xref: /linux/drivers/nvme/host/core.c (revision cfc4ca8986bb1f6182da6cd7bb57f228590b4643)
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/async.h>
8 #include <linux/blkdev.h>
9 #include <linux/blk-mq.h>
10 #include <linux/blk-integrity.h>
11 #include <linux/compat.h>
12 #include <linux/delay.h>
13 #include <linux/errno.h>
14 #include <linux/hdreg.h>
15 #include <linux/kernel.h>
16 #include <linux/module.h>
17 #include <linux/backing-dev.h>
18 #include <linux/slab.h>
19 #include <linux/types.h>
20 #include <linux/pr.h>
21 #include <linux/ptrace.h>
22 #include <linux/nvme_ioctl.h>
23 #include <linux/pm_qos.h>
24 #include <linux/ratelimit.h>
25 #include <linux/unaligned.h>
26 
27 #include "nvme.h"
28 #include "fabrics.h"
29 #include <linux/nvme-auth.h>
30 
31 #define CREATE_TRACE_POINTS
32 #include "trace.h"
33 
34 #define NVME_MINORS		(1U << MINORBITS)
35 
36 struct nvme_ns_info {
37 	struct nvme_ns_ids ids;
38 	u32 nsid;
39 	__le32 anagrpid;
40 	u8 pi_offset;
41 	u16 endgid;
42 	u64 runs;
43 	bool is_shared;
44 	bool is_readonly;
45 	bool is_ready;
46 	bool is_removed;
47 	bool is_rotational;
48 	bool no_vwc;
49 };
50 
51 unsigned int admin_timeout = 60;
52 module_param(admin_timeout, uint, 0644);
53 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
54 EXPORT_SYMBOL_GPL(admin_timeout);
55 
56 unsigned int nvme_io_timeout = 30;
57 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
58 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
59 EXPORT_SYMBOL_GPL(nvme_io_timeout);
60 
61 static unsigned char shutdown_timeout = 5;
62 module_param(shutdown_timeout, byte, 0644);
63 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
64 
65 static u8 nvme_max_retries = 5;
66 module_param_named(max_retries, nvme_max_retries, byte, 0644);
67 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
68 
69 static unsigned long default_ps_max_latency_us = 100000;
70 module_param(default_ps_max_latency_us, ulong, 0644);
71 MODULE_PARM_DESC(default_ps_max_latency_us,
72 		 "max power saving latency for new devices; use PM QOS to change per device");
73 
74 static bool force_apst;
75 module_param(force_apst, bool, 0644);
76 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
77 
78 static unsigned long apst_primary_timeout_ms = 100;
79 module_param(apst_primary_timeout_ms, ulong, 0644);
80 MODULE_PARM_DESC(apst_primary_timeout_ms,
81 	"primary APST timeout in ms");
82 
83 static unsigned long apst_secondary_timeout_ms = 2000;
84 module_param(apst_secondary_timeout_ms, ulong, 0644);
85 MODULE_PARM_DESC(apst_secondary_timeout_ms,
86 	"secondary APST timeout in ms");
87 
88 static unsigned long apst_primary_latency_tol_us = 15000;
89 module_param(apst_primary_latency_tol_us, ulong, 0644);
90 MODULE_PARM_DESC(apst_primary_latency_tol_us,
91 	"primary APST latency tolerance in us");
92 
93 static unsigned long apst_secondary_latency_tol_us = 100000;
94 module_param(apst_secondary_latency_tol_us, ulong, 0644);
95 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
96 	"secondary APST latency tolerance in us");
97 
98 /*
99  * Older kernels didn't enable protection information if it was at an offset.
100  * Newer kernels do, so it breaks reads on the upgrade if such formats were
101  * used in prior kernels since the metadata written did not contain a valid
102  * checksum.
103  */
104 static bool disable_pi_offsets = false;
105 module_param(disable_pi_offsets, bool, 0444);
106 MODULE_PARM_DESC(disable_pi_offsets,
107 	"disable protection information if it has an offset");
108 
109 /*
110  * nvme_wq - hosts nvme related works that are not reset or delete
111  * nvme_reset_wq - hosts nvme reset works
112  * nvme_delete_wq - hosts nvme delete works
113  *
114  * nvme_wq will host works such as scan, aen handling, fw activation,
115  * keep-alive, periodic reconnects etc. nvme_reset_wq
116  * runs reset works which also flush works hosted on nvme_wq for
117  * serialization purposes. nvme_delete_wq host controller deletion
118  * works which flush reset works for serialization.
119  */
120 struct workqueue_struct *nvme_wq;
121 EXPORT_SYMBOL_GPL(nvme_wq);
122 
123 struct workqueue_struct *nvme_reset_wq;
124 EXPORT_SYMBOL_GPL(nvme_reset_wq);
125 
126 struct workqueue_struct *nvme_delete_wq;
127 EXPORT_SYMBOL_GPL(nvme_delete_wq);
128 
129 static LIST_HEAD(nvme_subsystems);
130 DEFINE_MUTEX(nvme_subsystems_lock);
131 
132 static DEFINE_IDA(nvme_instance_ida);
133 static dev_t nvme_ctrl_base_chr_devt;
134 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env);
135 static const struct class nvme_class = {
136 	.name = "nvme",
137 	.dev_uevent = nvme_class_uevent,
138 };
139 
140 static const struct class nvme_subsys_class = {
141 	.name = "nvme-subsystem",
142 };
143 
144 static DEFINE_IDA(nvme_ns_chr_minor_ida);
145 static dev_t nvme_ns_chr_devt;
146 static const struct class nvme_ns_chr_class = {
147 	.name = "nvme-generic",
148 };
149 
150 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
151 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
152 					   unsigned nsid);
153 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
154 				   struct nvme_command *cmd);
155 static int nvme_get_log_lsi(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page,
156 		u8 lsp, u8 csi, void *log, size_t size, u64 offset, u16 lsi);
157 
158 void nvme_queue_scan(struct nvme_ctrl *ctrl)
159 {
160 	/*
161 	 * Only new queue scan work when admin and IO queues are both alive
162 	 */
163 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE && ctrl->tagset)
164 		queue_work(nvme_wq, &ctrl->scan_work);
165 }
166 
167 /*
168  * Use this function to proceed with scheduling reset_work for a controller
169  * that had previously been set to the resetting state. This is intended for
170  * code paths that can't be interrupted by other reset attempts. A hot removal
171  * may prevent this from succeeding.
172  */
173 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
174 {
175 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_RESETTING)
176 		return -EBUSY;
177 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
178 		return -EBUSY;
179 	return 0;
180 }
181 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
182 
183 static void nvme_failfast_work(struct work_struct *work)
184 {
185 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
186 			struct nvme_ctrl, failfast_work);
187 
188 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_CONNECTING)
189 		return;
190 
191 	set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
192 	dev_info(ctrl->device, "failfast expired\n");
193 	nvme_kick_requeue_lists(ctrl);
194 }
195 
196 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
197 {
198 	if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
199 		return;
200 
201 	schedule_delayed_work(&ctrl->failfast_work,
202 			      ctrl->opts->fast_io_fail_tmo * HZ);
203 }
204 
205 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
206 {
207 	if (!ctrl->opts)
208 		return;
209 
210 	cancel_delayed_work_sync(&ctrl->failfast_work);
211 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
212 }
213 
214 
215 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
216 {
217 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
218 		return -EBUSY;
219 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
220 		return -EBUSY;
221 	return 0;
222 }
223 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
224 
225 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
226 {
227 	int ret;
228 
229 	ret = nvme_reset_ctrl(ctrl);
230 	if (!ret) {
231 		flush_work(&ctrl->reset_work);
232 		if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE)
233 			ret = -ENETRESET;
234 	}
235 
236 	return ret;
237 }
238 
239 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
240 {
241 	dev_info(ctrl->device,
242 		 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
243 
244 	flush_work(&ctrl->reset_work);
245 	nvme_stop_ctrl(ctrl);
246 	nvme_remove_namespaces(ctrl);
247 	ctrl->ops->delete_ctrl(ctrl);
248 	nvme_uninit_ctrl(ctrl);
249 }
250 
251 static void nvme_delete_ctrl_work(struct work_struct *work)
252 {
253 	struct nvme_ctrl *ctrl =
254 		container_of(work, struct nvme_ctrl, delete_work);
255 
256 	nvme_do_delete_ctrl(ctrl);
257 }
258 
259 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
260 {
261 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
262 		return -EBUSY;
263 	if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
264 		return -EBUSY;
265 	return 0;
266 }
267 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
268 
269 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
270 {
271 	/*
272 	 * Keep a reference until nvme_do_delete_ctrl() complete,
273 	 * since ->delete_ctrl can free the controller.
274 	 */
275 	nvme_get_ctrl(ctrl);
276 	if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
277 		nvme_do_delete_ctrl(ctrl);
278 	nvme_put_ctrl(ctrl);
279 }
280 
281 static blk_status_t nvme_error_status(u16 status)
282 {
283 	switch (status & NVME_SCT_SC_MASK) {
284 	case NVME_SC_SUCCESS:
285 		return BLK_STS_OK;
286 	case NVME_SC_CAP_EXCEEDED:
287 		return BLK_STS_NOSPC;
288 	case NVME_SC_LBA_RANGE:
289 	case NVME_SC_CMD_INTERRUPTED:
290 	case NVME_SC_NS_NOT_READY:
291 		return BLK_STS_TARGET;
292 	case NVME_SC_BAD_ATTRIBUTES:
293 	case NVME_SC_ONCS_NOT_SUPPORTED:
294 	case NVME_SC_INVALID_OPCODE:
295 	case NVME_SC_INVALID_FIELD:
296 	case NVME_SC_INVALID_NS:
297 		return BLK_STS_NOTSUPP;
298 	case NVME_SC_WRITE_FAULT:
299 	case NVME_SC_READ_ERROR:
300 	case NVME_SC_UNWRITTEN_BLOCK:
301 	case NVME_SC_ACCESS_DENIED:
302 	case NVME_SC_READ_ONLY:
303 	case NVME_SC_COMPARE_FAILED:
304 		return BLK_STS_MEDIUM;
305 	case NVME_SC_GUARD_CHECK:
306 	case NVME_SC_APPTAG_CHECK:
307 	case NVME_SC_REFTAG_CHECK:
308 	case NVME_SC_INVALID_PI:
309 		return BLK_STS_PROTECTION;
310 	case NVME_SC_RESERVATION_CONFLICT:
311 		return BLK_STS_RESV_CONFLICT;
312 	case NVME_SC_HOST_PATH_ERROR:
313 		return BLK_STS_TRANSPORT;
314 	case NVME_SC_ZONE_TOO_MANY_ACTIVE:
315 		return BLK_STS_ZONE_ACTIVE_RESOURCE;
316 	case NVME_SC_ZONE_TOO_MANY_OPEN:
317 		return BLK_STS_ZONE_OPEN_RESOURCE;
318 	default:
319 		return BLK_STS_IOERR;
320 	}
321 }
322 
323 static void nvme_retry_req(struct request *req)
324 {
325 	unsigned long delay = 0;
326 	u16 crd;
327 
328 	/* The mask and shift result must be <= 3 */
329 	crd = (nvme_req(req)->status & NVME_STATUS_CRD) >> 11;
330 	if (crd)
331 		delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
332 
333 	nvme_req(req)->retries++;
334 	blk_mq_requeue_request(req, false);
335 	blk_mq_delay_kick_requeue_list(req->q, delay);
336 }
337 
338 static void nvme_log_error(struct request *req)
339 {
340 	struct nvme_ns *ns = req->q->queuedata;
341 	struct nvme_request *nr = nvme_req(req);
342 
343 	if (ns) {
344 		pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %u blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
345 		       ns->disk ? ns->disk->disk_name : "?",
346 		       nvme_get_opcode_str(nr->cmd->common.opcode),
347 		       nr->cmd->common.opcode,
348 		       nvme_sect_to_lba(ns->head, blk_rq_pos(req)),
349 		       blk_rq_bytes(req) >> ns->head->lba_shift,
350 		       nvme_get_error_status_str(nr->status),
351 		       NVME_SCT(nr->status),		/* Status Code Type */
352 		       nr->status & NVME_SC_MASK,	/* Status Code */
353 		       nr->status & NVME_STATUS_MORE ? "MORE " : "",
354 		       nr->status & NVME_STATUS_DNR  ? "DNR "  : "");
355 		return;
356 	}
357 
358 	pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
359 			   dev_name(nr->ctrl->device),
360 			   nvme_get_admin_opcode_str(nr->cmd->common.opcode),
361 			   nr->cmd->common.opcode,
362 			   nvme_get_error_status_str(nr->status),
363 			   NVME_SCT(nr->status),	/* Status Code Type */
364 			   nr->status & NVME_SC_MASK,	/* Status Code */
365 			   nr->status & NVME_STATUS_MORE ? "MORE " : "",
366 			   nr->status & NVME_STATUS_DNR  ? "DNR "  : "");
367 }
368 
369 static void nvme_log_err_passthru(struct request *req)
370 {
371 	struct nvme_ns *ns = req->q->queuedata;
372 	struct nvme_request *nr = nvme_req(req);
373 
374 	pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s"
375 		"cdw10=0x%x cdw11=0x%x cdw12=0x%x cdw13=0x%x cdw14=0x%x cdw15=0x%x\n",
376 		ns ? ns->disk->disk_name : dev_name(nr->ctrl->device),
377 		ns ? nvme_get_opcode_str(nr->cmd->common.opcode) :
378 		     nvme_get_admin_opcode_str(nr->cmd->common.opcode),
379 		nr->cmd->common.opcode,
380 		nvme_get_error_status_str(nr->status),
381 		NVME_SCT(nr->status),		/* Status Code Type */
382 		nr->status & NVME_SC_MASK,	/* Status Code */
383 		nr->status & NVME_STATUS_MORE ? "MORE " : "",
384 		nr->status & NVME_STATUS_DNR  ? "DNR "  : "",
385 		nr->cmd->common.cdw10,
386 		nr->cmd->common.cdw11,
387 		nr->cmd->common.cdw12,
388 		nr->cmd->common.cdw13,
389 		nr->cmd->common.cdw14,
390 		nr->cmd->common.cdw14);
391 }
392 
393 enum nvme_disposition {
394 	COMPLETE,
395 	RETRY,
396 	FAILOVER,
397 	AUTHENTICATE,
398 };
399 
400 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
401 {
402 	if (likely(nvme_req(req)->status == 0))
403 		return COMPLETE;
404 
405 	if (blk_noretry_request(req) ||
406 	    (nvme_req(req)->status & NVME_STATUS_DNR) ||
407 	    nvme_req(req)->retries >= nvme_max_retries)
408 		return COMPLETE;
409 
410 	if ((nvme_req(req)->status & NVME_SCT_SC_MASK) == NVME_SC_AUTH_REQUIRED)
411 		return AUTHENTICATE;
412 
413 	if (req->cmd_flags & REQ_NVME_MPATH) {
414 		if (nvme_is_path_error(nvme_req(req)->status) ||
415 		    blk_queue_dying(req->q))
416 			return FAILOVER;
417 	} else {
418 		if (blk_queue_dying(req->q))
419 			return COMPLETE;
420 	}
421 
422 	return RETRY;
423 }
424 
425 static inline void nvme_end_req_zoned(struct request *req)
426 {
427 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
428 	    req_op(req) == REQ_OP_ZONE_APPEND) {
429 		struct nvme_ns *ns = req->q->queuedata;
430 
431 		req->__sector = nvme_lba_to_sect(ns->head,
432 			le64_to_cpu(nvme_req(req)->result.u64));
433 	}
434 }
435 
436 static inline void __nvme_end_req(struct request *req)
437 {
438 	if (unlikely(nvme_req(req)->status && !(req->rq_flags & RQF_QUIET))) {
439 		if (blk_rq_is_passthrough(req))
440 			nvme_log_err_passthru(req);
441 		else
442 			nvme_log_error(req);
443 	}
444 	nvme_end_req_zoned(req);
445 	nvme_trace_bio_complete(req);
446 	if (req->cmd_flags & REQ_NVME_MPATH)
447 		nvme_mpath_end_request(req);
448 }
449 
450 void nvme_end_req(struct request *req)
451 {
452 	blk_status_t status = nvme_error_status(nvme_req(req)->status);
453 
454 	__nvme_end_req(req);
455 	blk_mq_end_request(req, status);
456 }
457 
458 void nvme_complete_rq(struct request *req)
459 {
460 	struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
461 
462 	trace_nvme_complete_rq(req);
463 	nvme_cleanup_cmd(req);
464 
465 	/*
466 	 * Completions of long-running commands should not be able to
467 	 * defer sending of periodic keep alives, since the controller
468 	 * may have completed processing such commands a long time ago
469 	 * (arbitrarily close to command submission time).
470 	 * req->deadline - req->timeout is the command submission time
471 	 * in jiffies.
472 	 */
473 	if (ctrl->kas &&
474 	    req->deadline - req->timeout >= ctrl->ka_last_check_time)
475 		ctrl->comp_seen = true;
476 
477 	switch (nvme_decide_disposition(req)) {
478 	case COMPLETE:
479 		nvme_end_req(req);
480 		return;
481 	case RETRY:
482 		nvme_retry_req(req);
483 		return;
484 	case FAILOVER:
485 		nvme_failover_req(req);
486 		return;
487 	case AUTHENTICATE:
488 #ifdef CONFIG_NVME_HOST_AUTH
489 		queue_work(nvme_wq, &ctrl->dhchap_auth_work);
490 		nvme_retry_req(req);
491 #else
492 		nvme_end_req(req);
493 #endif
494 		return;
495 	}
496 }
497 EXPORT_SYMBOL_GPL(nvme_complete_rq);
498 
499 void nvme_complete_batch_req(struct request *req)
500 {
501 	trace_nvme_complete_rq(req);
502 	nvme_cleanup_cmd(req);
503 	__nvme_end_req(req);
504 }
505 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
506 
507 /*
508  * Called to unwind from ->queue_rq on a failed command submission so that the
509  * multipathing code gets called to potentially failover to another path.
510  * The caller needs to unwind all transport specific resource allocations and
511  * must return propagate the return value.
512  */
513 blk_status_t nvme_host_path_error(struct request *req)
514 {
515 	nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
516 	blk_mq_set_request_complete(req);
517 	nvme_complete_rq(req);
518 	return BLK_STS_OK;
519 }
520 EXPORT_SYMBOL_GPL(nvme_host_path_error);
521 
522 bool nvme_cancel_request(struct request *req, void *data)
523 {
524 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
525 				"Cancelling I/O %d", req->tag);
526 
527 	/* don't abort one completed or idle request */
528 	if (blk_mq_rq_state(req) != MQ_RQ_IN_FLIGHT)
529 		return true;
530 
531 	nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
532 	nvme_req(req)->flags |= NVME_REQ_CANCELLED;
533 	blk_mq_complete_request(req);
534 	return true;
535 }
536 EXPORT_SYMBOL_GPL(nvme_cancel_request);
537 
538 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
539 {
540 	if (ctrl->tagset) {
541 		blk_mq_tagset_busy_iter(ctrl->tagset,
542 				nvme_cancel_request, ctrl);
543 		blk_mq_tagset_wait_completed_request(ctrl->tagset);
544 	}
545 }
546 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
547 
548 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
549 {
550 	if (ctrl->admin_tagset) {
551 		blk_mq_tagset_busy_iter(ctrl->admin_tagset,
552 				nvme_cancel_request, ctrl);
553 		blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
554 	}
555 }
556 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
557 
558 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
559 		enum nvme_ctrl_state new_state)
560 {
561 	enum nvme_ctrl_state old_state;
562 	unsigned long flags;
563 	bool changed = false;
564 
565 	spin_lock_irqsave(&ctrl->lock, flags);
566 
567 	old_state = nvme_ctrl_state(ctrl);
568 	switch (new_state) {
569 	case NVME_CTRL_LIVE:
570 		switch (old_state) {
571 		case NVME_CTRL_CONNECTING:
572 			changed = true;
573 			fallthrough;
574 		default:
575 			break;
576 		}
577 		break;
578 	case NVME_CTRL_RESETTING:
579 		switch (old_state) {
580 		case NVME_CTRL_NEW:
581 		case NVME_CTRL_LIVE:
582 			changed = true;
583 			fallthrough;
584 		default:
585 			break;
586 		}
587 		break;
588 	case NVME_CTRL_CONNECTING:
589 		switch (old_state) {
590 		case NVME_CTRL_NEW:
591 		case NVME_CTRL_RESETTING:
592 			changed = true;
593 			fallthrough;
594 		default:
595 			break;
596 		}
597 		break;
598 	case NVME_CTRL_DELETING:
599 		switch (old_state) {
600 		case NVME_CTRL_LIVE:
601 		case NVME_CTRL_RESETTING:
602 		case NVME_CTRL_CONNECTING:
603 			changed = true;
604 			fallthrough;
605 		default:
606 			break;
607 		}
608 		break;
609 	case NVME_CTRL_DELETING_NOIO:
610 		switch (old_state) {
611 		case NVME_CTRL_DELETING:
612 		case NVME_CTRL_DEAD:
613 			changed = true;
614 			fallthrough;
615 		default:
616 			break;
617 		}
618 		break;
619 	case NVME_CTRL_DEAD:
620 		switch (old_state) {
621 		case NVME_CTRL_DELETING:
622 			changed = true;
623 			fallthrough;
624 		default:
625 			break;
626 		}
627 		break;
628 	default:
629 		break;
630 	}
631 
632 	if (changed) {
633 		WRITE_ONCE(ctrl->state, new_state);
634 		wake_up_all(&ctrl->state_wq);
635 	}
636 
637 	spin_unlock_irqrestore(&ctrl->lock, flags);
638 	if (!changed)
639 		return false;
640 
641 	if (new_state == NVME_CTRL_LIVE) {
642 		if (old_state == NVME_CTRL_CONNECTING)
643 			nvme_stop_failfast_work(ctrl);
644 		nvme_kick_requeue_lists(ctrl);
645 	} else if (new_state == NVME_CTRL_CONNECTING &&
646 		old_state == NVME_CTRL_RESETTING) {
647 		nvme_start_failfast_work(ctrl);
648 	}
649 	return changed;
650 }
651 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
652 
653 /*
654  * Waits for the controller state to be resetting, or returns false if it is
655  * not possible to ever transition to that state.
656  */
657 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
658 {
659 	wait_event(ctrl->state_wq,
660 		   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
661 		   nvme_state_terminal(ctrl));
662 	return nvme_ctrl_state(ctrl) == NVME_CTRL_RESETTING;
663 }
664 EXPORT_SYMBOL_GPL(nvme_wait_reset);
665 
666 static void nvme_free_ns_head(struct kref *ref)
667 {
668 	struct nvme_ns_head *head =
669 		container_of(ref, struct nvme_ns_head, ref);
670 
671 	nvme_mpath_put_disk(head);
672 	ida_free(&head->subsys->ns_ida, head->instance);
673 	cleanup_srcu_struct(&head->srcu);
674 	nvme_put_subsystem(head->subsys);
675 	kfree(head->plids);
676 	kfree(head);
677 }
678 
679 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
680 {
681 	return kref_get_unless_zero(&head->ref);
682 }
683 
684 void nvme_put_ns_head(struct nvme_ns_head *head)
685 {
686 	kref_put(&head->ref, nvme_free_ns_head);
687 }
688 
689 static void nvme_free_ns(struct kref *kref)
690 {
691 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
692 
693 	put_disk(ns->disk);
694 	nvme_put_ns_head(ns->head);
695 	nvme_put_ctrl(ns->ctrl);
696 	kfree(ns);
697 }
698 
699 bool nvme_get_ns(struct nvme_ns *ns)
700 {
701 	return kref_get_unless_zero(&ns->kref);
702 }
703 
704 void nvme_put_ns(struct nvme_ns *ns)
705 {
706 	kref_put(&ns->kref, nvme_free_ns);
707 }
708 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, "NVME_TARGET_PASSTHRU");
709 
710 static inline void nvme_clear_nvme_request(struct request *req)
711 {
712 	nvme_req(req)->status = 0;
713 	nvme_req(req)->retries = 0;
714 	nvme_req(req)->flags = 0;
715 	req->rq_flags |= RQF_DONTPREP;
716 }
717 
718 /* initialize a passthrough request */
719 void nvme_init_request(struct request *req, struct nvme_command *cmd)
720 {
721 	struct nvme_request *nr = nvme_req(req);
722 	bool logging_enabled;
723 
724 	if (req->q->queuedata) {
725 		struct nvme_ns *ns = req->q->disk->private_data;
726 
727 		logging_enabled = ns->head->passthru_err_log_enabled;
728 		req->timeout = NVME_IO_TIMEOUT;
729 	} else { /* no queuedata implies admin queue */
730 		logging_enabled = nr->ctrl->passthru_err_log_enabled;
731 		req->timeout = NVME_ADMIN_TIMEOUT;
732 	}
733 
734 	if (!logging_enabled)
735 		req->rq_flags |= RQF_QUIET;
736 
737 	/* passthru commands should let the driver set the SGL flags */
738 	cmd->common.flags &= ~NVME_CMD_SGL_ALL;
739 
740 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
741 	if (req->mq_hctx->type == HCTX_TYPE_POLL)
742 		req->cmd_flags |= REQ_POLLED;
743 	nvme_clear_nvme_request(req);
744 	memcpy(nr->cmd, cmd, sizeof(*cmd));
745 }
746 EXPORT_SYMBOL_GPL(nvme_init_request);
747 
748 /*
749  * For something we're not in a state to send to the device the default action
750  * is to busy it and retry it after the controller state is recovered.  However,
751  * if the controller is deleting or if anything is marked for failfast or
752  * nvme multipath it is immediately failed.
753  *
754  * Note: commands used to initialize the controller will be marked for failfast.
755  * Note: nvme cli/ioctl commands are marked for failfast.
756  */
757 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
758 		struct request *rq)
759 {
760 	enum nvme_ctrl_state state = nvme_ctrl_state(ctrl);
761 
762 	if (state != NVME_CTRL_DELETING_NOIO &&
763 	    state != NVME_CTRL_DELETING &&
764 	    state != NVME_CTRL_DEAD &&
765 	    !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
766 	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
767 		return BLK_STS_RESOURCE;
768 	return nvme_host_path_error(rq);
769 }
770 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
771 
772 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
773 		bool queue_live, enum nvme_ctrl_state state)
774 {
775 	struct nvme_request *req = nvme_req(rq);
776 
777 	/*
778 	 * currently we have a problem sending passthru commands
779 	 * on the admin_q if the controller is not LIVE because we can't
780 	 * make sure that they are going out after the admin connect,
781 	 * controller enable and/or other commands in the initialization
782 	 * sequence. until the controller will be LIVE, fail with
783 	 * BLK_STS_RESOURCE so that they will be rescheduled.
784 	 */
785 	if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
786 		return false;
787 
788 	if (ctrl->ops->flags & NVME_F_FABRICS) {
789 		/*
790 		 * Only allow commands on a live queue, except for the connect
791 		 * command, which is require to set the queue live in the
792 		 * appropinquate states.
793 		 */
794 		switch (state) {
795 		case NVME_CTRL_CONNECTING:
796 			if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
797 			    (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
798 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
799 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
800 				return true;
801 			break;
802 		default:
803 			break;
804 		case NVME_CTRL_DEAD:
805 			return false;
806 		}
807 	}
808 
809 	return queue_live;
810 }
811 EXPORT_SYMBOL_GPL(__nvme_check_ready);
812 
813 static inline void nvme_setup_flush(struct nvme_ns *ns,
814 		struct nvme_command *cmnd)
815 {
816 	memset(cmnd, 0, sizeof(*cmnd));
817 	cmnd->common.opcode = nvme_cmd_flush;
818 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
819 }
820 
821 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
822 		struct nvme_command *cmnd)
823 {
824 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
825 	struct nvme_dsm_range *range;
826 	struct bio *bio;
827 
828 	/*
829 	 * Some devices do not consider the DSM 'Number of Ranges' field when
830 	 * determining how much data to DMA. Always allocate memory for maximum
831 	 * number of segments to prevent device reading beyond end of buffer.
832 	 */
833 	static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
834 
835 	range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
836 	if (!range) {
837 		/*
838 		 * If we fail allocation our range, fallback to the controller
839 		 * discard page. If that's also busy, it's safe to return
840 		 * busy, as we know we can make progress once that's freed.
841 		 */
842 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
843 			return BLK_STS_RESOURCE;
844 
845 		range = page_address(ns->ctrl->discard_page);
846 	}
847 
848 	if (queue_max_discard_segments(req->q) == 1) {
849 		u64 slba = nvme_sect_to_lba(ns->head, blk_rq_pos(req));
850 		u32 nlb = blk_rq_sectors(req) >> (ns->head->lba_shift - 9);
851 
852 		range[0].cattr = cpu_to_le32(0);
853 		range[0].nlb = cpu_to_le32(nlb);
854 		range[0].slba = cpu_to_le64(slba);
855 		n = 1;
856 	} else {
857 		__rq_for_each_bio(bio, req) {
858 			u64 slba = nvme_sect_to_lba(ns->head,
859 						    bio->bi_iter.bi_sector);
860 			u32 nlb = bio->bi_iter.bi_size >> ns->head->lba_shift;
861 
862 			if (n < segments) {
863 				range[n].cattr = cpu_to_le32(0);
864 				range[n].nlb = cpu_to_le32(nlb);
865 				range[n].slba = cpu_to_le64(slba);
866 			}
867 			n++;
868 		}
869 	}
870 
871 	if (WARN_ON_ONCE(n != segments)) {
872 		if (virt_to_page(range) == ns->ctrl->discard_page)
873 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
874 		else
875 			kfree(range);
876 		return BLK_STS_IOERR;
877 	}
878 
879 	memset(cmnd, 0, sizeof(*cmnd));
880 	cmnd->dsm.opcode = nvme_cmd_dsm;
881 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
882 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
883 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
884 
885 	bvec_set_virt(&req->special_vec, range, alloc_size);
886 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
887 
888 	return BLK_STS_OK;
889 }
890 
891 static void nvme_set_app_tag(struct request *req, struct nvme_command *cmnd)
892 {
893 	cmnd->rw.lbat = cpu_to_le16(bio_integrity(req->bio)->app_tag);
894 	cmnd->rw.lbatm = cpu_to_le16(0xffff);
895 }
896 
897 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
898 			      struct request *req)
899 {
900 	u32 upper, lower;
901 	u64 ref48;
902 
903 	/* both rw and write zeroes share the same reftag format */
904 	switch (ns->head->guard_type) {
905 	case NVME_NVM_NS_16B_GUARD:
906 		cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
907 		break;
908 	case NVME_NVM_NS_64B_GUARD:
909 		ref48 = ext_pi_ref_tag(req);
910 		lower = lower_32_bits(ref48);
911 		upper = upper_32_bits(ref48);
912 
913 		cmnd->rw.reftag = cpu_to_le32(lower);
914 		cmnd->rw.cdw3 = cpu_to_le32(upper);
915 		break;
916 	default:
917 		break;
918 	}
919 }
920 
921 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
922 		struct request *req, struct nvme_command *cmnd)
923 {
924 	memset(cmnd, 0, sizeof(*cmnd));
925 
926 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
927 		return nvme_setup_discard(ns, req, cmnd);
928 
929 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
930 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
931 	cmnd->write_zeroes.slba =
932 		cpu_to_le64(nvme_sect_to_lba(ns->head, blk_rq_pos(req)));
933 	cmnd->write_zeroes.length =
934 		cpu_to_le16((blk_rq_bytes(req) >> ns->head->lba_shift) - 1);
935 
936 	if (!(req->cmd_flags & REQ_NOUNMAP) &&
937 	    (ns->head->features & NVME_NS_DEAC))
938 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_WZ_DEAC);
939 
940 	if (nvme_ns_has_pi(ns->head)) {
941 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_RW_PRINFO_PRACT);
942 
943 		switch (ns->head->pi_type) {
944 		case NVME_NS_DPS_PI_TYPE1:
945 		case NVME_NS_DPS_PI_TYPE2:
946 			nvme_set_ref_tag(ns, cmnd, req);
947 			break;
948 		}
949 	}
950 
951 	return BLK_STS_OK;
952 }
953 
954 /*
955  * NVMe does not support a dedicated command to issue an atomic write. A write
956  * which does adhere to the device atomic limits will silently be executed
957  * non-atomically. The request issuer should ensure that the write is within
958  * the queue atomic writes limits, but just validate this in case it is not.
959  */
960 static bool nvme_valid_atomic_write(struct request *req)
961 {
962 	struct request_queue *q = req->q;
963 	u32 boundary_bytes = queue_atomic_write_boundary_bytes(q);
964 
965 	if (blk_rq_bytes(req) > queue_atomic_write_unit_max_bytes(q))
966 		return false;
967 
968 	if (boundary_bytes) {
969 		u64 mask = boundary_bytes - 1, imask = ~mask;
970 		u64 start = blk_rq_pos(req) << SECTOR_SHIFT;
971 		u64 end = start + blk_rq_bytes(req) - 1;
972 
973 		/* If greater then must be crossing a boundary */
974 		if (blk_rq_bytes(req) > boundary_bytes)
975 			return false;
976 
977 		if ((start & imask) != (end & imask))
978 			return false;
979 	}
980 
981 	return true;
982 }
983 
984 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
985 		struct request *req, struct nvme_command *cmnd,
986 		enum nvme_opcode op)
987 {
988 	u16 control = 0;
989 	u32 dsmgmt = 0;
990 
991 	if (req->cmd_flags & REQ_FUA)
992 		control |= NVME_RW_FUA;
993 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
994 		control |= NVME_RW_LR;
995 
996 	if (req->cmd_flags & REQ_RAHEAD)
997 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
998 
999 	if (op == nvme_cmd_write && ns->head->nr_plids) {
1000 		u16 write_stream = req->bio->bi_write_stream;
1001 
1002 		if (WARN_ON_ONCE(write_stream > ns->head->nr_plids))
1003 			return BLK_STS_INVAL;
1004 
1005 		if (write_stream) {
1006 			dsmgmt |= ns->head->plids[write_stream - 1] << 16;
1007 			control |= NVME_RW_DTYPE_DPLCMT;
1008 		}
1009 	}
1010 
1011 	if (req->cmd_flags & REQ_ATOMIC && !nvme_valid_atomic_write(req))
1012 		return BLK_STS_INVAL;
1013 
1014 	cmnd->rw.opcode = op;
1015 	cmnd->rw.flags = 0;
1016 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
1017 	cmnd->rw.cdw2 = 0;
1018 	cmnd->rw.cdw3 = 0;
1019 	cmnd->rw.metadata = 0;
1020 	cmnd->rw.slba =
1021 		cpu_to_le64(nvme_sect_to_lba(ns->head, blk_rq_pos(req)));
1022 	cmnd->rw.length =
1023 		cpu_to_le16((blk_rq_bytes(req) >> ns->head->lba_shift) - 1);
1024 	cmnd->rw.reftag = 0;
1025 	cmnd->rw.lbat = 0;
1026 	cmnd->rw.lbatm = 0;
1027 
1028 	if (ns->head->ms) {
1029 		/*
1030 		 * If formated with metadata, the block layer always provides a
1031 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
1032 		 * we enable the PRACT bit for protection information or set the
1033 		 * namespace capacity to zero to prevent any I/O.
1034 		 */
1035 		if (!blk_integrity_rq(req)) {
1036 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns->head)))
1037 				return BLK_STS_NOTSUPP;
1038 			control |= NVME_RW_PRINFO_PRACT;
1039 		}
1040 
1041 		if (bio_integrity_flagged(req->bio, BIP_CHECK_GUARD))
1042 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
1043 		if (bio_integrity_flagged(req->bio, BIP_CHECK_REFTAG)) {
1044 			control |= NVME_RW_PRINFO_PRCHK_REF;
1045 			if (op == nvme_cmd_zone_append)
1046 				control |= NVME_RW_APPEND_PIREMAP;
1047 			nvme_set_ref_tag(ns, cmnd, req);
1048 		}
1049 		if (bio_integrity_flagged(req->bio, BIP_CHECK_APPTAG)) {
1050 			control |= NVME_RW_PRINFO_PRCHK_APP;
1051 			nvme_set_app_tag(req, cmnd);
1052 		}
1053 	}
1054 
1055 	cmnd->rw.control = cpu_to_le16(control);
1056 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
1057 	return 0;
1058 }
1059 
1060 void nvme_cleanup_cmd(struct request *req)
1061 {
1062 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
1063 		struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
1064 
1065 		if (req->special_vec.bv_page == ctrl->discard_page)
1066 			clear_bit_unlock(0, &ctrl->discard_page_busy);
1067 		else
1068 			kfree(bvec_virt(&req->special_vec));
1069 		req->rq_flags &= ~RQF_SPECIAL_PAYLOAD;
1070 	}
1071 }
1072 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
1073 
1074 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
1075 {
1076 	struct nvme_command *cmd = nvme_req(req)->cmd;
1077 	blk_status_t ret = BLK_STS_OK;
1078 
1079 	if (!(req->rq_flags & RQF_DONTPREP))
1080 		nvme_clear_nvme_request(req);
1081 
1082 	switch (req_op(req)) {
1083 	case REQ_OP_DRV_IN:
1084 	case REQ_OP_DRV_OUT:
1085 		/* these are setup prior to execution in nvme_init_request() */
1086 		break;
1087 	case REQ_OP_FLUSH:
1088 		nvme_setup_flush(ns, cmd);
1089 		break;
1090 	case REQ_OP_ZONE_RESET_ALL:
1091 	case REQ_OP_ZONE_RESET:
1092 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
1093 		break;
1094 	case REQ_OP_ZONE_OPEN:
1095 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
1096 		break;
1097 	case REQ_OP_ZONE_CLOSE:
1098 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
1099 		break;
1100 	case REQ_OP_ZONE_FINISH:
1101 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
1102 		break;
1103 	case REQ_OP_WRITE_ZEROES:
1104 		ret = nvme_setup_write_zeroes(ns, req, cmd);
1105 		break;
1106 	case REQ_OP_DISCARD:
1107 		ret = nvme_setup_discard(ns, req, cmd);
1108 		break;
1109 	case REQ_OP_READ:
1110 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1111 		break;
1112 	case REQ_OP_WRITE:
1113 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1114 		break;
1115 	case REQ_OP_ZONE_APPEND:
1116 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1117 		break;
1118 	default:
1119 		WARN_ON_ONCE(1);
1120 		return BLK_STS_IOERR;
1121 	}
1122 
1123 	cmd->common.command_id = nvme_cid(req);
1124 	trace_nvme_setup_cmd(req, cmd);
1125 	return ret;
1126 }
1127 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1128 
1129 /*
1130  * Return values:
1131  * 0:  success
1132  * >0: nvme controller's cqe status response
1133  * <0: kernel error in lieu of controller response
1134  */
1135 int nvme_execute_rq(struct request *rq, bool at_head)
1136 {
1137 	blk_status_t status;
1138 
1139 	status = blk_execute_rq(rq, at_head);
1140 	if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1141 		return -EINTR;
1142 	if (nvme_req(rq)->status)
1143 		return nvme_req(rq)->status;
1144 	return blk_status_to_errno(status);
1145 }
1146 EXPORT_SYMBOL_NS_GPL(nvme_execute_rq, "NVME_TARGET_PASSTHRU");
1147 
1148 /*
1149  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1150  * if the result is positive, it's an NVM Express status code
1151  */
1152 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1153 		union nvme_result *result, void *buffer, unsigned bufflen,
1154 		int qid, nvme_submit_flags_t flags)
1155 {
1156 	struct request *req;
1157 	int ret;
1158 	blk_mq_req_flags_t blk_flags = 0;
1159 
1160 	if (flags & NVME_SUBMIT_NOWAIT)
1161 		blk_flags |= BLK_MQ_REQ_NOWAIT;
1162 	if (flags & NVME_SUBMIT_RESERVED)
1163 		blk_flags |= BLK_MQ_REQ_RESERVED;
1164 	if (qid == NVME_QID_ANY)
1165 		req = blk_mq_alloc_request(q, nvme_req_op(cmd), blk_flags);
1166 	else
1167 		req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), blk_flags,
1168 						qid - 1);
1169 
1170 	if (IS_ERR(req))
1171 		return PTR_ERR(req);
1172 	nvme_init_request(req, cmd);
1173 	if (flags & NVME_SUBMIT_RETRY)
1174 		req->cmd_flags &= ~REQ_FAILFAST_DRIVER;
1175 
1176 	if (buffer && bufflen) {
1177 		ret = blk_rq_map_kern(req, buffer, bufflen, GFP_KERNEL);
1178 		if (ret)
1179 			goto out;
1180 	}
1181 
1182 	ret = nvme_execute_rq(req, flags & NVME_SUBMIT_AT_HEAD);
1183 	if (result && ret >= 0)
1184 		*result = nvme_req(req)->result;
1185  out:
1186 	blk_mq_free_request(req);
1187 	return ret;
1188 }
1189 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1190 
1191 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1192 		void *buffer, unsigned bufflen)
1193 {
1194 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1195 			NVME_QID_ANY, 0);
1196 }
1197 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1198 
1199 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1200 {
1201 	u32 effects = 0;
1202 
1203 	if (ns) {
1204 		effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1205 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1206 			dev_warn_once(ctrl->device,
1207 				"IO command:%02x has unusual effects:%08x\n",
1208 				opcode, effects);
1209 
1210 		/*
1211 		 * NVME_CMD_EFFECTS_CSE_MASK causes a freeze all I/O queues,
1212 		 * which would deadlock when done on an I/O command.  Note that
1213 		 * We already warn about an unusual effect above.
1214 		 */
1215 		effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1216 	} else {
1217 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1218 
1219 		/* Ignore execution restrictions if any relaxation bits are set */
1220 		if (effects & NVME_CMD_EFFECTS_CSER_MASK)
1221 			effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1222 	}
1223 
1224 	return effects;
1225 }
1226 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, "NVME_TARGET_PASSTHRU");
1227 
1228 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1229 {
1230 	u32 effects = nvme_command_effects(ctrl, ns, opcode);
1231 
1232 	/*
1233 	 * For simplicity, IO to all namespaces is quiesced even if the command
1234 	 * effects say only one namespace is affected.
1235 	 */
1236 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1237 		mutex_lock(&ctrl->scan_lock);
1238 		mutex_lock(&ctrl->subsys->lock);
1239 		nvme_mpath_start_freeze(ctrl->subsys);
1240 		nvme_mpath_wait_freeze(ctrl->subsys);
1241 		nvme_start_freeze(ctrl);
1242 		nvme_wait_freeze(ctrl);
1243 	}
1244 	return effects;
1245 }
1246 EXPORT_SYMBOL_NS_GPL(nvme_passthru_start, "NVME_TARGET_PASSTHRU");
1247 
1248 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1249 		       struct nvme_command *cmd, int status)
1250 {
1251 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1252 		nvme_unfreeze(ctrl);
1253 		nvme_mpath_unfreeze(ctrl->subsys);
1254 		mutex_unlock(&ctrl->subsys->lock);
1255 		mutex_unlock(&ctrl->scan_lock);
1256 	}
1257 	if (effects & NVME_CMD_EFFECTS_CCC) {
1258 		if (!test_and_set_bit(NVME_CTRL_DIRTY_CAPABILITY,
1259 				      &ctrl->flags)) {
1260 			dev_info(ctrl->device,
1261 "controller capabilities changed, reset may be required to take effect.\n");
1262 		}
1263 	}
1264 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1265 		nvme_queue_scan(ctrl);
1266 		flush_work(&ctrl->scan_work);
1267 	}
1268 	if (ns)
1269 		return;
1270 
1271 	switch (cmd->common.opcode) {
1272 	case nvme_admin_set_features:
1273 		switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1274 		case NVME_FEAT_KATO:
1275 			/*
1276 			 * Keep alive commands interval on the host should be
1277 			 * updated when KATO is modified by Set Features
1278 			 * commands.
1279 			 */
1280 			if (!status)
1281 				nvme_update_keep_alive(ctrl, cmd);
1282 			break;
1283 		default:
1284 			break;
1285 		}
1286 		break;
1287 	default:
1288 		break;
1289 	}
1290 }
1291 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, "NVME_TARGET_PASSTHRU");
1292 
1293 /*
1294  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1295  *
1296  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1297  *   accounting for transport roundtrip times [..].
1298  */
1299 static unsigned long nvme_keep_alive_work_period(struct nvme_ctrl *ctrl)
1300 {
1301 	unsigned long delay = ctrl->kato * HZ / 2;
1302 
1303 	/*
1304 	 * When using Traffic Based Keep Alive, we need to run
1305 	 * nvme_keep_alive_work at twice the normal frequency, as one
1306 	 * command completion can postpone sending a keep alive command
1307 	 * by up to twice the delay between runs.
1308 	 */
1309 	if (ctrl->ctratt & NVME_CTRL_ATTR_TBKAS)
1310 		delay /= 2;
1311 	return delay;
1312 }
1313 
1314 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1315 {
1316 	unsigned long now = jiffies;
1317 	unsigned long delay = nvme_keep_alive_work_period(ctrl);
1318 	unsigned long ka_next_check_tm = ctrl->ka_last_check_time + delay;
1319 
1320 	if (time_after(now, ka_next_check_tm))
1321 		delay = 0;
1322 	else
1323 		delay = ka_next_check_tm - now;
1324 
1325 	queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1326 }
1327 
1328 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1329 						 blk_status_t status)
1330 {
1331 	struct nvme_ctrl *ctrl = rq->end_io_data;
1332 	unsigned long rtt = jiffies - (rq->deadline - rq->timeout);
1333 	unsigned long delay = nvme_keep_alive_work_period(ctrl);
1334 	enum nvme_ctrl_state state = nvme_ctrl_state(ctrl);
1335 
1336 	/*
1337 	 * Subtract off the keepalive RTT so nvme_keep_alive_work runs
1338 	 * at the desired frequency.
1339 	 */
1340 	if (rtt <= delay) {
1341 		delay -= rtt;
1342 	} else {
1343 		dev_warn(ctrl->device, "long keepalive RTT (%u ms)\n",
1344 			 jiffies_to_msecs(rtt));
1345 		delay = 0;
1346 	}
1347 
1348 	blk_mq_free_request(rq);
1349 
1350 	if (status) {
1351 		dev_err(ctrl->device,
1352 			"failed nvme_keep_alive_end_io error=%d\n",
1353 				status);
1354 		return RQ_END_IO_NONE;
1355 	}
1356 
1357 	ctrl->ka_last_check_time = jiffies;
1358 	ctrl->comp_seen = false;
1359 	if (state == NVME_CTRL_LIVE || state == NVME_CTRL_CONNECTING)
1360 		queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1361 	return RQ_END_IO_NONE;
1362 }
1363 
1364 static void nvme_keep_alive_work(struct work_struct *work)
1365 {
1366 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1367 			struct nvme_ctrl, ka_work);
1368 	bool comp_seen = ctrl->comp_seen;
1369 	struct request *rq;
1370 
1371 	ctrl->ka_last_check_time = jiffies;
1372 
1373 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1374 		dev_dbg(ctrl->device,
1375 			"reschedule traffic based keep-alive timer\n");
1376 		ctrl->comp_seen = false;
1377 		nvme_queue_keep_alive_work(ctrl);
1378 		return;
1379 	}
1380 
1381 	rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1382 				  BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1383 	if (IS_ERR(rq)) {
1384 		/* allocation failure, reset the controller */
1385 		dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1386 		nvme_reset_ctrl(ctrl);
1387 		return;
1388 	}
1389 	nvme_init_request(rq, &ctrl->ka_cmd);
1390 
1391 	rq->timeout = ctrl->kato * HZ;
1392 	rq->end_io = nvme_keep_alive_end_io;
1393 	rq->end_io_data = ctrl;
1394 	blk_execute_rq_nowait(rq, false);
1395 }
1396 
1397 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1398 {
1399 	if (unlikely(ctrl->kato == 0))
1400 		return;
1401 
1402 	nvme_queue_keep_alive_work(ctrl);
1403 }
1404 
1405 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1406 {
1407 	if (unlikely(ctrl->kato == 0))
1408 		return;
1409 
1410 	cancel_delayed_work_sync(&ctrl->ka_work);
1411 }
1412 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1413 
1414 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1415 				   struct nvme_command *cmd)
1416 {
1417 	unsigned int new_kato =
1418 		DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1419 
1420 	dev_info(ctrl->device,
1421 		 "keep alive interval updated from %u ms to %u ms\n",
1422 		 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1423 
1424 	nvme_stop_keep_alive(ctrl);
1425 	ctrl->kato = new_kato;
1426 	nvme_start_keep_alive(ctrl);
1427 }
1428 
1429 static bool nvme_id_cns_ok(struct nvme_ctrl *ctrl, u8 cns)
1430 {
1431 	/*
1432 	 * The CNS field occupies a full byte starting with NVMe 1.2
1433 	 */
1434 	if (ctrl->vs >= NVME_VS(1, 2, 0))
1435 		return true;
1436 
1437 	/*
1438 	 * NVMe 1.1 expanded the CNS value to two bits, which means values
1439 	 * larger than that could get truncated and treated as an incorrect
1440 	 * value.
1441 	 *
1442 	 * Qemu implemented 1.0 behavior for controllers claiming 1.1
1443 	 * compliance, so they need to be quirked here.
1444 	 */
1445 	if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1446 	    !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS))
1447 		return cns <= 3;
1448 
1449 	/*
1450 	 * NVMe 1.0 used a single bit for the CNS value.
1451 	 */
1452 	return cns <= 1;
1453 }
1454 
1455 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1456 {
1457 	struct nvme_command c = { };
1458 	int error;
1459 
1460 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1461 	c.identify.opcode = nvme_admin_identify;
1462 	c.identify.cns = NVME_ID_CNS_CTRL;
1463 
1464 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1465 	if (!*id)
1466 		return -ENOMEM;
1467 
1468 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1469 			sizeof(struct nvme_id_ctrl));
1470 	if (error) {
1471 		kfree(*id);
1472 		*id = NULL;
1473 	}
1474 	return error;
1475 }
1476 
1477 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1478 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1479 {
1480 	const char *warn_str = "ctrl returned bogus length:";
1481 	void *data = cur;
1482 
1483 	switch (cur->nidt) {
1484 	case NVME_NIDT_EUI64:
1485 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1486 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1487 				 warn_str, cur->nidl);
1488 			return -1;
1489 		}
1490 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1491 			return NVME_NIDT_EUI64_LEN;
1492 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1493 		return NVME_NIDT_EUI64_LEN;
1494 	case NVME_NIDT_NGUID:
1495 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1496 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1497 				 warn_str, cur->nidl);
1498 			return -1;
1499 		}
1500 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1501 			return NVME_NIDT_NGUID_LEN;
1502 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1503 		return NVME_NIDT_NGUID_LEN;
1504 	case NVME_NIDT_UUID:
1505 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1506 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1507 				 warn_str, cur->nidl);
1508 			return -1;
1509 		}
1510 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1511 			return NVME_NIDT_UUID_LEN;
1512 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1513 		return NVME_NIDT_UUID_LEN;
1514 	case NVME_NIDT_CSI:
1515 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1516 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1517 				 warn_str, cur->nidl);
1518 			return -1;
1519 		}
1520 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1521 		*csi_seen = true;
1522 		return NVME_NIDT_CSI_LEN;
1523 	default:
1524 		/* Skip unknown types */
1525 		return cur->nidl;
1526 	}
1527 }
1528 
1529 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1530 		struct nvme_ns_info *info)
1531 {
1532 	struct nvme_command c = { };
1533 	bool csi_seen = false;
1534 	int status, pos, len;
1535 	void *data;
1536 
1537 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1538 		return 0;
1539 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1540 		return 0;
1541 
1542 	c.identify.opcode = nvme_admin_identify;
1543 	c.identify.nsid = cpu_to_le32(info->nsid);
1544 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1545 
1546 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1547 	if (!data)
1548 		return -ENOMEM;
1549 
1550 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1551 				      NVME_IDENTIFY_DATA_SIZE);
1552 	if (status) {
1553 		dev_warn(ctrl->device,
1554 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1555 			info->nsid, status);
1556 		goto free_data;
1557 	}
1558 
1559 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1560 		struct nvme_ns_id_desc *cur = data + pos;
1561 
1562 		if (cur->nidl == 0)
1563 			break;
1564 
1565 		len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1566 		if (len < 0)
1567 			break;
1568 
1569 		len += sizeof(*cur);
1570 	}
1571 
1572 	if (nvme_multi_css(ctrl) && !csi_seen) {
1573 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1574 			 info->nsid);
1575 		status = -EINVAL;
1576 	}
1577 
1578 free_data:
1579 	kfree(data);
1580 	return status;
1581 }
1582 
1583 int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1584 			struct nvme_id_ns **id)
1585 {
1586 	struct nvme_command c = { };
1587 	int error;
1588 
1589 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1590 	c.identify.opcode = nvme_admin_identify;
1591 	c.identify.nsid = cpu_to_le32(nsid);
1592 	c.identify.cns = NVME_ID_CNS_NS;
1593 
1594 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1595 	if (!*id)
1596 		return -ENOMEM;
1597 
1598 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1599 	if (error) {
1600 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1601 		kfree(*id);
1602 		*id = NULL;
1603 	}
1604 	return error;
1605 }
1606 
1607 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1608 		struct nvme_ns_info *info)
1609 {
1610 	struct nvme_ns_ids *ids = &info->ids;
1611 	struct nvme_id_ns *id;
1612 	int ret;
1613 
1614 	ret = nvme_identify_ns(ctrl, info->nsid, &id);
1615 	if (ret)
1616 		return ret;
1617 
1618 	if (id->ncap == 0) {
1619 		/* namespace not allocated or attached */
1620 		info->is_removed = true;
1621 		ret = -ENODEV;
1622 		goto error;
1623 	}
1624 
1625 	info->anagrpid = id->anagrpid;
1626 	info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1627 	info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1628 	info->is_ready = true;
1629 	info->endgid = le16_to_cpu(id->endgid);
1630 	if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1631 		dev_info(ctrl->device,
1632 			 "Ignoring bogus Namespace Identifiers\n");
1633 	} else {
1634 		if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1635 		    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1636 			memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1637 		if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1638 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1639 			memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1640 	}
1641 
1642 error:
1643 	kfree(id);
1644 	return ret;
1645 }
1646 
1647 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1648 		struct nvme_ns_info *info)
1649 {
1650 	struct nvme_id_ns_cs_indep *id;
1651 	struct nvme_command c = {
1652 		.identify.opcode	= nvme_admin_identify,
1653 		.identify.nsid		= cpu_to_le32(info->nsid),
1654 		.identify.cns		= NVME_ID_CNS_NS_CS_INDEP,
1655 	};
1656 	int ret;
1657 
1658 	id = kmalloc(sizeof(*id), GFP_KERNEL);
1659 	if (!id)
1660 		return -ENOMEM;
1661 
1662 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1663 	if (!ret) {
1664 		info->anagrpid = id->anagrpid;
1665 		info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1666 		info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1667 		info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1668 		info->is_rotational = id->nsfeat & NVME_NS_ROTATIONAL;
1669 		info->no_vwc = id->nsfeat & NVME_NS_VWC_NOT_PRESENT;
1670 		info->endgid = le16_to_cpu(id->endgid);
1671 	}
1672 	kfree(id);
1673 	return ret;
1674 }
1675 
1676 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1677 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1678 {
1679 	union nvme_result res = { 0 };
1680 	struct nvme_command c = { };
1681 	int ret;
1682 
1683 	c.features.opcode = op;
1684 	c.features.fid = cpu_to_le32(fid);
1685 	c.features.dword11 = cpu_to_le32(dword11);
1686 
1687 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1688 			buffer, buflen, NVME_QID_ANY, 0);
1689 	if (ret >= 0 && result)
1690 		*result = le32_to_cpu(res.u32);
1691 	return ret;
1692 }
1693 
1694 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1695 		      unsigned int dword11, void *buffer, size_t buflen,
1696 		      void *result)
1697 {
1698 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1699 			     buflen, result);
1700 }
1701 EXPORT_SYMBOL_GPL(nvme_set_features);
1702 
1703 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1704 		      unsigned int dword11, void *buffer, size_t buflen,
1705 		      void *result)
1706 {
1707 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1708 			     buflen, result);
1709 }
1710 EXPORT_SYMBOL_GPL(nvme_get_features);
1711 
1712 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1713 {
1714 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1715 	u32 result;
1716 	int status, nr_io_queues;
1717 
1718 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1719 			&result);
1720 
1721 	/*
1722 	 * It's either a kernel error or the host observed a connection
1723 	 * lost. In either case it's not possible communicate with the
1724 	 * controller and thus enter the error code path.
1725 	 */
1726 	if (status < 0 || status == NVME_SC_HOST_PATH_ERROR)
1727 		return status;
1728 
1729 	/*
1730 	 * Degraded controllers might return an error when setting the queue
1731 	 * count.  We still want to be able to bring them online and offer
1732 	 * access to the admin queue, as that might be only way to fix them up.
1733 	 */
1734 	if (status > 0) {
1735 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1736 		*count = 0;
1737 	} else {
1738 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1739 		*count = min(*count, nr_io_queues);
1740 	}
1741 
1742 	return 0;
1743 }
1744 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1745 
1746 #define NVME_AEN_SUPPORTED \
1747 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1748 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1749 
1750 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1751 {
1752 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1753 	int status;
1754 
1755 	if (!supported_aens)
1756 		return;
1757 
1758 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1759 			NULL, 0, &result);
1760 	if (status)
1761 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1762 			 supported_aens);
1763 
1764 	queue_work(nvme_wq, &ctrl->async_event_work);
1765 }
1766 
1767 static int nvme_ns_open(struct nvme_ns *ns)
1768 {
1769 
1770 	/* should never be called due to GENHD_FL_HIDDEN */
1771 	if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1772 		goto fail;
1773 	if (!nvme_get_ns(ns))
1774 		goto fail;
1775 	if (!try_module_get(ns->ctrl->ops->module))
1776 		goto fail_put_ns;
1777 
1778 	return 0;
1779 
1780 fail_put_ns:
1781 	nvme_put_ns(ns);
1782 fail:
1783 	return -ENXIO;
1784 }
1785 
1786 static void nvme_ns_release(struct nvme_ns *ns)
1787 {
1788 
1789 	module_put(ns->ctrl->ops->module);
1790 	nvme_put_ns(ns);
1791 }
1792 
1793 static int nvme_open(struct gendisk *disk, blk_mode_t mode)
1794 {
1795 	return nvme_ns_open(disk->private_data);
1796 }
1797 
1798 static void nvme_release(struct gendisk *disk)
1799 {
1800 	nvme_ns_release(disk->private_data);
1801 }
1802 
1803 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1804 {
1805 	/* some standard values */
1806 	geo->heads = 1 << 6;
1807 	geo->sectors = 1 << 5;
1808 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1809 	return 0;
1810 }
1811 
1812 static bool nvme_init_integrity(struct nvme_ns_head *head,
1813 		struct queue_limits *lim, struct nvme_ns_info *info)
1814 {
1815 	struct blk_integrity *bi = &lim->integrity;
1816 
1817 	memset(bi, 0, sizeof(*bi));
1818 
1819 	if (!head->ms)
1820 		return true;
1821 
1822 	/*
1823 	 * PI can always be supported as we can ask the controller to simply
1824 	 * insert/strip it, which is not possible for other kinds of metadata.
1825 	 */
1826 	if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) ||
1827 	    !(head->features & NVME_NS_METADATA_SUPPORTED))
1828 		return nvme_ns_has_pi(head);
1829 
1830 	switch (head->pi_type) {
1831 	case NVME_NS_DPS_PI_TYPE3:
1832 		switch (head->guard_type) {
1833 		case NVME_NVM_NS_16B_GUARD:
1834 			bi->csum_type = BLK_INTEGRITY_CSUM_CRC;
1835 			bi->tag_size = sizeof(u16) + sizeof(u32);
1836 			bi->flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1837 			break;
1838 		case NVME_NVM_NS_64B_GUARD:
1839 			bi->csum_type = BLK_INTEGRITY_CSUM_CRC64;
1840 			bi->tag_size = sizeof(u16) + 6;
1841 			bi->flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1842 			break;
1843 		default:
1844 			break;
1845 		}
1846 		break;
1847 	case NVME_NS_DPS_PI_TYPE1:
1848 	case NVME_NS_DPS_PI_TYPE2:
1849 		switch (head->guard_type) {
1850 		case NVME_NVM_NS_16B_GUARD:
1851 			bi->csum_type = BLK_INTEGRITY_CSUM_CRC;
1852 			bi->tag_size = sizeof(u16);
1853 			bi->flags |= BLK_INTEGRITY_DEVICE_CAPABLE |
1854 				     BLK_INTEGRITY_REF_TAG;
1855 			break;
1856 		case NVME_NVM_NS_64B_GUARD:
1857 			bi->csum_type = BLK_INTEGRITY_CSUM_CRC64;
1858 			bi->tag_size = sizeof(u16);
1859 			bi->flags |= BLK_INTEGRITY_DEVICE_CAPABLE |
1860 				     BLK_INTEGRITY_REF_TAG;
1861 			break;
1862 		default:
1863 			break;
1864 		}
1865 		break;
1866 	default:
1867 		break;
1868 	}
1869 
1870 	bi->tuple_size = head->ms;
1871 	bi->pi_offset = info->pi_offset;
1872 	return true;
1873 }
1874 
1875 static void nvme_config_discard(struct nvme_ns *ns, struct queue_limits *lim)
1876 {
1877 	struct nvme_ctrl *ctrl = ns->ctrl;
1878 
1879 	if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns->head, UINT_MAX))
1880 		lim->max_hw_discard_sectors =
1881 			nvme_lba_to_sect(ns->head, ctrl->dmrsl);
1882 	else if (ctrl->oncs & NVME_CTRL_ONCS_DSM)
1883 		lim->max_hw_discard_sectors = UINT_MAX;
1884 	else
1885 		lim->max_hw_discard_sectors = 0;
1886 
1887 	lim->discard_granularity = lim->logical_block_size;
1888 
1889 	if (ctrl->dmrl)
1890 		lim->max_discard_segments = ctrl->dmrl;
1891 	else
1892 		lim->max_discard_segments = NVME_DSM_MAX_RANGES;
1893 }
1894 
1895 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1896 {
1897 	return uuid_equal(&a->uuid, &b->uuid) &&
1898 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1899 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1900 		a->csi == b->csi;
1901 }
1902 
1903 static int nvme_identify_ns_nvm(struct nvme_ctrl *ctrl, unsigned int nsid,
1904 		struct nvme_id_ns_nvm **nvmp)
1905 {
1906 	struct nvme_command c = {
1907 		.identify.opcode	= nvme_admin_identify,
1908 		.identify.nsid		= cpu_to_le32(nsid),
1909 		.identify.cns		= NVME_ID_CNS_CS_NS,
1910 		.identify.csi		= NVME_CSI_NVM,
1911 	};
1912 	struct nvme_id_ns_nvm *nvm;
1913 	int ret;
1914 
1915 	nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1916 	if (!nvm)
1917 		return -ENOMEM;
1918 
1919 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, nvm, sizeof(*nvm));
1920 	if (ret)
1921 		kfree(nvm);
1922 	else
1923 		*nvmp = nvm;
1924 	return ret;
1925 }
1926 
1927 static void nvme_configure_pi_elbas(struct nvme_ns_head *head,
1928 		struct nvme_id_ns *id, struct nvme_id_ns_nvm *nvm)
1929 {
1930 	u32 elbaf = le32_to_cpu(nvm->elbaf[nvme_lbaf_index(id->flbas)]);
1931 	u8 guard_type;
1932 
1933 	/* no support for storage tag formats right now */
1934 	if (nvme_elbaf_sts(elbaf))
1935 		return;
1936 
1937 	guard_type = nvme_elbaf_guard_type(elbaf);
1938 	if ((nvm->pic & NVME_ID_NS_NVM_QPIFS) &&
1939 	     guard_type == NVME_NVM_NS_QTYPE_GUARD)
1940 		guard_type = nvme_elbaf_qualified_guard_type(elbaf);
1941 
1942 	head->guard_type = guard_type;
1943 	switch (head->guard_type) {
1944 	case NVME_NVM_NS_64B_GUARD:
1945 		head->pi_size = sizeof(struct crc64_pi_tuple);
1946 		break;
1947 	case NVME_NVM_NS_16B_GUARD:
1948 		head->pi_size = sizeof(struct t10_pi_tuple);
1949 		break;
1950 	default:
1951 		break;
1952 	}
1953 }
1954 
1955 static void nvme_configure_metadata(struct nvme_ctrl *ctrl,
1956 		struct nvme_ns_head *head, struct nvme_id_ns *id,
1957 		struct nvme_id_ns_nvm *nvm, struct nvme_ns_info *info)
1958 {
1959 	head->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1960 	head->pi_type = 0;
1961 	head->pi_size = 0;
1962 	head->ms = le16_to_cpu(id->lbaf[nvme_lbaf_index(id->flbas)].ms);
1963 	if (!head->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1964 		return;
1965 
1966 	if (nvm && (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1967 		nvme_configure_pi_elbas(head, id, nvm);
1968 	} else {
1969 		head->pi_size = sizeof(struct t10_pi_tuple);
1970 		head->guard_type = NVME_NVM_NS_16B_GUARD;
1971 	}
1972 
1973 	if (head->pi_size && head->ms >= head->pi_size)
1974 		head->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1975 	if (!(id->dps & NVME_NS_DPS_PI_FIRST)) {
1976 		if (disable_pi_offsets)
1977 			head->pi_type = 0;
1978 		else
1979 			info->pi_offset = head->ms - head->pi_size;
1980 	}
1981 
1982 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1983 		/*
1984 		 * The NVMe over Fabrics specification only supports metadata as
1985 		 * part of the extended data LBA.  We rely on HCA/HBA support to
1986 		 * remap the separate metadata buffer from the block layer.
1987 		 */
1988 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1989 			return;
1990 
1991 		head->features |= NVME_NS_EXT_LBAS;
1992 
1993 		/*
1994 		 * The current fabrics transport drivers support namespace
1995 		 * metadata formats only if nvme_ns_has_pi() returns true.
1996 		 * Suppress support for all other formats so the namespace will
1997 		 * have a 0 capacity and not be usable through the block stack.
1998 		 *
1999 		 * Note, this check will need to be modified if any drivers
2000 		 * gain the ability to use other metadata formats.
2001 		 */
2002 		if (ctrl->max_integrity_segments && nvme_ns_has_pi(head))
2003 			head->features |= NVME_NS_METADATA_SUPPORTED;
2004 	} else {
2005 		/*
2006 		 * For PCIe controllers, we can't easily remap the separate
2007 		 * metadata buffer from the block layer and thus require a
2008 		 * separate metadata buffer for block layer metadata/PI support.
2009 		 * We allow extended LBAs for the passthrough interface, though.
2010 		 */
2011 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
2012 			head->features |= NVME_NS_EXT_LBAS;
2013 		else
2014 			head->features |= NVME_NS_METADATA_SUPPORTED;
2015 	}
2016 }
2017 
2018 
2019 static void nvme_update_atomic_write_disk_info(struct nvme_ns *ns,
2020 			struct nvme_id_ns *id, struct queue_limits *lim,
2021 			u32 bs, u32 atomic_bs)
2022 {
2023 	unsigned int boundary = 0;
2024 
2025 	if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf) {
2026 		if (le16_to_cpu(id->nabspf))
2027 			boundary = (le16_to_cpu(id->nabspf) + 1) * bs;
2028 	}
2029 	lim->atomic_write_hw_max = atomic_bs;
2030 	lim->atomic_write_hw_boundary = boundary;
2031 	lim->atomic_write_hw_unit_min = bs;
2032 	lim->atomic_write_hw_unit_max = rounddown_pow_of_two(atomic_bs);
2033 	lim->features |= BLK_FEAT_ATOMIC_WRITES;
2034 }
2035 
2036 static u32 nvme_max_drv_segments(struct nvme_ctrl *ctrl)
2037 {
2038 	return ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> SECTOR_SHIFT) + 1;
2039 }
2040 
2041 static void nvme_set_ctrl_limits(struct nvme_ctrl *ctrl,
2042 		struct queue_limits *lim)
2043 {
2044 	lim->max_hw_sectors = ctrl->max_hw_sectors;
2045 	lim->max_segments = min_t(u32, USHRT_MAX,
2046 		min_not_zero(nvme_max_drv_segments(ctrl), ctrl->max_segments));
2047 	lim->max_integrity_segments = ctrl->max_integrity_segments;
2048 	lim->virt_boundary_mask = NVME_CTRL_PAGE_SIZE - 1;
2049 	lim->max_segment_size = UINT_MAX;
2050 	lim->dma_alignment = 3;
2051 }
2052 
2053 static bool nvme_update_disk_info(struct nvme_ns *ns, struct nvme_id_ns *id,
2054 		struct queue_limits *lim)
2055 {
2056 	struct nvme_ns_head *head = ns->head;
2057 	u32 bs = 1U << head->lba_shift;
2058 	u32 atomic_bs, phys_bs, io_opt = 0;
2059 	bool valid = true;
2060 
2061 	/*
2062 	 * The block layer can't support LBA sizes larger than the page size
2063 	 * or smaller than a sector size yet, so catch this early and don't
2064 	 * allow block I/O.
2065 	 */
2066 	if (blk_validate_block_size(bs)) {
2067 		bs = (1 << 9);
2068 		valid = false;
2069 	}
2070 
2071 	atomic_bs = phys_bs = bs;
2072 	if (id->nabo == 0) {
2073 		/*
2074 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
2075 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
2076 		 * 0 then AWUPF must be used instead.
2077 		 */
2078 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2079 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2080 		else
2081 			atomic_bs = (1 + ns->ctrl->awupf) * bs;
2082 
2083 		/*
2084 		 * Set subsystem atomic bs.
2085 		 */
2086 		if (ns->ctrl->subsys->atomic_bs) {
2087 			if (atomic_bs != ns->ctrl->subsys->atomic_bs) {
2088 				dev_err_ratelimited(ns->ctrl->device,
2089 					"%s: Inconsistent Atomic Write Size, Namespace will not be added: Subsystem=%d bytes, Controller/Namespace=%d bytes\n",
2090 					ns->disk ? ns->disk->disk_name : "?",
2091 					ns->ctrl->subsys->atomic_bs,
2092 					atomic_bs);
2093 			}
2094 		} else
2095 			ns->ctrl->subsys->atomic_bs = atomic_bs;
2096 
2097 		nvme_update_atomic_write_disk_info(ns, id, lim, bs, atomic_bs);
2098 	}
2099 
2100 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2101 		/* NPWG = Namespace Preferred Write Granularity */
2102 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2103 		/* NOWS = Namespace Optimal Write Size */
2104 		if (id->nows)
2105 			io_opt = bs * (1 + le16_to_cpu(id->nows));
2106 	}
2107 
2108 	/*
2109 	 * Linux filesystems assume writing a single physical block is
2110 	 * an atomic operation. Hence limit the physical block size to the
2111 	 * value of the Atomic Write Unit Power Fail parameter.
2112 	 */
2113 	lim->logical_block_size = bs;
2114 	lim->physical_block_size = min(phys_bs, atomic_bs);
2115 	lim->io_min = phys_bs;
2116 	lim->io_opt = io_opt;
2117 	if ((ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES) &&
2118 	    (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM))
2119 		lim->max_write_zeroes_sectors = UINT_MAX;
2120 	else
2121 		lim->max_write_zeroes_sectors = ns->ctrl->max_zeroes_sectors;
2122 	return valid;
2123 }
2124 
2125 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
2126 {
2127 	return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
2128 }
2129 
2130 static inline bool nvme_first_scan(struct gendisk *disk)
2131 {
2132 	/* nvme_alloc_ns() scans the disk prior to adding it */
2133 	return !disk_live(disk);
2134 }
2135 
2136 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id,
2137 		struct queue_limits *lim)
2138 {
2139 	struct nvme_ctrl *ctrl = ns->ctrl;
2140 	u32 iob;
2141 
2142 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2143 	    is_power_of_2(ctrl->max_hw_sectors))
2144 		iob = ctrl->max_hw_sectors;
2145 	else
2146 		iob = nvme_lba_to_sect(ns->head, le16_to_cpu(id->noiob));
2147 
2148 	if (!iob)
2149 		return;
2150 
2151 	if (!is_power_of_2(iob)) {
2152 		if (nvme_first_scan(ns->disk))
2153 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2154 				ns->disk->disk_name, iob);
2155 		return;
2156 	}
2157 
2158 	if (blk_queue_is_zoned(ns->disk->queue)) {
2159 		if (nvme_first_scan(ns->disk))
2160 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
2161 				ns->disk->disk_name);
2162 		return;
2163 	}
2164 
2165 	lim->chunk_sectors = iob;
2166 }
2167 
2168 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
2169 		struct nvme_ns_info *info)
2170 {
2171 	struct queue_limits lim;
2172 	unsigned int memflags;
2173 	int ret;
2174 
2175 	lim = queue_limits_start_update(ns->disk->queue);
2176 	nvme_set_ctrl_limits(ns->ctrl, &lim);
2177 
2178 	memflags = blk_mq_freeze_queue(ns->disk->queue);
2179 	ret = queue_limits_commit_update(ns->disk->queue, &lim);
2180 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2181 	blk_mq_unfreeze_queue(ns->disk->queue, memflags);
2182 
2183 	/* Hide the block-interface for these devices */
2184 	if (!ret)
2185 		ret = -ENODEV;
2186 	return ret;
2187 }
2188 
2189 static int nvme_query_fdp_granularity(struct nvme_ctrl *ctrl,
2190 				      struct nvme_ns_info *info, u8 fdp_idx)
2191 {
2192 	struct nvme_fdp_config_log hdr, *h;
2193 	struct nvme_fdp_config_desc *desc;
2194 	size_t size = sizeof(hdr);
2195 	void *log, *end;
2196 	int i, n, ret;
2197 
2198 	ret = nvme_get_log_lsi(ctrl, 0, NVME_LOG_FDP_CONFIGS, 0,
2199 			       NVME_CSI_NVM, &hdr, size, 0, info->endgid);
2200 	if (ret) {
2201 		dev_warn(ctrl->device,
2202 			 "FDP configs log header status:0x%x endgid:%d\n", ret,
2203 			 info->endgid);
2204 		return ret;
2205 	}
2206 
2207 	size = le32_to_cpu(hdr.sze);
2208 	if (size > PAGE_SIZE * MAX_ORDER_NR_PAGES) {
2209 		dev_warn(ctrl->device, "FDP config size too large:%zu\n",
2210 			 size);
2211 		return 0;
2212 	}
2213 
2214 	h = kvmalloc(size, GFP_KERNEL);
2215 	if (!h)
2216 		return -ENOMEM;
2217 
2218 	ret = nvme_get_log_lsi(ctrl, 0, NVME_LOG_FDP_CONFIGS, 0,
2219 			       NVME_CSI_NVM, h, size, 0, info->endgid);
2220 	if (ret) {
2221 		dev_warn(ctrl->device,
2222 			 "FDP configs log status:0x%x endgid:%d\n", ret,
2223 			 info->endgid);
2224 		goto out;
2225 	}
2226 
2227 	n = le16_to_cpu(h->numfdpc) + 1;
2228 	if (fdp_idx > n) {
2229 		dev_warn(ctrl->device, "FDP index:%d out of range:%d\n",
2230 			 fdp_idx, n);
2231 		/* Proceed without registering FDP streams */
2232 		ret = 0;
2233 		goto out;
2234 	}
2235 
2236 	log = h + 1;
2237 	desc = log;
2238 	end = log + size - sizeof(*h);
2239 	for (i = 0; i < fdp_idx; i++) {
2240 		log += le16_to_cpu(desc->dsze);
2241 		desc = log;
2242 		if (log >= end) {
2243 			dev_warn(ctrl->device,
2244 				 "FDP invalid config descriptor list\n");
2245 			ret = 0;
2246 			goto out;
2247 		}
2248 	}
2249 
2250 	if (le32_to_cpu(desc->nrg) > 1) {
2251 		dev_warn(ctrl->device, "FDP NRG > 1 not supported\n");
2252 		ret = 0;
2253 		goto out;
2254 	}
2255 
2256 	info->runs = le64_to_cpu(desc->runs);
2257 out:
2258 	kvfree(h);
2259 	return ret;
2260 }
2261 
2262 static int nvme_query_fdp_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2263 {
2264 	struct nvme_ns_head *head = ns->head;
2265 	struct nvme_ctrl *ctrl = ns->ctrl;
2266 	struct nvme_fdp_ruh_status *ruhs;
2267 	struct nvme_fdp_config fdp;
2268 	struct nvme_command c = {};
2269 	size_t size;
2270 	int i, ret;
2271 
2272 	/*
2273 	 * The FDP configuration is static for the lifetime of the namespace,
2274 	 * so return immediately if we've already registered this namespace's
2275 	 * streams.
2276 	 */
2277 	if (head->nr_plids)
2278 		return 0;
2279 
2280 	ret = nvme_get_features(ctrl, NVME_FEAT_FDP, info->endgid, NULL, 0,
2281 				&fdp);
2282 	if (ret) {
2283 		dev_warn(ctrl->device, "FDP get feature status:0x%x\n", ret);
2284 		return ret;
2285 	}
2286 
2287 	if (!(fdp.flags & FDPCFG_FDPE))
2288 		return 0;
2289 
2290 	ret = nvme_query_fdp_granularity(ctrl, info, fdp.fdpcidx);
2291 	if (!info->runs)
2292 		return ret;
2293 
2294 	size = struct_size(ruhs, ruhsd, S8_MAX - 1);
2295 	ruhs = kzalloc(size, GFP_KERNEL);
2296 	if (!ruhs)
2297 		return -ENOMEM;
2298 
2299 	c.imr.opcode = nvme_cmd_io_mgmt_recv;
2300 	c.imr.nsid = cpu_to_le32(head->ns_id);
2301 	c.imr.mo = NVME_IO_MGMT_RECV_MO_RUHS;
2302 	c.imr.numd = cpu_to_le32(nvme_bytes_to_numd(size));
2303 	ret = nvme_submit_sync_cmd(ns->queue, &c, ruhs, size);
2304 	if (ret) {
2305 		dev_warn(ctrl->device, "FDP io-mgmt status:0x%x\n", ret);
2306 		goto free;
2307 	}
2308 
2309 	head->nr_plids = le16_to_cpu(ruhs->nruhsd);
2310 	if (!head->nr_plids)
2311 		goto free;
2312 
2313 	head->plids = kcalloc(head->nr_plids, sizeof(*head->plids),
2314 			      GFP_KERNEL);
2315 	if (!head->plids) {
2316 		dev_warn(ctrl->device,
2317 			 "failed to allocate %u FDP placement IDs\n",
2318 			 head->nr_plids);
2319 		head->nr_plids = 0;
2320 		ret = -ENOMEM;
2321 		goto free;
2322 	}
2323 
2324 	for (i = 0; i < head->nr_plids; i++)
2325 		head->plids[i] = le16_to_cpu(ruhs->ruhsd[i].pid);
2326 free:
2327 	kfree(ruhs);
2328 	return ret;
2329 }
2330 
2331 static int nvme_update_ns_info_block(struct nvme_ns *ns,
2332 		struct nvme_ns_info *info)
2333 {
2334 	struct queue_limits lim;
2335 	struct nvme_id_ns_nvm *nvm = NULL;
2336 	struct nvme_zone_info zi = {};
2337 	struct nvme_id_ns *id;
2338 	unsigned int memflags;
2339 	sector_t capacity;
2340 	unsigned lbaf;
2341 	int ret;
2342 
2343 	ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
2344 	if (ret)
2345 		return ret;
2346 
2347 	if (id->ncap == 0) {
2348 		/* namespace not allocated or attached */
2349 		info->is_removed = true;
2350 		ret = -ENXIO;
2351 		goto out;
2352 	}
2353 	lbaf = nvme_lbaf_index(id->flbas);
2354 
2355 	if (ns->ctrl->ctratt & NVME_CTRL_ATTR_ELBAS) {
2356 		ret = nvme_identify_ns_nvm(ns->ctrl, info->nsid, &nvm);
2357 		if (ret < 0)
2358 			goto out;
2359 	}
2360 
2361 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
2362 	    ns->head->ids.csi == NVME_CSI_ZNS) {
2363 		ret = nvme_query_zone_info(ns, lbaf, &zi);
2364 		if (ret < 0)
2365 			goto out;
2366 	}
2367 
2368 	if (ns->ctrl->ctratt & NVME_CTRL_ATTR_FDPS) {
2369 		ret = nvme_query_fdp_info(ns, info);
2370 		if (ret < 0)
2371 			goto out;
2372 	}
2373 
2374 	lim = queue_limits_start_update(ns->disk->queue);
2375 
2376 	memflags = blk_mq_freeze_queue(ns->disk->queue);
2377 	ns->head->lba_shift = id->lbaf[lbaf].ds;
2378 	ns->head->nuse = le64_to_cpu(id->nuse);
2379 	capacity = nvme_lba_to_sect(ns->head, le64_to_cpu(id->nsze));
2380 	nvme_set_ctrl_limits(ns->ctrl, &lim);
2381 	nvme_configure_metadata(ns->ctrl, ns->head, id, nvm, info);
2382 	nvme_set_chunk_sectors(ns, id, &lim);
2383 	if (!nvme_update_disk_info(ns, id, &lim))
2384 		capacity = 0;
2385 
2386 	/*
2387 	 * Validate the max atomic write size fits within the subsystem's
2388 	 * atomic write capabilities.
2389 	 */
2390 	if (lim.atomic_write_hw_max > ns->ctrl->subsys->atomic_bs) {
2391 		blk_mq_unfreeze_queue(ns->disk->queue, memflags);
2392 		ret = -ENXIO;
2393 		goto out;
2394 	}
2395 
2396 	nvme_config_discard(ns, &lim);
2397 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
2398 	    ns->head->ids.csi == NVME_CSI_ZNS)
2399 		nvme_update_zone_info(ns, &lim, &zi);
2400 
2401 	if ((ns->ctrl->vwc & NVME_CTRL_VWC_PRESENT) && !info->no_vwc)
2402 		lim.features |= BLK_FEAT_WRITE_CACHE | BLK_FEAT_FUA;
2403 	else
2404 		lim.features &= ~(BLK_FEAT_WRITE_CACHE | BLK_FEAT_FUA);
2405 
2406 	if (info->is_rotational)
2407 		lim.features |= BLK_FEAT_ROTATIONAL;
2408 
2409 	/*
2410 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
2411 	 * metadata masquerading as Type 0 if supported, otherwise reject block
2412 	 * I/O to namespaces with metadata except when the namespace supports
2413 	 * PI, as it can strip/insert in that case.
2414 	 */
2415 	if (!nvme_init_integrity(ns->head, &lim, info))
2416 		capacity = 0;
2417 
2418 	lim.max_write_streams = ns->head->nr_plids;
2419 	if (lim.max_write_streams)
2420 		lim.write_stream_granularity = min(info->runs, U32_MAX);
2421 	else
2422 		lim.write_stream_granularity = 0;
2423 
2424 	ret = queue_limits_commit_update(ns->disk->queue, &lim);
2425 	if (ret) {
2426 		blk_mq_unfreeze_queue(ns->disk->queue, memflags);
2427 		goto out;
2428 	}
2429 
2430 	set_capacity_and_notify(ns->disk, capacity);
2431 
2432 	/*
2433 	 * Only set the DEAC bit if the device guarantees that reads from
2434 	 * deallocated data return zeroes.  While the DEAC bit does not
2435 	 * require that, it must be a no-op if reads from deallocated data
2436 	 * do not return zeroes.
2437 	 */
2438 	if ((id->dlfeat & 0x7) == 0x1 && (id->dlfeat & (1 << 3)))
2439 		ns->head->features |= NVME_NS_DEAC;
2440 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2441 	set_bit(NVME_NS_READY, &ns->flags);
2442 	blk_mq_unfreeze_queue(ns->disk->queue, memflags);
2443 
2444 	if (blk_queue_is_zoned(ns->queue)) {
2445 		ret = blk_revalidate_disk_zones(ns->disk);
2446 		if (ret && !nvme_first_scan(ns->disk))
2447 			goto out;
2448 	}
2449 
2450 	ret = 0;
2451 out:
2452 	kfree(nvm);
2453 	kfree(id);
2454 	return ret;
2455 }
2456 
2457 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2458 {
2459 	bool unsupported = false;
2460 	int ret;
2461 
2462 	switch (info->ids.csi) {
2463 	case NVME_CSI_ZNS:
2464 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2465 			dev_info(ns->ctrl->device,
2466 	"block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2467 				info->nsid);
2468 			ret = nvme_update_ns_info_generic(ns, info);
2469 			break;
2470 		}
2471 		ret = nvme_update_ns_info_block(ns, info);
2472 		break;
2473 	case NVME_CSI_NVM:
2474 		ret = nvme_update_ns_info_block(ns, info);
2475 		break;
2476 	default:
2477 		dev_info(ns->ctrl->device,
2478 			"block device for nsid %u not supported (csi %u)\n",
2479 			info->nsid, info->ids.csi);
2480 		ret = nvme_update_ns_info_generic(ns, info);
2481 		break;
2482 	}
2483 
2484 	/*
2485 	 * If probing fails due an unsupported feature, hide the block device,
2486 	 * but still allow other access.
2487 	 */
2488 	if (ret == -ENODEV) {
2489 		ns->disk->flags |= GENHD_FL_HIDDEN;
2490 		set_bit(NVME_NS_READY, &ns->flags);
2491 		unsupported = true;
2492 		ret = 0;
2493 	}
2494 
2495 	if (!ret && nvme_ns_head_multipath(ns->head)) {
2496 		struct queue_limits *ns_lim = &ns->disk->queue->limits;
2497 		struct queue_limits lim;
2498 		unsigned int memflags;
2499 
2500 		lim = queue_limits_start_update(ns->head->disk->queue);
2501 		memflags = blk_mq_freeze_queue(ns->head->disk->queue);
2502 		/*
2503 		 * queue_limits mixes values that are the hardware limitations
2504 		 * for bio splitting with what is the device configuration.
2505 		 *
2506 		 * For NVMe the device configuration can change after e.g. a
2507 		 * Format command, and we really want to pick up the new format
2508 		 * value here.  But we must still stack the queue limits to the
2509 		 * least common denominator for multipathing to split the bios
2510 		 * properly.
2511 		 *
2512 		 * To work around this, we explicitly set the device
2513 		 * configuration to those that we just queried, but only stack
2514 		 * the splitting limits in to make sure we still obey possibly
2515 		 * lower limitations of other controllers.
2516 		 */
2517 		lim.logical_block_size = ns_lim->logical_block_size;
2518 		lim.physical_block_size = ns_lim->physical_block_size;
2519 		lim.io_min = ns_lim->io_min;
2520 		lim.io_opt = ns_lim->io_opt;
2521 		queue_limits_stack_bdev(&lim, ns->disk->part0, 0,
2522 					ns->head->disk->disk_name);
2523 		if (unsupported)
2524 			ns->head->disk->flags |= GENHD_FL_HIDDEN;
2525 		else
2526 			nvme_init_integrity(ns->head, &lim, info);
2527 		lim.max_write_streams = ns_lim->max_write_streams;
2528 		lim.write_stream_granularity = ns_lim->write_stream_granularity;
2529 		ret = queue_limits_commit_update(ns->head->disk->queue, &lim);
2530 
2531 		set_capacity_and_notify(ns->head->disk, get_capacity(ns->disk));
2532 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2533 		nvme_mpath_revalidate_paths(ns);
2534 
2535 		blk_mq_unfreeze_queue(ns->head->disk->queue, memflags);
2536 	}
2537 
2538 	return ret;
2539 }
2540 
2541 int nvme_ns_get_unique_id(struct nvme_ns *ns, u8 id[16],
2542 		enum blk_unique_id type)
2543 {
2544 	struct nvme_ns_ids *ids = &ns->head->ids;
2545 
2546 	if (type != BLK_UID_EUI64)
2547 		return -EINVAL;
2548 
2549 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid))) {
2550 		memcpy(id, &ids->nguid, sizeof(ids->nguid));
2551 		return sizeof(ids->nguid);
2552 	}
2553 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64))) {
2554 		memcpy(id, &ids->eui64, sizeof(ids->eui64));
2555 		return sizeof(ids->eui64);
2556 	}
2557 
2558 	return -EINVAL;
2559 }
2560 
2561 static int nvme_get_unique_id(struct gendisk *disk, u8 id[16],
2562 		enum blk_unique_id type)
2563 {
2564 	return nvme_ns_get_unique_id(disk->private_data, id, type);
2565 }
2566 
2567 #ifdef CONFIG_BLK_SED_OPAL
2568 static int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2569 		bool send)
2570 {
2571 	struct nvme_ctrl *ctrl = data;
2572 	struct nvme_command cmd = { };
2573 
2574 	if (send)
2575 		cmd.common.opcode = nvme_admin_security_send;
2576 	else
2577 		cmd.common.opcode = nvme_admin_security_recv;
2578 	cmd.common.nsid = 0;
2579 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2580 	cmd.common.cdw11 = cpu_to_le32(len);
2581 
2582 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2583 			NVME_QID_ANY, NVME_SUBMIT_AT_HEAD);
2584 }
2585 
2586 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2587 {
2588 	if (ctrl->oacs & NVME_CTRL_OACS_SEC_SUPP) {
2589 		if (!ctrl->opal_dev)
2590 			ctrl->opal_dev = init_opal_dev(ctrl, &nvme_sec_submit);
2591 		else if (was_suspended)
2592 			opal_unlock_from_suspend(ctrl->opal_dev);
2593 	} else {
2594 		free_opal_dev(ctrl->opal_dev);
2595 		ctrl->opal_dev = NULL;
2596 	}
2597 }
2598 #else
2599 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2600 {
2601 }
2602 #endif /* CONFIG_BLK_SED_OPAL */
2603 
2604 #ifdef CONFIG_BLK_DEV_ZONED
2605 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2606 		unsigned int nr_zones, report_zones_cb cb, void *data)
2607 {
2608 	return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2609 			data);
2610 }
2611 #else
2612 #define nvme_report_zones	NULL
2613 #endif /* CONFIG_BLK_DEV_ZONED */
2614 
2615 const struct block_device_operations nvme_bdev_ops = {
2616 	.owner		= THIS_MODULE,
2617 	.ioctl		= nvme_ioctl,
2618 	.compat_ioctl	= blkdev_compat_ptr_ioctl,
2619 	.open		= nvme_open,
2620 	.release	= nvme_release,
2621 	.getgeo		= nvme_getgeo,
2622 	.get_unique_id	= nvme_get_unique_id,
2623 	.report_zones	= nvme_report_zones,
2624 	.pr_ops		= &nvme_pr_ops,
2625 };
2626 
2627 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 mask, u32 val,
2628 		u32 timeout, const char *op)
2629 {
2630 	unsigned long timeout_jiffies = jiffies + timeout * HZ;
2631 	u32 csts;
2632 	int ret;
2633 
2634 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2635 		if (csts == ~0)
2636 			return -ENODEV;
2637 		if ((csts & mask) == val)
2638 			break;
2639 
2640 		usleep_range(1000, 2000);
2641 		if (fatal_signal_pending(current))
2642 			return -EINTR;
2643 		if (time_after(jiffies, timeout_jiffies)) {
2644 			dev_err(ctrl->device,
2645 				"Device not ready; aborting %s, CSTS=0x%x\n",
2646 				op, csts);
2647 			return -ENODEV;
2648 		}
2649 	}
2650 
2651 	return ret;
2652 }
2653 
2654 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown)
2655 {
2656 	int ret;
2657 
2658 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2659 	if (shutdown)
2660 		ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2661 	else
2662 		ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2663 
2664 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2665 	if (ret)
2666 		return ret;
2667 
2668 	if (shutdown) {
2669 		return nvme_wait_ready(ctrl, NVME_CSTS_SHST_MASK,
2670 				       NVME_CSTS_SHST_CMPLT,
2671 				       ctrl->shutdown_timeout, "shutdown");
2672 	}
2673 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2674 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2675 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, 0,
2676 			       (NVME_CAP_TIMEOUT(ctrl->cap) + 1) / 2, "reset");
2677 }
2678 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2679 
2680 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2681 {
2682 	unsigned dev_page_min;
2683 	u32 timeout;
2684 	int ret;
2685 
2686 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2687 	if (ret) {
2688 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2689 		return ret;
2690 	}
2691 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2692 
2693 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2694 		dev_err(ctrl->device,
2695 			"Minimum device page size %u too large for host (%u)\n",
2696 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2697 		return -ENODEV;
2698 	}
2699 
2700 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2701 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2702 	else
2703 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2704 
2705 	/*
2706 	 * Setting CRIME results in CSTS.RDY before the media is ready. This
2707 	 * makes it possible for media related commands to return the error
2708 	 * NVME_SC_ADMIN_COMMAND_MEDIA_NOT_READY. Until the driver is
2709 	 * restructured to handle retries, disable CC.CRIME.
2710 	 */
2711 	ctrl->ctrl_config &= ~NVME_CC_CRIME;
2712 
2713 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2714 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2715 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2716 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2717 	if (ret)
2718 		return ret;
2719 
2720 	/* CAP value may change after initial CC write */
2721 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2722 	if (ret)
2723 		return ret;
2724 
2725 	timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2726 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2727 		u32 crto, ready_timeout;
2728 
2729 		ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2730 		if (ret) {
2731 			dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2732 				ret);
2733 			return ret;
2734 		}
2735 
2736 		/*
2737 		 * CRTO should always be greater or equal to CAP.TO, but some
2738 		 * devices are known to get this wrong. Use the larger of the
2739 		 * two values.
2740 		 */
2741 		ready_timeout = NVME_CRTO_CRWMT(crto);
2742 
2743 		if (ready_timeout < timeout)
2744 			dev_warn_once(ctrl->device, "bad crto:%x cap:%llx\n",
2745 				      crto, ctrl->cap);
2746 		else
2747 			timeout = ready_timeout;
2748 	}
2749 
2750 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2751 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2752 	if (ret)
2753 		return ret;
2754 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, NVME_CSTS_RDY,
2755 			       (timeout + 1) / 2, "initialisation");
2756 }
2757 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2758 
2759 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2760 {
2761 	__le64 ts;
2762 	int ret;
2763 
2764 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2765 		return 0;
2766 
2767 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2768 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2769 			NULL);
2770 	if (ret)
2771 		dev_warn_once(ctrl->device,
2772 			"could not set timestamp (%d)\n", ret);
2773 	return ret;
2774 }
2775 
2776 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2777 {
2778 	struct nvme_feat_host_behavior *host;
2779 	u8 acre = 0, lbafee = 0;
2780 	int ret;
2781 
2782 	/* Don't bother enabling the feature if retry delay is not reported */
2783 	if (ctrl->crdt[0])
2784 		acre = NVME_ENABLE_ACRE;
2785 	if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2786 		lbafee = NVME_ENABLE_LBAFEE;
2787 
2788 	if (!acre && !lbafee)
2789 		return 0;
2790 
2791 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2792 	if (!host)
2793 		return 0;
2794 
2795 	host->acre = acre;
2796 	host->lbafee = lbafee;
2797 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2798 				host, sizeof(*host), NULL);
2799 	kfree(host);
2800 	return ret;
2801 }
2802 
2803 /*
2804  * The function checks whether the given total (exlat + enlat) latency of
2805  * a power state allows the latter to be used as an APST transition target.
2806  * It does so by comparing the latency to the primary and secondary latency
2807  * tolerances defined by module params. If there's a match, the corresponding
2808  * timeout value is returned and the matching tolerance index (1 or 2) is
2809  * reported.
2810  */
2811 static bool nvme_apst_get_transition_time(u64 total_latency,
2812 		u64 *transition_time, unsigned *last_index)
2813 {
2814 	if (total_latency <= apst_primary_latency_tol_us) {
2815 		if (*last_index == 1)
2816 			return false;
2817 		*last_index = 1;
2818 		*transition_time = apst_primary_timeout_ms;
2819 		return true;
2820 	}
2821 	if (apst_secondary_timeout_ms &&
2822 		total_latency <= apst_secondary_latency_tol_us) {
2823 		if (*last_index <= 2)
2824 			return false;
2825 		*last_index = 2;
2826 		*transition_time = apst_secondary_timeout_ms;
2827 		return true;
2828 	}
2829 	return false;
2830 }
2831 
2832 /*
2833  * APST (Autonomous Power State Transition) lets us program a table of power
2834  * state transitions that the controller will perform automatically.
2835  *
2836  * Depending on module params, one of the two supported techniques will be used:
2837  *
2838  * - If the parameters provide explicit timeouts and tolerances, they will be
2839  *   used to build a table with up to 2 non-operational states to transition to.
2840  *   The default parameter values were selected based on the values used by
2841  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2842  *   regeneration of the APST table in the event of switching between external
2843  *   and battery power, the timeouts and tolerances reflect a compromise
2844  *   between values used by Microsoft for AC and battery scenarios.
2845  * - If not, we'll configure the table with a simple heuristic: we are willing
2846  *   to spend at most 2% of the time transitioning between power states.
2847  *   Therefore, when running in any given state, we will enter the next
2848  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2849  *   microseconds, as long as that state's exit latency is under the requested
2850  *   maximum latency.
2851  *
2852  * We will not autonomously enter any non-operational state for which the total
2853  * latency exceeds ps_max_latency_us.
2854  *
2855  * Users can set ps_max_latency_us to zero to turn off APST.
2856  */
2857 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2858 {
2859 	struct nvme_feat_auto_pst *table;
2860 	unsigned apste = 0;
2861 	u64 max_lat_us = 0;
2862 	__le64 target = 0;
2863 	int max_ps = -1;
2864 	int state;
2865 	int ret;
2866 	unsigned last_lt_index = UINT_MAX;
2867 
2868 	/*
2869 	 * If APST isn't supported or if we haven't been initialized yet,
2870 	 * then don't do anything.
2871 	 */
2872 	if (!ctrl->apsta)
2873 		return 0;
2874 
2875 	if (ctrl->npss > 31) {
2876 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2877 		return 0;
2878 	}
2879 
2880 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2881 	if (!table)
2882 		return 0;
2883 
2884 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2885 		/* Turn off APST. */
2886 		dev_dbg(ctrl->device, "APST disabled\n");
2887 		goto done;
2888 	}
2889 
2890 	/*
2891 	 * Walk through all states from lowest- to highest-power.
2892 	 * According to the spec, lower-numbered states use more power.  NPSS,
2893 	 * despite the name, is the index of the lowest-power state, not the
2894 	 * number of states.
2895 	 */
2896 	for (state = (int)ctrl->npss; state >= 0; state--) {
2897 		u64 total_latency_us, exit_latency_us, transition_ms;
2898 
2899 		if (target)
2900 			table->entries[state] = target;
2901 
2902 		/*
2903 		 * Don't allow transitions to the deepest state if it's quirked
2904 		 * off.
2905 		 */
2906 		if (state == ctrl->npss &&
2907 		    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2908 			continue;
2909 
2910 		/*
2911 		 * Is this state a useful non-operational state for higher-power
2912 		 * states to autonomously transition to?
2913 		 */
2914 		if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2915 			continue;
2916 
2917 		exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2918 		if (exit_latency_us > ctrl->ps_max_latency_us)
2919 			continue;
2920 
2921 		total_latency_us = exit_latency_us +
2922 			le32_to_cpu(ctrl->psd[state].entry_lat);
2923 
2924 		/*
2925 		 * This state is good. It can be used as the APST idle target
2926 		 * for higher power states.
2927 		 */
2928 		if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2929 			if (!nvme_apst_get_transition_time(total_latency_us,
2930 					&transition_ms, &last_lt_index))
2931 				continue;
2932 		} else {
2933 			transition_ms = total_latency_us + 19;
2934 			do_div(transition_ms, 20);
2935 			if (transition_ms > (1 << 24) - 1)
2936 				transition_ms = (1 << 24) - 1;
2937 		}
2938 
2939 		target = cpu_to_le64((state << 3) | (transition_ms << 8));
2940 		if (max_ps == -1)
2941 			max_ps = state;
2942 		if (total_latency_us > max_lat_us)
2943 			max_lat_us = total_latency_us;
2944 	}
2945 
2946 	if (max_ps == -1)
2947 		dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2948 	else
2949 		dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2950 			max_ps, max_lat_us, (int)sizeof(*table), table);
2951 	apste = 1;
2952 
2953 done:
2954 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2955 				table, sizeof(*table), NULL);
2956 	if (ret)
2957 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2958 	kfree(table);
2959 	return ret;
2960 }
2961 
2962 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2963 {
2964 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2965 	u64 latency;
2966 
2967 	switch (val) {
2968 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2969 	case PM_QOS_LATENCY_ANY:
2970 		latency = U64_MAX;
2971 		break;
2972 
2973 	default:
2974 		latency = val;
2975 	}
2976 
2977 	if (ctrl->ps_max_latency_us != latency) {
2978 		ctrl->ps_max_latency_us = latency;
2979 		if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
2980 			nvme_configure_apst(ctrl);
2981 	}
2982 }
2983 
2984 struct nvme_core_quirk_entry {
2985 	/*
2986 	 * NVMe model and firmware strings are padded with spaces.  For
2987 	 * simplicity, strings in the quirk table are padded with NULLs
2988 	 * instead.
2989 	 */
2990 	u16 vid;
2991 	const char *mn;
2992 	const char *fr;
2993 	unsigned long quirks;
2994 };
2995 
2996 static const struct nvme_core_quirk_entry core_quirks[] = {
2997 	{
2998 		/*
2999 		 * This Toshiba device seems to die using any APST states.  See:
3000 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
3001 		 */
3002 		.vid = 0x1179,
3003 		.mn = "THNSF5256GPUK TOSHIBA",
3004 		.quirks = NVME_QUIRK_NO_APST,
3005 	},
3006 	{
3007 		/*
3008 		 * This LiteON CL1-3D*-Q11 firmware version has a race
3009 		 * condition associated with actions related to suspend to idle
3010 		 * LiteON has resolved the problem in future firmware
3011 		 */
3012 		.vid = 0x14a4,
3013 		.fr = "22301111",
3014 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
3015 	},
3016 	{
3017 		/*
3018 		 * This Kioxia CD6-V Series / HPE PE8030 device times out and
3019 		 * aborts I/O during any load, but more easily reproducible
3020 		 * with discards (fstrim).
3021 		 *
3022 		 * The device is left in a state where it is also not possible
3023 		 * to use "nvme set-feature" to disable APST, but booting with
3024 		 * nvme_core.default_ps_max_latency=0 works.
3025 		 */
3026 		.vid = 0x1e0f,
3027 		.mn = "KCD6XVUL6T40",
3028 		.quirks = NVME_QUIRK_NO_APST,
3029 	},
3030 	{
3031 		/*
3032 		 * The external Samsung X5 SSD fails initialization without a
3033 		 * delay before checking if it is ready and has a whole set of
3034 		 * other problems.  To make this even more interesting, it
3035 		 * shares the PCI ID with internal Samsung 970 Evo Plus that
3036 		 * does not need or want these quirks.
3037 		 */
3038 		.vid = 0x144d,
3039 		.mn = "Samsung Portable SSD X5",
3040 		.quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
3041 			  NVME_QUIRK_NO_DEEPEST_PS |
3042 			  NVME_QUIRK_IGNORE_DEV_SUBNQN,
3043 	}
3044 };
3045 
3046 /* match is null-terminated but idstr is space-padded. */
3047 static bool string_matches(const char *idstr, const char *match, size_t len)
3048 {
3049 	size_t matchlen;
3050 
3051 	if (!match)
3052 		return true;
3053 
3054 	matchlen = strlen(match);
3055 	WARN_ON_ONCE(matchlen > len);
3056 
3057 	if (memcmp(idstr, match, matchlen))
3058 		return false;
3059 
3060 	for (; matchlen < len; matchlen++)
3061 		if (idstr[matchlen] != ' ')
3062 			return false;
3063 
3064 	return true;
3065 }
3066 
3067 static bool quirk_matches(const struct nvme_id_ctrl *id,
3068 			  const struct nvme_core_quirk_entry *q)
3069 {
3070 	return q->vid == le16_to_cpu(id->vid) &&
3071 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
3072 		string_matches(id->fr, q->fr, sizeof(id->fr));
3073 }
3074 
3075 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
3076 		struct nvme_id_ctrl *id)
3077 {
3078 	size_t nqnlen;
3079 	int off;
3080 
3081 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
3082 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
3083 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
3084 			strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
3085 			return;
3086 		}
3087 
3088 		if (ctrl->vs >= NVME_VS(1, 2, 1))
3089 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
3090 	}
3091 
3092 	/*
3093 	 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
3094 	 * Base Specification 2.0.  It is slightly different from the format
3095 	 * specified there due to historic reasons, and we can't change it now.
3096 	 */
3097 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
3098 			"nqn.2014.08.org.nvmexpress:%04x%04x",
3099 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
3100 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
3101 	off += sizeof(id->sn);
3102 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
3103 	off += sizeof(id->mn);
3104 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
3105 }
3106 
3107 static void nvme_release_subsystem(struct device *dev)
3108 {
3109 	struct nvme_subsystem *subsys =
3110 		container_of(dev, struct nvme_subsystem, dev);
3111 
3112 	if (subsys->instance >= 0)
3113 		ida_free(&nvme_instance_ida, subsys->instance);
3114 	kfree(subsys);
3115 }
3116 
3117 static void nvme_destroy_subsystem(struct kref *ref)
3118 {
3119 	struct nvme_subsystem *subsys =
3120 			container_of(ref, struct nvme_subsystem, ref);
3121 
3122 	mutex_lock(&nvme_subsystems_lock);
3123 	list_del(&subsys->entry);
3124 	mutex_unlock(&nvme_subsystems_lock);
3125 
3126 	ida_destroy(&subsys->ns_ida);
3127 	device_del(&subsys->dev);
3128 	put_device(&subsys->dev);
3129 }
3130 
3131 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
3132 {
3133 	kref_put(&subsys->ref, nvme_destroy_subsystem);
3134 }
3135 
3136 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
3137 {
3138 	struct nvme_subsystem *subsys;
3139 
3140 	lockdep_assert_held(&nvme_subsystems_lock);
3141 
3142 	/*
3143 	 * Fail matches for discovery subsystems. This results
3144 	 * in each discovery controller bound to a unique subsystem.
3145 	 * This avoids issues with validating controller values
3146 	 * that can only be true when there is a single unique subsystem.
3147 	 * There may be multiple and completely independent entities
3148 	 * that provide discovery controllers.
3149 	 */
3150 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
3151 		return NULL;
3152 
3153 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
3154 		if (strcmp(subsys->subnqn, subsysnqn))
3155 			continue;
3156 		if (!kref_get_unless_zero(&subsys->ref))
3157 			continue;
3158 		return subsys;
3159 	}
3160 
3161 	return NULL;
3162 }
3163 
3164 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
3165 {
3166 	return ctrl->opts && ctrl->opts->discovery_nqn;
3167 }
3168 
3169 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
3170 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
3171 {
3172 	struct nvme_ctrl *tmp;
3173 
3174 	lockdep_assert_held(&nvme_subsystems_lock);
3175 
3176 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
3177 		if (nvme_state_terminal(tmp))
3178 			continue;
3179 
3180 		if (tmp->cntlid == ctrl->cntlid) {
3181 			dev_err(ctrl->device,
3182 				"Duplicate cntlid %u with %s, subsys %s, rejecting\n",
3183 				ctrl->cntlid, dev_name(tmp->device),
3184 				subsys->subnqn);
3185 			return false;
3186 		}
3187 
3188 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
3189 		    nvme_discovery_ctrl(ctrl))
3190 			continue;
3191 
3192 		dev_err(ctrl->device,
3193 			"Subsystem does not support multiple controllers\n");
3194 		return false;
3195 	}
3196 
3197 	return true;
3198 }
3199 
3200 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
3201 {
3202 	struct nvme_subsystem *subsys, *found;
3203 	int ret;
3204 
3205 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
3206 	if (!subsys)
3207 		return -ENOMEM;
3208 
3209 	subsys->instance = -1;
3210 	mutex_init(&subsys->lock);
3211 	kref_init(&subsys->ref);
3212 	INIT_LIST_HEAD(&subsys->ctrls);
3213 	INIT_LIST_HEAD(&subsys->nsheads);
3214 	nvme_init_subnqn(subsys, ctrl, id);
3215 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
3216 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
3217 	subsys->vendor_id = le16_to_cpu(id->vid);
3218 	subsys->cmic = id->cmic;
3219 
3220 	/* Versions prior to 1.4 don't necessarily report a valid type */
3221 	if (id->cntrltype == NVME_CTRL_DISC ||
3222 	    !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
3223 		subsys->subtype = NVME_NQN_DISC;
3224 	else
3225 		subsys->subtype = NVME_NQN_NVME;
3226 
3227 	if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
3228 		dev_err(ctrl->device,
3229 			"Subsystem %s is not a discovery controller",
3230 			subsys->subnqn);
3231 		kfree(subsys);
3232 		return -EINVAL;
3233 	}
3234 	nvme_mpath_default_iopolicy(subsys);
3235 
3236 	subsys->dev.class = &nvme_subsys_class;
3237 	subsys->dev.release = nvme_release_subsystem;
3238 	subsys->dev.groups = nvme_subsys_attrs_groups;
3239 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
3240 	device_initialize(&subsys->dev);
3241 
3242 	mutex_lock(&nvme_subsystems_lock);
3243 	found = __nvme_find_get_subsystem(subsys->subnqn);
3244 	if (found) {
3245 		put_device(&subsys->dev);
3246 		subsys = found;
3247 
3248 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
3249 			ret = -EINVAL;
3250 			goto out_put_subsystem;
3251 		}
3252 	} else {
3253 		ret = device_add(&subsys->dev);
3254 		if (ret) {
3255 			dev_err(ctrl->device,
3256 				"failed to register subsystem device.\n");
3257 			put_device(&subsys->dev);
3258 			goto out_unlock;
3259 		}
3260 		ida_init(&subsys->ns_ida);
3261 		list_add_tail(&subsys->entry, &nvme_subsystems);
3262 	}
3263 
3264 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
3265 				dev_name(ctrl->device));
3266 	if (ret) {
3267 		dev_err(ctrl->device,
3268 			"failed to create sysfs link from subsystem.\n");
3269 		goto out_put_subsystem;
3270 	}
3271 
3272 	if (!found)
3273 		subsys->instance = ctrl->instance;
3274 	ctrl->subsys = subsys;
3275 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
3276 	mutex_unlock(&nvme_subsystems_lock);
3277 	return 0;
3278 
3279 out_put_subsystem:
3280 	nvme_put_subsystem(subsys);
3281 out_unlock:
3282 	mutex_unlock(&nvme_subsystems_lock);
3283 	return ret;
3284 }
3285 
3286 static int nvme_get_log_lsi(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page,
3287 		u8 lsp, u8 csi, void *log, size_t size, u64 offset, u16 lsi)
3288 {
3289 	struct nvme_command c = { };
3290 	u32 dwlen = nvme_bytes_to_numd(size);
3291 
3292 	c.get_log_page.opcode = nvme_admin_get_log_page;
3293 	c.get_log_page.nsid = cpu_to_le32(nsid);
3294 	c.get_log_page.lid = log_page;
3295 	c.get_log_page.lsp = lsp;
3296 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3297 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3298 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3299 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3300 	c.get_log_page.csi = csi;
3301 	c.get_log_page.lsi = cpu_to_le16(lsi);
3302 
3303 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3304 }
3305 
3306 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
3307 		void *log, size_t size, u64 offset)
3308 {
3309 	return nvme_get_log_lsi(ctrl, nsid, log_page, lsp, csi, log, size,
3310 			offset, 0);
3311 }
3312 
3313 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3314 				struct nvme_effects_log **log)
3315 {
3316 	struct nvme_effects_log *old, *cel = xa_load(&ctrl->cels, csi);
3317 	int ret;
3318 
3319 	if (cel)
3320 		goto out;
3321 
3322 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3323 	if (!cel)
3324 		return -ENOMEM;
3325 
3326 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3327 			cel, sizeof(*cel), 0);
3328 	if (ret) {
3329 		kfree(cel);
3330 		return ret;
3331 	}
3332 
3333 	old = xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3334 	if (xa_is_err(old)) {
3335 		kfree(cel);
3336 		return xa_err(old);
3337 	}
3338 out:
3339 	*log = cel;
3340 	return 0;
3341 }
3342 
3343 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
3344 {
3345 	u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
3346 
3347 	if (check_shl_overflow(1U, units + page_shift - 9, &val))
3348 		return UINT_MAX;
3349 	return val;
3350 }
3351 
3352 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
3353 {
3354 	struct nvme_command c = { };
3355 	struct nvme_id_ctrl_nvm *id;
3356 	int ret;
3357 
3358 	/*
3359 	 * Even though NVMe spec explicitly states that MDTS is not applicable
3360 	 * to the write-zeroes, we are cautious and limit the size to the
3361 	 * controllers max_hw_sectors value, which is based on the MDTS field
3362 	 * and possibly other limiting factors.
3363 	 */
3364 	if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
3365 	    !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
3366 		ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
3367 	else
3368 		ctrl->max_zeroes_sectors = 0;
3369 
3370 	if (ctrl->subsys->subtype != NVME_NQN_NVME ||
3371 	    !nvme_id_cns_ok(ctrl, NVME_ID_CNS_CS_CTRL) ||
3372 	    test_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags))
3373 		return 0;
3374 
3375 	id = kzalloc(sizeof(*id), GFP_KERNEL);
3376 	if (!id)
3377 		return -ENOMEM;
3378 
3379 	c.identify.opcode = nvme_admin_identify;
3380 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
3381 	c.identify.csi = NVME_CSI_NVM;
3382 
3383 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
3384 	if (ret)
3385 		goto free_data;
3386 
3387 	ctrl->dmrl = id->dmrl;
3388 	ctrl->dmrsl = le32_to_cpu(id->dmrsl);
3389 	if (id->wzsl)
3390 		ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
3391 
3392 free_data:
3393 	if (ret > 0)
3394 		set_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags);
3395 	kfree(id);
3396 	return ret;
3397 }
3398 
3399 static int nvme_init_effects_log(struct nvme_ctrl *ctrl,
3400 		u8 csi, struct nvme_effects_log **log)
3401 {
3402 	struct nvme_effects_log *effects, *old;
3403 
3404 	effects = kzalloc(sizeof(*effects), GFP_KERNEL);
3405 	if (!effects)
3406 		return -ENOMEM;
3407 
3408 	old = xa_store(&ctrl->cels, csi, effects, GFP_KERNEL);
3409 	if (xa_is_err(old)) {
3410 		kfree(effects);
3411 		return xa_err(old);
3412 	}
3413 
3414 	*log = effects;
3415 	return 0;
3416 }
3417 
3418 static void nvme_init_known_nvm_effects(struct nvme_ctrl *ctrl)
3419 {
3420 	struct nvme_effects_log	*log = ctrl->effects;
3421 
3422 	log->acs[nvme_admin_format_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
3423 						NVME_CMD_EFFECTS_NCC |
3424 						NVME_CMD_EFFECTS_CSE_MASK);
3425 	log->acs[nvme_admin_sanitize_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
3426 						NVME_CMD_EFFECTS_CSE_MASK);
3427 
3428 	/*
3429 	 * The spec says the result of a security receive command depends on
3430 	 * the previous security send command. As such, many vendors log this
3431 	 * command as one to submitted only when no other commands to the same
3432 	 * namespace are outstanding. The intention is to tell the host to
3433 	 * prevent mixing security send and receive.
3434 	 *
3435 	 * This driver can only enforce such exclusive access against IO
3436 	 * queues, though. We are not readily able to enforce such a rule for
3437 	 * two commands to the admin queue, which is the only queue that
3438 	 * matters for this command.
3439 	 *
3440 	 * Rather than blindly freezing the IO queues for this effect that
3441 	 * doesn't even apply to IO, mask it off.
3442 	 */
3443 	log->acs[nvme_admin_security_recv] &= cpu_to_le32(~NVME_CMD_EFFECTS_CSE_MASK);
3444 
3445 	log->iocs[nvme_cmd_write] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3446 	log->iocs[nvme_cmd_write_zeroes] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3447 	log->iocs[nvme_cmd_write_uncor] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3448 }
3449 
3450 static int nvme_init_effects(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
3451 {
3452 	int ret = 0;
3453 
3454 	if (ctrl->effects)
3455 		return 0;
3456 
3457 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3458 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3459 		if (ret < 0)
3460 			return ret;
3461 	}
3462 
3463 	if (!ctrl->effects) {
3464 		ret = nvme_init_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3465 		if (ret < 0)
3466 			return ret;
3467 	}
3468 
3469 	nvme_init_known_nvm_effects(ctrl);
3470 	return 0;
3471 }
3472 
3473 static int nvme_check_ctrl_fabric_info(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
3474 {
3475 	/*
3476 	 * In fabrics we need to verify the cntlid matches the
3477 	 * admin connect
3478 	 */
3479 	if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3480 		dev_err(ctrl->device,
3481 			"Mismatching cntlid: Connect %u vs Identify %u, rejecting\n",
3482 			ctrl->cntlid, le16_to_cpu(id->cntlid));
3483 		return -EINVAL;
3484 	}
3485 
3486 	if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3487 		dev_err(ctrl->device,
3488 			"keep-alive support is mandatory for fabrics\n");
3489 		return -EINVAL;
3490 	}
3491 
3492 	if (!nvme_discovery_ctrl(ctrl) && ctrl->ioccsz < 4) {
3493 		dev_err(ctrl->device,
3494 			"I/O queue command capsule supported size %d < 4\n",
3495 			ctrl->ioccsz);
3496 		return -EINVAL;
3497 	}
3498 
3499 	if (!nvme_discovery_ctrl(ctrl) && ctrl->iorcsz < 1) {
3500 		dev_err(ctrl->device,
3501 			"I/O queue response capsule supported size %d < 1\n",
3502 			ctrl->iorcsz);
3503 		return -EINVAL;
3504 	}
3505 
3506 	if (!ctrl->maxcmd) {
3507 		dev_warn(ctrl->device,
3508 			"Firmware bug: maximum outstanding commands is 0\n");
3509 		ctrl->maxcmd = ctrl->sqsize + 1;
3510 	}
3511 
3512 	return 0;
3513 }
3514 
3515 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3516 {
3517 	struct queue_limits lim;
3518 	struct nvme_id_ctrl *id;
3519 	u32 max_hw_sectors;
3520 	bool prev_apst_enabled;
3521 	int ret;
3522 
3523 	ret = nvme_identify_ctrl(ctrl, &id);
3524 	if (ret) {
3525 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3526 		return -EIO;
3527 	}
3528 
3529 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
3530 		ctrl->cntlid = le16_to_cpu(id->cntlid);
3531 
3532 	if (!ctrl->identified) {
3533 		unsigned int i;
3534 
3535 		/*
3536 		 * Check for quirks.  Quirk can depend on firmware version,
3537 		 * so, in principle, the set of quirks present can change
3538 		 * across a reset.  As a possible future enhancement, we
3539 		 * could re-scan for quirks every time we reinitialize
3540 		 * the device, but we'd have to make sure that the driver
3541 		 * behaves intelligently if the quirks change.
3542 		 */
3543 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3544 			if (quirk_matches(id, &core_quirks[i]))
3545 				ctrl->quirks |= core_quirks[i].quirks;
3546 		}
3547 
3548 		ret = nvme_init_subsystem(ctrl, id);
3549 		if (ret)
3550 			goto out_free;
3551 
3552 		ret = nvme_init_effects(ctrl, id);
3553 		if (ret)
3554 			goto out_free;
3555 	}
3556 	memcpy(ctrl->subsys->firmware_rev, id->fr,
3557 	       sizeof(ctrl->subsys->firmware_rev));
3558 
3559 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3560 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3561 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3562 	}
3563 
3564 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3565 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3566 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3567 
3568 	ctrl->oacs = le16_to_cpu(id->oacs);
3569 	ctrl->oncs = le16_to_cpu(id->oncs);
3570 	ctrl->mtfa = le16_to_cpu(id->mtfa);
3571 	ctrl->oaes = le32_to_cpu(id->oaes);
3572 	ctrl->wctemp = le16_to_cpu(id->wctemp);
3573 	ctrl->cctemp = le16_to_cpu(id->cctemp);
3574 
3575 	atomic_set(&ctrl->abort_limit, id->acl + 1);
3576 	ctrl->vwc = id->vwc;
3577 	if (id->mdts)
3578 		max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3579 	else
3580 		max_hw_sectors = UINT_MAX;
3581 	ctrl->max_hw_sectors =
3582 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3583 
3584 	lim = queue_limits_start_update(ctrl->admin_q);
3585 	nvme_set_ctrl_limits(ctrl, &lim);
3586 	ret = queue_limits_commit_update(ctrl->admin_q, &lim);
3587 	if (ret)
3588 		goto out_free;
3589 
3590 	ctrl->sgls = le32_to_cpu(id->sgls);
3591 	ctrl->kas = le16_to_cpu(id->kas);
3592 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
3593 	ctrl->ctratt = le32_to_cpu(id->ctratt);
3594 
3595 	ctrl->cntrltype = id->cntrltype;
3596 	ctrl->dctype = id->dctype;
3597 
3598 	if (id->rtd3e) {
3599 		/* us -> s */
3600 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3601 
3602 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3603 						 shutdown_timeout, 60);
3604 
3605 		if (ctrl->shutdown_timeout != shutdown_timeout)
3606 			dev_info(ctrl->device,
3607 				 "D3 entry latency set to %u seconds\n",
3608 				 ctrl->shutdown_timeout);
3609 	} else
3610 		ctrl->shutdown_timeout = shutdown_timeout;
3611 
3612 	ctrl->npss = id->npss;
3613 	ctrl->apsta = id->apsta;
3614 	prev_apst_enabled = ctrl->apst_enabled;
3615 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3616 		if (force_apst && id->apsta) {
3617 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3618 			ctrl->apst_enabled = true;
3619 		} else {
3620 			ctrl->apst_enabled = false;
3621 		}
3622 	} else {
3623 		ctrl->apst_enabled = id->apsta;
3624 	}
3625 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3626 
3627 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3628 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3629 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3630 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3631 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3632 
3633 		ret = nvme_check_ctrl_fabric_info(ctrl, id);
3634 		if (ret)
3635 			goto out_free;
3636 	} else {
3637 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3638 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3639 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3640 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3641 	}
3642 
3643 	ret = nvme_mpath_init_identify(ctrl, id);
3644 	if (ret < 0)
3645 		goto out_free;
3646 
3647 	if (ctrl->apst_enabled && !prev_apst_enabled)
3648 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3649 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3650 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3651 	ctrl->awupf = le16_to_cpu(id->awupf);
3652 out_free:
3653 	kfree(id);
3654 	return ret;
3655 }
3656 
3657 /*
3658  * Initialize the cached copies of the Identify data and various controller
3659  * register in our nvme_ctrl structure.  This should be called as soon as
3660  * the admin queue is fully up and running.
3661  */
3662 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended)
3663 {
3664 	int ret;
3665 
3666 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3667 	if (ret) {
3668 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3669 		return ret;
3670 	}
3671 
3672 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3673 
3674 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3675 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3676 
3677 	ret = nvme_init_identify(ctrl);
3678 	if (ret)
3679 		return ret;
3680 
3681 	ret = nvme_configure_apst(ctrl);
3682 	if (ret < 0)
3683 		return ret;
3684 
3685 	ret = nvme_configure_timestamp(ctrl);
3686 	if (ret < 0)
3687 		return ret;
3688 
3689 	ret = nvme_configure_host_options(ctrl);
3690 	if (ret < 0)
3691 		return ret;
3692 
3693 	nvme_configure_opal(ctrl, was_suspended);
3694 
3695 	if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3696 		/*
3697 		 * Do not return errors unless we are in a controller reset,
3698 		 * the controller works perfectly fine without hwmon.
3699 		 */
3700 		ret = nvme_hwmon_init(ctrl);
3701 		if (ret == -EINTR)
3702 			return ret;
3703 	}
3704 
3705 	clear_bit(NVME_CTRL_DIRTY_CAPABILITY, &ctrl->flags);
3706 	ctrl->identified = true;
3707 
3708 	nvme_start_keep_alive(ctrl);
3709 
3710 	return 0;
3711 }
3712 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3713 
3714 static int nvme_dev_open(struct inode *inode, struct file *file)
3715 {
3716 	struct nvme_ctrl *ctrl =
3717 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3718 
3719 	switch (nvme_ctrl_state(ctrl)) {
3720 	case NVME_CTRL_LIVE:
3721 		break;
3722 	default:
3723 		return -EWOULDBLOCK;
3724 	}
3725 
3726 	nvme_get_ctrl(ctrl);
3727 	if (!try_module_get(ctrl->ops->module)) {
3728 		nvme_put_ctrl(ctrl);
3729 		return -EINVAL;
3730 	}
3731 
3732 	file->private_data = ctrl;
3733 	return 0;
3734 }
3735 
3736 static int nvme_dev_release(struct inode *inode, struct file *file)
3737 {
3738 	struct nvme_ctrl *ctrl =
3739 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3740 
3741 	module_put(ctrl->ops->module);
3742 	nvme_put_ctrl(ctrl);
3743 	return 0;
3744 }
3745 
3746 static const struct file_operations nvme_dev_fops = {
3747 	.owner		= THIS_MODULE,
3748 	.open		= nvme_dev_open,
3749 	.release	= nvme_dev_release,
3750 	.unlocked_ioctl	= nvme_dev_ioctl,
3751 	.compat_ioctl	= compat_ptr_ioctl,
3752 	.uring_cmd	= nvme_dev_uring_cmd,
3753 };
3754 
3755 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3756 		unsigned nsid)
3757 {
3758 	struct nvme_ns_head *h;
3759 
3760 	lockdep_assert_held(&ctrl->subsys->lock);
3761 
3762 	list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3763 		/*
3764 		 * Private namespaces can share NSIDs under some conditions.
3765 		 * In that case we can't use the same ns_head for namespaces
3766 		 * with the same NSID.
3767 		 */
3768 		if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3769 			continue;
3770 		if (nvme_tryget_ns_head(h))
3771 			return h;
3772 	}
3773 
3774 	return NULL;
3775 }
3776 
3777 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3778 		struct nvme_ns_ids *ids)
3779 {
3780 	bool has_uuid = !uuid_is_null(&ids->uuid);
3781 	bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3782 	bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3783 	struct nvme_ns_head *h;
3784 
3785 	lockdep_assert_held(&subsys->lock);
3786 
3787 	list_for_each_entry(h, &subsys->nsheads, entry) {
3788 		if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3789 			return -EINVAL;
3790 		if (has_nguid &&
3791 		    memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3792 			return -EINVAL;
3793 		if (has_eui64 &&
3794 		    memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3795 			return -EINVAL;
3796 	}
3797 
3798 	return 0;
3799 }
3800 
3801 static void nvme_cdev_rel(struct device *dev)
3802 {
3803 	ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3804 }
3805 
3806 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3807 {
3808 	cdev_device_del(cdev, cdev_device);
3809 	put_device(cdev_device);
3810 }
3811 
3812 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3813 		const struct file_operations *fops, struct module *owner)
3814 {
3815 	int minor, ret;
3816 
3817 	minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3818 	if (minor < 0)
3819 		return minor;
3820 	cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3821 	cdev_device->class = &nvme_ns_chr_class;
3822 	cdev_device->release = nvme_cdev_rel;
3823 	device_initialize(cdev_device);
3824 	cdev_init(cdev, fops);
3825 	cdev->owner = owner;
3826 	ret = cdev_device_add(cdev, cdev_device);
3827 	if (ret)
3828 		put_device(cdev_device);
3829 
3830 	return ret;
3831 }
3832 
3833 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3834 {
3835 	return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3836 }
3837 
3838 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3839 {
3840 	nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3841 	return 0;
3842 }
3843 
3844 static const struct file_operations nvme_ns_chr_fops = {
3845 	.owner		= THIS_MODULE,
3846 	.open		= nvme_ns_chr_open,
3847 	.release	= nvme_ns_chr_release,
3848 	.unlocked_ioctl	= nvme_ns_chr_ioctl,
3849 	.compat_ioctl	= compat_ptr_ioctl,
3850 	.uring_cmd	= nvme_ns_chr_uring_cmd,
3851 	.uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
3852 };
3853 
3854 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3855 {
3856 	int ret;
3857 
3858 	ns->cdev_device.parent = ns->ctrl->device;
3859 	ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3860 			   ns->ctrl->instance, ns->head->instance);
3861 	if (ret)
3862 		return ret;
3863 
3864 	return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3865 			     ns->ctrl->ops->module);
3866 }
3867 
3868 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3869 		struct nvme_ns_info *info)
3870 {
3871 	struct nvme_ns_head *head;
3872 	size_t size = sizeof(*head);
3873 	int ret = -ENOMEM;
3874 
3875 #ifdef CONFIG_NVME_MULTIPATH
3876 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3877 #endif
3878 
3879 	head = kzalloc(size, GFP_KERNEL);
3880 	if (!head)
3881 		goto out;
3882 	ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3883 	if (ret < 0)
3884 		goto out_free_head;
3885 	head->instance = ret;
3886 	INIT_LIST_HEAD(&head->list);
3887 	ret = init_srcu_struct(&head->srcu);
3888 	if (ret)
3889 		goto out_ida_remove;
3890 	head->subsys = ctrl->subsys;
3891 	head->ns_id = info->nsid;
3892 	head->ids = info->ids;
3893 	head->shared = info->is_shared;
3894 	head->rotational = info->is_rotational;
3895 	ratelimit_state_init(&head->rs_nuse, 5 * HZ, 1);
3896 	ratelimit_set_flags(&head->rs_nuse, RATELIMIT_MSG_ON_RELEASE);
3897 	kref_init(&head->ref);
3898 
3899 	if (head->ids.csi) {
3900 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3901 		if (ret)
3902 			goto out_cleanup_srcu;
3903 	} else
3904 		head->effects = ctrl->effects;
3905 
3906 	ret = nvme_mpath_alloc_disk(ctrl, head);
3907 	if (ret)
3908 		goto out_cleanup_srcu;
3909 
3910 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3911 
3912 	kref_get(&ctrl->subsys->ref);
3913 
3914 	return head;
3915 out_cleanup_srcu:
3916 	cleanup_srcu_struct(&head->srcu);
3917 out_ida_remove:
3918 	ida_free(&ctrl->subsys->ns_ida, head->instance);
3919 out_free_head:
3920 	kfree(head);
3921 out:
3922 	if (ret > 0)
3923 		ret = blk_status_to_errno(nvme_error_status(ret));
3924 	return ERR_PTR(ret);
3925 }
3926 
3927 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3928 		struct nvme_ns_ids *ids)
3929 {
3930 	struct nvme_subsystem *s;
3931 	int ret = 0;
3932 
3933 	/*
3934 	 * Note that this check is racy as we try to avoid holding the global
3935 	 * lock over the whole ns_head creation.  But it is only intended as
3936 	 * a sanity check anyway.
3937 	 */
3938 	mutex_lock(&nvme_subsystems_lock);
3939 	list_for_each_entry(s, &nvme_subsystems, entry) {
3940 		if (s == this)
3941 			continue;
3942 		mutex_lock(&s->lock);
3943 		ret = nvme_subsys_check_duplicate_ids(s, ids);
3944 		mutex_unlock(&s->lock);
3945 		if (ret)
3946 			break;
3947 	}
3948 	mutex_unlock(&nvme_subsystems_lock);
3949 
3950 	return ret;
3951 }
3952 
3953 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
3954 {
3955 	struct nvme_ctrl *ctrl = ns->ctrl;
3956 	struct nvme_ns_head *head = NULL;
3957 	int ret;
3958 
3959 	ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
3960 	if (ret) {
3961 		/*
3962 		 * We've found two different namespaces on two different
3963 		 * subsystems that report the same ID.  This is pretty nasty
3964 		 * for anything that actually requires unique device
3965 		 * identification.  In the kernel we need this for multipathing,
3966 		 * and in user space the /dev/disk/by-id/ links rely on it.
3967 		 *
3968 		 * If the device also claims to be multi-path capable back off
3969 		 * here now and refuse the probe the second device as this is a
3970 		 * recipe for data corruption.  If not this is probably a
3971 		 * cheap consumer device if on the PCIe bus, so let the user
3972 		 * proceed and use the shiny toy, but warn that with changing
3973 		 * probing order (which due to our async probing could just be
3974 		 * device taking longer to startup) the other device could show
3975 		 * up at any time.
3976 		 */
3977 		nvme_print_device_info(ctrl);
3978 		if ((ns->ctrl->ops->flags & NVME_F_FABRICS) || /* !PCIe */
3979 		    ((ns->ctrl->subsys->cmic & NVME_CTRL_CMIC_MULTI_CTRL) &&
3980 		     info->is_shared)) {
3981 			dev_err(ctrl->device,
3982 				"ignoring nsid %d because of duplicate IDs\n",
3983 				info->nsid);
3984 			return ret;
3985 		}
3986 
3987 		dev_err(ctrl->device,
3988 			"clearing duplicate IDs for nsid %d\n", info->nsid);
3989 		dev_err(ctrl->device,
3990 			"use of /dev/disk/by-id/ may cause data corruption\n");
3991 		memset(&info->ids.nguid, 0, sizeof(info->ids.nguid));
3992 		memset(&info->ids.uuid, 0, sizeof(info->ids.uuid));
3993 		memset(&info->ids.eui64, 0, sizeof(info->ids.eui64));
3994 		ctrl->quirks |= NVME_QUIRK_BOGUS_NID;
3995 	}
3996 
3997 	mutex_lock(&ctrl->subsys->lock);
3998 	head = nvme_find_ns_head(ctrl, info->nsid);
3999 	if (!head) {
4000 		ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
4001 		if (ret) {
4002 			dev_err(ctrl->device,
4003 				"duplicate IDs in subsystem for nsid %d\n",
4004 				info->nsid);
4005 			goto out_unlock;
4006 		}
4007 		head = nvme_alloc_ns_head(ctrl, info);
4008 		if (IS_ERR(head)) {
4009 			ret = PTR_ERR(head);
4010 			goto out_unlock;
4011 		}
4012 	} else {
4013 		ret = -EINVAL;
4014 		if ((!info->is_shared || !head->shared) &&
4015 		    !list_empty(&head->list)) {
4016 			dev_err(ctrl->device,
4017 				"Duplicate unshared namespace %d\n",
4018 				info->nsid);
4019 			goto out_put_ns_head;
4020 		}
4021 		if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
4022 			dev_err(ctrl->device,
4023 				"IDs don't match for shared namespace %d\n",
4024 					info->nsid);
4025 			goto out_put_ns_head;
4026 		}
4027 
4028 		if (!multipath) {
4029 			dev_warn(ctrl->device,
4030 				"Found shared namespace %d, but multipathing not supported.\n",
4031 				info->nsid);
4032 			dev_warn_once(ctrl->device,
4033 				"Shared namespace support requires core_nvme.multipath=Y.\n");
4034 		}
4035 	}
4036 
4037 	list_add_tail_rcu(&ns->siblings, &head->list);
4038 	ns->head = head;
4039 	mutex_unlock(&ctrl->subsys->lock);
4040 	return 0;
4041 
4042 out_put_ns_head:
4043 	nvme_put_ns_head(head);
4044 out_unlock:
4045 	mutex_unlock(&ctrl->subsys->lock);
4046 	return ret;
4047 }
4048 
4049 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4050 {
4051 	struct nvme_ns *ns, *ret = NULL;
4052 	int srcu_idx;
4053 
4054 	srcu_idx = srcu_read_lock(&ctrl->srcu);
4055 	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
4056 				 srcu_read_lock_held(&ctrl->srcu)) {
4057 		if (ns->head->ns_id == nsid) {
4058 			if (!nvme_get_ns(ns))
4059 				continue;
4060 			ret = ns;
4061 			break;
4062 		}
4063 		if (ns->head->ns_id > nsid)
4064 			break;
4065 	}
4066 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
4067 	return ret;
4068 }
4069 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, "NVME_TARGET_PASSTHRU");
4070 
4071 /*
4072  * Add the namespace to the controller list while keeping the list ordered.
4073  */
4074 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
4075 {
4076 	struct nvme_ns *tmp;
4077 
4078 	list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
4079 		if (tmp->head->ns_id < ns->head->ns_id) {
4080 			list_add_rcu(&ns->list, &tmp->list);
4081 			return;
4082 		}
4083 	}
4084 	list_add(&ns->list, &ns->ctrl->namespaces);
4085 }
4086 
4087 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
4088 {
4089 	struct queue_limits lim = { };
4090 	struct nvme_ns *ns;
4091 	struct gendisk *disk;
4092 	int node = ctrl->numa_node;
4093 
4094 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
4095 	if (!ns)
4096 		return;
4097 
4098 	if (ctrl->opts && ctrl->opts->data_digest)
4099 		lim.features |= BLK_FEAT_STABLE_WRITES;
4100 	if (ctrl->ops->supports_pci_p2pdma &&
4101 	    ctrl->ops->supports_pci_p2pdma(ctrl))
4102 		lim.features |= BLK_FEAT_PCI_P2PDMA;
4103 
4104 	disk = blk_mq_alloc_disk(ctrl->tagset, &lim, ns);
4105 	if (IS_ERR(disk))
4106 		goto out_free_ns;
4107 	disk->fops = &nvme_bdev_ops;
4108 	disk->private_data = ns;
4109 
4110 	ns->disk = disk;
4111 	ns->queue = disk->queue;
4112 	ns->ctrl = ctrl;
4113 	kref_init(&ns->kref);
4114 
4115 	if (nvme_init_ns_head(ns, info))
4116 		goto out_cleanup_disk;
4117 
4118 	/*
4119 	 * If multipathing is enabled, the device name for all disks and not
4120 	 * just those that represent shared namespaces needs to be based on the
4121 	 * subsystem instance.  Using the controller instance for private
4122 	 * namespaces could lead to naming collisions between shared and private
4123 	 * namespaces if they don't use a common numbering scheme.
4124 	 *
4125 	 * If multipathing is not enabled, disk names must use the controller
4126 	 * instance as shared namespaces will show up as multiple block
4127 	 * devices.
4128 	 */
4129 	if (nvme_ns_head_multipath(ns->head)) {
4130 		sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
4131 			ctrl->instance, ns->head->instance);
4132 		disk->flags |= GENHD_FL_HIDDEN;
4133 	} else if (multipath) {
4134 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
4135 			ns->head->instance);
4136 	} else {
4137 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
4138 			ns->head->instance);
4139 	}
4140 
4141 	if (nvme_update_ns_info(ns, info))
4142 		goto out_unlink_ns;
4143 
4144 	mutex_lock(&ctrl->namespaces_lock);
4145 	/*
4146 	 * Ensure that no namespaces are added to the ctrl list after the queues
4147 	 * are frozen, thereby avoiding a deadlock between scan and reset.
4148 	 */
4149 	if (test_bit(NVME_CTRL_FROZEN, &ctrl->flags)) {
4150 		mutex_unlock(&ctrl->namespaces_lock);
4151 		goto out_unlink_ns;
4152 	}
4153 	nvme_ns_add_to_ctrl_list(ns);
4154 	mutex_unlock(&ctrl->namespaces_lock);
4155 	synchronize_srcu(&ctrl->srcu);
4156 	nvme_get_ctrl(ctrl);
4157 
4158 	if (device_add_disk(ctrl->device, ns->disk, nvme_ns_attr_groups))
4159 		goto out_cleanup_ns_from_list;
4160 
4161 	if (!nvme_ns_head_multipath(ns->head))
4162 		nvme_add_ns_cdev(ns);
4163 
4164 	nvme_mpath_add_disk(ns, info->anagrpid);
4165 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
4166 
4167 	/*
4168 	 * Set ns->disk->device->driver_data to ns so we can access
4169 	 * ns->head->passthru_err_log_enabled in
4170 	 * nvme_io_passthru_err_log_enabled_[store | show]().
4171 	 */
4172 	dev_set_drvdata(disk_to_dev(ns->disk), ns);
4173 
4174 	return;
4175 
4176  out_cleanup_ns_from_list:
4177 	nvme_put_ctrl(ctrl);
4178 	mutex_lock(&ctrl->namespaces_lock);
4179 	list_del_rcu(&ns->list);
4180 	mutex_unlock(&ctrl->namespaces_lock);
4181 	synchronize_srcu(&ctrl->srcu);
4182  out_unlink_ns:
4183 	mutex_lock(&ctrl->subsys->lock);
4184 	list_del_rcu(&ns->siblings);
4185 	if (list_empty(&ns->head->list))
4186 		list_del_init(&ns->head->entry);
4187 	mutex_unlock(&ctrl->subsys->lock);
4188 	nvme_put_ns_head(ns->head);
4189  out_cleanup_disk:
4190 	put_disk(disk);
4191  out_free_ns:
4192 	kfree(ns);
4193 }
4194 
4195 static void nvme_ns_remove(struct nvme_ns *ns)
4196 {
4197 	bool last_path = false;
4198 
4199 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4200 		return;
4201 
4202 	clear_bit(NVME_NS_READY, &ns->flags);
4203 	set_capacity(ns->disk, 0);
4204 	nvme_fault_inject_fini(&ns->fault_inject);
4205 
4206 	/*
4207 	 * Ensure that !NVME_NS_READY is seen by other threads to prevent
4208 	 * this ns going back into current_path.
4209 	 */
4210 	synchronize_srcu(&ns->head->srcu);
4211 
4212 	/* wait for concurrent submissions */
4213 	if (nvme_mpath_clear_current_path(ns))
4214 		synchronize_srcu(&ns->head->srcu);
4215 
4216 	mutex_lock(&ns->ctrl->subsys->lock);
4217 	list_del_rcu(&ns->siblings);
4218 	if (list_empty(&ns->head->list)) {
4219 		if (!nvme_mpath_queue_if_no_path(ns->head))
4220 			list_del_init(&ns->head->entry);
4221 		last_path = true;
4222 	}
4223 	mutex_unlock(&ns->ctrl->subsys->lock);
4224 
4225 	/* guarantee not available in head->list */
4226 	synchronize_srcu(&ns->head->srcu);
4227 
4228 	if (!nvme_ns_head_multipath(ns->head))
4229 		nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4230 
4231 	nvme_mpath_remove_sysfs_link(ns);
4232 
4233 	del_gendisk(ns->disk);
4234 
4235 	mutex_lock(&ns->ctrl->namespaces_lock);
4236 	list_del_rcu(&ns->list);
4237 	mutex_unlock(&ns->ctrl->namespaces_lock);
4238 	synchronize_srcu(&ns->ctrl->srcu);
4239 
4240 	if (last_path)
4241 		nvme_mpath_remove_disk(ns->head);
4242 	nvme_put_ns(ns);
4243 }
4244 
4245 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4246 {
4247 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4248 
4249 	if (ns) {
4250 		nvme_ns_remove(ns);
4251 		nvme_put_ns(ns);
4252 	}
4253 }
4254 
4255 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
4256 {
4257 	int ret = NVME_SC_INVALID_NS | NVME_STATUS_DNR;
4258 
4259 	if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
4260 		dev_err(ns->ctrl->device,
4261 			"identifiers changed for nsid %d\n", ns->head->ns_id);
4262 		goto out;
4263 	}
4264 
4265 	ret = nvme_update_ns_info(ns, info);
4266 out:
4267 	/*
4268 	 * Only remove the namespace if we got a fatal error back from the
4269 	 * device, otherwise ignore the error and just move on.
4270 	 *
4271 	 * TODO: we should probably schedule a delayed retry here.
4272 	 */
4273 	if (ret > 0 && (ret & NVME_STATUS_DNR))
4274 		nvme_ns_remove(ns);
4275 }
4276 
4277 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4278 {
4279 	struct nvme_ns_info info = { .nsid = nsid };
4280 	struct nvme_ns *ns;
4281 	int ret = 1;
4282 
4283 	if (nvme_identify_ns_descs(ctrl, &info))
4284 		return;
4285 
4286 	if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
4287 		dev_warn(ctrl->device,
4288 			"command set not reported for nsid: %d\n", nsid);
4289 		return;
4290 	}
4291 
4292 	/*
4293 	 * If available try to use the Command Set Idependent Identify Namespace
4294 	 * data structure to find all the generic information that is needed to
4295 	 * set up a namespace.  If not fall back to the legacy version.
4296 	 */
4297 	if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
4298 	    (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS) ||
4299 	    ctrl->vs >= NVME_VS(2, 0, 0))
4300 		ret = nvme_ns_info_from_id_cs_indep(ctrl, &info);
4301 	if (ret > 0)
4302 		ret = nvme_ns_info_from_identify(ctrl, &info);
4303 
4304 	if (info.is_removed)
4305 		nvme_ns_remove_by_nsid(ctrl, nsid);
4306 
4307 	/*
4308 	 * Ignore the namespace if it is not ready. We will get an AEN once it
4309 	 * becomes ready and restart the scan.
4310 	 */
4311 	if (ret || !info.is_ready)
4312 		return;
4313 
4314 	ns = nvme_find_get_ns(ctrl, nsid);
4315 	if (ns) {
4316 		nvme_validate_ns(ns, &info);
4317 		nvme_put_ns(ns);
4318 	} else {
4319 		nvme_alloc_ns(ctrl, &info);
4320 	}
4321 }
4322 
4323 /**
4324  * struct async_scan_info - keeps track of controller & NSIDs to scan
4325  * @ctrl:	Controller on which namespaces are being scanned
4326  * @next_nsid:	Index of next NSID to scan in ns_list
4327  * @ns_list:	Pointer to list of NSIDs to scan
4328  *
4329  * Note: There is a single async_scan_info structure shared by all instances
4330  * of nvme_scan_ns_async() scanning a given controller, so the atomic
4331  * operations on next_nsid are critical to ensure each instance scans a unique
4332  * NSID.
4333  */
4334 struct async_scan_info {
4335 	struct nvme_ctrl *ctrl;
4336 	atomic_t next_nsid;
4337 	__le32 *ns_list;
4338 };
4339 
4340 static void nvme_scan_ns_async(void *data, async_cookie_t cookie)
4341 {
4342 	struct async_scan_info *scan_info = data;
4343 	int idx;
4344 	u32 nsid;
4345 
4346 	idx = (u32)atomic_fetch_inc(&scan_info->next_nsid);
4347 	nsid = le32_to_cpu(scan_info->ns_list[idx]);
4348 
4349 	nvme_scan_ns(scan_info->ctrl, nsid);
4350 }
4351 
4352 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4353 					unsigned nsid)
4354 {
4355 	struct nvme_ns *ns, *next;
4356 	LIST_HEAD(rm_list);
4357 
4358 	mutex_lock(&ctrl->namespaces_lock);
4359 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4360 		if (ns->head->ns_id > nsid) {
4361 			list_del_rcu(&ns->list);
4362 			synchronize_srcu(&ctrl->srcu);
4363 			list_add_tail_rcu(&ns->list, &rm_list);
4364 		}
4365 	}
4366 	mutex_unlock(&ctrl->namespaces_lock);
4367 
4368 	list_for_each_entry_safe(ns, next, &rm_list, list)
4369 		nvme_ns_remove(ns);
4370 }
4371 
4372 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4373 {
4374 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4375 	__le32 *ns_list;
4376 	u32 prev = 0;
4377 	int ret = 0, i;
4378 	ASYNC_DOMAIN(domain);
4379 	struct async_scan_info scan_info;
4380 
4381 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4382 	if (!ns_list)
4383 		return -ENOMEM;
4384 
4385 	scan_info.ctrl = ctrl;
4386 	scan_info.ns_list = ns_list;
4387 	for (;;) {
4388 		struct nvme_command cmd = {
4389 			.identify.opcode	= nvme_admin_identify,
4390 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
4391 			.identify.nsid		= cpu_to_le32(prev),
4392 		};
4393 
4394 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4395 					    NVME_IDENTIFY_DATA_SIZE);
4396 		if (ret) {
4397 			dev_warn(ctrl->device,
4398 				"Identify NS List failed (status=0x%x)\n", ret);
4399 			goto free;
4400 		}
4401 
4402 		atomic_set(&scan_info.next_nsid, 0);
4403 		for (i = 0; i < nr_entries; i++) {
4404 			u32 nsid = le32_to_cpu(ns_list[i]);
4405 
4406 			if (!nsid)	/* end of the list? */
4407 				goto out;
4408 			async_schedule_domain(nvme_scan_ns_async, &scan_info,
4409 						&domain);
4410 			while (++prev < nsid)
4411 				nvme_ns_remove_by_nsid(ctrl, prev);
4412 		}
4413 		async_synchronize_full_domain(&domain);
4414 	}
4415  out:
4416 	nvme_remove_invalid_namespaces(ctrl, prev);
4417  free:
4418 	async_synchronize_full_domain(&domain);
4419 	kfree(ns_list);
4420 	return ret;
4421 }
4422 
4423 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4424 {
4425 	struct nvme_id_ctrl *id;
4426 	u32 nn, i;
4427 
4428 	if (nvme_identify_ctrl(ctrl, &id))
4429 		return;
4430 	nn = le32_to_cpu(id->nn);
4431 	kfree(id);
4432 
4433 	for (i = 1; i <= nn; i++)
4434 		nvme_scan_ns(ctrl, i);
4435 
4436 	nvme_remove_invalid_namespaces(ctrl, nn);
4437 }
4438 
4439 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4440 {
4441 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4442 	__le32 *log;
4443 	int error;
4444 
4445 	log = kzalloc(log_size, GFP_KERNEL);
4446 	if (!log)
4447 		return;
4448 
4449 	/*
4450 	 * We need to read the log to clear the AEN, but we don't want to rely
4451 	 * on it for the changed namespace information as userspace could have
4452 	 * raced with us in reading the log page, which could cause us to miss
4453 	 * updates.
4454 	 */
4455 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4456 			NVME_CSI_NVM, log, log_size, 0);
4457 	if (error)
4458 		dev_warn(ctrl->device,
4459 			"reading changed ns log failed: %d\n", error);
4460 
4461 	kfree(log);
4462 }
4463 
4464 static void nvme_scan_work(struct work_struct *work)
4465 {
4466 	struct nvme_ctrl *ctrl =
4467 		container_of(work, struct nvme_ctrl, scan_work);
4468 	int ret;
4469 
4470 	/* No tagset on a live ctrl means IO queues could not created */
4471 	if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE || !ctrl->tagset)
4472 		return;
4473 
4474 	/*
4475 	 * Identify controller limits can change at controller reset due to
4476 	 * new firmware download, even though it is not common we cannot ignore
4477 	 * such scenario. Controller's non-mdts limits are reported in the unit
4478 	 * of logical blocks that is dependent on the format of attached
4479 	 * namespace. Hence re-read the limits at the time of ns allocation.
4480 	 */
4481 	ret = nvme_init_non_mdts_limits(ctrl);
4482 	if (ret < 0) {
4483 		dev_warn(ctrl->device,
4484 			"reading non-mdts-limits failed: %d\n", ret);
4485 		return;
4486 	}
4487 
4488 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4489 		dev_info(ctrl->device, "rescanning namespaces.\n");
4490 		nvme_clear_changed_ns_log(ctrl);
4491 	}
4492 
4493 	mutex_lock(&ctrl->scan_lock);
4494 	if (!nvme_id_cns_ok(ctrl, NVME_ID_CNS_NS_ACTIVE_LIST)) {
4495 		nvme_scan_ns_sequential(ctrl);
4496 	} else {
4497 		/*
4498 		 * Fall back to sequential scan if DNR is set to handle broken
4499 		 * devices which should support Identify NS List (as per the VS
4500 		 * they report) but don't actually support it.
4501 		 */
4502 		ret = nvme_scan_ns_list(ctrl);
4503 		if (ret > 0 && ret & NVME_STATUS_DNR)
4504 			nvme_scan_ns_sequential(ctrl);
4505 	}
4506 	mutex_unlock(&ctrl->scan_lock);
4507 
4508 	/* Requeue if we have missed AENs */
4509 	if (test_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events))
4510 		nvme_queue_scan(ctrl);
4511 #ifdef CONFIG_NVME_MULTIPATH
4512 	else if (ctrl->ana_log_buf)
4513 		/* Re-read the ANA log page to not miss updates */
4514 		queue_work(nvme_wq, &ctrl->ana_work);
4515 #endif
4516 }
4517 
4518 /*
4519  * This function iterates the namespace list unlocked to allow recovery from
4520  * controller failure. It is up to the caller to ensure the namespace list is
4521  * not modified by scan work while this function is executing.
4522  */
4523 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4524 {
4525 	struct nvme_ns *ns, *next;
4526 	LIST_HEAD(ns_list);
4527 
4528 	/*
4529 	 * make sure to requeue I/O to all namespaces as these
4530 	 * might result from the scan itself and must complete
4531 	 * for the scan_work to make progress
4532 	 */
4533 	nvme_mpath_clear_ctrl_paths(ctrl);
4534 
4535 	/*
4536 	 * Unquiesce io queues so any pending IO won't hang, especially
4537 	 * those submitted from scan work
4538 	 */
4539 	nvme_unquiesce_io_queues(ctrl);
4540 
4541 	/* prevent racing with ns scanning */
4542 	flush_work(&ctrl->scan_work);
4543 
4544 	/*
4545 	 * The dead states indicates the controller was not gracefully
4546 	 * disconnected. In that case, we won't be able to flush any data while
4547 	 * removing the namespaces' disks; fail all the queues now to avoid
4548 	 * potentially having to clean up the failed sync later.
4549 	 */
4550 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_DEAD)
4551 		nvme_mark_namespaces_dead(ctrl);
4552 
4553 	/* this is a no-op when called from the controller reset handler */
4554 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4555 
4556 	mutex_lock(&ctrl->namespaces_lock);
4557 	list_splice_init_rcu(&ctrl->namespaces, &ns_list, synchronize_rcu);
4558 	mutex_unlock(&ctrl->namespaces_lock);
4559 	synchronize_srcu(&ctrl->srcu);
4560 
4561 	list_for_each_entry_safe(ns, next, &ns_list, list)
4562 		nvme_ns_remove(ns);
4563 }
4564 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4565 
4566 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env)
4567 {
4568 	const struct nvme_ctrl *ctrl =
4569 		container_of(dev, struct nvme_ctrl, ctrl_device);
4570 	struct nvmf_ctrl_options *opts = ctrl->opts;
4571 	int ret;
4572 
4573 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4574 	if (ret)
4575 		return ret;
4576 
4577 	if (opts) {
4578 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4579 		if (ret)
4580 			return ret;
4581 
4582 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4583 				opts->trsvcid ?: "none");
4584 		if (ret)
4585 			return ret;
4586 
4587 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4588 				opts->host_traddr ?: "none");
4589 		if (ret)
4590 			return ret;
4591 
4592 		ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4593 				opts->host_iface ?: "none");
4594 	}
4595 	return ret;
4596 }
4597 
4598 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4599 {
4600 	char *envp[2] = { envdata, NULL };
4601 
4602 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4603 }
4604 
4605 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4606 {
4607 	char *envp[2] = { NULL, NULL };
4608 	u32 aen_result = ctrl->aen_result;
4609 
4610 	ctrl->aen_result = 0;
4611 	if (!aen_result)
4612 		return;
4613 
4614 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4615 	if (!envp[0])
4616 		return;
4617 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4618 	kfree(envp[0]);
4619 }
4620 
4621 static void nvme_async_event_work(struct work_struct *work)
4622 {
4623 	struct nvme_ctrl *ctrl =
4624 		container_of(work, struct nvme_ctrl, async_event_work);
4625 
4626 	nvme_aen_uevent(ctrl);
4627 
4628 	/*
4629 	 * The transport drivers must guarantee AER submission here is safe by
4630 	 * flushing ctrl async_event_work after changing the controller state
4631 	 * from LIVE and before freeing the admin queue.
4632 	*/
4633 	if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
4634 		ctrl->ops->submit_async_event(ctrl);
4635 }
4636 
4637 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4638 {
4639 
4640 	u32 csts;
4641 
4642 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4643 		return false;
4644 
4645 	if (csts == ~0)
4646 		return false;
4647 
4648 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4649 }
4650 
4651 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4652 {
4653 	struct nvme_fw_slot_info_log *log;
4654 	u8 next_fw_slot, cur_fw_slot;
4655 
4656 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4657 	if (!log)
4658 		return;
4659 
4660 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4661 			 log, sizeof(*log), 0)) {
4662 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4663 		goto out_free_log;
4664 	}
4665 
4666 	cur_fw_slot = log->afi & 0x7;
4667 	next_fw_slot = (log->afi & 0x70) >> 4;
4668 	if (!cur_fw_slot || (next_fw_slot && (cur_fw_slot != next_fw_slot))) {
4669 		dev_info(ctrl->device,
4670 			 "Firmware is activated after next Controller Level Reset\n");
4671 		goto out_free_log;
4672 	}
4673 
4674 	memcpy(ctrl->subsys->firmware_rev, &log->frs[cur_fw_slot - 1],
4675 		sizeof(ctrl->subsys->firmware_rev));
4676 
4677 out_free_log:
4678 	kfree(log);
4679 }
4680 
4681 static void nvme_fw_act_work(struct work_struct *work)
4682 {
4683 	struct nvme_ctrl *ctrl = container_of(work,
4684 				struct nvme_ctrl, fw_act_work);
4685 	unsigned long fw_act_timeout;
4686 
4687 	nvme_auth_stop(ctrl);
4688 
4689 	if (ctrl->mtfa)
4690 		fw_act_timeout = jiffies + msecs_to_jiffies(ctrl->mtfa * 100);
4691 	else
4692 		fw_act_timeout = jiffies + secs_to_jiffies(admin_timeout);
4693 
4694 	nvme_quiesce_io_queues(ctrl);
4695 	while (nvme_ctrl_pp_status(ctrl)) {
4696 		if (time_after(jiffies, fw_act_timeout)) {
4697 			dev_warn(ctrl->device,
4698 				"Fw activation timeout, reset controller\n");
4699 			nvme_try_sched_reset(ctrl);
4700 			return;
4701 		}
4702 		msleep(100);
4703 	}
4704 
4705 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_CONNECTING) ||
4706 	    !nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4707 		return;
4708 
4709 	nvme_unquiesce_io_queues(ctrl);
4710 	/* read FW slot information to clear the AER */
4711 	nvme_get_fw_slot_info(ctrl);
4712 
4713 	queue_work(nvme_wq, &ctrl->async_event_work);
4714 }
4715 
4716 static u32 nvme_aer_type(u32 result)
4717 {
4718 	return result & 0x7;
4719 }
4720 
4721 static u32 nvme_aer_subtype(u32 result)
4722 {
4723 	return (result & 0xff00) >> 8;
4724 }
4725 
4726 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4727 {
4728 	u32 aer_notice_type = nvme_aer_subtype(result);
4729 	bool requeue = true;
4730 
4731 	switch (aer_notice_type) {
4732 	case NVME_AER_NOTICE_NS_CHANGED:
4733 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4734 		nvme_queue_scan(ctrl);
4735 		break;
4736 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4737 		/*
4738 		 * We are (ab)using the RESETTING state to prevent subsequent
4739 		 * recovery actions from interfering with the controller's
4740 		 * firmware activation.
4741 		 */
4742 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4743 			requeue = false;
4744 			queue_work(nvme_wq, &ctrl->fw_act_work);
4745 		}
4746 		break;
4747 #ifdef CONFIG_NVME_MULTIPATH
4748 	case NVME_AER_NOTICE_ANA:
4749 		if (!ctrl->ana_log_buf)
4750 			break;
4751 		queue_work(nvme_wq, &ctrl->ana_work);
4752 		break;
4753 #endif
4754 	case NVME_AER_NOTICE_DISC_CHANGED:
4755 		ctrl->aen_result = result;
4756 		break;
4757 	default:
4758 		dev_warn(ctrl->device, "async event result %08x\n", result);
4759 	}
4760 	return requeue;
4761 }
4762 
4763 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4764 {
4765 	dev_warn(ctrl->device,
4766 		"resetting controller due to persistent internal error\n");
4767 	nvme_reset_ctrl(ctrl);
4768 }
4769 
4770 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4771 		volatile union nvme_result *res)
4772 {
4773 	u32 result = le32_to_cpu(res->u32);
4774 	u32 aer_type = nvme_aer_type(result);
4775 	u32 aer_subtype = nvme_aer_subtype(result);
4776 	bool requeue = true;
4777 
4778 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4779 		return;
4780 
4781 	trace_nvme_async_event(ctrl, result);
4782 	switch (aer_type) {
4783 	case NVME_AER_NOTICE:
4784 		requeue = nvme_handle_aen_notice(ctrl, result);
4785 		break;
4786 	case NVME_AER_ERROR:
4787 		/*
4788 		 * For a persistent internal error, don't run async_event_work
4789 		 * to submit a new AER. The controller reset will do it.
4790 		 */
4791 		if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4792 			nvme_handle_aer_persistent_error(ctrl);
4793 			return;
4794 		}
4795 		fallthrough;
4796 	case NVME_AER_SMART:
4797 	case NVME_AER_CSS:
4798 	case NVME_AER_VS:
4799 		ctrl->aen_result = result;
4800 		break;
4801 	default:
4802 		break;
4803 	}
4804 
4805 	if (requeue)
4806 		queue_work(nvme_wq, &ctrl->async_event_work);
4807 }
4808 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4809 
4810 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4811 		const struct blk_mq_ops *ops, unsigned int cmd_size)
4812 {
4813 	struct queue_limits lim = {};
4814 	int ret;
4815 
4816 	memset(set, 0, sizeof(*set));
4817 	set->ops = ops;
4818 	set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4819 	if (ctrl->ops->flags & NVME_F_FABRICS)
4820 		/* Reserved for fabric connect and keep alive */
4821 		set->reserved_tags = 2;
4822 	set->numa_node = ctrl->numa_node;
4823 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4824 		set->flags |= BLK_MQ_F_BLOCKING;
4825 	set->cmd_size = cmd_size;
4826 	set->driver_data = ctrl;
4827 	set->nr_hw_queues = 1;
4828 	set->timeout = NVME_ADMIN_TIMEOUT;
4829 	ret = blk_mq_alloc_tag_set(set);
4830 	if (ret)
4831 		return ret;
4832 
4833 	ctrl->admin_q = blk_mq_alloc_queue(set, &lim, NULL);
4834 	if (IS_ERR(ctrl->admin_q)) {
4835 		ret = PTR_ERR(ctrl->admin_q);
4836 		goto out_free_tagset;
4837 	}
4838 
4839 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4840 		ctrl->fabrics_q = blk_mq_alloc_queue(set, NULL, NULL);
4841 		if (IS_ERR(ctrl->fabrics_q)) {
4842 			ret = PTR_ERR(ctrl->fabrics_q);
4843 			goto out_cleanup_admin_q;
4844 		}
4845 	}
4846 
4847 	ctrl->admin_tagset = set;
4848 	return 0;
4849 
4850 out_cleanup_admin_q:
4851 	blk_mq_destroy_queue(ctrl->admin_q);
4852 	blk_put_queue(ctrl->admin_q);
4853 out_free_tagset:
4854 	blk_mq_free_tag_set(set);
4855 	ctrl->admin_q = NULL;
4856 	ctrl->fabrics_q = NULL;
4857 	return ret;
4858 }
4859 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4860 
4861 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4862 {
4863 	/*
4864 	 * As we're about to destroy the queue and free tagset
4865 	 * we can not have keep-alive work running.
4866 	 */
4867 	nvme_stop_keep_alive(ctrl);
4868 	blk_mq_destroy_queue(ctrl->admin_q);
4869 	blk_put_queue(ctrl->admin_q);
4870 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4871 		blk_mq_destroy_queue(ctrl->fabrics_q);
4872 		blk_put_queue(ctrl->fabrics_q);
4873 	}
4874 	blk_mq_free_tag_set(ctrl->admin_tagset);
4875 }
4876 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4877 
4878 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4879 		const struct blk_mq_ops *ops, unsigned int nr_maps,
4880 		unsigned int cmd_size)
4881 {
4882 	int ret;
4883 
4884 	memset(set, 0, sizeof(*set));
4885 	set->ops = ops;
4886 	set->queue_depth = min_t(unsigned, ctrl->sqsize, BLK_MQ_MAX_DEPTH - 1);
4887 	/*
4888 	 * Some Apple controllers requires tags to be unique across admin and
4889 	 * the (only) I/O queue, so reserve the first 32 tags of the I/O queue.
4890 	 */
4891 	if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
4892 		set->reserved_tags = NVME_AQ_DEPTH;
4893 	else if (ctrl->ops->flags & NVME_F_FABRICS)
4894 		/* Reserved for fabric connect */
4895 		set->reserved_tags = 1;
4896 	set->numa_node = ctrl->numa_node;
4897 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4898 		set->flags |= BLK_MQ_F_BLOCKING;
4899 	set->cmd_size = cmd_size;
4900 	set->driver_data = ctrl;
4901 	set->nr_hw_queues = ctrl->queue_count - 1;
4902 	set->timeout = NVME_IO_TIMEOUT;
4903 	set->nr_maps = nr_maps;
4904 	ret = blk_mq_alloc_tag_set(set);
4905 	if (ret)
4906 		return ret;
4907 
4908 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4909 		struct queue_limits lim = {
4910 			.features	= BLK_FEAT_SKIP_TAGSET_QUIESCE,
4911 		};
4912 
4913 		ctrl->connect_q = blk_mq_alloc_queue(set, &lim, NULL);
4914         	if (IS_ERR(ctrl->connect_q)) {
4915 			ret = PTR_ERR(ctrl->connect_q);
4916 			goto out_free_tag_set;
4917 		}
4918 	}
4919 
4920 	ctrl->tagset = set;
4921 	return 0;
4922 
4923 out_free_tag_set:
4924 	blk_mq_free_tag_set(set);
4925 	ctrl->connect_q = NULL;
4926 	return ret;
4927 }
4928 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
4929 
4930 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
4931 {
4932 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4933 		blk_mq_destroy_queue(ctrl->connect_q);
4934 		blk_put_queue(ctrl->connect_q);
4935 	}
4936 	blk_mq_free_tag_set(ctrl->tagset);
4937 }
4938 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
4939 
4940 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4941 {
4942 	nvme_mpath_stop(ctrl);
4943 	nvme_auth_stop(ctrl);
4944 	nvme_stop_failfast_work(ctrl);
4945 	flush_work(&ctrl->async_event_work);
4946 	cancel_work_sync(&ctrl->fw_act_work);
4947 	if (ctrl->ops->stop_ctrl)
4948 		ctrl->ops->stop_ctrl(ctrl);
4949 }
4950 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4951 
4952 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4953 {
4954 	nvme_enable_aen(ctrl);
4955 
4956 	/*
4957 	 * persistent discovery controllers need to send indication to userspace
4958 	 * to re-read the discovery log page to learn about possible changes
4959 	 * that were missed. We identify persistent discovery controllers by
4960 	 * checking that they started once before, hence are reconnecting back.
4961 	 */
4962 	if (test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
4963 	    nvme_discovery_ctrl(ctrl))
4964 		nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
4965 
4966 	if (ctrl->queue_count > 1) {
4967 		nvme_queue_scan(ctrl);
4968 		nvme_unquiesce_io_queues(ctrl);
4969 		nvme_mpath_update(ctrl);
4970 	}
4971 
4972 	nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4973 	set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags);
4974 }
4975 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4976 
4977 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4978 {
4979 	nvme_stop_keep_alive(ctrl);
4980 	nvme_hwmon_exit(ctrl);
4981 	nvme_fault_inject_fini(&ctrl->fault_inject);
4982 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4983 	cdev_device_del(&ctrl->cdev, ctrl->device);
4984 	nvme_put_ctrl(ctrl);
4985 }
4986 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4987 
4988 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4989 {
4990 	struct nvme_effects_log	*cel;
4991 	unsigned long i;
4992 
4993 	xa_for_each(&ctrl->cels, i, cel) {
4994 		xa_erase(&ctrl->cels, i);
4995 		kfree(cel);
4996 	}
4997 
4998 	xa_destroy(&ctrl->cels);
4999 }
5000 
5001 static void nvme_free_ctrl(struct device *dev)
5002 {
5003 	struct nvme_ctrl *ctrl =
5004 		container_of(dev, struct nvme_ctrl, ctrl_device);
5005 	struct nvme_subsystem *subsys = ctrl->subsys;
5006 
5007 	if (!subsys || ctrl->instance != subsys->instance)
5008 		ida_free(&nvme_instance_ida, ctrl->instance);
5009 	nvme_free_cels(ctrl);
5010 	nvme_mpath_uninit(ctrl);
5011 	cleanup_srcu_struct(&ctrl->srcu);
5012 	nvme_auth_stop(ctrl);
5013 	nvme_auth_free(ctrl);
5014 	__free_page(ctrl->discard_page);
5015 	free_opal_dev(ctrl->opal_dev);
5016 
5017 	if (subsys) {
5018 		mutex_lock(&nvme_subsystems_lock);
5019 		list_del(&ctrl->subsys_entry);
5020 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
5021 		mutex_unlock(&nvme_subsystems_lock);
5022 	}
5023 
5024 	ctrl->ops->free_ctrl(ctrl);
5025 
5026 	if (subsys)
5027 		nvme_put_subsystem(subsys);
5028 }
5029 
5030 /*
5031  * Initialize a NVMe controller structures.  This needs to be called during
5032  * earliest initialization so that we have the initialized structured around
5033  * during probing.
5034  *
5035  * On success, the caller must use the nvme_put_ctrl() to release this when
5036  * needed, which also invokes the ops->free_ctrl() callback.
5037  */
5038 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
5039 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
5040 {
5041 	int ret;
5042 
5043 	WRITE_ONCE(ctrl->state, NVME_CTRL_NEW);
5044 	ctrl->passthru_err_log_enabled = false;
5045 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
5046 	spin_lock_init(&ctrl->lock);
5047 	mutex_init(&ctrl->namespaces_lock);
5048 
5049 	ret = init_srcu_struct(&ctrl->srcu);
5050 	if (ret)
5051 		return ret;
5052 
5053 	mutex_init(&ctrl->scan_lock);
5054 	INIT_LIST_HEAD(&ctrl->namespaces);
5055 	xa_init(&ctrl->cels);
5056 	ctrl->dev = dev;
5057 	ctrl->ops = ops;
5058 	ctrl->quirks = quirks;
5059 	ctrl->numa_node = NUMA_NO_NODE;
5060 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
5061 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
5062 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
5063 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
5064 	init_waitqueue_head(&ctrl->state_wq);
5065 
5066 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
5067 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
5068 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
5069 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
5070 	ctrl->ka_last_check_time = jiffies;
5071 
5072 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
5073 			PAGE_SIZE);
5074 	ctrl->discard_page = alloc_page(GFP_KERNEL);
5075 	if (!ctrl->discard_page) {
5076 		ret = -ENOMEM;
5077 		goto out;
5078 	}
5079 
5080 	ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
5081 	if (ret < 0)
5082 		goto out;
5083 	ctrl->instance = ret;
5084 
5085 	ret = nvme_auth_init_ctrl(ctrl);
5086 	if (ret)
5087 		goto out_release_instance;
5088 
5089 	nvme_mpath_init_ctrl(ctrl);
5090 
5091 	device_initialize(&ctrl->ctrl_device);
5092 	ctrl->device = &ctrl->ctrl_device;
5093 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
5094 			ctrl->instance);
5095 	ctrl->device->class = &nvme_class;
5096 	ctrl->device->parent = ctrl->dev;
5097 	if (ops->dev_attr_groups)
5098 		ctrl->device->groups = ops->dev_attr_groups;
5099 	else
5100 		ctrl->device->groups = nvme_dev_attr_groups;
5101 	ctrl->device->release = nvme_free_ctrl;
5102 	dev_set_drvdata(ctrl->device, ctrl);
5103 
5104 	return ret;
5105 
5106 out_release_instance:
5107 	ida_free(&nvme_instance_ida, ctrl->instance);
5108 out:
5109 	if (ctrl->discard_page)
5110 		__free_page(ctrl->discard_page);
5111 	cleanup_srcu_struct(&ctrl->srcu);
5112 	return ret;
5113 }
5114 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
5115 
5116 /*
5117  * On success, returns with an elevated controller reference and caller must
5118  * use nvme_uninit_ctrl() to properly free resources associated with the ctrl.
5119  */
5120 int nvme_add_ctrl(struct nvme_ctrl *ctrl)
5121 {
5122 	int ret;
5123 
5124 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
5125 	if (ret)
5126 		return ret;
5127 
5128 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
5129 	ctrl->cdev.owner = ctrl->ops->module;
5130 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
5131 	if (ret)
5132 		return ret;
5133 
5134 	/*
5135 	 * Initialize latency tolerance controls.  The sysfs files won't
5136 	 * be visible to userspace unless the device actually supports APST.
5137 	 */
5138 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
5139 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
5140 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
5141 
5142 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
5143 	nvme_get_ctrl(ctrl);
5144 
5145 	return 0;
5146 }
5147 EXPORT_SYMBOL_GPL(nvme_add_ctrl);
5148 
5149 /* let I/O to all namespaces fail in preparation for surprise removal */
5150 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
5151 {
5152 	struct nvme_ns *ns;
5153 	int srcu_idx;
5154 
5155 	srcu_idx = srcu_read_lock(&ctrl->srcu);
5156 	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
5157 				 srcu_read_lock_held(&ctrl->srcu))
5158 		blk_mark_disk_dead(ns->disk);
5159 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
5160 }
5161 EXPORT_SYMBOL_GPL(nvme_mark_namespaces_dead);
5162 
5163 void nvme_unfreeze(struct nvme_ctrl *ctrl)
5164 {
5165 	struct nvme_ns *ns;
5166 	int srcu_idx;
5167 
5168 	srcu_idx = srcu_read_lock(&ctrl->srcu);
5169 	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
5170 				 srcu_read_lock_held(&ctrl->srcu))
5171 		blk_mq_unfreeze_queue_non_owner(ns->queue);
5172 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
5173 	clear_bit(NVME_CTRL_FROZEN, &ctrl->flags);
5174 }
5175 EXPORT_SYMBOL_GPL(nvme_unfreeze);
5176 
5177 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
5178 {
5179 	struct nvme_ns *ns;
5180 	int srcu_idx;
5181 
5182 	srcu_idx = srcu_read_lock(&ctrl->srcu);
5183 	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
5184 				 srcu_read_lock_held(&ctrl->srcu)) {
5185 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
5186 		if (timeout <= 0)
5187 			break;
5188 	}
5189 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
5190 	return timeout;
5191 }
5192 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
5193 
5194 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
5195 {
5196 	struct nvme_ns *ns;
5197 	int srcu_idx;
5198 
5199 	srcu_idx = srcu_read_lock(&ctrl->srcu);
5200 	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
5201 				 srcu_read_lock_held(&ctrl->srcu))
5202 		blk_mq_freeze_queue_wait(ns->queue);
5203 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
5204 }
5205 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
5206 
5207 void nvme_start_freeze(struct nvme_ctrl *ctrl)
5208 {
5209 	struct nvme_ns *ns;
5210 	int srcu_idx;
5211 
5212 	set_bit(NVME_CTRL_FROZEN, &ctrl->flags);
5213 	srcu_idx = srcu_read_lock(&ctrl->srcu);
5214 	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
5215 				 srcu_read_lock_held(&ctrl->srcu))
5216 		/*
5217 		 * Typical non_owner use case is from pci driver, in which
5218 		 * start_freeze is called from timeout work function, but
5219 		 * unfreeze is done in reset work context
5220 		 */
5221 		blk_freeze_queue_start_non_owner(ns->queue);
5222 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
5223 }
5224 EXPORT_SYMBOL_GPL(nvme_start_freeze);
5225 
5226 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl)
5227 {
5228 	if (!ctrl->tagset)
5229 		return;
5230 	if (!test_and_set_bit(NVME_CTRL_STOPPED, &ctrl->flags))
5231 		blk_mq_quiesce_tagset(ctrl->tagset);
5232 	else
5233 		blk_mq_wait_quiesce_done(ctrl->tagset);
5234 }
5235 EXPORT_SYMBOL_GPL(nvme_quiesce_io_queues);
5236 
5237 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl)
5238 {
5239 	if (!ctrl->tagset)
5240 		return;
5241 	if (test_and_clear_bit(NVME_CTRL_STOPPED, &ctrl->flags))
5242 		blk_mq_unquiesce_tagset(ctrl->tagset);
5243 }
5244 EXPORT_SYMBOL_GPL(nvme_unquiesce_io_queues);
5245 
5246 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl)
5247 {
5248 	if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5249 		blk_mq_quiesce_queue(ctrl->admin_q);
5250 	else
5251 		blk_mq_wait_quiesce_done(ctrl->admin_q->tag_set);
5252 }
5253 EXPORT_SYMBOL_GPL(nvme_quiesce_admin_queue);
5254 
5255 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl)
5256 {
5257 	if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5258 		blk_mq_unquiesce_queue(ctrl->admin_q);
5259 }
5260 EXPORT_SYMBOL_GPL(nvme_unquiesce_admin_queue);
5261 
5262 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
5263 {
5264 	struct nvme_ns *ns;
5265 	int srcu_idx;
5266 
5267 	srcu_idx = srcu_read_lock(&ctrl->srcu);
5268 	list_for_each_entry_srcu(ns, &ctrl->namespaces, list,
5269 				 srcu_read_lock_held(&ctrl->srcu))
5270 		blk_sync_queue(ns->queue);
5271 	srcu_read_unlock(&ctrl->srcu, srcu_idx);
5272 }
5273 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
5274 
5275 void nvme_sync_queues(struct nvme_ctrl *ctrl)
5276 {
5277 	nvme_sync_io_queues(ctrl);
5278 	if (ctrl->admin_q)
5279 		blk_sync_queue(ctrl->admin_q);
5280 }
5281 EXPORT_SYMBOL_GPL(nvme_sync_queues);
5282 
5283 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
5284 {
5285 	if (file->f_op != &nvme_dev_fops)
5286 		return NULL;
5287 	return file->private_data;
5288 }
5289 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, "NVME_TARGET_PASSTHRU");
5290 
5291 /*
5292  * Check we didn't inadvertently grow the command structure sizes:
5293  */
5294 static inline void _nvme_check_size(void)
5295 {
5296 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
5297 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
5298 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
5299 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
5300 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
5301 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
5302 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
5303 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
5304 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
5305 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
5306 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
5307 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
5308 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
5309 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
5310 			NVME_IDENTIFY_DATA_SIZE);
5311 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
5312 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
5313 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
5314 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
5315 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
5316 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
5317 	BUILD_BUG_ON(sizeof(struct nvme_endurance_group_log) != 512);
5318 	BUILD_BUG_ON(sizeof(struct nvme_rotational_media_log) != 512);
5319 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
5320 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
5321 	BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
5322 }
5323 
5324 
5325 static int __init nvme_core_init(void)
5326 {
5327 	unsigned int wq_flags = WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS;
5328 	int result = -ENOMEM;
5329 
5330 	_nvme_check_size();
5331 
5332 	nvme_wq = alloc_workqueue("nvme-wq", wq_flags, 0);
5333 	if (!nvme_wq)
5334 		goto out;
5335 
5336 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq", wq_flags, 0);
5337 	if (!nvme_reset_wq)
5338 		goto destroy_wq;
5339 
5340 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq", wq_flags, 0);
5341 	if (!nvme_delete_wq)
5342 		goto destroy_reset_wq;
5343 
5344 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
5345 			NVME_MINORS, "nvme");
5346 	if (result < 0)
5347 		goto destroy_delete_wq;
5348 
5349 	result = class_register(&nvme_class);
5350 	if (result)
5351 		goto unregister_chrdev;
5352 
5353 	result = class_register(&nvme_subsys_class);
5354 	if (result)
5355 		goto destroy_class;
5356 
5357 	result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
5358 				     "nvme-generic");
5359 	if (result < 0)
5360 		goto destroy_subsys_class;
5361 
5362 	result = class_register(&nvme_ns_chr_class);
5363 	if (result)
5364 		goto unregister_generic_ns;
5365 
5366 	result = nvme_init_auth();
5367 	if (result)
5368 		goto destroy_ns_chr;
5369 	return 0;
5370 
5371 destroy_ns_chr:
5372 	class_unregister(&nvme_ns_chr_class);
5373 unregister_generic_ns:
5374 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5375 destroy_subsys_class:
5376 	class_unregister(&nvme_subsys_class);
5377 destroy_class:
5378 	class_unregister(&nvme_class);
5379 unregister_chrdev:
5380 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5381 destroy_delete_wq:
5382 	destroy_workqueue(nvme_delete_wq);
5383 destroy_reset_wq:
5384 	destroy_workqueue(nvme_reset_wq);
5385 destroy_wq:
5386 	destroy_workqueue(nvme_wq);
5387 out:
5388 	return result;
5389 }
5390 
5391 static void __exit nvme_core_exit(void)
5392 {
5393 	nvme_exit_auth();
5394 	class_unregister(&nvme_ns_chr_class);
5395 	class_unregister(&nvme_subsys_class);
5396 	class_unregister(&nvme_class);
5397 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5398 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5399 	destroy_workqueue(nvme_delete_wq);
5400 	destroy_workqueue(nvme_reset_wq);
5401 	destroy_workqueue(nvme_wq);
5402 	ida_destroy(&nvme_ns_chr_minor_ida);
5403 	ida_destroy(&nvme_instance_ida);
5404 }
5405 
5406 MODULE_LICENSE("GPL");
5407 MODULE_VERSION("1.0");
5408 MODULE_DESCRIPTION("NVMe host core framework");
5409 module_init(nvme_core_init);
5410 module_exit(nvme_core_exit);
5411