xref: /openbmc/linux/drivers/nvme/host/core.c (revision bd336e63)
1 /*
2  * NVM Express device driver
3  * Copyright (c) 2011-2014, Intel Corporation.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms and conditions of the GNU General Public License,
7  * version 2, as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12  * more details.
13  */
14 
15 #include <linux/blkdev.h>
16 #include <linux/blk-mq.h>
17 #include <linux/delay.h>
18 #include <linux/errno.h>
19 #include <linux/hdreg.h>
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/list_sort.h>
23 #include <linux/slab.h>
24 #include <linux/types.h>
25 #include <linux/pr.h>
26 #include <linux/ptrace.h>
27 #include <linux/nvme_ioctl.h>
28 #include <linux/t10-pi.h>
29 #include <scsi/sg.h>
30 #include <asm/unaligned.h>
31 
32 #include "nvme.h"
33 #include "fabrics.h"
34 
35 #define NVME_MINORS		(1U << MINORBITS)
36 
37 unsigned char admin_timeout = 60;
38 module_param(admin_timeout, byte, 0644);
39 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
40 EXPORT_SYMBOL_GPL(admin_timeout);
41 
42 unsigned char nvme_io_timeout = 30;
43 module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
44 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
45 EXPORT_SYMBOL_GPL(nvme_io_timeout);
46 
47 unsigned char shutdown_timeout = 5;
48 module_param(shutdown_timeout, byte, 0644);
49 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
50 
51 unsigned int nvme_max_retries = 5;
52 module_param_named(max_retries, nvme_max_retries, uint, 0644);
53 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
54 EXPORT_SYMBOL_GPL(nvme_max_retries);
55 
56 static int nvme_char_major;
57 module_param(nvme_char_major, int, 0);
58 
59 static LIST_HEAD(nvme_ctrl_list);
60 static DEFINE_SPINLOCK(dev_list_lock);
61 
62 static struct class *nvme_class;
63 
64 void nvme_cancel_request(struct request *req, void *data, bool reserved)
65 {
66 	int status;
67 
68 	if (!blk_mq_request_started(req))
69 		return;
70 
71 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
72 				"Cancelling I/O %d", req->tag);
73 
74 	status = NVME_SC_ABORT_REQ;
75 	if (blk_queue_dying(req->q))
76 		status |= NVME_SC_DNR;
77 	blk_mq_complete_request(req, status);
78 }
79 EXPORT_SYMBOL_GPL(nvme_cancel_request);
80 
81 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
82 		enum nvme_ctrl_state new_state)
83 {
84 	enum nvme_ctrl_state old_state;
85 	bool changed = false;
86 
87 	spin_lock_irq(&ctrl->lock);
88 
89 	old_state = ctrl->state;
90 	switch (new_state) {
91 	case NVME_CTRL_LIVE:
92 		switch (old_state) {
93 		case NVME_CTRL_NEW:
94 		case NVME_CTRL_RESETTING:
95 		case NVME_CTRL_RECONNECTING:
96 			changed = true;
97 			/* FALLTHRU */
98 		default:
99 			break;
100 		}
101 		break;
102 	case NVME_CTRL_RESETTING:
103 		switch (old_state) {
104 		case NVME_CTRL_NEW:
105 		case NVME_CTRL_LIVE:
106 		case NVME_CTRL_RECONNECTING:
107 			changed = true;
108 			/* FALLTHRU */
109 		default:
110 			break;
111 		}
112 		break;
113 	case NVME_CTRL_RECONNECTING:
114 		switch (old_state) {
115 		case NVME_CTRL_LIVE:
116 			changed = true;
117 			/* FALLTHRU */
118 		default:
119 			break;
120 		}
121 		break;
122 	case NVME_CTRL_DELETING:
123 		switch (old_state) {
124 		case NVME_CTRL_LIVE:
125 		case NVME_CTRL_RESETTING:
126 		case NVME_CTRL_RECONNECTING:
127 			changed = true;
128 			/* FALLTHRU */
129 		default:
130 			break;
131 		}
132 		break;
133 	case NVME_CTRL_DEAD:
134 		switch (old_state) {
135 		case NVME_CTRL_DELETING:
136 			changed = true;
137 			/* FALLTHRU */
138 		default:
139 			break;
140 		}
141 		break;
142 	default:
143 		break;
144 	}
145 
146 	if (changed)
147 		ctrl->state = new_state;
148 
149 	spin_unlock_irq(&ctrl->lock);
150 
151 	return changed;
152 }
153 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
154 
155 static void nvme_free_ns(struct kref *kref)
156 {
157 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
158 
159 	if (ns->ndev)
160 		nvme_nvm_unregister(ns);
161 
162 	if (ns->disk) {
163 		spin_lock(&dev_list_lock);
164 		ns->disk->private_data = NULL;
165 		spin_unlock(&dev_list_lock);
166 	}
167 
168 	put_disk(ns->disk);
169 	ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
170 	nvme_put_ctrl(ns->ctrl);
171 	kfree(ns);
172 }
173 
174 static void nvme_put_ns(struct nvme_ns *ns)
175 {
176 	kref_put(&ns->kref, nvme_free_ns);
177 }
178 
179 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
180 {
181 	struct nvme_ns *ns;
182 
183 	spin_lock(&dev_list_lock);
184 	ns = disk->private_data;
185 	if (ns) {
186 		if (!kref_get_unless_zero(&ns->kref))
187 			goto fail;
188 		if (!try_module_get(ns->ctrl->ops->module))
189 			goto fail_put_ns;
190 	}
191 	spin_unlock(&dev_list_lock);
192 
193 	return ns;
194 
195 fail_put_ns:
196 	kref_put(&ns->kref, nvme_free_ns);
197 fail:
198 	spin_unlock(&dev_list_lock);
199 	return NULL;
200 }
201 
202 void nvme_requeue_req(struct request *req)
203 {
204 	unsigned long flags;
205 
206 	blk_mq_requeue_request(req);
207 	spin_lock_irqsave(req->q->queue_lock, flags);
208 	if (!blk_queue_stopped(req->q))
209 		blk_mq_kick_requeue_list(req->q);
210 	spin_unlock_irqrestore(req->q->queue_lock, flags);
211 }
212 EXPORT_SYMBOL_GPL(nvme_requeue_req);
213 
214 struct request *nvme_alloc_request(struct request_queue *q,
215 		struct nvme_command *cmd, unsigned int flags, int qid)
216 {
217 	struct request *req;
218 
219 	if (qid == NVME_QID_ANY) {
220 		req = blk_mq_alloc_request(q, nvme_is_write(cmd), flags);
221 	} else {
222 		req = blk_mq_alloc_request_hctx(q, nvme_is_write(cmd), flags,
223 				qid ? qid - 1 : 0);
224 	}
225 	if (IS_ERR(req))
226 		return req;
227 
228 	req->cmd_type = REQ_TYPE_DRV_PRIV;
229 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
230 	req->cmd = (unsigned char *)cmd;
231 	req->cmd_len = sizeof(struct nvme_command);
232 
233 	return req;
234 }
235 EXPORT_SYMBOL_GPL(nvme_alloc_request);
236 
237 static inline void nvme_setup_flush(struct nvme_ns *ns,
238 		struct nvme_command *cmnd)
239 {
240 	memset(cmnd, 0, sizeof(*cmnd));
241 	cmnd->common.opcode = nvme_cmd_flush;
242 	cmnd->common.nsid = cpu_to_le32(ns->ns_id);
243 }
244 
245 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
246 		struct nvme_command *cmnd)
247 {
248 	struct nvme_dsm_range *range;
249 	struct page *page;
250 	int offset;
251 	unsigned int nr_bytes = blk_rq_bytes(req);
252 
253 	range = kmalloc(sizeof(*range), GFP_ATOMIC);
254 	if (!range)
255 		return BLK_MQ_RQ_QUEUE_BUSY;
256 
257 	range->cattr = cpu_to_le32(0);
258 	range->nlb = cpu_to_le32(nr_bytes >> ns->lba_shift);
259 	range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
260 
261 	memset(cmnd, 0, sizeof(*cmnd));
262 	cmnd->dsm.opcode = nvme_cmd_dsm;
263 	cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
264 	cmnd->dsm.nr = 0;
265 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
266 
267 	req->completion_data = range;
268 	page = virt_to_page(range);
269 	offset = offset_in_page(range);
270 	blk_add_request_payload(req, page, offset, sizeof(*range));
271 
272 	/*
273 	 * we set __data_len back to the size of the area to be discarded
274 	 * on disk. This allows us to report completion on the full amount
275 	 * of blocks described by the request.
276 	 */
277 	req->__data_len = nr_bytes;
278 
279 	return 0;
280 }
281 
282 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
283 		struct nvme_command *cmnd)
284 {
285 	u16 control = 0;
286 	u32 dsmgmt = 0;
287 
288 	if (req->cmd_flags & REQ_FUA)
289 		control |= NVME_RW_FUA;
290 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
291 		control |= NVME_RW_LR;
292 
293 	if (req->cmd_flags & REQ_RAHEAD)
294 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
295 
296 	memset(cmnd, 0, sizeof(*cmnd));
297 	cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
298 	cmnd->rw.command_id = req->tag;
299 	cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
300 	cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
301 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
302 
303 	if (ns->ms) {
304 		switch (ns->pi_type) {
305 		case NVME_NS_DPS_PI_TYPE3:
306 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
307 			break;
308 		case NVME_NS_DPS_PI_TYPE1:
309 		case NVME_NS_DPS_PI_TYPE2:
310 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
311 					NVME_RW_PRINFO_PRCHK_REF;
312 			cmnd->rw.reftag = cpu_to_le32(
313 					nvme_block_nr(ns, blk_rq_pos(req)));
314 			break;
315 		}
316 		if (!blk_integrity_rq(req))
317 			control |= NVME_RW_PRINFO_PRACT;
318 	}
319 
320 	cmnd->rw.control = cpu_to_le16(control);
321 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
322 }
323 
324 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
325 		struct nvme_command *cmd)
326 {
327 	int ret = 0;
328 
329 	if (req->cmd_type == REQ_TYPE_DRV_PRIV)
330 		memcpy(cmd, req->cmd, sizeof(*cmd));
331 	else if (req_op(req) == REQ_OP_FLUSH)
332 		nvme_setup_flush(ns, cmd);
333 	else if (req_op(req) == REQ_OP_DISCARD)
334 		ret = nvme_setup_discard(ns, req, cmd);
335 	else
336 		nvme_setup_rw(ns, req, cmd);
337 
338 	return ret;
339 }
340 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
341 
342 /*
343  * Returns 0 on success.  If the result is negative, it's a Linux error code;
344  * if the result is positive, it's an NVM Express status code
345  */
346 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
347 		struct nvme_completion *cqe, void *buffer, unsigned bufflen,
348 		unsigned timeout, int qid, int at_head, int flags)
349 {
350 	struct request *req;
351 	int ret;
352 
353 	req = nvme_alloc_request(q, cmd, flags, qid);
354 	if (IS_ERR(req))
355 		return PTR_ERR(req);
356 
357 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
358 	req->special = cqe;
359 
360 	if (buffer && bufflen) {
361 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
362 		if (ret)
363 			goto out;
364 	}
365 
366 	blk_execute_rq(req->q, NULL, req, at_head);
367 	ret = req->errors;
368  out:
369 	blk_mq_free_request(req);
370 	return ret;
371 }
372 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
373 
374 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
375 		void *buffer, unsigned bufflen)
376 {
377 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
378 			NVME_QID_ANY, 0, 0);
379 }
380 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
381 
382 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
383 		void __user *ubuffer, unsigned bufflen,
384 		void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
385 		u32 *result, unsigned timeout)
386 {
387 	bool write = nvme_is_write(cmd);
388 	struct nvme_completion cqe;
389 	struct nvme_ns *ns = q->queuedata;
390 	struct gendisk *disk = ns ? ns->disk : NULL;
391 	struct request *req;
392 	struct bio *bio = NULL;
393 	void *meta = NULL;
394 	int ret;
395 
396 	req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
397 	if (IS_ERR(req))
398 		return PTR_ERR(req);
399 
400 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
401 	req->special = &cqe;
402 
403 	if (ubuffer && bufflen) {
404 		ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
405 				GFP_KERNEL);
406 		if (ret)
407 			goto out;
408 		bio = req->bio;
409 
410 		if (!disk)
411 			goto submit;
412 		bio->bi_bdev = bdget_disk(disk, 0);
413 		if (!bio->bi_bdev) {
414 			ret = -ENODEV;
415 			goto out_unmap;
416 		}
417 
418 		if (meta_buffer && meta_len) {
419 			struct bio_integrity_payload *bip;
420 
421 			meta = kmalloc(meta_len, GFP_KERNEL);
422 			if (!meta) {
423 				ret = -ENOMEM;
424 				goto out_unmap;
425 			}
426 
427 			if (write) {
428 				if (copy_from_user(meta, meta_buffer,
429 						meta_len)) {
430 					ret = -EFAULT;
431 					goto out_free_meta;
432 				}
433 			}
434 
435 			bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
436 			if (IS_ERR(bip)) {
437 				ret = PTR_ERR(bip);
438 				goto out_free_meta;
439 			}
440 
441 			bip->bip_iter.bi_size = meta_len;
442 			bip->bip_iter.bi_sector = meta_seed;
443 
444 			ret = bio_integrity_add_page(bio, virt_to_page(meta),
445 					meta_len, offset_in_page(meta));
446 			if (ret != meta_len) {
447 				ret = -ENOMEM;
448 				goto out_free_meta;
449 			}
450 		}
451 	}
452  submit:
453 	blk_execute_rq(req->q, disk, req, 0);
454 	ret = req->errors;
455 	if (result)
456 		*result = le32_to_cpu(cqe.result);
457 	if (meta && !ret && !write) {
458 		if (copy_to_user(meta_buffer, meta, meta_len))
459 			ret = -EFAULT;
460 	}
461  out_free_meta:
462 	kfree(meta);
463  out_unmap:
464 	if (bio) {
465 		if (disk && bio->bi_bdev)
466 			bdput(bio->bi_bdev);
467 		blk_rq_unmap_user(bio);
468 	}
469  out:
470 	blk_mq_free_request(req);
471 	return ret;
472 }
473 
474 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
475 		void __user *ubuffer, unsigned bufflen, u32 *result,
476 		unsigned timeout)
477 {
478 	return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
479 			result, timeout);
480 }
481 
482 static void nvme_keep_alive_end_io(struct request *rq, int error)
483 {
484 	struct nvme_ctrl *ctrl = rq->end_io_data;
485 
486 	blk_mq_free_request(rq);
487 
488 	if (error) {
489 		dev_err(ctrl->device,
490 			"failed nvme_keep_alive_end_io error=%d\n", error);
491 		return;
492 	}
493 
494 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
495 }
496 
497 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
498 {
499 	struct nvme_command c;
500 	struct request *rq;
501 
502 	memset(&c, 0, sizeof(c));
503 	c.common.opcode = nvme_admin_keep_alive;
504 
505 	rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
506 			NVME_QID_ANY);
507 	if (IS_ERR(rq))
508 		return PTR_ERR(rq);
509 
510 	rq->timeout = ctrl->kato * HZ;
511 	rq->end_io_data = ctrl;
512 
513 	blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
514 
515 	return 0;
516 }
517 
518 static void nvme_keep_alive_work(struct work_struct *work)
519 {
520 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
521 			struct nvme_ctrl, ka_work);
522 
523 	if (nvme_keep_alive(ctrl)) {
524 		/* allocation failure, reset the controller */
525 		dev_err(ctrl->device, "keep-alive failed\n");
526 		ctrl->ops->reset_ctrl(ctrl);
527 		return;
528 	}
529 }
530 
531 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
532 {
533 	if (unlikely(ctrl->kato == 0))
534 		return;
535 
536 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
537 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
538 }
539 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
540 
541 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
542 {
543 	if (unlikely(ctrl->kato == 0))
544 		return;
545 
546 	cancel_delayed_work_sync(&ctrl->ka_work);
547 }
548 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
549 
550 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
551 {
552 	struct nvme_command c = { };
553 	int error;
554 
555 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
556 	c.identify.opcode = nvme_admin_identify;
557 	c.identify.cns = cpu_to_le32(NVME_ID_CNS_CTRL);
558 
559 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
560 	if (!*id)
561 		return -ENOMEM;
562 
563 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
564 			sizeof(struct nvme_id_ctrl));
565 	if (error)
566 		kfree(*id);
567 	return error;
568 }
569 
570 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
571 {
572 	struct nvme_command c = { };
573 
574 	c.identify.opcode = nvme_admin_identify;
575 	c.identify.cns = cpu_to_le32(NVME_ID_CNS_NS_ACTIVE_LIST);
576 	c.identify.nsid = cpu_to_le32(nsid);
577 	return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
578 }
579 
580 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
581 		struct nvme_id_ns **id)
582 {
583 	struct nvme_command c = { };
584 	int error;
585 
586 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
587 	c.identify.opcode = nvme_admin_identify,
588 	c.identify.nsid = cpu_to_le32(nsid),
589 
590 	*id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
591 	if (!*id)
592 		return -ENOMEM;
593 
594 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
595 			sizeof(struct nvme_id_ns));
596 	if (error)
597 		kfree(*id);
598 	return error;
599 }
600 
601 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
602 		      void *buffer, size_t buflen, u32 *result)
603 {
604 	struct nvme_command c;
605 	struct nvme_completion cqe;
606 	int ret;
607 
608 	memset(&c, 0, sizeof(c));
609 	c.features.opcode = nvme_admin_get_features;
610 	c.features.nsid = cpu_to_le32(nsid);
611 	c.features.fid = cpu_to_le32(fid);
612 
613 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, buffer, buflen, 0,
614 			NVME_QID_ANY, 0, 0);
615 	if (ret >= 0 && result)
616 		*result = le32_to_cpu(cqe.result);
617 	return ret;
618 }
619 
620 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
621 		      void *buffer, size_t buflen, u32 *result)
622 {
623 	struct nvme_command c;
624 	struct nvme_completion cqe;
625 	int ret;
626 
627 	memset(&c, 0, sizeof(c));
628 	c.features.opcode = nvme_admin_set_features;
629 	c.features.fid = cpu_to_le32(fid);
630 	c.features.dword11 = cpu_to_le32(dword11);
631 
632 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe,
633 			buffer, buflen, 0, NVME_QID_ANY, 0, 0);
634 	if (ret >= 0 && result)
635 		*result = le32_to_cpu(cqe.result);
636 	return ret;
637 }
638 
639 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
640 {
641 	struct nvme_command c = { };
642 	int error;
643 
644 	c.common.opcode = nvme_admin_get_log_page,
645 	c.common.nsid = cpu_to_le32(0xFFFFFFFF),
646 	c.common.cdw10[0] = cpu_to_le32(
647 			(((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
648 			 NVME_LOG_SMART),
649 
650 	*log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
651 	if (!*log)
652 		return -ENOMEM;
653 
654 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
655 			sizeof(struct nvme_smart_log));
656 	if (error)
657 		kfree(*log);
658 	return error;
659 }
660 
661 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
662 {
663 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
664 	u32 result;
665 	int status, nr_io_queues;
666 
667 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
668 			&result);
669 	if (status < 0)
670 		return status;
671 
672 	/*
673 	 * Degraded controllers might return an error when setting the queue
674 	 * count.  We still want to be able to bring them online and offer
675 	 * access to the admin queue, as that might be only way to fix them up.
676 	 */
677 	if (status > 0) {
678 		dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
679 		*count = 0;
680 	} else {
681 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
682 		*count = min(*count, nr_io_queues);
683 	}
684 
685 	return 0;
686 }
687 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
688 
689 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
690 {
691 	struct nvme_user_io io;
692 	struct nvme_command c;
693 	unsigned length, meta_len;
694 	void __user *metadata;
695 
696 	if (copy_from_user(&io, uio, sizeof(io)))
697 		return -EFAULT;
698 	if (io.flags)
699 		return -EINVAL;
700 
701 	switch (io.opcode) {
702 	case nvme_cmd_write:
703 	case nvme_cmd_read:
704 	case nvme_cmd_compare:
705 		break;
706 	default:
707 		return -EINVAL;
708 	}
709 
710 	length = (io.nblocks + 1) << ns->lba_shift;
711 	meta_len = (io.nblocks + 1) * ns->ms;
712 	metadata = (void __user *)(uintptr_t)io.metadata;
713 
714 	if (ns->ext) {
715 		length += meta_len;
716 		meta_len = 0;
717 	} else if (meta_len) {
718 		if ((io.metadata & 3) || !io.metadata)
719 			return -EINVAL;
720 	}
721 
722 	memset(&c, 0, sizeof(c));
723 	c.rw.opcode = io.opcode;
724 	c.rw.flags = io.flags;
725 	c.rw.nsid = cpu_to_le32(ns->ns_id);
726 	c.rw.slba = cpu_to_le64(io.slba);
727 	c.rw.length = cpu_to_le16(io.nblocks);
728 	c.rw.control = cpu_to_le16(io.control);
729 	c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
730 	c.rw.reftag = cpu_to_le32(io.reftag);
731 	c.rw.apptag = cpu_to_le16(io.apptag);
732 	c.rw.appmask = cpu_to_le16(io.appmask);
733 
734 	return __nvme_submit_user_cmd(ns->queue, &c,
735 			(void __user *)(uintptr_t)io.addr, length,
736 			metadata, meta_len, io.slba, NULL, 0);
737 }
738 
739 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
740 			struct nvme_passthru_cmd __user *ucmd)
741 {
742 	struct nvme_passthru_cmd cmd;
743 	struct nvme_command c;
744 	unsigned timeout = 0;
745 	int status;
746 
747 	if (!capable(CAP_SYS_ADMIN))
748 		return -EACCES;
749 	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
750 		return -EFAULT;
751 	if (cmd.flags)
752 		return -EINVAL;
753 
754 	memset(&c, 0, sizeof(c));
755 	c.common.opcode = cmd.opcode;
756 	c.common.flags = cmd.flags;
757 	c.common.nsid = cpu_to_le32(cmd.nsid);
758 	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
759 	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
760 	c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
761 	c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
762 	c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
763 	c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
764 	c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
765 	c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
766 
767 	if (cmd.timeout_ms)
768 		timeout = msecs_to_jiffies(cmd.timeout_ms);
769 
770 	status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
771 			(void __user *)(uintptr_t)cmd.addr, cmd.data_len,
772 			&cmd.result, timeout);
773 	if (status >= 0) {
774 		if (put_user(cmd.result, &ucmd->result))
775 			return -EFAULT;
776 	}
777 
778 	return status;
779 }
780 
781 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
782 		unsigned int cmd, unsigned long arg)
783 {
784 	struct nvme_ns *ns = bdev->bd_disk->private_data;
785 
786 	switch (cmd) {
787 	case NVME_IOCTL_ID:
788 		force_successful_syscall_return();
789 		return ns->ns_id;
790 	case NVME_IOCTL_ADMIN_CMD:
791 		return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
792 	case NVME_IOCTL_IO_CMD:
793 		return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
794 	case NVME_IOCTL_SUBMIT_IO:
795 		return nvme_submit_io(ns, (void __user *)arg);
796 #ifdef CONFIG_BLK_DEV_NVME_SCSI
797 	case SG_GET_VERSION_NUM:
798 		return nvme_sg_get_version_num((void __user *)arg);
799 	case SG_IO:
800 		return nvme_sg_io(ns, (void __user *)arg);
801 #endif
802 	default:
803 		return -ENOTTY;
804 	}
805 }
806 
807 #ifdef CONFIG_COMPAT
808 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
809 			unsigned int cmd, unsigned long arg)
810 {
811 	switch (cmd) {
812 	case SG_IO:
813 		return -ENOIOCTLCMD;
814 	}
815 	return nvme_ioctl(bdev, mode, cmd, arg);
816 }
817 #else
818 #define nvme_compat_ioctl	NULL
819 #endif
820 
821 static int nvme_open(struct block_device *bdev, fmode_t mode)
822 {
823 	return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
824 }
825 
826 static void nvme_release(struct gendisk *disk, fmode_t mode)
827 {
828 	struct nvme_ns *ns = disk->private_data;
829 
830 	module_put(ns->ctrl->ops->module);
831 	nvme_put_ns(ns);
832 }
833 
834 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
835 {
836 	/* some standard values */
837 	geo->heads = 1 << 6;
838 	geo->sectors = 1 << 5;
839 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
840 	return 0;
841 }
842 
843 #ifdef CONFIG_BLK_DEV_INTEGRITY
844 static void nvme_init_integrity(struct nvme_ns *ns)
845 {
846 	struct blk_integrity integrity;
847 
848 	memset(&integrity, 0, sizeof(integrity));
849 	switch (ns->pi_type) {
850 	case NVME_NS_DPS_PI_TYPE3:
851 		integrity.profile = &t10_pi_type3_crc;
852 		integrity.tag_size = sizeof(u16) + sizeof(u32);
853 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
854 		break;
855 	case NVME_NS_DPS_PI_TYPE1:
856 	case NVME_NS_DPS_PI_TYPE2:
857 		integrity.profile = &t10_pi_type1_crc;
858 		integrity.tag_size = sizeof(u16);
859 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
860 		break;
861 	default:
862 		integrity.profile = NULL;
863 		break;
864 	}
865 	integrity.tuple_size = ns->ms;
866 	blk_integrity_register(ns->disk, &integrity);
867 	blk_queue_max_integrity_segments(ns->queue, 1);
868 }
869 #else
870 static void nvme_init_integrity(struct nvme_ns *ns)
871 {
872 }
873 #endif /* CONFIG_BLK_DEV_INTEGRITY */
874 
875 static void nvme_config_discard(struct nvme_ns *ns)
876 {
877 	struct nvme_ctrl *ctrl = ns->ctrl;
878 	u32 logical_block_size = queue_logical_block_size(ns->queue);
879 
880 	if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
881 		ns->queue->limits.discard_zeroes_data = 1;
882 	else
883 		ns->queue->limits.discard_zeroes_data = 0;
884 
885 	ns->queue->limits.discard_alignment = logical_block_size;
886 	ns->queue->limits.discard_granularity = logical_block_size;
887 	blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
888 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
889 }
890 
891 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
892 {
893 	if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
894 		dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
895 		return -ENODEV;
896 	}
897 
898 	if ((*id)->ncap == 0) {
899 		kfree(*id);
900 		return -ENODEV;
901 	}
902 
903 	if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
904 		memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
905 	if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
906 		memcpy(ns->uuid, (*id)->nguid, sizeof(ns->uuid));
907 
908 	return 0;
909 }
910 
911 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
912 {
913 	struct nvme_ns *ns = disk->private_data;
914 	u8 lbaf, pi_type;
915 	u16 old_ms;
916 	unsigned short bs;
917 
918 	old_ms = ns->ms;
919 	lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
920 	ns->lba_shift = id->lbaf[lbaf].ds;
921 	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
922 	ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
923 
924 	/*
925 	 * If identify namespace failed, use default 512 byte block size so
926 	 * block layer can use before failing read/write for 0 capacity.
927 	 */
928 	if (ns->lba_shift == 0)
929 		ns->lba_shift = 9;
930 	bs = 1 << ns->lba_shift;
931 	/* XXX: PI implementation requires metadata equal t10 pi tuple size */
932 	pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
933 					id->dps & NVME_NS_DPS_PI_MASK : 0;
934 
935 	blk_mq_freeze_queue(disk->queue);
936 	if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
937 				ns->ms != old_ms ||
938 				bs != queue_logical_block_size(disk->queue) ||
939 				(ns->ms && ns->ext)))
940 		blk_integrity_unregister(disk);
941 
942 	ns->pi_type = pi_type;
943 	blk_queue_logical_block_size(ns->queue, bs);
944 
945 	if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
946 		nvme_init_integrity(ns);
947 	if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
948 		set_capacity(disk, 0);
949 	else
950 		set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
951 
952 	if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
953 		nvme_config_discard(ns);
954 	blk_mq_unfreeze_queue(disk->queue);
955 }
956 
957 static int nvme_revalidate_disk(struct gendisk *disk)
958 {
959 	struct nvme_ns *ns = disk->private_data;
960 	struct nvme_id_ns *id = NULL;
961 	int ret;
962 
963 	if (test_bit(NVME_NS_DEAD, &ns->flags)) {
964 		set_capacity(disk, 0);
965 		return -ENODEV;
966 	}
967 
968 	ret = nvme_revalidate_ns(ns, &id);
969 	if (ret)
970 		return ret;
971 
972 	__nvme_revalidate_disk(disk, id);
973 	kfree(id);
974 
975 	return 0;
976 }
977 
978 static char nvme_pr_type(enum pr_type type)
979 {
980 	switch (type) {
981 	case PR_WRITE_EXCLUSIVE:
982 		return 1;
983 	case PR_EXCLUSIVE_ACCESS:
984 		return 2;
985 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
986 		return 3;
987 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
988 		return 4;
989 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
990 		return 5;
991 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
992 		return 6;
993 	default:
994 		return 0;
995 	}
996 };
997 
998 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
999 				u64 key, u64 sa_key, u8 op)
1000 {
1001 	struct nvme_ns *ns = bdev->bd_disk->private_data;
1002 	struct nvme_command c;
1003 	u8 data[16] = { 0, };
1004 
1005 	put_unaligned_le64(key, &data[0]);
1006 	put_unaligned_le64(sa_key, &data[8]);
1007 
1008 	memset(&c, 0, sizeof(c));
1009 	c.common.opcode = op;
1010 	c.common.nsid = cpu_to_le32(ns->ns_id);
1011 	c.common.cdw10[0] = cpu_to_le32(cdw10);
1012 
1013 	return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1014 }
1015 
1016 static int nvme_pr_register(struct block_device *bdev, u64 old,
1017 		u64 new, unsigned flags)
1018 {
1019 	u32 cdw10;
1020 
1021 	if (flags & ~PR_FL_IGNORE_KEY)
1022 		return -EOPNOTSUPP;
1023 
1024 	cdw10 = old ? 2 : 0;
1025 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1026 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1027 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1028 }
1029 
1030 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1031 		enum pr_type type, unsigned flags)
1032 {
1033 	u32 cdw10;
1034 
1035 	if (flags & ~PR_FL_IGNORE_KEY)
1036 		return -EOPNOTSUPP;
1037 
1038 	cdw10 = nvme_pr_type(type) << 8;
1039 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1040 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1041 }
1042 
1043 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1044 		enum pr_type type, bool abort)
1045 {
1046 	u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1047 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1048 }
1049 
1050 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1051 {
1052 	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1053 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1054 }
1055 
1056 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1057 {
1058 	u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1059 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1060 }
1061 
1062 static const struct pr_ops nvme_pr_ops = {
1063 	.pr_register	= nvme_pr_register,
1064 	.pr_reserve	= nvme_pr_reserve,
1065 	.pr_release	= nvme_pr_release,
1066 	.pr_preempt	= nvme_pr_preempt,
1067 	.pr_clear	= nvme_pr_clear,
1068 };
1069 
1070 static const struct block_device_operations nvme_fops = {
1071 	.owner		= THIS_MODULE,
1072 	.ioctl		= nvme_ioctl,
1073 	.compat_ioctl	= nvme_compat_ioctl,
1074 	.open		= nvme_open,
1075 	.release	= nvme_release,
1076 	.getgeo		= nvme_getgeo,
1077 	.revalidate_disk= nvme_revalidate_disk,
1078 	.pr_ops		= &nvme_pr_ops,
1079 };
1080 
1081 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1082 {
1083 	unsigned long timeout =
1084 		((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1085 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1086 	int ret;
1087 
1088 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1089 		if (csts == ~0)
1090 			return -ENODEV;
1091 		if ((csts & NVME_CSTS_RDY) == bit)
1092 			break;
1093 
1094 		msleep(100);
1095 		if (fatal_signal_pending(current))
1096 			return -EINTR;
1097 		if (time_after(jiffies, timeout)) {
1098 			dev_err(ctrl->device,
1099 				"Device not ready; aborting %s\n", enabled ?
1100 						"initialisation" : "reset");
1101 			return -ENODEV;
1102 		}
1103 	}
1104 
1105 	return ret;
1106 }
1107 
1108 /*
1109  * If the device has been passed off to us in an enabled state, just clear
1110  * the enabled bit.  The spec says we should set the 'shutdown notification
1111  * bits', but doing so may cause the device to complete commands to the
1112  * admin queue ... and we don't know what memory that might be pointing at!
1113  */
1114 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1115 {
1116 	int ret;
1117 
1118 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1119 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1120 
1121 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1122 	if (ret)
1123 		return ret;
1124 
1125 	/* Checking for ctrl->tagset is a trick to avoid sleeping on module
1126 	 * load, since we only need the quirk on reset_controller. Notice
1127 	 * that the HGST device needs this delay only in firmware activation
1128 	 * procedure; unfortunately we have no (easy) way to verify this.
1129 	 */
1130 	if ((ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY) && ctrl->tagset)
1131 		msleep(NVME_QUIRK_DELAY_AMOUNT);
1132 
1133 	return nvme_wait_ready(ctrl, cap, false);
1134 }
1135 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1136 
1137 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1138 {
1139 	/*
1140 	 * Default to a 4K page size, with the intention to update this
1141 	 * path in the future to accomodate architectures with differing
1142 	 * kernel and IO page sizes.
1143 	 */
1144 	unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1145 	int ret;
1146 
1147 	if (page_shift < dev_page_min) {
1148 		dev_err(ctrl->device,
1149 			"Minimum device page size %u too large for host (%u)\n",
1150 			1 << dev_page_min, 1 << page_shift);
1151 		return -ENODEV;
1152 	}
1153 
1154 	ctrl->page_size = 1 << page_shift;
1155 
1156 	ctrl->ctrl_config = NVME_CC_CSS_NVM;
1157 	ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1158 	ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1159 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1160 	ctrl->ctrl_config |= NVME_CC_ENABLE;
1161 
1162 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1163 	if (ret)
1164 		return ret;
1165 	return nvme_wait_ready(ctrl, cap, true);
1166 }
1167 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1168 
1169 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1170 {
1171 	unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1172 	u32 csts;
1173 	int ret;
1174 
1175 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1176 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1177 
1178 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1179 	if (ret)
1180 		return ret;
1181 
1182 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1183 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1184 			break;
1185 
1186 		msleep(100);
1187 		if (fatal_signal_pending(current))
1188 			return -EINTR;
1189 		if (time_after(jiffies, timeout)) {
1190 			dev_err(ctrl->device,
1191 				"Device shutdown incomplete; abort shutdown\n");
1192 			return -ENODEV;
1193 		}
1194 	}
1195 
1196 	return ret;
1197 }
1198 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1199 
1200 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1201 		struct request_queue *q)
1202 {
1203 	bool vwc = false;
1204 
1205 	if (ctrl->max_hw_sectors) {
1206 		u32 max_segments =
1207 			(ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1208 
1209 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1210 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1211 	}
1212 	if (ctrl->stripe_size)
1213 		blk_queue_chunk_sectors(q, ctrl->stripe_size >> 9);
1214 	blk_queue_virt_boundary(q, ctrl->page_size - 1);
1215 	if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1216 		vwc = true;
1217 	blk_queue_write_cache(q, vwc, vwc);
1218 }
1219 
1220 /*
1221  * Initialize the cached copies of the Identify data and various controller
1222  * register in our nvme_ctrl structure.  This should be called as soon as
1223  * the admin queue is fully up and running.
1224  */
1225 int nvme_init_identify(struct nvme_ctrl *ctrl)
1226 {
1227 	struct nvme_id_ctrl *id;
1228 	u64 cap;
1229 	int ret, page_shift;
1230 	u32 max_hw_sectors;
1231 
1232 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1233 	if (ret) {
1234 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1235 		return ret;
1236 	}
1237 
1238 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1239 	if (ret) {
1240 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1241 		return ret;
1242 	}
1243 	page_shift = NVME_CAP_MPSMIN(cap) + 12;
1244 
1245 	if (ctrl->vs >= NVME_VS(1, 1, 0))
1246 		ctrl->subsystem = NVME_CAP_NSSRC(cap);
1247 
1248 	ret = nvme_identify_ctrl(ctrl, &id);
1249 	if (ret) {
1250 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1251 		return -EIO;
1252 	}
1253 
1254 	ctrl->vid = le16_to_cpu(id->vid);
1255 	ctrl->oncs = le16_to_cpup(&id->oncs);
1256 	atomic_set(&ctrl->abort_limit, id->acl + 1);
1257 	ctrl->vwc = id->vwc;
1258 	ctrl->cntlid = le16_to_cpup(&id->cntlid);
1259 	memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1260 	memcpy(ctrl->model, id->mn, sizeof(id->mn));
1261 	memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1262 	if (id->mdts)
1263 		max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1264 	else
1265 		max_hw_sectors = UINT_MAX;
1266 	ctrl->max_hw_sectors =
1267 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1268 
1269 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
1270 		unsigned int max_hw_sectors;
1271 
1272 		ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
1273 		max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
1274 		if (ctrl->max_hw_sectors) {
1275 			ctrl->max_hw_sectors = min(max_hw_sectors,
1276 							ctrl->max_hw_sectors);
1277 		} else {
1278 			ctrl->max_hw_sectors = max_hw_sectors;
1279 		}
1280 	}
1281 
1282 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
1283 	ctrl->sgls = le32_to_cpu(id->sgls);
1284 	ctrl->kas = le16_to_cpu(id->kas);
1285 
1286 	if (ctrl->ops->is_fabrics) {
1287 		ctrl->icdoff = le16_to_cpu(id->icdoff);
1288 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1289 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1290 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1291 
1292 		/*
1293 		 * In fabrics we need to verify the cntlid matches the
1294 		 * admin connect
1295 		 */
1296 		if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1297 			ret = -EINVAL;
1298 
1299 		if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1300 			dev_err(ctrl->dev,
1301 				"keep-alive support is mandatory for fabrics\n");
1302 			ret = -EINVAL;
1303 		}
1304 	} else {
1305 		ctrl->cntlid = le16_to_cpu(id->cntlid);
1306 	}
1307 
1308 	kfree(id);
1309 	return ret;
1310 }
1311 EXPORT_SYMBOL_GPL(nvme_init_identify);
1312 
1313 static int nvme_dev_open(struct inode *inode, struct file *file)
1314 {
1315 	struct nvme_ctrl *ctrl;
1316 	int instance = iminor(inode);
1317 	int ret = -ENODEV;
1318 
1319 	spin_lock(&dev_list_lock);
1320 	list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1321 		if (ctrl->instance != instance)
1322 			continue;
1323 
1324 		if (!ctrl->admin_q) {
1325 			ret = -EWOULDBLOCK;
1326 			break;
1327 		}
1328 		if (!kref_get_unless_zero(&ctrl->kref))
1329 			break;
1330 		file->private_data = ctrl;
1331 		ret = 0;
1332 		break;
1333 	}
1334 	spin_unlock(&dev_list_lock);
1335 
1336 	return ret;
1337 }
1338 
1339 static int nvme_dev_release(struct inode *inode, struct file *file)
1340 {
1341 	nvme_put_ctrl(file->private_data);
1342 	return 0;
1343 }
1344 
1345 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1346 {
1347 	struct nvme_ns *ns;
1348 	int ret;
1349 
1350 	mutex_lock(&ctrl->namespaces_mutex);
1351 	if (list_empty(&ctrl->namespaces)) {
1352 		ret = -ENOTTY;
1353 		goto out_unlock;
1354 	}
1355 
1356 	ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1357 	if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1358 		dev_warn(ctrl->device,
1359 			"NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1360 		ret = -EINVAL;
1361 		goto out_unlock;
1362 	}
1363 
1364 	dev_warn(ctrl->device,
1365 		"using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1366 	kref_get(&ns->kref);
1367 	mutex_unlock(&ctrl->namespaces_mutex);
1368 
1369 	ret = nvme_user_cmd(ctrl, ns, argp);
1370 	nvme_put_ns(ns);
1371 	return ret;
1372 
1373 out_unlock:
1374 	mutex_unlock(&ctrl->namespaces_mutex);
1375 	return ret;
1376 }
1377 
1378 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1379 		unsigned long arg)
1380 {
1381 	struct nvme_ctrl *ctrl = file->private_data;
1382 	void __user *argp = (void __user *)arg;
1383 
1384 	switch (cmd) {
1385 	case NVME_IOCTL_ADMIN_CMD:
1386 		return nvme_user_cmd(ctrl, NULL, argp);
1387 	case NVME_IOCTL_IO_CMD:
1388 		return nvme_dev_user_cmd(ctrl, argp);
1389 	case NVME_IOCTL_RESET:
1390 		dev_warn(ctrl->device, "resetting controller\n");
1391 		return ctrl->ops->reset_ctrl(ctrl);
1392 	case NVME_IOCTL_SUBSYS_RESET:
1393 		return nvme_reset_subsystem(ctrl);
1394 	case NVME_IOCTL_RESCAN:
1395 		nvme_queue_scan(ctrl);
1396 		return 0;
1397 	default:
1398 		return -ENOTTY;
1399 	}
1400 }
1401 
1402 static const struct file_operations nvme_dev_fops = {
1403 	.owner		= THIS_MODULE,
1404 	.open		= nvme_dev_open,
1405 	.release	= nvme_dev_release,
1406 	.unlocked_ioctl	= nvme_dev_ioctl,
1407 	.compat_ioctl	= nvme_dev_ioctl,
1408 };
1409 
1410 static ssize_t nvme_sysfs_reset(struct device *dev,
1411 				struct device_attribute *attr, const char *buf,
1412 				size_t count)
1413 {
1414 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1415 	int ret;
1416 
1417 	ret = ctrl->ops->reset_ctrl(ctrl);
1418 	if (ret < 0)
1419 		return ret;
1420 	return count;
1421 }
1422 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1423 
1424 static ssize_t nvme_sysfs_rescan(struct device *dev,
1425 				struct device_attribute *attr, const char *buf,
1426 				size_t count)
1427 {
1428 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1429 
1430 	nvme_queue_scan(ctrl);
1431 	return count;
1432 }
1433 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1434 
1435 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1436 								char *buf)
1437 {
1438 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1439 	struct nvme_ctrl *ctrl = ns->ctrl;
1440 	int serial_len = sizeof(ctrl->serial);
1441 	int model_len = sizeof(ctrl->model);
1442 
1443 	if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1444 		return sprintf(buf, "eui.%16phN\n", ns->uuid);
1445 
1446 	if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1447 		return sprintf(buf, "eui.%8phN\n", ns->eui);
1448 
1449 	while (ctrl->serial[serial_len - 1] == ' ')
1450 		serial_len--;
1451 	while (ctrl->model[model_len - 1] == ' ')
1452 		model_len--;
1453 
1454 	return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1455 		serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1456 }
1457 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1458 
1459 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1460 								char *buf)
1461 {
1462 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1463 	return sprintf(buf, "%pU\n", ns->uuid);
1464 }
1465 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1466 
1467 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1468 								char *buf)
1469 {
1470 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1471 	return sprintf(buf, "%8phd\n", ns->eui);
1472 }
1473 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1474 
1475 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1476 								char *buf)
1477 {
1478 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1479 	return sprintf(buf, "%d\n", ns->ns_id);
1480 }
1481 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1482 
1483 static struct attribute *nvme_ns_attrs[] = {
1484 	&dev_attr_wwid.attr,
1485 	&dev_attr_uuid.attr,
1486 	&dev_attr_eui.attr,
1487 	&dev_attr_nsid.attr,
1488 	NULL,
1489 };
1490 
1491 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1492 		struct attribute *a, int n)
1493 {
1494 	struct device *dev = container_of(kobj, struct device, kobj);
1495 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1496 
1497 	if (a == &dev_attr_uuid.attr) {
1498 		if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1499 			return 0;
1500 	}
1501 	if (a == &dev_attr_eui.attr) {
1502 		if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1503 			return 0;
1504 	}
1505 	return a->mode;
1506 }
1507 
1508 static const struct attribute_group nvme_ns_attr_group = {
1509 	.attrs		= nvme_ns_attrs,
1510 	.is_visible	= nvme_ns_attrs_are_visible,
1511 };
1512 
1513 #define nvme_show_str_function(field)						\
1514 static ssize_t  field##_show(struct device *dev,				\
1515 			    struct device_attribute *attr, char *buf)		\
1516 {										\
1517         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
1518         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);	\
1519 }										\
1520 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1521 
1522 #define nvme_show_int_function(field)						\
1523 static ssize_t  field##_show(struct device *dev,				\
1524 			    struct device_attribute *attr, char *buf)		\
1525 {										\
1526         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
1527         return sprintf(buf, "%d\n", ctrl->field);	\
1528 }										\
1529 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1530 
1531 nvme_show_str_function(model);
1532 nvme_show_str_function(serial);
1533 nvme_show_str_function(firmware_rev);
1534 nvme_show_int_function(cntlid);
1535 
1536 static ssize_t nvme_sysfs_delete(struct device *dev,
1537 				struct device_attribute *attr, const char *buf,
1538 				size_t count)
1539 {
1540 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1541 
1542 	if (device_remove_file_self(dev, attr))
1543 		ctrl->ops->delete_ctrl(ctrl);
1544 	return count;
1545 }
1546 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1547 
1548 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1549 					 struct device_attribute *attr,
1550 					 char *buf)
1551 {
1552 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1553 
1554 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1555 }
1556 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1557 
1558 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1559 					 struct device_attribute *attr,
1560 					 char *buf)
1561 {
1562 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1563 
1564 	return snprintf(buf, PAGE_SIZE, "%s\n",
1565 			ctrl->ops->get_subsysnqn(ctrl));
1566 }
1567 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1568 
1569 static ssize_t nvme_sysfs_show_address(struct device *dev,
1570 					 struct device_attribute *attr,
1571 					 char *buf)
1572 {
1573 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1574 
1575 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1576 }
1577 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1578 
1579 static struct attribute *nvme_dev_attrs[] = {
1580 	&dev_attr_reset_controller.attr,
1581 	&dev_attr_rescan_controller.attr,
1582 	&dev_attr_model.attr,
1583 	&dev_attr_serial.attr,
1584 	&dev_attr_firmware_rev.attr,
1585 	&dev_attr_cntlid.attr,
1586 	&dev_attr_delete_controller.attr,
1587 	&dev_attr_transport.attr,
1588 	&dev_attr_subsysnqn.attr,
1589 	&dev_attr_address.attr,
1590 	NULL
1591 };
1592 
1593 #define CHECK_ATTR(ctrl, a, name)		\
1594 	if ((a) == &dev_attr_##name.attr &&	\
1595 	    !(ctrl)->ops->get_##name)		\
1596 		return 0
1597 
1598 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1599 		struct attribute *a, int n)
1600 {
1601 	struct device *dev = container_of(kobj, struct device, kobj);
1602 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1603 
1604 	if (a == &dev_attr_delete_controller.attr) {
1605 		if (!ctrl->ops->delete_ctrl)
1606 			return 0;
1607 	}
1608 
1609 	CHECK_ATTR(ctrl, a, subsysnqn);
1610 	CHECK_ATTR(ctrl, a, address);
1611 
1612 	return a->mode;
1613 }
1614 
1615 static struct attribute_group nvme_dev_attrs_group = {
1616 	.attrs		= nvme_dev_attrs,
1617 	.is_visible	= nvme_dev_attrs_are_visible,
1618 };
1619 
1620 static const struct attribute_group *nvme_dev_attr_groups[] = {
1621 	&nvme_dev_attrs_group,
1622 	NULL,
1623 };
1624 
1625 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1626 {
1627 	struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1628 	struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1629 
1630 	return nsa->ns_id - nsb->ns_id;
1631 }
1632 
1633 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1634 {
1635 	struct nvme_ns *ns, *ret = NULL;
1636 
1637 	mutex_lock(&ctrl->namespaces_mutex);
1638 	list_for_each_entry(ns, &ctrl->namespaces, list) {
1639 		if (ns->ns_id == nsid) {
1640 			kref_get(&ns->kref);
1641 			ret = ns;
1642 			break;
1643 		}
1644 		if (ns->ns_id > nsid)
1645 			break;
1646 	}
1647 	mutex_unlock(&ctrl->namespaces_mutex);
1648 	return ret;
1649 }
1650 
1651 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1652 {
1653 	struct nvme_ns *ns;
1654 	struct gendisk *disk;
1655 	struct nvme_id_ns *id;
1656 	char disk_name[DISK_NAME_LEN];
1657 	int node = dev_to_node(ctrl->dev);
1658 
1659 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1660 	if (!ns)
1661 		return;
1662 
1663 	ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1664 	if (ns->instance < 0)
1665 		goto out_free_ns;
1666 
1667 	ns->queue = blk_mq_init_queue(ctrl->tagset);
1668 	if (IS_ERR(ns->queue))
1669 		goto out_release_instance;
1670 	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1671 	ns->queue->queuedata = ns;
1672 	ns->ctrl = ctrl;
1673 
1674 	kref_init(&ns->kref);
1675 	ns->ns_id = nsid;
1676 	ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1677 
1678 	blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1679 	nvme_set_queue_limits(ctrl, ns->queue);
1680 
1681 	sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1682 
1683 	if (nvme_revalidate_ns(ns, &id))
1684 		goto out_free_queue;
1685 
1686 	if (nvme_nvm_ns_supported(ns, id)) {
1687 		if (nvme_nvm_register(ns, disk_name, node,
1688 							&nvme_ns_attr_group)) {
1689 			dev_warn(ctrl->dev, "%s: LightNVM init failure\n",
1690 								__func__);
1691 			goto out_free_id;
1692 		}
1693 	} else {
1694 		disk = alloc_disk_node(0, node);
1695 		if (!disk)
1696 			goto out_free_id;
1697 
1698 		disk->fops = &nvme_fops;
1699 		disk->private_data = ns;
1700 		disk->queue = ns->queue;
1701 		disk->flags = GENHD_FL_EXT_DEVT;
1702 		memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
1703 		ns->disk = disk;
1704 
1705 		__nvme_revalidate_disk(disk, id);
1706 	}
1707 
1708 	mutex_lock(&ctrl->namespaces_mutex);
1709 	list_add_tail(&ns->list, &ctrl->namespaces);
1710 	mutex_unlock(&ctrl->namespaces_mutex);
1711 
1712 	kref_get(&ctrl->kref);
1713 
1714 	kfree(id);
1715 
1716 	if (ns->ndev)
1717 		return;
1718 
1719 	device_add_disk(ctrl->device, ns->disk);
1720 	if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1721 					&nvme_ns_attr_group))
1722 		pr_warn("%s: failed to create sysfs group for identification\n",
1723 			ns->disk->disk_name);
1724 	return;
1725  out_free_id:
1726 	kfree(id);
1727  out_free_queue:
1728 	blk_cleanup_queue(ns->queue);
1729  out_release_instance:
1730 	ida_simple_remove(&ctrl->ns_ida, ns->instance);
1731  out_free_ns:
1732 	kfree(ns);
1733 }
1734 
1735 static void nvme_ns_remove(struct nvme_ns *ns)
1736 {
1737 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1738 		return;
1739 
1740 	if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
1741 		if (blk_get_integrity(ns->disk))
1742 			blk_integrity_unregister(ns->disk);
1743 		sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1744 					&nvme_ns_attr_group);
1745 		del_gendisk(ns->disk);
1746 		blk_mq_abort_requeue_list(ns->queue);
1747 		blk_cleanup_queue(ns->queue);
1748 	}
1749 
1750 	mutex_lock(&ns->ctrl->namespaces_mutex);
1751 	list_del_init(&ns->list);
1752 	mutex_unlock(&ns->ctrl->namespaces_mutex);
1753 
1754 	nvme_put_ns(ns);
1755 }
1756 
1757 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1758 {
1759 	struct nvme_ns *ns;
1760 
1761 	ns = nvme_find_get_ns(ctrl, nsid);
1762 	if (ns) {
1763 		if (ns->disk && revalidate_disk(ns->disk))
1764 			nvme_ns_remove(ns);
1765 		nvme_put_ns(ns);
1766 	} else
1767 		nvme_alloc_ns(ctrl, nsid);
1768 }
1769 
1770 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
1771 					unsigned nsid)
1772 {
1773 	struct nvme_ns *ns, *next;
1774 
1775 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1776 		if (ns->ns_id > nsid)
1777 			nvme_ns_remove(ns);
1778 	}
1779 }
1780 
1781 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1782 {
1783 	struct nvme_ns *ns;
1784 	__le32 *ns_list;
1785 	unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1786 	int ret = 0;
1787 
1788 	ns_list = kzalloc(0x1000, GFP_KERNEL);
1789 	if (!ns_list)
1790 		return -ENOMEM;
1791 
1792 	for (i = 0; i < num_lists; i++) {
1793 		ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1794 		if (ret)
1795 			goto free;
1796 
1797 		for (j = 0; j < min(nn, 1024U); j++) {
1798 			nsid = le32_to_cpu(ns_list[j]);
1799 			if (!nsid)
1800 				goto out;
1801 
1802 			nvme_validate_ns(ctrl, nsid);
1803 
1804 			while (++prev < nsid) {
1805 				ns = nvme_find_get_ns(ctrl, prev);
1806 				if (ns) {
1807 					nvme_ns_remove(ns);
1808 					nvme_put_ns(ns);
1809 				}
1810 			}
1811 		}
1812 		nn -= j;
1813 	}
1814  out:
1815 	nvme_remove_invalid_namespaces(ctrl, prev);
1816  free:
1817 	kfree(ns_list);
1818 	return ret;
1819 }
1820 
1821 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
1822 {
1823 	unsigned i;
1824 
1825 	for (i = 1; i <= nn; i++)
1826 		nvme_validate_ns(ctrl, i);
1827 
1828 	nvme_remove_invalid_namespaces(ctrl, nn);
1829 }
1830 
1831 static void nvme_scan_work(struct work_struct *work)
1832 {
1833 	struct nvme_ctrl *ctrl =
1834 		container_of(work, struct nvme_ctrl, scan_work);
1835 	struct nvme_id_ctrl *id;
1836 	unsigned nn;
1837 
1838 	if (ctrl->state != NVME_CTRL_LIVE)
1839 		return;
1840 
1841 	if (nvme_identify_ctrl(ctrl, &id))
1842 		return;
1843 
1844 	nn = le32_to_cpu(id->nn);
1845 	if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1846 	    !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1847 		if (!nvme_scan_ns_list(ctrl, nn))
1848 			goto done;
1849 	}
1850 	nvme_scan_ns_sequential(ctrl, nn);
1851  done:
1852 	mutex_lock(&ctrl->namespaces_mutex);
1853 	list_sort(NULL, &ctrl->namespaces, ns_cmp);
1854 	mutex_unlock(&ctrl->namespaces_mutex);
1855 	kfree(id);
1856 }
1857 
1858 void nvme_queue_scan(struct nvme_ctrl *ctrl)
1859 {
1860 	/*
1861 	 * Do not queue new scan work when a controller is reset during
1862 	 * removal.
1863 	 */
1864 	if (ctrl->state == NVME_CTRL_LIVE)
1865 		schedule_work(&ctrl->scan_work);
1866 }
1867 EXPORT_SYMBOL_GPL(nvme_queue_scan);
1868 
1869 /*
1870  * This function iterates the namespace list unlocked to allow recovery from
1871  * controller failure. It is up to the caller to ensure the namespace list is
1872  * not modified by scan work while this function is executing.
1873  */
1874 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1875 {
1876 	struct nvme_ns *ns, *next;
1877 
1878 	/*
1879 	 * The dead states indicates the controller was not gracefully
1880 	 * disconnected. In that case, we won't be able to flush any data while
1881 	 * removing the namespaces' disks; fail all the queues now to avoid
1882 	 * potentially having to clean up the failed sync later.
1883 	 */
1884 	if (ctrl->state == NVME_CTRL_DEAD)
1885 		nvme_kill_queues(ctrl);
1886 
1887 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1888 		nvme_ns_remove(ns);
1889 }
1890 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1891 
1892 static void nvme_async_event_work(struct work_struct *work)
1893 {
1894 	struct nvme_ctrl *ctrl =
1895 		container_of(work, struct nvme_ctrl, async_event_work);
1896 
1897 	spin_lock_irq(&ctrl->lock);
1898 	while (ctrl->event_limit > 0) {
1899 		int aer_idx = --ctrl->event_limit;
1900 
1901 		spin_unlock_irq(&ctrl->lock);
1902 		ctrl->ops->submit_async_event(ctrl, aer_idx);
1903 		spin_lock_irq(&ctrl->lock);
1904 	}
1905 	spin_unlock_irq(&ctrl->lock);
1906 }
1907 
1908 void nvme_complete_async_event(struct nvme_ctrl *ctrl,
1909 		struct nvme_completion *cqe)
1910 {
1911 	u16 status = le16_to_cpu(cqe->status) >> 1;
1912 	u32 result = le32_to_cpu(cqe->result);
1913 
1914 	if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ) {
1915 		++ctrl->event_limit;
1916 		schedule_work(&ctrl->async_event_work);
1917 	}
1918 
1919 	if (status != NVME_SC_SUCCESS)
1920 		return;
1921 
1922 	switch (result & 0xff07) {
1923 	case NVME_AER_NOTICE_NS_CHANGED:
1924 		dev_info(ctrl->device, "rescanning\n");
1925 		nvme_queue_scan(ctrl);
1926 		break;
1927 	default:
1928 		dev_warn(ctrl->device, "async event result %08x\n", result);
1929 	}
1930 }
1931 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
1932 
1933 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
1934 {
1935 	ctrl->event_limit = NVME_NR_AERS;
1936 	schedule_work(&ctrl->async_event_work);
1937 }
1938 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
1939 
1940 static DEFINE_IDA(nvme_instance_ida);
1941 
1942 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1943 {
1944 	int instance, error;
1945 
1946 	do {
1947 		if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1948 			return -ENODEV;
1949 
1950 		spin_lock(&dev_list_lock);
1951 		error = ida_get_new(&nvme_instance_ida, &instance);
1952 		spin_unlock(&dev_list_lock);
1953 	} while (error == -EAGAIN);
1954 
1955 	if (error)
1956 		return -ENODEV;
1957 
1958 	ctrl->instance = instance;
1959 	return 0;
1960 }
1961 
1962 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1963 {
1964 	spin_lock(&dev_list_lock);
1965 	ida_remove(&nvme_instance_ida, ctrl->instance);
1966 	spin_unlock(&dev_list_lock);
1967 }
1968 
1969 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1970 {
1971 	flush_work(&ctrl->async_event_work);
1972 	flush_work(&ctrl->scan_work);
1973 	nvme_remove_namespaces(ctrl);
1974 
1975 	device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1976 
1977 	spin_lock(&dev_list_lock);
1978 	list_del(&ctrl->node);
1979 	spin_unlock(&dev_list_lock);
1980 }
1981 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
1982 
1983 static void nvme_free_ctrl(struct kref *kref)
1984 {
1985 	struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1986 
1987 	put_device(ctrl->device);
1988 	nvme_release_instance(ctrl);
1989 	ida_destroy(&ctrl->ns_ida);
1990 
1991 	ctrl->ops->free_ctrl(ctrl);
1992 }
1993 
1994 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1995 {
1996 	kref_put(&ctrl->kref, nvme_free_ctrl);
1997 }
1998 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
1999 
2000 /*
2001  * Initialize a NVMe controller structures.  This needs to be called during
2002  * earliest initialization so that we have the initialized structured around
2003  * during probing.
2004  */
2005 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2006 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
2007 {
2008 	int ret;
2009 
2010 	ctrl->state = NVME_CTRL_NEW;
2011 	spin_lock_init(&ctrl->lock);
2012 	INIT_LIST_HEAD(&ctrl->namespaces);
2013 	mutex_init(&ctrl->namespaces_mutex);
2014 	kref_init(&ctrl->kref);
2015 	ctrl->dev = dev;
2016 	ctrl->ops = ops;
2017 	ctrl->quirks = quirks;
2018 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2019 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2020 
2021 	ret = nvme_set_instance(ctrl);
2022 	if (ret)
2023 		goto out;
2024 
2025 	ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2026 				MKDEV(nvme_char_major, ctrl->instance),
2027 				ctrl, nvme_dev_attr_groups,
2028 				"nvme%d", ctrl->instance);
2029 	if (IS_ERR(ctrl->device)) {
2030 		ret = PTR_ERR(ctrl->device);
2031 		goto out_release_instance;
2032 	}
2033 	get_device(ctrl->device);
2034 	ida_init(&ctrl->ns_ida);
2035 
2036 	spin_lock(&dev_list_lock);
2037 	list_add_tail(&ctrl->node, &nvme_ctrl_list);
2038 	spin_unlock(&dev_list_lock);
2039 
2040 	return 0;
2041 out_release_instance:
2042 	nvme_release_instance(ctrl);
2043 out:
2044 	return ret;
2045 }
2046 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2047 
2048 /**
2049  * nvme_kill_queues(): Ends all namespace queues
2050  * @ctrl: the dead controller that needs to end
2051  *
2052  * Call this function when the driver determines it is unable to get the
2053  * controller in a state capable of servicing IO.
2054  */
2055 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2056 {
2057 	struct nvme_ns *ns;
2058 
2059 	mutex_lock(&ctrl->namespaces_mutex);
2060 	list_for_each_entry(ns, &ctrl->namespaces, list) {
2061 		/*
2062 		 * Revalidating a dead namespace sets capacity to 0. This will
2063 		 * end buffered writers dirtying pages that can't be synced.
2064 		 */
2065 		if (ns->disk && !test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2066 			revalidate_disk(ns->disk);
2067 
2068 		blk_set_queue_dying(ns->queue);
2069 		blk_mq_abort_requeue_list(ns->queue);
2070 		blk_mq_start_stopped_hw_queues(ns->queue, true);
2071 	}
2072 	mutex_unlock(&ctrl->namespaces_mutex);
2073 }
2074 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2075 
2076 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2077 {
2078 	struct nvme_ns *ns;
2079 
2080 	mutex_lock(&ctrl->namespaces_mutex);
2081 	list_for_each_entry(ns, &ctrl->namespaces, list) {
2082 		spin_lock_irq(ns->queue->queue_lock);
2083 		queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
2084 		spin_unlock_irq(ns->queue->queue_lock);
2085 
2086 		blk_mq_cancel_requeue_work(ns->queue);
2087 		blk_mq_stop_hw_queues(ns->queue);
2088 	}
2089 	mutex_unlock(&ctrl->namespaces_mutex);
2090 }
2091 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2092 
2093 void nvme_start_queues(struct nvme_ctrl *ctrl)
2094 {
2095 	struct nvme_ns *ns;
2096 
2097 	mutex_lock(&ctrl->namespaces_mutex);
2098 	list_for_each_entry(ns, &ctrl->namespaces, list) {
2099 		queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
2100 		blk_mq_start_stopped_hw_queues(ns->queue, true);
2101 		blk_mq_kick_requeue_list(ns->queue);
2102 	}
2103 	mutex_unlock(&ctrl->namespaces_mutex);
2104 }
2105 EXPORT_SYMBOL_GPL(nvme_start_queues);
2106 
2107 int __init nvme_core_init(void)
2108 {
2109 	int result;
2110 
2111 	result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2112 							&nvme_dev_fops);
2113 	if (result < 0)
2114 		return result;
2115 	else if (result > 0)
2116 		nvme_char_major = result;
2117 
2118 	nvme_class = class_create(THIS_MODULE, "nvme");
2119 	if (IS_ERR(nvme_class)) {
2120 		result = PTR_ERR(nvme_class);
2121 		goto unregister_chrdev;
2122 	}
2123 
2124 	return 0;
2125 
2126  unregister_chrdev:
2127 	__unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2128 	return result;
2129 }
2130 
2131 void nvme_core_exit(void)
2132 {
2133 	class_destroy(nvme_class);
2134 	__unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2135 }
2136 
2137 MODULE_LICENSE("GPL");
2138 MODULE_VERSION("1.0");
2139 module_init(nvme_core_init);
2140 module_exit(nvme_core_exit);
2141