xref: /linux/drivers/nvme/host/core.c (revision c1aac62f36c1e37ee81c9e09ee9ee733eef05dcb)
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 	blk_mq_requeue_request(req, !blk_mq_queue_stopped(req->q));
205 }
206 EXPORT_SYMBOL_GPL(nvme_requeue_req);
207 
208 struct request *nvme_alloc_request(struct request_queue *q,
209 		struct nvme_command *cmd, unsigned int flags, int qid)
210 {
211 	unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
212 	struct request *req;
213 
214 	if (qid == NVME_QID_ANY) {
215 		req = blk_mq_alloc_request(q, op, flags);
216 	} else {
217 		req = blk_mq_alloc_request_hctx(q, op, flags,
218 				qid ? qid - 1 : 0);
219 	}
220 	if (IS_ERR(req))
221 		return req;
222 
223 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
224 	nvme_req(req)->cmd = cmd;
225 
226 	return req;
227 }
228 EXPORT_SYMBOL_GPL(nvme_alloc_request);
229 
230 static inline void nvme_setup_flush(struct nvme_ns *ns,
231 		struct nvme_command *cmnd)
232 {
233 	memset(cmnd, 0, sizeof(*cmnd));
234 	cmnd->common.opcode = nvme_cmd_flush;
235 	cmnd->common.nsid = cpu_to_le32(ns->ns_id);
236 }
237 
238 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
239 		struct nvme_command *cmnd)
240 {
241 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
242 	struct nvme_dsm_range *range;
243 	struct bio *bio;
244 
245 	range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
246 	if (!range)
247 		return BLK_MQ_RQ_QUEUE_BUSY;
248 
249 	__rq_for_each_bio(bio, req) {
250 		u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
251 		u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
252 
253 		range[n].cattr = cpu_to_le32(0);
254 		range[n].nlb = cpu_to_le32(nlb);
255 		range[n].slba = cpu_to_le64(slba);
256 		n++;
257 	}
258 
259 	if (WARN_ON_ONCE(n != segments)) {
260 		kfree(range);
261 		return BLK_MQ_RQ_QUEUE_ERROR;
262 	}
263 
264 	memset(cmnd, 0, sizeof(*cmnd));
265 	cmnd->dsm.opcode = nvme_cmd_dsm;
266 	cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
267 	cmnd->dsm.nr = segments - 1;
268 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
269 
270 	req->special_vec.bv_page = virt_to_page(range);
271 	req->special_vec.bv_offset = offset_in_page(range);
272 	req->special_vec.bv_len = sizeof(*range) * segments;
273 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
274 
275 	return BLK_MQ_RQ_QUEUE_OK;
276 }
277 
278 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
279 		struct nvme_command *cmnd)
280 {
281 	u16 control = 0;
282 	u32 dsmgmt = 0;
283 
284 	if (req->cmd_flags & REQ_FUA)
285 		control |= NVME_RW_FUA;
286 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
287 		control |= NVME_RW_LR;
288 
289 	if (req->cmd_flags & REQ_RAHEAD)
290 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
291 
292 	memset(cmnd, 0, sizeof(*cmnd));
293 	cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
294 	cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
295 	cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
296 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
297 
298 	if (ns->ms) {
299 		switch (ns->pi_type) {
300 		case NVME_NS_DPS_PI_TYPE3:
301 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
302 			break;
303 		case NVME_NS_DPS_PI_TYPE1:
304 		case NVME_NS_DPS_PI_TYPE2:
305 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
306 					NVME_RW_PRINFO_PRCHK_REF;
307 			cmnd->rw.reftag = cpu_to_le32(
308 					nvme_block_nr(ns, blk_rq_pos(req)));
309 			break;
310 		}
311 		if (!blk_integrity_rq(req))
312 			control |= NVME_RW_PRINFO_PRACT;
313 	}
314 
315 	cmnd->rw.control = cpu_to_le16(control);
316 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
317 }
318 
319 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
320 		struct nvme_command *cmd)
321 {
322 	int ret = BLK_MQ_RQ_QUEUE_OK;
323 
324 	switch (req_op(req)) {
325 	case REQ_OP_DRV_IN:
326 	case REQ_OP_DRV_OUT:
327 		memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
328 		break;
329 	case REQ_OP_FLUSH:
330 		nvme_setup_flush(ns, cmd);
331 		break;
332 	case REQ_OP_DISCARD:
333 		ret = nvme_setup_discard(ns, req, cmd);
334 		break;
335 	case REQ_OP_READ:
336 	case REQ_OP_WRITE:
337 		nvme_setup_rw(ns, req, cmd);
338 		break;
339 	default:
340 		WARN_ON_ONCE(1);
341 		return BLK_MQ_RQ_QUEUE_ERROR;
342 	}
343 
344 	cmd->common.command_id = req->tag;
345 	return ret;
346 }
347 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
348 
349 /*
350  * Returns 0 on success.  If the result is negative, it's a Linux error code;
351  * if the result is positive, it's an NVM Express status code
352  */
353 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
354 		union nvme_result *result, void *buffer, unsigned bufflen,
355 		unsigned timeout, int qid, int at_head, int flags)
356 {
357 	struct request *req;
358 	int ret;
359 
360 	req = nvme_alloc_request(q, cmd, flags, qid);
361 	if (IS_ERR(req))
362 		return PTR_ERR(req);
363 
364 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
365 
366 	if (buffer && bufflen) {
367 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
368 		if (ret)
369 			goto out;
370 	}
371 
372 	blk_execute_rq(req->q, NULL, req, at_head);
373 	if (result)
374 		*result = nvme_req(req)->result;
375 	ret = req->errors;
376  out:
377 	blk_mq_free_request(req);
378 	return ret;
379 }
380 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
381 
382 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
383 		void *buffer, unsigned bufflen)
384 {
385 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
386 			NVME_QID_ANY, 0, 0);
387 }
388 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
389 
390 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
391 		void __user *ubuffer, unsigned bufflen,
392 		void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
393 		u32 *result, unsigned timeout)
394 {
395 	bool write = nvme_is_write(cmd);
396 	struct nvme_ns *ns = q->queuedata;
397 	struct gendisk *disk = ns ? ns->disk : NULL;
398 	struct request *req;
399 	struct bio *bio = NULL;
400 	void *meta = NULL;
401 	int ret;
402 
403 	req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
404 	if (IS_ERR(req))
405 		return PTR_ERR(req);
406 
407 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
408 
409 	if (ubuffer && bufflen) {
410 		ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
411 				GFP_KERNEL);
412 		if (ret)
413 			goto out;
414 		bio = req->bio;
415 
416 		if (!disk)
417 			goto submit;
418 		bio->bi_bdev = bdget_disk(disk, 0);
419 		if (!bio->bi_bdev) {
420 			ret = -ENODEV;
421 			goto out_unmap;
422 		}
423 
424 		if (meta_buffer && meta_len) {
425 			struct bio_integrity_payload *bip;
426 
427 			meta = kmalloc(meta_len, GFP_KERNEL);
428 			if (!meta) {
429 				ret = -ENOMEM;
430 				goto out_unmap;
431 			}
432 
433 			if (write) {
434 				if (copy_from_user(meta, meta_buffer,
435 						meta_len)) {
436 					ret = -EFAULT;
437 					goto out_free_meta;
438 				}
439 			}
440 
441 			bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
442 			if (IS_ERR(bip)) {
443 				ret = PTR_ERR(bip);
444 				goto out_free_meta;
445 			}
446 
447 			bip->bip_iter.bi_size = meta_len;
448 			bip->bip_iter.bi_sector = meta_seed;
449 
450 			ret = bio_integrity_add_page(bio, virt_to_page(meta),
451 					meta_len, offset_in_page(meta));
452 			if (ret != meta_len) {
453 				ret = -ENOMEM;
454 				goto out_free_meta;
455 			}
456 		}
457 	}
458  submit:
459 	blk_execute_rq(req->q, disk, req, 0);
460 	ret = req->errors;
461 	if (result)
462 		*result = le32_to_cpu(nvme_req(req)->result.u32);
463 	if (meta && !ret && !write) {
464 		if (copy_to_user(meta_buffer, meta, meta_len))
465 			ret = -EFAULT;
466 	}
467  out_free_meta:
468 	kfree(meta);
469  out_unmap:
470 	if (bio) {
471 		if (disk && bio->bi_bdev)
472 			bdput(bio->bi_bdev);
473 		blk_rq_unmap_user(bio);
474 	}
475  out:
476 	blk_mq_free_request(req);
477 	return ret;
478 }
479 
480 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
481 		void __user *ubuffer, unsigned bufflen, u32 *result,
482 		unsigned timeout)
483 {
484 	return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
485 			result, timeout);
486 }
487 
488 static void nvme_keep_alive_end_io(struct request *rq, int error)
489 {
490 	struct nvme_ctrl *ctrl = rq->end_io_data;
491 
492 	blk_mq_free_request(rq);
493 
494 	if (error) {
495 		dev_err(ctrl->device,
496 			"failed nvme_keep_alive_end_io error=%d\n", error);
497 		return;
498 	}
499 
500 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
501 }
502 
503 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
504 {
505 	struct nvme_command c;
506 	struct request *rq;
507 
508 	memset(&c, 0, sizeof(c));
509 	c.common.opcode = nvme_admin_keep_alive;
510 
511 	rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
512 			NVME_QID_ANY);
513 	if (IS_ERR(rq))
514 		return PTR_ERR(rq);
515 
516 	rq->timeout = ctrl->kato * HZ;
517 	rq->end_io_data = ctrl;
518 
519 	blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
520 
521 	return 0;
522 }
523 
524 static void nvme_keep_alive_work(struct work_struct *work)
525 {
526 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
527 			struct nvme_ctrl, ka_work);
528 
529 	if (nvme_keep_alive(ctrl)) {
530 		/* allocation failure, reset the controller */
531 		dev_err(ctrl->device, "keep-alive failed\n");
532 		ctrl->ops->reset_ctrl(ctrl);
533 		return;
534 	}
535 }
536 
537 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
538 {
539 	if (unlikely(ctrl->kato == 0))
540 		return;
541 
542 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
543 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
544 }
545 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
546 
547 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
548 {
549 	if (unlikely(ctrl->kato == 0))
550 		return;
551 
552 	cancel_delayed_work_sync(&ctrl->ka_work);
553 }
554 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
555 
556 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
557 {
558 	struct nvme_command c = { };
559 	int error;
560 
561 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
562 	c.identify.opcode = nvme_admin_identify;
563 	c.identify.cns = cpu_to_le32(NVME_ID_CNS_CTRL);
564 
565 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
566 	if (!*id)
567 		return -ENOMEM;
568 
569 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
570 			sizeof(struct nvme_id_ctrl));
571 	if (error)
572 		kfree(*id);
573 	return error;
574 }
575 
576 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
577 {
578 	struct nvme_command c = { };
579 
580 	c.identify.opcode = nvme_admin_identify;
581 	c.identify.cns = cpu_to_le32(NVME_ID_CNS_NS_ACTIVE_LIST);
582 	c.identify.nsid = cpu_to_le32(nsid);
583 	return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
584 }
585 
586 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
587 		struct nvme_id_ns **id)
588 {
589 	struct nvme_command c = { };
590 	int error;
591 
592 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
593 	c.identify.opcode = nvme_admin_identify,
594 	c.identify.nsid = cpu_to_le32(nsid),
595 
596 	*id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
597 	if (!*id)
598 		return -ENOMEM;
599 
600 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
601 			sizeof(struct nvme_id_ns));
602 	if (error)
603 		kfree(*id);
604 	return error;
605 }
606 
607 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
608 		      void *buffer, size_t buflen, u32 *result)
609 {
610 	struct nvme_command c;
611 	union nvme_result res;
612 	int ret;
613 
614 	memset(&c, 0, sizeof(c));
615 	c.features.opcode = nvme_admin_get_features;
616 	c.features.nsid = cpu_to_le32(nsid);
617 	c.features.fid = cpu_to_le32(fid);
618 
619 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, buffer, buflen, 0,
620 			NVME_QID_ANY, 0, 0);
621 	if (ret >= 0 && result)
622 		*result = le32_to_cpu(res.u32);
623 	return ret;
624 }
625 
626 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
627 		      void *buffer, size_t buflen, u32 *result)
628 {
629 	struct nvme_command c;
630 	union nvme_result res;
631 	int ret;
632 
633 	memset(&c, 0, sizeof(c));
634 	c.features.opcode = nvme_admin_set_features;
635 	c.features.fid = cpu_to_le32(fid);
636 	c.features.dword11 = cpu_to_le32(dword11);
637 
638 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
639 			buffer, buflen, 0, NVME_QID_ANY, 0, 0);
640 	if (ret >= 0 && result)
641 		*result = le32_to_cpu(res.u32);
642 	return ret;
643 }
644 
645 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
646 {
647 	struct nvme_command c = { };
648 	int error;
649 
650 	c.common.opcode = nvme_admin_get_log_page,
651 	c.common.nsid = cpu_to_le32(0xFFFFFFFF),
652 	c.common.cdw10[0] = cpu_to_le32(
653 			(((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
654 			 NVME_LOG_SMART),
655 
656 	*log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
657 	if (!*log)
658 		return -ENOMEM;
659 
660 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
661 			sizeof(struct nvme_smart_log));
662 	if (error)
663 		kfree(*log);
664 	return error;
665 }
666 
667 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
668 {
669 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
670 	u32 result;
671 	int status, nr_io_queues;
672 
673 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
674 			&result);
675 	if (status < 0)
676 		return status;
677 
678 	/*
679 	 * Degraded controllers might return an error when setting the queue
680 	 * count.  We still want to be able to bring them online and offer
681 	 * access to the admin queue, as that might be only way to fix them up.
682 	 */
683 	if (status > 0) {
684 		dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
685 		*count = 0;
686 	} else {
687 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
688 		*count = min(*count, nr_io_queues);
689 	}
690 
691 	return 0;
692 }
693 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
694 
695 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
696 {
697 	struct nvme_user_io io;
698 	struct nvme_command c;
699 	unsigned length, meta_len;
700 	void __user *metadata;
701 
702 	if (copy_from_user(&io, uio, sizeof(io)))
703 		return -EFAULT;
704 	if (io.flags)
705 		return -EINVAL;
706 
707 	switch (io.opcode) {
708 	case nvme_cmd_write:
709 	case nvme_cmd_read:
710 	case nvme_cmd_compare:
711 		break;
712 	default:
713 		return -EINVAL;
714 	}
715 
716 	length = (io.nblocks + 1) << ns->lba_shift;
717 	meta_len = (io.nblocks + 1) * ns->ms;
718 	metadata = (void __user *)(uintptr_t)io.metadata;
719 
720 	if (ns->ext) {
721 		length += meta_len;
722 		meta_len = 0;
723 	} else if (meta_len) {
724 		if ((io.metadata & 3) || !io.metadata)
725 			return -EINVAL;
726 	}
727 
728 	memset(&c, 0, sizeof(c));
729 	c.rw.opcode = io.opcode;
730 	c.rw.flags = io.flags;
731 	c.rw.nsid = cpu_to_le32(ns->ns_id);
732 	c.rw.slba = cpu_to_le64(io.slba);
733 	c.rw.length = cpu_to_le16(io.nblocks);
734 	c.rw.control = cpu_to_le16(io.control);
735 	c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
736 	c.rw.reftag = cpu_to_le32(io.reftag);
737 	c.rw.apptag = cpu_to_le16(io.apptag);
738 	c.rw.appmask = cpu_to_le16(io.appmask);
739 
740 	return __nvme_submit_user_cmd(ns->queue, &c,
741 			(void __user *)(uintptr_t)io.addr, length,
742 			metadata, meta_len, io.slba, NULL, 0);
743 }
744 
745 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
746 			struct nvme_passthru_cmd __user *ucmd)
747 {
748 	struct nvme_passthru_cmd cmd;
749 	struct nvme_command c;
750 	unsigned timeout = 0;
751 	int status;
752 
753 	if (!capable(CAP_SYS_ADMIN))
754 		return -EACCES;
755 	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
756 		return -EFAULT;
757 	if (cmd.flags)
758 		return -EINVAL;
759 
760 	memset(&c, 0, sizeof(c));
761 	c.common.opcode = cmd.opcode;
762 	c.common.flags = cmd.flags;
763 	c.common.nsid = cpu_to_le32(cmd.nsid);
764 	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
765 	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
766 	c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
767 	c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
768 	c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
769 	c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
770 	c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
771 	c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
772 
773 	if (cmd.timeout_ms)
774 		timeout = msecs_to_jiffies(cmd.timeout_ms);
775 
776 	status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
777 			(void __user *)(uintptr_t)cmd.addr, cmd.data_len,
778 			&cmd.result, timeout);
779 	if (status >= 0) {
780 		if (put_user(cmd.result, &ucmd->result))
781 			return -EFAULT;
782 	}
783 
784 	return status;
785 }
786 
787 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
788 		unsigned int cmd, unsigned long arg)
789 {
790 	struct nvme_ns *ns = bdev->bd_disk->private_data;
791 
792 	switch (cmd) {
793 	case NVME_IOCTL_ID:
794 		force_successful_syscall_return();
795 		return ns->ns_id;
796 	case NVME_IOCTL_ADMIN_CMD:
797 		return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
798 	case NVME_IOCTL_IO_CMD:
799 		return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
800 	case NVME_IOCTL_SUBMIT_IO:
801 		return nvme_submit_io(ns, (void __user *)arg);
802 #ifdef CONFIG_BLK_DEV_NVME_SCSI
803 	case SG_GET_VERSION_NUM:
804 		return nvme_sg_get_version_num((void __user *)arg);
805 	case SG_IO:
806 		return nvme_sg_io(ns, (void __user *)arg);
807 #endif
808 	default:
809 #ifdef CONFIG_NVM
810 		if (ns->ndev)
811 			return nvme_nvm_ioctl(ns, cmd, arg);
812 #endif
813 		if (is_sed_ioctl(cmd))
814 			return sed_ioctl(ns->ctrl->opal_dev, cmd,
815 					 (void __user *) arg);
816 		return -ENOTTY;
817 	}
818 }
819 
820 #ifdef CONFIG_COMPAT
821 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
822 			unsigned int cmd, unsigned long arg)
823 {
824 	switch (cmd) {
825 	case SG_IO:
826 		return -ENOIOCTLCMD;
827 	}
828 	return nvme_ioctl(bdev, mode, cmd, arg);
829 }
830 #else
831 #define nvme_compat_ioctl	NULL
832 #endif
833 
834 static int nvme_open(struct block_device *bdev, fmode_t mode)
835 {
836 	return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
837 }
838 
839 static void nvme_release(struct gendisk *disk, fmode_t mode)
840 {
841 	struct nvme_ns *ns = disk->private_data;
842 
843 	module_put(ns->ctrl->ops->module);
844 	nvme_put_ns(ns);
845 }
846 
847 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
848 {
849 	/* some standard values */
850 	geo->heads = 1 << 6;
851 	geo->sectors = 1 << 5;
852 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
853 	return 0;
854 }
855 
856 #ifdef CONFIG_BLK_DEV_INTEGRITY
857 static void nvme_init_integrity(struct nvme_ns *ns)
858 {
859 	struct blk_integrity integrity;
860 
861 	memset(&integrity, 0, sizeof(integrity));
862 	switch (ns->pi_type) {
863 	case NVME_NS_DPS_PI_TYPE3:
864 		integrity.profile = &t10_pi_type3_crc;
865 		integrity.tag_size = sizeof(u16) + sizeof(u32);
866 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
867 		break;
868 	case NVME_NS_DPS_PI_TYPE1:
869 	case NVME_NS_DPS_PI_TYPE2:
870 		integrity.profile = &t10_pi_type1_crc;
871 		integrity.tag_size = sizeof(u16);
872 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
873 		break;
874 	default:
875 		integrity.profile = NULL;
876 		break;
877 	}
878 	integrity.tuple_size = ns->ms;
879 	blk_integrity_register(ns->disk, &integrity);
880 	blk_queue_max_integrity_segments(ns->queue, 1);
881 }
882 #else
883 static void nvme_init_integrity(struct nvme_ns *ns)
884 {
885 }
886 #endif /* CONFIG_BLK_DEV_INTEGRITY */
887 
888 static void nvme_config_discard(struct nvme_ns *ns)
889 {
890 	struct nvme_ctrl *ctrl = ns->ctrl;
891 	u32 logical_block_size = queue_logical_block_size(ns->queue);
892 
893 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
894 			NVME_DSM_MAX_RANGES);
895 
896 	if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
897 		ns->queue->limits.discard_zeroes_data = 1;
898 	else
899 		ns->queue->limits.discard_zeroes_data = 0;
900 
901 	ns->queue->limits.discard_alignment = logical_block_size;
902 	ns->queue->limits.discard_granularity = logical_block_size;
903 	blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
904 	blk_queue_max_discard_segments(ns->queue, NVME_DSM_MAX_RANGES);
905 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
906 }
907 
908 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
909 {
910 	if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
911 		dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
912 		return -ENODEV;
913 	}
914 
915 	if ((*id)->ncap == 0) {
916 		kfree(*id);
917 		return -ENODEV;
918 	}
919 
920 	if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
921 		memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
922 	if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
923 		memcpy(ns->uuid, (*id)->nguid, sizeof(ns->uuid));
924 
925 	return 0;
926 }
927 
928 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
929 {
930 	struct nvme_ns *ns = disk->private_data;
931 	u8 lbaf, pi_type;
932 	u16 old_ms;
933 	unsigned short bs;
934 
935 	old_ms = ns->ms;
936 	lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
937 	ns->lba_shift = id->lbaf[lbaf].ds;
938 	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
939 	ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
940 
941 	/*
942 	 * If identify namespace failed, use default 512 byte block size so
943 	 * block layer can use before failing read/write for 0 capacity.
944 	 */
945 	if (ns->lba_shift == 0)
946 		ns->lba_shift = 9;
947 	bs = 1 << ns->lba_shift;
948 	/* XXX: PI implementation requires metadata equal t10 pi tuple size */
949 	pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
950 					id->dps & NVME_NS_DPS_PI_MASK : 0;
951 
952 	blk_mq_freeze_queue(disk->queue);
953 	if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
954 				ns->ms != old_ms ||
955 				bs != queue_logical_block_size(disk->queue) ||
956 				(ns->ms && ns->ext)))
957 		blk_integrity_unregister(disk);
958 
959 	ns->pi_type = pi_type;
960 	blk_queue_logical_block_size(ns->queue, bs);
961 
962 	if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
963 		nvme_init_integrity(ns);
964 	if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
965 		set_capacity(disk, 0);
966 	else
967 		set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
968 
969 	if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
970 		nvme_config_discard(ns);
971 	blk_mq_unfreeze_queue(disk->queue);
972 }
973 
974 static int nvme_revalidate_disk(struct gendisk *disk)
975 {
976 	struct nvme_ns *ns = disk->private_data;
977 	struct nvme_id_ns *id = NULL;
978 	int ret;
979 
980 	if (test_bit(NVME_NS_DEAD, &ns->flags)) {
981 		set_capacity(disk, 0);
982 		return -ENODEV;
983 	}
984 
985 	ret = nvme_revalidate_ns(ns, &id);
986 	if (ret)
987 		return ret;
988 
989 	__nvme_revalidate_disk(disk, id);
990 	kfree(id);
991 
992 	return 0;
993 }
994 
995 static char nvme_pr_type(enum pr_type type)
996 {
997 	switch (type) {
998 	case PR_WRITE_EXCLUSIVE:
999 		return 1;
1000 	case PR_EXCLUSIVE_ACCESS:
1001 		return 2;
1002 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1003 		return 3;
1004 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1005 		return 4;
1006 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1007 		return 5;
1008 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1009 		return 6;
1010 	default:
1011 		return 0;
1012 	}
1013 };
1014 
1015 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1016 				u64 key, u64 sa_key, u8 op)
1017 {
1018 	struct nvme_ns *ns = bdev->bd_disk->private_data;
1019 	struct nvme_command c;
1020 	u8 data[16] = { 0, };
1021 
1022 	put_unaligned_le64(key, &data[0]);
1023 	put_unaligned_le64(sa_key, &data[8]);
1024 
1025 	memset(&c, 0, sizeof(c));
1026 	c.common.opcode = op;
1027 	c.common.nsid = cpu_to_le32(ns->ns_id);
1028 	c.common.cdw10[0] = cpu_to_le32(cdw10);
1029 
1030 	return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1031 }
1032 
1033 static int nvme_pr_register(struct block_device *bdev, u64 old,
1034 		u64 new, unsigned flags)
1035 {
1036 	u32 cdw10;
1037 
1038 	if (flags & ~PR_FL_IGNORE_KEY)
1039 		return -EOPNOTSUPP;
1040 
1041 	cdw10 = old ? 2 : 0;
1042 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1043 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1044 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1045 }
1046 
1047 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1048 		enum pr_type type, unsigned flags)
1049 {
1050 	u32 cdw10;
1051 
1052 	if (flags & ~PR_FL_IGNORE_KEY)
1053 		return -EOPNOTSUPP;
1054 
1055 	cdw10 = nvme_pr_type(type) << 8;
1056 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1057 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1058 }
1059 
1060 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1061 		enum pr_type type, bool abort)
1062 {
1063 	u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1064 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1065 }
1066 
1067 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1068 {
1069 	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1070 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1071 }
1072 
1073 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1074 {
1075 	u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1076 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1077 }
1078 
1079 static const struct pr_ops nvme_pr_ops = {
1080 	.pr_register	= nvme_pr_register,
1081 	.pr_reserve	= nvme_pr_reserve,
1082 	.pr_release	= nvme_pr_release,
1083 	.pr_preempt	= nvme_pr_preempt,
1084 	.pr_clear	= nvme_pr_clear,
1085 };
1086 
1087 #ifdef CONFIG_BLK_SED_OPAL
1088 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1089 		bool send)
1090 {
1091 	struct nvme_ctrl *ctrl = data;
1092 	struct nvme_command cmd;
1093 
1094 	memset(&cmd, 0, sizeof(cmd));
1095 	if (send)
1096 		cmd.common.opcode = nvme_admin_security_send;
1097 	else
1098 		cmd.common.opcode = nvme_admin_security_recv;
1099 	cmd.common.nsid = 0;
1100 	cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1101 	cmd.common.cdw10[1] = cpu_to_le32(len);
1102 
1103 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1104 				      ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1105 }
1106 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1107 #endif /* CONFIG_BLK_SED_OPAL */
1108 
1109 static const struct block_device_operations nvme_fops = {
1110 	.owner		= THIS_MODULE,
1111 	.ioctl		= nvme_ioctl,
1112 	.compat_ioctl	= nvme_compat_ioctl,
1113 	.open		= nvme_open,
1114 	.release	= nvme_release,
1115 	.getgeo		= nvme_getgeo,
1116 	.revalidate_disk= nvme_revalidate_disk,
1117 	.pr_ops		= &nvme_pr_ops,
1118 };
1119 
1120 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1121 {
1122 	unsigned long timeout =
1123 		((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1124 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1125 	int ret;
1126 
1127 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1128 		if (csts == ~0)
1129 			return -ENODEV;
1130 		if ((csts & NVME_CSTS_RDY) == bit)
1131 			break;
1132 
1133 		msleep(100);
1134 		if (fatal_signal_pending(current))
1135 			return -EINTR;
1136 		if (time_after(jiffies, timeout)) {
1137 			dev_err(ctrl->device,
1138 				"Device not ready; aborting %s\n", enabled ?
1139 						"initialisation" : "reset");
1140 			return -ENODEV;
1141 		}
1142 	}
1143 
1144 	return ret;
1145 }
1146 
1147 /*
1148  * If the device has been passed off to us in an enabled state, just clear
1149  * the enabled bit.  The spec says we should set the 'shutdown notification
1150  * bits', but doing so may cause the device to complete commands to the
1151  * admin queue ... and we don't know what memory that might be pointing at!
1152  */
1153 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1154 {
1155 	int ret;
1156 
1157 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1158 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1159 
1160 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1161 	if (ret)
1162 		return ret;
1163 
1164 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1165 		msleep(NVME_QUIRK_DELAY_AMOUNT);
1166 
1167 	return nvme_wait_ready(ctrl, cap, false);
1168 }
1169 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1170 
1171 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1172 {
1173 	/*
1174 	 * Default to a 4K page size, with the intention to update this
1175 	 * path in the future to accomodate architectures with differing
1176 	 * kernel and IO page sizes.
1177 	 */
1178 	unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1179 	int ret;
1180 
1181 	if (page_shift < dev_page_min) {
1182 		dev_err(ctrl->device,
1183 			"Minimum device page size %u too large for host (%u)\n",
1184 			1 << dev_page_min, 1 << page_shift);
1185 		return -ENODEV;
1186 	}
1187 
1188 	ctrl->page_size = 1 << page_shift;
1189 
1190 	ctrl->ctrl_config = NVME_CC_CSS_NVM;
1191 	ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1192 	ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1193 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1194 	ctrl->ctrl_config |= NVME_CC_ENABLE;
1195 
1196 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1197 	if (ret)
1198 		return ret;
1199 	return nvme_wait_ready(ctrl, cap, true);
1200 }
1201 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1202 
1203 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1204 {
1205 	unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1206 	u32 csts;
1207 	int ret;
1208 
1209 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1210 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1211 
1212 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1213 	if (ret)
1214 		return ret;
1215 
1216 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1217 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1218 			break;
1219 
1220 		msleep(100);
1221 		if (fatal_signal_pending(current))
1222 			return -EINTR;
1223 		if (time_after(jiffies, timeout)) {
1224 			dev_err(ctrl->device,
1225 				"Device shutdown incomplete; abort shutdown\n");
1226 			return -ENODEV;
1227 		}
1228 	}
1229 
1230 	return ret;
1231 }
1232 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1233 
1234 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1235 		struct request_queue *q)
1236 {
1237 	bool vwc = false;
1238 
1239 	if (ctrl->max_hw_sectors) {
1240 		u32 max_segments =
1241 			(ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1242 
1243 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1244 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1245 	}
1246 	if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1247 		blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1248 	blk_queue_virt_boundary(q, ctrl->page_size - 1);
1249 	if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1250 		vwc = true;
1251 	blk_queue_write_cache(q, vwc, vwc);
1252 }
1253 
1254 /*
1255  * Initialize the cached copies of the Identify data and various controller
1256  * register in our nvme_ctrl structure.  This should be called as soon as
1257  * the admin queue is fully up and running.
1258  */
1259 int nvme_init_identify(struct nvme_ctrl *ctrl)
1260 {
1261 	struct nvme_id_ctrl *id;
1262 	u64 cap;
1263 	int ret, page_shift;
1264 	u32 max_hw_sectors;
1265 
1266 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1267 	if (ret) {
1268 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1269 		return ret;
1270 	}
1271 
1272 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1273 	if (ret) {
1274 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1275 		return ret;
1276 	}
1277 	page_shift = NVME_CAP_MPSMIN(cap) + 12;
1278 
1279 	if (ctrl->vs >= NVME_VS(1, 1, 0))
1280 		ctrl->subsystem = NVME_CAP_NSSRC(cap);
1281 
1282 	ret = nvme_identify_ctrl(ctrl, &id);
1283 	if (ret) {
1284 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1285 		return -EIO;
1286 	}
1287 
1288 	ctrl->oacs = le16_to_cpu(id->oacs);
1289 	ctrl->vid = le16_to_cpu(id->vid);
1290 	ctrl->oncs = le16_to_cpup(&id->oncs);
1291 	atomic_set(&ctrl->abort_limit, id->acl + 1);
1292 	ctrl->vwc = id->vwc;
1293 	ctrl->cntlid = le16_to_cpup(&id->cntlid);
1294 	memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1295 	memcpy(ctrl->model, id->mn, sizeof(id->mn));
1296 	memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1297 	if (id->mdts)
1298 		max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1299 	else
1300 		max_hw_sectors = UINT_MAX;
1301 	ctrl->max_hw_sectors =
1302 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1303 
1304 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
1305 	ctrl->sgls = le32_to_cpu(id->sgls);
1306 	ctrl->kas = le16_to_cpu(id->kas);
1307 
1308 	if (ctrl->ops->is_fabrics) {
1309 		ctrl->icdoff = le16_to_cpu(id->icdoff);
1310 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1311 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1312 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1313 
1314 		/*
1315 		 * In fabrics we need to verify the cntlid matches the
1316 		 * admin connect
1317 		 */
1318 		if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1319 			ret = -EINVAL;
1320 
1321 		if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1322 			dev_err(ctrl->dev,
1323 				"keep-alive support is mandatory for fabrics\n");
1324 			ret = -EINVAL;
1325 		}
1326 	} else {
1327 		ctrl->cntlid = le16_to_cpu(id->cntlid);
1328 	}
1329 
1330 	kfree(id);
1331 	return ret;
1332 }
1333 EXPORT_SYMBOL_GPL(nvme_init_identify);
1334 
1335 static int nvme_dev_open(struct inode *inode, struct file *file)
1336 {
1337 	struct nvme_ctrl *ctrl;
1338 	int instance = iminor(inode);
1339 	int ret = -ENODEV;
1340 
1341 	spin_lock(&dev_list_lock);
1342 	list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1343 		if (ctrl->instance != instance)
1344 			continue;
1345 
1346 		if (!ctrl->admin_q) {
1347 			ret = -EWOULDBLOCK;
1348 			break;
1349 		}
1350 		if (!kref_get_unless_zero(&ctrl->kref))
1351 			break;
1352 		file->private_data = ctrl;
1353 		ret = 0;
1354 		break;
1355 	}
1356 	spin_unlock(&dev_list_lock);
1357 
1358 	return ret;
1359 }
1360 
1361 static int nvme_dev_release(struct inode *inode, struct file *file)
1362 {
1363 	nvme_put_ctrl(file->private_data);
1364 	return 0;
1365 }
1366 
1367 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1368 {
1369 	struct nvme_ns *ns;
1370 	int ret;
1371 
1372 	mutex_lock(&ctrl->namespaces_mutex);
1373 	if (list_empty(&ctrl->namespaces)) {
1374 		ret = -ENOTTY;
1375 		goto out_unlock;
1376 	}
1377 
1378 	ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1379 	if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1380 		dev_warn(ctrl->device,
1381 			"NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1382 		ret = -EINVAL;
1383 		goto out_unlock;
1384 	}
1385 
1386 	dev_warn(ctrl->device,
1387 		"using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1388 	kref_get(&ns->kref);
1389 	mutex_unlock(&ctrl->namespaces_mutex);
1390 
1391 	ret = nvme_user_cmd(ctrl, ns, argp);
1392 	nvme_put_ns(ns);
1393 	return ret;
1394 
1395 out_unlock:
1396 	mutex_unlock(&ctrl->namespaces_mutex);
1397 	return ret;
1398 }
1399 
1400 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1401 		unsigned long arg)
1402 {
1403 	struct nvme_ctrl *ctrl = file->private_data;
1404 	void __user *argp = (void __user *)arg;
1405 
1406 	switch (cmd) {
1407 	case NVME_IOCTL_ADMIN_CMD:
1408 		return nvme_user_cmd(ctrl, NULL, argp);
1409 	case NVME_IOCTL_IO_CMD:
1410 		return nvme_dev_user_cmd(ctrl, argp);
1411 	case NVME_IOCTL_RESET:
1412 		dev_warn(ctrl->device, "resetting controller\n");
1413 		return ctrl->ops->reset_ctrl(ctrl);
1414 	case NVME_IOCTL_SUBSYS_RESET:
1415 		return nvme_reset_subsystem(ctrl);
1416 	case NVME_IOCTL_RESCAN:
1417 		nvme_queue_scan(ctrl);
1418 		return 0;
1419 	default:
1420 		return -ENOTTY;
1421 	}
1422 }
1423 
1424 static const struct file_operations nvme_dev_fops = {
1425 	.owner		= THIS_MODULE,
1426 	.open		= nvme_dev_open,
1427 	.release	= nvme_dev_release,
1428 	.unlocked_ioctl	= nvme_dev_ioctl,
1429 	.compat_ioctl	= nvme_dev_ioctl,
1430 };
1431 
1432 static ssize_t nvme_sysfs_reset(struct device *dev,
1433 				struct device_attribute *attr, const char *buf,
1434 				size_t count)
1435 {
1436 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1437 	int ret;
1438 
1439 	ret = ctrl->ops->reset_ctrl(ctrl);
1440 	if (ret < 0)
1441 		return ret;
1442 	return count;
1443 }
1444 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1445 
1446 static ssize_t nvme_sysfs_rescan(struct device *dev,
1447 				struct device_attribute *attr, const char *buf,
1448 				size_t count)
1449 {
1450 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1451 
1452 	nvme_queue_scan(ctrl);
1453 	return count;
1454 }
1455 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1456 
1457 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1458 								char *buf)
1459 {
1460 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1461 	struct nvme_ctrl *ctrl = ns->ctrl;
1462 	int serial_len = sizeof(ctrl->serial);
1463 	int model_len = sizeof(ctrl->model);
1464 
1465 	if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1466 		return sprintf(buf, "eui.%16phN\n", ns->uuid);
1467 
1468 	if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1469 		return sprintf(buf, "eui.%8phN\n", ns->eui);
1470 
1471 	while (ctrl->serial[serial_len - 1] == ' ')
1472 		serial_len--;
1473 	while (ctrl->model[model_len - 1] == ' ')
1474 		model_len--;
1475 
1476 	return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1477 		serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1478 }
1479 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1480 
1481 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1482 								char *buf)
1483 {
1484 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1485 	return sprintf(buf, "%pU\n", ns->uuid);
1486 }
1487 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1488 
1489 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1490 								char *buf)
1491 {
1492 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1493 	return sprintf(buf, "%8phd\n", ns->eui);
1494 }
1495 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1496 
1497 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1498 								char *buf)
1499 {
1500 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1501 	return sprintf(buf, "%d\n", ns->ns_id);
1502 }
1503 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1504 
1505 static struct attribute *nvme_ns_attrs[] = {
1506 	&dev_attr_wwid.attr,
1507 	&dev_attr_uuid.attr,
1508 	&dev_attr_eui.attr,
1509 	&dev_attr_nsid.attr,
1510 	NULL,
1511 };
1512 
1513 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1514 		struct attribute *a, int n)
1515 {
1516 	struct device *dev = container_of(kobj, struct device, kobj);
1517 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1518 
1519 	if (a == &dev_attr_uuid.attr) {
1520 		if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1521 			return 0;
1522 	}
1523 	if (a == &dev_attr_eui.attr) {
1524 		if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1525 			return 0;
1526 	}
1527 	return a->mode;
1528 }
1529 
1530 static const struct attribute_group nvme_ns_attr_group = {
1531 	.attrs		= nvme_ns_attrs,
1532 	.is_visible	= nvme_ns_attrs_are_visible,
1533 };
1534 
1535 #define nvme_show_str_function(field)						\
1536 static ssize_t  field##_show(struct device *dev,				\
1537 			    struct device_attribute *attr, char *buf)		\
1538 {										\
1539         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
1540         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);	\
1541 }										\
1542 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1543 
1544 #define nvme_show_int_function(field)						\
1545 static ssize_t  field##_show(struct device *dev,				\
1546 			    struct device_attribute *attr, char *buf)		\
1547 {										\
1548         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
1549         return sprintf(buf, "%d\n", ctrl->field);	\
1550 }										\
1551 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1552 
1553 nvme_show_str_function(model);
1554 nvme_show_str_function(serial);
1555 nvme_show_str_function(firmware_rev);
1556 nvme_show_int_function(cntlid);
1557 
1558 static ssize_t nvme_sysfs_delete(struct device *dev,
1559 				struct device_attribute *attr, const char *buf,
1560 				size_t count)
1561 {
1562 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1563 
1564 	if (device_remove_file_self(dev, attr))
1565 		ctrl->ops->delete_ctrl(ctrl);
1566 	return count;
1567 }
1568 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1569 
1570 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1571 					 struct device_attribute *attr,
1572 					 char *buf)
1573 {
1574 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1575 
1576 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1577 }
1578 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1579 
1580 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1581 					 struct device_attribute *attr,
1582 					 char *buf)
1583 {
1584 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1585 
1586 	return snprintf(buf, PAGE_SIZE, "%s\n",
1587 			ctrl->ops->get_subsysnqn(ctrl));
1588 }
1589 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1590 
1591 static ssize_t nvme_sysfs_show_address(struct device *dev,
1592 					 struct device_attribute *attr,
1593 					 char *buf)
1594 {
1595 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1596 
1597 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1598 }
1599 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1600 
1601 static struct attribute *nvme_dev_attrs[] = {
1602 	&dev_attr_reset_controller.attr,
1603 	&dev_attr_rescan_controller.attr,
1604 	&dev_attr_model.attr,
1605 	&dev_attr_serial.attr,
1606 	&dev_attr_firmware_rev.attr,
1607 	&dev_attr_cntlid.attr,
1608 	&dev_attr_delete_controller.attr,
1609 	&dev_attr_transport.attr,
1610 	&dev_attr_subsysnqn.attr,
1611 	&dev_attr_address.attr,
1612 	NULL
1613 };
1614 
1615 #define CHECK_ATTR(ctrl, a, name)		\
1616 	if ((a) == &dev_attr_##name.attr &&	\
1617 	    !(ctrl)->ops->get_##name)		\
1618 		return 0
1619 
1620 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1621 		struct attribute *a, int n)
1622 {
1623 	struct device *dev = container_of(kobj, struct device, kobj);
1624 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1625 
1626 	if (a == &dev_attr_delete_controller.attr) {
1627 		if (!ctrl->ops->delete_ctrl)
1628 			return 0;
1629 	}
1630 
1631 	CHECK_ATTR(ctrl, a, subsysnqn);
1632 	CHECK_ATTR(ctrl, a, address);
1633 
1634 	return a->mode;
1635 }
1636 
1637 static struct attribute_group nvme_dev_attrs_group = {
1638 	.attrs		= nvme_dev_attrs,
1639 	.is_visible	= nvme_dev_attrs_are_visible,
1640 };
1641 
1642 static const struct attribute_group *nvme_dev_attr_groups[] = {
1643 	&nvme_dev_attrs_group,
1644 	NULL,
1645 };
1646 
1647 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1648 {
1649 	struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1650 	struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1651 
1652 	return nsa->ns_id - nsb->ns_id;
1653 }
1654 
1655 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1656 {
1657 	struct nvme_ns *ns, *ret = NULL;
1658 
1659 	mutex_lock(&ctrl->namespaces_mutex);
1660 	list_for_each_entry(ns, &ctrl->namespaces, list) {
1661 		if (ns->ns_id == nsid) {
1662 			kref_get(&ns->kref);
1663 			ret = ns;
1664 			break;
1665 		}
1666 		if (ns->ns_id > nsid)
1667 			break;
1668 	}
1669 	mutex_unlock(&ctrl->namespaces_mutex);
1670 	return ret;
1671 }
1672 
1673 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1674 {
1675 	struct nvme_ns *ns;
1676 	struct gendisk *disk;
1677 	struct nvme_id_ns *id;
1678 	char disk_name[DISK_NAME_LEN];
1679 	int node = dev_to_node(ctrl->dev);
1680 
1681 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1682 	if (!ns)
1683 		return;
1684 
1685 	ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1686 	if (ns->instance < 0)
1687 		goto out_free_ns;
1688 
1689 	ns->queue = blk_mq_init_queue(ctrl->tagset);
1690 	if (IS_ERR(ns->queue))
1691 		goto out_release_instance;
1692 	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1693 	ns->queue->queuedata = ns;
1694 	ns->ctrl = ctrl;
1695 
1696 	kref_init(&ns->kref);
1697 	ns->ns_id = nsid;
1698 	ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1699 
1700 	blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1701 	nvme_set_queue_limits(ctrl, ns->queue);
1702 
1703 	sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1704 
1705 	if (nvme_revalidate_ns(ns, &id))
1706 		goto out_free_queue;
1707 
1708 	if (nvme_nvm_ns_supported(ns, id) &&
1709 				nvme_nvm_register(ns, disk_name, node)) {
1710 		dev_warn(ctrl->dev, "%s: LightNVM init failure\n", __func__);
1711 		goto out_free_id;
1712 	}
1713 
1714 	disk = alloc_disk_node(0, node);
1715 	if (!disk)
1716 		goto out_free_id;
1717 
1718 	disk->fops = &nvme_fops;
1719 	disk->private_data = ns;
1720 	disk->queue = ns->queue;
1721 	disk->flags = GENHD_FL_EXT_DEVT;
1722 	memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
1723 	ns->disk = disk;
1724 
1725 	__nvme_revalidate_disk(disk, id);
1726 
1727 	mutex_lock(&ctrl->namespaces_mutex);
1728 	list_add_tail(&ns->list, &ctrl->namespaces);
1729 	mutex_unlock(&ctrl->namespaces_mutex);
1730 
1731 	kref_get(&ctrl->kref);
1732 
1733 	kfree(id);
1734 
1735 	device_add_disk(ctrl->device, ns->disk);
1736 	if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1737 					&nvme_ns_attr_group))
1738 		pr_warn("%s: failed to create sysfs group for identification\n",
1739 			ns->disk->disk_name);
1740 	if (ns->ndev && nvme_nvm_register_sysfs(ns))
1741 		pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
1742 			ns->disk->disk_name);
1743 	return;
1744  out_free_id:
1745 	kfree(id);
1746  out_free_queue:
1747 	blk_cleanup_queue(ns->queue);
1748  out_release_instance:
1749 	ida_simple_remove(&ctrl->ns_ida, ns->instance);
1750  out_free_ns:
1751 	kfree(ns);
1752 }
1753 
1754 static void nvme_ns_remove(struct nvme_ns *ns)
1755 {
1756 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1757 		return;
1758 
1759 	if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
1760 		if (blk_get_integrity(ns->disk))
1761 			blk_integrity_unregister(ns->disk);
1762 		sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1763 					&nvme_ns_attr_group);
1764 		if (ns->ndev)
1765 			nvme_nvm_unregister_sysfs(ns);
1766 		del_gendisk(ns->disk);
1767 		blk_mq_abort_requeue_list(ns->queue);
1768 		blk_cleanup_queue(ns->queue);
1769 	}
1770 
1771 	mutex_lock(&ns->ctrl->namespaces_mutex);
1772 	list_del_init(&ns->list);
1773 	mutex_unlock(&ns->ctrl->namespaces_mutex);
1774 
1775 	nvme_put_ns(ns);
1776 }
1777 
1778 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1779 {
1780 	struct nvme_ns *ns;
1781 
1782 	ns = nvme_find_get_ns(ctrl, nsid);
1783 	if (ns) {
1784 		if (ns->disk && revalidate_disk(ns->disk))
1785 			nvme_ns_remove(ns);
1786 		nvme_put_ns(ns);
1787 	} else
1788 		nvme_alloc_ns(ctrl, nsid);
1789 }
1790 
1791 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
1792 					unsigned nsid)
1793 {
1794 	struct nvme_ns *ns, *next;
1795 
1796 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1797 		if (ns->ns_id > nsid)
1798 			nvme_ns_remove(ns);
1799 	}
1800 }
1801 
1802 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1803 {
1804 	struct nvme_ns *ns;
1805 	__le32 *ns_list;
1806 	unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1807 	int ret = 0;
1808 
1809 	ns_list = kzalloc(0x1000, GFP_KERNEL);
1810 	if (!ns_list)
1811 		return -ENOMEM;
1812 
1813 	for (i = 0; i < num_lists; i++) {
1814 		ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1815 		if (ret)
1816 			goto free;
1817 
1818 		for (j = 0; j < min(nn, 1024U); j++) {
1819 			nsid = le32_to_cpu(ns_list[j]);
1820 			if (!nsid)
1821 				goto out;
1822 
1823 			nvme_validate_ns(ctrl, nsid);
1824 
1825 			while (++prev < nsid) {
1826 				ns = nvme_find_get_ns(ctrl, prev);
1827 				if (ns) {
1828 					nvme_ns_remove(ns);
1829 					nvme_put_ns(ns);
1830 				}
1831 			}
1832 		}
1833 		nn -= j;
1834 	}
1835  out:
1836 	nvme_remove_invalid_namespaces(ctrl, prev);
1837  free:
1838 	kfree(ns_list);
1839 	return ret;
1840 }
1841 
1842 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
1843 {
1844 	unsigned i;
1845 
1846 	for (i = 1; i <= nn; i++)
1847 		nvme_validate_ns(ctrl, i);
1848 
1849 	nvme_remove_invalid_namespaces(ctrl, nn);
1850 }
1851 
1852 static void nvme_scan_work(struct work_struct *work)
1853 {
1854 	struct nvme_ctrl *ctrl =
1855 		container_of(work, struct nvme_ctrl, scan_work);
1856 	struct nvme_id_ctrl *id;
1857 	unsigned nn;
1858 
1859 	if (ctrl->state != NVME_CTRL_LIVE)
1860 		return;
1861 
1862 	if (nvme_identify_ctrl(ctrl, &id))
1863 		return;
1864 
1865 	nn = le32_to_cpu(id->nn);
1866 	if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1867 	    !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1868 		if (!nvme_scan_ns_list(ctrl, nn))
1869 			goto done;
1870 	}
1871 	nvme_scan_ns_sequential(ctrl, nn);
1872  done:
1873 	mutex_lock(&ctrl->namespaces_mutex);
1874 	list_sort(NULL, &ctrl->namespaces, ns_cmp);
1875 	mutex_unlock(&ctrl->namespaces_mutex);
1876 	kfree(id);
1877 }
1878 
1879 void nvme_queue_scan(struct nvme_ctrl *ctrl)
1880 {
1881 	/*
1882 	 * Do not queue new scan work when a controller is reset during
1883 	 * removal.
1884 	 */
1885 	if (ctrl->state == NVME_CTRL_LIVE)
1886 		schedule_work(&ctrl->scan_work);
1887 }
1888 EXPORT_SYMBOL_GPL(nvme_queue_scan);
1889 
1890 /*
1891  * This function iterates the namespace list unlocked to allow recovery from
1892  * controller failure. It is up to the caller to ensure the namespace list is
1893  * not modified by scan work while this function is executing.
1894  */
1895 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1896 {
1897 	struct nvme_ns *ns, *next;
1898 
1899 	/*
1900 	 * The dead states indicates the controller was not gracefully
1901 	 * disconnected. In that case, we won't be able to flush any data while
1902 	 * removing the namespaces' disks; fail all the queues now to avoid
1903 	 * potentially having to clean up the failed sync later.
1904 	 */
1905 	if (ctrl->state == NVME_CTRL_DEAD)
1906 		nvme_kill_queues(ctrl);
1907 
1908 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1909 		nvme_ns_remove(ns);
1910 }
1911 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1912 
1913 static void nvme_async_event_work(struct work_struct *work)
1914 {
1915 	struct nvme_ctrl *ctrl =
1916 		container_of(work, struct nvme_ctrl, async_event_work);
1917 
1918 	spin_lock_irq(&ctrl->lock);
1919 	while (ctrl->event_limit > 0) {
1920 		int aer_idx = --ctrl->event_limit;
1921 
1922 		spin_unlock_irq(&ctrl->lock);
1923 		ctrl->ops->submit_async_event(ctrl, aer_idx);
1924 		spin_lock_irq(&ctrl->lock);
1925 	}
1926 	spin_unlock_irq(&ctrl->lock);
1927 }
1928 
1929 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
1930 		union nvme_result *res)
1931 {
1932 	u32 result = le32_to_cpu(res->u32);
1933 	bool done = true;
1934 
1935 	switch (le16_to_cpu(status) >> 1) {
1936 	case NVME_SC_SUCCESS:
1937 		done = false;
1938 		/*FALLTHRU*/
1939 	case NVME_SC_ABORT_REQ:
1940 		++ctrl->event_limit;
1941 		schedule_work(&ctrl->async_event_work);
1942 		break;
1943 	default:
1944 		break;
1945 	}
1946 
1947 	if (done)
1948 		return;
1949 
1950 	switch (result & 0xff07) {
1951 	case NVME_AER_NOTICE_NS_CHANGED:
1952 		dev_info(ctrl->device, "rescanning\n");
1953 		nvme_queue_scan(ctrl);
1954 		break;
1955 	default:
1956 		dev_warn(ctrl->device, "async event result %08x\n", result);
1957 	}
1958 }
1959 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
1960 
1961 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
1962 {
1963 	ctrl->event_limit = NVME_NR_AERS;
1964 	schedule_work(&ctrl->async_event_work);
1965 }
1966 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
1967 
1968 static DEFINE_IDA(nvme_instance_ida);
1969 
1970 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1971 {
1972 	int instance, error;
1973 
1974 	do {
1975 		if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1976 			return -ENODEV;
1977 
1978 		spin_lock(&dev_list_lock);
1979 		error = ida_get_new(&nvme_instance_ida, &instance);
1980 		spin_unlock(&dev_list_lock);
1981 	} while (error == -EAGAIN);
1982 
1983 	if (error)
1984 		return -ENODEV;
1985 
1986 	ctrl->instance = instance;
1987 	return 0;
1988 }
1989 
1990 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1991 {
1992 	spin_lock(&dev_list_lock);
1993 	ida_remove(&nvme_instance_ida, ctrl->instance);
1994 	spin_unlock(&dev_list_lock);
1995 }
1996 
1997 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1998 {
1999 	flush_work(&ctrl->async_event_work);
2000 	flush_work(&ctrl->scan_work);
2001 	nvme_remove_namespaces(ctrl);
2002 
2003 	device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
2004 
2005 	spin_lock(&dev_list_lock);
2006 	list_del(&ctrl->node);
2007 	spin_unlock(&dev_list_lock);
2008 }
2009 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
2010 
2011 static void nvme_free_ctrl(struct kref *kref)
2012 {
2013 	struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
2014 
2015 	put_device(ctrl->device);
2016 	nvme_release_instance(ctrl);
2017 	ida_destroy(&ctrl->ns_ida);
2018 
2019 	ctrl->ops->free_ctrl(ctrl);
2020 }
2021 
2022 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
2023 {
2024 	kref_put(&ctrl->kref, nvme_free_ctrl);
2025 }
2026 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2027 
2028 /*
2029  * Initialize a NVMe controller structures.  This needs to be called during
2030  * earliest initialization so that we have the initialized structured around
2031  * during probing.
2032  */
2033 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2034 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
2035 {
2036 	int ret;
2037 
2038 	ctrl->state = NVME_CTRL_NEW;
2039 	spin_lock_init(&ctrl->lock);
2040 	INIT_LIST_HEAD(&ctrl->namespaces);
2041 	mutex_init(&ctrl->namespaces_mutex);
2042 	kref_init(&ctrl->kref);
2043 	ctrl->dev = dev;
2044 	ctrl->ops = ops;
2045 	ctrl->quirks = quirks;
2046 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2047 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2048 
2049 	ret = nvme_set_instance(ctrl);
2050 	if (ret)
2051 		goto out;
2052 
2053 	ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2054 				MKDEV(nvme_char_major, ctrl->instance),
2055 				ctrl, nvme_dev_attr_groups,
2056 				"nvme%d", ctrl->instance);
2057 	if (IS_ERR(ctrl->device)) {
2058 		ret = PTR_ERR(ctrl->device);
2059 		goto out_release_instance;
2060 	}
2061 	get_device(ctrl->device);
2062 	ida_init(&ctrl->ns_ida);
2063 
2064 	spin_lock(&dev_list_lock);
2065 	list_add_tail(&ctrl->node, &nvme_ctrl_list);
2066 	spin_unlock(&dev_list_lock);
2067 
2068 	return 0;
2069 out_release_instance:
2070 	nvme_release_instance(ctrl);
2071 out:
2072 	return ret;
2073 }
2074 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2075 
2076 /**
2077  * nvme_kill_queues(): Ends all namespace queues
2078  * @ctrl: the dead controller that needs to end
2079  *
2080  * Call this function when the driver determines it is unable to get the
2081  * controller in a state capable of servicing IO.
2082  */
2083 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2084 {
2085 	struct nvme_ns *ns;
2086 
2087 	mutex_lock(&ctrl->namespaces_mutex);
2088 	list_for_each_entry(ns, &ctrl->namespaces, list) {
2089 		/*
2090 		 * Revalidating a dead namespace sets capacity to 0. This will
2091 		 * end buffered writers dirtying pages that can't be synced.
2092 		 */
2093 		if (ns->disk && !test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2094 			revalidate_disk(ns->disk);
2095 
2096 		blk_set_queue_dying(ns->queue);
2097 		blk_mq_abort_requeue_list(ns->queue);
2098 		blk_mq_start_stopped_hw_queues(ns->queue, true);
2099 	}
2100 	mutex_unlock(&ctrl->namespaces_mutex);
2101 }
2102 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2103 
2104 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2105 {
2106 	struct nvme_ns *ns;
2107 
2108 	mutex_lock(&ctrl->namespaces_mutex);
2109 	list_for_each_entry(ns, &ctrl->namespaces, list)
2110 		blk_mq_quiesce_queue(ns->queue);
2111 	mutex_unlock(&ctrl->namespaces_mutex);
2112 }
2113 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2114 
2115 void nvme_start_queues(struct nvme_ctrl *ctrl)
2116 {
2117 	struct nvme_ns *ns;
2118 
2119 	mutex_lock(&ctrl->namespaces_mutex);
2120 	list_for_each_entry(ns, &ctrl->namespaces, list) {
2121 		blk_mq_start_stopped_hw_queues(ns->queue, true);
2122 		blk_mq_kick_requeue_list(ns->queue);
2123 	}
2124 	mutex_unlock(&ctrl->namespaces_mutex);
2125 }
2126 EXPORT_SYMBOL_GPL(nvme_start_queues);
2127 
2128 int __init nvme_core_init(void)
2129 {
2130 	int result;
2131 
2132 	result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2133 							&nvme_dev_fops);
2134 	if (result < 0)
2135 		return result;
2136 	else if (result > 0)
2137 		nvme_char_major = result;
2138 
2139 	nvme_class = class_create(THIS_MODULE, "nvme");
2140 	if (IS_ERR(nvme_class)) {
2141 		result = PTR_ERR(nvme_class);
2142 		goto unregister_chrdev;
2143 	}
2144 
2145 	return 0;
2146 
2147  unregister_chrdev:
2148 	__unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2149 	return result;
2150 }
2151 
2152 void nvme_core_exit(void)
2153 {
2154 	class_destroy(nvme_class);
2155 	__unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2156 }
2157 
2158 MODULE_LICENSE("GPL");
2159 MODULE_VERSION("1.0");
2160 module_init(nvme_core_init);
2161 module_exit(nvme_core_exit);
2162