xref: /openbmc/linux/drivers/nvme/host/core.c (revision 53e8558837be58c1d44d50ad87247a8c56c95c13)
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 static bool streams;
81 module_param(streams, bool, 0644);
82 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
83 
84 /*
85  * nvme_wq - hosts nvme related works that are not reset or delete
86  * nvme_reset_wq - hosts nvme reset works
87  * nvme_delete_wq - hosts nvme delete works
88  *
89  * nvme_wq will host works such as scan, aen handling, fw activation,
90  * keep-alive, periodic reconnects etc. nvme_reset_wq
91  * runs reset works which also flush works hosted on nvme_wq for
92  * serialization purposes. nvme_delete_wq host controller deletion
93  * works which flush reset works for serialization.
94  */
95 struct workqueue_struct *nvme_wq;
96 EXPORT_SYMBOL_GPL(nvme_wq);
97 
98 struct workqueue_struct *nvme_reset_wq;
99 EXPORT_SYMBOL_GPL(nvme_reset_wq);
100 
101 struct workqueue_struct *nvme_delete_wq;
102 EXPORT_SYMBOL_GPL(nvme_delete_wq);
103 
104 static LIST_HEAD(nvme_subsystems);
105 static DEFINE_MUTEX(nvme_subsystems_lock);
106 
107 static DEFINE_IDA(nvme_instance_ida);
108 static dev_t nvme_ctrl_base_chr_devt;
109 static struct class *nvme_class;
110 static struct class *nvme_subsys_class;
111 
112 static DEFINE_IDA(nvme_ns_chr_minor_ida);
113 static dev_t nvme_ns_chr_devt;
114 static struct class *nvme_ns_chr_class;
115 
116 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
117 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
118 					   unsigned nsid);
119 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
120 				   struct nvme_command *cmd);
121 
122 void nvme_queue_scan(struct nvme_ctrl *ctrl)
123 {
124 	/*
125 	 * Only new queue scan work when admin and IO queues are both alive
126 	 */
127 	if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
128 		queue_work(nvme_wq, &ctrl->scan_work);
129 }
130 
131 /*
132  * Use this function to proceed with scheduling reset_work for a controller
133  * that had previously been set to the resetting state. This is intended for
134  * code paths that can't be interrupted by other reset attempts. A hot removal
135  * may prevent this from succeeding.
136  */
137 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
138 {
139 	if (ctrl->state != NVME_CTRL_RESETTING)
140 		return -EBUSY;
141 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
142 		return -EBUSY;
143 	return 0;
144 }
145 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
146 
147 static void nvme_failfast_work(struct work_struct *work)
148 {
149 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
150 			struct nvme_ctrl, failfast_work);
151 
152 	if (ctrl->state != NVME_CTRL_CONNECTING)
153 		return;
154 
155 	set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
156 	dev_info(ctrl->device, "failfast expired\n");
157 	nvme_kick_requeue_lists(ctrl);
158 }
159 
160 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
161 {
162 	if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
163 		return;
164 
165 	schedule_delayed_work(&ctrl->failfast_work,
166 			      ctrl->opts->fast_io_fail_tmo * HZ);
167 }
168 
169 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
170 {
171 	if (!ctrl->opts)
172 		return;
173 
174 	cancel_delayed_work_sync(&ctrl->failfast_work);
175 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
176 }
177 
178 
179 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
180 {
181 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
182 		return -EBUSY;
183 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
184 		return -EBUSY;
185 	return 0;
186 }
187 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
188 
189 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
190 {
191 	int ret;
192 
193 	ret = nvme_reset_ctrl(ctrl);
194 	if (!ret) {
195 		flush_work(&ctrl->reset_work);
196 		if (ctrl->state != NVME_CTRL_LIVE)
197 			ret = -ENETRESET;
198 	}
199 
200 	return ret;
201 }
202 
203 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
204 {
205 	dev_info(ctrl->device,
206 		 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
207 
208 	flush_work(&ctrl->reset_work);
209 	nvme_stop_ctrl(ctrl);
210 	nvme_remove_namespaces(ctrl);
211 	ctrl->ops->delete_ctrl(ctrl);
212 	nvme_uninit_ctrl(ctrl);
213 }
214 
215 static void nvme_delete_ctrl_work(struct work_struct *work)
216 {
217 	struct nvme_ctrl *ctrl =
218 		container_of(work, struct nvme_ctrl, delete_work);
219 
220 	nvme_do_delete_ctrl(ctrl);
221 }
222 
223 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
224 {
225 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
226 		return -EBUSY;
227 	if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
228 		return -EBUSY;
229 	return 0;
230 }
231 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
232 
233 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
234 {
235 	/*
236 	 * Keep a reference until nvme_do_delete_ctrl() complete,
237 	 * since ->delete_ctrl can free the controller.
238 	 */
239 	nvme_get_ctrl(ctrl);
240 	if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
241 		nvme_do_delete_ctrl(ctrl);
242 	nvme_put_ctrl(ctrl);
243 }
244 
245 static blk_status_t nvme_error_status(u16 status)
246 {
247 	switch (status & 0x7ff) {
248 	case NVME_SC_SUCCESS:
249 		return BLK_STS_OK;
250 	case NVME_SC_CAP_EXCEEDED:
251 		return BLK_STS_NOSPC;
252 	case NVME_SC_LBA_RANGE:
253 	case NVME_SC_CMD_INTERRUPTED:
254 	case NVME_SC_NS_NOT_READY:
255 		return BLK_STS_TARGET;
256 	case NVME_SC_BAD_ATTRIBUTES:
257 	case NVME_SC_ONCS_NOT_SUPPORTED:
258 	case NVME_SC_INVALID_OPCODE:
259 	case NVME_SC_INVALID_FIELD:
260 	case NVME_SC_INVALID_NS:
261 		return BLK_STS_NOTSUPP;
262 	case NVME_SC_WRITE_FAULT:
263 	case NVME_SC_READ_ERROR:
264 	case NVME_SC_UNWRITTEN_BLOCK:
265 	case NVME_SC_ACCESS_DENIED:
266 	case NVME_SC_READ_ONLY:
267 	case NVME_SC_COMPARE_FAILED:
268 		return BLK_STS_MEDIUM;
269 	case NVME_SC_GUARD_CHECK:
270 	case NVME_SC_APPTAG_CHECK:
271 	case NVME_SC_REFTAG_CHECK:
272 	case NVME_SC_INVALID_PI:
273 		return BLK_STS_PROTECTION;
274 	case NVME_SC_RESERVATION_CONFLICT:
275 		return BLK_STS_NEXUS;
276 	case NVME_SC_HOST_PATH_ERROR:
277 		return BLK_STS_TRANSPORT;
278 	case NVME_SC_ZONE_TOO_MANY_ACTIVE:
279 		return BLK_STS_ZONE_ACTIVE_RESOURCE;
280 	case NVME_SC_ZONE_TOO_MANY_OPEN:
281 		return BLK_STS_ZONE_OPEN_RESOURCE;
282 	default:
283 		return BLK_STS_IOERR;
284 	}
285 }
286 
287 static void nvme_retry_req(struct request *req)
288 {
289 	unsigned long delay = 0;
290 	u16 crd;
291 
292 	/* The mask and shift result must be <= 3 */
293 	crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
294 	if (crd)
295 		delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
296 
297 	nvme_req(req)->retries++;
298 	blk_mq_requeue_request(req, false);
299 	blk_mq_delay_kick_requeue_list(req->q, delay);
300 }
301 
302 enum nvme_disposition {
303 	COMPLETE,
304 	RETRY,
305 	FAILOVER,
306 };
307 
308 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
309 {
310 	if (likely(nvme_req(req)->status == 0))
311 		return COMPLETE;
312 
313 	if (blk_noretry_request(req) ||
314 	    (nvme_req(req)->status & NVME_SC_DNR) ||
315 	    nvme_req(req)->retries >= nvme_max_retries)
316 		return COMPLETE;
317 
318 	if (req->cmd_flags & REQ_NVME_MPATH) {
319 		if (nvme_is_path_error(nvme_req(req)->status) ||
320 		    blk_queue_dying(req->q))
321 			return FAILOVER;
322 	} else {
323 		if (blk_queue_dying(req->q))
324 			return COMPLETE;
325 	}
326 
327 	return RETRY;
328 }
329 
330 static inline void nvme_end_req_zoned(struct request *req)
331 {
332 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
333 	    req_op(req) == REQ_OP_ZONE_APPEND)
334 		req->__sector = nvme_lba_to_sect(req->q->queuedata,
335 			le64_to_cpu(nvme_req(req)->result.u64));
336 }
337 
338 static inline void nvme_end_req(struct request *req)
339 {
340 	blk_status_t status = nvme_error_status(nvme_req(req)->status);
341 
342 	nvme_end_req_zoned(req);
343 	nvme_trace_bio_complete(req);
344 	blk_mq_end_request(req, status);
345 }
346 
347 void nvme_complete_rq(struct request *req)
348 {
349 	trace_nvme_complete_rq(req);
350 	nvme_cleanup_cmd(req);
351 
352 	if (nvme_req(req)->ctrl->kas)
353 		nvme_req(req)->ctrl->comp_seen = true;
354 
355 	switch (nvme_decide_disposition(req)) {
356 	case COMPLETE:
357 		nvme_end_req(req);
358 		return;
359 	case RETRY:
360 		nvme_retry_req(req);
361 		return;
362 	case FAILOVER:
363 		nvme_failover_req(req);
364 		return;
365 	}
366 }
367 EXPORT_SYMBOL_GPL(nvme_complete_rq);
368 
369 void nvme_complete_batch_req(struct request *req)
370 {
371 	nvme_cleanup_cmd(req);
372 	nvme_end_req_zoned(req);
373 }
374 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
375 
376 /*
377  * Called to unwind from ->queue_rq on a failed command submission so that the
378  * multipathing code gets called to potentially failover to another path.
379  * The caller needs to unwind all transport specific resource allocations and
380  * must return propagate the return value.
381  */
382 blk_status_t nvme_host_path_error(struct request *req)
383 {
384 	nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
385 	blk_mq_set_request_complete(req);
386 	nvme_complete_rq(req);
387 	return BLK_STS_OK;
388 }
389 EXPORT_SYMBOL_GPL(nvme_host_path_error);
390 
391 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
392 {
393 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
394 				"Cancelling I/O %d", req->tag);
395 
396 	/* don't abort one completed request */
397 	if (blk_mq_request_completed(req))
398 		return true;
399 
400 	nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
401 	nvme_req(req)->flags |= NVME_REQ_CANCELLED;
402 	blk_mq_complete_request(req);
403 	return true;
404 }
405 EXPORT_SYMBOL_GPL(nvme_cancel_request);
406 
407 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
408 {
409 	if (ctrl->tagset) {
410 		blk_mq_tagset_busy_iter(ctrl->tagset,
411 				nvme_cancel_request, ctrl);
412 		blk_mq_tagset_wait_completed_request(ctrl->tagset);
413 	}
414 }
415 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
416 
417 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
418 {
419 	if (ctrl->admin_tagset) {
420 		blk_mq_tagset_busy_iter(ctrl->admin_tagset,
421 				nvme_cancel_request, ctrl);
422 		blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
423 	}
424 }
425 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
426 
427 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
428 		enum nvme_ctrl_state new_state)
429 {
430 	enum nvme_ctrl_state old_state;
431 	unsigned long flags;
432 	bool changed = false;
433 
434 	spin_lock_irqsave(&ctrl->lock, flags);
435 
436 	old_state = ctrl->state;
437 	switch (new_state) {
438 	case NVME_CTRL_LIVE:
439 		switch (old_state) {
440 		case NVME_CTRL_NEW:
441 		case NVME_CTRL_RESETTING:
442 		case NVME_CTRL_CONNECTING:
443 			changed = true;
444 			fallthrough;
445 		default:
446 			break;
447 		}
448 		break;
449 	case NVME_CTRL_RESETTING:
450 		switch (old_state) {
451 		case NVME_CTRL_NEW:
452 		case NVME_CTRL_LIVE:
453 			changed = true;
454 			fallthrough;
455 		default:
456 			break;
457 		}
458 		break;
459 	case NVME_CTRL_CONNECTING:
460 		switch (old_state) {
461 		case NVME_CTRL_NEW:
462 		case NVME_CTRL_RESETTING:
463 			changed = true;
464 			fallthrough;
465 		default:
466 			break;
467 		}
468 		break;
469 	case NVME_CTRL_DELETING:
470 		switch (old_state) {
471 		case NVME_CTRL_LIVE:
472 		case NVME_CTRL_RESETTING:
473 		case NVME_CTRL_CONNECTING:
474 			changed = true;
475 			fallthrough;
476 		default:
477 			break;
478 		}
479 		break;
480 	case NVME_CTRL_DELETING_NOIO:
481 		switch (old_state) {
482 		case NVME_CTRL_DELETING:
483 		case NVME_CTRL_DEAD:
484 			changed = true;
485 			fallthrough;
486 		default:
487 			break;
488 		}
489 		break;
490 	case NVME_CTRL_DEAD:
491 		switch (old_state) {
492 		case NVME_CTRL_DELETING:
493 			changed = true;
494 			fallthrough;
495 		default:
496 			break;
497 		}
498 		break;
499 	default:
500 		break;
501 	}
502 
503 	if (changed) {
504 		ctrl->state = new_state;
505 		wake_up_all(&ctrl->state_wq);
506 	}
507 
508 	spin_unlock_irqrestore(&ctrl->lock, flags);
509 	if (!changed)
510 		return false;
511 
512 	if (ctrl->state == NVME_CTRL_LIVE) {
513 		if (old_state == NVME_CTRL_CONNECTING)
514 			nvme_stop_failfast_work(ctrl);
515 		nvme_kick_requeue_lists(ctrl);
516 	} else if (ctrl->state == NVME_CTRL_CONNECTING &&
517 		old_state == NVME_CTRL_RESETTING) {
518 		nvme_start_failfast_work(ctrl);
519 	}
520 	return changed;
521 }
522 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
523 
524 /*
525  * Returns true for sink states that can't ever transition back to live.
526  */
527 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
528 {
529 	switch (ctrl->state) {
530 	case NVME_CTRL_NEW:
531 	case NVME_CTRL_LIVE:
532 	case NVME_CTRL_RESETTING:
533 	case NVME_CTRL_CONNECTING:
534 		return false;
535 	case NVME_CTRL_DELETING:
536 	case NVME_CTRL_DELETING_NOIO:
537 	case NVME_CTRL_DEAD:
538 		return true;
539 	default:
540 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
541 		return true;
542 	}
543 }
544 
545 /*
546  * Waits for the controller state to be resetting, or returns false if it is
547  * not possible to ever transition to that state.
548  */
549 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
550 {
551 	wait_event(ctrl->state_wq,
552 		   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
553 		   nvme_state_terminal(ctrl));
554 	return ctrl->state == NVME_CTRL_RESETTING;
555 }
556 EXPORT_SYMBOL_GPL(nvme_wait_reset);
557 
558 static void nvme_free_ns_head(struct kref *ref)
559 {
560 	struct nvme_ns_head *head =
561 		container_of(ref, struct nvme_ns_head, ref);
562 
563 	nvme_mpath_remove_disk(head);
564 	ida_simple_remove(&head->subsys->ns_ida, head->instance);
565 	cleanup_srcu_struct(&head->srcu);
566 	nvme_put_subsystem(head->subsys);
567 	kfree(head);
568 }
569 
570 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
571 {
572 	return kref_get_unless_zero(&head->ref);
573 }
574 
575 void nvme_put_ns_head(struct nvme_ns_head *head)
576 {
577 	kref_put(&head->ref, nvme_free_ns_head);
578 }
579 
580 static void nvme_free_ns(struct kref *kref)
581 {
582 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
583 
584 	put_disk(ns->disk);
585 	nvme_put_ns_head(ns->head);
586 	nvme_put_ctrl(ns->ctrl);
587 	kfree(ns);
588 }
589 
590 static inline bool nvme_get_ns(struct nvme_ns *ns)
591 {
592 	return kref_get_unless_zero(&ns->kref);
593 }
594 
595 void nvme_put_ns(struct nvme_ns *ns)
596 {
597 	kref_put(&ns->kref, nvme_free_ns);
598 }
599 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
600 
601 static inline void nvme_clear_nvme_request(struct request *req)
602 {
603 	nvme_req(req)->status = 0;
604 	nvme_req(req)->retries = 0;
605 	nvme_req(req)->flags = 0;
606 	req->rq_flags |= RQF_DONTPREP;
607 }
608 
609 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
610 {
611 	return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
612 }
613 
614 static inline void nvme_init_request(struct request *req,
615 		struct nvme_command *cmd)
616 {
617 	if (req->q->queuedata)
618 		req->timeout = NVME_IO_TIMEOUT;
619 	else /* no queuedata implies admin queue */
620 		req->timeout = NVME_ADMIN_TIMEOUT;
621 
622 	/* passthru commands should let the driver set the SGL flags */
623 	cmd->common.flags &= ~NVME_CMD_SGL_ALL;
624 
625 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
626 	if (req->mq_hctx->type == HCTX_TYPE_POLL)
627 		req->cmd_flags |= REQ_POLLED;
628 	nvme_clear_nvme_request(req);
629 	memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
630 }
631 
632 struct request *nvme_alloc_request(struct request_queue *q,
633 		struct nvme_command *cmd, blk_mq_req_flags_t flags)
634 {
635 	struct request *req;
636 
637 	req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
638 	if (!IS_ERR(req))
639 		nvme_init_request(req, cmd);
640 	return req;
641 }
642 EXPORT_SYMBOL_GPL(nvme_alloc_request);
643 
644 static struct request *nvme_alloc_request_qid(struct request_queue *q,
645 		struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
646 {
647 	struct request *req;
648 
649 	req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
650 			qid ? qid - 1 : 0);
651 	if (!IS_ERR(req))
652 		nvme_init_request(req, cmd);
653 	return req;
654 }
655 
656 /*
657  * For something we're not in a state to send to the device the default action
658  * is to busy it and retry it after the controller state is recovered.  However,
659  * if the controller is deleting or if anything is marked for failfast or
660  * nvme multipath it is immediately failed.
661  *
662  * Note: commands used to initialize the controller will be marked for failfast.
663  * Note: nvme cli/ioctl commands are marked for failfast.
664  */
665 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
666 		struct request *rq)
667 {
668 	if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
669 	    ctrl->state != NVME_CTRL_DELETING &&
670 	    ctrl->state != NVME_CTRL_DEAD &&
671 	    !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
672 	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
673 		return BLK_STS_RESOURCE;
674 	return nvme_host_path_error(rq);
675 }
676 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
677 
678 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
679 		bool queue_live)
680 {
681 	struct nvme_request *req = nvme_req(rq);
682 
683 	/*
684 	 * currently we have a problem sending passthru commands
685 	 * on the admin_q if the controller is not LIVE because we can't
686 	 * make sure that they are going out after the admin connect,
687 	 * controller enable and/or other commands in the initialization
688 	 * sequence. until the controller will be LIVE, fail with
689 	 * BLK_STS_RESOURCE so that they will be rescheduled.
690 	 */
691 	if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
692 		return false;
693 
694 	if (ctrl->ops->flags & NVME_F_FABRICS) {
695 		/*
696 		 * Only allow commands on a live queue, except for the connect
697 		 * command, which is require to set the queue live in the
698 		 * appropinquate states.
699 		 */
700 		switch (ctrl->state) {
701 		case NVME_CTRL_CONNECTING:
702 			if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
703 			    req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
704 				return true;
705 			break;
706 		default:
707 			break;
708 		case NVME_CTRL_DEAD:
709 			return false;
710 		}
711 	}
712 
713 	return queue_live;
714 }
715 EXPORT_SYMBOL_GPL(__nvme_check_ready);
716 
717 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
718 {
719 	struct nvme_command c = { };
720 
721 	c.directive.opcode = nvme_admin_directive_send;
722 	c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
723 	c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
724 	c.directive.dtype = NVME_DIR_IDENTIFY;
725 	c.directive.tdtype = NVME_DIR_STREAMS;
726 	c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
727 
728 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
729 }
730 
731 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
732 {
733 	return nvme_toggle_streams(ctrl, false);
734 }
735 
736 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
737 {
738 	return nvme_toggle_streams(ctrl, true);
739 }
740 
741 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
742 				  struct streams_directive_params *s, u32 nsid)
743 {
744 	struct nvme_command c = { };
745 
746 	memset(s, 0, sizeof(*s));
747 
748 	c.directive.opcode = nvme_admin_directive_recv;
749 	c.directive.nsid = cpu_to_le32(nsid);
750 	c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
751 	c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
752 	c.directive.dtype = NVME_DIR_STREAMS;
753 
754 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
755 }
756 
757 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
758 {
759 	struct streams_directive_params s;
760 	int ret;
761 
762 	if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
763 		return 0;
764 	if (!streams)
765 		return 0;
766 
767 	ret = nvme_enable_streams(ctrl);
768 	if (ret)
769 		return ret;
770 
771 	ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
772 	if (ret)
773 		goto out_disable_stream;
774 
775 	ctrl->nssa = le16_to_cpu(s.nssa);
776 	if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
777 		dev_info(ctrl->device, "too few streams (%u) available\n",
778 					ctrl->nssa);
779 		goto out_disable_stream;
780 	}
781 
782 	ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
783 	dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
784 	return 0;
785 
786 out_disable_stream:
787 	nvme_disable_streams(ctrl);
788 	return ret;
789 }
790 
791 /*
792  * Check if 'req' has a write hint associated with it. If it does, assign
793  * a valid namespace stream to the write.
794  */
795 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
796 				     struct request *req, u16 *control,
797 				     u32 *dsmgmt)
798 {
799 	enum rw_hint streamid = req->write_hint;
800 
801 	if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
802 		streamid = 0;
803 	else {
804 		streamid--;
805 		if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
806 			return;
807 
808 		*control |= NVME_RW_DTYPE_STREAMS;
809 		*dsmgmt |= streamid << 16;
810 	}
811 
812 	if (streamid < ARRAY_SIZE(req->q->write_hints))
813 		req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
814 }
815 
816 static inline void nvme_setup_flush(struct nvme_ns *ns,
817 		struct nvme_command *cmnd)
818 {
819 	memset(cmnd, 0, sizeof(*cmnd));
820 	cmnd->common.opcode = nvme_cmd_flush;
821 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
822 }
823 
824 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
825 		struct nvme_command *cmnd)
826 {
827 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
828 	struct nvme_dsm_range *range;
829 	struct bio *bio;
830 
831 	/*
832 	 * Some devices do not consider the DSM 'Number of Ranges' field when
833 	 * determining how much data to DMA. Always allocate memory for maximum
834 	 * number of segments to prevent device reading beyond end of buffer.
835 	 */
836 	static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
837 
838 	range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
839 	if (!range) {
840 		/*
841 		 * If we fail allocation our range, fallback to the controller
842 		 * discard page. If that's also busy, it's safe to return
843 		 * busy, as we know we can make progress once that's freed.
844 		 */
845 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
846 			return BLK_STS_RESOURCE;
847 
848 		range = page_address(ns->ctrl->discard_page);
849 	}
850 
851 	__rq_for_each_bio(bio, req) {
852 		u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
853 		u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
854 
855 		if (n < segments) {
856 			range[n].cattr = cpu_to_le32(0);
857 			range[n].nlb = cpu_to_le32(nlb);
858 			range[n].slba = cpu_to_le64(slba);
859 		}
860 		n++;
861 	}
862 
863 	if (WARN_ON_ONCE(n != segments)) {
864 		if (virt_to_page(range) == ns->ctrl->discard_page)
865 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
866 		else
867 			kfree(range);
868 		return BLK_STS_IOERR;
869 	}
870 
871 	memset(cmnd, 0, sizeof(*cmnd));
872 	cmnd->dsm.opcode = nvme_cmd_dsm;
873 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
874 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
875 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
876 
877 	req->special_vec.bv_page = virt_to_page(range);
878 	req->special_vec.bv_offset = offset_in_page(range);
879 	req->special_vec.bv_len = alloc_size;
880 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
881 
882 	return BLK_STS_OK;
883 }
884 
885 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
886 		struct request *req, struct nvme_command *cmnd)
887 {
888 	memset(cmnd, 0, sizeof(*cmnd));
889 
890 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
891 		return nvme_setup_discard(ns, req, cmnd);
892 
893 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
894 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
895 	cmnd->write_zeroes.slba =
896 		cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
897 	cmnd->write_zeroes.length =
898 		cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
899 
900 	if (nvme_ns_has_pi(ns)) {
901 		cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
902 
903 		switch (ns->pi_type) {
904 		case NVME_NS_DPS_PI_TYPE1:
905 		case NVME_NS_DPS_PI_TYPE2:
906 			cmnd->write_zeroes.reftag =
907 				cpu_to_le32(t10_pi_ref_tag(req));
908 			break;
909 		}
910 	}
911 
912 	return BLK_STS_OK;
913 }
914 
915 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
916 		struct request *req, struct nvme_command *cmnd,
917 		enum nvme_opcode op)
918 {
919 	struct nvme_ctrl *ctrl = ns->ctrl;
920 	u16 control = 0;
921 	u32 dsmgmt = 0;
922 
923 	if (req->cmd_flags & REQ_FUA)
924 		control |= NVME_RW_FUA;
925 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
926 		control |= NVME_RW_LR;
927 
928 	if (req->cmd_flags & REQ_RAHEAD)
929 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
930 
931 	cmnd->rw.opcode = op;
932 	cmnd->rw.flags = 0;
933 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
934 	cmnd->rw.rsvd2 = 0;
935 	cmnd->rw.metadata = 0;
936 	cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
937 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
938 	cmnd->rw.reftag = 0;
939 	cmnd->rw.apptag = 0;
940 	cmnd->rw.appmask = 0;
941 
942 	if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
943 		nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
944 
945 	if (ns->ms) {
946 		/*
947 		 * If formated with metadata, the block layer always provides a
948 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
949 		 * we enable the PRACT bit for protection information or set the
950 		 * namespace capacity to zero to prevent any I/O.
951 		 */
952 		if (!blk_integrity_rq(req)) {
953 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
954 				return BLK_STS_NOTSUPP;
955 			control |= NVME_RW_PRINFO_PRACT;
956 		}
957 
958 		switch (ns->pi_type) {
959 		case NVME_NS_DPS_PI_TYPE3:
960 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
961 			break;
962 		case NVME_NS_DPS_PI_TYPE1:
963 		case NVME_NS_DPS_PI_TYPE2:
964 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
965 					NVME_RW_PRINFO_PRCHK_REF;
966 			if (op == nvme_cmd_zone_append)
967 				control |= NVME_RW_APPEND_PIREMAP;
968 			cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
969 			break;
970 		}
971 	}
972 
973 	cmnd->rw.control = cpu_to_le16(control);
974 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
975 	return 0;
976 }
977 
978 void nvme_cleanup_cmd(struct request *req)
979 {
980 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
981 		struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
982 
983 		if (req->special_vec.bv_page == ctrl->discard_page)
984 			clear_bit_unlock(0, &ctrl->discard_page_busy);
985 		else
986 			kfree(bvec_virt(&req->special_vec));
987 	}
988 }
989 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
990 
991 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
992 {
993 	struct nvme_command *cmd = nvme_req(req)->cmd;
994 	struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
995 	blk_status_t ret = BLK_STS_OK;
996 
997 	if (!(req->rq_flags & RQF_DONTPREP))
998 		nvme_clear_nvme_request(req);
999 
1000 	switch (req_op(req)) {
1001 	case REQ_OP_DRV_IN:
1002 	case REQ_OP_DRV_OUT:
1003 		/* these are setup prior to execution in nvme_init_request() */
1004 		break;
1005 	case REQ_OP_FLUSH:
1006 		nvme_setup_flush(ns, cmd);
1007 		break;
1008 	case REQ_OP_ZONE_RESET_ALL:
1009 	case REQ_OP_ZONE_RESET:
1010 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
1011 		break;
1012 	case REQ_OP_ZONE_OPEN:
1013 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
1014 		break;
1015 	case REQ_OP_ZONE_CLOSE:
1016 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
1017 		break;
1018 	case REQ_OP_ZONE_FINISH:
1019 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
1020 		break;
1021 	case REQ_OP_WRITE_ZEROES:
1022 		ret = nvme_setup_write_zeroes(ns, req, cmd);
1023 		break;
1024 	case REQ_OP_DISCARD:
1025 		ret = nvme_setup_discard(ns, req, cmd);
1026 		break;
1027 	case REQ_OP_READ:
1028 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1029 		break;
1030 	case REQ_OP_WRITE:
1031 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1032 		break;
1033 	case REQ_OP_ZONE_APPEND:
1034 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1035 		break;
1036 	default:
1037 		WARN_ON_ONCE(1);
1038 		return BLK_STS_IOERR;
1039 	}
1040 
1041 	if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
1042 		nvme_req(req)->genctr++;
1043 	cmd->common.command_id = nvme_cid(req);
1044 	trace_nvme_setup_cmd(req, cmd);
1045 	return ret;
1046 }
1047 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1048 
1049 /*
1050  * Return values:
1051  * 0:  success
1052  * >0: nvme controller's cqe status response
1053  * <0: kernel error in lieu of controller response
1054  */
1055 static int nvme_execute_rq(struct gendisk *disk, struct request *rq,
1056 		bool at_head)
1057 {
1058 	blk_status_t status;
1059 
1060 	status = blk_execute_rq(disk, rq, at_head);
1061 	if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1062 		return -EINTR;
1063 	if (nvme_req(rq)->status)
1064 		return nvme_req(rq)->status;
1065 	return blk_status_to_errno(status);
1066 }
1067 
1068 /*
1069  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1070  * if the result is positive, it's an NVM Express status code
1071  */
1072 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1073 		union nvme_result *result, void *buffer, unsigned bufflen,
1074 		unsigned timeout, int qid, int at_head,
1075 		blk_mq_req_flags_t flags)
1076 {
1077 	struct request *req;
1078 	int ret;
1079 
1080 	if (qid == NVME_QID_ANY)
1081 		req = nvme_alloc_request(q, cmd, flags);
1082 	else
1083 		req = nvme_alloc_request_qid(q, cmd, flags, qid);
1084 	if (IS_ERR(req))
1085 		return PTR_ERR(req);
1086 
1087 	if (timeout)
1088 		req->timeout = timeout;
1089 
1090 	if (buffer && bufflen) {
1091 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1092 		if (ret)
1093 			goto out;
1094 	}
1095 
1096 	ret = nvme_execute_rq(NULL, req, at_head);
1097 	if (result && ret >= 0)
1098 		*result = nvme_req(req)->result;
1099  out:
1100 	blk_mq_free_request(req);
1101 	return ret;
1102 }
1103 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1104 
1105 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1106 		void *buffer, unsigned bufflen)
1107 {
1108 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1109 			NVME_QID_ANY, 0, 0);
1110 }
1111 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1112 
1113 static u32 nvme_known_admin_effects(u8 opcode)
1114 {
1115 	switch (opcode) {
1116 	case nvme_admin_format_nvm:
1117 		return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1118 			NVME_CMD_EFFECTS_CSE_MASK;
1119 	case nvme_admin_sanitize_nvm:
1120 		return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1121 	default:
1122 		break;
1123 	}
1124 	return 0;
1125 }
1126 
1127 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1128 {
1129 	u32 effects = 0;
1130 
1131 	if (ns) {
1132 		if (ns->head->effects)
1133 			effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1134 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1135 			dev_warn_once(ctrl->device,
1136 				"IO command:%02x has unhandled effects:%08x\n",
1137 				opcode, effects);
1138 		return 0;
1139 	}
1140 
1141 	if (ctrl->effects)
1142 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1143 	effects |= nvme_known_admin_effects(opcode);
1144 
1145 	return effects;
1146 }
1147 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1148 
1149 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1150 			       u8 opcode)
1151 {
1152 	u32 effects = nvme_command_effects(ctrl, ns, opcode);
1153 
1154 	/*
1155 	 * For simplicity, IO to all namespaces is quiesced even if the command
1156 	 * effects say only one namespace is affected.
1157 	 */
1158 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1159 		mutex_lock(&ctrl->scan_lock);
1160 		mutex_lock(&ctrl->subsys->lock);
1161 		nvme_mpath_start_freeze(ctrl->subsys);
1162 		nvme_mpath_wait_freeze(ctrl->subsys);
1163 		nvme_start_freeze(ctrl);
1164 		nvme_wait_freeze(ctrl);
1165 	}
1166 	return effects;
1167 }
1168 
1169 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1170 			      struct nvme_command *cmd, int status)
1171 {
1172 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1173 		nvme_unfreeze(ctrl);
1174 		nvme_mpath_unfreeze(ctrl->subsys);
1175 		mutex_unlock(&ctrl->subsys->lock);
1176 		nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1177 		mutex_unlock(&ctrl->scan_lock);
1178 	}
1179 	if (effects & NVME_CMD_EFFECTS_CCC)
1180 		nvme_init_ctrl_finish(ctrl);
1181 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1182 		nvme_queue_scan(ctrl);
1183 		flush_work(&ctrl->scan_work);
1184 	}
1185 
1186 	switch (cmd->common.opcode) {
1187 	case nvme_admin_set_features:
1188 		switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1189 		case NVME_FEAT_KATO:
1190 			/*
1191 			 * Keep alive commands interval on the host should be
1192 			 * updated when KATO is modified by Set Features
1193 			 * commands.
1194 			 */
1195 			if (!status)
1196 				nvme_update_keep_alive(ctrl, cmd);
1197 			break;
1198 		default:
1199 			break;
1200 		}
1201 		break;
1202 	default:
1203 		break;
1204 	}
1205 }
1206 
1207 int nvme_execute_passthru_rq(struct request *rq)
1208 {
1209 	struct nvme_command *cmd = nvme_req(rq)->cmd;
1210 	struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1211 	struct nvme_ns *ns = rq->q->queuedata;
1212 	struct gendisk *disk = ns ? ns->disk : NULL;
1213 	u32 effects;
1214 	int  ret;
1215 
1216 	effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1217 	ret = nvme_execute_rq(disk, rq, false);
1218 	if (effects) /* nothing to be done for zero cmd effects */
1219 		nvme_passthru_end(ctrl, effects, cmd, ret);
1220 
1221 	return ret;
1222 }
1223 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1224 
1225 /*
1226  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1227  *
1228  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1229  *   accounting for transport roundtrip times [..].
1230  */
1231 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1232 {
1233 	queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1234 }
1235 
1236 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1237 {
1238 	struct nvme_ctrl *ctrl = rq->end_io_data;
1239 	unsigned long flags;
1240 	bool startka = false;
1241 
1242 	blk_mq_free_request(rq);
1243 
1244 	if (status) {
1245 		dev_err(ctrl->device,
1246 			"failed nvme_keep_alive_end_io error=%d\n",
1247 				status);
1248 		return;
1249 	}
1250 
1251 	ctrl->comp_seen = false;
1252 	spin_lock_irqsave(&ctrl->lock, flags);
1253 	if (ctrl->state == NVME_CTRL_LIVE ||
1254 	    ctrl->state == NVME_CTRL_CONNECTING)
1255 		startka = true;
1256 	spin_unlock_irqrestore(&ctrl->lock, flags);
1257 	if (startka)
1258 		nvme_queue_keep_alive_work(ctrl);
1259 }
1260 
1261 static void nvme_keep_alive_work(struct work_struct *work)
1262 {
1263 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1264 			struct nvme_ctrl, ka_work);
1265 	bool comp_seen = ctrl->comp_seen;
1266 	struct request *rq;
1267 
1268 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1269 		dev_dbg(ctrl->device,
1270 			"reschedule traffic based keep-alive timer\n");
1271 		ctrl->comp_seen = false;
1272 		nvme_queue_keep_alive_work(ctrl);
1273 		return;
1274 	}
1275 
1276 	rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1277 				BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1278 	if (IS_ERR(rq)) {
1279 		/* allocation failure, reset the controller */
1280 		dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1281 		nvme_reset_ctrl(ctrl);
1282 		return;
1283 	}
1284 
1285 	rq->timeout = ctrl->kato * HZ;
1286 	rq->end_io_data = ctrl;
1287 	blk_execute_rq_nowait(NULL, rq, 0, nvme_keep_alive_end_io);
1288 }
1289 
1290 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1291 {
1292 	if (unlikely(ctrl->kato == 0))
1293 		return;
1294 
1295 	nvme_queue_keep_alive_work(ctrl);
1296 }
1297 
1298 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1299 {
1300 	if (unlikely(ctrl->kato == 0))
1301 		return;
1302 
1303 	cancel_delayed_work_sync(&ctrl->ka_work);
1304 }
1305 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1306 
1307 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1308 				   struct nvme_command *cmd)
1309 {
1310 	unsigned int new_kato =
1311 		DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1312 
1313 	dev_info(ctrl->device,
1314 		 "keep alive interval updated from %u ms to %u ms\n",
1315 		 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1316 
1317 	nvme_stop_keep_alive(ctrl);
1318 	ctrl->kato = new_kato;
1319 	nvme_start_keep_alive(ctrl);
1320 }
1321 
1322 /*
1323  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1324  * flag, thus sending any new CNS opcodes has a big chance of not working.
1325  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1326  * (but not for any later version).
1327  */
1328 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1329 {
1330 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1331 		return ctrl->vs < NVME_VS(1, 2, 0);
1332 	return ctrl->vs < NVME_VS(1, 1, 0);
1333 }
1334 
1335 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1336 {
1337 	struct nvme_command c = { };
1338 	int error;
1339 
1340 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1341 	c.identify.opcode = nvme_admin_identify;
1342 	c.identify.cns = NVME_ID_CNS_CTRL;
1343 
1344 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1345 	if (!*id)
1346 		return -ENOMEM;
1347 
1348 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1349 			sizeof(struct nvme_id_ctrl));
1350 	if (error)
1351 		kfree(*id);
1352 	return error;
1353 }
1354 
1355 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1356 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1357 {
1358 	const char *warn_str = "ctrl returned bogus length:";
1359 	void *data = cur;
1360 
1361 	switch (cur->nidt) {
1362 	case NVME_NIDT_EUI64:
1363 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1364 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1365 				 warn_str, cur->nidl);
1366 			return -1;
1367 		}
1368 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1369 		return NVME_NIDT_EUI64_LEN;
1370 	case NVME_NIDT_NGUID:
1371 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1372 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1373 				 warn_str, cur->nidl);
1374 			return -1;
1375 		}
1376 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1377 		return NVME_NIDT_NGUID_LEN;
1378 	case NVME_NIDT_UUID:
1379 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1380 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1381 				 warn_str, cur->nidl);
1382 			return -1;
1383 		}
1384 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1385 		return NVME_NIDT_UUID_LEN;
1386 	case NVME_NIDT_CSI:
1387 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1388 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1389 				 warn_str, cur->nidl);
1390 			return -1;
1391 		}
1392 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1393 		*csi_seen = true;
1394 		return NVME_NIDT_CSI_LEN;
1395 	default:
1396 		/* Skip unknown types */
1397 		return cur->nidl;
1398 	}
1399 }
1400 
1401 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1402 		struct nvme_ns_ids *ids)
1403 {
1404 	struct nvme_command c = { };
1405 	bool csi_seen = false;
1406 	int status, pos, len;
1407 	void *data;
1408 
1409 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1410 		return 0;
1411 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1412 		return 0;
1413 
1414 	c.identify.opcode = nvme_admin_identify;
1415 	c.identify.nsid = cpu_to_le32(nsid);
1416 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1417 
1418 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1419 	if (!data)
1420 		return -ENOMEM;
1421 
1422 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1423 				      NVME_IDENTIFY_DATA_SIZE);
1424 	if (status) {
1425 		dev_warn(ctrl->device,
1426 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1427 			nsid, status);
1428 		goto free_data;
1429 	}
1430 
1431 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1432 		struct nvme_ns_id_desc *cur = data + pos;
1433 
1434 		if (cur->nidl == 0)
1435 			break;
1436 
1437 		len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1438 		if (len < 0)
1439 			break;
1440 
1441 		len += sizeof(*cur);
1442 	}
1443 
1444 	if (nvme_multi_css(ctrl) && !csi_seen) {
1445 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1446 			 nsid);
1447 		status = -EINVAL;
1448 	}
1449 
1450 free_data:
1451 	kfree(data);
1452 	return status;
1453 }
1454 
1455 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1456 			struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1457 {
1458 	struct nvme_command c = { };
1459 	int error;
1460 
1461 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1462 	c.identify.opcode = nvme_admin_identify;
1463 	c.identify.nsid = cpu_to_le32(nsid);
1464 	c.identify.cns = NVME_ID_CNS_NS;
1465 
1466 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1467 	if (!*id)
1468 		return -ENOMEM;
1469 
1470 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1471 	if (error) {
1472 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1473 		goto out_free_id;
1474 	}
1475 
1476 	error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1477 	if ((*id)->ncap == 0) /* namespace not allocated or attached */
1478 		goto out_free_id;
1479 
1480 	if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1481 	    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1482 		memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1483 	if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1484 	    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1485 		memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1486 
1487 	return 0;
1488 
1489 out_free_id:
1490 	kfree(*id);
1491 	return error;
1492 }
1493 
1494 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1495 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1496 {
1497 	union nvme_result res = { 0 };
1498 	struct nvme_command c = { };
1499 	int ret;
1500 
1501 	c.features.opcode = op;
1502 	c.features.fid = cpu_to_le32(fid);
1503 	c.features.dword11 = cpu_to_le32(dword11);
1504 
1505 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1506 			buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1507 	if (ret >= 0 && result)
1508 		*result = le32_to_cpu(res.u32);
1509 	return ret;
1510 }
1511 
1512 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1513 		      unsigned int dword11, void *buffer, size_t buflen,
1514 		      u32 *result)
1515 {
1516 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1517 			     buflen, result);
1518 }
1519 EXPORT_SYMBOL_GPL(nvme_set_features);
1520 
1521 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1522 		      unsigned int dword11, void *buffer, size_t buflen,
1523 		      u32 *result)
1524 {
1525 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1526 			     buflen, result);
1527 }
1528 EXPORT_SYMBOL_GPL(nvme_get_features);
1529 
1530 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1531 {
1532 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1533 	u32 result;
1534 	int status, nr_io_queues;
1535 
1536 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1537 			&result);
1538 	if (status < 0)
1539 		return status;
1540 
1541 	/*
1542 	 * Degraded controllers might return an error when setting the queue
1543 	 * count.  We still want to be able to bring them online and offer
1544 	 * access to the admin queue, as that might be only way to fix them up.
1545 	 */
1546 	if (status > 0) {
1547 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1548 		*count = 0;
1549 	} else {
1550 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1551 		*count = min(*count, nr_io_queues);
1552 	}
1553 
1554 	return 0;
1555 }
1556 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1557 
1558 #define NVME_AEN_SUPPORTED \
1559 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1560 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1561 
1562 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1563 {
1564 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1565 	int status;
1566 
1567 	if (!supported_aens)
1568 		return;
1569 
1570 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1571 			NULL, 0, &result);
1572 	if (status)
1573 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1574 			 supported_aens);
1575 
1576 	queue_work(nvme_wq, &ctrl->async_event_work);
1577 }
1578 
1579 static int nvme_ns_open(struct nvme_ns *ns)
1580 {
1581 
1582 	/* should never be called due to GENHD_FL_HIDDEN */
1583 	if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1584 		goto fail;
1585 	if (!nvme_get_ns(ns))
1586 		goto fail;
1587 	if (!try_module_get(ns->ctrl->ops->module))
1588 		goto fail_put_ns;
1589 
1590 	return 0;
1591 
1592 fail_put_ns:
1593 	nvme_put_ns(ns);
1594 fail:
1595 	return -ENXIO;
1596 }
1597 
1598 static void nvme_ns_release(struct nvme_ns *ns)
1599 {
1600 
1601 	module_put(ns->ctrl->ops->module);
1602 	nvme_put_ns(ns);
1603 }
1604 
1605 static int nvme_open(struct block_device *bdev, fmode_t mode)
1606 {
1607 	return nvme_ns_open(bdev->bd_disk->private_data);
1608 }
1609 
1610 static void nvme_release(struct gendisk *disk, fmode_t mode)
1611 {
1612 	nvme_ns_release(disk->private_data);
1613 }
1614 
1615 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1616 {
1617 	/* some standard values */
1618 	geo->heads = 1 << 6;
1619 	geo->sectors = 1 << 5;
1620 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1621 	return 0;
1622 }
1623 
1624 #ifdef CONFIG_BLK_DEV_INTEGRITY
1625 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1626 				u32 max_integrity_segments)
1627 {
1628 	struct blk_integrity integrity = { };
1629 
1630 	switch (pi_type) {
1631 	case NVME_NS_DPS_PI_TYPE3:
1632 		integrity.profile = &t10_pi_type3_crc;
1633 		integrity.tag_size = sizeof(u16) + sizeof(u32);
1634 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1635 		break;
1636 	case NVME_NS_DPS_PI_TYPE1:
1637 	case NVME_NS_DPS_PI_TYPE2:
1638 		integrity.profile = &t10_pi_type1_crc;
1639 		integrity.tag_size = sizeof(u16);
1640 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1641 		break;
1642 	default:
1643 		integrity.profile = NULL;
1644 		break;
1645 	}
1646 	integrity.tuple_size = ms;
1647 	blk_integrity_register(disk, &integrity);
1648 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1649 }
1650 #else
1651 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1652 				u32 max_integrity_segments)
1653 {
1654 }
1655 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1656 
1657 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1658 {
1659 	struct nvme_ctrl *ctrl = ns->ctrl;
1660 	struct request_queue *queue = disk->queue;
1661 	u32 size = queue_logical_block_size(queue);
1662 
1663 	if (ctrl->max_discard_sectors == 0) {
1664 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1665 		return;
1666 	}
1667 
1668 	if (ctrl->nr_streams && ns->sws && ns->sgs)
1669 		size *= ns->sws * ns->sgs;
1670 
1671 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1672 			NVME_DSM_MAX_RANGES);
1673 
1674 	queue->limits.discard_alignment = 0;
1675 	queue->limits.discard_granularity = size;
1676 
1677 	/* If discard is already enabled, don't reset queue limits */
1678 	if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1679 		return;
1680 
1681 	blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1682 	blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1683 
1684 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1685 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1686 }
1687 
1688 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1689 {
1690 	return !uuid_is_null(&ids->uuid) ||
1691 		memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1692 		memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1693 }
1694 
1695 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1696 {
1697 	return uuid_equal(&a->uuid, &b->uuid) &&
1698 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1699 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1700 		a->csi == b->csi;
1701 }
1702 
1703 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1704 				 u32 *phys_bs, u32 *io_opt)
1705 {
1706 	struct streams_directive_params s;
1707 	int ret;
1708 
1709 	if (!ctrl->nr_streams)
1710 		return 0;
1711 
1712 	ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1713 	if (ret)
1714 		return ret;
1715 
1716 	ns->sws = le32_to_cpu(s.sws);
1717 	ns->sgs = le16_to_cpu(s.sgs);
1718 
1719 	if (ns->sws) {
1720 		*phys_bs = ns->sws * (1 << ns->lba_shift);
1721 		if (ns->sgs)
1722 			*io_opt = *phys_bs * ns->sgs;
1723 	}
1724 
1725 	return 0;
1726 }
1727 
1728 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1729 {
1730 	struct nvme_ctrl *ctrl = ns->ctrl;
1731 
1732 	/*
1733 	 * The PI implementation requires the metadata size to be equal to the
1734 	 * t10 pi tuple size.
1735 	 */
1736 	ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1737 	if (ns->ms == sizeof(struct t10_pi_tuple))
1738 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1739 	else
1740 		ns->pi_type = 0;
1741 
1742 	ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1743 	if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1744 		return 0;
1745 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1746 		/*
1747 		 * The NVMe over Fabrics specification only supports metadata as
1748 		 * part of the extended data LBA.  We rely on HCA/HBA support to
1749 		 * remap the separate metadata buffer from the block layer.
1750 		 */
1751 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1752 			return -EINVAL;
1753 
1754 		ns->features |= NVME_NS_EXT_LBAS;
1755 
1756 		/*
1757 		 * The current fabrics transport drivers support namespace
1758 		 * metadata formats only if nvme_ns_has_pi() returns true.
1759 		 * Suppress support for all other formats so the namespace will
1760 		 * have a 0 capacity and not be usable through the block stack.
1761 		 *
1762 		 * Note, this check will need to be modified if any drivers
1763 		 * gain the ability to use other metadata formats.
1764 		 */
1765 		if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1766 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1767 	} else {
1768 		/*
1769 		 * For PCIe controllers, we can't easily remap the separate
1770 		 * metadata buffer from the block layer and thus require a
1771 		 * separate metadata buffer for block layer metadata/PI support.
1772 		 * We allow extended LBAs for the passthrough interface, though.
1773 		 */
1774 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
1775 			ns->features |= NVME_NS_EXT_LBAS;
1776 		else
1777 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1778 	}
1779 
1780 	return 0;
1781 }
1782 
1783 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1784 		struct request_queue *q)
1785 {
1786 	bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1787 
1788 	if (ctrl->max_hw_sectors) {
1789 		u32 max_segments =
1790 			(ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1791 
1792 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
1793 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1794 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1795 	}
1796 	blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1797 	blk_queue_dma_alignment(q, 7);
1798 	blk_queue_write_cache(q, vwc, vwc);
1799 }
1800 
1801 static void nvme_update_disk_info(struct gendisk *disk,
1802 		struct nvme_ns *ns, struct nvme_id_ns *id)
1803 {
1804 	sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1805 	unsigned short bs = 1 << ns->lba_shift;
1806 	u32 atomic_bs, phys_bs, io_opt = 0;
1807 
1808 	/*
1809 	 * The block layer can't support LBA sizes larger than the page size
1810 	 * yet, so catch this early and don't allow block I/O.
1811 	 */
1812 	if (ns->lba_shift > PAGE_SHIFT) {
1813 		capacity = 0;
1814 		bs = (1 << 9);
1815 	}
1816 
1817 	blk_integrity_unregister(disk);
1818 
1819 	atomic_bs = phys_bs = bs;
1820 	nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1821 	if (id->nabo == 0) {
1822 		/*
1823 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1824 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1825 		 * 0 then AWUPF must be used instead.
1826 		 */
1827 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1828 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1829 		else
1830 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1831 	}
1832 
1833 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1834 		/* NPWG = Namespace Preferred Write Granularity */
1835 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1836 		/* NOWS = Namespace Optimal Write Size */
1837 		io_opt = bs * (1 + le16_to_cpu(id->nows));
1838 	}
1839 
1840 	blk_queue_logical_block_size(disk->queue, bs);
1841 	/*
1842 	 * Linux filesystems assume writing a single physical block is
1843 	 * an atomic operation. Hence limit the physical block size to the
1844 	 * value of the Atomic Write Unit Power Fail parameter.
1845 	 */
1846 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1847 	blk_queue_io_min(disk->queue, phys_bs);
1848 	blk_queue_io_opt(disk->queue, io_opt);
1849 
1850 	/*
1851 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
1852 	 * metadata masquerading as Type 0 if supported, otherwise reject block
1853 	 * I/O to namespaces with metadata except when the namespace supports
1854 	 * PI, as it can strip/insert in that case.
1855 	 */
1856 	if (ns->ms) {
1857 		if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1858 		    (ns->features & NVME_NS_METADATA_SUPPORTED))
1859 			nvme_init_integrity(disk, ns->ms, ns->pi_type,
1860 					    ns->ctrl->max_integrity_segments);
1861 		else if (!nvme_ns_has_pi(ns))
1862 			capacity = 0;
1863 	}
1864 
1865 	set_capacity_and_notify(disk, capacity);
1866 
1867 	nvme_config_discard(disk, ns);
1868 	blk_queue_max_write_zeroes_sectors(disk->queue,
1869 					   ns->ctrl->max_zeroes_sectors);
1870 
1871 	set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1872 		test_bit(NVME_NS_FORCE_RO, &ns->flags));
1873 }
1874 
1875 static inline bool nvme_first_scan(struct gendisk *disk)
1876 {
1877 	/* nvme_alloc_ns() scans the disk prior to adding it */
1878 	return !disk_live(disk);
1879 }
1880 
1881 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1882 {
1883 	struct nvme_ctrl *ctrl = ns->ctrl;
1884 	u32 iob;
1885 
1886 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1887 	    is_power_of_2(ctrl->max_hw_sectors))
1888 		iob = ctrl->max_hw_sectors;
1889 	else
1890 		iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1891 
1892 	if (!iob)
1893 		return;
1894 
1895 	if (!is_power_of_2(iob)) {
1896 		if (nvme_first_scan(ns->disk))
1897 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1898 				ns->disk->disk_name, iob);
1899 		return;
1900 	}
1901 
1902 	if (blk_queue_is_zoned(ns->disk->queue)) {
1903 		if (nvme_first_scan(ns->disk))
1904 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
1905 				ns->disk->disk_name);
1906 		return;
1907 	}
1908 
1909 	blk_queue_chunk_sectors(ns->queue, iob);
1910 }
1911 
1912 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1913 {
1914 	unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1915 	int ret;
1916 
1917 	blk_mq_freeze_queue(ns->disk->queue);
1918 	ns->lba_shift = id->lbaf[lbaf].ds;
1919 	nvme_set_queue_limits(ns->ctrl, ns->queue);
1920 
1921 	ret = nvme_configure_metadata(ns, id);
1922 	if (ret)
1923 		goto out_unfreeze;
1924 	nvme_set_chunk_sectors(ns, id);
1925 	nvme_update_disk_info(ns->disk, ns, id);
1926 
1927 	if (ns->head->ids.csi == NVME_CSI_ZNS) {
1928 		ret = nvme_update_zone_info(ns, lbaf);
1929 		if (ret)
1930 			goto out_unfreeze;
1931 	}
1932 
1933 	set_bit(NVME_NS_READY, &ns->flags);
1934 	blk_mq_unfreeze_queue(ns->disk->queue);
1935 
1936 	if (blk_queue_is_zoned(ns->queue)) {
1937 		ret = nvme_revalidate_zones(ns);
1938 		if (ret && !nvme_first_scan(ns->disk))
1939 			goto out;
1940 	}
1941 
1942 	if (nvme_ns_head_multipath(ns->head)) {
1943 		blk_mq_freeze_queue(ns->head->disk->queue);
1944 		nvme_update_disk_info(ns->head->disk, ns, id);
1945 		nvme_mpath_revalidate_paths(ns);
1946 		blk_stack_limits(&ns->head->disk->queue->limits,
1947 				 &ns->queue->limits, 0);
1948 		disk_update_readahead(ns->head->disk);
1949 		blk_mq_unfreeze_queue(ns->head->disk->queue);
1950 	}
1951 	return 0;
1952 
1953 out_unfreeze:
1954 	blk_mq_unfreeze_queue(ns->disk->queue);
1955 out:
1956 	/*
1957 	 * If probing fails due an unsupported feature, hide the block device,
1958 	 * but still allow other access.
1959 	 */
1960 	if (ret == -ENODEV) {
1961 		ns->disk->flags |= GENHD_FL_HIDDEN;
1962 		ret = 0;
1963 	}
1964 	return ret;
1965 }
1966 
1967 static char nvme_pr_type(enum pr_type type)
1968 {
1969 	switch (type) {
1970 	case PR_WRITE_EXCLUSIVE:
1971 		return 1;
1972 	case PR_EXCLUSIVE_ACCESS:
1973 		return 2;
1974 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1975 		return 3;
1976 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1977 		return 4;
1978 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1979 		return 5;
1980 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1981 		return 6;
1982 	default:
1983 		return 0;
1984 	}
1985 };
1986 
1987 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
1988 		struct nvme_command *c, u8 data[16])
1989 {
1990 	struct nvme_ns_head *head = bdev->bd_disk->private_data;
1991 	int srcu_idx = srcu_read_lock(&head->srcu);
1992 	struct nvme_ns *ns = nvme_find_path(head);
1993 	int ret = -EWOULDBLOCK;
1994 
1995 	if (ns) {
1996 		c->common.nsid = cpu_to_le32(ns->head->ns_id);
1997 		ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
1998 	}
1999 	srcu_read_unlock(&head->srcu, srcu_idx);
2000 	return ret;
2001 }
2002 
2003 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
2004 		u8 data[16])
2005 {
2006 	c->common.nsid = cpu_to_le32(ns->head->ns_id);
2007 	return nvme_submit_sync_cmd(ns->queue, c, data, 16);
2008 }
2009 
2010 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2011 				u64 key, u64 sa_key, u8 op)
2012 {
2013 	struct nvme_command c = { };
2014 	u8 data[16] = { 0, };
2015 
2016 	put_unaligned_le64(key, &data[0]);
2017 	put_unaligned_le64(sa_key, &data[8]);
2018 
2019 	c.common.opcode = op;
2020 	c.common.cdw10 = cpu_to_le32(cdw10);
2021 
2022 	if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2023 	    bdev->bd_disk->fops == &nvme_ns_head_ops)
2024 		return nvme_send_ns_head_pr_command(bdev, &c, data);
2025 	return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
2026 }
2027 
2028 static int nvme_pr_register(struct block_device *bdev, u64 old,
2029 		u64 new, unsigned flags)
2030 {
2031 	u32 cdw10;
2032 
2033 	if (flags & ~PR_FL_IGNORE_KEY)
2034 		return -EOPNOTSUPP;
2035 
2036 	cdw10 = old ? 2 : 0;
2037 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2038 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2039 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2040 }
2041 
2042 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2043 		enum pr_type type, unsigned flags)
2044 {
2045 	u32 cdw10;
2046 
2047 	if (flags & ~PR_FL_IGNORE_KEY)
2048 		return -EOPNOTSUPP;
2049 
2050 	cdw10 = nvme_pr_type(type) << 8;
2051 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2052 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2053 }
2054 
2055 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2056 		enum pr_type type, bool abort)
2057 {
2058 	u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2059 
2060 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2061 }
2062 
2063 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2064 {
2065 	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2066 
2067 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2068 }
2069 
2070 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2071 {
2072 	u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2073 
2074 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2075 }
2076 
2077 const struct pr_ops nvme_pr_ops = {
2078 	.pr_register	= nvme_pr_register,
2079 	.pr_reserve	= nvme_pr_reserve,
2080 	.pr_release	= nvme_pr_release,
2081 	.pr_preempt	= nvme_pr_preempt,
2082 	.pr_clear	= nvme_pr_clear,
2083 };
2084 
2085 #ifdef CONFIG_BLK_SED_OPAL
2086 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2087 		bool send)
2088 {
2089 	struct nvme_ctrl *ctrl = data;
2090 	struct nvme_command cmd = { };
2091 
2092 	if (send)
2093 		cmd.common.opcode = nvme_admin_security_send;
2094 	else
2095 		cmd.common.opcode = nvme_admin_security_recv;
2096 	cmd.common.nsid = 0;
2097 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2098 	cmd.common.cdw11 = cpu_to_le32(len);
2099 
2100 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2101 			NVME_QID_ANY, 1, 0);
2102 }
2103 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2104 #endif /* CONFIG_BLK_SED_OPAL */
2105 
2106 #ifdef CONFIG_BLK_DEV_ZONED
2107 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2108 		unsigned int nr_zones, report_zones_cb cb, void *data)
2109 {
2110 	return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2111 			data);
2112 }
2113 #else
2114 #define nvme_report_zones	NULL
2115 #endif /* CONFIG_BLK_DEV_ZONED */
2116 
2117 static const struct block_device_operations nvme_bdev_ops = {
2118 	.owner		= THIS_MODULE,
2119 	.ioctl		= nvme_ioctl,
2120 	.open		= nvme_open,
2121 	.release	= nvme_release,
2122 	.getgeo		= nvme_getgeo,
2123 	.report_zones	= nvme_report_zones,
2124 	.pr_ops		= &nvme_pr_ops,
2125 };
2126 
2127 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2128 {
2129 	unsigned long timeout =
2130 		((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2131 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2132 	int ret;
2133 
2134 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2135 		if (csts == ~0)
2136 			return -ENODEV;
2137 		if ((csts & NVME_CSTS_RDY) == bit)
2138 			break;
2139 
2140 		usleep_range(1000, 2000);
2141 		if (fatal_signal_pending(current))
2142 			return -EINTR;
2143 		if (time_after(jiffies, timeout)) {
2144 			dev_err(ctrl->device,
2145 				"Device not ready; aborting %s, CSTS=0x%x\n",
2146 				enabled ? "initialisation" : "reset", csts);
2147 			return -ENODEV;
2148 		}
2149 	}
2150 
2151 	return ret;
2152 }
2153 
2154 /*
2155  * If the device has been passed off to us in an enabled state, just clear
2156  * the enabled bit.  The spec says we should set the 'shutdown notification
2157  * bits', but doing so may cause the device to complete commands to the
2158  * admin queue ... and we don't know what memory that might be pointing at!
2159  */
2160 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2161 {
2162 	int ret;
2163 
2164 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2165 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2166 
2167 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2168 	if (ret)
2169 		return ret;
2170 
2171 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2172 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2173 
2174 	return nvme_wait_ready(ctrl, ctrl->cap, false);
2175 }
2176 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2177 
2178 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2179 {
2180 	unsigned dev_page_min;
2181 	int ret;
2182 
2183 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2184 	if (ret) {
2185 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2186 		return ret;
2187 	}
2188 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2189 
2190 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2191 		dev_err(ctrl->device,
2192 			"Minimum device page size %u too large for host (%u)\n",
2193 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2194 		return -ENODEV;
2195 	}
2196 
2197 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2198 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2199 	else
2200 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2201 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2202 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2203 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2204 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2205 
2206 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2207 	if (ret)
2208 		return ret;
2209 	return nvme_wait_ready(ctrl, ctrl->cap, true);
2210 }
2211 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2212 
2213 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2214 {
2215 	unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2216 	u32 csts;
2217 	int ret;
2218 
2219 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2220 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2221 
2222 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2223 	if (ret)
2224 		return ret;
2225 
2226 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2227 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2228 			break;
2229 
2230 		msleep(100);
2231 		if (fatal_signal_pending(current))
2232 			return -EINTR;
2233 		if (time_after(jiffies, timeout)) {
2234 			dev_err(ctrl->device,
2235 				"Device shutdown incomplete; abort shutdown\n");
2236 			return -ENODEV;
2237 		}
2238 	}
2239 
2240 	return ret;
2241 }
2242 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2243 
2244 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2245 {
2246 	__le64 ts;
2247 	int ret;
2248 
2249 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2250 		return 0;
2251 
2252 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2253 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2254 			NULL);
2255 	if (ret)
2256 		dev_warn_once(ctrl->device,
2257 			"could not set timestamp (%d)\n", ret);
2258 	return ret;
2259 }
2260 
2261 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2262 {
2263 	struct nvme_feat_host_behavior *host;
2264 	int ret;
2265 
2266 	/* Don't bother enabling the feature if retry delay is not reported */
2267 	if (!ctrl->crdt[0])
2268 		return 0;
2269 
2270 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2271 	if (!host)
2272 		return 0;
2273 
2274 	host->acre = NVME_ENABLE_ACRE;
2275 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2276 				host, sizeof(*host), NULL);
2277 	kfree(host);
2278 	return ret;
2279 }
2280 
2281 /*
2282  * The function checks whether the given total (exlat + enlat) latency of
2283  * a power state allows the latter to be used as an APST transition target.
2284  * It does so by comparing the latency to the primary and secondary latency
2285  * tolerances defined by module params. If there's a match, the corresponding
2286  * timeout value is returned and the matching tolerance index (1 or 2) is
2287  * reported.
2288  */
2289 static bool nvme_apst_get_transition_time(u64 total_latency,
2290 		u64 *transition_time, unsigned *last_index)
2291 {
2292 	if (total_latency <= apst_primary_latency_tol_us) {
2293 		if (*last_index == 1)
2294 			return false;
2295 		*last_index = 1;
2296 		*transition_time = apst_primary_timeout_ms;
2297 		return true;
2298 	}
2299 	if (apst_secondary_timeout_ms &&
2300 		total_latency <= apst_secondary_latency_tol_us) {
2301 		if (*last_index <= 2)
2302 			return false;
2303 		*last_index = 2;
2304 		*transition_time = apst_secondary_timeout_ms;
2305 		return true;
2306 	}
2307 	return false;
2308 }
2309 
2310 /*
2311  * APST (Autonomous Power State Transition) lets us program a table of power
2312  * state transitions that the controller will perform automatically.
2313  *
2314  * Depending on module params, one of the two supported techniques will be used:
2315  *
2316  * - If the parameters provide explicit timeouts and tolerances, they will be
2317  *   used to build a table with up to 2 non-operational states to transition to.
2318  *   The default parameter values were selected based on the values used by
2319  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2320  *   regeneration of the APST table in the event of switching between external
2321  *   and battery power, the timeouts and tolerances reflect a compromise
2322  *   between values used by Microsoft for AC and battery scenarios.
2323  * - If not, we'll configure the table with a simple heuristic: we are willing
2324  *   to spend at most 2% of the time transitioning between power states.
2325  *   Therefore, when running in any given state, we will enter the next
2326  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2327  *   microseconds, as long as that state's exit latency is under the requested
2328  *   maximum latency.
2329  *
2330  * We will not autonomously enter any non-operational state for which the total
2331  * latency exceeds ps_max_latency_us.
2332  *
2333  * Users can set ps_max_latency_us to zero to turn off APST.
2334  */
2335 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2336 {
2337 	struct nvme_feat_auto_pst *table;
2338 	unsigned apste = 0;
2339 	u64 max_lat_us = 0;
2340 	__le64 target = 0;
2341 	int max_ps = -1;
2342 	int state;
2343 	int ret;
2344 	unsigned last_lt_index = UINT_MAX;
2345 
2346 	/*
2347 	 * If APST isn't supported or if we haven't been initialized yet,
2348 	 * then don't do anything.
2349 	 */
2350 	if (!ctrl->apsta)
2351 		return 0;
2352 
2353 	if (ctrl->npss > 31) {
2354 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2355 		return 0;
2356 	}
2357 
2358 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2359 	if (!table)
2360 		return 0;
2361 
2362 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2363 		/* Turn off APST. */
2364 		dev_dbg(ctrl->device, "APST disabled\n");
2365 		goto done;
2366 	}
2367 
2368 	/*
2369 	 * Walk through all states from lowest- to highest-power.
2370 	 * According to the spec, lower-numbered states use more power.  NPSS,
2371 	 * despite the name, is the index of the lowest-power state, not the
2372 	 * number of states.
2373 	 */
2374 	for (state = (int)ctrl->npss; state >= 0; state--) {
2375 		u64 total_latency_us, exit_latency_us, transition_ms;
2376 
2377 		if (target)
2378 			table->entries[state] = target;
2379 
2380 		/*
2381 		 * Don't allow transitions to the deepest state if it's quirked
2382 		 * off.
2383 		 */
2384 		if (state == ctrl->npss &&
2385 		    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2386 			continue;
2387 
2388 		/*
2389 		 * Is this state a useful non-operational state for higher-power
2390 		 * states to autonomously transition to?
2391 		 */
2392 		if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2393 			continue;
2394 
2395 		exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2396 		if (exit_latency_us > ctrl->ps_max_latency_us)
2397 			continue;
2398 
2399 		total_latency_us = exit_latency_us +
2400 			le32_to_cpu(ctrl->psd[state].entry_lat);
2401 
2402 		/*
2403 		 * This state is good. It can be used as the APST idle target
2404 		 * for higher power states.
2405 		 */
2406 		if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2407 			if (!nvme_apst_get_transition_time(total_latency_us,
2408 					&transition_ms, &last_lt_index))
2409 				continue;
2410 		} else {
2411 			transition_ms = total_latency_us + 19;
2412 			do_div(transition_ms, 20);
2413 			if (transition_ms > (1 << 24) - 1)
2414 				transition_ms = (1 << 24) - 1;
2415 		}
2416 
2417 		target = cpu_to_le64((state << 3) | (transition_ms << 8));
2418 		if (max_ps == -1)
2419 			max_ps = state;
2420 		if (total_latency_us > max_lat_us)
2421 			max_lat_us = total_latency_us;
2422 	}
2423 
2424 	if (max_ps == -1)
2425 		dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2426 	else
2427 		dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2428 			max_ps, max_lat_us, (int)sizeof(*table), table);
2429 	apste = 1;
2430 
2431 done:
2432 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2433 				table, sizeof(*table), NULL);
2434 	if (ret)
2435 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2436 	kfree(table);
2437 	return ret;
2438 }
2439 
2440 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2441 {
2442 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2443 	u64 latency;
2444 
2445 	switch (val) {
2446 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2447 	case PM_QOS_LATENCY_ANY:
2448 		latency = U64_MAX;
2449 		break;
2450 
2451 	default:
2452 		latency = val;
2453 	}
2454 
2455 	if (ctrl->ps_max_latency_us != latency) {
2456 		ctrl->ps_max_latency_us = latency;
2457 		if (ctrl->state == NVME_CTRL_LIVE)
2458 			nvme_configure_apst(ctrl);
2459 	}
2460 }
2461 
2462 struct nvme_core_quirk_entry {
2463 	/*
2464 	 * NVMe model and firmware strings are padded with spaces.  For
2465 	 * simplicity, strings in the quirk table are padded with NULLs
2466 	 * instead.
2467 	 */
2468 	u16 vid;
2469 	const char *mn;
2470 	const char *fr;
2471 	unsigned long quirks;
2472 };
2473 
2474 static const struct nvme_core_quirk_entry core_quirks[] = {
2475 	{
2476 		/*
2477 		 * This Toshiba device seems to die using any APST states.  See:
2478 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2479 		 */
2480 		.vid = 0x1179,
2481 		.mn = "THNSF5256GPUK TOSHIBA",
2482 		.quirks = NVME_QUIRK_NO_APST,
2483 	},
2484 	{
2485 		/*
2486 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2487 		 * condition associated with actions related to suspend to idle
2488 		 * LiteON has resolved the problem in future firmware
2489 		 */
2490 		.vid = 0x14a4,
2491 		.fr = "22301111",
2492 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2493 	},
2494 	{
2495 		/*
2496 		 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2497 		 * aborts I/O during any load, but more easily reproducible
2498 		 * with discards (fstrim).
2499 		 *
2500 		 * The device is left in a state where it is also not possible
2501 		 * to use "nvme set-feature" to disable APST, but booting with
2502 		 * nvme_core.default_ps_max_latency=0 works.
2503 		 */
2504 		.vid = 0x1e0f,
2505 		.mn = "KCD6XVUL6T40",
2506 		.quirks = NVME_QUIRK_NO_APST,
2507 	}
2508 };
2509 
2510 /* match is null-terminated but idstr is space-padded. */
2511 static bool string_matches(const char *idstr, const char *match, size_t len)
2512 {
2513 	size_t matchlen;
2514 
2515 	if (!match)
2516 		return true;
2517 
2518 	matchlen = strlen(match);
2519 	WARN_ON_ONCE(matchlen > len);
2520 
2521 	if (memcmp(idstr, match, matchlen))
2522 		return false;
2523 
2524 	for (; matchlen < len; matchlen++)
2525 		if (idstr[matchlen] != ' ')
2526 			return false;
2527 
2528 	return true;
2529 }
2530 
2531 static bool quirk_matches(const struct nvme_id_ctrl *id,
2532 			  const struct nvme_core_quirk_entry *q)
2533 {
2534 	return q->vid == le16_to_cpu(id->vid) &&
2535 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2536 		string_matches(id->fr, q->fr, sizeof(id->fr));
2537 }
2538 
2539 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2540 		struct nvme_id_ctrl *id)
2541 {
2542 	size_t nqnlen;
2543 	int off;
2544 
2545 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2546 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2547 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2548 			strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2549 			return;
2550 		}
2551 
2552 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2553 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2554 	}
2555 
2556 	/* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2557 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2558 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2559 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2560 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2561 	off += sizeof(id->sn);
2562 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2563 	off += sizeof(id->mn);
2564 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2565 }
2566 
2567 static void nvme_release_subsystem(struct device *dev)
2568 {
2569 	struct nvme_subsystem *subsys =
2570 		container_of(dev, struct nvme_subsystem, dev);
2571 
2572 	if (subsys->instance >= 0)
2573 		ida_simple_remove(&nvme_instance_ida, subsys->instance);
2574 	kfree(subsys);
2575 }
2576 
2577 static void nvme_destroy_subsystem(struct kref *ref)
2578 {
2579 	struct nvme_subsystem *subsys =
2580 			container_of(ref, struct nvme_subsystem, ref);
2581 
2582 	mutex_lock(&nvme_subsystems_lock);
2583 	list_del(&subsys->entry);
2584 	mutex_unlock(&nvme_subsystems_lock);
2585 
2586 	ida_destroy(&subsys->ns_ida);
2587 	device_del(&subsys->dev);
2588 	put_device(&subsys->dev);
2589 }
2590 
2591 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2592 {
2593 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2594 }
2595 
2596 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2597 {
2598 	struct nvme_subsystem *subsys;
2599 
2600 	lockdep_assert_held(&nvme_subsystems_lock);
2601 
2602 	/*
2603 	 * Fail matches for discovery subsystems. This results
2604 	 * in each discovery controller bound to a unique subsystem.
2605 	 * This avoids issues with validating controller values
2606 	 * that can only be true when there is a single unique subsystem.
2607 	 * There may be multiple and completely independent entities
2608 	 * that provide discovery controllers.
2609 	 */
2610 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2611 		return NULL;
2612 
2613 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2614 		if (strcmp(subsys->subnqn, subsysnqn))
2615 			continue;
2616 		if (!kref_get_unless_zero(&subsys->ref))
2617 			continue;
2618 		return subsys;
2619 	}
2620 
2621 	return NULL;
2622 }
2623 
2624 #define SUBSYS_ATTR_RO(_name, _mode, _show)			\
2625 	struct device_attribute subsys_attr_##_name = \
2626 		__ATTR(_name, _mode, _show, NULL)
2627 
2628 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2629 				    struct device_attribute *attr,
2630 				    char *buf)
2631 {
2632 	struct nvme_subsystem *subsys =
2633 		container_of(dev, struct nvme_subsystem, dev);
2634 
2635 	return sysfs_emit(buf, "%s\n", subsys->subnqn);
2636 }
2637 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2638 
2639 static ssize_t nvme_subsys_show_type(struct device *dev,
2640 				    struct device_attribute *attr,
2641 				    char *buf)
2642 {
2643 	struct nvme_subsystem *subsys =
2644 		container_of(dev, struct nvme_subsystem, dev);
2645 
2646 	switch (subsys->subtype) {
2647 	case NVME_NQN_DISC:
2648 		return sysfs_emit(buf, "discovery\n");
2649 	case NVME_NQN_NVME:
2650 		return sysfs_emit(buf, "nvm\n");
2651 	default:
2652 		return sysfs_emit(buf, "reserved\n");
2653 	}
2654 }
2655 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2656 
2657 #define nvme_subsys_show_str_function(field)				\
2658 static ssize_t subsys_##field##_show(struct device *dev,		\
2659 			    struct device_attribute *attr, char *buf)	\
2660 {									\
2661 	struct nvme_subsystem *subsys =					\
2662 		container_of(dev, struct nvme_subsystem, dev);		\
2663 	return sysfs_emit(buf, "%.*s\n",				\
2664 			   (int)sizeof(subsys->field), subsys->field);	\
2665 }									\
2666 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2667 
2668 nvme_subsys_show_str_function(model);
2669 nvme_subsys_show_str_function(serial);
2670 nvme_subsys_show_str_function(firmware_rev);
2671 
2672 static struct attribute *nvme_subsys_attrs[] = {
2673 	&subsys_attr_model.attr,
2674 	&subsys_attr_serial.attr,
2675 	&subsys_attr_firmware_rev.attr,
2676 	&subsys_attr_subsysnqn.attr,
2677 	&subsys_attr_subsystype.attr,
2678 #ifdef CONFIG_NVME_MULTIPATH
2679 	&subsys_attr_iopolicy.attr,
2680 #endif
2681 	NULL,
2682 };
2683 
2684 static const struct attribute_group nvme_subsys_attrs_group = {
2685 	.attrs = nvme_subsys_attrs,
2686 };
2687 
2688 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2689 	&nvme_subsys_attrs_group,
2690 	NULL,
2691 };
2692 
2693 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2694 {
2695 	return ctrl->opts && ctrl->opts->discovery_nqn;
2696 }
2697 
2698 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2699 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2700 {
2701 	struct nvme_ctrl *tmp;
2702 
2703 	lockdep_assert_held(&nvme_subsystems_lock);
2704 
2705 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2706 		if (nvme_state_terminal(tmp))
2707 			continue;
2708 
2709 		if (tmp->cntlid == ctrl->cntlid) {
2710 			dev_err(ctrl->device,
2711 				"Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2712 				ctrl->cntlid, dev_name(tmp->device),
2713 				subsys->subnqn);
2714 			return false;
2715 		}
2716 
2717 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2718 		    nvme_discovery_ctrl(ctrl))
2719 			continue;
2720 
2721 		dev_err(ctrl->device,
2722 			"Subsystem does not support multiple controllers\n");
2723 		return false;
2724 	}
2725 
2726 	return true;
2727 }
2728 
2729 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2730 {
2731 	struct nvme_subsystem *subsys, *found;
2732 	int ret;
2733 
2734 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2735 	if (!subsys)
2736 		return -ENOMEM;
2737 
2738 	subsys->instance = -1;
2739 	mutex_init(&subsys->lock);
2740 	kref_init(&subsys->ref);
2741 	INIT_LIST_HEAD(&subsys->ctrls);
2742 	INIT_LIST_HEAD(&subsys->nsheads);
2743 	nvme_init_subnqn(subsys, ctrl, id);
2744 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2745 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2746 	memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2747 	subsys->vendor_id = le16_to_cpu(id->vid);
2748 	subsys->cmic = id->cmic;
2749 
2750 	/* Versions prior to 1.4 don't necessarily report a valid type */
2751 	if (id->cntrltype == NVME_CTRL_DISC ||
2752 	    !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2753 		subsys->subtype = NVME_NQN_DISC;
2754 	else
2755 		subsys->subtype = NVME_NQN_NVME;
2756 
2757 	if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2758 		dev_err(ctrl->device,
2759 			"Subsystem %s is not a discovery controller",
2760 			subsys->subnqn);
2761 		kfree(subsys);
2762 		return -EINVAL;
2763 	}
2764 	subsys->awupf = le16_to_cpu(id->awupf);
2765 #ifdef CONFIG_NVME_MULTIPATH
2766 	subsys->iopolicy = NVME_IOPOLICY_NUMA;
2767 #endif
2768 
2769 	subsys->dev.class = nvme_subsys_class;
2770 	subsys->dev.release = nvme_release_subsystem;
2771 	subsys->dev.groups = nvme_subsys_attrs_groups;
2772 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2773 	device_initialize(&subsys->dev);
2774 
2775 	mutex_lock(&nvme_subsystems_lock);
2776 	found = __nvme_find_get_subsystem(subsys->subnqn);
2777 	if (found) {
2778 		put_device(&subsys->dev);
2779 		subsys = found;
2780 
2781 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2782 			ret = -EINVAL;
2783 			goto out_put_subsystem;
2784 		}
2785 	} else {
2786 		ret = device_add(&subsys->dev);
2787 		if (ret) {
2788 			dev_err(ctrl->device,
2789 				"failed to register subsystem device.\n");
2790 			put_device(&subsys->dev);
2791 			goto out_unlock;
2792 		}
2793 		ida_init(&subsys->ns_ida);
2794 		list_add_tail(&subsys->entry, &nvme_subsystems);
2795 	}
2796 
2797 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2798 				dev_name(ctrl->device));
2799 	if (ret) {
2800 		dev_err(ctrl->device,
2801 			"failed to create sysfs link from subsystem.\n");
2802 		goto out_put_subsystem;
2803 	}
2804 
2805 	if (!found)
2806 		subsys->instance = ctrl->instance;
2807 	ctrl->subsys = subsys;
2808 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2809 	mutex_unlock(&nvme_subsystems_lock);
2810 	return 0;
2811 
2812 out_put_subsystem:
2813 	nvme_put_subsystem(subsys);
2814 out_unlock:
2815 	mutex_unlock(&nvme_subsystems_lock);
2816 	return ret;
2817 }
2818 
2819 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2820 		void *log, size_t size, u64 offset)
2821 {
2822 	struct nvme_command c = { };
2823 	u32 dwlen = nvme_bytes_to_numd(size);
2824 
2825 	c.get_log_page.opcode = nvme_admin_get_log_page;
2826 	c.get_log_page.nsid = cpu_to_le32(nsid);
2827 	c.get_log_page.lid = log_page;
2828 	c.get_log_page.lsp = lsp;
2829 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2830 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2831 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2832 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2833 	c.get_log_page.csi = csi;
2834 
2835 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2836 }
2837 
2838 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2839 				struct nvme_effects_log **log)
2840 {
2841 	struct nvme_effects_log	*cel = xa_load(&ctrl->cels, csi);
2842 	int ret;
2843 
2844 	if (cel)
2845 		goto out;
2846 
2847 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2848 	if (!cel)
2849 		return -ENOMEM;
2850 
2851 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2852 			cel, sizeof(*cel), 0);
2853 	if (ret) {
2854 		kfree(cel);
2855 		return ret;
2856 	}
2857 
2858 	xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2859 out:
2860 	*log = cel;
2861 	return 0;
2862 }
2863 
2864 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2865 {
2866 	u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2867 
2868 	if (check_shl_overflow(1U, units + page_shift - 9, &val))
2869 		return UINT_MAX;
2870 	return val;
2871 }
2872 
2873 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2874 {
2875 	struct nvme_command c = { };
2876 	struct nvme_id_ctrl_nvm *id;
2877 	int ret;
2878 
2879 	if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2880 		ctrl->max_discard_sectors = UINT_MAX;
2881 		ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2882 	} else {
2883 		ctrl->max_discard_sectors = 0;
2884 		ctrl->max_discard_segments = 0;
2885 	}
2886 
2887 	/*
2888 	 * Even though NVMe spec explicitly states that MDTS is not applicable
2889 	 * to the write-zeroes, we are cautious and limit the size to the
2890 	 * controllers max_hw_sectors value, which is based on the MDTS field
2891 	 * and possibly other limiting factors.
2892 	 */
2893 	if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2894 	    !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2895 		ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2896 	else
2897 		ctrl->max_zeroes_sectors = 0;
2898 
2899 	if (nvme_ctrl_limited_cns(ctrl))
2900 		return 0;
2901 
2902 	id = kzalloc(sizeof(*id), GFP_KERNEL);
2903 	if (!id)
2904 		return 0;
2905 
2906 	c.identify.opcode = nvme_admin_identify;
2907 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
2908 	c.identify.csi = NVME_CSI_NVM;
2909 
2910 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2911 	if (ret)
2912 		goto free_data;
2913 
2914 	if (id->dmrl)
2915 		ctrl->max_discard_segments = id->dmrl;
2916 	if (id->dmrsl)
2917 		ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
2918 	if (id->wzsl)
2919 		ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2920 
2921 free_data:
2922 	kfree(id);
2923 	return ret;
2924 }
2925 
2926 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2927 {
2928 	struct nvme_id_ctrl *id;
2929 	u32 max_hw_sectors;
2930 	bool prev_apst_enabled;
2931 	int ret;
2932 
2933 	ret = nvme_identify_ctrl(ctrl, &id);
2934 	if (ret) {
2935 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2936 		return -EIO;
2937 	}
2938 
2939 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2940 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2941 		if (ret < 0)
2942 			goto out_free;
2943 	}
2944 
2945 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
2946 		ctrl->cntlid = le16_to_cpu(id->cntlid);
2947 
2948 	if (!ctrl->identified) {
2949 		unsigned int i;
2950 
2951 		ret = nvme_init_subsystem(ctrl, id);
2952 		if (ret)
2953 			goto out_free;
2954 
2955 		/*
2956 		 * Check for quirks.  Quirk can depend on firmware version,
2957 		 * so, in principle, the set of quirks present can change
2958 		 * across a reset.  As a possible future enhancement, we
2959 		 * could re-scan for quirks every time we reinitialize
2960 		 * the device, but we'd have to make sure that the driver
2961 		 * behaves intelligently if the quirks change.
2962 		 */
2963 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2964 			if (quirk_matches(id, &core_quirks[i]))
2965 				ctrl->quirks |= core_quirks[i].quirks;
2966 		}
2967 	}
2968 
2969 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2970 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2971 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2972 	}
2973 
2974 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2975 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2976 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2977 
2978 	ctrl->oacs = le16_to_cpu(id->oacs);
2979 	ctrl->oncs = le16_to_cpu(id->oncs);
2980 	ctrl->mtfa = le16_to_cpu(id->mtfa);
2981 	ctrl->oaes = le32_to_cpu(id->oaes);
2982 	ctrl->wctemp = le16_to_cpu(id->wctemp);
2983 	ctrl->cctemp = le16_to_cpu(id->cctemp);
2984 
2985 	atomic_set(&ctrl->abort_limit, id->acl + 1);
2986 	ctrl->vwc = id->vwc;
2987 	if (id->mdts)
2988 		max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
2989 	else
2990 		max_hw_sectors = UINT_MAX;
2991 	ctrl->max_hw_sectors =
2992 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2993 
2994 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
2995 	ctrl->sgls = le32_to_cpu(id->sgls);
2996 	ctrl->kas = le16_to_cpu(id->kas);
2997 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
2998 	ctrl->ctratt = le32_to_cpu(id->ctratt);
2999 
3000 	if (id->rtd3e) {
3001 		/* us -> s */
3002 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3003 
3004 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3005 						 shutdown_timeout, 60);
3006 
3007 		if (ctrl->shutdown_timeout != shutdown_timeout)
3008 			dev_info(ctrl->device,
3009 				 "Shutdown timeout set to %u seconds\n",
3010 				 ctrl->shutdown_timeout);
3011 	} else
3012 		ctrl->shutdown_timeout = shutdown_timeout;
3013 
3014 	ctrl->npss = id->npss;
3015 	ctrl->apsta = id->apsta;
3016 	prev_apst_enabled = ctrl->apst_enabled;
3017 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3018 		if (force_apst && id->apsta) {
3019 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3020 			ctrl->apst_enabled = true;
3021 		} else {
3022 			ctrl->apst_enabled = false;
3023 		}
3024 	} else {
3025 		ctrl->apst_enabled = id->apsta;
3026 	}
3027 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3028 
3029 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3030 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3031 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3032 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3033 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3034 
3035 		/*
3036 		 * In fabrics we need to verify the cntlid matches the
3037 		 * admin connect
3038 		 */
3039 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3040 			dev_err(ctrl->device,
3041 				"Mismatching cntlid: Connect %u vs Identify "
3042 				"%u, rejecting\n",
3043 				ctrl->cntlid, le16_to_cpu(id->cntlid));
3044 			ret = -EINVAL;
3045 			goto out_free;
3046 		}
3047 
3048 		if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3049 			dev_err(ctrl->device,
3050 				"keep-alive support is mandatory for fabrics\n");
3051 			ret = -EINVAL;
3052 			goto out_free;
3053 		}
3054 	} else {
3055 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3056 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3057 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3058 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3059 	}
3060 
3061 	ret = nvme_mpath_init_identify(ctrl, id);
3062 	if (ret < 0)
3063 		goto out_free;
3064 
3065 	if (ctrl->apst_enabled && !prev_apst_enabled)
3066 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3067 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3068 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3069 
3070 out_free:
3071 	kfree(id);
3072 	return ret;
3073 }
3074 
3075 /*
3076  * Initialize the cached copies of the Identify data and various controller
3077  * register in our nvme_ctrl structure.  This should be called as soon as
3078  * the admin queue is fully up and running.
3079  */
3080 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3081 {
3082 	int ret;
3083 
3084 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3085 	if (ret) {
3086 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3087 		return ret;
3088 	}
3089 
3090 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3091 
3092 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3093 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3094 
3095 	ret = nvme_init_identify(ctrl);
3096 	if (ret)
3097 		return ret;
3098 
3099 	ret = nvme_init_non_mdts_limits(ctrl);
3100 	if (ret < 0)
3101 		return ret;
3102 
3103 	ret = nvme_configure_apst(ctrl);
3104 	if (ret < 0)
3105 		return ret;
3106 
3107 	ret = nvme_configure_timestamp(ctrl);
3108 	if (ret < 0)
3109 		return ret;
3110 
3111 	ret = nvme_configure_directives(ctrl);
3112 	if (ret < 0)
3113 		return ret;
3114 
3115 	ret = nvme_configure_acre(ctrl);
3116 	if (ret < 0)
3117 		return ret;
3118 
3119 	if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3120 		ret = nvme_hwmon_init(ctrl);
3121 		if (ret < 0)
3122 			return ret;
3123 	}
3124 
3125 	ctrl->identified = true;
3126 
3127 	return 0;
3128 }
3129 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3130 
3131 static int nvme_dev_open(struct inode *inode, struct file *file)
3132 {
3133 	struct nvme_ctrl *ctrl =
3134 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3135 
3136 	switch (ctrl->state) {
3137 	case NVME_CTRL_LIVE:
3138 		break;
3139 	default:
3140 		return -EWOULDBLOCK;
3141 	}
3142 
3143 	nvme_get_ctrl(ctrl);
3144 	if (!try_module_get(ctrl->ops->module)) {
3145 		nvme_put_ctrl(ctrl);
3146 		return -EINVAL;
3147 	}
3148 
3149 	file->private_data = ctrl;
3150 	return 0;
3151 }
3152 
3153 static int nvme_dev_release(struct inode *inode, struct file *file)
3154 {
3155 	struct nvme_ctrl *ctrl =
3156 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3157 
3158 	module_put(ctrl->ops->module);
3159 	nvme_put_ctrl(ctrl);
3160 	return 0;
3161 }
3162 
3163 static const struct file_operations nvme_dev_fops = {
3164 	.owner		= THIS_MODULE,
3165 	.open		= nvme_dev_open,
3166 	.release	= nvme_dev_release,
3167 	.unlocked_ioctl	= nvme_dev_ioctl,
3168 	.compat_ioctl	= compat_ptr_ioctl,
3169 };
3170 
3171 static ssize_t nvme_sysfs_reset(struct device *dev,
3172 				struct device_attribute *attr, const char *buf,
3173 				size_t count)
3174 {
3175 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3176 	int ret;
3177 
3178 	ret = nvme_reset_ctrl_sync(ctrl);
3179 	if (ret < 0)
3180 		return ret;
3181 	return count;
3182 }
3183 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3184 
3185 static ssize_t nvme_sysfs_rescan(struct device *dev,
3186 				struct device_attribute *attr, const char *buf,
3187 				size_t count)
3188 {
3189 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3190 
3191 	nvme_queue_scan(ctrl);
3192 	return count;
3193 }
3194 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3195 
3196 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3197 {
3198 	struct gendisk *disk = dev_to_disk(dev);
3199 
3200 	if (disk->fops == &nvme_bdev_ops)
3201 		return nvme_get_ns_from_dev(dev)->head;
3202 	else
3203 		return disk->private_data;
3204 }
3205 
3206 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3207 		char *buf)
3208 {
3209 	struct nvme_ns_head *head = dev_to_ns_head(dev);
3210 	struct nvme_ns_ids *ids = &head->ids;
3211 	struct nvme_subsystem *subsys = head->subsys;
3212 	int serial_len = sizeof(subsys->serial);
3213 	int model_len = sizeof(subsys->model);
3214 
3215 	if (!uuid_is_null(&ids->uuid))
3216 		return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3217 
3218 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3219 		return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3220 
3221 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3222 		return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3223 
3224 	while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3225 				  subsys->serial[serial_len - 1] == '\0'))
3226 		serial_len--;
3227 	while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3228 				 subsys->model[model_len - 1] == '\0'))
3229 		model_len--;
3230 
3231 	return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3232 		serial_len, subsys->serial, model_len, subsys->model,
3233 		head->ns_id);
3234 }
3235 static DEVICE_ATTR_RO(wwid);
3236 
3237 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3238 		char *buf)
3239 {
3240 	return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3241 }
3242 static DEVICE_ATTR_RO(nguid);
3243 
3244 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3245 		char *buf)
3246 {
3247 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3248 
3249 	/* For backward compatibility expose the NGUID to userspace if
3250 	 * we have no UUID set
3251 	 */
3252 	if (uuid_is_null(&ids->uuid)) {
3253 		printk_ratelimited(KERN_WARNING
3254 				   "No UUID available providing old NGUID\n");
3255 		return sysfs_emit(buf, "%pU\n", ids->nguid);
3256 	}
3257 	return sysfs_emit(buf, "%pU\n", &ids->uuid);
3258 }
3259 static DEVICE_ATTR_RO(uuid);
3260 
3261 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3262 		char *buf)
3263 {
3264 	return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3265 }
3266 static DEVICE_ATTR_RO(eui);
3267 
3268 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3269 		char *buf)
3270 {
3271 	return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3272 }
3273 static DEVICE_ATTR_RO(nsid);
3274 
3275 static struct attribute *nvme_ns_id_attrs[] = {
3276 	&dev_attr_wwid.attr,
3277 	&dev_attr_uuid.attr,
3278 	&dev_attr_nguid.attr,
3279 	&dev_attr_eui.attr,
3280 	&dev_attr_nsid.attr,
3281 #ifdef CONFIG_NVME_MULTIPATH
3282 	&dev_attr_ana_grpid.attr,
3283 	&dev_attr_ana_state.attr,
3284 #endif
3285 	NULL,
3286 };
3287 
3288 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3289 		struct attribute *a, int n)
3290 {
3291 	struct device *dev = container_of(kobj, struct device, kobj);
3292 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3293 
3294 	if (a == &dev_attr_uuid.attr) {
3295 		if (uuid_is_null(&ids->uuid) &&
3296 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3297 			return 0;
3298 	}
3299 	if (a == &dev_attr_nguid.attr) {
3300 		if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3301 			return 0;
3302 	}
3303 	if (a == &dev_attr_eui.attr) {
3304 		if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3305 			return 0;
3306 	}
3307 #ifdef CONFIG_NVME_MULTIPATH
3308 	if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3309 		if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3310 			return 0;
3311 		if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3312 			return 0;
3313 	}
3314 #endif
3315 	return a->mode;
3316 }
3317 
3318 static const struct attribute_group nvme_ns_id_attr_group = {
3319 	.attrs		= nvme_ns_id_attrs,
3320 	.is_visible	= nvme_ns_id_attrs_are_visible,
3321 };
3322 
3323 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3324 	&nvme_ns_id_attr_group,
3325 	NULL,
3326 };
3327 
3328 #define nvme_show_str_function(field)						\
3329 static ssize_t  field##_show(struct device *dev,				\
3330 			    struct device_attribute *attr, char *buf)		\
3331 {										\
3332         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3333         return sysfs_emit(buf, "%.*s\n",					\
3334 		(int)sizeof(ctrl->subsys->field), ctrl->subsys->field);		\
3335 }										\
3336 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3337 
3338 nvme_show_str_function(model);
3339 nvme_show_str_function(serial);
3340 nvme_show_str_function(firmware_rev);
3341 
3342 #define nvme_show_int_function(field)						\
3343 static ssize_t  field##_show(struct device *dev,				\
3344 			    struct device_attribute *attr, char *buf)		\
3345 {										\
3346         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3347         return sysfs_emit(buf, "%d\n", ctrl->field);				\
3348 }										\
3349 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3350 
3351 nvme_show_int_function(cntlid);
3352 nvme_show_int_function(numa_node);
3353 nvme_show_int_function(queue_count);
3354 nvme_show_int_function(sqsize);
3355 nvme_show_int_function(kato);
3356 
3357 static ssize_t nvme_sysfs_delete(struct device *dev,
3358 				struct device_attribute *attr, const char *buf,
3359 				size_t count)
3360 {
3361 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3362 
3363 	if (device_remove_file_self(dev, attr))
3364 		nvme_delete_ctrl_sync(ctrl);
3365 	return count;
3366 }
3367 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3368 
3369 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3370 					 struct device_attribute *attr,
3371 					 char *buf)
3372 {
3373 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3374 
3375 	return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3376 }
3377 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3378 
3379 static ssize_t nvme_sysfs_show_state(struct device *dev,
3380 				     struct device_attribute *attr,
3381 				     char *buf)
3382 {
3383 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3384 	static const char *const state_name[] = {
3385 		[NVME_CTRL_NEW]		= "new",
3386 		[NVME_CTRL_LIVE]	= "live",
3387 		[NVME_CTRL_RESETTING]	= "resetting",
3388 		[NVME_CTRL_CONNECTING]	= "connecting",
3389 		[NVME_CTRL_DELETING]	= "deleting",
3390 		[NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3391 		[NVME_CTRL_DEAD]	= "dead",
3392 	};
3393 
3394 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3395 	    state_name[ctrl->state])
3396 		return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3397 
3398 	return sysfs_emit(buf, "unknown state\n");
3399 }
3400 
3401 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3402 
3403 static ssize_t nvme_sysfs_show_subsysnqn(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, "%s\n", ctrl->subsys->subnqn);
3410 }
3411 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3412 
3413 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3414 					struct device_attribute *attr,
3415 					char *buf)
3416 {
3417 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3418 
3419 	return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3420 }
3421 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3422 
3423 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3424 					struct device_attribute *attr,
3425 					char *buf)
3426 {
3427 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3428 
3429 	return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3430 }
3431 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3432 
3433 static ssize_t nvme_sysfs_show_address(struct device *dev,
3434 					 struct device_attribute *attr,
3435 					 char *buf)
3436 {
3437 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3438 
3439 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3440 }
3441 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3442 
3443 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3444 		struct device_attribute *attr, char *buf)
3445 {
3446 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3447 	struct nvmf_ctrl_options *opts = ctrl->opts;
3448 
3449 	if (ctrl->opts->max_reconnects == -1)
3450 		return sysfs_emit(buf, "off\n");
3451 	return sysfs_emit(buf, "%d\n",
3452 			  opts->max_reconnects * opts->reconnect_delay);
3453 }
3454 
3455 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3456 		struct device_attribute *attr, const char *buf, size_t count)
3457 {
3458 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3459 	struct nvmf_ctrl_options *opts = ctrl->opts;
3460 	int ctrl_loss_tmo, err;
3461 
3462 	err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3463 	if (err)
3464 		return -EINVAL;
3465 
3466 	if (ctrl_loss_tmo < 0)
3467 		opts->max_reconnects = -1;
3468 	else
3469 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3470 						opts->reconnect_delay);
3471 	return count;
3472 }
3473 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3474 	nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3475 
3476 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3477 		struct device_attribute *attr, char *buf)
3478 {
3479 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3480 
3481 	if (ctrl->opts->reconnect_delay == -1)
3482 		return sysfs_emit(buf, "off\n");
3483 	return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3484 }
3485 
3486 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3487 		struct device_attribute *attr, const char *buf, size_t count)
3488 {
3489 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3490 	unsigned int v;
3491 	int err;
3492 
3493 	err = kstrtou32(buf, 10, &v);
3494 	if (err)
3495 		return err;
3496 
3497 	ctrl->opts->reconnect_delay = v;
3498 	return count;
3499 }
3500 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3501 	nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3502 
3503 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3504 		struct device_attribute *attr, char *buf)
3505 {
3506 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3507 
3508 	if (ctrl->opts->fast_io_fail_tmo == -1)
3509 		return sysfs_emit(buf, "off\n");
3510 	return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3511 }
3512 
3513 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3514 		struct device_attribute *attr, const char *buf, size_t count)
3515 {
3516 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3517 	struct nvmf_ctrl_options *opts = ctrl->opts;
3518 	int fast_io_fail_tmo, err;
3519 
3520 	err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3521 	if (err)
3522 		return -EINVAL;
3523 
3524 	if (fast_io_fail_tmo < 0)
3525 		opts->fast_io_fail_tmo = -1;
3526 	else
3527 		opts->fast_io_fail_tmo = fast_io_fail_tmo;
3528 	return count;
3529 }
3530 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3531 	nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3532 
3533 static struct attribute *nvme_dev_attrs[] = {
3534 	&dev_attr_reset_controller.attr,
3535 	&dev_attr_rescan_controller.attr,
3536 	&dev_attr_model.attr,
3537 	&dev_attr_serial.attr,
3538 	&dev_attr_firmware_rev.attr,
3539 	&dev_attr_cntlid.attr,
3540 	&dev_attr_delete_controller.attr,
3541 	&dev_attr_transport.attr,
3542 	&dev_attr_subsysnqn.attr,
3543 	&dev_attr_address.attr,
3544 	&dev_attr_state.attr,
3545 	&dev_attr_numa_node.attr,
3546 	&dev_attr_queue_count.attr,
3547 	&dev_attr_sqsize.attr,
3548 	&dev_attr_hostnqn.attr,
3549 	&dev_attr_hostid.attr,
3550 	&dev_attr_ctrl_loss_tmo.attr,
3551 	&dev_attr_reconnect_delay.attr,
3552 	&dev_attr_fast_io_fail_tmo.attr,
3553 	&dev_attr_kato.attr,
3554 	NULL
3555 };
3556 
3557 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3558 		struct attribute *a, int n)
3559 {
3560 	struct device *dev = container_of(kobj, struct device, kobj);
3561 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3562 
3563 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3564 		return 0;
3565 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3566 		return 0;
3567 	if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3568 		return 0;
3569 	if (a == &dev_attr_hostid.attr && !ctrl->opts)
3570 		return 0;
3571 	if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3572 		return 0;
3573 	if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3574 		return 0;
3575 	if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3576 		return 0;
3577 
3578 	return a->mode;
3579 }
3580 
3581 static const struct attribute_group nvme_dev_attrs_group = {
3582 	.attrs		= nvme_dev_attrs,
3583 	.is_visible	= nvme_dev_attrs_are_visible,
3584 };
3585 
3586 static const struct attribute_group *nvme_dev_attr_groups[] = {
3587 	&nvme_dev_attrs_group,
3588 	NULL,
3589 };
3590 
3591 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3592 		unsigned nsid)
3593 {
3594 	struct nvme_ns_head *h;
3595 
3596 	lockdep_assert_held(&subsys->lock);
3597 
3598 	list_for_each_entry(h, &subsys->nsheads, entry) {
3599 		if (h->ns_id != nsid)
3600 			continue;
3601 		if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3602 			return h;
3603 	}
3604 
3605 	return NULL;
3606 }
3607 
3608 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3609 		struct nvme_ns_head *new)
3610 {
3611 	struct nvme_ns_head *h;
3612 
3613 	lockdep_assert_held(&subsys->lock);
3614 
3615 	list_for_each_entry(h, &subsys->nsheads, entry) {
3616 		if (nvme_ns_ids_valid(&new->ids) &&
3617 		    nvme_ns_ids_equal(&new->ids, &h->ids))
3618 			return -EINVAL;
3619 	}
3620 
3621 	return 0;
3622 }
3623 
3624 static void nvme_cdev_rel(struct device *dev)
3625 {
3626 	ida_simple_remove(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3627 }
3628 
3629 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3630 {
3631 	cdev_device_del(cdev, cdev_device);
3632 	put_device(cdev_device);
3633 }
3634 
3635 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3636 		const struct file_operations *fops, struct module *owner)
3637 {
3638 	int minor, ret;
3639 
3640 	minor = ida_simple_get(&nvme_ns_chr_minor_ida, 0, 0, GFP_KERNEL);
3641 	if (minor < 0)
3642 		return minor;
3643 	cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3644 	cdev_device->class = nvme_ns_chr_class;
3645 	cdev_device->release = nvme_cdev_rel;
3646 	device_initialize(cdev_device);
3647 	cdev_init(cdev, fops);
3648 	cdev->owner = owner;
3649 	ret = cdev_device_add(cdev, cdev_device);
3650 	if (ret)
3651 		put_device(cdev_device);
3652 
3653 	return ret;
3654 }
3655 
3656 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3657 {
3658 	return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3659 }
3660 
3661 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3662 {
3663 	nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3664 	return 0;
3665 }
3666 
3667 static const struct file_operations nvme_ns_chr_fops = {
3668 	.owner		= THIS_MODULE,
3669 	.open		= nvme_ns_chr_open,
3670 	.release	= nvme_ns_chr_release,
3671 	.unlocked_ioctl	= nvme_ns_chr_ioctl,
3672 	.compat_ioctl	= compat_ptr_ioctl,
3673 };
3674 
3675 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3676 {
3677 	int ret;
3678 
3679 	ns->cdev_device.parent = ns->ctrl->device;
3680 	ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3681 			   ns->ctrl->instance, ns->head->instance);
3682 	if (ret)
3683 		return ret;
3684 
3685 	return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3686 			     ns->ctrl->ops->module);
3687 }
3688 
3689 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3690 		unsigned nsid, struct nvme_ns_ids *ids)
3691 {
3692 	struct nvme_ns_head *head;
3693 	size_t size = sizeof(*head);
3694 	int ret = -ENOMEM;
3695 
3696 #ifdef CONFIG_NVME_MULTIPATH
3697 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3698 #endif
3699 
3700 	head = kzalloc(size, GFP_KERNEL);
3701 	if (!head)
3702 		goto out;
3703 	ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3704 	if (ret < 0)
3705 		goto out_free_head;
3706 	head->instance = ret;
3707 	INIT_LIST_HEAD(&head->list);
3708 	ret = init_srcu_struct(&head->srcu);
3709 	if (ret)
3710 		goto out_ida_remove;
3711 	head->subsys = ctrl->subsys;
3712 	head->ns_id = nsid;
3713 	head->ids = *ids;
3714 	kref_init(&head->ref);
3715 
3716 	ret = __nvme_check_ids(ctrl->subsys, head);
3717 	if (ret) {
3718 		dev_err(ctrl->device,
3719 			"duplicate IDs for nsid %d\n", nsid);
3720 		goto out_cleanup_srcu;
3721 	}
3722 
3723 	if (head->ids.csi) {
3724 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3725 		if (ret)
3726 			goto out_cleanup_srcu;
3727 	} else
3728 		head->effects = ctrl->effects;
3729 
3730 	ret = nvme_mpath_alloc_disk(ctrl, head);
3731 	if (ret)
3732 		goto out_cleanup_srcu;
3733 
3734 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3735 
3736 	kref_get(&ctrl->subsys->ref);
3737 
3738 	return head;
3739 out_cleanup_srcu:
3740 	cleanup_srcu_struct(&head->srcu);
3741 out_ida_remove:
3742 	ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3743 out_free_head:
3744 	kfree(head);
3745 out:
3746 	if (ret > 0)
3747 		ret = blk_status_to_errno(nvme_error_status(ret));
3748 	return ERR_PTR(ret);
3749 }
3750 
3751 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3752 		struct nvme_ns_ids *ids, bool is_shared)
3753 {
3754 	struct nvme_ctrl *ctrl = ns->ctrl;
3755 	struct nvme_ns_head *head = NULL;
3756 	int ret = 0;
3757 
3758 	mutex_lock(&ctrl->subsys->lock);
3759 	head = nvme_find_ns_head(ctrl->subsys, nsid);
3760 	if (!head) {
3761 		head = nvme_alloc_ns_head(ctrl, nsid, ids);
3762 		if (IS_ERR(head)) {
3763 			ret = PTR_ERR(head);
3764 			goto out_unlock;
3765 		}
3766 		head->shared = is_shared;
3767 	} else {
3768 		ret = -EINVAL;
3769 		if (!is_shared || !head->shared) {
3770 			dev_err(ctrl->device,
3771 				"Duplicate unshared namespace %d\n", nsid);
3772 			goto out_put_ns_head;
3773 		}
3774 		if (!nvme_ns_ids_equal(&head->ids, ids)) {
3775 			dev_err(ctrl->device,
3776 				"IDs don't match for shared namespace %d\n",
3777 					nsid);
3778 			goto out_put_ns_head;
3779 		}
3780 	}
3781 
3782 	list_add_tail_rcu(&ns->siblings, &head->list);
3783 	ns->head = head;
3784 	mutex_unlock(&ctrl->subsys->lock);
3785 	return 0;
3786 
3787 out_put_ns_head:
3788 	nvme_put_ns_head(head);
3789 out_unlock:
3790 	mutex_unlock(&ctrl->subsys->lock);
3791 	return ret;
3792 }
3793 
3794 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3795 {
3796 	struct nvme_ns *ns, *ret = NULL;
3797 
3798 	down_read(&ctrl->namespaces_rwsem);
3799 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3800 		if (ns->head->ns_id == nsid) {
3801 			if (!nvme_get_ns(ns))
3802 				continue;
3803 			ret = ns;
3804 			break;
3805 		}
3806 		if (ns->head->ns_id > nsid)
3807 			break;
3808 	}
3809 	up_read(&ctrl->namespaces_rwsem);
3810 	return ret;
3811 }
3812 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3813 
3814 /*
3815  * Add the namespace to the controller list while keeping the list ordered.
3816  */
3817 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3818 {
3819 	struct nvme_ns *tmp;
3820 
3821 	list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3822 		if (tmp->head->ns_id < ns->head->ns_id) {
3823 			list_add(&ns->list, &tmp->list);
3824 			return;
3825 		}
3826 	}
3827 	list_add(&ns->list, &ns->ctrl->namespaces);
3828 }
3829 
3830 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3831 		struct nvme_ns_ids *ids)
3832 {
3833 	struct nvme_ns *ns;
3834 	struct gendisk *disk;
3835 	struct nvme_id_ns *id;
3836 	int node = ctrl->numa_node;
3837 
3838 	if (nvme_identify_ns(ctrl, nsid, ids, &id))
3839 		return;
3840 
3841 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3842 	if (!ns)
3843 		goto out_free_id;
3844 
3845 	disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3846 	if (IS_ERR(disk))
3847 		goto out_free_ns;
3848 	disk->fops = &nvme_bdev_ops;
3849 	disk->private_data = ns;
3850 
3851 	ns->disk = disk;
3852 	ns->queue = disk->queue;
3853 
3854 	if (ctrl->opts && ctrl->opts->data_digest)
3855 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3856 
3857 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3858 	if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3859 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3860 
3861 	ns->ctrl = ctrl;
3862 	kref_init(&ns->kref);
3863 
3864 	if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3865 		goto out_cleanup_disk;
3866 
3867 	/*
3868 	 * Without the multipath code enabled, multiple controller per
3869 	 * subsystems are visible as devices and thus we cannot use the
3870 	 * subsystem instance.
3871 	 */
3872 	if (!nvme_mpath_set_disk_name(ns, disk->disk_name, &disk->flags))
3873 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3874 			ns->head->instance);
3875 
3876 	if (nvme_update_ns_info(ns, id))
3877 		goto out_unlink_ns;
3878 
3879 	down_write(&ctrl->namespaces_rwsem);
3880 	nvme_ns_add_to_ctrl_list(ns);
3881 	up_write(&ctrl->namespaces_rwsem);
3882 	nvme_get_ctrl(ctrl);
3883 
3884 	if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
3885 		goto out_cleanup_ns_from_list;
3886 
3887 	if (!nvme_ns_head_multipath(ns->head))
3888 		nvme_add_ns_cdev(ns);
3889 
3890 	nvme_mpath_add_disk(ns, id);
3891 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3892 	kfree(id);
3893 
3894 	return;
3895 
3896  out_cleanup_ns_from_list:
3897 	nvme_put_ctrl(ctrl);
3898 	down_write(&ctrl->namespaces_rwsem);
3899 	list_del_init(&ns->list);
3900 	up_write(&ctrl->namespaces_rwsem);
3901  out_unlink_ns:
3902 	mutex_lock(&ctrl->subsys->lock);
3903 	list_del_rcu(&ns->siblings);
3904 	if (list_empty(&ns->head->list))
3905 		list_del_init(&ns->head->entry);
3906 	mutex_unlock(&ctrl->subsys->lock);
3907 	nvme_put_ns_head(ns->head);
3908  out_cleanup_disk:
3909 	blk_cleanup_disk(disk);
3910  out_free_ns:
3911 	kfree(ns);
3912  out_free_id:
3913 	kfree(id);
3914 }
3915 
3916 static void nvme_ns_remove(struct nvme_ns *ns)
3917 {
3918 	bool last_path = false;
3919 
3920 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3921 		return;
3922 
3923 	clear_bit(NVME_NS_READY, &ns->flags);
3924 	set_capacity(ns->disk, 0);
3925 	nvme_fault_inject_fini(&ns->fault_inject);
3926 
3927 	mutex_lock(&ns->ctrl->subsys->lock);
3928 	list_del_rcu(&ns->siblings);
3929 	if (list_empty(&ns->head->list)) {
3930 		list_del_init(&ns->head->entry);
3931 		last_path = true;
3932 	}
3933 	mutex_unlock(&ns->ctrl->subsys->lock);
3934 
3935 	/* guarantee not available in head->list */
3936 	synchronize_rcu();
3937 
3938 	/* wait for concurrent submissions */
3939 	if (nvme_mpath_clear_current_path(ns))
3940 		synchronize_srcu(&ns->head->srcu);
3941 
3942 	if (!nvme_ns_head_multipath(ns->head))
3943 		nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3944 	del_gendisk(ns->disk);
3945 	blk_cleanup_queue(ns->queue);
3946 
3947 	down_write(&ns->ctrl->namespaces_rwsem);
3948 	list_del_init(&ns->list);
3949 	up_write(&ns->ctrl->namespaces_rwsem);
3950 
3951 	if (last_path)
3952 		nvme_mpath_shutdown_disk(ns->head);
3953 	nvme_put_ns(ns);
3954 }
3955 
3956 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3957 {
3958 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3959 
3960 	if (ns) {
3961 		nvme_ns_remove(ns);
3962 		nvme_put_ns(ns);
3963 	}
3964 }
3965 
3966 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
3967 {
3968 	struct nvme_id_ns *id;
3969 	int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3970 
3971 	if (test_bit(NVME_NS_DEAD, &ns->flags))
3972 		goto out;
3973 
3974 	ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
3975 	if (ret)
3976 		goto out;
3977 
3978 	ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3979 	if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
3980 		dev_err(ns->ctrl->device,
3981 			"identifiers changed for nsid %d\n", ns->head->ns_id);
3982 		goto out_free_id;
3983 	}
3984 
3985 	ret = nvme_update_ns_info(ns, id);
3986 
3987 out_free_id:
3988 	kfree(id);
3989 out:
3990 	/*
3991 	 * Only remove the namespace if we got a fatal error back from the
3992 	 * device, otherwise ignore the error and just move on.
3993 	 *
3994 	 * TODO: we should probably schedule a delayed retry here.
3995 	 */
3996 	if (ret > 0 && (ret & NVME_SC_DNR))
3997 		nvme_ns_remove(ns);
3998 }
3999 
4000 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4001 {
4002 	struct nvme_ns_ids ids = { };
4003 	struct nvme_ns *ns;
4004 
4005 	if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4006 		return;
4007 
4008 	ns = nvme_find_get_ns(ctrl, nsid);
4009 	if (ns) {
4010 		nvme_validate_ns(ns, &ids);
4011 		nvme_put_ns(ns);
4012 		return;
4013 	}
4014 
4015 	switch (ids.csi) {
4016 	case NVME_CSI_NVM:
4017 		nvme_alloc_ns(ctrl, nsid, &ids);
4018 		break;
4019 	case NVME_CSI_ZNS:
4020 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4021 			dev_warn(ctrl->device,
4022 				"nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4023 				nsid);
4024 			break;
4025 		}
4026 		if (!nvme_multi_css(ctrl)) {
4027 			dev_warn(ctrl->device,
4028 				"command set not reported for nsid: %d\n",
4029 				nsid);
4030 			break;
4031 		}
4032 		nvme_alloc_ns(ctrl, nsid, &ids);
4033 		break;
4034 	default:
4035 		dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4036 			ids.csi, nsid);
4037 		break;
4038 	}
4039 }
4040 
4041 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4042 					unsigned nsid)
4043 {
4044 	struct nvme_ns *ns, *next;
4045 	LIST_HEAD(rm_list);
4046 
4047 	down_write(&ctrl->namespaces_rwsem);
4048 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4049 		if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4050 			list_move_tail(&ns->list, &rm_list);
4051 	}
4052 	up_write(&ctrl->namespaces_rwsem);
4053 
4054 	list_for_each_entry_safe(ns, next, &rm_list, list)
4055 		nvme_ns_remove(ns);
4056 
4057 }
4058 
4059 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4060 {
4061 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4062 	__le32 *ns_list;
4063 	u32 prev = 0;
4064 	int ret = 0, i;
4065 
4066 	if (nvme_ctrl_limited_cns(ctrl))
4067 		return -EOPNOTSUPP;
4068 
4069 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4070 	if (!ns_list)
4071 		return -ENOMEM;
4072 
4073 	for (;;) {
4074 		struct nvme_command cmd = {
4075 			.identify.opcode	= nvme_admin_identify,
4076 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
4077 			.identify.nsid		= cpu_to_le32(prev),
4078 		};
4079 
4080 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4081 					    NVME_IDENTIFY_DATA_SIZE);
4082 		if (ret) {
4083 			dev_warn(ctrl->device,
4084 				"Identify NS List failed (status=0x%x)\n", ret);
4085 			goto free;
4086 		}
4087 
4088 		for (i = 0; i < nr_entries; i++) {
4089 			u32 nsid = le32_to_cpu(ns_list[i]);
4090 
4091 			if (!nsid)	/* end of the list? */
4092 				goto out;
4093 			nvme_validate_or_alloc_ns(ctrl, nsid);
4094 			while (++prev < nsid)
4095 				nvme_ns_remove_by_nsid(ctrl, prev);
4096 		}
4097 	}
4098  out:
4099 	nvme_remove_invalid_namespaces(ctrl, prev);
4100  free:
4101 	kfree(ns_list);
4102 	return ret;
4103 }
4104 
4105 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4106 {
4107 	struct nvme_id_ctrl *id;
4108 	u32 nn, i;
4109 
4110 	if (nvme_identify_ctrl(ctrl, &id))
4111 		return;
4112 	nn = le32_to_cpu(id->nn);
4113 	kfree(id);
4114 
4115 	for (i = 1; i <= nn; i++)
4116 		nvme_validate_or_alloc_ns(ctrl, i);
4117 
4118 	nvme_remove_invalid_namespaces(ctrl, nn);
4119 }
4120 
4121 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4122 {
4123 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4124 	__le32 *log;
4125 	int error;
4126 
4127 	log = kzalloc(log_size, GFP_KERNEL);
4128 	if (!log)
4129 		return;
4130 
4131 	/*
4132 	 * We need to read the log to clear the AEN, but we don't want to rely
4133 	 * on it for the changed namespace information as userspace could have
4134 	 * raced with us in reading the log page, which could cause us to miss
4135 	 * updates.
4136 	 */
4137 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4138 			NVME_CSI_NVM, log, log_size, 0);
4139 	if (error)
4140 		dev_warn(ctrl->device,
4141 			"reading changed ns log failed: %d\n", error);
4142 
4143 	kfree(log);
4144 }
4145 
4146 static void nvme_scan_work(struct work_struct *work)
4147 {
4148 	struct nvme_ctrl *ctrl =
4149 		container_of(work, struct nvme_ctrl, scan_work);
4150 
4151 	/* No tagset on a live ctrl means IO queues could not created */
4152 	if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4153 		return;
4154 
4155 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4156 		dev_info(ctrl->device, "rescanning namespaces.\n");
4157 		nvme_clear_changed_ns_log(ctrl);
4158 	}
4159 
4160 	mutex_lock(&ctrl->scan_lock);
4161 	if (nvme_scan_ns_list(ctrl) != 0)
4162 		nvme_scan_ns_sequential(ctrl);
4163 	mutex_unlock(&ctrl->scan_lock);
4164 }
4165 
4166 /*
4167  * This function iterates the namespace list unlocked to allow recovery from
4168  * controller failure. It is up to the caller to ensure the namespace list is
4169  * not modified by scan work while this function is executing.
4170  */
4171 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4172 {
4173 	struct nvme_ns *ns, *next;
4174 	LIST_HEAD(ns_list);
4175 
4176 	/*
4177 	 * make sure to requeue I/O to all namespaces as these
4178 	 * might result from the scan itself and must complete
4179 	 * for the scan_work to make progress
4180 	 */
4181 	nvme_mpath_clear_ctrl_paths(ctrl);
4182 
4183 	/* prevent racing with ns scanning */
4184 	flush_work(&ctrl->scan_work);
4185 
4186 	/*
4187 	 * The dead states indicates the controller was not gracefully
4188 	 * disconnected. In that case, we won't be able to flush any data while
4189 	 * removing the namespaces' disks; fail all the queues now to avoid
4190 	 * potentially having to clean up the failed sync later.
4191 	 */
4192 	if (ctrl->state == NVME_CTRL_DEAD)
4193 		nvme_kill_queues(ctrl);
4194 
4195 	/* this is a no-op when called from the controller reset handler */
4196 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4197 
4198 	down_write(&ctrl->namespaces_rwsem);
4199 	list_splice_init(&ctrl->namespaces, &ns_list);
4200 	up_write(&ctrl->namespaces_rwsem);
4201 
4202 	list_for_each_entry_safe(ns, next, &ns_list, list)
4203 		nvme_ns_remove(ns);
4204 }
4205 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4206 
4207 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4208 {
4209 	struct nvme_ctrl *ctrl =
4210 		container_of(dev, struct nvme_ctrl, ctrl_device);
4211 	struct nvmf_ctrl_options *opts = ctrl->opts;
4212 	int ret;
4213 
4214 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4215 	if (ret)
4216 		return ret;
4217 
4218 	if (opts) {
4219 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4220 		if (ret)
4221 			return ret;
4222 
4223 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4224 				opts->trsvcid ?: "none");
4225 		if (ret)
4226 			return ret;
4227 
4228 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4229 				opts->host_traddr ?: "none");
4230 		if (ret)
4231 			return ret;
4232 
4233 		ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4234 				opts->host_iface ?: "none");
4235 	}
4236 	return ret;
4237 }
4238 
4239 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4240 {
4241 	char *envp[2] = { NULL, NULL };
4242 	u32 aen_result = ctrl->aen_result;
4243 
4244 	ctrl->aen_result = 0;
4245 	if (!aen_result)
4246 		return;
4247 
4248 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4249 	if (!envp[0])
4250 		return;
4251 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4252 	kfree(envp[0]);
4253 }
4254 
4255 static void nvme_async_event_work(struct work_struct *work)
4256 {
4257 	struct nvme_ctrl *ctrl =
4258 		container_of(work, struct nvme_ctrl, async_event_work);
4259 
4260 	nvme_aen_uevent(ctrl);
4261 	ctrl->ops->submit_async_event(ctrl);
4262 }
4263 
4264 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4265 {
4266 
4267 	u32 csts;
4268 
4269 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4270 		return false;
4271 
4272 	if (csts == ~0)
4273 		return false;
4274 
4275 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4276 }
4277 
4278 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4279 {
4280 	struct nvme_fw_slot_info_log *log;
4281 
4282 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4283 	if (!log)
4284 		return;
4285 
4286 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4287 			log, sizeof(*log), 0))
4288 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4289 	kfree(log);
4290 }
4291 
4292 static void nvme_fw_act_work(struct work_struct *work)
4293 {
4294 	struct nvme_ctrl *ctrl = container_of(work,
4295 				struct nvme_ctrl, fw_act_work);
4296 	unsigned long fw_act_timeout;
4297 
4298 	if (ctrl->mtfa)
4299 		fw_act_timeout = jiffies +
4300 				msecs_to_jiffies(ctrl->mtfa * 100);
4301 	else
4302 		fw_act_timeout = jiffies +
4303 				msecs_to_jiffies(admin_timeout * 1000);
4304 
4305 	nvme_stop_queues(ctrl);
4306 	while (nvme_ctrl_pp_status(ctrl)) {
4307 		if (time_after(jiffies, fw_act_timeout)) {
4308 			dev_warn(ctrl->device,
4309 				"Fw activation timeout, reset controller\n");
4310 			nvme_try_sched_reset(ctrl);
4311 			return;
4312 		}
4313 		msleep(100);
4314 	}
4315 
4316 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4317 		return;
4318 
4319 	nvme_start_queues(ctrl);
4320 	/* read FW slot information to clear the AER */
4321 	nvme_get_fw_slot_info(ctrl);
4322 }
4323 
4324 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4325 {
4326 	u32 aer_notice_type = (result & 0xff00) >> 8;
4327 
4328 	trace_nvme_async_event(ctrl, aer_notice_type);
4329 
4330 	switch (aer_notice_type) {
4331 	case NVME_AER_NOTICE_NS_CHANGED:
4332 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4333 		nvme_queue_scan(ctrl);
4334 		break;
4335 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4336 		/*
4337 		 * We are (ab)using the RESETTING state to prevent subsequent
4338 		 * recovery actions from interfering with the controller's
4339 		 * firmware activation.
4340 		 */
4341 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4342 			queue_work(nvme_wq, &ctrl->fw_act_work);
4343 		break;
4344 #ifdef CONFIG_NVME_MULTIPATH
4345 	case NVME_AER_NOTICE_ANA:
4346 		if (!ctrl->ana_log_buf)
4347 			break;
4348 		queue_work(nvme_wq, &ctrl->ana_work);
4349 		break;
4350 #endif
4351 	case NVME_AER_NOTICE_DISC_CHANGED:
4352 		ctrl->aen_result = result;
4353 		break;
4354 	default:
4355 		dev_warn(ctrl->device, "async event result %08x\n", result);
4356 	}
4357 }
4358 
4359 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4360 		volatile union nvme_result *res)
4361 {
4362 	u32 result = le32_to_cpu(res->u32);
4363 	u32 aer_type = result & 0x07;
4364 
4365 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4366 		return;
4367 
4368 	switch (aer_type) {
4369 	case NVME_AER_NOTICE:
4370 		nvme_handle_aen_notice(ctrl, result);
4371 		break;
4372 	case NVME_AER_ERROR:
4373 	case NVME_AER_SMART:
4374 	case NVME_AER_CSS:
4375 	case NVME_AER_VS:
4376 		trace_nvme_async_event(ctrl, aer_type);
4377 		ctrl->aen_result = result;
4378 		break;
4379 	default:
4380 		break;
4381 	}
4382 	queue_work(nvme_wq, &ctrl->async_event_work);
4383 }
4384 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4385 
4386 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4387 {
4388 	nvme_mpath_stop(ctrl);
4389 	nvme_stop_keep_alive(ctrl);
4390 	nvme_stop_failfast_work(ctrl);
4391 	flush_work(&ctrl->async_event_work);
4392 	cancel_work_sync(&ctrl->fw_act_work);
4393 }
4394 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4395 
4396 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4397 {
4398 	nvme_start_keep_alive(ctrl);
4399 
4400 	nvme_enable_aen(ctrl);
4401 
4402 	if (ctrl->queue_count > 1) {
4403 		nvme_queue_scan(ctrl);
4404 		nvme_start_queues(ctrl);
4405 	}
4406 }
4407 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4408 
4409 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4410 {
4411 	nvme_hwmon_exit(ctrl);
4412 	nvme_fault_inject_fini(&ctrl->fault_inject);
4413 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4414 	cdev_device_del(&ctrl->cdev, ctrl->device);
4415 	nvme_put_ctrl(ctrl);
4416 }
4417 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4418 
4419 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4420 {
4421 	struct nvme_effects_log	*cel;
4422 	unsigned long i;
4423 
4424 	xa_for_each(&ctrl->cels, i, cel) {
4425 		xa_erase(&ctrl->cels, i);
4426 		kfree(cel);
4427 	}
4428 
4429 	xa_destroy(&ctrl->cels);
4430 }
4431 
4432 static void nvme_free_ctrl(struct device *dev)
4433 {
4434 	struct nvme_ctrl *ctrl =
4435 		container_of(dev, struct nvme_ctrl, ctrl_device);
4436 	struct nvme_subsystem *subsys = ctrl->subsys;
4437 
4438 	if (!subsys || ctrl->instance != subsys->instance)
4439 		ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4440 
4441 	nvme_free_cels(ctrl);
4442 	nvme_mpath_uninit(ctrl);
4443 	__free_page(ctrl->discard_page);
4444 
4445 	if (subsys) {
4446 		mutex_lock(&nvme_subsystems_lock);
4447 		list_del(&ctrl->subsys_entry);
4448 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4449 		mutex_unlock(&nvme_subsystems_lock);
4450 	}
4451 
4452 	ctrl->ops->free_ctrl(ctrl);
4453 
4454 	if (subsys)
4455 		nvme_put_subsystem(subsys);
4456 }
4457 
4458 /*
4459  * Initialize a NVMe controller structures.  This needs to be called during
4460  * earliest initialization so that we have the initialized structured around
4461  * during probing.
4462  */
4463 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4464 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
4465 {
4466 	int ret;
4467 
4468 	ctrl->state = NVME_CTRL_NEW;
4469 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4470 	spin_lock_init(&ctrl->lock);
4471 	mutex_init(&ctrl->scan_lock);
4472 	INIT_LIST_HEAD(&ctrl->namespaces);
4473 	xa_init(&ctrl->cels);
4474 	init_rwsem(&ctrl->namespaces_rwsem);
4475 	ctrl->dev = dev;
4476 	ctrl->ops = ops;
4477 	ctrl->quirks = quirks;
4478 	ctrl->numa_node = NUMA_NO_NODE;
4479 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4480 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4481 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4482 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4483 	init_waitqueue_head(&ctrl->state_wq);
4484 
4485 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4486 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4487 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4488 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4489 
4490 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4491 			PAGE_SIZE);
4492 	ctrl->discard_page = alloc_page(GFP_KERNEL);
4493 	if (!ctrl->discard_page) {
4494 		ret = -ENOMEM;
4495 		goto out;
4496 	}
4497 
4498 	ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4499 	if (ret < 0)
4500 		goto out;
4501 	ctrl->instance = ret;
4502 
4503 	device_initialize(&ctrl->ctrl_device);
4504 	ctrl->device = &ctrl->ctrl_device;
4505 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4506 			ctrl->instance);
4507 	ctrl->device->class = nvme_class;
4508 	ctrl->device->parent = ctrl->dev;
4509 	ctrl->device->groups = nvme_dev_attr_groups;
4510 	ctrl->device->release = nvme_free_ctrl;
4511 	dev_set_drvdata(ctrl->device, ctrl);
4512 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4513 	if (ret)
4514 		goto out_release_instance;
4515 
4516 	nvme_get_ctrl(ctrl);
4517 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
4518 	ctrl->cdev.owner = ops->module;
4519 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4520 	if (ret)
4521 		goto out_free_name;
4522 
4523 	/*
4524 	 * Initialize latency tolerance controls.  The sysfs files won't
4525 	 * be visible to userspace unless the device actually supports APST.
4526 	 */
4527 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4528 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4529 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4530 
4531 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4532 	nvme_mpath_init_ctrl(ctrl);
4533 
4534 	return 0;
4535 out_free_name:
4536 	nvme_put_ctrl(ctrl);
4537 	kfree_const(ctrl->device->kobj.name);
4538 out_release_instance:
4539 	ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4540 out:
4541 	if (ctrl->discard_page)
4542 		__free_page(ctrl->discard_page);
4543 	return ret;
4544 }
4545 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4546 
4547 static void nvme_start_ns_queue(struct nvme_ns *ns)
4548 {
4549 	if (test_and_clear_bit(NVME_NS_STOPPED, &ns->flags))
4550 		blk_mq_unquiesce_queue(ns->queue);
4551 }
4552 
4553 static void nvme_stop_ns_queue(struct nvme_ns *ns)
4554 {
4555 	if (!test_and_set_bit(NVME_NS_STOPPED, &ns->flags))
4556 		blk_mq_quiesce_queue(ns->queue);
4557 	else
4558 		blk_mq_wait_quiesce_done(ns->queue);
4559 }
4560 
4561 /*
4562  * Prepare a queue for teardown.
4563  *
4564  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
4565  * the capacity to 0 after that to avoid blocking dispatchers that may be
4566  * holding bd_butex.  This will end buffered writers dirtying pages that can't
4567  * be synced.
4568  */
4569 static void nvme_set_queue_dying(struct nvme_ns *ns)
4570 {
4571 	if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
4572 		return;
4573 
4574 	blk_set_queue_dying(ns->queue);
4575 	nvme_start_ns_queue(ns);
4576 
4577 	set_capacity_and_notify(ns->disk, 0);
4578 }
4579 
4580 /**
4581  * nvme_kill_queues(): Ends all namespace queues
4582  * @ctrl: the dead controller that needs to end
4583  *
4584  * Call this function when the driver determines it is unable to get the
4585  * controller in a state capable of servicing IO.
4586  */
4587 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4588 {
4589 	struct nvme_ns *ns;
4590 
4591 	down_read(&ctrl->namespaces_rwsem);
4592 
4593 	/* Forcibly unquiesce queues to avoid blocking dispatch */
4594 	if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4595 		nvme_start_admin_queue(ctrl);
4596 
4597 	list_for_each_entry(ns, &ctrl->namespaces, list)
4598 		nvme_set_queue_dying(ns);
4599 
4600 	up_read(&ctrl->namespaces_rwsem);
4601 }
4602 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4603 
4604 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4605 {
4606 	struct nvme_ns *ns;
4607 
4608 	down_read(&ctrl->namespaces_rwsem);
4609 	list_for_each_entry(ns, &ctrl->namespaces, list)
4610 		blk_mq_unfreeze_queue(ns->queue);
4611 	up_read(&ctrl->namespaces_rwsem);
4612 }
4613 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4614 
4615 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4616 {
4617 	struct nvme_ns *ns;
4618 
4619 	down_read(&ctrl->namespaces_rwsem);
4620 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4621 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4622 		if (timeout <= 0)
4623 			break;
4624 	}
4625 	up_read(&ctrl->namespaces_rwsem);
4626 	return timeout;
4627 }
4628 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4629 
4630 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4631 {
4632 	struct nvme_ns *ns;
4633 
4634 	down_read(&ctrl->namespaces_rwsem);
4635 	list_for_each_entry(ns, &ctrl->namespaces, list)
4636 		blk_mq_freeze_queue_wait(ns->queue);
4637 	up_read(&ctrl->namespaces_rwsem);
4638 }
4639 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4640 
4641 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4642 {
4643 	struct nvme_ns *ns;
4644 
4645 	down_read(&ctrl->namespaces_rwsem);
4646 	list_for_each_entry(ns, &ctrl->namespaces, list)
4647 		blk_freeze_queue_start(ns->queue);
4648 	up_read(&ctrl->namespaces_rwsem);
4649 }
4650 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4651 
4652 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4653 {
4654 	struct nvme_ns *ns;
4655 
4656 	down_read(&ctrl->namespaces_rwsem);
4657 	list_for_each_entry(ns, &ctrl->namespaces, list)
4658 		nvme_stop_ns_queue(ns);
4659 	up_read(&ctrl->namespaces_rwsem);
4660 }
4661 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4662 
4663 void nvme_start_queues(struct nvme_ctrl *ctrl)
4664 {
4665 	struct nvme_ns *ns;
4666 
4667 	down_read(&ctrl->namespaces_rwsem);
4668 	list_for_each_entry(ns, &ctrl->namespaces, list)
4669 		nvme_start_ns_queue(ns);
4670 	up_read(&ctrl->namespaces_rwsem);
4671 }
4672 EXPORT_SYMBOL_GPL(nvme_start_queues);
4673 
4674 void nvme_stop_admin_queue(struct nvme_ctrl *ctrl)
4675 {
4676 	if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4677 		blk_mq_quiesce_queue(ctrl->admin_q);
4678 	else
4679 		blk_mq_wait_quiesce_done(ctrl->admin_q);
4680 }
4681 EXPORT_SYMBOL_GPL(nvme_stop_admin_queue);
4682 
4683 void nvme_start_admin_queue(struct nvme_ctrl *ctrl)
4684 {
4685 	if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4686 		blk_mq_unquiesce_queue(ctrl->admin_q);
4687 }
4688 EXPORT_SYMBOL_GPL(nvme_start_admin_queue);
4689 
4690 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4691 {
4692 	struct nvme_ns *ns;
4693 
4694 	down_read(&ctrl->namespaces_rwsem);
4695 	list_for_each_entry(ns, &ctrl->namespaces, list)
4696 		blk_sync_queue(ns->queue);
4697 	up_read(&ctrl->namespaces_rwsem);
4698 }
4699 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4700 
4701 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4702 {
4703 	nvme_sync_io_queues(ctrl);
4704 	if (ctrl->admin_q)
4705 		blk_sync_queue(ctrl->admin_q);
4706 }
4707 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4708 
4709 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4710 {
4711 	if (file->f_op != &nvme_dev_fops)
4712 		return NULL;
4713 	return file->private_data;
4714 }
4715 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4716 
4717 /*
4718  * Check we didn't inadvertently grow the command structure sizes:
4719  */
4720 static inline void _nvme_check_size(void)
4721 {
4722 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4723 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4724 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4725 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4726 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4727 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4728 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4729 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4730 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4731 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4732 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4733 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4734 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4735 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4736 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4737 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4738 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4739 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4740 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4741 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4742 }
4743 
4744 
4745 static int __init nvme_core_init(void)
4746 {
4747 	int result = -ENOMEM;
4748 
4749 	_nvme_check_size();
4750 
4751 	nvme_wq = alloc_workqueue("nvme-wq",
4752 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4753 	if (!nvme_wq)
4754 		goto out;
4755 
4756 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4757 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4758 	if (!nvme_reset_wq)
4759 		goto destroy_wq;
4760 
4761 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4762 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4763 	if (!nvme_delete_wq)
4764 		goto destroy_reset_wq;
4765 
4766 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4767 			NVME_MINORS, "nvme");
4768 	if (result < 0)
4769 		goto destroy_delete_wq;
4770 
4771 	nvme_class = class_create(THIS_MODULE, "nvme");
4772 	if (IS_ERR(nvme_class)) {
4773 		result = PTR_ERR(nvme_class);
4774 		goto unregister_chrdev;
4775 	}
4776 	nvme_class->dev_uevent = nvme_class_uevent;
4777 
4778 	nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4779 	if (IS_ERR(nvme_subsys_class)) {
4780 		result = PTR_ERR(nvme_subsys_class);
4781 		goto destroy_class;
4782 	}
4783 
4784 	result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4785 				     "nvme-generic");
4786 	if (result < 0)
4787 		goto destroy_subsys_class;
4788 
4789 	nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
4790 	if (IS_ERR(nvme_ns_chr_class)) {
4791 		result = PTR_ERR(nvme_ns_chr_class);
4792 		goto unregister_generic_ns;
4793 	}
4794 
4795 	return 0;
4796 
4797 unregister_generic_ns:
4798 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4799 destroy_subsys_class:
4800 	class_destroy(nvme_subsys_class);
4801 destroy_class:
4802 	class_destroy(nvme_class);
4803 unregister_chrdev:
4804 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4805 destroy_delete_wq:
4806 	destroy_workqueue(nvme_delete_wq);
4807 destroy_reset_wq:
4808 	destroy_workqueue(nvme_reset_wq);
4809 destroy_wq:
4810 	destroy_workqueue(nvme_wq);
4811 out:
4812 	return result;
4813 }
4814 
4815 static void __exit nvme_core_exit(void)
4816 {
4817 	class_destroy(nvme_ns_chr_class);
4818 	class_destroy(nvme_subsys_class);
4819 	class_destroy(nvme_class);
4820 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4821 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4822 	destroy_workqueue(nvme_delete_wq);
4823 	destroy_workqueue(nvme_reset_wq);
4824 	destroy_workqueue(nvme_wq);
4825 	ida_destroy(&nvme_ns_chr_minor_ida);
4826 	ida_destroy(&nvme_instance_ida);
4827 }
4828 
4829 MODULE_LICENSE("GPL");
4830 MODULE_VERSION("1.0");
4831 module_init(nvme_core_init);
4832 module_exit(nvme_core_exit);
4833