands.
1117 			 */
1118 			if (!status)
1119 				nvme_update_keep_alive(ctrl, cmd);
1120 			break;
1121 		default:
1122 			break;
1123 		}
1124 		break;
1125 	default:
1126 		break;
1127 	}
1128 }
1129 
1130 int nvme_execute_passthru_rq(struct request *rq)
1131 {
1132 	struct nvme_command *cmd = nvme_req(rq)->cmd;
1133 	struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1134 	struct nvme_ns *ns = rq->q->queuedata;
1135 	u32 effects;
1136 	int  ret;
1137 
1138 	effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1139 	ret = nvme_execute_rq(rq, false);
1140 	if (effects) /* nothing to be done for zero cmd effects */
1141 		nvme_passthru_end(ctrl, effects, cmd, ret);
1142 
1143 	return ret;
1144 }
1145 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1146 
1147 /*
1148  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1149  *
1150  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1151  *   accounting for transport roundtrip times [..].
1152  */
1153 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1154 {
1155 	queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1156 }
1157 
1158 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1159 {
1160 	struct nvme_ctrl *ctrl = rq->end_io_data;
1161 	unsigned long flags;
1162 	bool startka = false;
1163 
1164 	blk_mq_free_request(rq);
1165 
1166 	if (status) {
1167 		dev_err(ctrl->device,
1168 			"failed nvme_keep_alive_end_io error=%d\n",
1169 				status);
1170 		return;
1171 	}
1172 
1173 	ctrl->comp_seen = false;
1174 	spin_lock_irqsave(&ctrl->lock, flags);
1175 	if (ctrl->state == NVME_CTRL_LIVE ||
1176 	    ctrl->state == NVME_CTRL_CONNECTING)
1177 		startka = true;
1178 	spin_unlock_irqrestore(&ctrl->lock, flags);
1179 	if (startka)
1180 		nvme_queue_keep_alive_work(ctrl);
1181 }
1182 
1183 static void nvme_keep_alive_work(struct work_struct *work)
1184 {
1185 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1186 			struct nvme_ctrl, ka_work);
1187 	bool comp_seen = ctrl->comp_seen;
1188 	struct request *rq;
1189 
1190 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1191 		dev_dbg(ctrl->device,
1192 			"reschedule traffic based keep-alive timer\n");
1193 		ctrl->comp_seen = false;
1194 		nvme_queue_keep_alive_work(ctrl);
1195 		return;
1196 	}
1197 
1198 	rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1199 				  BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1200 	if (IS_ERR(rq)) {
1201 		/* allocation failure, reset the controller */
1202 		dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1203 		nvme_reset_ctrl(ctrl);
1204 		return;
1205 	}
1206 	nvme_init_request(rq, &ctrl->ka_cmd);
1207 
1208 	rq->timeout = ctrl->kato * HZ;
1209 	rq->end_io_data = ctrl;
1210 	rq->rq_flags |= RQF_QUIET;
1211 	blk_execute_rq_nowait(rq, false, nvme_keep_alive_end_io);
1212 }
1213 
1214 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1215 {
1216 	if (unlikely(ctrl->kato == 0))
1217 		return;
1218 
1219 	nvme_queue_keep_alive_work(ctrl);
1220 }
1221 
1222 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1223 {
1224 	if (unlikely(ctrl->kato == 0))
1225 		return;
1226 
1227 	cancel_delayed_work_sync(&ctrl->ka_work);
1228 }
1229 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1230 
1231 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1232 				   struct nvme_command *cmd)
1233 {
1234 	unsigned int new_kato =
1235 		DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1236 
1237 	dev_info(ctrl->device,
1238 		 "keep alive interval updated from %u ms to %u ms\n",
1239 		 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1240 
1241 	nvme_stop_keep_alive(ctrl);
1242 	ctrl->kato = new_kato;
1243 	nvme_start_keep_alive(ctrl);
1244 }
1245 
1246 /*
1247  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1248  * flag, thus sending any new CNS opcodes has a big chance of not working.
1249  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1250  * (but not for any later version).
1251  */
1252 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1253 {
1254 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1255 		return ctrl->vs < NVME_VS(1, 2, 0);
1256 	return ctrl->vs < NVME_VS(1, 1, 0);
1257 }
1258 
1259 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1260 {
1261 	struct nvme_command c = { };
1262 	int error;
1263 
1264 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1265 	c.identify.opcode = nvme_admin_identify;
1266 	c.identify.cns = NVME_ID_CNS_CTRL;
1267 
1268 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1269 	if (!*id)
1270 		return -ENOMEM;
1271 
1272 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1273 			sizeof(struct nvme_id_ctrl));
1274 	if (error)
1275 		kfree(*id);
1276 	return error;
1277 }
1278 
1279 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1280 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1281 {
1282 	const char *warn_str = "ctrl returned bogus length:";
1283 	void *data = cur;
1284 
1285 	switch (cur->nidt) {
1286 	case NVME_NIDT_EUI64:
1287 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1288 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1289 				 warn_str, cur->nidl);
1290 			return -1;
1291 		}
1292 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1293 			return NVME_NIDT_EUI64_LEN;
1294 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1295 		return NVME_NIDT_EUI64_LEN;
1296 	case NVME_NIDT_NGUID:
1297 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1298 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1299 				 warn_str, cur->nidl);
1300 			return -1;
1301 		}
1302 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1303 			return NVME_NIDT_NGUID_LEN;
1304 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1305 		return NVME_NIDT_NGUID_LEN;
1306 	case NVME_NIDT_UUID:
1307 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1308 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1309 				 warn_str, cur->nidl);
1310 			return -1;
1311 		}
1312 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1313 			return NVME_NIDT_UUID_LEN;
1314 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1315 		return NVME_NIDT_UUID_LEN;
1316 	case NVME_NIDT_CSI:
1317 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1318 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1319 				 warn_str, cur->nidl);
1320 			return -1;
1321 		}
1322 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1323 		*csi_seen = true;
1324 		return NVME_NIDT_CSI_LEN;
1325 	default:
1326 		/* Skip unknown types */
1327 		return cur->nidl;
1328 	}
1329 }
1330 
1331 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1332 		struct nvme_ns_ids *ids)
1333 {
1334 	struct nvme_command c = { };
1335 	bool csi_seen = false;
1336 	int status, pos, len;
1337 	void *data;
1338 
1339 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1340 		return 0;
1341 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1342 		return 0;
1343 
1344 	c.identify.opcode = nvme_admin_identify;
1345 	c.identify.nsid = cpu_to_le32(nsid);
1346 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1347 
1348 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1349 	if (!data)
1350 		return -ENOMEM;
1351 
1352 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1353 				      NVME_IDENTIFY_DATA_SIZE);
1354 	if (status) {
1355 		dev_warn(ctrl->device,
1356 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1357 			nsid, status);
1358 		goto free_data;
1359 	}
1360 
1361 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1362 		struct nvme_ns_id_desc *cur = data + pos;
1363 
1364 		if (cur->nidl == 0)
1365 			break;
1366 
1367 		len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1368 		if (len < 0)
1369 			break;
1370 
1371 		len += sizeof(*cur);
1372 	}
1373 
1374 	if (nvme_multi_css(ctrl) && !csi_seen) {
1375 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1376 			 nsid);
1377 		status = -EINVAL;
1378 	}
1379 
1380 free_data:
1381 	kfree(data);
1382 	return status;
1383 }
1384 
1385 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1386 			struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1387 {
1388 	struct nvme_command c = { };
1389 	int error;
1390 
1391 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1392 	c.identify.opcode = nvme_admin_identify;
1393 	c.identify.nsid = cpu_to_le32(nsid);
1394 	c.identify.cns = NVME_ID_CNS_NS;
1395 
1396 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1397 	if (!*id)
1398 		return -ENOMEM;
1399 
1400 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1401 	if (error) {
1402 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1403 		goto out_free_id;
1404 	}
1405 
1406 	error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1407 	if ((*id)->ncap == 0) /* namespace not allocated or attached */
1408 		goto out_free_id;
1409 
1410 
1411 	if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1412 		dev_info(ctrl->device,
1413 			 "Ignoring bogus Namespace Identifiers\n");
1414 	} else {
1415 		if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1416 		    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1417 			memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1418 		if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1419 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1420 			memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1421 	}
1422 
1423 	return 0;
1424 
1425 out_free_id:
1426 	kfree(*id);
1427 	return error;
1428 }
1429 
1430 static int nvme_identify_ns_cs_indep(struct nvme_ctrl *ctrl, unsigned nsid,
1431 			struct nvme_id_ns_cs_indep **id)
1432 {
1433 	struct nvme_command c = {
1434 		.identify.opcode	= nvme_admin_identify,
1435 		.identify.nsid		= cpu_to_le32(nsid),
1436 		.identify.cns		= NVME_ID_CNS_NS_CS_INDEP,
1437 	};
1438 	int ret;
1439 
1440 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1441 	if (!*id)
1442 		return -ENOMEM;
1443 
1444 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1445 	if (ret) {
1446 		dev_warn(ctrl->device,
1447 			 "Identify namespace (CS independent) failed (%d)\n",
1448 			 ret);
1449 		kfree(*id);
1450 		return ret;
1451 	}
1452 
1453 	return 0;
1454 }
1455 
1456 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1457 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1458 {
1459 	union nvme_result res = { 0 };
1460 	struct nvme_command c = { };
1461 	int ret;
1462 
1463 	c.features.opcode = op;
1464 	c.features.fid = cpu_to_le32(fid);
1465 	c.features.dword11 = cpu_to_le32(dword11);
1466 
1467 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1468 			buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1469 	if (ret >= 0 && result)
1470 		*result = le32_to_cpu(res.u32);
1471 	return ret;
1472 }
1473 
1474 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1475 		      unsigned int dword11, void *buffer, size_t buflen,
1476 		      u32 *result)
1477 {
1478 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1479 			     buflen, result);
1480 }
1481 EXPORT_SYMBOL_GPL(nvme_set_features);
1482 
1483 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1484 		      unsigned int dword11, void *buffer, size_t buflen,
1485 		      u32 *result)
1486 {
1487 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1488 			     buflen, result);
1489 }
1490 EXPORT_SYMBOL_GPL(nvme_get_features);
1491 
1492 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1493 {
1494 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1495 	u32 result;
1496 	int status, nr_io_queues;
1497 
1498 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1499 			&result);
1500 	if (status < 0)
1501 		return status;
1502 
1503 	/*
1504 	 * Degraded controllers might return an error when setting the queue
1505 	 * count.  We still want to be able to bring them online and offer
1506 	 * access to the admin queue, as that might be only way to fix them up.
1507 	 */
1508 	if (status > 0) {
1509 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1510 		*count = 0;
1511 	} else {
1512 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1513 		*count = min(*count, nr_io_queues);
1514 	}
1515 
1516 	return 0;
1517 }
1518 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1519 
1520 #define NVME_AEN_SUPPORTED \
1521 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1522 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1523 
1524 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1525 {
1526 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1527 	int status;
1528 
1529 	if (!supported_aens)
1530 		return;
1531 
1532 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1533 			NULL, 0, &result);
1534 	if (status)
1535 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1536 			 supported_aens);
1537 
1538 	queue_work(nvme_wq, &ctrl->async_event_work);
1539 }
1540 
1541 static int nvme_ns_open(struct nvme_ns *ns)
1542 {
1543 
1544 	/* should never be called due to GENHD_FL_HIDDEN */
1545 	if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1546 		goto fail;
1547 	if (!nvme_get_ns(ns))
1548 		goto fail;
1549 	if (!try_module_get(ns->ctrl->ops->module))
1550 		goto fail_put_ns;
1551 
1552 	return 0;
1553 
1554 fail_put_ns:
1555 	nvme_put_ns(ns);
1556 fail:
1557 	return -ENXIO;
1558 }
1559 
1560 static void nvme_ns_release(struct nvme_ns *ns)
1561 {
1562 
1563 	module_put(ns->ctrl->ops->module);
1564 	nvme_put_ns(ns);
1565 }
1566 
1567 static int nvme_open(struct block_device *bdev, fmode_t mode)
1568 {
1569 	return nvme_ns_open(bdev->bd_disk->private_data);
1570 }
1571 
1572 static void nvme_release(struct gendisk *disk, fmode_t mode)
1573 {
1574 	nvme_ns_release(disk->private_data);
1575 }
1576 
1577 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1578 {
1579 	/* some standard values */
1580 	geo->heads = 1 << 6;
1581 	geo->sectors = 1 << 5;
1582 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1583 	return 0;
1584 }
1585 
1586 #ifdef CONFIG_BLK_DEV_INTEGRITY
1587 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1588 				u32 max_integrity_segments)
1589 {
1590 	struct blk_integrity integrity = { };
1591 
1592 	switch (ns->pi_type) {
1593 	case NVME_NS_DPS_PI_TYPE3:
1594 		switch (ns->guard_type) {
1595 		case NVME_NVM_NS_16B_GUARD:
1596 			integrity.profile = &t10_pi_type3_crc;
1597 			integrity.tag_size = sizeof(u16) + sizeof(u32);
1598 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1599 			break;
1600 		case NVME_NVM_NS_64B_GUARD:
1601 			integrity.profile = &ext_pi_type3_crc64;
1602 			integrity.tag_size = sizeof(u16) + 6;
1603 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1604 			break;
1605 		default:
1606 			integrity.profile = NULL;
1607 			break;
1608 		}
1609 		break;
1610 	case NVME_NS_DPS_PI_TYPE1:
1611 	case NVME_NS_DPS_PI_TYPE2:
1612 		switch (ns->guard_type) {
1613 		case NVME_NVM_NS_16B_GUARD:
1614 			integrity.profile = &t10_pi_type1_crc;
1615 			integrity.tag_size = sizeof(u16);
1616 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1617 			break;
1618 		case NVME_NVM_NS_64B_GUARD:
1619 			integrity.profile = &ext_pi_type1_crc64;
1620 			integrity.tag_size = sizeof(u16);
1621 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1622 			break;
1623 		default:
1624 			integrity.profile = NULL;
1625 			break;
1626 		}
1627 		break;
1628 	default:
1629 		integrity.profile = NULL;
1630 		break;
1631 	}
1632 
1633 	integrity.tuple_size = ns->ms;
1634 	blk_integrity_register(disk, &integrity);
1635 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1636 }
1637 #else
1638 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1639 				u32 max_integrity_segments)
1640 {
1641 }
1642 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1643 
1644 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1645 {
1646 	struct nvme_ctrl *ctrl = ns->ctrl;
1647 	struct request_queue *queue = disk->queue;
1648 	u32 size = queue_logical_block_size(queue);
1649 
1650 	if (ctrl->max_discard_sectors == 0) {
1651 		blk_queue_max_discard_sectors(queue, 0);
1652 		return;
1653 	}
1654 
1655 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1656 			NVME_DSM_MAX_RANGES);
1657 
1658 	queue->limits.discard_granularity = size;
1659 
1660 	/* If discard is already enabled, don't reset queue limits */
1661 	if (queue->limits.max_discard_sectors)
1662 		return;
1663 
1664 	if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1665 		ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1666 
1667 	blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1668 	blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1669 
1670 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1671 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1672 }
1673 
1674 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1675 {
1676 	return uuid_equal(&a->uuid, &b->uuid) &&
1677 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1678 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1679 		a->csi == b->csi;
1680 }
1681 
1682 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1683 {
1684 	bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1685 	unsigned lbaf = nvme_lbaf_index(id->flbas);
1686 	struct nvme_ctrl *ctrl = ns->ctrl;
1687 	struct nvme_command c = { };
1688 	struct nvme_id_ns_nvm *nvm;
1689 	int ret = 0;
1690 	u32 elbaf;
1691 
1692 	ns->pi_size = 0;
1693 	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1694 	if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1695 		ns->pi_size = sizeof(struct t10_pi_tuple);
1696 		ns->guard_type = NVME_NVM_NS_16B_GUARD;
1697 		goto set_pi;
1698 	}
1699 
1700 	nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1701 	if (!nvm)
1702 		return -ENOMEM;
1703 
1704 	c.identify.opcode = nvme_admin_identify;
1705 	c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1706 	c.identify.cns = NVME_ID_CNS_CS_NS;
1707 	c.identify.csi = NVME_CSI_NVM;
1708 
1709 	ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1710 	if (ret)
1711 		goto free_data;
1712 
1713 	elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1714 
1715 	/* no support for storage tag formats right now */
1716 	if (nvme_elbaf_sts(elbaf))
1717 		goto free_data;
1718 
1719 	ns->guard_type = nvme_elbaf_guard_type(elbaf);
1720 	switch (ns->guard_type) {
1721 	case NVME_NVM_NS_64B_GUARD:
1722 		ns->pi_size = sizeof(struct crc64_pi_tuple);
1723 		break;
1724 	case NVME_NVM_NS_16B_GUARD:
1725 		ns->pi_size = sizeof(struct t10_pi_tuple);
1726 		break;
1727 	default:
1728 		break;
1729 	}
1730 
1731 free_data:
1732 	kfree(nvm);
1733 set_pi:
1734 	if (ns->pi_size && (first || ns->ms == ns->pi_size))
1735 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1736 	else
1737 		ns->pi_type = 0;
1738 
1739 	return ret;
1740 }
1741 
1742 static void nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1743 {
1744 	struct nvme_ctrl *ctrl = ns->ctrl;
1745 
1746 	if (nvme_init_ms(ns, id))
1747 		return;
1748 
1749 	ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1750 	if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1751 		return;
1752 
1753 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1754 		/*
1755 		 * The NVMe over Fabrics specification only supports metadata as
1756 		 * part of the extended data LBA.  We rely on HCA/HBA support to
1757 		 * remap the separate metadata buffer from the block layer.
1758 		 */
1759 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1760 			return;
1761 
1762 		ns->features |= NVME_NS_EXT_LBAS;
1763 
1764 		/*
1765 		 * The current fabrics transport drivers support namespace
1766 		 * metadata formats only if nvme_ns_has_pi() returns true.
1767 		 * Suppress support for all other formats so the namespace will
1768 		 * have a 0 capacity and not be usable through the block stack.
1769 		 *
1770 		 * Note, this check will need to be modified if any drivers
1771 		 * gain the ability to use other metadata formats.
1772 		 */
1773 		if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1774 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1775 	} else {
1776 		/*
1777 		 * For PCIe controllers, we can't easily remap the separate
1778 		 * metadata buffer from the block layer and thus require a
1779 		 * separate metadata buffer for block layer metadata/PI support.
1780 		 * We allow extended LBAs for the passthrough interface, though.
1781 		 */
1782 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
1783 			ns->features |= NVME_NS_EXT_LBAS;
1784 		else
1785 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1786 	}
1787 }
1788 
1789 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1790 		struct request_queue *q)
1791 {
1792 	bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1793 
1794 	if (ctrl->max_hw_sectors) {
1795 		u32 max_segments =
1796 			(ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1797 
1798 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
1799 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1800 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1801 	}
1802 	blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1803 	blk_queue_dma_alignment(q, 3);
1804 	blk_queue_write_cache(q, vwc, vwc);
1805 }
1806 
1807 static void nvme_update_disk_info(struct gendisk *disk,
1808 		struct nvme_ns *ns, struct nvme_id_ns *id)
1809 {
1810 	sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1811 	unsigned short bs = 1 << ns->lba_shift;
1812 	u32 atomic_bs, phys_bs, io_opt = 0;
1813 
1814 	/*
1815 	 * The block layer can't support LBA sizes larger than the page size
1816 	 * yet, so catch this early and don't allow block I/O.
1817 	 */
1818 	if (ns->lba_shift > PAGE_SHIFT) {
1819 		capacity = 0;
1820 		bs = (1 << 9);
1821 	}
1822 
1823 	blk_integrity_unregister(disk);
1824 
1825 	atomic_bs = phys_bs = bs;
1826 	if (id->nabo == 0) {
1827 		/*
1828 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1829 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1830 		 * 0 then AWUPF must be used instead.
1831 		 */
1832 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1833 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1834 		else
1835 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1836 	}
1837 
1838 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1839 		/* NPWG = Namespace Preferred Write Granularity */
1840 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1841 		/* NOWS = Namespace Optimal Write Size */
1842 		io_opt = bs * (1 + le16_to_cpu(id->nows));
1843 	}
1844 
1845 	blk_queue_logical_block_size(disk->queue, bs);
1846 	/*
1847 	 * Linux filesystems assume writing a single physical block is
1848 	 * an atomic operation. Hence limit the physical block size to the
1849 	 * value of the Atomic Write Unit Power Fail parameter.
1850 	 */
1851 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1852 	blk_queue_io_min(disk->queue, phys_bs);
1853 	blk_queue_io_opt(disk->queue, io_opt);
1854 
1855 	/*
1856 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
1857 	 * metadata masquerading as Type 0 if supported, otherwise reject block
1858 	 * I/O to namespaces with metadata except when the namespace supports
1859 	 * PI, as it can strip/insert in that case.
1860 	 */
1861 	if (ns->ms) {
1862 		if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1863 		    (ns->features & NVME_NS_METADATA_SUPPORTED))
1864 			nvme_init_integrity(disk, ns,
1865 					    ns->ctrl->max_integrity_segments);
1866 		else if (!nvme_ns_has_pi(ns))
1867 			capacity = 0;
1868 	}
1869 
1870 	set_capacity_and_notify(disk, capacity);
1871 
1872 	nvme_config_discard(disk, ns);
1873 	blk_queue_max_write_zeroes_sectors(disk->queue,
1874 					   ns->ctrl->max_zeroes_sectors);
1875 }
1876 
1877 static inline bool nvme_first_scan(struct gendisk *disk)
1878 {
1879 	/* nvme_alloc_ns() scans the disk prior to adding it */
1880 	return !disk_live(disk);
1881 }
1882 
1883 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1884 {
1885 	struct nvme_ctrl *ctrl = ns->ctrl;
1886 	u32 iob;
1887 
1888 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1889 	    is_power_of_2(ctrl->max_hw_sectors))
1890 		iob = ctrl->max_hw_sectors;
1891 	else
1892 		iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1893 
1894 	if (!iob)
1895 		return;
1896 
1897 	if (!is_power_of_2(iob)) {
1898 		if (nvme_first_scan(ns->disk))
1899 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1900 				ns->disk->disk_name, iob);
1901 		return;
1902 	}
1903 
1904 	if (blk_queue_is_zoned(ns->disk->queue)) {
1905 		if (nvme_first_scan(ns->disk))
1906 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
1907 				ns->disk->disk_name);
1908 		return;
1909 	}
1910 
1911 	blk_queue_chunk_sectors(ns->queue, iob);
1912 }
1913 
1914 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1915 {
1916 	unsigned lbaf = nvme_lbaf_index(id->flbas);
1917 	int ret;
1918 
1919 	blk_mq_freeze_queue(ns->disk->queue);
1920 	ns->lba_shift = id->lbaf[lbaf].ds;
1921 	nvme_set_queue_limits(ns->ctrl, ns->queue);
1922 
1923 	nvme_configure_metadata(ns, id);
1924 	nvme_set_chunk_sectors(ns, id);
1925 	nvme_update_disk_info(ns->disk, ns, id);
1926 
1927 	if (ns->head->ids.csi == NVME_CSI_ZNS) {
1928 		ret = nvme_update_zone_info(ns, lbaf);
1929 		if (ret)
1930 			goto out_unfreeze;
1931 	}
1932 
1933 	set_disk_ro(ns->disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1934 		test_bit(NVME_NS_FORCE_RO, &ns->flags));
1935 	set_bit(NVME_NS_READY, &ns->flags);
1936 	blk_mq_unfreeze_queue(ns->disk->queue);
1937 
1938 	if (blk_queue_is_zoned(ns->queue)) {
1939 		ret = nvme_revalidate_zones(ns);
1940 		if (ret && !nvme_first_scan(ns->disk))
1941 			return ret;
1942 	}
1943 
1944 	if (nvme_ns_head_multipath(ns->head)) {
1945 		blk_mq_freeze_queue(ns->head->disk->queue);
1946 		nvme_update_disk_info(ns->head->disk, ns, id);
1947 		set_disk_ro(ns->head->disk,
1948 			    (id->nsattr & NVME_NS_ATTR_RO) ||
1949 				    test_bit(NVME_NS_FORCE_RO, &ns->flags));
1950 		nvme_mpath_revalidate_paths(ns);
1951 		blk_stack_limits(&ns->head->disk->queue->limits,
1952 				 &ns->queue->limits, 0);
1953 		disk_update_readahead(ns->head->disk);
1954 		blk_mq_unfreeze_queue(ns->head->disk->queue);
1955 	}
1956 	return 0;
1957 
1958 out_unfreeze:
1959 	/*
1960 	 * If probing fails due an unsupported feature, hide the block device,
1961 	 * but still allow other access.
1962 	 */
1963 	if (ret == -ENODEV) {
1964 		ns->disk->flags |= GENHD_FL_HIDDEN;
1965 		set_bit(NVME_NS_READY, &ns->flags);
1966 		ret = 0;
1967 	}
1968 	blk_mq_unfreeze_queue(ns->disk->queue);
1969 	return ret;
1970 }
1971 
1972 static char nvme_pr_type(enum pr_type type)
1973 {
1974 	switch (type) {
1975 	case PR_WRITE_EXCLUSIVE:
1976 		return 1;
1977 	case PR_EXCLUSIVE_ACCESS:
1978 		return 2;
1979 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1980 		return 3;
1981 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1982 		return 4;
1983 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1984 		return 5;
1985 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1986 		return 6;
1987 	default:
1988 		return 0;
1989 	}
1990 }
1991 
1992 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
1993 		struct nvme_command *c, u8 data[16])
1994 {
1995 	struct nvme_ns_head *head = bdev->bd_disk->private_data;
1996 	int srcu_idx = srcu_read_lock(&head->srcu);
1997 	struct nvme_ns *ns = nvme_find_path(head);
1998 	int ret = -EWOULDBLOCK;
1999 
2000 	if (ns) {
2001 		c->common.nsid = cpu_to_le32(ns->head->ns_id);
2002 		ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
2003 	}
2004 	srcu_read_unlock(&head->srcu, srcu_idx);
2005 	return ret;
2006 }
2007 
2008 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
2009 		u8 data[16])
2010 {
2011 	c->common.nsid = cpu_to_le32(ns->head->ns_id);
2012 	return nvme_submit_sync_cmd(ns->queue, c, data, 16);
2013 }
2014 
2015 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2016 				u64 key, u64 sa_key, u8 op)
2017 {
2018 	struct nvme_command c = { };
2019 	u8 data[16] = { 0, };
2020 
2021 	put_unaligned_le64(key, &data[0]);
2022 	put_unaligned_le64(sa_key, &data[8]);
2023 
2024 	c.common.opcode = op;
2025 	c.common.cdw10 = cpu_to_le32(cdw10);
2026 
2027 	if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2028 	    bdev->bd_disk->fops == &nvme_ns_head_ops)
2029 		return nvme_send_ns_head_pr_command(bdev, &c, data);
2030 	return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
2031 }
2032 
2033 static int nvme_pr_register(struct block_device *bdev, u64 old,
2034 		u64 new, unsigned flags)
2035 {
2036 	u32 cdw10;
2037 
2038 	if (flags & ~PR_FL_IGNORE_KEY)
2039 		return -EOPNOTSUPP;
2040 
2041 	cdw10 = old ? 2 : 0;
2042 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2043 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2044 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2045 }
2046 
2047 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2048 		enum pr_type type, unsigned flags)
2049 {
2050 	u32 cdw10;
2051 
2052 	if (flags & ~PR_FL_IGNORE_KEY)
2053 		return -EOPNOTSUPP;
2054 
2055 	cdw10 = nvme_pr_type(type) << 8;
2056 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2057 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2058 }
2059 
2060 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2061 		enum pr_type type, bool abort)
2062 {
2063 	u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2064 
2065 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2066 }
2067 
2068 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2069 {
2070 	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2071 
2072 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2073 }
2074 
2075 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2076 {
2077 	u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2078 
2079 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2080 }
2081 
2082 const struct pr_ops nvme_pr_ops = {
2083 	.pr_register	= nvme_pr_register,
2084 	.pr_reserve	= nvme_pr_reserve,
2085 	.pr_release	= nvme_pr_release,
2086 	.pr_preempt	= nvme_pr_preempt,
2087 	.pr_clear	= nvme_pr_clear,
2088 };
2089 
2090 #ifdef CONFIG_BLK_SED_OPAL
2091 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2092 		bool send)
2093 {
2094 	struct nvme_ctrl *ctrl = data;
2095 	struct nvme_command cmd = { };
2096 
2097 	if (send)
2098 		cmd.common.opcode = nvme_admin_security_send;
2099 	else
2100 		cmd.common.opcode = nvme_admin_security_recv;
2101 	cmd.common.nsid = 0;
2102 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2103 	cmd.common.cdw11 = cpu_to_le32(len);
2104 
2105 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2106 			NVME_QID_ANY, 1, 0);
2107 }
2108 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2109 #endif /* CONFIG_BLK_SED_OPAL */
2110 
2111 #ifdef CONFIG_BLK_DEV_ZONED
2112 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2113 		unsigned int nr_zones, report_zones_cb cb, void *data)
2114 {
2115 	return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2116 			data);
2117 }
2118 #else
2119 #define nvme_report_zones	NULL
2120 #endif /* CONFIG_BLK_DEV_ZONED */
2121 
2122 static const struct block_device_operations nvme_bdev_ops = {
2123 	.owner		= THIS_MODULE,
2124 	.ioctl		= nvme_ioctl,
2125 	.open		= nvme_open,
2126 	.release	= nvme_release,
2127 	.getgeo		= nvme_getgeo,
2128 	.report_zones	= nvme_report_zones,
2129 	.pr_ops		= &nvme_pr_ops,
2130 };
2131 
2132 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 timeout, bool enabled)
2133 {
2134 	unsigned long timeout_jiffies = ((timeout + 1) * HZ / 2) + jiffies;
2135 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2136 	int ret;
2137 
2138 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2139 		if (csts == ~0)
2140 			return -ENODEV;
2141 		if ((csts & NVME_CSTS_RDY) == bit)
2142 			break;
2143 
2144 		usleep_range(1000, 2000);
2145 		if (fatal_signal_pending(current))
2146 			return -EINTR;
2147 		if (time_after(jiffies, timeout_jiffies)) {
2148 			dev_err(ctrl->device,
2149 				"Device not ready; aborting %s, CSTS=0x%x\n",
2150 				enabled ? "initialisation" : "reset", csts);
2151 			return -ENODEV;
2152 		}
2153 	}
2154 
2155 	return ret;
2156 }
2157 
2158 /*
2159  * If the device has been passed off to us in an enabled state, just clear
2160  * the enabled bit.  The spec says we should set the 'shutdown notification
2161  * bits', but doing so may cause the device to complete commands to the
2162  * admin queue ... and we don't know what memory that might be pointing at!
2163  */
2164 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2165 {
2166 	int ret;
2167 
2168 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2169 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2170 
2171 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2172 	if (ret)
2173 		return ret;
2174 
2175 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2176 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2177 
2178 	return nvme_wait_ready(ctrl, NVME_CAP_TIMEOUT(ctrl->cap), false);
2179 }
2180 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2181 
2182 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2183 {
2184 	unsigned dev_page_min;
2185 	u32 timeout;
2186 	int ret;
2187 
2188 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2189 	if (ret) {
2190 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2191 		return ret;
2192 	}
2193 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2194 
2195 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2196 		dev_err(ctrl->device,
2197 			"Minimum device page size %u too large for host (%u)\n",
2198 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2199 		return -ENODEV;
2200 	}
2201 
2202 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2203 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2204 	else
2205 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2206 
2207 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2208 		u32 crto;
2209 
2210 		ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2211 		if (ret) {
2212 			dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2213 				ret);
2214 			return ret;
2215 		}
2216 
2217 		if (ctrl->cap & NVME_CAP_CRMS_CRIMS) {
2218 			ctrl->ctrl_config |= NVME_CC_CRIME;
2219 			timeout = NVME_CRTO_CRIMT(crto);
2220 		} else {
2221 			timeout = NVME_CRTO_CRWMT(crto);
2222 		}
2223 	} else {
2224 		timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2225 	}
2226 
2227 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2228 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2229 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2230 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2231 
2232 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2233 	if (ret)
2234 		return ret;
2235 	return nvme_wait_ready(ctrl, timeout, true);
2236 }
2237 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2238 
2239 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2240 {
2241 	unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2242 	u32 csts;
2243 	int ret;
2244 
2245 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2246 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2247 
2248 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2249 	if (ret)
2250 		return ret;
2251 
2252 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2253 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2254 			break;
2255 
2256 		msleep(100);
2257 		if (fatal_signal_pending(current))
2258 			return -EINTR;
2259 		if (time_after(jiffies, timeout)) {
2260 			dev_err(ctrl->device,
2261 				"Device shutdown incomplete; abort shutdown\n");
2262 			return -ENODEV;
2263 		}
2264 	}
2265 
2266 	return ret;
2267 }
2268 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2269 
2270 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2271 {
2272 	__le64 ts;
2273 	int ret;
2274 
2275 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2276 		return 0;
2277 
2278 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2279 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2280 			NULL);
2281 	if (ret)
2282 		dev_warn_once(ctrl->device,
2283 			"could not set timestamp (%d)\n", ret);
2284 	return ret;
2285 }
2286 
2287 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2288 {
2289 	struct nvme_feat_host_behavior *host;
2290 	u8 acre = 0, lbafee = 0;
2291 	int ret;
2292 
2293 	/* Don't bother enabling the feature if retry delay is not reported */
2294 	if (ctrl->crdt[0])
2295 		acre = NVME_ENABLE_ACRE;
2296 	if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2297 		lbafee = NVME_ENABLE_LBAFEE;
2298 
2299 	if (!acre && !lbafee)
2300 		return 0;
2301 
2302 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2303 	if (!host)
2304 		return 0;
2305 
2306 	host->acre = acre;
2307 	host->lbafee = lbafee;
2308 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2309 				host, sizeof(*host), NULL);
2310 	kfree(host);
2311 	return ret;
2312 }
2313 
2314 /*
2315  * The function checks whether the given total (exlat + enlat) latency of
2316  * a power state allows the latter to be used as an APST transition target.
2317  * It does so by comparing the latency to the primary and secondary latency
2318  * tolerances defined by module params. If there's a match, the corresponding
2319  * timeout value is returned and the matching tolerance index (1 or 2) is
2320  * reported.
2321  */
2322 static bool nvme_apst_get_transition_time(u64 total_latency,
2323 		u64 *transition_time, unsigned *last_index)
2324 {
2325 	if (total_latency <= apst_primary_latency_tol_us) {
2326 		if (*last_index == 1)
2327 			return false;
2328 		*last_index = 1;
2329 		*transition_time = apst_primary_timeout_ms;
2330 		return true;
2331 	}
2332 	if (apst_secondary_timeout_ms &&
2333 		total_latency <= apst_secondary_latency_tol_us) {
2334 		if (*last_index <= 2)
2335 			return false;
2336 		*last_index = 2;
2337 		*transition_time = apst_secondary_timeout_ms;
2338 		return true;
2339 	}
2340 	return false;
2341 }
2342 
2343 /*
2344  * APST (Autonomous Power State Transition) lets us program a table of power
2345  * state transitions that the controller will perform automatically.
2346  *
2347  * Depending on module params, one of the two supported techniques will be used:
2348  *
2349  * - If the parameters provide explicit timeouts and tolerances, they will be
2350  *   used to build a table with up to 2 non-operational states to transition to.
2351  *   The default parameter values were selected based on the values used by
2352  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2353  *   regeneration of the APST table in the event of switching between external
2354  *   and battery power, the timeouts and tolerances reflect a compromise
2355  *   between values used by Microsoft for AC and battery scenarios.
2356  * - If not, we'll configure the table with a simple heuristic: we are willing
2357  *   to spend at most 2% of the time transitioning between power states.
2358  *   Therefore, when running in any given state, we will enter the next
2359  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2360  *   microseconds, as long as that state's exit latency is under the requested
2361  *   maximum latency.
2362  *
2363  * We will not autonomously enter any non-operational state for which the total
2364  * latency exceeds ps_max_latency_us.
2365  *
2366  * Users can set ps_max_latency_us to zero to turn off APST.
2367  */
2368 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2369 {
2370 	struct nvme_feat_auto_pst *table;
2371 	unsigned apste = 0;
2372 	u64 max_lat_us = 0;
2373 	__le64 target = 0;
2374 	int max_ps = -1;
2375 	int state;
2376 	int ret;
2377 	unsigned last_lt_index = UINT_MAX;
2378 
2379 	/*
2380 	 * If APST isn't supported or if we haven't been initialized yet,
2381 	 * then don't do anything.
2382 	 */
2383 	if (!ctrl->apsta)
2384 		return 0;
2385 
2386 	if (ctrl->npss > 31) {
2387 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2388 		return 0;
2389 	}
2390 
2391 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2392 	if (!table)
2393 		return 0;
2394 
2395 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2396 		/* Turn off APST. */
2397 		dev_dbg(ctrl->device, "APST disabled\n");
2398 		goto done;
2399 	}
2400 
2401 	/*
2402 	 * Walk through all states from lowest- to highest-power.
2403 	 * According to the spec, lower-numbered states use more power.  NPSS,
2404 	 * despite the name, is the index of the lowest-power state, not the
2405 	 * number of states.
2406 	 */
2407 	for (state = (int)ctrl->npss; state >= 0; state--) {
2408 		u64 total_latency_us, exit_latency_us, transition_ms;
2409 
2410 		if (target)
2411 			table->entries[state] = target;
2412 
2413 		/*
2414 		 * Don't allow transitions to the deepest state if it's quirked
2415 		 * off.
2416 		 */
2417 		if (state == ctrl->npss &&
2418 		    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2419 			continue;
2420 
2421 		/*
2422 		 * Is this state a useful non-operational state for higher-power
2423 		 * states to autonomously transition to?
2424 		 */
2425 		if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2426 			continue;
2427 
2428 		exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2429 		if (exit_latency_us > ctrl->ps_max_latency_us)
2430 			continue;
2431 
2432 		total_latency_us = exit_latency_us +
2433 			le32_to_cpu(ctrl->psd[state].entry_lat);
2434 
2435 		/*
2436 		 * This state is good. It can be used as the APST idle target
2437 		 * for higher power states.
2438 		 */
2439 		if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2440 			if (!nvme_apst_get_transition_time(total_latency_us,
2441 					&transition_ms, &last_lt_index))
2442 				continue;
2443 		} else {
2444 			transition_ms = total_latency_us + 19;
2445 			do_div(transition_ms, 20);
2446 			if (transition_ms > (1 << 24) - 1)
2447 				transition_ms = (1 << 24) - 1;
2448 		}
2449 
2450 		target = cpu_to_le64((state << 3) | (transition_ms << 8));
2451 		if (max_ps == -1)
2452 			max_ps = state;
2453 		if (total_latency_us > max_lat_us)
2454 			max_lat_us = total_latency_us;
2455 	}
2456 
2457 	if (max_ps == -1)
2458 		dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2459 	else
2460 		dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2461 			max_ps, max_lat_us, (int)sizeof(*table), table);
2462 	apste = 1;
2463 
2464 done:
2465 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2466 				table, sizeof(*table), NULL);
2467 	if (ret)
2468 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2469 	kfree(table);
2470 	return ret;
2471 }
2472 
2473 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2474 {
2475 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2476 	u64 latency;
2477 
2478 	switch (val) {
2479 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2480 	case PM_QOS_LATENCY_ANY:
2481 		latency = U64_MAX;
2482 		break;
2483 
2484 	default:
2485 		latency = val;
2486 	}
2487 
2488 	if (ctrl->ps_max_latency_us != latency) {
2489 		ctrl->ps_max_latency_us = latency;
2490 		if (ctrl->state == NVME_CTRL_LIVE)
2491 			nvme_configure_apst(ctrl);
2492 	}
2493 }
2494 
2495 struct nvme_core_quirk_entry {
2496 	/*
2497 	 * NVMe model and firmware strings are padded with spaces.  For
2498 	 * simplicity, strings in the quirk table are padded with NULLs
2499 	 * instead.
2500 	 */
2501 	u16 vid;
2502 	const char *mn;
2503 	const char *fr;
2504 	unsigned long quirks;
2505 };
2506 
2507 static const struct nvme_core_quirk_entry core_quirks[] = {
2508 	{
2509 		/*
2510 		 * This Toshiba device seems to die using any APST states.  See:
2511 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2512 		 */
2513 		.vid = 0x1179,
2514 		.mn = "THNSF5256GPUK TOSHIBA",
2515 		.quirks = NVME_QUIRK_NO_APST,
2516 	},
2517 	{
2518 		/*
2519 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2520 		 * condition associated with actions related to suspend to idle
2521 		 * LiteON has resolved the problem in future firmware
2522 		 */
2523 		.vid = 0x14a4,
2524 		.fr = "22301111",
2525 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2526 	},
2527 	{
2528 		/*
2529 		 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2530 		 * aborts I/O during any load, but more easily reproducible
2531 		 * with discards (fstrim).
2532 		 *
2533 		 * The device is left in a state where it is also not possible
2534 		 * to use "nvme set-feature" to disable APST, but booting with
2535 		 * nvme_core.default_ps_max_latency=0 works.
2536 		 */
2537 		.vid = 0x1e0f,
2538 		.mn = "KCD6XVUL6T40",
2539 		.quirks = NVME_QUIRK_NO_APST,
2540 	}
2541 };
2542 
2543 /* match is null-terminated but idstr is space-padded. */
2544 static bool string_matches(const char *idstr, const char *match, size_t len)
2545 {
2546 	size_t matchlen;
2547 
2548 	if (!match)
2549 		return true;
2550 
2551 	matchlen = strlen(match);
2552 	WARN_ON_ONCE(matchlen > len);
2553 
2554 	if (memcmp(idstr, match, matchlen))
2555 		return false;
2556 
2557 	for (; matchlen < len; matchlen++)
2558 		if (idstr[matchlen] != ' ')
2559 			return false;
2560 
2561 	return true;
2562 }
2563 
2564 static bool quirk_matches(const struct nvme_id_ctrl *id,
2565 			  const struct nvme_core_quirk_entry *q)
2566 {
2567 	return q->vid == le16_to_cpu(id->vid) &&
2568 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2569 		string_matches(id->fr, q->fr, sizeof(id->fr));
2570 }
2571 
2572 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2573 		struct nvme_id_ctrl *id)
2574 {
2575 	size_t nqnlen;
2576 	int off;
2577 
2578 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2579 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2580 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2581 			strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2582 			return;
2583 		}
2584 
2585 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2586 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2587 	}
2588 
2589 	/* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2590 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2591 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2592 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2593 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2594 	off += sizeof(id->sn);
2595 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2596 	off += sizeof(id->mn);
2597 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2598 }
2599 
2600 static void nvme_release_subsystem(struct device *dev)
2601 {
2602 	struct nvme_subsystem *subsys =
2603 		container_of(dev, struct nvme_subsystem, dev);
2604 
2605 	if (subsys->instance >= 0)
2606 		ida_free(&nvme_instance_ida, subsys->instance);
2607 	kfree(subsys);
2608 }
2609 
2610 static void nvme_destroy_subsystem(struct kref *ref)
2611 {
2612 	struct nvme_subsystem *subsys =
2613 			container_of(ref, struct nvme_subsystem, ref);
2614 
2615 	mutex_lock(&nvme_subsystems_lock);
2616 	list_del(&subsys->entry);
2617 	mutex_unlock(&nvme_subsystems_lock);
2618 
2619 	ida_destroy(&subsys->ns_ida);
2620 	device_del(&subsys->dev);
2621 	put_device(&subsys->dev);
2622 }
2623 
2624 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2625 {
2626 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2627 }
2628 
2629 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2630 {
2631 	struct nvme_subsystem *subsys;
2632 
2633 	lockdep_assert_held(&nvme_subsystems_lock);
2634 
2635 	/*
2636 	 * Fail matches for discovery subsystems. This results
2637 	 * in each discovery controller bound to a unique subsystem.
2638 	 * This avoids issues with validating controller values
2639 	 * that can only be true when there is a single unique subsystem.
2640 	 * There may be multiple and completely independent entities
2641 	 * that provide discovery controllers.
2642 	 */
2643 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2644 		return NULL;
2645 
2646 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2647 		if (strcmp(subsys->subnqn, subsysnqn))
2648 			continue;
2649 		if (!kref_get_unless_zero(&subsys->ref))
2650 			continue;
2651 		return subsys;
2652 	}
2653 
2654 	return NULL;
2655 }
2656 
2657 #define SUBSYS_ATTR_RO(_name, _mode, _show)			\
2658 	struct device_attribute subsys_attr_##_name = \
2659 		__ATTR(_name, _mode, _show, NULL)
2660 
2661 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2662 				    struct device_attribute *attr,
2663 				    char *buf)
2664 {
2665 	struct nvme_subsystem *subsys =
2666 		container_of(dev, struct nvme_subsystem, dev);
2667 
2668 	return sysfs_emit(buf, "%s\n", subsys->subnqn);
2669 }
2670 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2671 
2672 static ssize_t nvme_subsys_show_type(struct device *dev,
2673 				    struct device_attribute *attr,
2674 				    char *buf)
2675 {
2676 	struct nvme_subsystem *subsys =
2677 		container_of(dev, struct nvme_subsystem, dev);
2678 
2679 	switch (subsys->subtype) {
2680 	case NVME_NQN_DISC:
2681 		return sysfs_emit(buf, "discovery\n");
2682 	case NVME_NQN_NVME:
2683 		return sysfs_emit(buf, "nvm\n");
2684 	default:
2685 		return sysfs_emit(buf, "reserved\n");
2686 	}
2687 }
2688 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2689 
2690 #define nvme_subsys_show_str_function(field)				\
2691 static ssize_t subsys_##field##_show(struct device *dev,		\
2692 			    struct device_attribute *attr, char *buf)	\
2693 {									\
2694 	struct nvme_subsystem *subsys =					\
2695 		container_of(dev, struct nvme_subsystem, dev);		\
2696 	return sysfs_emit(buf, "%.*s\n",				\
2697 			   (int)sizeof(subsys->field), subsys->field);	\
2698 }									\
2699 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2700 
2701 nvme_subsys_show_str_function(model);
2702 nvme_subsys_show_str_function(serial);
2703 nvme_subsys_show_str_function(firmware_rev);
2704 
2705 static struct attribute *nvme_subsys_attrs[] = {
2706 	&subsys_attr_model.attr,
2707 	&subsys_attr_serial.attr,
2708 	&subsys_attr_firmware_rev.attr,
2709 	&subsys_attr_subsysnqn.attr,
2710 	&subsys_attr_subsystype.attr,
2711 #ifdef CONFIG_NVME_MULTIPATH
2712 	&subsys_attr_iopolicy.attr,
2713 #endif
2714 	NULL,
2715 };
2716 
2717 static const struct attribute_group nvme_subsys_attrs_group = {
2718 	.attrs = nvme_subsys_attrs,
2719 };
2720 
2721 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2722 	&nvme_subsys_attrs_group,
2723 	NULL,
2724 };
2725 
2726 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2727 {
2728 	return ctrl->opts && ctrl->opts->discovery_nqn;
2729 }
2730 
2731 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2732 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2733 {
2734 	struct nvme_ctrl *tmp;
2735 
2736 	lockdep_assert_held(&nvme_subsystems_lock);
2737 
2738 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2739 		if (nvme_state_terminal(tmp))
2740 			continue;
2741 
2742 		if (tmp->cntlid == ctrl->cntlid) {
2743 			dev_err(ctrl->device,
2744 				"Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2745 				ctrl->cntlid, dev_name(tmp->device),
2746 				subsys->subnqn);
2747 			return false;
2748 		}
2749 
2750 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2751 		    nvme_discovery_ctrl(ctrl))
2752 			continue;
2753 
2754 		dev_err(ctrl->device,
2755 			"Subsystem does not support multiple controllers\n");
2756 		return false;
2757 	}
2758 
2759 	return true;
2760 }
2761 
2762 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2763 {
2764 	struct nvme_subsystem *subsys, *found;
2765 	int ret;
2766 
2767 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2768 	if (!subsys)
2769 		return -ENOMEM;
2770 
2771 	subsys->instance = -1;
2772 	mutex_init(&subsys->lock);
2773 	kref_init(&subsys->ref);
2774 	INIT_LIST_HEAD(&subsys->ctrls);
2775 	INIT_LIST_HEAD(&subsys->nsheads);
2776 	nvme_init_subnqn(subsys, ctrl, id);
2777 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2778 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2779 	memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2780 	subsys->vendor_id = le16_to_cpu(id->vid);
2781 	subsys->cmic = id->cmic;
2782 
2783 	/* Versions prior to 1.4 don't necessarily report a valid type */
2784 	if (id->cntrltype == NVME_CTRL_DISC ||
2785 	    !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2786 		subsys->subtype = NVME_NQN_DISC;
2787 	else
2788 		subsys->subtype = NVME_NQN_NVME;
2789 
2790 	if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2791 		dev_err(ctrl->device,
2792 			"Subsystem %s is not a discovery controller",
2793 			subsys->subnqn);
2794 		kfree(subsys);
2795 		return -EINVAL;
2796 	}
2797 	subsys->awupf = le16_to_cpu(id->awupf);
2798 	nvme_mpath_default_iopolicy(subsys);
2799 
2800 	subsys->dev.class = nvme_subsys_class;
2801 	subsys->dev.release = nvme_release_subsystem;
2802 	subsys->dev.groups = nvme_subsys_attrs_groups;
2803 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2804 	device_initialize(&subsys->dev);
2805 
2806 	mutex_lock(&nvme_subsystems_lock);
2807 	found = __nvme_find_get_subsystem(subsys->subnqn);
2808 	if (found) {
2809 		put_device(&subsys->dev);
2810 		subsys = found;
2811 
2812 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2813 			ret = -EINVAL;
2814 			goto out_put_subsystem;
2815 		}
2816 	} else {
2817 		ret = device_add(&subsys->dev);
2818 		if (ret) {
2819 			dev_err(ctrl->device,
2820 				"failed to register subsystem device.\n");
2821 			put_device(&subsys->dev);
2822 			goto out_unlock;
2823 		}
2824 		ida_init(&subsys->ns_ida);
2825 		list_add_tail(&subsys->entry, &nvme_subsystems);
2826 	}
2827 
2828 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2829 				dev_name(ctrl->device));
2830 	if (ret) {
2831 		dev_err(ctrl->device,
2832 			"failed to create sysfs link from subsystem.\n");
2833 		goto out_put_subsystem;
2834 	}
2835 
2836 	if (!found)
2837 		subsys->instance = ctrl->instance;
2838 	ctrl->subsys = subsys;
2839 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2840 	mutex_unlock(&nvme_subsystems_lock);
2841 	return 0;
2842 
2843 out_put_subsystem:
2844 	nvme_put_subsystem(subsys);
2845 out_unlock:
2846 	mutex_unlock(&nvme_subsystems_lock);
2847 	return ret;
2848 }
2849 
2850 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2851 		void *log, size_t size, u64 offset)
2852 {
2853 	struct nvme_command c = { };
2854 	u32 dwlen = nvme_bytes_to_numd(size);
2855 
2856 	c.get_log_page.opcode = nvme_admin_get_log_page;
2857 	c.get_log_page.nsid = cpu_to_le32(nsid);
2858 	c.get_log_page.lid = log_page;
2859 	c.get_log_page.lsp = lsp;
2860 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2861 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2862 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2863 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2864 	c.get_log_page.csi = csi;
2865 
2866 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2867 }
2868 
2869 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2870 				struct nvme_effects_log **log)
2871 {
2872 	struct nvme_effects_log	*cel = xa_load(&ctrl->cels, csi);
2873 	int ret;
2874 
2875 	if (cel)
2876 		goto out;
2877 
2878 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2879 	if (!cel)
2880 		return -ENOMEM;
2881 
2882 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2883 			cel, sizeof(*cel), 0);
2884 	if (ret) {
2885 		kfree(cel);
2886 		return ret;
2887 	}
2888 
2889 	xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2890 out:
2891 	*log = cel;
2892 	return 0;
2893 }
2894 
2895 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2896 {
2897 	u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2898 
2899 	if (check_shl_overflow(1U, units + page_shift - 9, &val))
2900 		return UINT_MAX;
2901 	return val;
2902 }
2903 
2904 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2905 {
2906 	struct nvme_command c = { };
2907 	struct nvme_id_ctrl_nvm *id;
2908 	int ret;
2909 
2910 	if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2911 		ctrl->max_discard_sectors = UINT_MAX;
2912 		ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2913 	} else {
2914 		ctrl->max_discard_sectors = 0;
2915 		ctrl->max_discard_segments = 0;
2916 	}
2917 
2918 	/*
2919 	 * Even though NVMe spec explicitly states that MDTS is not applicable
2920 	 * to the write-zeroes, we are cautious and limit the size to the
2921 	 * controllers max_hw_sectors value, which is based on the MDTS field
2922 	 * and possibly other limiting factors.
2923 	 */
2924 	if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2925 	    !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2926 		ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2927 	else
2928 		ctrl->max_zeroes_sectors = 0;
2929 
2930 	if (nvme_ctrl_limited_cns(ctrl))
2931 		return 0;
2932 
2933 	id = kzalloc(sizeof(*id), GFP_KERNEL);
2934 	if (!id)
2935 		return 0;
2936 
2937 	c.identify.opcode = nvme_admin_identify;
2938 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
2939 	c.identify.csi = NVME_CSI_NVM;
2940 
2941 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2942 	if (ret)
2943 		goto free_data;
2944 
2945 	if (id->dmrl)
2946 		ctrl->max_discard_segments = id->dmrl;
2947 	ctrl->dmrsl = le32_to_cpu(id->dmrsl);
2948 	if (id->wzsl)
2949 		ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2950 
2951 free_data:
2952 	kfree(id);
2953 	return ret;
2954 }
2955 
2956 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2957 {
2958 	struct nvme_id_ctrl *id;
2959 	u32 max_hw_sectors;
2960 	bool prev_apst_enabled;
2961 	int ret;
2962 
2963 	ret = nvme_identify_ctrl(ctrl, &id);
2964 	if (ret) {
2965 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2966 		return -EIO;
2967 	}
2968 
2969 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2970 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2971 		if (ret < 0)
2972 			goto out_free;
2973 	}
2974 
2975 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
2976 		ctrl->cntlid = le16_to_cpu(id->cntlid);
2977 
2978 	if (!ctrl->identified) {
2979 		unsigned int i;
2980 
2981 		ret = nvme_init_subsystem(ctrl, id);
2982 		if (ret)
2983 			goto out_free;
2984 
2985 		/*
2986 		 * Check for quirks.  Quirk can depend on firmware version,
2987 		 * so, in principle, the set of quirks present can change
2988 		 * across a reset.  As a possible future enhancement, we
2989 		 * could re-scan for quirks every time we reinitialize
2990 		 * the device, but we'd have to make sure that the driver
2991 		 * behaves intelligently if the quirks change.
2992 		 */
2993 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2994 			if (quirk_matches(id, &core_quirks[i]))
2995 				ctrl->quirks |= core_quirks[i].quirks;
2996 		}
2997 	}
2998 
2999 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3000 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3001 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3002 	}
3003 
3004 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3005 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3006 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3007 
3008 	ctrl->oacs = le16_to_cpu(id->oacs);
3009 	ctrl->oncs = le16_to_cpu(id->oncs);
3010 	ctrl->mtfa = le16_to_cpu(id->mtfa);
3011 	ctrl->oaes = le32_to_cpu(id->oaes);
3012 	ctrl->wctemp = le16_to_cpu(id->wctemp);
3013 	ctrl->cctemp = le16_to_cpu(id->cctemp);
3014 
3015 	atomic_set(&ctrl->abort_limit, id->acl + 1);
3016 	ctrl->vwc = id->vwc;
3017 	if (id->mdts)
3018 		max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3019 	else
3020 		max_hw_sectors = UINT_MAX;
3021 	ctrl->max_hw_sectors =
3022 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3023 
3024 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
3025 	ctrl->sgls = le32_to_cpu(id->sgls);
3026 	ctrl->kas = le16_to_cpu(id->kas);
3027 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
3028 	ctrl->ctratt = le32_to_cpu(id->ctratt);
3029 
3030 	ctrl->cntrltype = id->cntrltype;
3031 	ctrl->dctype = id->dctype;
3032 
3033 	if (id->rtd3e) {
3034 		/* us -> s */
3035 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3036 
3037 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3038 						 shutdown_timeout, 60);
3039 
3040 		if (ctrl->shutdown_timeout != shutdown_timeout)
3041 			dev_info(ctrl->device,
3042 				 "Shutdown timeout set to %u seconds\n",
3043 				 ctrl->shutdown_timeout);
3044 	} else
3045 		ctrl->shutdown_timeout = shutdown_timeout;
3046 
3047 	ctrl->npss = id->npss;
3048 	ctrl->apsta = id->apsta;
3049 	prev_apst_enabled = ctrl->apst_enabled;
3050 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3051 		if (force_apst && id->apsta) {
3052 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3053 			ctrl->apst_enabled = true;
3054 		} else {
3055 			ctrl->apst_enabled = false;
3056 		}
3057 	} else {
3058 		ctrl->apst_enabled = id->apsta;
3059 	}
3060 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3061 
3062 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3063 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3064 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3065 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3066 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3067 
3068 		/*
3069 		 * In fabrics we need to verify the cntlid matches the
3070 		 * admin connect
3071 		 */
3072 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3073 			dev_err(ctrl->device,
3074 				"Mismatching cntlid: Connect %u vs Identify "
3075 				"%u, rejecting\n",
3076 				ctrl->cntlid, le16_to_cpu(id->cntlid));
3077 			ret = -EINVAL;
3078 			goto out_free;
3079 		}
3080 
3081 		if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3082 			dev_err(ctrl->device,
3083 				"keep-alive support is mandatory for fabrics\n");
3084 			ret = -EINVAL;
3085 			goto out_free;
3086 		}
3087 	} else {
3088 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3089 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3090 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3091 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3092 	}
3093 
3094 	ret = nvme_mpath_init_identify(ctrl, id);
3095 	if (ret < 0)
3096 		goto out_free;
3097 
3098 	if (ctrl->apst_enabled && !prev_apst_enabled)
3099 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3100 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3101 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3102 
3103 out_free:
3104 	kfree(id);
3105 	return ret;
3106 }
3107 
3108 /*
3109  * Initialize the cached copies of the Identify data and various controller
3110  * register in our nvme_ctrl structure.  This should be called as soon as
3111  * the admin queue is fully up and running.
3112  */
3113 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3114 {
3115 	int ret;
3116 
3117 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3118 	if (ret) {
3119 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3120 		return ret;
3121 	}
3122 
3123 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3124 
3125 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3126 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3127 
3128 	ret = nvme_init_identify(ctrl);
3129 	if (ret)
3130 		return ret;
3131 
3132 	ret = nvme_configure_apst(ctrl);
3133 	if (ret < 0)
3134 		return ret;
3135 
3136 	ret = nvme_configure_timestamp(ctrl);
3137 	if (ret < 0)
3138 		return ret;
3139 
3140 	ret = nvme_configure_host_options(ctrl);
3141 	if (ret < 0)
3142 		return ret;
3143 
3144 	if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3145 		ret = nvme_hwmon_init(ctrl);
3146 		if (ret < 0)
3147 			return ret;
3148 	}
3149 
3150 	ctrl->identified = true;
3151 
3152 	return 0;
3153 }
3154 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3155 
3156 static int nvme_dev_open(struct inode *inode, struct file *file)
3157 {
3158 	struct nvme_ctrl *ctrl =
3159 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3160 
3161 	switch (ctrl->state) {
3162 	case NVME_CTRL_LIVE:
3163 		break;
3164 	default:
3165 		return -EWOULDBLOCK;
3166 	}
3167 
3168 	nvme_get_ctrl(ctrl);
3169 	if (!try_module_get(ctrl->ops->module)) {
3170 		nvme_put_ctrl(ctrl);
3171 		return -EINVAL;
3172 	}
3173 
3174 	file->private_data = ctrl;
3175 	return 0;
3176 }
3177 
3178 static int nvme_dev_release(struct inode *inode, struct file *file)
3179 {
3180 	struct nvme_ctrl *ctrl =
3181 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3182 
3183 	module_put(ctrl->ops->module);
3184 	nvme_put_ctrl(ctrl);
3185 	return 0;
3186 }
3187 
3188 static const struct file_operations nvme_dev_fops = {
3189 	.owner		= THIS_MODULE,
3190 	.open		= nvme_dev_open,
3191 	.release	= nvme_dev_release,
3192 	.unlocked_ioctl	= nvme_dev_ioctl,
3193 	.compat_ioctl	= compat_ptr_ioctl,
3194 	.uring_cmd	= nvme_dev_uring_cmd,
3195 };
3196 
3197 static ssize_t nvme_sysfs_reset(struct device *dev,
3198 				struct device_attribute *attr, const char *buf,
3199 				size_t count)
3200 {
3201 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3202 	int ret;
3203 
3204 	ret = nvme_reset_ctrl_sync(ctrl);
3205 	if (ret < 0)
3206 		return ret;
3207 	return count;
3208 }
3209 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3210 
3211 static ssize_t nvme_sysfs_rescan(struct device *dev,
3212 				struct device_attribute *attr, const char *buf,
3213 				size_t count)
3214 {
3215 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3216 
3217 	nvme_queue_scan(ctrl);
3218 	return count;
3219 }
3220 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3221 
3222 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3223 {
3224 	struct gendisk *disk = dev_to_disk(dev);
3225 
3226 	if (disk->fops == &nvme_bdev_ops)
3227 		return nvme_get_ns_from_dev(dev)->head;
3228 	else
3229 		return disk->private_data;
3230 }
3231 
3232 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3233 		char *buf)
3234 {
3235 	struct nvme_ns_head *head = dev_to_ns_head(dev);
3236 	struct nvme_ns_ids *ids = &head->ids;
3237 	struct nvme_subsystem *subsys = head->subsys;
3238 	int serial_len = sizeof(subsys->serial);
3239 	int model_len = sizeof(subsys->model);
3240 
3241 	if (!uuid_is_null(&ids->uuid))
3242 		return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3243 
3244 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3245 		return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3246 
3247 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3248 		return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3249 
3250 	while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3251 				  subsys->serial[serial_len - 1] == '\0'))
3252 		serial_len--;
3253 	while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3254 				 subsys->model[model_len - 1] == '\0'))
3255 		model_len--;
3256 
3257 	return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3258 		serial_len, subsys->serial, model_len, subsys->model,
3259 		head->ns_id);
3260 }
3261 static DEVICE_ATTR_RO(wwid);
3262 
3263 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3264 		char *buf)
3265 {
3266 	return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3267 }
3268 static DEVICE_ATTR_RO(nguid);
3269 
3270 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3271 		char *buf)
3272 {
3273 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3274 
3275 	/* For backward compatibility expose the NGUID to userspace if
3276 	 * we have no UUID set
3277 	 */
3278 	if (uuid_is_null(&ids->uuid)) {
3279 		printk_ratelimited(KERN_WARNING
3280 				   "No UUID available providing old NGUID\n");
3281 		return sysfs_emit(buf, "%pU\n", ids->nguid);
3282 	}
3283 	return sysfs_emit(buf, "%pU\n", &ids->uuid);
3284 }
3285 static DEVICE_ATTR_RO(uuid);
3286 
3287 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3288 		char *buf)
3289 {
3290 	return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3291 }
3292 static DEVICE_ATTR_RO(eui);
3293 
3294 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3295 		char *buf)
3296 {
3297 	return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3298 }
3299 static DEVICE_ATTR_RO(nsid);
3300 
3301 static struct attribute *nvme_ns_id_attrs[] = {
3302 	&dev_attr_wwid.attr,
3303 	&dev_attr_uuid.attr,
3304 	&dev_attr_nguid.attr,
3305 	&dev_attr_eui.attr,
3306 	&dev_attr_nsid.attr,
3307 #ifdef CONFIG_NVME_MULTIPATH
3308 	&dev_attr_ana_grpid.attr,
3309 	&dev_attr_ana_state.attr,
3310 #endif
3311 	NULL,
3312 };
3313 
3314 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3315 		struct attribute *a, int n)
3316 {
3317 	struct device *dev = container_of(kobj, struct device, kobj);
3318 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3319 
3320 	if (a == &dev_attr_uuid.attr) {
3321 		if (uuid_is_null(&ids->uuid) &&
3322 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3323 			return 0;
3324 	}
3325 	if (a == &dev_attr_nguid.attr) {
3326 		if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3327 			return 0;
3328 	}
3329 	if (a == &dev_attr_eui.attr) {
3330 		if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3331 			return 0;
3332 	}
3333 #ifdef CONFIG_NVME_MULTIPATH
3334 	if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3335 		if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3336 			return 0;
3337 		if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3338 			return 0;
3339 	}
3340 #endif
3341 	return a->mode;
3342 }
3343 
3344 static const struct attribute_group nvme_ns_id_attr_group = {
3345 	.attrs		= nvme_ns_id_attrs,
3346 	.is_visible	= nvme_ns_id_attrs_are_visible,
3347 };
3348 
3349 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3350 	&nvme_ns_id_attr_group,
3351 	NULL,
3352 };
3353 
3354 #define nvme_show_str_function(field)						\
3355 static ssize_t  field##_show(struct device *dev,				\
3356 			    struct device_attribute *attr, char *buf)		\
3357 {										\
3358         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3359         return sysfs_emit(buf, "%.*s\n",					\
3360 		(int)sizeof(ctrl->subsys->field), ctrl->subsys->field);		\
3361 }										\
3362 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3363 
3364 nvme_show_str_function(model);
3365 nvme_show_str_function(serial);
3366 nvme_show_str_function(firmware_rev);
3367 
3368 #define nvme_show_int_function(field)						\
3369 static ssize_t  field##_show(struct device *dev,				\
3370 			    struct device_attribute *attr, char *buf)		\
3371 {										\
3372         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3373         return sysfs_emit(buf, "%d\n", ctrl->field);				\
3374 }										\
3375 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3376 
3377 nvme_show_int_function(cntlid);
3378 nvme_show_int_function(numa_node);
3379 nvme_show_int_function(queue_count);
3380 nvme_show_int_function(sqsize);
3381 nvme_show_int_function(kato);
3382 
3383 static ssize_t nvme_sysfs_delete(struct device *dev,
3384 				struct device_attribute *attr, const char *buf,
3385 				size_t count)
3386 {
3387 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3388 
3389 	if (device_remove_file_self(dev, attr))
3390 		nvme_delete_ctrl_sync(ctrl);
3391 	return count;
3392 }
3393 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3394 
3395 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3396 					 struct device_attribute *attr,
3397 					 char *buf)
3398 {
3399 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3400 
3401 	return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3402 }
3403 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3404 
3405 static ssize_t nvme_sysfs_show_state(struct device *dev,
3406 				     struct device_attribute *attr,
3407 				     char *buf)
3408 {
3409 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3410 	static const char *const state_name[] = {
3411 		[NVME_CTRL_NEW]		= "new",
3412 		[NVME_CTRL_LIVE]	= "live",
3413 		[NVME_CTRL_RESETTING]	= "resetting",
3414 		[NVME_CTRL_CONNECTING]	= "connecting",
3415 		[NVME_CTRL_DELETING]	= "deleting",
3416 		[NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3417 		[NVME_CTRL_DEAD]	= "dead",
3418 	};
3419 
3420 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3421 	    state_name[ctrl->state])
3422 		return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3423 
3424 	return sysfs_emit(buf, "unknown state\n");
3425 }
3426 
3427 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3428 
3429 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3430 					 struct device_attribute *attr,
3431 					 char *buf)
3432 {
3433 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3434 
3435 	return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3436 }
3437 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3438 
3439 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3440 					struct device_attribute *attr,
3441 					char *buf)
3442 {
3443 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3444 
3445 	return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3446 }
3447 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3448 
3449 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3450 					struct device_attribute *attr,
3451 					char *buf)
3452 {
3453 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3454 
3455 	return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3456 }
3457 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3458 
3459 static ssize_t nvme_sysfs_show_address(struct device *dev,
3460 					 struct device_attribute *attr,
3461 					 char *buf)
3462 {
3463 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3464 
3465 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3466 }
3467 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3468 
3469 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3470 		struct device_attribute *attr, char *buf)
3471 {
3472 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3473 	struct nvmf_ctrl_options *opts = ctrl->opts;
3474 
3475 	if (ctrl->opts->max_reconnects == -1)
3476 		return sysfs_emit(buf, "off\n");
3477 	return sysfs_emit(buf, "%d\n",
3478 			  opts->max_reconnects * opts->reconnect_delay);
3479 }
3480 
3481 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3482 		struct device_attribute *attr, const char *buf, size_t count)
3483 {
3484 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3485 	struct nvmf_ctrl_options *opts = ctrl->opts;
3486 	int ctrl_loss_tmo, err;
3487 
3488 	err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3489 	if (err)
3490 		return -EINVAL;
3491 
3492 	if (ctrl_loss_tmo < 0)
3493 		opts->max_reconnects = -1;
3494 	else
3495 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3496 						opts->reconnect_delay);
3497 	return count;
3498 }
3499 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3500 	nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3501 
3502 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3503 		struct device_attribute *attr, char *buf)
3504 {
3505 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3506 
3507 	if (ctrl->opts->reconnect_delay == -1)
3508 		return sysfs_emit(buf, "off\n");
3509 	return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3510 }
3511 
3512 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3513 		struct device_attribute *attr, const char *buf, size_t count)
3514 {
3515 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3516 	unsigned int v;
3517 	int err;
3518 
3519 	err = kstrtou32(buf, 10, &v);
3520 	if (err)
3521 		return err;
3522 
3523 	ctrl->opts->reconnect_delay = v;
3524 	return count;
3525 }
3526 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3527 	nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3528 
3529 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3530 		struct device_attribute *attr, char *buf)
3531 {
3532 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3533 
3534 	if (ctrl->opts->fast_io_fail_tmo == -1)
3535 		return sysfs_emit(buf, "off\n");
3536 	return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3537 }
3538 
3539 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3540 		struct device_attribute *attr, const char *buf, size_t count)
3541 {
3542 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3543 	struct nvmf_ctrl_options *opts = ctrl->opts;
3544 	int fast_io_fail_tmo, err;
3545 
3546 	err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3547 	if (err)
3548 		return -EINVAL;
3549 
3550 	if (fast_io_fail_tmo < 0)
3551 		opts->fast_io_fail_tmo = -1;
3552 	else
3553 		opts->fast_io_fail_tmo = fast_io_fail_tmo;
3554 	return count;
3555 }
3556 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3557 	nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3558 
3559 static ssize_t cntrltype_show(struct device *dev,
3560 			      struct device_attribute *attr, char *buf)
3561 {
3562 	static const char * const type[] = {
3563 		[NVME_CTRL_IO] = "io\n",
3564 		[NVME_CTRL_DISC] = "discovery\n",
3565 		[NVME_CTRL_ADMIN] = "admin\n",
3566 	};
3567 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3568 
3569 	if (ctrl->cntrltype > NVME_CTRL_ADMIN || !type[ctrl->cntrltype])
3570 		return sysfs_emit(buf, "reserved\n");
3571 
3572 	return sysfs_emit(buf, type[ctrl->cntrltype]);
3573 }
3574 static DEVICE_ATTR_RO(cntrltype);
3575 
3576 static ssize_t dctype_show(struct device *dev,
3577 			   struct device_attribute *attr, char *buf)
3578 {
3579 	static const char * const type[] = {
3580 		[NVME_DCTYPE_NOT_REPORTED] = "none\n",
3581 		[NVME_DCTYPE_DDC] = "ddc\n",
3582 		[NVME_DCTYPE_CDC] = "cdc\n",
3583 	};
3584 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3585 
3586 	if (ctrl->dctype > NVME_DCTYPE_CDC || !type[ctrl->dctype])
3587 		return sysfs_emit(buf, "reserved\n");
3588 
3589 	return sysfs_emit(buf, type[ctrl->dctype]);
3590 }
3591 static DEVICE_ATTR_RO(dctype);
3592 
3593 static struct attribute *nvme_dev_attrs[] = {
3594 	&dev_attr_reset_controller.attr,
3595 	&dev_attr_rescan_controller.attr,
3596 	&dev_attr_model.attr,
3597 	&dev_attr_serial.attr,
3598 	&dev_attr_firmware_rev.attr,
3599 	&dev_attr_cntlid.attr,
3600 	&dev_attr_delete_controller.attr,
3601 	&dev_attr_transport.attr,
3602 	&dev_attr_subsysnqn.attr,
3603 	&dev_attr_address.attr,
3604 	&dev_attr_state.attr,
3605 	&dev_attr_numa_node.attr,
3606 	&dev_attr_queue_count.attr,
3607 	&dev_attr_sqsize.attr,
3608 	&dev_attr_hostnqn.attr,
3609 	&dev_attr_hostid.attr,
3610 	&dev_attr_ctrl_loss_tmo.attr,
3611 	&dev_attr_reconnect_delay.attr,
3612 	&dev_attr_fast_io_fail_tmo.attr,
3613 	&dev_attr_kato.attr,
3614 	&dev_attr_cntrltype.attr,
3615 	&dev_attr_dctype.attr,
3616 	NULL
3617 };
3618 
3619 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3620 		struct attribute *a, int n)
3621 {
3622 	struct device *dev = container_of(kobj, struct device, kobj);
3623 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3624 
3625 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3626 		return 0;
3627 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3628 		return 0;
3629 	if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3630 		return 0;
3631 	if (a == &dev_attr_hostid.attr && !ctrl->opts)
3632 		return 0;
3633 	if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3634 		return 0;
3635 	if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3636 		return 0;
3637 	if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3638 		return 0;
3639 
3640 	return a->mode;
3641 }
3642 
3643 static const struct attribute_group nvme_dev_attrs_group = {
3644 	.attrs		= nvme_dev_attrs,
3645 	.is_visible	= nvme_dev_attrs_are_visible,
3646 };
3647 
3648 static const struct attribute_group *nvme_dev_attr_groups[] = {
3649 	&nvme_dev_attrs_group,
3650 	NULL,
3651 };
3652 
3653 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3654 		unsigned nsid)
3655 {
3656 	struct nvme_ns_head *h;
3657 
3658 	lockdep_assert_held(&ctrl->subsys->lock);
3659 
3660 	list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3661 		/*
3662 		 * Private namespaces can share NSIDs under some conditions.
3663 		 * In that case we can't use the same ns_head for namespaces
3664 		 * with the same NSID.
3665 		 */
3666 		if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3667 			continue;
3668 		if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3669 			return h;
3670 	}
3671 
3672 	return NULL;
3673 }
3674 
3675 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3676 		struct nvme_ns_ids *ids)
3677 {
3678 	bool has_uuid = !uuid_is_null(&ids->uuid);
3679 	bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3680 	bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3681 	struct nvme_ns_head *h;
3682 
3683 	lockdep_assert_held(&subsys->lock);
3684 
3685 	list_for_each_entry(h, &subsys->nsheads, entry) {
3686 		if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3687 			return -EINVAL;
3688 		if (has_nguid &&
3689 		    memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3690 			return -EINVAL;
3691 		if (has_eui64 &&
3692 		    memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3693 			return -EINVAL;
3694 	}
3695 
3696 	return 0;
3697 }
3698 
3699 static void nvme_cdev_rel(struct device *dev)
3700 {
3701 	ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3702 }
3703 
3704 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3705 {
3706 	cdev_device_del(cdev, cdev_device);
3707 	put_device(cdev_device);
3708 }
3709 
3710 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3711 		const struct file_operations *fops, struct module *owner)
3712 {
3713 	int minor, ret;
3714 
3715 	minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3716 	if (minor < 0)
3717 		return minor;
3718 	cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3719 	cdev_device->class = nvme_ns_chr_class;
3720 	cdev_device->release = nvme_cdev_rel;
3721 	device_initialize(cdev_device);
3722 	cdev_init(cdev, fops);
3723 	cdev->owner = owner;
3724 	ret = cdev_device_add(cdev, cdev_device);
3725 	if (ret)
3726 		put_device(cdev_device);
3727 
3728 	return ret;
3729 }
3730 
3731 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3732 {
3733 	return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3734 }
3735 
3736 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3737 {
3738 	nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3739 	return 0;
3740 }
3741 
3742 static const struct file_operations nvme_ns_chr_fops = {
3743 	.owner		= THIS_MODULE,
3744 	.open		= nvme_ns_chr_open,
3745 	.release	= nvme_ns_chr_release,
3746 	.unlocked_ioctl	= nvme_ns_chr_ioctl,
3747 	.compat_ioctl	= compat_ptr_ioctl,
3748 	.uring_cmd	= nvme_ns_chr_uring_cmd,
3749 };
3750 
3751 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3752 {
3753 	int ret;
3754 
3755 	ns->cdev_device.parent = ns->ctrl->device;
3756 	ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3757 			   ns->ctrl->instance, ns->head->instance);
3758 	if (ret)
3759 		return ret;
3760 
3761 	return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3762 			     ns->ctrl->ops->module);
3763 }
3764 
3765 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3766 		unsigned nsid, struct nvme_ns_ids *ids)
3767 {
3768 	struct nvme_ns_head *head;
3769 	size_t size = sizeof(*head);
3770 	int ret = -ENOMEM;
3771 
3772 #ifdef CONFIG_NVME_MULTIPATH
3773 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3774 #endif
3775 
3776 	head = kzalloc(size, GFP_KERNEL);
3777 	if (!head)
3778 		goto out;
3779 	ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3780 	if (ret < 0)
3781 		goto out_free_head;
3782 	head->instance = ret;
3783 	INIT_LIST_HEAD(&head->list);
3784 	ret = init_srcu_struct(&head->srcu);
3785 	if (ret)
3786 		goto out_ida_remove;
3787 	head->subsys = ctrl->subsys;
3788 	head->ns_id = nsid;
3789 	head->ids = *ids;
3790 	kref_init(&head->ref);
3791 
3792 	if (head->ids.csi) {
3793 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3794 		if (ret)
3795 			goto out_cleanup_srcu;
3796 	} else
3797 		head->effects = ctrl->effects;
3798 
3799 	ret = nvme_mpath_alloc_disk(ctrl, head);
3800 	if (ret)
3801 		goto out_cleanup_srcu;
3802 
3803 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3804 
3805 	kref_get(&ctrl->subsys->ref);
3806 
3807 	return head;
3808 out_cleanup_srcu:
3809 	cleanup_srcu_struct(&head->srcu);
3810 out_ida_remove:
3811 	ida_free(&ctrl->subsys->ns_ida, head->instance);
3812 out_free_head:
3813 	kfree(head);
3814 out:
3815 	if (ret > 0)
3816 		ret = blk_status_to_errno(nvme_error_status(ret));
3817 	return ERR_PTR(ret);
3818 }
3819 
3820 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3821 		struct nvme_ns_ids *ids)
3822 {
3823 	struct nvme_subsystem *s;
3824 	int ret = 0;
3825 
3826 	/*
3827 	 * Note that this check is racy as we try to avoid holding the global
3828 	 * lock over the whole ns_head creation.  But it is only intended as
3829 	 * a sanity check anyway.
3830 	 */
3831 	mutex_lock(&nvme_subsystems_lock);
3832 	list_for_each_entry(s, &nvme_subsystems, entry) {
3833 		if (s == this)
3834 			continue;
3835 		mutex_lock(&s->lock);
3836 		ret = nvme_subsys_check_duplicate_ids(s, ids);
3837 		mutex_unlock(&s->lock);
3838 		if (ret)
3839 			break;
3840 	}
3841 	mutex_unlock(&nvme_subsystems_lock);
3842 
3843 	return ret;
3844 }
3845 
3846 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3847 		struct nvme_ns_ids *ids, bool is_shared)
3848 {
3849 	struct nvme_ctrl *ctrl = ns->ctrl;
3850 	struct nvme_ns_head *head = NULL;
3851 	int ret;
3852 
3853 	ret = nvme_global_check_duplicate_ids(ctrl->subsys, ids);
3854 	if (ret) {
3855 		dev_err(ctrl->device,
3856 			"globally duplicate IDs for nsid %d\n", nsid);
3857 		return ret;
3858 	}
3859 
3860 	mutex_lock(&ctrl->subsys->lock);
3861 	head = nvme_find_ns_head(ctrl, nsid);
3862 	if (!head) {
3863 		ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, ids);
3864 		if (ret) {
3865 			dev_err(ctrl->device,
3866 				"duplicate IDs in subsystem for nsid %d\n",
3867 				nsid);
3868 			goto out_unlock;
3869 		}
3870 		head = nvme_alloc_ns_head(ctrl, nsid, ids);
3871 		if (IS_ERR(head)) {
3872 			ret = PTR_ERR(head);
3873 			goto out_unlock;
3874 		}
3875 		head->shared = is_shared;
3876 	} else {
3877 		ret = -EINVAL;
3878 		if (!is_shared || !head->shared) {
3879 			dev_err(ctrl->device,
3880 				"Duplicate unshared namespace %d\n", nsid);
3881 			goto out_put_ns_head;
3882 		}
3883 		if (!nvme_ns_ids_equal(&head->ids, ids)) {
3884 			dev_err(ctrl->device,
3885 				"IDs don't match for shared namespace %d\n",
3886 					nsid);
3887 			goto out_put_ns_head;
3888 		}
3889 
3890 		if (!multipath && !list_empty(&head->list)) {
3891 			dev_warn(ctrl->device,
3892 				"Found shared namespace %d, but multipathing not supported.\n",
3893 				nsid);
3894 			dev_warn_once(ctrl->device,
3895 				"Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
3896 		}
3897 	}
3898 
3899 	list_add_tail_rcu(&ns->siblings, &head->list);
3900 	ns->head = head;
3901 	mutex_unlock(&ctrl->subsys->lock);
3902 	return 0;
3903 
3904 out_put_ns_head:
3905 	nvme_put_ns_head(head);
3906 out_unlock:
3907 	mutex_unlock(&ctrl->subsys->lock);
3908 	return ret;
3909 }
3910 
3911 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3912 {
3913 	struct nvme_ns *ns, *ret = NULL;
3914 
3915 	down_read(&ctrl->namespaces_rwsem);
3916 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3917 		if (ns->head->ns_id == nsid) {
3918 			if (!nvme_get_ns(ns))
3919 				continue;
3920 			ret = ns;
3921 			break;
3922 		}
3923 		if (ns->head->ns_id > nsid)
3924 			break;
3925 	}
3926 	up_read(&ctrl->namespaces_rwsem);
3927 	return ret;
3928 }
3929 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3930 
3931 /*
3932  * Add the namespace to the controller list while keeping the list ordered.
3933  */
3934 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3935 {
3936 	struct nvme_ns *tmp;
3937 
3938 	list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3939 		if (tmp->head->ns_id < ns->head->ns_id) {
3940 			list_add(&ns->list, &tmp->list);
3941 			return;
3942 		}
3943 	}
3944 	list_add(&ns->list, &ns->ctrl->namespaces);
3945 }
3946 
3947 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3948 		struct nvme_ns_ids *ids)
3949 {
3950 	struct nvme_ns *ns;
3951 	struct gendisk *disk;
3952 	struct nvme_id_ns *id;
3953 	int node = ctrl->numa_node;
3954 
3955 	if (nvme_identify_ns(ctrl, nsid, ids, &id))
3956 		return;
3957 
3958 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3959 	if (!ns)
3960 		goto out_free_id;
3961 
3962 	disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3963 	if (IS_ERR(disk))
3964 		goto out_free_ns;
3965 	disk->fops = &nvme_bdev_ops;
3966 	disk->private_data = ns;
3967 
3968 	ns->disk = disk;
3969 	ns->queue = disk->queue;
3970 
3971 	if (ctrl->opts && ctrl->opts->data_digest)
3972 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3973 
3974 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3975 	if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3976 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3977 
3978 	ns->ctrl = ctrl;
3979 	kref_init(&ns->kref);
3980 
3981 	if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3982 		goto out_cleanup_disk;
3983 
3984 	/*
3985 	 * If multipathing is enabled, the device name for all disks and not
3986 	 * just those that represent shared namespaces needs to be based on the
3987 	 * subsystem instance.  Using the controller instance for private
3988 	 * namespaces could lead to naming collisions between shared and private
3989 	 * namespaces if they don't use a common numbering scheme.
3990 	 *
3991 	 * If multipathing is not enabled, disk names must use the controller
3992 	 * instance as shared namespaces will show up as multiple block
3993 	 * devices.
3994 	 */
3995 	if (ns->head->disk) {
3996 		sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
3997 			ctrl->instance, ns->head->instance);
3998 		disk->flags |= GENHD_FL_HIDDEN;
3999 	} else if (multipath) {
4000 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
4001 			ns->head->instance);
4002 	} else {
4003 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
4004 			ns->head->instance);
4005 	}
4006 
4007 	if (nvme_update_ns_info(ns, id))
4008 		goto out_unlink_ns;
4009 
4010 	down_write(&ctrl->namespaces_rwsem);
4011 	nvme_ns_add_to_ctrl_list(ns);
4012 	up_write(&ctrl->namespaces_rwsem);
4013 	nvme_get_ctrl(ctrl);
4014 
4015 	if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
4016 		goto out_cleanup_ns_from_list;
4017 
4018 	if (!nvme_ns_head_multipath(ns->head))
4019 		nvme_add_ns_cdev(ns);
4020 
4021 	nvme_mpath_add_disk(ns, id);
4022 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
4023 	kfree(id);
4024 
4025 	return;
4026 
4027  out_cleanup_ns_from_list:
4028 	nvme_put_ctrl(ctrl);
4029 	down_write(&ctrl->namespaces_rwsem);
4030 	list_del_init(&ns->list);
4031 	up_write(&ctrl->namespaces_rwsem);
4032  out_unlink_ns:
4033 	mutex_lock(&ctrl->subsys->lock);
4034 	list_del_rcu(&ns->siblings);
4035 	if (list_empty(&ns->head->list))
4036 		list_del_init(&ns->head->entry);
4037 	mutex_unlock(&ctrl->subsys->lock);
4038 	nvme_put_ns_head(ns->head);
4039  out_cleanup_disk:
4040 	blk_cleanup_disk(disk);
4041  out_free_ns:
4042 	kfree(ns);
4043  out_free_id:
4044 	kfree(id);
4045 }
4046 
4047 static void nvme_ns_remove(struct nvme_ns *ns)
4048 {
4049 	bool last_path = false;
4050 
4051 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4052 		return;
4053 
4054 	clear_bit(NVME_NS_READY, &ns->flags);
4055 	set_capacity(ns->disk, 0);
4056 	nvme_fault_inject_fini(&ns->fault_inject);
4057 
4058 	/*
4059 	 * Ensure that !NVME_NS_READY is seen by other threads to prevent
4060 	 * this ns going back into current_path.
4061 	 */
4062 	synchronize_srcu(&ns->head->srcu);
4063 
4064 	/* wait for concurrent submissions */
4065 	if (nvme_mpath_clear_current_path(ns))
4066 		synchronize_srcu(&ns->head->srcu);
4067 
4068 	mutex_lock(&ns->ctrl->subsys->lock);
4069 	list_del_rcu(&ns->siblings);
4070 	if (list_empty(&ns->head->list)) {
4071 		list_del_init(&ns->head->entry);
4072 		last_path = true;
4073 	}
4074 	mutex_unlock(&ns->ctrl->subsys->lock);
4075 
4076 	/* guarantee not available in head->list */
4077 	synchronize_rcu();
4078 
4079 	if (!nvme_ns_head_multipath(ns->head))
4080 		nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4081 	del_gendisk(ns->disk);
4082 	blk_cleanup_queue(ns->queue);
4083 
4084 	down_write(&ns->ctrl->namespaces_rwsem);
4085 	list_del_init(&ns->list);
4086 	up_write(&ns->ctrl->namespaces_rwsem);
4087 
4088 	if (last_path)
4089 		nvme_mpath_shutdown_disk(ns->head);
4090 	nvme_put_ns(ns);
4091 }
4092 
4093 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4094 {
4095 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4096 
4097 	if (ns) {
4098 		nvme_ns_remove(ns);
4099 		nvme_put_ns(ns);
4100 	}
4101 }
4102 
4103 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4104 {
4105 	struct nvme_id_ns *id;
4106 	int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4107 
4108 	if (test_bit(NVME_NS_DEAD, &ns->flags))
4109 		goto out;
4110 
4111 	ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4112 	if (ret)
4113 		goto out;
4114 
4115 	ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4116 	if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4117 		dev_err(ns->ctrl->device,
4118 			"identifiers changed for nsid %d\n", ns->head->ns_id);
4119 		goto out_free_id;
4120 	}
4121 
4122 	ret = nvme_update_ns_info(ns, id);
4123 
4124 out_free_id:
4125 	kfree(id);
4126 out:
4127 	/*
4128 	 * Only remove the namespace if we got a fatal error back from the
4129 	 * device, otherwise ignore the error and just move on.
4130 	 *
4131 	 * TODO: we should probably schedule a delayed retry here.
4132 	 */
4133 	if (ret > 0 && (ret & NVME_SC_DNR))
4134 		nvme_ns_remove(ns);
4135 }
4136 
4137 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4138 {
4139 	struct nvme_ns_ids ids = { };
4140 	struct nvme_id_ns_cs_indep *id;
4141 	struct nvme_ns *ns;
4142 	bool ready = true;
4143 
4144 	if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4145 		return;
4146 
4147 	/*
4148 	 * Check if the namespace is ready.  If not ignore it, we will get an
4149 	 * AEN once it becomes ready and restart the scan.
4150 	 */
4151 	if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) &&
4152 	    !nvme_identify_ns_cs_indep(ctrl, nsid, &id)) {
4153 		ready = id->nstat & NVME_NSTAT_NRDY;
4154 		kfree(id);
4155 	}
4156 
4157 	if (!ready)
4158 		return;
4159 
4160 	ns = nvme_find_get_ns(ctrl, nsid);
4161 	if (ns) {
4162 		nvme_validate_ns(ns, &ids);
4163 		nvme_put_ns(ns);
4164 		return;
4165 	}
4166 
4167 	switch (ids.csi) {
4168 	case NVME_CSI_NVM:
4169 		nvme_alloc_ns(ctrl, nsid, &ids);
4170 		break;
4171 	case NVME_CSI_ZNS:
4172 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4173 			dev_warn(ctrl->device,
4174 				"nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4175 				nsid);
4176 			break;
4177 		}
4178 		if (!nvme_multi_css(ctrl)) {
4179 			dev_warn(ctrl->device,
4180 				"command set not reported for nsid: %d\n",
4181 				nsid);
4182 			break;
4183 		}
4184 		nvme_alloc_ns(ctrl, nsid, &ids);
4185 		break;
4186 	default:
4187 		dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4188 			ids.csi, nsid);
4189 		break;
4190 	}
4191 }
4192 
4193 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4194 					unsigned nsid)
4195 {
4196 	struct nvme_ns *ns, *next;
4197 	LIST_HEAD(rm_list);
4198 
4199 	down_write(&ctrl->namespaces_rwsem);
4200 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4201 		if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4202 			list_move_tail(&ns->list, &rm_list);
4203 	}
4204 	up_write(&ctrl->namespaces_rwsem);
4205 
4206 	list_for_each_entry_safe(ns, next, &rm_list, list)
4207 		nvme_ns_remove(ns);
4208 
4209 }
4210 
4211 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4212 {
4213 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4214 	__le32 *ns_list;
4215 	u32 prev = 0;
4216 	int ret = 0, i;
4217 
4218 	if (nvme_ctrl_limited_cns(ctrl))
4219 		return -EOPNOTSUPP;
4220 
4221 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4222 	if (!ns_list)
4223 		return -ENOMEM;
4224 
4225 	for (;;) {
4226 		struct nvme_command cmd = {
4227 			.identify.opcode	= nvme_admin_identify,
4228 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
4229 			.identify.nsid		= cpu_to_le32(prev),
4230 		};
4231 
4232 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4233 					    NVME_IDENTIFY_DATA_SIZE);
4234 		if (ret) {
4235 			dev_warn(ctrl->device,
4236 				"Identify NS List failed (status=0x%x)\n", ret);
4237 			goto free;
4238 		}
4239 
4240 		for (i = 0; i < nr_entries; i++) {
4241 			u32 nsid = le32_to_cpu(ns_list[i]);
4242 
4243 			if (!nsid)	/* end of the list? */
4244 				goto out;
4245 			nvme_validate_or_alloc_ns(ctrl, nsid);
4246 			while (++prev < nsid)
4247 				nvme_ns_remove_by_nsid(ctrl, prev);
4248 		}
4249 	}
4250  out:
4251 	nvme_remove_invalid_namespaces(ctrl, prev);
4252  free:
4253 	kfree(ns_list);
4254 	return ret;
4255 }
4256 
4257 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4258 {
4259 	struct nvme_id_ctrl *id;
4260 	u32 nn, i;
4261 
4262 	if (nvme_identify_ctrl(ctrl, &id))
4263 		return;
4264 	nn = le32_to_cpu(id->nn);
4265 	kfree(id);
4266 
4267 	for (i = 1; i <= nn; i++)
4268 		nvme_validate_or_alloc_ns(ctrl, i);
4269 
4270 	nvme_remove_invalid_namespaces(ctrl, nn);
4271 }
4272 
4273 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4274 {
4275 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4276 	__le32 *log;
4277 	int error;
4278 
4279 	log = kzalloc(log_size, GFP_KERNEL);
4280 	if (!log)
4281 		return;
4282 
4283 	/*
4284 	 * We need to read the log to clear the AEN, but we don't want to rely
4285 	 * on it for the changed namespace information as userspace could have
4286 	 * raced with us in reading the log page, which could cause us to miss
4287 	 * updates.
4288 	 */
4289 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4290 			NVME_CSI_NVM, log, log_size, 0);
4291 	if (error)
4292 		dev_warn(ctrl->device,
4293 			"reading changed ns log failed: %d\n", error);
4294 
4295 	kfree(log);
4296 }
4297 
4298 static void nvme_scan_work(struct work_struct *work)
4299 {
4300 	struct nvme_ctrl *ctrl =
4301 		container_of(work, struct nvme_ctrl, scan_work);
4302 	int ret;
4303 
4304 	/* No tagset on a live ctrl means IO queues could not created */
4305 	if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4306 		return;
4307 
4308 	/*
4309 	 * Identify controller limits can change at controller reset due to
4310 	 * new firmware download, even though it is not common we cannot ignore
4311 	 * such scenario. Controller's non-mdts limits are reported in the unit
4312 	 * of logical blocks that is dependent on the format of attached
4313 	 * namespace. Hence re-read the limits at the time of ns allocation.
4314 	 */
4315 	ret = nvme_init_non_mdts_limits(ctrl);
4316 	if (ret < 0) {
4317 		dev_warn(ctrl->device,
4318 			"reading non-mdts-limits failed: %d\n", ret);
4319 		return;
4320 	}
4321 
4322 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4323 		dev_info(ctrl->device, "rescanning namespaces.\n");
4324 		nvme_clear_changed_ns_log(ctrl);
4325 	}
4326 
4327 	mutex_lock(&ctrl->scan_lock);
4328 	if (nvme_scan_ns_list(ctrl) != 0)
4329 		nvme_scan_ns_sequential(ctrl);
4330 	mutex_unlock(&ctrl->scan_lock);
4331 }
4332 
4333 /*
4334  * This function iterates the namespace list unlocked to allow recovery from
4335  * controller failure. It is up to the caller to ensure the namespace list is
4336  * not modified by scan work while this function is executing.
4337  */
4338 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4339 {
4340 	struct nvme_ns *ns, *next;
4341 	LIST_HEAD(ns_list);
4342 
4343 	/*
4344 	 * make sure to requeue I/O to all namespaces as these
4345 	 * might result from the scan itself and must complete
4346 	 * for the scan_work to make progress
4347 	 */
4348 	nvme_mpath_clear_ctrl_paths(ctrl);
4349 
4350 	/* prevent racing with ns scanning */
4351 	flush_work(&ctrl->scan_work);
4352 
4353 	/*
4354 	 * The dead states indicates the controller was not gracefully
4355 	 * disconnected. In that case, we won't be able to flush any data while
4356 	 * removing the namespaces' disks; fail all the queues now to avoid
4357 	 * potentially having to clean up the failed sync later.
4358 	 */
4359 	if (ctrl->state == NVME_CTRL_DEAD)
4360 		nvme_kill_queues(ctrl);
4361 
4362 	/* this is a no-op when called from the controller reset handler */
4363 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4364 
4365 	down_write(&ctrl->namespaces_rwsem);
4366 	list_splice_init(&ctrl->namespaces, &ns_list);
4367 	up_write(&ctrl->namespaces_rwsem);
4368 
4369 	list_for_each_entry_safe(ns, next, &ns_list, list)
4370 		nvme_ns_remove(ns);
4371 }
4372 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4373 
4374 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4375 {
4376 	struct nvme_ctrl *ctrl =
4377 		container_of(dev, struct nvme_ctrl, ctrl_device);
4378 	struct nvmf_ctrl_options *opts = ctrl->opts;
4379 	int ret;
4380 
4381 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4382 	if (ret)
4383 		return ret;
4384 
4385 	if (opts) {
4386 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4387 		if (ret)
4388 			return ret;
4389 
4390 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4391 				opts->trsvcid ?: "none");
4392 		if (ret)
4393 			return ret;
4394 
4395 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4396 				opts->host_traddr ?: "none");
4397 		if (ret)
4398 			return ret;
4399 
4400 		ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4401 				opts->host_iface ?: "none");
4402 	}
4403 	return ret;
4404 }
4405 
4406 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4407 {
4408 	char *envp[2] = { envdata, NULL };
4409 
4410 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4411 }
4412 
4413 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4414 {
4415 	char *envp[2] = { NULL, NULL };
4416 	u32 aen_result = ctrl->aen_result;
4417 
4418 	ctrl->aen_result = 0;
4419 	if (!aen_result)
4420 		return;
4421 
4422 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4423 	if (!envp[0])
4424 		return;
4425 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4426 	kfree(envp[0]);
4427 }
4428 
4429 static void nvme_async_event_work(struct work_struct *work)
4430 {
4431 	struct nvme_ctrl *ctrl =
4432 		container_of(work, struct nvme_ctrl, async_event_work);
4433 
4434 	nvme_aen_uevent(ctrl);
4435 
4436 	/*
4437 	 * The transport drivers must guarantee AER submission here is safe by
4438 	 * flushing ctrl async_event_work after changing the controller state
4439 	 * from LIVE and before freeing the admin queue.
4440 	*/
4441 	if (ctrl->state == NVME_CTRL_LIVE)
4442 		ctrl->ops->submit_async_event(ctrl);
4443 }
4444 
4445 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4446 {
4447 
4448 	u32 csts;
4449 
4450 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4451 		return false;
4452 
4453 	if (csts == ~0)
4454 		return false;
4455 
4456 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4457 }
4458 
4459 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4460 {
4461 	struct nvme_fw_slot_info_log *log;
4462 
4463 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4464 	if (!log)
4465 		return;
4466 
4467 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4468 			log, sizeof(*log), 0))
4469 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4470 	kfree(log);
4471 }
4472 
4473 static void nvme_fw_act_work(struct work_struct *work)
4474 {
4475 	struct nvme_ctrl *ctrl = container_of(work,
4476 				struct nvme_ctrl, fw_act_work);
4477 	unsigned long fw_act_timeout;
4478 
4479 	if (ctrl->mtfa)
4480 		fw_act_timeout = jiffies +
4481 				msecs_to_jiffies(ctrl->mtfa * 100);
4482 	else
4483 		fw_act_timeout = jiffies +
4484 				msecs_to_jiffies(admin_timeout * 1000);
4485 
4486 	nvme_stop_queues(ctrl);
4487 	while (nvme_ctrl_pp_status(ctrl)) {
4488 		if (time_after(jiffies, fw_act_timeout)) {
4489 			dev_warn(ctrl->device,
4490 				"Fw activation timeout, reset controller\n");
4491 			nvme_try_sched_reset(ctrl);
4492 			return;
4493 		}
4494 		msleep(100);
4495 	}
4496 
4497 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4498 		return;
4499 
4500 	nvme_start_queues(ctrl);
4501 	/* read FW slot information to clear the AER */
4502 	nvme_get_fw_slot_info(ctrl);
4503 }
4504 
4505 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4506 {
4507 	u32 aer_notice_type = (result & 0xff00) >> 8;
4508 
4509 	trace_nvme_async_event(ctrl, aer_notice_type);
4510 
4511 	switch (aer_notice_type) {
4512 	case NVME_AER_NOTICE_NS_CHANGED:
4513 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4514 		nvme_queue_scan(ctrl);
4515 		break;
4516 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4517 		/*
4518 		 * We are (ab)using the RESETTING state to prevent subsequent
4519 		 * recovery actions from interfering with the controller's
4520 		 * firmware activation.
4521 		 */
4522 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4523 			queue_work(nvme_wq, &ctrl->fw_act_work);
4524 		break;
4525 #ifdef CONFIG_NVME_MULTIPATH
4526 	case NVME_AER_NOTICE_ANA:
4527 		if (!ctrl->ana_log_buf)
4528 			break;
4529 		queue_work(nvme_wq, &ctrl->ana_work);
4530 		break;
4531 #endif
4532 	case NVME_AER_NOTICE_DISC_CHANGED:
4533 		ctrl->aen_result = result;
4534 		break;
4535 	default:
4536 		dev_warn(ctrl->device, "async event result %08x\n", result);
4537 	}
4538 }
4539 
4540 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4541 		volatile union nvme_result *res)
4542 {
4543 	u32 result = le32_to_cpu(res->u32);
4544 	u32 aer_type = result & 0x07;
4545 
4546 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4547 		return;
4548 
4549 	switch (aer_type) {
4550 	case NVME_AER_NOTICE:
4551 		nvme_handle_aen_notice(ctrl, result);
4552 		break;
4553 	case NVME_AER_ERROR:
4554 	case NVME_AER_SMART:
4555 	case NVME_AER_CSS:
4556 	case NVME_AER_VS:
4557 		trace_nvme_async_event(ctrl, aer_type);
4558 		ctrl->aen_result = result;
4559 		break;
4560 	default:
4561 		break;
4562 	}
4563 	queue_work(nvme_wq, &ctrl->async_event_work);
4564 }
4565 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4566 
4567 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4568 {
4569 	nvme_mpath_stop(ctrl);
4570 	nvme_stop_keep_alive(ctrl);
4571 	nvme_stop_failfast_work(ctrl);
4572 	flush_work(&ctrl->async_event_work);
4573 	cancel_work_sync(&ctrl->fw_act_work);
4574 }
4575 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4576 
4577 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4578 {
4579 	nvme_start_keep_alive(ctrl);
4580 
4581 	nvme_enable_aen(ctrl);
4582 
4583 	if (ctrl->queue_count > 1) {
4584 		nvme_queue_scan(ctrl);
4585 		nvme_start_queues(ctrl);
4586 		nvme_mpath_update(ctrl);
4587 	}
4588 
4589 	nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4590 }
4591 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4592 
4593 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4594 {
4595 	nvme_hwmon_exit(ctrl);
4596 	nvme_fault_inject_fini(&ctrl->fault_inject);
4597 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4598 	cdev_device_del(&ctrl->cdev, ctrl->device);
4599 	nvme_put_ctrl(ctrl);
4600 }
4601 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4602 
4603 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4604 {
4605 	struct nvme_effects_log	*cel;
4606 	unsigned long i;
4607 
4608 	xa_for_each(&ctrl->cels, i, cel) {
4609 		xa_erase(&ctrl->cels, i);
4610 		kfree(cel);
4611 	}
4612 
4613 	xa_destroy(&ctrl->cels);
4614 }
4615 
4616 static void nvme_free_ctrl(struct device *dev)
4617 {
4618 	struct nvme_ctrl *ctrl =
4619 		container_of(dev, struct nvme_ctrl, ctrl_device);
4620 	struct nvme_subsystem *subsys = ctrl->subsys;
4621 
4622 	if (!subsys || ctrl->instance != subsys->instance)
4623 		ida_free(&nvme_instance_ida, ctrl->instance);
4624 
4625 	nvme_free_cels(ctrl);
4626 	nvme_mpath_uninit(ctrl);
4627 	__free_page(ctrl->discard_page);
4628 
4629 	if (subsys) {
4630 		mutex_lock(&nvme_subsystems_lock);
4631 		list_del(&ctrl->subsys_entry);
4632 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4633 		mutex_unlock(&nvme_subsystems_lock);
4634 	}
4635 
4636 	ctrl->ops->free_ctrl(ctrl);
4637 
4638 	if (subsys)
4639 		nvme_put_subsystem(subsys);
4640 }
4641 
4642 /*
4643  * Initialize a NVMe controller structures.  This needs to be called during
4644  * earliest initialization so that we have the initialized structured around
4645  * during probing.
4646  */
4647 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4648 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
4649 {
4650 	int ret;
4651 
4652 	ctrl->state = NVME_CTRL_NEW;
4653 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4654 	spin_lock_init(&ctrl->lock);
4655 	mutex_init(&ctrl->scan_lock);
4656 	INIT_LIST_HEAD(&ctrl->namespaces);
4657 	xa_init(&ctrl->cels);
4658 	init_rwsem(&ctrl->namespaces_rwsem);
4659 	ctrl->dev = dev;
4660 	ctrl->ops = ops;
4661 	ctrl->quirks = quirks;
4662 	ctrl->numa_node = NUMA_NO_NODE;
4663 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4664 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4665 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4666 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4667 	init_waitqueue_head(&ctrl->state_wq);
4668 
4669 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4670 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4671 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4672 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4673 
4674 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4675 			PAGE_SIZE);
4676 	ctrl->discard_page = alloc_page(GFP_KERNEL);
4677 	if (!ctrl->discard_page) {
4678 		ret = -ENOMEM;
4679 		goto out;
4680 	}
4681 
4682 	ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4683 	if (ret < 0)
4684 		goto out;
4685 	ctrl->instance = ret;
4686 
4687 	device_initialize(&ctrl->ctrl_device);
4688 	ctrl->device = &ctrl->ctrl_device;
4689 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4690 			ctrl->instance);
4691 	ctrl->device->class = nvme_class;
4692 	ctrl->device->parent = ctrl->dev;
4693 	ctrl->device->groups = nvme_dev_attr_groups;
4694 	ctrl->device->release = nvme_free_ctrl;
4695 	dev_set_drvdata(ctrl->device, ctrl);
4696 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4697 	if (ret)
4698 		goto out_release_instance;
4699 
4700 	nvme_get_ctrl(ctrl);
4701 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
4702 	ctrl->cdev.owner = ops->module;
4703 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4704 	if (ret)
4705 		goto out_free_name;
4706 
4707 	/*
4708 	 * Initialize latency tolerance controls.  The sysfs files won't
4709 	 * be visible to userspace unless the device actually supports APST.
4710 	 */
4711 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4712 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4713 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4714 
4715 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4716 	nvme_mpath_init_ctrl(ctrl);
4717 
4718 	return 0;
4719 out_free_name:
4720 	nvme_put_ctrl(ctrl);
4721 	kfree_const(ctrl->device->kobj.name);
4722 out_release_instance:
4723 	ida_free(&nvme_instance_ida, ctrl->instance);
4724 out:
4725 	if (ctrl->discard_page)
4726 		__free_page(ctrl->discard_page);
4727 	return ret;
4728 }
4729 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4730 
4731 static void nvme_start_ns_queue(struct nvme_ns *ns)
4732 {
4733 	if (test_and_clear_bit(NVME_NS_STOPPED, &ns->flags))
4734 		blk_mq_unquiesce_queue(ns->queue);
4735 }
4736 
4737 static void nvme_stop_ns_queue(struct nvme_ns *ns)
4738 {
4739 	if (!test_and_set_bit(NVME_NS_STOPPED, &ns->flags))
4740 		blk_mq_quiesce_queue(ns->queue);
4741 	else
4742 		blk_mq_wait_quiesce_done(ns->queue);
4743 }
4744 
4745 /*
4746  * Prepare a queue for teardown.
4747  *
4748  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
4749  * the capacity to 0 after that to avoid blocking dispatchers that may be
4750  * holding bd_butex.  This will end buffered writers dirtying pages that can't
4751  * be synced.
4752  */
4753 static void nvme_set_queue_dying(struct nvme_ns *ns)
4754 {
4755 	if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
4756 		return;
4757 
4758 	blk_mark_disk_dead(ns->disk);
4759 	nvme_start_ns_queue(ns);
4760 
4761 	set_capacity_and_notify(ns->disk, 0);
4762 }
4763 
4764 /**
4765  * nvme_kill_queues(): Ends all namespace queues
4766  * @ctrl: the dead controller that needs to end
4767  *
4768  * Call this function when the driver determines it is unable to get the
4769  * controller in a state capable of servicing IO.
4770  */
4771 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4772 {
4773 	struct nvme_ns *ns;
4774 
4775 	down_read(&ctrl->namespaces_rwsem);
4776 
4777 	/* Forcibly unquiesce queues to avoid blocking dispatch */
4778 	if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4779 		nvme_start_admin_queue(ctrl);
4780 
4781 	list_for_each_entry(ns, &ctrl->namespaces, list)
4782 		nvme_set_queue_dying(ns);
4783 
4784 	up_read(&ctrl->namespaces_rwsem);
4785 }
4786 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4787 
4788 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4789 {
4790 	struct nvme_ns *ns;
4791 
4792 	down_read(&ctrl->namespaces_rwsem);
4793 	list_for_each_entry(ns, &ctrl->namespaces, list)
4794 		blk_mq_unfreeze_queue(ns->queue);
4795 	up_read(&ctrl->namespaces_rwsem);
4796 }
4797 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4798 
4799 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4800 {
4801 	struct nvme_ns *ns;
4802 
4803 	down_read(&ctrl->namespaces_rwsem);
4804 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4805 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4806 		if (timeout <= 0)
4807 			break;
4808 	}
4809 	up_read(&ctrl->namespaces_rwsem);
4810 	return timeout;
4811 }
4812 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4813 
4814 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4815 {
4816 	struct nvme_ns *ns;
4817 
4818 	down_read(&ctrl->namespaces_rwsem);
4819 	list_for_each_entry(ns, &ctrl->namespaces, list)
4820 		blk_mq_freeze_queue_wait(ns->queue);
4821 	up_read(&ctrl->namespaces_rwsem);
4822 }
4823 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4824 
4825 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4826 {
4827 	struct nvme_ns *ns;
4828 
4829 	down_read(&ctrl->namespaces_rwsem);
4830 	list_for_each_entry(ns, &ctrl->namespaces, list)
4831 		blk_freeze_queue_start(ns->queue);
4832 	up_read(&ctrl->namespaces_rwsem);
4833 }
4834 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4835 
4836 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4837 {
4838 	struct nvme_ns *ns;
4839 
4840 	down_read(&ctrl->namespaces_rwsem);
4841 	list_for_each_entry(ns, &ctrl->namespaces, list)
4842 		nvme_stop_ns_queue(ns);
4843 	up_read(&ctrl->namespaces_rwsem);
4844 }
4845 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4846 
4847 void nvme_start_queues(struct nvme_ctrl *ctrl)
4848 {
4849 	struct nvme_ns *ns;
4850 
4851 	down_read(&ctrl->namespaces_rwsem);
4852 	list_for_each_entry(ns, &ctrl->namespaces, list)
4853 		nvme_start_ns_queue(ns);
4854 	up_read(&ctrl->namespaces_rwsem);
4855 }
4856 EXPORT_SYMBOL_GPL(nvme_start_queues);
4857 
4858 void nvme_stop_admin_queue(struct nvme_ctrl *ctrl)
4859 {
4860 	if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4861 		blk_mq_quiesce_queue(ctrl->admin_q);
4862 	else
4863 		blk_mq_wait_quiesce_done(ctrl->admin_q);
4864 }
4865 EXPORT_SYMBOL_GPL(nvme_stop_admin_queue);
4866 
4867 void nvme_start_admin_queue(struct nvme_ctrl *ctrl)
4868 {
4869 	if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4870 		blk_mq_unquiesce_queue(ctrl->admin_q);
4871 }
4872 EXPORT_SYMBOL_GPL(nvme_start_admin_queue);
4873 
4874 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4875 {
4876 	struct nvme_ns *ns;
4877 
4878 	down_read(&ctrl->namespaces_rwsem);
4879 	list_for_each_entry(ns, &ctrl->namespaces, list)
4880 		blk_sync_queue(ns->queue);
4881 	up_read(&ctrl->namespaces_rwsem);
4882 }
4883 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4884 
4885 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4886 {
4887 	nvme_sync_io_queues(ctrl);
4888 	if (ctrl->admin_q)
4889 		blk_sync_queue(ctrl->admin_q);
4890 }
4891 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4892 
4893 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4894 {
4895 	if (file->f_op != &nvme_dev_fops)
4896 		return NULL;
4897 	return file->private_data;
4898 }
4899 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4900 
4901 /*
4902  * Check we didn't inadvertently grow the command structure sizes:
4903  */
4904 static inline void _nvme_check_size(void)
4905 {
4906 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4907 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4908 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4909 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4910 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4911 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4912 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4913 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4914 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4915 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4916 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4917 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4918 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4919 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
4920 			NVME_IDENTIFY_DATA_SIZE);
4921 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4922 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
4923 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4924 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4925 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4926 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4927 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4928 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4929 	BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
4930 }
4931 
4932 
4933 static int __init nvme_core_init(void)
4934 {
4935 	int result = -ENOMEM;
4936 
4937 	_nvme_check_size();
4938 
4939 	nvme_wq = alloc_workqueue("nvme-wq",
4940 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4941 	if (!nvme_wq)
4942 		goto out;
4943 
4944 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4945 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4946 	if (!nvme_reset_wq)
4947 		goto destroy_wq;
4948 
4949 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4950 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4951 	if (!nvme_delete_wq)
4952 		goto destroy_reset_wq;
4953 
4954 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4955 			NVME_MINORS, "nvme");
4956 	if (result < 0)
4957 		goto destroy_delete_wq;
4958 
4959 	nvme_class = class_create(THIS_MODULE, "nvme");
4960 	if (IS_ERR(nvme_class)) {
4961 		result = PTR_ERR(nvme_class);
4962 		goto unregister_chrdev;
4963 	}
4964 	nvme_class->dev_uevent = nvme_class_uevent;
4965 
4966 	nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4967 	if (IS_ERR(nvme_subsys_class)) {
4968 		result = PTR_ERR(nvme_subsys_class);
4969 		goto destroy_class;
4970 	}
4971 
4972 	result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4973 				     "nvme-generic");
4974 	if (result < 0)
4975 		goto destroy_subsys_class;
4976 
4977 	nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
4978 	if (IS_ERR(nvme_ns_chr_class)) {
4979 		result = PTR_ERR(nvme_ns_chr_class);
4980 		goto unregister_generic_ns;
4981 	}
4982 
4983 	return 0;
4984 
4985 unregister_generic_ns:
4986 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4987 destroy_subsys_class:
4988 	class_destroy(nvme_subsys_class);
4989 destroy_class:
4990 	class_destroy(nvme_class);
4991 unregister_chrdev:
4992 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4993 destroy_delete_wq:
4994 	destroy_workqueue(nvme_delete_wq);
4995 destroy_reset_wq:
4996 	destroy_workqueue(nvme_reset_wq);
4997 destroy_wq:
4998 	destroy_workqueue(nvme_wq);
4999 out:
5000 	return result;
5001 }
5002 
5003 static void __exit nvme_core_exit(void)
5004 {
5005 	class_destroy(nvme_ns_chr_class);
5006 	class_destroy(nvme_subsys_class);
5007 	class_destroy(nvme_class);
5008 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5009 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5010 	destroy_workqueue(nvme_delete_wq);
5011 	destroy_workqueue(nvme_reset_wq);
5012 	destroy_workqueue(nvme_wq);
5013 	ida_destroy(&nvme_ns_chr_minor_ida);
5014 	ida_destroy(&nvme_instance_ida);
5015 }
5016 
5017 MODULE_LICENSE("GPL");
5018 MODULE_VERSION("1.0");
5019 module_init(nvme_core_init);
5020 module_exit(nvme_core_exit);
5021