xref: /openbmc/linux/drivers/nvme/host/core.c (revision cef69974)
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 	blk_execute_rq_nowait(rq, false, nvme_keep_alive_end_io);
1211 }
1212 
1213 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1214 {
1215 	if (unlikely(ctrl->kato == 0))
1216 		return;
1217 
1218 	nvme_queue_keep_alive_work(ctrl);
1219 }
1220 
1221 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1222 {
1223 	if (unlikely(ctrl->kato == 0))
1224 		return;
1225 
1226 	cancel_delayed_work_sync(&ctrl->ka_work);
1227 }
1228 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1229 
1230 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1231 				   struct nvme_command *cmd)
1232 {
1233 	unsigned int new_kato =
1234 		DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1235 
1236 	dev_info(ctrl->device,
1237 		 "keep alive interval updated from %u ms to %u ms\n",
1238 		 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1239 
1240 	nvme_stop_keep_alive(ctrl);
1241 	ctrl->kato = new_kato;
1242 	nvme_start_keep_alive(ctrl);
1243 }
1244 
1245 /*
1246  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1247  * flag, thus sending any new CNS opcodes has a big chance of not working.
1248  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1249  * (but not for any later version).
1250  */
1251 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1252 {
1253 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1254 		return ctrl->vs < NVME_VS(1, 2, 0);
1255 	return ctrl->vs < NVME_VS(1, 1, 0);
1256 }
1257 
1258 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1259 {
1260 	struct nvme_command c = { };
1261 	int error;
1262 
1263 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1264 	c.identify.opcode = nvme_admin_identify;
1265 	c.identify.cns = NVME_ID_CNS_CTRL;
1266 
1267 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1268 	if (!*id)
1269 		return -ENOMEM;
1270 
1271 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1272 			sizeof(struct nvme_id_ctrl));
1273 	if (error)
1274 		kfree(*id);
1275 	return error;
1276 }
1277 
1278 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1279 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1280 {
1281 	const char *warn_str = "ctrl returned bogus length:";
1282 	void *data = cur;
1283 
1284 	switch (cur->nidt) {
1285 	case NVME_NIDT_EUI64:
1286 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1287 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1288 				 warn_str, cur->nidl);
1289 			return -1;
1290 		}
1291 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1292 			return NVME_NIDT_EUI64_LEN;
1293 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1294 		return NVME_NIDT_EUI64_LEN;
1295 	case NVME_NIDT_NGUID:
1296 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1297 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1298 				 warn_str, cur->nidl);
1299 			return -1;
1300 		}
1301 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1302 			return NVME_NIDT_NGUID_LEN;
1303 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1304 		return NVME_NIDT_NGUID_LEN;
1305 	case NVME_NIDT_UUID:
1306 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1307 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1308 				 warn_str, cur->nidl);
1309 			return -1;
1310 		}
1311 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1312 			return NVME_NIDT_UUID_LEN;
1313 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1314 		return NVME_NIDT_UUID_LEN;
1315 	case NVME_NIDT_CSI:
1316 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1317 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1318 				 warn_str, cur->nidl);
1319 			return -1;
1320 		}
1321 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1322 		*csi_seen = true;
1323 		return NVME_NIDT_CSI_LEN;
1324 	default:
1325 		/* Skip unknown types */
1326 		return cur->nidl;
1327 	}
1328 }
1329 
1330 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1331 		struct nvme_ns_ids *ids)
1332 {
1333 	struct nvme_command c = { };
1334 	bool csi_seen = false;
1335 	int status, pos, len;
1336 	void *data;
1337 
1338 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1339 		return 0;
1340 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1341 		return 0;
1342 
1343 	c.identify.opcode = nvme_admin_identify;
1344 	c.identify.nsid = cpu_to_le32(nsid);
1345 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1346 
1347 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1348 	if (!data)
1349 		return -ENOMEM;
1350 
1351 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1352 				      NVME_IDENTIFY_DATA_SIZE);
1353 	if (status) {
1354 		dev_warn(ctrl->device,
1355 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1356 			nsid, status);
1357 		goto free_data;
1358 	}
1359 
1360 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1361 		struct nvme_ns_id_desc *cur = data + pos;
1362 
1363 		if (cur->nidl == 0)
1364 			break;
1365 
1366 		len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1367 		if (len < 0)
1368 			break;
1369 
1370 		len += sizeof(*cur);
1371 	}
1372 
1373 	if (nvme_multi_css(ctrl) && !csi_seen) {
1374 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1375 			 nsid);
1376 		status = -EINVAL;
1377 	}
1378 
1379 free_data:
1380 	kfree(data);
1381 	return status;
1382 }
1383 
1384 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1385 			struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1386 {
1387 	struct nvme_command c = { };
1388 	int error;
1389 
1390 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1391 	c.identify.opcode = nvme_admin_identify;
1392 	c.identify.nsid = cpu_to_le32(nsid);
1393 	c.identify.cns = NVME_ID_CNS_NS;
1394 
1395 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1396 	if (!*id)
1397 		return -ENOMEM;
1398 
1399 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1400 	if (error) {
1401 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1402 		goto out_free_id;
1403 	}
1404 
1405 	error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1406 	if ((*id)->ncap == 0) /* namespace not allocated or attached */
1407 		goto out_free_id;
1408 
1409 
1410 	if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1411 		dev_info(ctrl->device,
1412 			 "Ignoring bogus Namespace Identifiers\n");
1413 	} else {
1414 		if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1415 		    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1416 			memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1417 		if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1418 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1419 			memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1420 	}
1421 
1422 	return 0;
1423 
1424 out_free_id:
1425 	kfree(*id);
1426 	return error;
1427 }
1428 
1429 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1430 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1431 {
1432 	union nvme_result res = { 0 };
1433 	struct nvme_command c = { };
1434 	int ret;
1435 
1436 	c.features.opcode = op;
1437 	c.features.fid = cpu_to_le32(fid);
1438 	c.features.dword11 = cpu_to_le32(dword11);
1439 
1440 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1441 			buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1442 	if (ret >= 0 && result)
1443 		*result = le32_to_cpu(res.u32);
1444 	return ret;
1445 }
1446 
1447 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1448 		      unsigned int dword11, void *buffer, size_t buflen,
1449 		      u32 *result)
1450 {
1451 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1452 			     buflen, result);
1453 }
1454 EXPORT_SYMBOL_GPL(nvme_set_features);
1455 
1456 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1457 		      unsigned int dword11, void *buffer, size_t buflen,
1458 		      u32 *result)
1459 {
1460 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1461 			     buflen, result);
1462 }
1463 EXPORT_SYMBOL_GPL(nvme_get_features);
1464 
1465 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1466 {
1467 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1468 	u32 result;
1469 	int status, nr_io_queues;
1470 
1471 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1472 			&result);
1473 	if (status < 0)
1474 		return status;
1475 
1476 	/*
1477 	 * Degraded controllers might return an error when setting the queue
1478 	 * count.  We still want to be able to bring them online and offer
1479 	 * access to the admin queue, as that might be only way to fix them up.
1480 	 */
1481 	if (status > 0) {
1482 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1483 		*count = 0;
1484 	} else {
1485 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1486 		*count = min(*count, nr_io_queues);
1487 	}
1488 
1489 	return 0;
1490 }
1491 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1492 
1493 #define NVME_AEN_SUPPORTED \
1494 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1495 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1496 
1497 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1498 {
1499 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1500 	int status;
1501 
1502 	if (!supported_aens)
1503 		return;
1504 
1505 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1506 			NULL, 0, &result);
1507 	if (status)
1508 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1509 			 supported_aens);
1510 
1511 	queue_work(nvme_wq, &ctrl->async_event_work);
1512 }
1513 
1514 static int nvme_ns_open(struct nvme_ns *ns)
1515 {
1516 
1517 	/* should never be called due to GENHD_FL_HIDDEN */
1518 	if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1519 		goto fail;
1520 	if (!nvme_get_ns(ns))
1521 		goto fail;
1522 	if (!try_module_get(ns->ctrl->ops->module))
1523 		goto fail_put_ns;
1524 
1525 	return 0;
1526 
1527 fail_put_ns:
1528 	nvme_put_ns(ns);
1529 fail:
1530 	return -ENXIO;
1531 }
1532 
1533 static void nvme_ns_release(struct nvme_ns *ns)
1534 {
1535 
1536 	module_put(ns->ctrl->ops->module);
1537 	nvme_put_ns(ns);
1538 }
1539 
1540 static int nvme_open(struct block_device *bdev, fmode_t mode)
1541 {
1542 	return nvme_ns_open(bdev->bd_disk->private_data);
1543 }
1544 
1545 static void nvme_release(struct gendisk *disk, fmode_t mode)
1546 {
1547 	nvme_ns_release(disk->private_data);
1548 }
1549 
1550 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1551 {
1552 	/* some standard values */
1553 	geo->heads = 1 << 6;
1554 	geo->sectors = 1 << 5;
1555 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1556 	return 0;
1557 }
1558 
1559 #ifdef CONFIG_BLK_DEV_INTEGRITY
1560 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1561 				u32 max_integrity_segments)
1562 {
1563 	struct blk_integrity integrity = { };
1564 
1565 	switch (ns->pi_type) {
1566 	case NVME_NS_DPS_PI_TYPE3:
1567 		switch (ns->guard_type) {
1568 		case NVME_NVM_NS_16B_GUARD:
1569 			integrity.profile = &t10_pi_type3_crc;
1570 			integrity.tag_size = sizeof(u16) + sizeof(u32);
1571 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1572 			break;
1573 		case NVME_NVM_NS_64B_GUARD:
1574 			integrity.profile = &ext_pi_type3_crc64;
1575 			integrity.tag_size = sizeof(u16) + 6;
1576 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1577 			break;
1578 		default:
1579 			integrity.profile = NULL;
1580 			break;
1581 		}
1582 		break;
1583 	case NVME_NS_DPS_PI_TYPE1:
1584 	case NVME_NS_DPS_PI_TYPE2:
1585 		switch (ns->guard_type) {
1586 		case NVME_NVM_NS_16B_GUARD:
1587 			integrity.profile = &t10_pi_type1_crc;
1588 			integrity.tag_size = sizeof(u16);
1589 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1590 			break;
1591 		case NVME_NVM_NS_64B_GUARD:
1592 			integrity.profile = &ext_pi_type1_crc64;
1593 			integrity.tag_size = sizeof(u16);
1594 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1595 			break;
1596 		default:
1597 			integrity.profile = NULL;
1598 			break;
1599 		}
1600 		break;
1601 	default:
1602 		integrity.profile = NULL;
1603 		break;
1604 	}
1605 
1606 	integrity.tuple_size = ns->ms;
1607 	blk_integrity_register(disk, &integrity);
1608 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1609 }
1610 #else
1611 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1612 				u32 max_integrity_segments)
1613 {
1614 }
1615 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1616 
1617 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1618 {
1619 	struct nvme_ctrl *ctrl = ns->ctrl;
1620 	struct request_queue *queue = disk->queue;
1621 	u32 size = queue_logical_block_size(queue);
1622 
1623 	if (ctrl->max_discard_sectors == 0) {
1624 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1625 		return;
1626 	}
1627 
1628 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1629 			NVME_DSM_MAX_RANGES);
1630 
1631 	queue->limits.discard_alignment = 0;
1632 	queue->limits.discard_granularity = size;
1633 
1634 	/* If discard is already enabled, don't reset queue limits */
1635 	if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1636 		return;
1637 
1638 	blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1639 	blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1640 
1641 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1642 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1643 }
1644 
1645 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1646 {
1647 	return uuid_equal(&a->uuid, &b->uuid) &&
1648 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1649 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1650 		a->csi == b->csi;
1651 }
1652 
1653 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1654 {
1655 	bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1656 	unsigned lbaf = nvme_lbaf_index(id->flbas);
1657 	struct nvme_ctrl *ctrl = ns->ctrl;
1658 	struct nvme_command c = { };
1659 	struct nvme_id_ns_nvm *nvm;
1660 	int ret = 0;
1661 	u32 elbaf;
1662 
1663 	ns->pi_size = 0;
1664 	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1665 	if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1666 		ns->pi_size = sizeof(struct t10_pi_tuple);
1667 		ns->guard_type = NVME_NVM_NS_16B_GUARD;
1668 		goto set_pi;
1669 	}
1670 
1671 	nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1672 	if (!nvm)
1673 		return -ENOMEM;
1674 
1675 	c.identify.opcode = nvme_admin_identify;
1676 	c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1677 	c.identify.cns = NVME_ID_CNS_CS_NS;
1678 	c.identify.csi = NVME_CSI_NVM;
1679 
1680 	ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1681 	if (ret)
1682 		goto free_data;
1683 
1684 	elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1685 
1686 	/* no support for storage tag formats right now */
1687 	if (nvme_elbaf_sts(elbaf))
1688 		goto free_data;
1689 
1690 	ns->guard_type = nvme_elbaf_guard_type(elbaf);
1691 	switch (ns->guard_type) {
1692 	case NVME_NVM_NS_64B_GUARD:
1693 		ns->pi_size = sizeof(struct crc64_pi_tuple);
1694 		break;
1695 	case NVME_NVM_NS_16B_GUARD:
1696 		ns->pi_size = sizeof(struct t10_pi_tuple);
1697 		break;
1698 	default:
1699 		break;
1700 	}
1701 
1702 free_data:
1703 	kfree(nvm);
1704 set_pi:
1705 	if (ns->pi_size && (first || ns->ms == ns->pi_size))
1706 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1707 	else
1708 		ns->pi_type = 0;
1709 
1710 	return ret;
1711 }
1712 
1713 static void nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1714 {
1715 	struct nvme_ctrl *ctrl = ns->ctrl;
1716 
1717 	if (nvme_init_ms(ns, id))
1718 		return;
1719 
1720 	ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1721 	if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1722 		return;
1723 
1724 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1725 		/*
1726 		 * The NVMe over Fabrics specification only supports metadata as
1727 		 * part of the extended data LBA.  We rely on HCA/HBA support to
1728 		 * remap the separate metadata buffer from the block layer.
1729 		 */
1730 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1731 			return;
1732 
1733 		ns->features |= NVME_NS_EXT_LBAS;
1734 
1735 		/*
1736 		 * The current fabrics transport drivers support namespace
1737 		 * metadata formats only if nvme_ns_has_pi() returns true.
1738 		 * Suppress support for all other formats so the namespace will
1739 		 * have a 0 capacity and not be usable through the block stack.
1740 		 *
1741 		 * Note, this check will need to be modified if any drivers
1742 		 * gain the ability to use other metadata formats.
1743 		 */
1744 		if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1745 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1746 	} else {
1747 		/*
1748 		 * For PCIe controllers, we can't easily remap the separate
1749 		 * metadata buffer from the block layer and thus require a
1750 		 * separate metadata buffer for block layer metadata/PI support.
1751 		 * We allow extended LBAs for the passthrough interface, though.
1752 		 */
1753 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
1754 			ns->features |= NVME_NS_EXT_LBAS;
1755 		else
1756 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1757 	}
1758 }
1759 
1760 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1761 		struct request_queue *q)
1762 {
1763 	bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1764 
1765 	if (ctrl->max_hw_sectors) {
1766 		u32 max_segments =
1767 			(ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1768 
1769 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
1770 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1771 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1772 	}
1773 	blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1774 	blk_queue_dma_alignment(q, 7);
1775 	blk_queue_write_cache(q, vwc, vwc);
1776 }
1777 
1778 static void nvme_update_disk_info(struct gendisk *disk,
1779 		struct nvme_ns *ns, struct nvme_id_ns *id)
1780 {
1781 	sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1782 	unsigned short bs = 1 << ns->lba_shift;
1783 	u32 atomic_bs, phys_bs, io_opt = 0;
1784 
1785 	/*
1786 	 * The block layer can't support LBA sizes larger than the page size
1787 	 * yet, so catch this early and don't allow block I/O.
1788 	 */
1789 	if (ns->lba_shift > PAGE_SHIFT) {
1790 		capacity = 0;
1791 		bs = (1 << 9);
1792 	}
1793 
1794 	blk_integrity_unregister(disk);
1795 
1796 	atomic_bs = phys_bs = bs;
1797 	if (id->nabo == 0) {
1798 		/*
1799 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1800 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1801 		 * 0 then AWUPF must be used instead.
1802 		 */
1803 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1804 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1805 		else
1806 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1807 	}
1808 
1809 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1810 		/* NPWG = Namespace Preferred Write Granularity */
1811 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1812 		/* NOWS = Namespace Optimal Write Size */
1813 		io_opt = bs * (1 + le16_to_cpu(id->nows));
1814 	}
1815 
1816 	blk_queue_logical_block_size(disk->queue, bs);
1817 	/*
1818 	 * Linux filesystems assume writing a single physical block is
1819 	 * an atomic operation. Hence limit the physical block size to the
1820 	 * value of the Atomic Write Unit Power Fail parameter.
1821 	 */
1822 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1823 	blk_queue_io_min(disk->queue, phys_bs);
1824 	blk_queue_io_opt(disk->queue, io_opt);
1825 
1826 	/*
1827 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
1828 	 * metadata masquerading as Type 0 if supported, otherwise reject block
1829 	 * I/O to namespaces with metadata except when the namespace supports
1830 	 * PI, as it can strip/insert in that case.
1831 	 */
1832 	if (ns->ms) {
1833 		if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1834 		    (ns->features & NVME_NS_METADATA_SUPPORTED))
1835 			nvme_init_integrity(disk, ns,
1836 					    ns->ctrl->max_integrity_segments);
1837 		else if (!nvme_ns_has_pi(ns))
1838 			capacity = 0;
1839 	}
1840 
1841 	set_capacity_and_notify(disk, capacity);
1842 
1843 	nvme_config_discard(disk, ns);
1844 	blk_queue_max_write_zeroes_sectors(disk->queue,
1845 					   ns->ctrl->max_zeroes_sectors);
1846 }
1847 
1848 static inline bool nvme_first_scan(struct gendisk *disk)
1849 {
1850 	/* nvme_alloc_ns() scans the disk prior to adding it */
1851 	return !disk_live(disk);
1852 }
1853 
1854 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1855 {
1856 	struct nvme_ctrl *ctrl = ns->ctrl;
1857 	u32 iob;
1858 
1859 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1860 	    is_power_of_2(ctrl->max_hw_sectors))
1861 		iob = ctrl->max_hw_sectors;
1862 	else
1863 		iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1864 
1865 	if (!iob)
1866 		return;
1867 
1868 	if (!is_power_of_2(iob)) {
1869 		if (nvme_first_scan(ns->disk))
1870 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1871 				ns->disk->disk_name, iob);
1872 		return;
1873 	}
1874 
1875 	if (blk_queue_is_zoned(ns->disk->queue)) {
1876 		if (nvme_first_scan(ns->disk))
1877 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
1878 				ns->disk->disk_name);
1879 		return;
1880 	}
1881 
1882 	blk_queue_chunk_sectors(ns->queue, iob);
1883 }
1884 
1885 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1886 {
1887 	unsigned lbaf = nvme_lbaf_index(id->flbas);
1888 	int ret;
1889 
1890 	blk_mq_freeze_queue(ns->disk->queue);
1891 	ns->lba_shift = id->lbaf[lbaf].ds;
1892 	nvme_set_queue_limits(ns->ctrl, ns->queue);
1893 
1894 	nvme_configure_metadata(ns, id);
1895 	nvme_set_chunk_sectors(ns, id);
1896 	nvme_update_disk_info(ns->disk, ns, id);
1897 
1898 	if (ns->head->ids.csi == NVME_CSI_ZNS) {
1899 		ret = nvme_update_zone_info(ns, lbaf);
1900 		if (ret)
1901 			goto out_unfreeze;
1902 	}
1903 
1904 	set_disk_ro(ns->disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1905 		test_bit(NVME_NS_FORCE_RO, &ns->flags));
1906 	set_bit(NVME_NS_READY, &ns->flags);
1907 	blk_mq_unfreeze_queue(ns->disk->queue);
1908 
1909 	if (blk_queue_is_zoned(ns->queue)) {
1910 		ret = nvme_revalidate_zones(ns);
1911 		if (ret && !nvme_first_scan(ns->disk))
1912 			return ret;
1913 	}
1914 
1915 	if (nvme_ns_head_multipath(ns->head)) {
1916 		blk_mq_freeze_queue(ns->head->disk->queue);
1917 		nvme_update_disk_info(ns->head->disk, ns, id);
1918 		set_disk_ro(ns->head->disk,
1919 			    (id->nsattr & NVME_NS_ATTR_RO) ||
1920 				    test_bit(NVME_NS_FORCE_RO, &ns->flags));
1921 		nvme_mpath_revalidate_paths(ns);
1922 		blk_stack_limits(&ns->head->disk->queue->limits,
1923 				 &ns->queue->limits, 0);
1924 		disk_update_readahead(ns->head->disk);
1925 		blk_mq_unfreeze_queue(ns->head->disk->queue);
1926 	}
1927 	return 0;
1928 
1929 out_unfreeze:
1930 	/*
1931 	 * If probing fails due an unsupported feature, hide the block device,
1932 	 * but still allow other access.
1933 	 */
1934 	if (ret == -ENODEV) {
1935 		ns->disk->flags |= GENHD_FL_HIDDEN;
1936 		set_bit(NVME_NS_READY, &ns->flags);
1937 		ret = 0;
1938 	}
1939 	blk_mq_unfreeze_queue(ns->disk->queue);
1940 	return ret;
1941 }
1942 
1943 static char nvme_pr_type(enum pr_type type)
1944 {
1945 	switch (type) {
1946 	case PR_WRITE_EXCLUSIVE:
1947 		return 1;
1948 	case PR_EXCLUSIVE_ACCESS:
1949 		return 2;
1950 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1951 		return 3;
1952 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1953 		return 4;
1954 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1955 		return 5;
1956 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1957 		return 6;
1958 	default:
1959 		return 0;
1960 	}
1961 }
1962 
1963 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
1964 		struct nvme_command *c, u8 data[16])
1965 {
1966 	struct nvme_ns_head *head = bdev->bd_disk->private_data;
1967 	int srcu_idx = srcu_read_lock(&head->srcu);
1968 	struct nvme_ns *ns = nvme_find_path(head);
1969 	int ret = -EWOULDBLOCK;
1970 
1971 	if (ns) {
1972 		c->common.nsid = cpu_to_le32(ns->head->ns_id);
1973 		ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
1974 	}
1975 	srcu_read_unlock(&head->srcu, srcu_idx);
1976 	return ret;
1977 }
1978 
1979 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
1980 		u8 data[16])
1981 {
1982 	c->common.nsid = cpu_to_le32(ns->head->ns_id);
1983 	return nvme_submit_sync_cmd(ns->queue, c, data, 16);
1984 }
1985 
1986 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1987 				u64 key, u64 sa_key, u8 op)
1988 {
1989 	struct nvme_command c = { };
1990 	u8 data[16] = { 0, };
1991 
1992 	put_unaligned_le64(key, &data[0]);
1993 	put_unaligned_le64(sa_key, &data[8]);
1994 
1995 	c.common.opcode = op;
1996 	c.common.cdw10 = cpu_to_le32(cdw10);
1997 
1998 	if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
1999 	    bdev->bd_disk->fops == &nvme_ns_head_ops)
2000 		return nvme_send_ns_head_pr_command(bdev, &c, data);
2001 	return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
2002 }
2003 
2004 static int nvme_pr_register(struct block_device *bdev, u64 old,
2005 		u64 new, unsigned flags)
2006 {
2007 	u32 cdw10;
2008 
2009 	if (flags & ~PR_FL_IGNORE_KEY)
2010 		return -EOPNOTSUPP;
2011 
2012 	cdw10 = old ? 2 : 0;
2013 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2014 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2015 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2016 }
2017 
2018 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2019 		enum pr_type type, unsigned flags)
2020 {
2021 	u32 cdw10;
2022 
2023 	if (flags & ~PR_FL_IGNORE_KEY)
2024 		return -EOPNOTSUPP;
2025 
2026 	cdw10 = nvme_pr_type(type) << 8;
2027 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2028 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2029 }
2030 
2031 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2032 		enum pr_type type, bool abort)
2033 {
2034 	u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2035 
2036 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2037 }
2038 
2039 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2040 {
2041 	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2042 
2043 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2044 }
2045 
2046 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2047 {
2048 	u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2049 
2050 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2051 }
2052 
2053 const struct pr_ops nvme_pr_ops = {
2054 	.pr_register	= nvme_pr_register,
2055 	.pr_reserve	= nvme_pr_reserve,
2056 	.pr_release	= nvme_pr_release,
2057 	.pr_preempt	= nvme_pr_preempt,
2058 	.pr_clear	= nvme_pr_clear,
2059 };
2060 
2061 #ifdef CONFIG_BLK_SED_OPAL
2062 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2063 		bool send)
2064 {
2065 	struct nvme_ctrl *ctrl = data;
2066 	struct nvme_command cmd = { };
2067 
2068 	if (send)
2069 		cmd.common.opcode = nvme_admin_security_send;
2070 	else
2071 		cmd.common.opcode = nvme_admin_security_recv;
2072 	cmd.common.nsid = 0;
2073 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2074 	cmd.common.cdw11 = cpu_to_le32(len);
2075 
2076 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2077 			NVME_QID_ANY, 1, 0);
2078 }
2079 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2080 #endif /* CONFIG_BLK_SED_OPAL */
2081 
2082 #ifdef CONFIG_BLK_DEV_ZONED
2083 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2084 		unsigned int nr_zones, report_zones_cb cb, void *data)
2085 {
2086 	return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2087 			data);
2088 }
2089 #else
2090 #define nvme_report_zones	NULL
2091 #endif /* CONFIG_BLK_DEV_ZONED */
2092 
2093 static const struct block_device_operations nvme_bdev_ops = {
2094 	.owner		= THIS_MODULE,
2095 	.ioctl		= nvme_ioctl,
2096 	.open		= nvme_open,
2097 	.release	= nvme_release,
2098 	.getgeo		= nvme_getgeo,
2099 	.report_zones	= nvme_report_zones,
2100 	.pr_ops		= &nvme_pr_ops,
2101 };
2102 
2103 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2104 {
2105 	unsigned long timeout =
2106 		((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2107 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2108 	int ret;
2109 
2110 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2111 		if (csts == ~0)
2112 			return -ENODEV;
2113 		if ((csts & NVME_CSTS_RDY) == bit)
2114 			break;
2115 
2116 		usleep_range(1000, 2000);
2117 		if (fatal_signal_pending(current))
2118 			return -EINTR;
2119 		if (time_after(jiffies, timeout)) {
2120 			dev_err(ctrl->device,
2121 				"Device not ready; aborting %s, CSTS=0x%x\n",
2122 				enabled ? "initialisation" : "reset", csts);
2123 			return -ENODEV;
2124 		}
2125 	}
2126 
2127 	return ret;
2128 }
2129 
2130 /*
2131  * If the device has been passed off to us in an enabled state, just clear
2132  * the enabled bit.  The spec says we should set the 'shutdown notification
2133  * bits', but doing so may cause the device to complete commands to the
2134  * admin queue ... and we don't know what memory that might be pointing at!
2135  */
2136 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2137 {
2138 	int ret;
2139 
2140 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2141 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2142 
2143 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2144 	if (ret)
2145 		return ret;
2146 
2147 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2148 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2149 
2150 	return nvme_wait_ready(ctrl, ctrl->cap, false);
2151 }
2152 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2153 
2154 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2155 {
2156 	unsigned dev_page_min;
2157 	int ret;
2158 
2159 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2160 	if (ret) {
2161 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2162 		return ret;
2163 	}
2164 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2165 
2166 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2167 		dev_err(ctrl->device,
2168 			"Minimum device page size %u too large for host (%u)\n",
2169 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2170 		return -ENODEV;
2171 	}
2172 
2173 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2174 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2175 	else
2176 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2177 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2178 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2179 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2180 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2181 
2182 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2183 	if (ret)
2184 		return ret;
2185 	return nvme_wait_ready(ctrl, ctrl->cap, true);
2186 }
2187 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2188 
2189 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2190 {
2191 	unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2192 	u32 csts;
2193 	int ret;
2194 
2195 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2196 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2197 
2198 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2199 	if (ret)
2200 		return ret;
2201 
2202 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2203 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2204 			break;
2205 
2206 		msleep(100);
2207 		if (fatal_signal_pending(current))
2208 			return -EINTR;
2209 		if (time_after(jiffies, timeout)) {
2210 			dev_err(ctrl->device,
2211 				"Device shutdown incomplete; abort shutdown\n");
2212 			return -ENODEV;
2213 		}
2214 	}
2215 
2216 	return ret;
2217 }
2218 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2219 
2220 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2221 {
2222 	__le64 ts;
2223 	int ret;
2224 
2225 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2226 		return 0;
2227 
2228 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2229 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2230 			NULL);
2231 	if (ret)
2232 		dev_warn_once(ctrl->device,
2233 			"could not set timestamp (%d)\n", ret);
2234 	return ret;
2235 }
2236 
2237 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2238 {
2239 	struct nvme_feat_host_behavior *host;
2240 	u8 acre = 0, lbafee = 0;
2241 	int ret;
2242 
2243 	/* Don't bother enabling the feature if retry delay is not reported */
2244 	if (ctrl->crdt[0])
2245 		acre = NVME_ENABLE_ACRE;
2246 	if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2247 		lbafee = NVME_ENABLE_LBAFEE;
2248 
2249 	if (!acre && !lbafee)
2250 		return 0;
2251 
2252 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2253 	if (!host)
2254 		return 0;
2255 
2256 	host->acre = acre;
2257 	host->lbafee = lbafee;
2258 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2259 				host, sizeof(*host), NULL);
2260 	kfree(host);
2261 	return ret;
2262 }
2263 
2264 /*
2265  * The function checks whether the given total (exlat + enlat) latency of
2266  * a power state allows the latter to be used as an APST transition target.
2267  * It does so by comparing the latency to the primary and secondary latency
2268  * tolerances defined by module params. If there's a match, the corresponding
2269  * timeout value is returned and the matching tolerance index (1 or 2) is
2270  * reported.
2271  */
2272 static bool nvme_apst_get_transition_time(u64 total_latency,
2273 		u64 *transition_time, unsigned *last_index)
2274 {
2275 	if (total_latency <= apst_primary_latency_tol_us) {
2276 		if (*last_index == 1)
2277 			return false;
2278 		*last_index = 1;
2279 		*transition_time = apst_primary_timeout_ms;
2280 		return true;
2281 	}
2282 	if (apst_secondary_timeout_ms &&
2283 		total_latency <= apst_secondary_latency_tol_us) {
2284 		if (*last_index <= 2)
2285 			return false;
2286 		*last_index = 2;
2287 		*transition_time = apst_secondary_timeout_ms;
2288 		return true;
2289 	}
2290 	return false;
2291 }
2292 
2293 /*
2294  * APST (Autonomous Power State Transition) lets us program a table of power
2295  * state transitions that the controller will perform automatically.
2296  *
2297  * Depending on module params, one of the two supported techniques will be used:
2298  *
2299  * - If the parameters provide explicit timeouts and tolerances, they will be
2300  *   used to build a table with up to 2 non-operational states to transition to.
2301  *   The default parameter values were selected based on the values used by
2302  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2303  *   regeneration of the APST table in the event of switching between external
2304  *   and battery power, the timeouts and tolerances reflect a compromise
2305  *   between values used by Microsoft for AC and battery scenarios.
2306  * - If not, we'll configure the table with a simple heuristic: we are willing
2307  *   to spend at most 2% of the time transitioning between power states.
2308  *   Therefore, when running in any given state, we will enter the next
2309  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2310  *   microseconds, as long as that state's exit latency is under the requested
2311  *   maximum latency.
2312  *
2313  * We will not autonomously enter any non-operational state for which the total
2314  * latency exceeds ps_max_latency_us.
2315  *
2316  * Users can set ps_max_latency_us to zero to turn off APST.
2317  */
2318 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2319 {
2320 	struct nvme_feat_auto_pst *table;
2321 	unsigned apste = 0;
2322 	u64 max_lat_us = 0;
2323 	__le64 target = 0;
2324 	int max_ps = -1;
2325 	int state;
2326 	int ret;
2327 	unsigned last_lt_index = UINT_MAX;
2328 
2329 	/*
2330 	 * If APST isn't supported or if we haven't been initialized yet,
2331 	 * then don't do anything.
2332 	 */
2333 	if (!ctrl->apsta)
2334 		return 0;
2335 
2336 	if (ctrl->npss > 31) {
2337 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2338 		return 0;
2339 	}
2340 
2341 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2342 	if (!table)
2343 		return 0;
2344 
2345 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2346 		/* Turn off APST. */
2347 		dev_dbg(ctrl->device, "APST disabled\n");
2348 		goto done;
2349 	}
2350 
2351 	/*
2352 	 * Walk through all states from lowest- to highest-power.
2353 	 * According to the spec, lower-numbered states use more power.  NPSS,
2354 	 * despite the name, is the index of the lowest-power state, not the
2355 	 * number of states.
2356 	 */
2357 	for (state = (int)ctrl->npss; state >= 0; state--) {
2358 		u64 total_latency_us, exit_latency_us, transition_ms;
2359 
2360 		if (target)
2361 			table->entries[state] = target;
2362 
2363 		/*
2364 		 * Don't allow transitions to the deepest state if it's quirked
2365 		 * off.
2366 		 */
2367 		if (state == ctrl->npss &&
2368 		    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2369 			continue;
2370 
2371 		/*
2372 		 * Is this state a useful non-operational state for higher-power
2373 		 * states to autonomously transition to?
2374 		 */
2375 		if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2376 			continue;
2377 
2378 		exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2379 		if (exit_latency_us > ctrl->ps_max_latency_us)
2380 			continue;
2381 
2382 		total_latency_us = exit_latency_us +
2383 			le32_to_cpu(ctrl->psd[state].entry_lat);
2384 
2385 		/*
2386 		 * This state is good. It can be used as the APST idle target
2387 		 * for higher power states.
2388 		 */
2389 		if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2390 			if (!nvme_apst_get_transition_time(total_latency_us,
2391 					&transition_ms, &last_lt_index))
2392 				continue;
2393 		} else {
2394 			transition_ms = total_latency_us + 19;
2395 			do_div(transition_ms, 20);
2396 			if (transition_ms > (1 << 24) - 1)
2397 				transition_ms = (1 << 24) - 1;
2398 		}
2399 
2400 		target = cpu_to_le64((state << 3) | (transition_ms << 8));
2401 		if (max_ps == -1)
2402 			max_ps = state;
2403 		if (total_latency_us > max_lat_us)
2404 			max_lat_us = total_latency_us;
2405 	}
2406 
2407 	if (max_ps == -1)
2408 		dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2409 	else
2410 		dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2411 			max_ps, max_lat_us, (int)sizeof(*table), table);
2412 	apste = 1;
2413 
2414 done:
2415 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2416 				table, sizeof(*table), NULL);
2417 	if (ret)
2418 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2419 	kfree(table);
2420 	return ret;
2421 }
2422 
2423 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2424 {
2425 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2426 	u64 latency;
2427 
2428 	switch (val) {
2429 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2430 	case PM_QOS_LATENCY_ANY:
2431 		latency = U64_MAX;
2432 		break;
2433 
2434 	default:
2435 		latency = val;
2436 	}
2437 
2438 	if (ctrl->ps_max_latency_us != latency) {
2439 		ctrl->ps_max_latency_us = latency;
2440 		if (ctrl->state == NVME_CTRL_LIVE)
2441 			nvme_configure_apst(ctrl);
2442 	}
2443 }
2444 
2445 struct nvme_core_quirk_entry {
2446 	/*
2447 	 * NVMe model and firmware strings are padded with spaces.  For
2448 	 * simplicity, strings in the quirk table are padded with NULLs
2449 	 * instead.
2450 	 */
2451 	u16 vid;
2452 	const char *mn;
2453 	const char *fr;
2454 	unsigned long quirks;
2455 };
2456 
2457 static const struct nvme_core_quirk_entry core_quirks[] = {
2458 	{
2459 		/*
2460 		 * This Toshiba device seems to die using any APST states.  See:
2461 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2462 		 */
2463 		.vid = 0x1179,
2464 		.mn = "THNSF5256GPUK TOSHIBA",
2465 		.quirks = NVME_QUIRK_NO_APST,
2466 	},
2467 	{
2468 		/*
2469 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2470 		 * condition associated with actions related to suspend to idle
2471 		 * LiteON has resolved the problem in future firmware
2472 		 */
2473 		.vid = 0x14a4,
2474 		.fr = "22301111",
2475 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2476 	},
2477 	{
2478 		/*
2479 		 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2480 		 * aborts I/O during any load, but more easily reproducible
2481 		 * with discards (fstrim).
2482 		 *
2483 		 * The device is left in a state where it is also not possible
2484 		 * to use "nvme set-feature" to disable APST, but booting with
2485 		 * nvme_core.default_ps_max_latency=0 works.
2486 		 */
2487 		.vid = 0x1e0f,
2488 		.mn = "KCD6XVUL6T40",
2489 		.quirks = NVME_QUIRK_NO_APST,
2490 	}
2491 };
2492 
2493 /* match is null-terminated but idstr is space-padded. */
2494 static bool string_matches(const char *idstr, const char *match, size_t len)
2495 {
2496 	size_t matchlen;
2497 
2498 	if (!match)
2499 		return true;
2500 
2501 	matchlen = strlen(match);
2502 	WARN_ON_ONCE(matchlen > len);
2503 
2504 	if (memcmp(idstr, match, matchlen))
2505 		return false;
2506 
2507 	for (; matchlen < len; matchlen++)
2508 		if (idstr[matchlen] != ' ')
2509 			return false;
2510 
2511 	return true;
2512 }
2513 
2514 static bool quirk_matches(const struct nvme_id_ctrl *id,
2515 			  const struct nvme_core_quirk_entry *q)
2516 {
2517 	return q->vid == le16_to_cpu(id->vid) &&
2518 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2519 		string_matches(id->fr, q->fr, sizeof(id->fr));
2520 }
2521 
2522 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2523 		struct nvme_id_ctrl *id)
2524 {
2525 	size_t nqnlen;
2526 	int off;
2527 
2528 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2529 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2530 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2531 			strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2532 			return;
2533 		}
2534 
2535 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2536 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2537 	}
2538 
2539 	/* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2540 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2541 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2542 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2543 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2544 	off += sizeof(id->sn);
2545 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2546 	off += sizeof(id->mn);
2547 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2548 }
2549 
2550 static void nvme_release_subsystem(struct device *dev)
2551 {
2552 	struct nvme_subsystem *subsys =
2553 		container_of(dev, struct nvme_subsystem, dev);
2554 
2555 	if (subsys->instance >= 0)
2556 		ida_free(&nvme_instance_ida, subsys->instance);
2557 	kfree(subsys);
2558 }
2559 
2560 static void nvme_destroy_subsystem(struct kref *ref)
2561 {
2562 	struct nvme_subsystem *subsys =
2563 			container_of(ref, struct nvme_subsystem, ref);
2564 
2565 	mutex_lock(&nvme_subsystems_lock);
2566 	list_del(&subsys->entry);
2567 	mutex_unlock(&nvme_subsystems_lock);
2568 
2569 	ida_destroy(&subsys->ns_ida);
2570 	device_del(&subsys->dev);
2571 	put_device(&subsys->dev);
2572 }
2573 
2574 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2575 {
2576 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2577 }
2578 
2579 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2580 {
2581 	struct nvme_subsystem *subsys;
2582 
2583 	lockdep_assert_held(&nvme_subsystems_lock);
2584 
2585 	/*
2586 	 * Fail matches for discovery subsystems. This results
2587 	 * in each discovery controller bound to a unique subsystem.
2588 	 * This avoids issues with validating controller values
2589 	 * that can only be true when there is a single unique subsystem.
2590 	 * There may be multiple and completely independent entities
2591 	 * that provide discovery controllers.
2592 	 */
2593 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2594 		return NULL;
2595 
2596 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2597 		if (strcmp(subsys->subnqn, subsysnqn))
2598 			continue;
2599 		if (!kref_get_unless_zero(&subsys->ref))
2600 			continue;
2601 		return subsys;
2602 	}
2603 
2604 	return NULL;
2605 }
2606 
2607 #define SUBSYS_ATTR_RO(_name, _mode, _show)			\
2608 	struct device_attribute subsys_attr_##_name = \
2609 		__ATTR(_name, _mode, _show, NULL)
2610 
2611 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2612 				    struct device_attribute *attr,
2613 				    char *buf)
2614 {
2615 	struct nvme_subsystem *subsys =
2616 		container_of(dev, struct nvme_subsystem, dev);
2617 
2618 	return sysfs_emit(buf, "%s\n", subsys->subnqn);
2619 }
2620 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2621 
2622 static ssize_t nvme_subsys_show_type(struct device *dev,
2623 				    struct device_attribute *attr,
2624 				    char *buf)
2625 {
2626 	struct nvme_subsystem *subsys =
2627 		container_of(dev, struct nvme_subsystem, dev);
2628 
2629 	switch (subsys->subtype) {
2630 	case NVME_NQN_DISC:
2631 		return sysfs_emit(buf, "discovery\n");
2632 	case NVME_NQN_NVME:
2633 		return sysfs_emit(buf, "nvm\n");
2634 	default:
2635 		return sysfs_emit(buf, "reserved\n");
2636 	}
2637 }
2638 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2639 
2640 #define nvme_subsys_show_str_function(field)				\
2641 static ssize_t subsys_##field##_show(struct device *dev,		\
2642 			    struct device_attribute *attr, char *buf)	\
2643 {									\
2644 	struct nvme_subsystem *subsys =					\
2645 		container_of(dev, struct nvme_subsystem, dev);		\
2646 	return sysfs_emit(buf, "%.*s\n",				\
2647 			   (int)sizeof(subsys->field), subsys->field);	\
2648 }									\
2649 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2650 
2651 nvme_subsys_show_str_function(model);
2652 nvme_subsys_show_str_function(serial);
2653 nvme_subsys_show_str_function(firmware_rev);
2654 
2655 static struct attribute *nvme_subsys_attrs[] = {
2656 	&subsys_attr_model.attr,
2657 	&subsys_attr_serial.attr,
2658 	&subsys_attr_firmware_rev.attr,
2659 	&subsys_attr_subsysnqn.attr,
2660 	&subsys_attr_subsystype.attr,
2661 #ifdef CONFIG_NVME_MULTIPATH
2662 	&subsys_attr_iopolicy.attr,
2663 #endif
2664 	NULL,
2665 };
2666 
2667 static const struct attribute_group nvme_subsys_attrs_group = {
2668 	.attrs = nvme_subsys_attrs,
2669 };
2670 
2671 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2672 	&nvme_subsys_attrs_group,
2673 	NULL,
2674 };
2675 
2676 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2677 {
2678 	return ctrl->opts && ctrl->opts->discovery_nqn;
2679 }
2680 
2681 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2682 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2683 {
2684 	struct nvme_ctrl *tmp;
2685 
2686 	lockdep_assert_held(&nvme_subsystems_lock);
2687 
2688 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2689 		if (nvme_state_terminal(tmp))
2690 			continue;
2691 
2692 		if (tmp->cntlid == ctrl->cntlid) {
2693 			dev_err(ctrl->device,
2694 				"Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2695 				ctrl->cntlid, dev_name(tmp->device),
2696 				subsys->subnqn);
2697 			return false;
2698 		}
2699 
2700 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2701 		    nvme_discovery_ctrl(ctrl))
2702 			continue;
2703 
2704 		dev_err(ctrl->device,
2705 			"Subsystem does not support multiple controllers\n");
2706 		return false;
2707 	}
2708 
2709 	return true;
2710 }
2711 
2712 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2713 {
2714 	struct nvme_subsystem *subsys, *found;
2715 	int ret;
2716 
2717 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2718 	if (!subsys)
2719 		return -ENOMEM;
2720 
2721 	subsys->instance = -1;
2722 	mutex_init(&subsys->lock);
2723 	kref_init(&subsys->ref);
2724 	INIT_LIST_HEAD(&subsys->ctrls);
2725 	INIT_LIST_HEAD(&subsys->nsheads);
2726 	nvme_init_subnqn(subsys, ctrl, id);
2727 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2728 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2729 	memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2730 	subsys->vendor_id = le16_to_cpu(id->vid);
2731 	subsys->cmic = id->cmic;
2732 
2733 	/* Versions prior to 1.4 don't necessarily report a valid type */
2734 	if (id->cntrltype == NVME_CTRL_DISC ||
2735 	    !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2736 		subsys->subtype = NVME_NQN_DISC;
2737 	else
2738 		subsys->subtype = NVME_NQN_NVME;
2739 
2740 	if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2741 		dev_err(ctrl->device,
2742 			"Subsystem %s is not a discovery controller",
2743 			subsys->subnqn);
2744 		kfree(subsys);
2745 		return -EINVAL;
2746 	}
2747 	subsys->awupf = le16_to_cpu(id->awupf);
2748 	nvme_mpath_default_iopolicy(subsys);
2749 
2750 	subsys->dev.class = nvme_subsys_class;
2751 	subsys->dev.release = nvme_release_subsystem;
2752 	subsys->dev.groups = nvme_subsys_attrs_groups;
2753 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2754 	device_initialize(&subsys->dev);
2755 
2756 	mutex_lock(&nvme_subsystems_lock);
2757 	found = __nvme_find_get_subsystem(subsys->subnqn);
2758 	if (found) {
2759 		put_device(&subsys->dev);
2760 		subsys = found;
2761 
2762 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2763 			ret = -EINVAL;
2764 			goto out_put_subsystem;
2765 		}
2766 	} else {
2767 		ret = device_add(&subsys->dev);
2768 		if (ret) {
2769 			dev_err(ctrl->device,
2770 				"failed to register subsystem device.\n");
2771 			put_device(&subsys->dev);
2772 			goto out_unlock;
2773 		}
2774 		ida_init(&subsys->ns_ida);
2775 		list_add_tail(&subsys->entry, &nvme_subsystems);
2776 	}
2777 
2778 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2779 				dev_name(ctrl->device));
2780 	if (ret) {
2781 		dev_err(ctrl->device,
2782 			"failed to create sysfs link from subsystem.\n");
2783 		goto out_put_subsystem;
2784 	}
2785 
2786 	if (!found)
2787 		subsys->instance = ctrl->instance;
2788 	ctrl->subsys = subsys;
2789 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2790 	mutex_unlock(&nvme_subsystems_lock);
2791 	return 0;
2792 
2793 out_put_subsystem:
2794 	nvme_put_subsystem(subsys);
2795 out_unlock:
2796 	mutex_unlock(&nvme_subsystems_lock);
2797 	return ret;
2798 }
2799 
2800 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2801 		void *log, size_t size, u64 offset)
2802 {
2803 	struct nvme_command c = { };
2804 	u32 dwlen = nvme_bytes_to_numd(size);
2805 
2806 	c.get_log_page.opcode = nvme_admin_get_log_page;
2807 	c.get_log_page.nsid = cpu_to_le32(nsid);
2808 	c.get_log_page.lid = log_page;
2809 	c.get_log_page.lsp = lsp;
2810 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2811 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2812 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2813 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2814 	c.get_log_page.csi = csi;
2815 
2816 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2817 }
2818 
2819 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2820 				struct nvme_effects_log **log)
2821 {
2822 	struct nvme_effects_log	*cel = xa_load(&ctrl->cels, csi);
2823 	int ret;
2824 
2825 	if (cel)
2826 		goto out;
2827 
2828 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2829 	if (!cel)
2830 		return -ENOMEM;
2831 
2832 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2833 			cel, sizeof(*cel), 0);
2834 	if (ret) {
2835 		kfree(cel);
2836 		return ret;
2837 	}
2838 
2839 	xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2840 out:
2841 	*log = cel;
2842 	return 0;
2843 }
2844 
2845 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2846 {
2847 	u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2848 
2849 	if (check_shl_overflow(1U, units + page_shift - 9, &val))
2850 		return UINT_MAX;
2851 	return val;
2852 }
2853 
2854 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2855 {
2856 	struct nvme_command c = { };
2857 	struct nvme_id_ctrl_nvm *id;
2858 	int ret;
2859 
2860 	if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2861 		ctrl->max_discard_sectors = UINT_MAX;
2862 		ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2863 	} else {
2864 		ctrl->max_discard_sectors = 0;
2865 		ctrl->max_discard_segments = 0;
2866 	}
2867 
2868 	/*
2869 	 * Even though NVMe spec explicitly states that MDTS is not applicable
2870 	 * to the write-zeroes, we are cautious and limit the size to the
2871 	 * controllers max_hw_sectors value, which is based on the MDTS field
2872 	 * and possibly other limiting factors.
2873 	 */
2874 	if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2875 	    !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2876 		ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2877 	else
2878 		ctrl->max_zeroes_sectors = 0;
2879 
2880 	if (nvme_ctrl_limited_cns(ctrl))
2881 		return 0;
2882 
2883 	id = kzalloc(sizeof(*id), GFP_KERNEL);
2884 	if (!id)
2885 		return 0;
2886 
2887 	c.identify.opcode = nvme_admin_identify;
2888 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
2889 	c.identify.csi = NVME_CSI_NVM;
2890 
2891 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2892 	if (ret)
2893 		goto free_data;
2894 
2895 	if (id->dmrl)
2896 		ctrl->max_discard_segments = id->dmrl;
2897 	if (id->dmrsl)
2898 		ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
2899 	if (id->wzsl)
2900 		ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2901 
2902 free_data:
2903 	kfree(id);
2904 	return ret;
2905 }
2906 
2907 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2908 {
2909 	struct nvme_id_ctrl *id;
2910 	u32 max_hw_sectors;
2911 	bool prev_apst_enabled;
2912 	int ret;
2913 
2914 	ret = nvme_identify_ctrl(ctrl, &id);
2915 	if (ret) {
2916 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2917 		return -EIO;
2918 	}
2919 
2920 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2921 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2922 		if (ret < 0)
2923 			goto out_free;
2924 	}
2925 
2926 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
2927 		ctrl->cntlid = le16_to_cpu(id->cntlid);
2928 
2929 	if (!ctrl->identified) {
2930 		unsigned int i;
2931 
2932 		ret = nvme_init_subsystem(ctrl, id);
2933 		if (ret)
2934 			goto out_free;
2935 
2936 		/*
2937 		 * Check for quirks.  Quirk can depend on firmware version,
2938 		 * so, in principle, the set of quirks present can change
2939 		 * across a reset.  As a possible future enhancement, we
2940 		 * could re-scan for quirks every time we reinitialize
2941 		 * the device, but we'd have to make sure that the driver
2942 		 * behaves intelligently if the quirks change.
2943 		 */
2944 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2945 			if (quirk_matches(id, &core_quirks[i]))
2946 				ctrl->quirks |= core_quirks[i].quirks;
2947 		}
2948 	}
2949 
2950 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2951 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2952 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2953 	}
2954 
2955 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2956 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2957 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2958 
2959 	ctrl->oacs = le16_to_cpu(id->oacs);
2960 	ctrl->oncs = le16_to_cpu(id->oncs);
2961 	ctrl->mtfa = le16_to_cpu(id->mtfa);
2962 	ctrl->oaes = le32_to_cpu(id->oaes);
2963 	ctrl->wctemp = le16_to_cpu(id->wctemp);
2964 	ctrl->cctemp = le16_to_cpu(id->cctemp);
2965 
2966 	atomic_set(&ctrl->abort_limit, id->acl + 1);
2967 	ctrl->vwc = id->vwc;
2968 	if (id->mdts)
2969 		max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
2970 	else
2971 		max_hw_sectors = UINT_MAX;
2972 	ctrl->max_hw_sectors =
2973 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2974 
2975 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
2976 	ctrl->sgls = le32_to_cpu(id->sgls);
2977 	ctrl->kas = le16_to_cpu(id->kas);
2978 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
2979 	ctrl->ctratt = le32_to_cpu(id->ctratt);
2980 
2981 	ctrl->cntrltype = id->cntrltype;
2982 	ctrl->dctype = id->dctype;
2983 
2984 	if (id->rtd3e) {
2985 		/* us -> s */
2986 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
2987 
2988 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2989 						 shutdown_timeout, 60);
2990 
2991 		if (ctrl->shutdown_timeout != shutdown_timeout)
2992 			dev_info(ctrl->device,
2993 				 "Shutdown timeout set to %u seconds\n",
2994 				 ctrl->shutdown_timeout);
2995 	} else
2996 		ctrl->shutdown_timeout = shutdown_timeout;
2997 
2998 	ctrl->npss = id->npss;
2999 	ctrl->apsta = id->apsta;
3000 	prev_apst_enabled = ctrl->apst_enabled;
3001 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3002 		if (force_apst && id->apsta) {
3003 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3004 			ctrl->apst_enabled = true;
3005 		} else {
3006 			ctrl->apst_enabled = false;
3007 		}
3008 	} else {
3009 		ctrl->apst_enabled = id->apsta;
3010 	}
3011 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3012 
3013 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3014 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3015 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3016 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3017 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3018 
3019 		/*
3020 		 * In fabrics we need to verify the cntlid matches the
3021 		 * admin connect
3022 		 */
3023 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3024 			dev_err(ctrl->device,
3025 				"Mismatching cntlid: Connect %u vs Identify "
3026 				"%u, rejecting\n",
3027 				ctrl->cntlid, le16_to_cpu(id->cntlid));
3028 			ret = -EINVAL;
3029 			goto out_free;
3030 		}
3031 
3032 		if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3033 			dev_err(ctrl->device,
3034 				"keep-alive support is mandatory for fabrics\n");
3035 			ret = -EINVAL;
3036 			goto out_free;
3037 		}
3038 	} else {
3039 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3040 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3041 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3042 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3043 	}
3044 
3045 	ret = nvme_mpath_init_identify(ctrl, id);
3046 	if (ret < 0)
3047 		goto out_free;
3048 
3049 	if (ctrl->apst_enabled && !prev_apst_enabled)
3050 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3051 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3052 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3053 
3054 out_free:
3055 	kfree(id);
3056 	return ret;
3057 }
3058 
3059 /*
3060  * Initialize the cached copies of the Identify data and various controller
3061  * register in our nvme_ctrl structure.  This should be called as soon as
3062  * the admin queue is fully up and running.
3063  */
3064 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3065 {
3066 	int ret;
3067 
3068 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3069 	if (ret) {
3070 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3071 		return ret;
3072 	}
3073 
3074 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3075 
3076 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3077 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3078 
3079 	ret = nvme_init_identify(ctrl);
3080 	if (ret)
3081 		return ret;
3082 
3083 	ret = nvme_init_non_mdts_limits(ctrl);
3084 	if (ret < 0)
3085 		return ret;
3086 
3087 	ret = nvme_configure_apst(ctrl);
3088 	if (ret < 0)
3089 		return ret;
3090 
3091 	ret = nvme_configure_timestamp(ctrl);
3092 	if (ret < 0)
3093 		return ret;
3094 
3095 	ret = nvme_configure_host_options(ctrl);
3096 	if (ret < 0)
3097 		return ret;
3098 
3099 	if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3100 		ret = nvme_hwmon_init(ctrl);
3101 		if (ret < 0)
3102 			return ret;
3103 	}
3104 
3105 	ctrl->identified = true;
3106 
3107 	return 0;
3108 }
3109 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3110 
3111 static int nvme_dev_open(struct inode *inode, struct file *file)
3112 {
3113 	struct nvme_ctrl *ctrl =
3114 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3115 
3116 	switch (ctrl->state) {
3117 	case NVME_CTRL_LIVE:
3118 		break;
3119 	default:
3120 		return -EWOULDBLOCK;
3121 	}
3122 
3123 	nvme_get_ctrl(ctrl);
3124 	if (!try_module_get(ctrl->ops->module)) {
3125 		nvme_put_ctrl(ctrl);
3126 		return -EINVAL;
3127 	}
3128 
3129 	file->private_data = ctrl;
3130 	return 0;
3131 }
3132 
3133 static int nvme_dev_release(struct inode *inode, struct file *file)
3134 {
3135 	struct nvme_ctrl *ctrl =
3136 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3137 
3138 	module_put(ctrl->ops->module);
3139 	nvme_put_ctrl(ctrl);
3140 	return 0;
3141 }
3142 
3143 static const struct file_operations nvme_dev_fops = {
3144 	.owner		= THIS_MODULE,
3145 	.open		= nvme_dev_open,
3146 	.release	= nvme_dev_release,
3147 	.unlocked_ioctl	= nvme_dev_ioctl,
3148 	.compat_ioctl	= compat_ptr_ioctl,
3149 };
3150 
3151 static ssize_t nvme_sysfs_reset(struct device *dev,
3152 				struct device_attribute *attr, const char *buf,
3153 				size_t count)
3154 {
3155 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3156 	int ret;
3157 
3158 	ret = nvme_reset_ctrl_sync(ctrl);
3159 	if (ret < 0)
3160 		return ret;
3161 	return count;
3162 }
3163 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3164 
3165 static ssize_t nvme_sysfs_rescan(struct device *dev,
3166 				struct device_attribute *attr, const char *buf,
3167 				size_t count)
3168 {
3169 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3170 
3171 	nvme_queue_scan(ctrl);
3172 	return count;
3173 }
3174 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3175 
3176 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3177 {
3178 	struct gendisk *disk = dev_to_disk(dev);
3179 
3180 	if (disk->fops == &nvme_bdev_ops)
3181 		return nvme_get_ns_from_dev(dev)->head;
3182 	else
3183 		return disk->private_data;
3184 }
3185 
3186 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3187 		char *buf)
3188 {
3189 	struct nvme_ns_head *head = dev_to_ns_head(dev);
3190 	struct nvme_ns_ids *ids = &head->ids;
3191 	struct nvme_subsystem *subsys = head->subsys;
3192 	int serial_len = sizeof(subsys->serial);
3193 	int model_len = sizeof(subsys->model);
3194 
3195 	if (!uuid_is_null(&ids->uuid))
3196 		return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3197 
3198 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3199 		return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3200 
3201 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3202 		return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3203 
3204 	while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3205 				  subsys->serial[serial_len - 1] == '\0'))
3206 		serial_len--;
3207 	while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3208 				 subsys->model[model_len - 1] == '\0'))
3209 		model_len--;
3210 
3211 	return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3212 		serial_len, subsys->serial, model_len, subsys->model,
3213 		head->ns_id);
3214 }
3215 static DEVICE_ATTR_RO(wwid);
3216 
3217 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3218 		char *buf)
3219 {
3220 	return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3221 }
3222 static DEVICE_ATTR_RO(nguid);
3223 
3224 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3225 		char *buf)
3226 {
3227 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3228 
3229 	/* For backward compatibility expose the NGUID to userspace if
3230 	 * we have no UUID set
3231 	 */
3232 	if (uuid_is_null(&ids->uuid)) {
3233 		printk_ratelimited(KERN_WARNING
3234 				   "No UUID available providing old NGUID\n");
3235 		return sysfs_emit(buf, "%pU\n", ids->nguid);
3236 	}
3237 	return sysfs_emit(buf, "%pU\n", &ids->uuid);
3238 }
3239 static DEVICE_ATTR_RO(uuid);
3240 
3241 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3242 		char *buf)
3243 {
3244 	return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3245 }
3246 static DEVICE_ATTR_RO(eui);
3247 
3248 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3249 		char *buf)
3250 {
3251 	return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3252 }
3253 static DEVICE_ATTR_RO(nsid);
3254 
3255 static struct attribute *nvme_ns_id_attrs[] = {
3256 	&dev_attr_wwid.attr,
3257 	&dev_attr_uuid.attr,
3258 	&dev_attr_nguid.attr,
3259 	&dev_attr_eui.attr,
3260 	&dev_attr_nsid.attr,
3261 #ifdef CONFIG_NVME_MULTIPATH
3262 	&dev_attr_ana_grpid.attr,
3263 	&dev_attr_ana_state.attr,
3264 #endif
3265 	NULL,
3266 };
3267 
3268 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3269 		struct attribute *a, int n)
3270 {
3271 	struct device *dev = container_of(kobj, struct device, kobj);
3272 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3273 
3274 	if (a == &dev_attr_uuid.attr) {
3275 		if (uuid_is_null(&ids->uuid) &&
3276 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3277 			return 0;
3278 	}
3279 	if (a == &dev_attr_nguid.attr) {
3280 		if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3281 			return 0;
3282 	}
3283 	if (a == &dev_attr_eui.attr) {
3284 		if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3285 			return 0;
3286 	}
3287 #ifdef CONFIG_NVME_MULTIPATH
3288 	if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3289 		if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3290 			return 0;
3291 		if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3292 			return 0;
3293 	}
3294 #endif
3295 	return a->mode;
3296 }
3297 
3298 static const struct attribute_group nvme_ns_id_attr_group = {
3299 	.attrs		= nvme_ns_id_attrs,
3300 	.is_visible	= nvme_ns_id_attrs_are_visible,
3301 };
3302 
3303 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3304 	&nvme_ns_id_attr_group,
3305 	NULL,
3306 };
3307 
3308 #define nvme_show_str_function(field)						\
3309 static ssize_t  field##_show(struct device *dev,				\
3310 			    struct device_attribute *attr, char *buf)		\
3311 {										\
3312         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3313         return sysfs_emit(buf, "%.*s\n",					\
3314 		(int)sizeof(ctrl->subsys->field), ctrl->subsys->field);		\
3315 }										\
3316 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3317 
3318 nvme_show_str_function(model);
3319 nvme_show_str_function(serial);
3320 nvme_show_str_function(firmware_rev);
3321 
3322 #define nvme_show_int_function(field)						\
3323 static ssize_t  field##_show(struct device *dev,				\
3324 			    struct device_attribute *attr, char *buf)		\
3325 {										\
3326         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3327         return sysfs_emit(buf, "%d\n", ctrl->field);				\
3328 }										\
3329 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3330 
3331 nvme_show_int_function(cntlid);
3332 nvme_show_int_function(numa_node);
3333 nvme_show_int_function(queue_count);
3334 nvme_show_int_function(sqsize);
3335 nvme_show_int_function(kato);
3336 
3337 static ssize_t nvme_sysfs_delete(struct device *dev,
3338 				struct device_attribute *attr, const char *buf,
3339 				size_t count)
3340 {
3341 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3342 
3343 	if (device_remove_file_self(dev, attr))
3344 		nvme_delete_ctrl_sync(ctrl);
3345 	return count;
3346 }
3347 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3348 
3349 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3350 					 struct device_attribute *attr,
3351 					 char *buf)
3352 {
3353 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3354 
3355 	return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3356 }
3357 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3358 
3359 static ssize_t nvme_sysfs_show_state(struct device *dev,
3360 				     struct device_attribute *attr,
3361 				     char *buf)
3362 {
3363 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3364 	static const char *const state_name[] = {
3365 		[NVME_CTRL_NEW]		= "new",
3366 		[NVME_CTRL_LIVE]	= "live",
3367 		[NVME_CTRL_RESETTING]	= "resetting",
3368 		[NVME_CTRL_CONNECTING]	= "connecting",
3369 		[NVME_CTRL_DELETING]	= "deleting",
3370 		[NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3371 		[NVME_CTRL_DEAD]	= "dead",
3372 	};
3373 
3374 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3375 	    state_name[ctrl->state])
3376 		return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3377 
3378 	return sysfs_emit(buf, "unknown state\n");
3379 }
3380 
3381 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3382 
3383 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3384 					 struct device_attribute *attr,
3385 					 char *buf)
3386 {
3387 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3388 
3389 	return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3390 }
3391 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3392 
3393 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3394 					struct device_attribute *attr,
3395 					char *buf)
3396 {
3397 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3398 
3399 	return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3400 }
3401 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3402 
3403 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3404 					struct device_attribute *attr,
3405 					char *buf)
3406 {
3407 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3408 
3409 	return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3410 }
3411 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3412 
3413 static ssize_t nvme_sysfs_show_address(struct device *dev,
3414 					 struct device_attribute *attr,
3415 					 char *buf)
3416 {
3417 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3418 
3419 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3420 }
3421 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3422 
3423 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3424 		struct device_attribute *attr, char *buf)
3425 {
3426 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3427 	struct nvmf_ctrl_options *opts = ctrl->opts;
3428 
3429 	if (ctrl->opts->max_reconnects == -1)
3430 		return sysfs_emit(buf, "off\n");
3431 	return sysfs_emit(buf, "%d\n",
3432 			  opts->max_reconnects * opts->reconnect_delay);
3433 }
3434 
3435 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3436 		struct device_attribute *attr, const char *buf, size_t count)
3437 {
3438 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3439 	struct nvmf_ctrl_options *opts = ctrl->opts;
3440 	int ctrl_loss_tmo, err;
3441 
3442 	err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3443 	if (err)
3444 		return -EINVAL;
3445 
3446 	if (ctrl_loss_tmo < 0)
3447 		opts->max_reconnects = -1;
3448 	else
3449 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3450 						opts->reconnect_delay);
3451 	return count;
3452 }
3453 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3454 	nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3455 
3456 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3457 		struct device_attribute *attr, char *buf)
3458 {
3459 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3460 
3461 	if (ctrl->opts->reconnect_delay == -1)
3462 		return sysfs_emit(buf, "off\n");
3463 	return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3464 }
3465 
3466 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3467 		struct device_attribute *attr, const char *buf, size_t count)
3468 {
3469 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3470 	unsigned int v;
3471 	int err;
3472 
3473 	err = kstrtou32(buf, 10, &v);
3474 	if (err)
3475 		return err;
3476 
3477 	ctrl->opts->reconnect_delay = v;
3478 	return count;
3479 }
3480 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3481 	nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3482 
3483 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3484 		struct device_attribute *attr, char *buf)
3485 {
3486 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3487 
3488 	if (ctrl->opts->fast_io_fail_tmo == -1)
3489 		return sysfs_emit(buf, "off\n");
3490 	return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3491 }
3492 
3493 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3494 		struct device_attribute *attr, const char *buf, size_t count)
3495 {
3496 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3497 	struct nvmf_ctrl_options *opts = ctrl->opts;
3498 	int fast_io_fail_tmo, err;
3499 
3500 	err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3501 	if (err)
3502 		return -EINVAL;
3503 
3504 	if (fast_io_fail_tmo < 0)
3505 		opts->fast_io_fail_tmo = -1;
3506 	else
3507 		opts->fast_io_fail_tmo = fast_io_fail_tmo;
3508 	return count;
3509 }
3510 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3511 	nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3512 
3513 static ssize_t cntrltype_show(struct device *dev,
3514 			      struct device_attribute *attr, char *buf)
3515 {
3516 	static const char * const type[] = {
3517 		[NVME_CTRL_IO] = "io\n",
3518 		[NVME_CTRL_DISC] = "discovery\n",
3519 		[NVME_CTRL_ADMIN] = "admin\n",
3520 	};
3521 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3522 
3523 	if (ctrl->cntrltype > NVME_CTRL_ADMIN || !type[ctrl->cntrltype])
3524 		return sysfs_emit(buf, "reserved\n");
3525 
3526 	return sysfs_emit(buf, type[ctrl->cntrltype]);
3527 }
3528 static DEVICE_ATTR_RO(cntrltype);
3529 
3530 static ssize_t dctype_show(struct device *dev,
3531 			   struct device_attribute *attr, char *buf)
3532 {
3533 	static const char * const type[] = {
3534 		[NVME_DCTYPE_NOT_REPORTED] = "none\n",
3535 		[NVME_DCTYPE_DDC] = "ddc\n",
3536 		[NVME_DCTYPE_CDC] = "cdc\n",
3537 	};
3538 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3539 
3540 	if (ctrl->dctype > NVME_DCTYPE_CDC || !type[ctrl->dctype])
3541 		return sysfs_emit(buf, "reserved\n");
3542 
3543 	return sysfs_emit(buf, type[ctrl->dctype]);
3544 }
3545 static DEVICE_ATTR_RO(dctype);
3546 
3547 static struct attribute *nvme_dev_attrs[] = {
3548 	&dev_attr_reset_controller.attr,
3549 	&dev_attr_rescan_controller.attr,
3550 	&dev_attr_model.attr,
3551 	&dev_attr_serial.attr,
3552 	&dev_attr_firmware_rev.attr,
3553 	&dev_attr_cntlid.attr,
3554 	&dev_attr_delete_controller.attr,
3555 	&dev_attr_transport.attr,
3556 	&dev_attr_subsysnqn.attr,
3557 	&dev_attr_address.attr,
3558 	&dev_attr_state.attr,
3559 	&dev_attr_numa_node.attr,
3560 	&dev_attr_queue_count.attr,
3561 	&dev_attr_sqsize.attr,
3562 	&dev_attr_hostnqn.attr,
3563 	&dev_attr_hostid.attr,
3564 	&dev_attr_ctrl_loss_tmo.attr,
3565 	&dev_attr_reconnect_delay.attr,
3566 	&dev_attr_fast_io_fail_tmo.attr,
3567 	&dev_attr_kato.attr,
3568 	&dev_attr_cntrltype.attr,
3569 	&dev_attr_dctype.attr,
3570 	NULL
3571 };
3572 
3573 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3574 		struct attribute *a, int n)
3575 {
3576 	struct device *dev = container_of(kobj, struct device, kobj);
3577 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3578 
3579 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3580 		return 0;
3581 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3582 		return 0;
3583 	if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3584 		return 0;
3585 	if (a == &dev_attr_hostid.attr && !ctrl->opts)
3586 		return 0;
3587 	if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3588 		return 0;
3589 	if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3590 		return 0;
3591 	if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3592 		return 0;
3593 
3594 	return a->mode;
3595 }
3596 
3597 static const struct attribute_group nvme_dev_attrs_group = {
3598 	.attrs		= nvme_dev_attrs,
3599 	.is_visible	= nvme_dev_attrs_are_visible,
3600 };
3601 
3602 static const struct attribute_group *nvme_dev_attr_groups[] = {
3603 	&nvme_dev_attrs_group,
3604 	NULL,
3605 };
3606 
3607 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3608 		unsigned nsid)
3609 {
3610 	struct nvme_ns_head *h;
3611 
3612 	lockdep_assert_held(&ctrl->subsys->lock);
3613 
3614 	list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3615 		/*
3616 		 * Private namespaces can share NSIDs under some conditions.
3617 		 * In that case we can't use the same ns_head for namespaces
3618 		 * with the same NSID.
3619 		 */
3620 		if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3621 			continue;
3622 		if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3623 			return h;
3624 	}
3625 
3626 	return NULL;
3627 }
3628 
3629 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3630 		struct nvme_ns_ids *ids)
3631 {
3632 	bool has_uuid = !uuid_is_null(&ids->uuid);
3633 	bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3634 	bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3635 	struct nvme_ns_head *h;
3636 
3637 	lockdep_assert_held(&subsys->lock);
3638 
3639 	list_for_each_entry(h, &subsys->nsheads, entry) {
3640 		if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3641 			return -EINVAL;
3642 		if (has_nguid &&
3643 		    memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3644 			return -EINVAL;
3645 		if (has_eui64 &&
3646 		    memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3647 			return -EINVAL;
3648 	}
3649 
3650 	return 0;
3651 }
3652 
3653 static void nvme_cdev_rel(struct device *dev)
3654 {
3655 	ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3656 }
3657 
3658 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3659 {
3660 	cdev_device_del(cdev, cdev_device);
3661 	put_device(cdev_device);
3662 }
3663 
3664 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3665 		const struct file_operations *fops, struct module *owner)
3666 {
3667 	int minor, ret;
3668 
3669 	minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3670 	if (minor < 0)
3671 		return minor;
3672 	cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3673 	cdev_device->class = nvme_ns_chr_class;
3674 	cdev_device->release = nvme_cdev_rel;
3675 	device_initialize(cdev_device);
3676 	cdev_init(cdev, fops);
3677 	cdev->owner = owner;
3678 	ret = cdev_device_add(cdev, cdev_device);
3679 	if (ret)
3680 		put_device(cdev_device);
3681 
3682 	return ret;
3683 }
3684 
3685 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3686 {
3687 	return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3688 }
3689 
3690 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3691 {
3692 	nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3693 	return 0;
3694 }
3695 
3696 static const struct file_operations nvme_ns_chr_fops = {
3697 	.owner		= THIS_MODULE,
3698 	.open		= nvme_ns_chr_open,
3699 	.release	= nvme_ns_chr_release,
3700 	.unlocked_ioctl	= nvme_ns_chr_ioctl,
3701 	.compat_ioctl	= compat_ptr_ioctl,
3702 };
3703 
3704 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3705 {
3706 	int ret;
3707 
3708 	ns->cdev_device.parent = ns->ctrl->device;
3709 	ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3710 			   ns->ctrl->instance, ns->head->instance);
3711 	if (ret)
3712 		return ret;
3713 
3714 	return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3715 			     ns->ctrl->ops->module);
3716 }
3717 
3718 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3719 		unsigned nsid, struct nvme_ns_ids *ids)
3720 {
3721 	struct nvme_ns_head *head;
3722 	size_t size = sizeof(*head);
3723 	int ret = -ENOMEM;
3724 
3725 #ifdef CONFIG_NVME_MULTIPATH
3726 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3727 #endif
3728 
3729 	head = kzalloc(size, GFP_KERNEL);
3730 	if (!head)
3731 		goto out;
3732 	ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3733 	if (ret < 0)
3734 		goto out_free_head;
3735 	head->instance = ret;
3736 	INIT_LIST_HEAD(&head->list);
3737 	ret = init_srcu_struct(&head->srcu);
3738 	if (ret)
3739 		goto out_ida_remove;
3740 	head->subsys = ctrl->subsys;
3741 	head->ns_id = nsid;
3742 	head->ids = *ids;
3743 	kref_init(&head->ref);
3744 
3745 	if (head->ids.csi) {
3746 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3747 		if (ret)
3748 			goto out_cleanup_srcu;
3749 	} else
3750 		head->effects = ctrl->effects;
3751 
3752 	ret = nvme_mpath_alloc_disk(ctrl, head);
3753 	if (ret)
3754 		goto out_cleanup_srcu;
3755 
3756 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3757 
3758 	kref_get(&ctrl->subsys->ref);
3759 
3760 	return head;
3761 out_cleanup_srcu:
3762 	cleanup_srcu_struct(&head->srcu);
3763 out_ida_remove:
3764 	ida_free(&ctrl->subsys->ns_ida, head->instance);
3765 out_free_head:
3766 	kfree(head);
3767 out:
3768 	if (ret > 0)
3769 		ret = blk_status_to_errno(nvme_error_status(ret));
3770 	return ERR_PTR(ret);
3771 }
3772 
3773 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3774 		struct nvme_ns_ids *ids)
3775 {
3776 	struct nvme_subsystem *s;
3777 	int ret = 0;
3778 
3779 	/*
3780 	 * Note that this check is racy as we try to avoid holding the global
3781 	 * lock over the whole ns_head creation.  But it is only intended as
3782 	 * a sanity check anyway.
3783 	 */
3784 	mutex_lock(&nvme_subsystems_lock);
3785 	list_for_each_entry(s, &nvme_subsystems, entry) {
3786 		if (s == this)
3787 			continue;
3788 		mutex_lock(&s->lock);
3789 		ret = nvme_subsys_check_duplicate_ids(s, ids);
3790 		mutex_unlock(&s->lock);
3791 		if (ret)
3792 			break;
3793 	}
3794 	mutex_unlock(&nvme_subsystems_lock);
3795 
3796 	return ret;
3797 }
3798 
3799 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3800 		struct nvme_ns_ids *ids, bool is_shared)
3801 {
3802 	struct nvme_ctrl *ctrl = ns->ctrl;
3803 	struct nvme_ns_head *head = NULL;
3804 	int ret;
3805 
3806 	ret = nvme_global_check_duplicate_ids(ctrl->subsys, ids);
3807 	if (ret) {
3808 		dev_err(ctrl->device,
3809 			"globally duplicate IDs for nsid %d\n", nsid);
3810 		return ret;
3811 	}
3812 
3813 	mutex_lock(&ctrl->subsys->lock);
3814 	head = nvme_find_ns_head(ctrl, nsid);
3815 	if (!head) {
3816 		ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, ids);
3817 		if (ret) {
3818 			dev_err(ctrl->device,
3819 				"duplicate IDs in subsystem for nsid %d\n",
3820 				nsid);
3821 			goto out_unlock;
3822 		}
3823 		head = nvme_alloc_ns_head(ctrl, nsid, ids);
3824 		if (IS_ERR(head)) {
3825 			ret = PTR_ERR(head);
3826 			goto out_unlock;
3827 		}
3828 		head->shared = is_shared;
3829 	} else {
3830 		ret = -EINVAL;
3831 		if (!is_shared || !head->shared) {
3832 			dev_err(ctrl->device,
3833 				"Duplicate unshared namespace %d\n", nsid);
3834 			goto out_put_ns_head;
3835 		}
3836 		if (!nvme_ns_ids_equal(&head->ids, ids)) {
3837 			dev_err(ctrl->device,
3838 				"IDs don't match for shared namespace %d\n",
3839 					nsid);
3840 			goto out_put_ns_head;
3841 		}
3842 
3843 		if (!multipath && !list_empty(&head->list)) {
3844 			dev_warn(ctrl->device,
3845 				"Found shared namespace %d, but multipathing not supported.\n",
3846 				nsid);
3847 			dev_warn_once(ctrl->device,
3848 				"Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
3849 		}
3850 	}
3851 
3852 	list_add_tail_rcu(&ns->siblings, &head->list);
3853 	ns->head = head;
3854 	mutex_unlock(&ctrl->subsys->lock);
3855 	return 0;
3856 
3857 out_put_ns_head:
3858 	nvme_put_ns_head(head);
3859 out_unlock:
3860 	mutex_unlock(&ctrl->subsys->lock);
3861 	return ret;
3862 }
3863 
3864 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3865 {
3866 	struct nvme_ns *ns, *ret = NULL;
3867 
3868 	down_read(&ctrl->namespaces_rwsem);
3869 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3870 		if (ns->head->ns_id == nsid) {
3871 			if (!nvme_get_ns(ns))
3872 				continue;
3873 			ret = ns;
3874 			break;
3875 		}
3876 		if (ns->head->ns_id > nsid)
3877 			break;
3878 	}
3879 	up_read(&ctrl->namespaces_rwsem);
3880 	return ret;
3881 }
3882 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3883 
3884 /*
3885  * Add the namespace to the controller list while keeping the list ordered.
3886  */
3887 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3888 {
3889 	struct nvme_ns *tmp;
3890 
3891 	list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3892 		if (tmp->head->ns_id < ns->head->ns_id) {
3893 			list_add(&ns->list, &tmp->list);
3894 			return;
3895 		}
3896 	}
3897 	list_add(&ns->list, &ns->ctrl->namespaces);
3898 }
3899 
3900 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3901 		struct nvme_ns_ids *ids)
3902 {
3903 	struct nvme_ns *ns;
3904 	struct gendisk *disk;
3905 	struct nvme_id_ns *id;
3906 	int node = ctrl->numa_node;
3907 
3908 	if (nvme_identify_ns(ctrl, nsid, ids, &id))
3909 		return;
3910 
3911 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3912 	if (!ns)
3913 		goto out_free_id;
3914 
3915 	disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3916 	if (IS_ERR(disk))
3917 		goto out_free_ns;
3918 	disk->fops = &nvme_bdev_ops;
3919 	disk->private_data = ns;
3920 
3921 	ns->disk = disk;
3922 	ns->queue = disk->queue;
3923 
3924 	if (ctrl->opts && ctrl->opts->data_digest)
3925 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3926 
3927 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3928 	if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3929 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3930 
3931 	ns->ctrl = ctrl;
3932 	kref_init(&ns->kref);
3933 
3934 	if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3935 		goto out_cleanup_disk;
3936 
3937 	/*
3938 	 * If multipathing is enabled, the device name for all disks and not
3939 	 * just those that represent shared namespaces needs to be based on the
3940 	 * subsystem instance.  Using the controller instance for private
3941 	 * namespaces could lead to naming collisions between shared and private
3942 	 * namespaces if they don't use a common numbering scheme.
3943 	 *
3944 	 * If multipathing is not enabled, disk names must use the controller
3945 	 * instance as shared namespaces will show up as multiple block
3946 	 * devices.
3947 	 */
3948 	if (ns->head->disk) {
3949 		sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
3950 			ctrl->instance, ns->head->instance);
3951 		disk->flags |= GENHD_FL_HIDDEN;
3952 	} else if (multipath) {
3953 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
3954 			ns->head->instance);
3955 	} else {
3956 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3957 			ns->head->instance);
3958 	}
3959 
3960 	if (nvme_update_ns_info(ns, id))
3961 		goto out_unlink_ns;
3962 
3963 	down_write(&ctrl->namespaces_rwsem);
3964 	nvme_ns_add_to_ctrl_list(ns);
3965 	up_write(&ctrl->namespaces_rwsem);
3966 	nvme_get_ctrl(ctrl);
3967 
3968 	if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
3969 		goto out_cleanup_ns_from_list;
3970 
3971 	if (!nvme_ns_head_multipath(ns->head))
3972 		nvme_add_ns_cdev(ns);
3973 
3974 	nvme_mpath_add_disk(ns, id);
3975 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3976 	kfree(id);
3977 
3978 	return;
3979 
3980  out_cleanup_ns_from_list:
3981 	nvme_put_ctrl(ctrl);
3982 	down_write(&ctrl->namespaces_rwsem);
3983 	list_del_init(&ns->list);
3984 	up_write(&ctrl->namespaces_rwsem);
3985  out_unlink_ns:
3986 	mutex_lock(&ctrl->subsys->lock);
3987 	list_del_rcu(&ns->siblings);
3988 	if (list_empty(&ns->head->list))
3989 		list_del_init(&ns->head->entry);
3990 	mutex_unlock(&ctrl->subsys->lock);
3991 	nvme_put_ns_head(ns->head);
3992  out_cleanup_disk:
3993 	blk_cleanup_disk(disk);
3994  out_free_ns:
3995 	kfree(ns);
3996  out_free_id:
3997 	kfree(id);
3998 }
3999 
4000 static void nvme_ns_remove(struct nvme_ns *ns)
4001 {
4002 	bool last_path = false;
4003 
4004 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4005 		return;
4006 
4007 	clear_bit(NVME_NS_READY, &ns->flags);
4008 	set_capacity(ns->disk, 0);
4009 	nvme_fault_inject_fini(&ns->fault_inject);
4010 
4011 	/*
4012 	 * Ensure that !NVME_NS_READY is seen by other threads to prevent
4013 	 * this ns going back into current_path.
4014 	 */
4015 	synchronize_srcu(&ns->head->srcu);
4016 
4017 	/* wait for concurrent submissions */
4018 	if (nvme_mpath_clear_current_path(ns))
4019 		synchronize_srcu(&ns->head->srcu);
4020 
4021 	mutex_lock(&ns->ctrl->subsys->lock);
4022 	list_del_rcu(&ns->siblings);
4023 	if (list_empty(&ns->head->list)) {
4024 		list_del_init(&ns->head->entry);
4025 		last_path = true;
4026 	}
4027 	mutex_unlock(&ns->ctrl->subsys->lock);
4028 
4029 	/* guarantee not available in head->list */
4030 	synchronize_rcu();
4031 
4032 	if (!nvme_ns_head_multipath(ns->head))
4033 		nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4034 	del_gendisk(ns->disk);
4035 	blk_cleanup_queue(ns->queue);
4036 
4037 	down_write(&ns->ctrl->namespaces_rwsem);
4038 	list_del_init(&ns->list);
4039 	up_write(&ns->ctrl->namespaces_rwsem);
4040 
4041 	if (last_path)
4042 		nvme_mpath_shutdown_disk(ns->head);
4043 	nvme_put_ns(ns);
4044 }
4045 
4046 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4047 {
4048 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4049 
4050 	if (ns) {
4051 		nvme_ns_remove(ns);
4052 		nvme_put_ns(ns);
4053 	}
4054 }
4055 
4056 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4057 {
4058 	struct nvme_id_ns *id;
4059 	int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4060 
4061 	if (test_bit(NVME_NS_DEAD, &ns->flags))
4062 		goto out;
4063 
4064 	ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4065 	if (ret)
4066 		goto out;
4067 
4068 	ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4069 	if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4070 		dev_err(ns->ctrl->device,
4071 			"identifiers changed for nsid %d\n", ns->head->ns_id);
4072 		goto out_free_id;
4073 	}
4074 
4075 	ret = nvme_update_ns_info(ns, id);
4076 
4077 out_free_id:
4078 	kfree(id);
4079 out:
4080 	/*
4081 	 * Only remove the namespace if we got a fatal error back from the
4082 	 * device, otherwise ignore the error and just move on.
4083 	 *
4084 	 * TODO: we should probably schedule a delayed retry here.
4085 	 */
4086 	if (ret > 0 && (ret & NVME_SC_DNR))
4087 		nvme_ns_remove(ns);
4088 }
4089 
4090 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4091 {
4092 	struct nvme_ns_ids ids = { };
4093 	struct nvme_ns *ns;
4094 
4095 	if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4096 		return;
4097 
4098 	ns = nvme_find_get_ns(ctrl, nsid);
4099 	if (ns) {
4100 		nvme_validate_ns(ns, &ids);
4101 		nvme_put_ns(ns);
4102 		return;
4103 	}
4104 
4105 	switch (ids.csi) {
4106 	case NVME_CSI_NVM:
4107 		nvme_alloc_ns(ctrl, nsid, &ids);
4108 		break;
4109 	case NVME_CSI_ZNS:
4110 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4111 			dev_warn(ctrl->device,
4112 				"nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4113 				nsid);
4114 			break;
4115 		}
4116 		if (!nvme_multi_css(ctrl)) {
4117 			dev_warn(ctrl->device,
4118 				"command set not reported for nsid: %d\n",
4119 				nsid);
4120 			break;
4121 		}
4122 		nvme_alloc_ns(ctrl, nsid, &ids);
4123 		break;
4124 	default:
4125 		dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4126 			ids.csi, nsid);
4127 		break;
4128 	}
4129 }
4130 
4131 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4132 					unsigned nsid)
4133 {
4134 	struct nvme_ns *ns, *next;
4135 	LIST_HEAD(rm_list);
4136 
4137 	down_write(&ctrl->namespaces_rwsem);
4138 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4139 		if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4140 			list_move_tail(&ns->list, &rm_list);
4141 	}
4142 	up_write(&ctrl->namespaces_rwsem);
4143 
4144 	list_for_each_entry_safe(ns, next, &rm_list, list)
4145 		nvme_ns_remove(ns);
4146 
4147 }
4148 
4149 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4150 {
4151 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4152 	__le32 *ns_list;
4153 	u32 prev = 0;
4154 	int ret = 0, i;
4155 
4156 	if (nvme_ctrl_limited_cns(ctrl))
4157 		return -EOPNOTSUPP;
4158 
4159 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4160 	if (!ns_list)
4161 		return -ENOMEM;
4162 
4163 	for (;;) {
4164 		struct nvme_command cmd = {
4165 			.identify.opcode	= nvme_admin_identify,
4166 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
4167 			.identify.nsid		= cpu_to_le32(prev),
4168 		};
4169 
4170 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4171 					    NVME_IDENTIFY_DATA_SIZE);
4172 		if (ret) {
4173 			dev_warn(ctrl->device,
4174 				"Identify NS List failed (status=0x%x)\n", ret);
4175 			goto free;
4176 		}
4177 
4178 		for (i = 0; i < nr_entries; i++) {
4179 			u32 nsid = le32_to_cpu(ns_list[i]);
4180 
4181 			if (!nsid)	/* end of the list? */
4182 				goto out;
4183 			nvme_validate_or_alloc_ns(ctrl, nsid);
4184 			while (++prev < nsid)
4185 				nvme_ns_remove_by_nsid(ctrl, prev);
4186 		}
4187 	}
4188  out:
4189 	nvme_remove_invalid_namespaces(ctrl, prev);
4190  free:
4191 	kfree(ns_list);
4192 	return ret;
4193 }
4194 
4195 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4196 {
4197 	struct nvme_id_ctrl *id;
4198 	u32 nn, i;
4199 
4200 	if (nvme_identify_ctrl(ctrl, &id))
4201 		return;
4202 	nn = le32_to_cpu(id->nn);
4203 	kfree(id);
4204 
4205 	for (i = 1; i <= nn; i++)
4206 		nvme_validate_or_alloc_ns(ctrl, i);
4207 
4208 	nvme_remove_invalid_namespaces(ctrl, nn);
4209 }
4210 
4211 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4212 {
4213 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4214 	__le32 *log;
4215 	int error;
4216 
4217 	log = kzalloc(log_size, GFP_KERNEL);
4218 	if (!log)
4219 		return;
4220 
4221 	/*
4222 	 * We need to read the log to clear the AEN, but we don't want to rely
4223 	 * on it for the changed namespace information as userspace could have
4224 	 * raced with us in reading the log page, which could cause us to miss
4225 	 * updates.
4226 	 */
4227 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4228 			NVME_CSI_NVM, log, log_size, 0);
4229 	if (error)
4230 		dev_warn(ctrl->device,
4231 			"reading changed ns log failed: %d\n", error);
4232 
4233 	kfree(log);
4234 }
4235 
4236 static void nvme_scan_work(struct work_struct *work)
4237 {
4238 	struct nvme_ctrl *ctrl =
4239 		container_of(work, struct nvme_ctrl, scan_work);
4240 
4241 	/* No tagset on a live ctrl means IO queues could not created */
4242 	if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4243 		return;
4244 
4245 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4246 		dev_info(ctrl->device, "rescanning namespaces.\n");
4247 		nvme_clear_changed_ns_log(ctrl);
4248 	}
4249 
4250 	mutex_lock(&ctrl->scan_lock);
4251 	if (nvme_scan_ns_list(ctrl) != 0)
4252 		nvme_scan_ns_sequential(ctrl);
4253 	mutex_unlock(&ctrl->scan_lock);
4254 }
4255 
4256 /*
4257  * This function iterates the namespace list unlocked to allow recovery from
4258  * controller failure. It is up to the caller to ensure the namespace list is
4259  * not modified by scan work while this function is executing.
4260  */
4261 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4262 {
4263 	struct nvme_ns *ns, *next;
4264 	LIST_HEAD(ns_list);
4265 
4266 	/*
4267 	 * make sure to requeue I/O to all namespaces as these
4268 	 * might result from the scan itself and must complete
4269 	 * for the scan_work to make progress
4270 	 */
4271 	nvme_mpath_clear_ctrl_paths(ctrl);
4272 
4273 	/* prevent racing with ns scanning */
4274 	flush_work(&ctrl->scan_work);
4275 
4276 	/*
4277 	 * The dead states indicates the controller was not gracefully
4278 	 * disconnected. In that case, we won't be able to flush any data while
4279 	 * removing the namespaces' disks; fail all the queues now to avoid
4280 	 * potentially having to clean up the failed sync later.
4281 	 */
4282 	if (ctrl->state == NVME_CTRL_DEAD)
4283 		nvme_kill_queues(ctrl);
4284 
4285 	/* this is a no-op when called from the controller reset handler */
4286 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4287 
4288 	down_write(&ctrl->namespaces_rwsem);
4289 	list_splice_init(&ctrl->namespaces, &ns_list);
4290 	up_write(&ctrl->namespaces_rwsem);
4291 
4292 	list_for_each_entry_safe(ns, next, &ns_list, list)
4293 		nvme_ns_remove(ns);
4294 }
4295 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4296 
4297 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4298 {
4299 	struct nvme_ctrl *ctrl =
4300 		container_of(dev, struct nvme_ctrl, ctrl_device);
4301 	struct nvmf_ctrl_options *opts = ctrl->opts;
4302 	int ret;
4303 
4304 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4305 	if (ret)
4306 		return ret;
4307 
4308 	if (opts) {
4309 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4310 		if (ret)
4311 			return ret;
4312 
4313 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4314 				opts->trsvcid ?: "none");
4315 		if (ret)
4316 			return ret;
4317 
4318 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4319 				opts->host_traddr ?: "none");
4320 		if (ret)
4321 			return ret;
4322 
4323 		ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4324 				opts->host_iface ?: "none");
4325 	}
4326 	return ret;
4327 }
4328 
4329 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4330 {
4331 	char *envp[2] = { envdata, NULL };
4332 
4333 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4334 }
4335 
4336 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4337 {
4338 	char *envp[2] = { NULL, NULL };
4339 	u32 aen_result = ctrl->aen_result;
4340 
4341 	ctrl->aen_result = 0;
4342 	if (!aen_result)
4343 		return;
4344 
4345 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4346 	if (!envp[0])
4347 		return;
4348 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4349 	kfree(envp[0]);
4350 }
4351 
4352 static void nvme_async_event_work(struct work_struct *work)
4353 {
4354 	struct nvme_ctrl *ctrl =
4355 		container_of(work, struct nvme_ctrl, async_event_work);
4356 
4357 	nvme_aen_uevent(ctrl);
4358 
4359 	/*
4360 	 * The transport drivers must guarantee AER submission here is safe by
4361 	 * flushing ctrl async_event_work after changing the controller state
4362 	 * from LIVE and before freeing the admin queue.
4363 	*/
4364 	if (ctrl->state == NVME_CTRL_LIVE)
4365 		ctrl->ops->submit_async_event(ctrl);
4366 }
4367 
4368 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4369 {
4370 
4371 	u32 csts;
4372 
4373 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4374 		return false;
4375 
4376 	if (csts == ~0)
4377 		return false;
4378 
4379 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4380 }
4381 
4382 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4383 {
4384 	struct nvme_fw_slot_info_log *log;
4385 
4386 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4387 	if (!log)
4388 		return;
4389 
4390 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4391 			log, sizeof(*log), 0))
4392 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4393 	kfree(log);
4394 }
4395 
4396 static void nvme_fw_act_work(struct work_struct *work)
4397 {
4398 	struct nvme_ctrl *ctrl = container_of(work,
4399 				struct nvme_ctrl, fw_act_work);
4400 	unsigned long fw_act_timeout;
4401 
4402 	if (ctrl->mtfa)
4403 		fw_act_timeout = jiffies +
4404 				msecs_to_jiffies(ctrl->mtfa * 100);
4405 	else
4406 		fw_act_timeout = jiffies +
4407 				msecs_to_jiffies(admin_timeout * 1000);
4408 
4409 	nvme_stop_queues(ctrl);
4410 	while (nvme_ctrl_pp_status(ctrl)) {
4411 		if (time_after(jiffies, fw_act_timeout)) {
4412 			dev_warn(ctrl->device,
4413 				"Fw activation timeout, reset controller\n");
4414 			nvme_try_sched_reset(ctrl);
4415 			return;
4416 		}
4417 		msleep(100);
4418 	}
4419 
4420 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4421 		return;
4422 
4423 	nvme_start_queues(ctrl);
4424 	/* read FW slot information to clear the AER */
4425 	nvme_get_fw_slot_info(ctrl);
4426 }
4427 
4428 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4429 {
4430 	u32 aer_notice_type = (result & 0xff00) >> 8;
4431 
4432 	trace_nvme_async_event(ctrl, aer_notice_type);
4433 
4434 	switch (aer_notice_type) {
4435 	case NVME_AER_NOTICE_NS_CHANGED:
4436 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4437 		nvme_queue_scan(ctrl);
4438 		break;
4439 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4440 		/*
4441 		 * We are (ab)using the RESETTING state to prevent subsequent
4442 		 * recovery actions from interfering with the controller's
4443 		 * firmware activation.
4444 		 */
4445 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4446 			queue_work(nvme_wq, &ctrl->fw_act_work);
4447 		break;
4448 #ifdef CONFIG_NVME_MULTIPATH
4449 	case NVME_AER_NOTICE_ANA:
4450 		if (!ctrl->ana_log_buf)
4451 			break;
4452 		queue_work(nvme_wq, &ctrl->ana_work);
4453 		break;
4454 #endif
4455 	case NVME_AER_NOTICE_DISC_CHANGED:
4456 		ctrl->aen_result = result;
4457 		break;
4458 	default:
4459 		dev_warn(ctrl->device, "async event result %08x\n", result);
4460 	}
4461 }
4462 
4463 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4464 		volatile union nvme_result *res)
4465 {
4466 	u32 result = le32_to_cpu(res->u32);
4467 	u32 aer_type = result & 0x07;
4468 
4469 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4470 		return;
4471 
4472 	switch (aer_type) {
4473 	case NVME_AER_NOTICE:
4474 		nvme_handle_aen_notice(ctrl, result);
4475 		break;
4476 	case NVME_AER_ERROR:
4477 	case NVME_AER_SMART:
4478 	case NVME_AER_CSS:
4479 	case NVME_AER_VS:
4480 		trace_nvme_async_event(ctrl, aer_type);
4481 		ctrl->aen_result = result;
4482 		break;
4483 	default:
4484 		break;
4485 	}
4486 	queue_work(nvme_wq, &ctrl->async_event_work);
4487 }
4488 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4489 
4490 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4491 {
4492 	nvme_mpath_stop(ctrl);
4493 	nvme_stop_keep_alive(ctrl);
4494 	nvme_stop_failfast_work(ctrl);
4495 	flush_work(&ctrl->async_event_work);
4496 	cancel_work_sync(&ctrl->fw_act_work);
4497 }
4498 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4499 
4500 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4501 {
4502 	nvme_start_keep_alive(ctrl);
4503 
4504 	nvme_enable_aen(ctrl);
4505 
4506 	if (ctrl->queue_count > 1) {
4507 		nvme_queue_scan(ctrl);
4508 		nvme_start_queues(ctrl);
4509 		nvme_mpath_update(ctrl);
4510 	}
4511 
4512 	nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4513 }
4514 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4515 
4516 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4517 {
4518 	nvme_hwmon_exit(ctrl);
4519 	nvme_fault_inject_fini(&ctrl->fault_inject);
4520 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4521 	cdev_device_del(&ctrl->cdev, ctrl->device);
4522 	nvme_put_ctrl(ctrl);
4523 }
4524 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4525 
4526 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4527 {
4528 	struct nvme_effects_log	*cel;
4529 	unsigned long i;
4530 
4531 	xa_for_each(&ctrl->cels, i, cel) {
4532 		xa_erase(&ctrl->cels, i);
4533 		kfree(cel);
4534 	}
4535 
4536 	xa_destroy(&ctrl->cels);
4537 }
4538 
4539 static void nvme_free_ctrl(struct device *dev)
4540 {
4541 	struct nvme_ctrl *ctrl =
4542 		container_of(dev, struct nvme_ctrl, ctrl_device);
4543 	struct nvme_subsystem *subsys = ctrl->subsys;
4544 
4545 	if (!subsys || ctrl->instance != subsys->instance)
4546 		ida_free(&nvme_instance_ida, ctrl->instance);
4547 
4548 	nvme_free_cels(ctrl);
4549 	nvme_mpath_uninit(ctrl);
4550 	__free_page(ctrl->discard_page);
4551 
4552 	if (subsys) {
4553 		mutex_lock(&nvme_subsystems_lock);
4554 		list_del(&ctrl->subsys_entry);
4555 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4556 		mutex_unlock(&nvme_subsystems_lock);
4557 	}
4558 
4559 	ctrl->ops->free_ctrl(ctrl);
4560 
4561 	if (subsys)
4562 		nvme_put_subsystem(subsys);
4563 }
4564 
4565 /*
4566  * Initialize a NVMe controller structures.  This needs to be called during
4567  * earliest initialization so that we have the initialized structured around
4568  * during probing.
4569  */
4570 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4571 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
4572 {
4573 	int ret;
4574 
4575 	ctrl->state = NVME_CTRL_NEW;
4576 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4577 	spin_lock_init(&ctrl->lock);
4578 	mutex_init(&ctrl->scan_lock);
4579 	INIT_LIST_HEAD(&ctrl->namespaces);
4580 	xa_init(&ctrl->cels);
4581 	init_rwsem(&ctrl->namespaces_rwsem);
4582 	ctrl->dev = dev;
4583 	ctrl->ops = ops;
4584 	ctrl->quirks = quirks;
4585 	ctrl->numa_node = NUMA_NO_NODE;
4586 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4587 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4588 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4589 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4590 	init_waitqueue_head(&ctrl->state_wq);
4591 
4592 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4593 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4594 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4595 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4596 
4597 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4598 			PAGE_SIZE);
4599 	ctrl->discard_page = alloc_page(GFP_KERNEL);
4600 	if (!ctrl->discard_page) {
4601 		ret = -ENOMEM;
4602 		goto out;
4603 	}
4604 
4605 	ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4606 	if (ret < 0)
4607 		goto out;
4608 	ctrl->instance = ret;
4609 
4610 	device_initialize(&ctrl->ctrl_device);
4611 	ctrl->device = &ctrl->ctrl_device;
4612 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4613 			ctrl->instance);
4614 	ctrl->device->class = nvme_class;
4615 	ctrl->device->parent = ctrl->dev;
4616 	ctrl->device->groups = nvme_dev_attr_groups;
4617 	ctrl->device->release = nvme_free_ctrl;
4618 	dev_set_drvdata(ctrl->device, ctrl);
4619 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4620 	if (ret)
4621 		goto out_release_instance;
4622 
4623 	nvme_get_ctrl(ctrl);
4624 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
4625 	ctrl->cdev.owner = ops->module;
4626 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4627 	if (ret)
4628 		goto out_free_name;
4629 
4630 	/*
4631 	 * Initialize latency tolerance controls.  The sysfs files won't
4632 	 * be visible to userspace unless the device actually supports APST.
4633 	 */
4634 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4635 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4636 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4637 
4638 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4639 	nvme_mpath_init_ctrl(ctrl);
4640 
4641 	return 0;
4642 out_free_name:
4643 	nvme_put_ctrl(ctrl);
4644 	kfree_const(ctrl->device->kobj.name);
4645 out_release_instance:
4646 	ida_free(&nvme_instance_ida, ctrl->instance);
4647 out:
4648 	if (ctrl->discard_page)
4649 		__free_page(ctrl->discard_page);
4650 	return ret;
4651 }
4652 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4653 
4654 static void nvme_start_ns_queue(struct nvme_ns *ns)
4655 {
4656 	if (test_and_clear_bit(NVME_NS_STOPPED, &ns->flags))
4657 		blk_mq_unquiesce_queue(ns->queue);
4658 }
4659 
4660 static void nvme_stop_ns_queue(struct nvme_ns *ns)
4661 {
4662 	if (!test_and_set_bit(NVME_NS_STOPPED, &ns->flags))
4663 		blk_mq_quiesce_queue(ns->queue);
4664 	else
4665 		blk_mq_wait_quiesce_done(ns->queue);
4666 }
4667 
4668 /*
4669  * Prepare a queue for teardown.
4670  *
4671  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
4672  * the capacity to 0 after that to avoid blocking dispatchers that may be
4673  * holding bd_butex.  This will end buffered writers dirtying pages that can't
4674  * be synced.
4675  */
4676 static void nvme_set_queue_dying(struct nvme_ns *ns)
4677 {
4678 	if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
4679 		return;
4680 
4681 	blk_mark_disk_dead(ns->disk);
4682 	nvme_start_ns_queue(ns);
4683 
4684 	set_capacity_and_notify(ns->disk, 0);
4685 }
4686 
4687 /**
4688  * nvme_kill_queues(): Ends all namespace queues
4689  * @ctrl: the dead controller that needs to end
4690  *
4691  * Call this function when the driver determines it is unable to get the
4692  * controller in a state capable of servicing IO.
4693  */
4694 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4695 {
4696 	struct nvme_ns *ns;
4697 
4698 	down_read(&ctrl->namespaces_rwsem);
4699 
4700 	/* Forcibly unquiesce queues to avoid blocking dispatch */
4701 	if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4702 		nvme_start_admin_queue(ctrl);
4703 
4704 	list_for_each_entry(ns, &ctrl->namespaces, list)
4705 		nvme_set_queue_dying(ns);
4706 
4707 	up_read(&ctrl->namespaces_rwsem);
4708 }
4709 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4710 
4711 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4712 {
4713 	struct nvme_ns *ns;
4714 
4715 	down_read(&ctrl->namespaces_rwsem);
4716 	list_for_each_entry(ns, &ctrl->namespaces, list)
4717 		blk_mq_unfreeze_queue(ns->queue);
4718 	up_read(&ctrl->namespaces_rwsem);
4719 }
4720 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4721 
4722 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4723 {
4724 	struct nvme_ns *ns;
4725 
4726 	down_read(&ctrl->namespaces_rwsem);
4727 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4728 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4729 		if (timeout <= 0)
4730 			break;
4731 	}
4732 	up_read(&ctrl->namespaces_rwsem);
4733 	return timeout;
4734 }
4735 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4736 
4737 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4738 {
4739 	struct nvme_ns *ns;
4740 
4741 	down_read(&ctrl->namespaces_rwsem);
4742 	list_for_each_entry(ns, &ctrl->namespaces, list)
4743 		blk_mq_freeze_queue_wait(ns->queue);
4744 	up_read(&ctrl->namespaces_rwsem);
4745 }
4746 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4747 
4748 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4749 {
4750 	struct nvme_ns *ns;
4751 
4752 	down_read(&ctrl->namespaces_rwsem);
4753 	list_for_each_entry(ns, &ctrl->namespaces, list)
4754 		blk_freeze_queue_start(ns->queue);
4755 	up_read(&ctrl->namespaces_rwsem);
4756 }
4757 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4758 
4759 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4760 {
4761 	struct nvme_ns *ns;
4762 
4763 	down_read(&ctrl->namespaces_rwsem);
4764 	list_for_each_entry(ns, &ctrl->namespaces, list)
4765 		nvme_stop_ns_queue(ns);
4766 	up_read(&ctrl->namespaces_rwsem);
4767 }
4768 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4769 
4770 void nvme_start_queues(struct nvme_ctrl *ctrl)
4771 {
4772 	struct nvme_ns *ns;
4773 
4774 	down_read(&ctrl->namespaces_rwsem);
4775 	list_for_each_entry(ns, &ctrl->namespaces, list)
4776 		nvme_start_ns_queue(ns);
4777 	up_read(&ctrl->namespaces_rwsem);
4778 }
4779 EXPORT_SYMBOL_GPL(nvme_start_queues);
4780 
4781 void nvme_stop_admin_queue(struct nvme_ctrl *ctrl)
4782 {
4783 	if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4784 		blk_mq_quiesce_queue(ctrl->admin_q);
4785 	else
4786 		blk_mq_wait_quiesce_done(ctrl->admin_q);
4787 }
4788 EXPORT_SYMBOL_GPL(nvme_stop_admin_queue);
4789 
4790 void nvme_start_admin_queue(struct nvme_ctrl *ctrl)
4791 {
4792 	if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4793 		blk_mq_unquiesce_queue(ctrl->admin_q);
4794 }
4795 EXPORT_SYMBOL_GPL(nvme_start_admin_queue);
4796 
4797 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4798 {
4799 	struct nvme_ns *ns;
4800 
4801 	down_read(&ctrl->namespaces_rwsem);
4802 	list_for_each_entry(ns, &ctrl->namespaces, list)
4803 		blk_sync_queue(ns->queue);
4804 	up_read(&ctrl->namespaces_rwsem);
4805 }
4806 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4807 
4808 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4809 {
4810 	nvme_sync_io_queues(ctrl);
4811 	if (ctrl->admin_q)
4812 		blk_sync_queue(ctrl->admin_q);
4813 }
4814 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4815 
4816 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4817 {
4818 	if (file->f_op != &nvme_dev_fops)
4819 		return NULL;
4820 	return file->private_data;
4821 }
4822 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4823 
4824 /*
4825  * Check we didn't inadvertently grow the command structure sizes:
4826  */
4827 static inline void _nvme_check_size(void)
4828 {
4829 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4830 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4831 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4832 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4833 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4834 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4835 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4836 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4837 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4838 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4839 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4840 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4841 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4842 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4843 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
4844 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4845 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4846 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4847 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4848 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4849 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4850 	BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
4851 }
4852 
4853 
4854 static int __init nvme_core_init(void)
4855 {
4856 	int result = -ENOMEM;
4857 
4858 	_nvme_check_size();
4859 
4860 	nvme_wq = alloc_workqueue("nvme-wq",
4861 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4862 	if (!nvme_wq)
4863 		goto out;
4864 
4865 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4866 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4867 	if (!nvme_reset_wq)
4868 		goto destroy_wq;
4869 
4870 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4871 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4872 	if (!nvme_delete_wq)
4873 		goto destroy_reset_wq;
4874 
4875 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4876 			NVME_MINORS, "nvme");
4877 	if (result < 0)
4878 		goto destroy_delete_wq;
4879 
4880 	nvme_class = class_create(THIS_MODULE, "nvme");
4881 	if (IS_ERR(nvme_class)) {
4882 		result = PTR_ERR(nvme_class);
4883 		goto unregister_chrdev;
4884 	}
4885 	nvme_class->dev_uevent = nvme_class_uevent;
4886 
4887 	nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4888 	if (IS_ERR(nvme_subsys_class)) {
4889 		result = PTR_ERR(nvme_subsys_class);
4890 		goto destroy_class;
4891 	}
4892 
4893 	result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4894 				     "nvme-generic");
4895 	if (result < 0)
4896 		goto destroy_subsys_class;
4897 
4898 	nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
4899 	if (IS_ERR(nvme_ns_chr_class)) {
4900 		result = PTR_ERR(nvme_ns_chr_class);
4901 		goto unregister_generic_ns;
4902 	}
4903 
4904 	return 0;
4905 
4906 unregister_generic_ns:
4907 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4908 destroy_subsys_class:
4909 	class_destroy(nvme_subsys_class);
4910 destroy_class:
4911 	class_destroy(nvme_class);
4912 unregister_chrdev:
4913 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4914 destroy_delete_wq:
4915 	destroy_workqueue(nvme_delete_wq);
4916 destroy_reset_wq:
4917 	destroy_workqueue(nvme_reset_wq);
4918 destroy_wq:
4919 	destroy_workqueue(nvme_wq);
4920 out:
4921 	return result;
4922 }
4923 
4924 static void __exit nvme_core_exit(void)
4925 {
4926 	class_destroy(nvme_ns_chr_class);
4927 	class_destroy(nvme_subsys_class);
4928 	class_destroy(nvme_class);
4929 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4930 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4931 	destroy_workqueue(nvme_delete_wq);
4932 	destroy_workqueue(nvme_reset_wq);
4933 	destroy_workqueue(nvme_wq);
4934 	ida_destroy(&nvme_ns_chr_minor_ida);
4935 	ida_destroy(&nvme_instance_ida);
4936 }
4937 
4938 MODULE_LICENSE("GPL");
4939 MODULE_VERSION("1.0");
4940 module_init(nvme_core_init);
4941 module_exit(nvme_core_exit);
4942