xref: /openbmc/linux/drivers/nvme/host/core.c (revision 7bcae826)
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 <linux/pm_qos.h>
30 #include <scsi/sg.h>
31 #include <asm/unaligned.h>
32 
33 #include "nvme.h"
34 #include "fabrics.h"
35 
36 #define NVME_MINORS		(1U << MINORBITS)
37 
38 unsigned char admin_timeout = 60;
39 module_param(admin_timeout, byte, 0644);
40 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
41 EXPORT_SYMBOL_GPL(admin_timeout);
42 
43 unsigned char nvme_io_timeout = 30;
44 module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
45 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
46 EXPORT_SYMBOL_GPL(nvme_io_timeout);
47 
48 unsigned char shutdown_timeout = 5;
49 module_param(shutdown_timeout, byte, 0644);
50 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
51 
52 unsigned int nvme_max_retries = 5;
53 module_param_named(max_retries, nvme_max_retries, uint, 0644);
54 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
55 EXPORT_SYMBOL_GPL(nvme_max_retries);
56 
57 static int nvme_char_major;
58 module_param(nvme_char_major, int, 0);
59 
60 static unsigned long default_ps_max_latency_us = 25000;
61 module_param(default_ps_max_latency_us, ulong, 0644);
62 MODULE_PARM_DESC(default_ps_max_latency_us,
63 		 "max power saving latency for new devices; use PM QOS to change per device");
64 
65 static LIST_HEAD(nvme_ctrl_list);
66 static DEFINE_SPINLOCK(dev_list_lock);
67 
68 static struct class *nvme_class;
69 
70 void nvme_cancel_request(struct request *req, void *data, bool reserved)
71 {
72 	int status;
73 
74 	if (!blk_mq_request_started(req))
75 		return;
76 
77 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
78 				"Cancelling I/O %d", req->tag);
79 
80 	status = NVME_SC_ABORT_REQ;
81 	if (blk_queue_dying(req->q))
82 		status |= NVME_SC_DNR;
83 	blk_mq_complete_request(req, status);
84 }
85 EXPORT_SYMBOL_GPL(nvme_cancel_request);
86 
87 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
88 		enum nvme_ctrl_state new_state)
89 {
90 	enum nvme_ctrl_state old_state;
91 	bool changed = false;
92 
93 	spin_lock_irq(&ctrl->lock);
94 
95 	old_state = ctrl->state;
96 	switch (new_state) {
97 	case NVME_CTRL_LIVE:
98 		switch (old_state) {
99 		case NVME_CTRL_NEW:
100 		case NVME_CTRL_RESETTING:
101 		case NVME_CTRL_RECONNECTING:
102 			changed = true;
103 			/* FALLTHRU */
104 		default:
105 			break;
106 		}
107 		break;
108 	case NVME_CTRL_RESETTING:
109 		switch (old_state) {
110 		case NVME_CTRL_NEW:
111 		case NVME_CTRL_LIVE:
112 		case NVME_CTRL_RECONNECTING:
113 			changed = true;
114 			/* FALLTHRU */
115 		default:
116 			break;
117 		}
118 		break;
119 	case NVME_CTRL_RECONNECTING:
120 		switch (old_state) {
121 		case NVME_CTRL_LIVE:
122 			changed = true;
123 			/* FALLTHRU */
124 		default:
125 			break;
126 		}
127 		break;
128 	case NVME_CTRL_DELETING:
129 		switch (old_state) {
130 		case NVME_CTRL_LIVE:
131 		case NVME_CTRL_RESETTING:
132 		case NVME_CTRL_RECONNECTING:
133 			changed = true;
134 			/* FALLTHRU */
135 		default:
136 			break;
137 		}
138 		break;
139 	case NVME_CTRL_DEAD:
140 		switch (old_state) {
141 		case NVME_CTRL_DELETING:
142 			changed = true;
143 			/* FALLTHRU */
144 		default:
145 			break;
146 		}
147 		break;
148 	default:
149 		break;
150 	}
151 
152 	if (changed)
153 		ctrl->state = new_state;
154 
155 	spin_unlock_irq(&ctrl->lock);
156 
157 	return changed;
158 }
159 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
160 
161 static void nvme_free_ns(struct kref *kref)
162 {
163 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
164 
165 	if (ns->ndev)
166 		nvme_nvm_unregister(ns);
167 
168 	if (ns->disk) {
169 		spin_lock(&dev_list_lock);
170 		ns->disk->private_data = NULL;
171 		spin_unlock(&dev_list_lock);
172 	}
173 
174 	put_disk(ns->disk);
175 	ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
176 	nvme_put_ctrl(ns->ctrl);
177 	kfree(ns);
178 }
179 
180 static void nvme_put_ns(struct nvme_ns *ns)
181 {
182 	kref_put(&ns->kref, nvme_free_ns);
183 }
184 
185 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
186 {
187 	struct nvme_ns *ns;
188 
189 	spin_lock(&dev_list_lock);
190 	ns = disk->private_data;
191 	if (ns) {
192 		if (!kref_get_unless_zero(&ns->kref))
193 			goto fail;
194 		if (!try_module_get(ns->ctrl->ops->module))
195 			goto fail_put_ns;
196 	}
197 	spin_unlock(&dev_list_lock);
198 
199 	return ns;
200 
201 fail_put_ns:
202 	kref_put(&ns->kref, nvme_free_ns);
203 fail:
204 	spin_unlock(&dev_list_lock);
205 	return NULL;
206 }
207 
208 void nvme_requeue_req(struct request *req)
209 {
210 	blk_mq_requeue_request(req, !blk_mq_queue_stopped(req->q));
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 	unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
218 	struct request *req;
219 
220 	if (qid == NVME_QID_ANY) {
221 		req = blk_mq_alloc_request(q, op, flags);
222 	} else {
223 		req = blk_mq_alloc_request_hctx(q, op, flags,
224 				qid ? qid - 1 : 0);
225 	}
226 	if (IS_ERR(req))
227 		return req;
228 
229 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
230 	nvme_req(req)->cmd = cmd;
231 
232 	return req;
233 }
234 EXPORT_SYMBOL_GPL(nvme_alloc_request);
235 
236 static inline void nvme_setup_flush(struct nvme_ns *ns,
237 		struct nvme_command *cmnd)
238 {
239 	memset(cmnd, 0, sizeof(*cmnd));
240 	cmnd->common.opcode = nvme_cmd_flush;
241 	cmnd->common.nsid = cpu_to_le32(ns->ns_id);
242 }
243 
244 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
245 		struct nvme_command *cmnd)
246 {
247 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
248 	struct nvme_dsm_range *range;
249 	struct bio *bio;
250 
251 	range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
252 	if (!range)
253 		return BLK_MQ_RQ_QUEUE_BUSY;
254 
255 	__rq_for_each_bio(bio, req) {
256 		u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
257 		u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
258 
259 		range[n].cattr = cpu_to_le32(0);
260 		range[n].nlb = cpu_to_le32(nlb);
261 		range[n].slba = cpu_to_le64(slba);
262 		n++;
263 	}
264 
265 	if (WARN_ON_ONCE(n != segments)) {
266 		kfree(range);
267 		return BLK_MQ_RQ_QUEUE_ERROR;
268 	}
269 
270 	memset(cmnd, 0, sizeof(*cmnd));
271 	cmnd->dsm.opcode = nvme_cmd_dsm;
272 	cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
273 	cmnd->dsm.nr = segments - 1;
274 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
275 
276 	req->special_vec.bv_page = virt_to_page(range);
277 	req->special_vec.bv_offset = offset_in_page(range);
278 	req->special_vec.bv_len = sizeof(*range) * segments;
279 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
280 
281 	return BLK_MQ_RQ_QUEUE_OK;
282 }
283 
284 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
285 		struct nvme_command *cmnd)
286 {
287 	u16 control = 0;
288 	u32 dsmgmt = 0;
289 
290 	if (req->cmd_flags & REQ_FUA)
291 		control |= NVME_RW_FUA;
292 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
293 		control |= NVME_RW_LR;
294 
295 	if (req->cmd_flags & REQ_RAHEAD)
296 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
297 
298 	memset(cmnd, 0, sizeof(*cmnd));
299 	cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
300 	cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
301 	cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
302 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
303 
304 	if (ns->ms) {
305 		switch (ns->pi_type) {
306 		case NVME_NS_DPS_PI_TYPE3:
307 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
308 			break;
309 		case NVME_NS_DPS_PI_TYPE1:
310 		case NVME_NS_DPS_PI_TYPE2:
311 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
312 					NVME_RW_PRINFO_PRCHK_REF;
313 			cmnd->rw.reftag = cpu_to_le32(
314 					nvme_block_nr(ns, blk_rq_pos(req)));
315 			break;
316 		}
317 		if (!blk_integrity_rq(req))
318 			control |= NVME_RW_PRINFO_PRACT;
319 	}
320 
321 	cmnd->rw.control = cpu_to_le16(control);
322 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
323 }
324 
325 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
326 		struct nvme_command *cmd)
327 {
328 	int ret = BLK_MQ_RQ_QUEUE_OK;
329 
330 	switch (req_op(req)) {
331 	case REQ_OP_DRV_IN:
332 	case REQ_OP_DRV_OUT:
333 		memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
334 		break;
335 	case REQ_OP_FLUSH:
336 		nvme_setup_flush(ns, cmd);
337 		break;
338 	case REQ_OP_DISCARD:
339 		ret = nvme_setup_discard(ns, req, cmd);
340 		break;
341 	case REQ_OP_READ:
342 	case REQ_OP_WRITE:
343 		nvme_setup_rw(ns, req, cmd);
344 		break;
345 	default:
346 		WARN_ON_ONCE(1);
347 		return BLK_MQ_RQ_QUEUE_ERROR;
348 	}
349 
350 	cmd->common.command_id = req->tag;
351 	return ret;
352 }
353 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
354 
355 /*
356  * Returns 0 on success.  If the result is negative, it's a Linux error code;
357  * if the result is positive, it's an NVM Express status code
358  */
359 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
360 		union nvme_result *result, void *buffer, unsigned bufflen,
361 		unsigned timeout, int qid, int at_head, int flags)
362 {
363 	struct request *req;
364 	int ret;
365 
366 	req = nvme_alloc_request(q, cmd, flags, qid);
367 	if (IS_ERR(req))
368 		return PTR_ERR(req);
369 
370 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
371 
372 	if (buffer && bufflen) {
373 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
374 		if (ret)
375 			goto out;
376 	}
377 
378 	blk_execute_rq(req->q, NULL, req, at_head);
379 	if (result)
380 		*result = nvme_req(req)->result;
381 	ret = req->errors;
382  out:
383 	blk_mq_free_request(req);
384 	return ret;
385 }
386 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
387 
388 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
389 		void *buffer, unsigned bufflen)
390 {
391 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
392 			NVME_QID_ANY, 0, 0);
393 }
394 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
395 
396 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
397 		void __user *ubuffer, unsigned bufflen,
398 		void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
399 		u32 *result, unsigned timeout)
400 {
401 	bool write = nvme_is_write(cmd);
402 	struct nvme_ns *ns = q->queuedata;
403 	struct gendisk *disk = ns ? ns->disk : NULL;
404 	struct request *req;
405 	struct bio *bio = NULL;
406 	void *meta = NULL;
407 	int ret;
408 
409 	req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
410 	if (IS_ERR(req))
411 		return PTR_ERR(req);
412 
413 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
414 
415 	if (ubuffer && bufflen) {
416 		ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
417 				GFP_KERNEL);
418 		if (ret)
419 			goto out;
420 		bio = req->bio;
421 
422 		if (!disk)
423 			goto submit;
424 		bio->bi_bdev = bdget_disk(disk, 0);
425 		if (!bio->bi_bdev) {
426 			ret = -ENODEV;
427 			goto out_unmap;
428 		}
429 
430 		if (meta_buffer && meta_len) {
431 			struct bio_integrity_payload *bip;
432 
433 			meta = kmalloc(meta_len, GFP_KERNEL);
434 			if (!meta) {
435 				ret = -ENOMEM;
436 				goto out_unmap;
437 			}
438 
439 			if (write) {
440 				if (copy_from_user(meta, meta_buffer,
441 						meta_len)) {
442 					ret = -EFAULT;
443 					goto out_free_meta;
444 				}
445 			}
446 
447 			bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
448 			if (IS_ERR(bip)) {
449 				ret = PTR_ERR(bip);
450 				goto out_free_meta;
451 			}
452 
453 			bip->bip_iter.bi_size = meta_len;
454 			bip->bip_iter.bi_sector = meta_seed;
455 
456 			ret = bio_integrity_add_page(bio, virt_to_page(meta),
457 					meta_len, offset_in_page(meta));
458 			if (ret != meta_len) {
459 				ret = -ENOMEM;
460 				goto out_free_meta;
461 			}
462 		}
463 	}
464  submit:
465 	blk_execute_rq(req->q, disk, req, 0);
466 	ret = req->errors;
467 	if (result)
468 		*result = le32_to_cpu(nvme_req(req)->result.u32);
469 	if (meta && !ret && !write) {
470 		if (copy_to_user(meta_buffer, meta, meta_len))
471 			ret = -EFAULT;
472 	}
473  out_free_meta:
474 	kfree(meta);
475  out_unmap:
476 	if (bio) {
477 		if (disk && bio->bi_bdev)
478 			bdput(bio->bi_bdev);
479 		blk_rq_unmap_user(bio);
480 	}
481  out:
482 	blk_mq_free_request(req);
483 	return ret;
484 }
485 
486 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
487 		void __user *ubuffer, unsigned bufflen, u32 *result,
488 		unsigned timeout)
489 {
490 	return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
491 			result, timeout);
492 }
493 
494 static void nvme_keep_alive_end_io(struct request *rq, int error)
495 {
496 	struct nvme_ctrl *ctrl = rq->end_io_data;
497 
498 	blk_mq_free_request(rq);
499 
500 	if (error) {
501 		dev_err(ctrl->device,
502 			"failed nvme_keep_alive_end_io error=%d\n", error);
503 		return;
504 	}
505 
506 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
507 }
508 
509 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
510 {
511 	struct nvme_command c;
512 	struct request *rq;
513 
514 	memset(&c, 0, sizeof(c));
515 	c.common.opcode = nvme_admin_keep_alive;
516 
517 	rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
518 			NVME_QID_ANY);
519 	if (IS_ERR(rq))
520 		return PTR_ERR(rq);
521 
522 	rq->timeout = ctrl->kato * HZ;
523 	rq->end_io_data = ctrl;
524 
525 	blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
526 
527 	return 0;
528 }
529 
530 static void nvme_keep_alive_work(struct work_struct *work)
531 {
532 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
533 			struct nvme_ctrl, ka_work);
534 
535 	if (nvme_keep_alive(ctrl)) {
536 		/* allocation failure, reset the controller */
537 		dev_err(ctrl->device, "keep-alive failed\n");
538 		ctrl->ops->reset_ctrl(ctrl);
539 		return;
540 	}
541 }
542 
543 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
544 {
545 	if (unlikely(ctrl->kato == 0))
546 		return;
547 
548 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
549 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
550 }
551 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
552 
553 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
554 {
555 	if (unlikely(ctrl->kato == 0))
556 		return;
557 
558 	cancel_delayed_work_sync(&ctrl->ka_work);
559 }
560 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
561 
562 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
563 {
564 	struct nvme_command c = { };
565 	int error;
566 
567 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
568 	c.identify.opcode = nvme_admin_identify;
569 	c.identify.cns = NVME_ID_CNS_CTRL;
570 
571 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
572 	if (!*id)
573 		return -ENOMEM;
574 
575 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
576 			sizeof(struct nvme_id_ctrl));
577 	if (error)
578 		kfree(*id);
579 	return error;
580 }
581 
582 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
583 {
584 	struct nvme_command c = { };
585 
586 	c.identify.opcode = nvme_admin_identify;
587 	c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
588 	c.identify.nsid = cpu_to_le32(nsid);
589 	return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
590 }
591 
592 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
593 		struct nvme_id_ns **id)
594 {
595 	struct nvme_command c = { };
596 	int error;
597 
598 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
599 	c.identify.opcode = nvme_admin_identify;
600 	c.identify.nsid = cpu_to_le32(nsid);
601 	c.identify.cns = NVME_ID_CNS_NS;
602 
603 	*id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
604 	if (!*id)
605 		return -ENOMEM;
606 
607 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
608 			sizeof(struct nvme_id_ns));
609 	if (error)
610 		kfree(*id);
611 	return error;
612 }
613 
614 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
615 		      void *buffer, size_t buflen, u32 *result)
616 {
617 	struct nvme_command c;
618 	union nvme_result res;
619 	int ret;
620 
621 	memset(&c, 0, sizeof(c));
622 	c.features.opcode = nvme_admin_get_features;
623 	c.features.nsid = cpu_to_le32(nsid);
624 	c.features.fid = cpu_to_le32(fid);
625 
626 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, buffer, buflen, 0,
627 			NVME_QID_ANY, 0, 0);
628 	if (ret >= 0 && result)
629 		*result = le32_to_cpu(res.u32);
630 	return ret;
631 }
632 
633 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
634 		      void *buffer, size_t buflen, u32 *result)
635 {
636 	struct nvme_command c;
637 	union nvme_result res;
638 	int ret;
639 
640 	memset(&c, 0, sizeof(c));
641 	c.features.opcode = nvme_admin_set_features;
642 	c.features.fid = cpu_to_le32(fid);
643 	c.features.dword11 = cpu_to_le32(dword11);
644 
645 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
646 			buffer, buflen, 0, NVME_QID_ANY, 0, 0);
647 	if (ret >= 0 && result)
648 		*result = le32_to_cpu(res.u32);
649 	return ret;
650 }
651 
652 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
653 {
654 	struct nvme_command c = { };
655 	int error;
656 
657 	c.common.opcode = nvme_admin_get_log_page,
658 	c.common.nsid = cpu_to_le32(0xFFFFFFFF),
659 	c.common.cdw10[0] = cpu_to_le32(
660 			(((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
661 			 NVME_LOG_SMART),
662 
663 	*log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
664 	if (!*log)
665 		return -ENOMEM;
666 
667 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
668 			sizeof(struct nvme_smart_log));
669 	if (error)
670 		kfree(*log);
671 	return error;
672 }
673 
674 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
675 {
676 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
677 	u32 result;
678 	int status, nr_io_queues;
679 
680 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
681 			&result);
682 	if (status < 0)
683 		return status;
684 
685 	/*
686 	 * Degraded controllers might return an error when setting the queue
687 	 * count.  We still want to be able to bring them online and offer
688 	 * access to the admin queue, as that might be only way to fix them up.
689 	 */
690 	if (status > 0) {
691 		dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
692 		*count = 0;
693 	} else {
694 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
695 		*count = min(*count, nr_io_queues);
696 	}
697 
698 	return 0;
699 }
700 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
701 
702 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
703 {
704 	struct nvme_user_io io;
705 	struct nvme_command c;
706 	unsigned length, meta_len;
707 	void __user *metadata;
708 
709 	if (copy_from_user(&io, uio, sizeof(io)))
710 		return -EFAULT;
711 	if (io.flags)
712 		return -EINVAL;
713 
714 	switch (io.opcode) {
715 	case nvme_cmd_write:
716 	case nvme_cmd_read:
717 	case nvme_cmd_compare:
718 		break;
719 	default:
720 		return -EINVAL;
721 	}
722 
723 	length = (io.nblocks + 1) << ns->lba_shift;
724 	meta_len = (io.nblocks + 1) * ns->ms;
725 	metadata = (void __user *)(uintptr_t)io.metadata;
726 
727 	if (ns->ext) {
728 		length += meta_len;
729 		meta_len = 0;
730 	} else if (meta_len) {
731 		if ((io.metadata & 3) || !io.metadata)
732 			return -EINVAL;
733 	}
734 
735 	memset(&c, 0, sizeof(c));
736 	c.rw.opcode = io.opcode;
737 	c.rw.flags = io.flags;
738 	c.rw.nsid = cpu_to_le32(ns->ns_id);
739 	c.rw.slba = cpu_to_le64(io.slba);
740 	c.rw.length = cpu_to_le16(io.nblocks);
741 	c.rw.control = cpu_to_le16(io.control);
742 	c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
743 	c.rw.reftag = cpu_to_le32(io.reftag);
744 	c.rw.apptag = cpu_to_le16(io.apptag);
745 	c.rw.appmask = cpu_to_le16(io.appmask);
746 
747 	return __nvme_submit_user_cmd(ns->queue, &c,
748 			(void __user *)(uintptr_t)io.addr, length,
749 			metadata, meta_len, io.slba, NULL, 0);
750 }
751 
752 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
753 			struct nvme_passthru_cmd __user *ucmd)
754 {
755 	struct nvme_passthru_cmd cmd;
756 	struct nvme_command c;
757 	unsigned timeout = 0;
758 	int status;
759 
760 	if (!capable(CAP_SYS_ADMIN))
761 		return -EACCES;
762 	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
763 		return -EFAULT;
764 	if (cmd.flags)
765 		return -EINVAL;
766 
767 	memset(&c, 0, sizeof(c));
768 	c.common.opcode = cmd.opcode;
769 	c.common.flags = cmd.flags;
770 	c.common.nsid = cpu_to_le32(cmd.nsid);
771 	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
772 	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
773 	c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
774 	c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
775 	c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
776 	c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
777 	c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
778 	c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
779 
780 	if (cmd.timeout_ms)
781 		timeout = msecs_to_jiffies(cmd.timeout_ms);
782 
783 	status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
784 			(void __user *)(uintptr_t)cmd.addr, cmd.data_len,
785 			&cmd.result, timeout);
786 	if (status >= 0) {
787 		if (put_user(cmd.result, &ucmd->result))
788 			return -EFAULT;
789 	}
790 
791 	return status;
792 }
793 
794 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
795 		unsigned int cmd, unsigned long arg)
796 {
797 	struct nvme_ns *ns = bdev->bd_disk->private_data;
798 
799 	switch (cmd) {
800 	case NVME_IOCTL_ID:
801 		force_successful_syscall_return();
802 		return ns->ns_id;
803 	case NVME_IOCTL_ADMIN_CMD:
804 		return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
805 	case NVME_IOCTL_IO_CMD:
806 		return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
807 	case NVME_IOCTL_SUBMIT_IO:
808 		return nvme_submit_io(ns, (void __user *)arg);
809 #ifdef CONFIG_BLK_DEV_NVME_SCSI
810 	case SG_GET_VERSION_NUM:
811 		return nvme_sg_get_version_num((void __user *)arg);
812 	case SG_IO:
813 		return nvme_sg_io(ns, (void __user *)arg);
814 #endif
815 	default:
816 #ifdef CONFIG_NVM
817 		if (ns->ndev)
818 			return nvme_nvm_ioctl(ns, cmd, arg);
819 #endif
820 		if (is_sed_ioctl(cmd))
821 			return sed_ioctl(ns->ctrl->opal_dev, cmd,
822 					 (void __user *) arg);
823 		return -ENOTTY;
824 	}
825 }
826 
827 #ifdef CONFIG_COMPAT
828 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
829 			unsigned int cmd, unsigned long arg)
830 {
831 	switch (cmd) {
832 	case SG_IO:
833 		return -ENOIOCTLCMD;
834 	}
835 	return nvme_ioctl(bdev, mode, cmd, arg);
836 }
837 #else
838 #define nvme_compat_ioctl	NULL
839 #endif
840 
841 static int nvme_open(struct block_device *bdev, fmode_t mode)
842 {
843 	return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
844 }
845 
846 static void nvme_release(struct gendisk *disk, fmode_t mode)
847 {
848 	struct nvme_ns *ns = disk->private_data;
849 
850 	module_put(ns->ctrl->ops->module);
851 	nvme_put_ns(ns);
852 }
853 
854 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
855 {
856 	/* some standard values */
857 	geo->heads = 1 << 6;
858 	geo->sectors = 1 << 5;
859 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
860 	return 0;
861 }
862 
863 #ifdef CONFIG_BLK_DEV_INTEGRITY
864 static void nvme_init_integrity(struct nvme_ns *ns)
865 {
866 	struct blk_integrity integrity;
867 
868 	memset(&integrity, 0, sizeof(integrity));
869 	switch (ns->pi_type) {
870 	case NVME_NS_DPS_PI_TYPE3:
871 		integrity.profile = &t10_pi_type3_crc;
872 		integrity.tag_size = sizeof(u16) + sizeof(u32);
873 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
874 		break;
875 	case NVME_NS_DPS_PI_TYPE1:
876 	case NVME_NS_DPS_PI_TYPE2:
877 		integrity.profile = &t10_pi_type1_crc;
878 		integrity.tag_size = sizeof(u16);
879 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
880 		break;
881 	default:
882 		integrity.profile = NULL;
883 		break;
884 	}
885 	integrity.tuple_size = ns->ms;
886 	blk_integrity_register(ns->disk, &integrity);
887 	blk_queue_max_integrity_segments(ns->queue, 1);
888 }
889 #else
890 static void nvme_init_integrity(struct nvme_ns *ns)
891 {
892 }
893 #endif /* CONFIG_BLK_DEV_INTEGRITY */
894 
895 static void nvme_config_discard(struct nvme_ns *ns)
896 {
897 	struct nvme_ctrl *ctrl = ns->ctrl;
898 	u32 logical_block_size = queue_logical_block_size(ns->queue);
899 
900 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
901 			NVME_DSM_MAX_RANGES);
902 
903 	if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
904 		ns->queue->limits.discard_zeroes_data = 1;
905 	else
906 		ns->queue->limits.discard_zeroes_data = 0;
907 
908 	ns->queue->limits.discard_alignment = logical_block_size;
909 	ns->queue->limits.discard_granularity = logical_block_size;
910 	blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
911 	blk_queue_max_discard_segments(ns->queue, NVME_DSM_MAX_RANGES);
912 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
913 }
914 
915 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
916 {
917 	if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
918 		dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
919 		return -ENODEV;
920 	}
921 
922 	if ((*id)->ncap == 0) {
923 		kfree(*id);
924 		return -ENODEV;
925 	}
926 
927 	if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
928 		memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
929 	if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
930 		memcpy(ns->uuid, (*id)->nguid, sizeof(ns->uuid));
931 
932 	return 0;
933 }
934 
935 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
936 {
937 	struct nvme_ns *ns = disk->private_data;
938 	u8 lbaf, pi_type;
939 	u16 old_ms;
940 	unsigned short bs;
941 
942 	old_ms = ns->ms;
943 	lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
944 	ns->lba_shift = id->lbaf[lbaf].ds;
945 	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
946 	ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
947 
948 	/*
949 	 * If identify namespace failed, use default 512 byte block size so
950 	 * block layer can use before failing read/write for 0 capacity.
951 	 */
952 	if (ns->lba_shift == 0)
953 		ns->lba_shift = 9;
954 	bs = 1 << ns->lba_shift;
955 	/* XXX: PI implementation requires metadata equal t10 pi tuple size */
956 	pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
957 					id->dps & NVME_NS_DPS_PI_MASK : 0;
958 
959 	blk_mq_freeze_queue(disk->queue);
960 	if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
961 				ns->ms != old_ms ||
962 				bs != queue_logical_block_size(disk->queue) ||
963 				(ns->ms && ns->ext)))
964 		blk_integrity_unregister(disk);
965 
966 	ns->pi_type = pi_type;
967 	blk_queue_logical_block_size(ns->queue, bs);
968 
969 	if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
970 		nvme_init_integrity(ns);
971 	if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
972 		set_capacity(disk, 0);
973 	else
974 		set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
975 
976 	if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
977 		nvme_config_discard(ns);
978 	blk_mq_unfreeze_queue(disk->queue);
979 }
980 
981 static int nvme_revalidate_disk(struct gendisk *disk)
982 {
983 	struct nvme_ns *ns = disk->private_data;
984 	struct nvme_id_ns *id = NULL;
985 	int ret;
986 
987 	if (test_bit(NVME_NS_DEAD, &ns->flags)) {
988 		set_capacity(disk, 0);
989 		return -ENODEV;
990 	}
991 
992 	ret = nvme_revalidate_ns(ns, &id);
993 	if (ret)
994 		return ret;
995 
996 	__nvme_revalidate_disk(disk, id);
997 	kfree(id);
998 
999 	return 0;
1000 }
1001 
1002 static char nvme_pr_type(enum pr_type type)
1003 {
1004 	switch (type) {
1005 	case PR_WRITE_EXCLUSIVE:
1006 		return 1;
1007 	case PR_EXCLUSIVE_ACCESS:
1008 		return 2;
1009 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1010 		return 3;
1011 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1012 		return 4;
1013 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1014 		return 5;
1015 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1016 		return 6;
1017 	default:
1018 		return 0;
1019 	}
1020 };
1021 
1022 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1023 				u64 key, u64 sa_key, u8 op)
1024 {
1025 	struct nvme_ns *ns = bdev->bd_disk->private_data;
1026 	struct nvme_command c;
1027 	u8 data[16] = { 0, };
1028 
1029 	put_unaligned_le64(key, &data[0]);
1030 	put_unaligned_le64(sa_key, &data[8]);
1031 
1032 	memset(&c, 0, sizeof(c));
1033 	c.common.opcode = op;
1034 	c.common.nsid = cpu_to_le32(ns->ns_id);
1035 	c.common.cdw10[0] = cpu_to_le32(cdw10);
1036 
1037 	return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1038 }
1039 
1040 static int nvme_pr_register(struct block_device *bdev, u64 old,
1041 		u64 new, unsigned flags)
1042 {
1043 	u32 cdw10;
1044 
1045 	if (flags & ~PR_FL_IGNORE_KEY)
1046 		return -EOPNOTSUPP;
1047 
1048 	cdw10 = old ? 2 : 0;
1049 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1050 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1051 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1052 }
1053 
1054 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1055 		enum pr_type type, unsigned flags)
1056 {
1057 	u32 cdw10;
1058 
1059 	if (flags & ~PR_FL_IGNORE_KEY)
1060 		return -EOPNOTSUPP;
1061 
1062 	cdw10 = nvme_pr_type(type) << 8;
1063 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1064 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1065 }
1066 
1067 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1068 		enum pr_type type, bool abort)
1069 {
1070 	u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1071 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1072 }
1073 
1074 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1075 {
1076 	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1077 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1078 }
1079 
1080 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1081 {
1082 	u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1083 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1084 }
1085 
1086 static const struct pr_ops nvme_pr_ops = {
1087 	.pr_register	= nvme_pr_register,
1088 	.pr_reserve	= nvme_pr_reserve,
1089 	.pr_release	= nvme_pr_release,
1090 	.pr_preempt	= nvme_pr_preempt,
1091 	.pr_clear	= nvme_pr_clear,
1092 };
1093 
1094 #ifdef CONFIG_BLK_SED_OPAL
1095 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1096 		bool send)
1097 {
1098 	struct nvme_ctrl *ctrl = data;
1099 	struct nvme_command cmd;
1100 
1101 	memset(&cmd, 0, sizeof(cmd));
1102 	if (send)
1103 		cmd.common.opcode = nvme_admin_security_send;
1104 	else
1105 		cmd.common.opcode = nvme_admin_security_recv;
1106 	cmd.common.nsid = 0;
1107 	cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1108 	cmd.common.cdw10[1] = cpu_to_le32(len);
1109 
1110 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1111 				      ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1112 }
1113 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1114 #endif /* CONFIG_BLK_SED_OPAL */
1115 
1116 static const struct block_device_operations nvme_fops = {
1117 	.owner		= THIS_MODULE,
1118 	.ioctl		= nvme_ioctl,
1119 	.compat_ioctl	= nvme_compat_ioctl,
1120 	.open		= nvme_open,
1121 	.release	= nvme_release,
1122 	.getgeo		= nvme_getgeo,
1123 	.revalidate_disk= nvme_revalidate_disk,
1124 	.pr_ops		= &nvme_pr_ops,
1125 };
1126 
1127 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1128 {
1129 	unsigned long timeout =
1130 		((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1131 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1132 	int ret;
1133 
1134 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1135 		if (csts == ~0)
1136 			return -ENODEV;
1137 		if ((csts & NVME_CSTS_RDY) == bit)
1138 			break;
1139 
1140 		msleep(100);
1141 		if (fatal_signal_pending(current))
1142 			return -EINTR;
1143 		if (time_after(jiffies, timeout)) {
1144 			dev_err(ctrl->device,
1145 				"Device not ready; aborting %s\n", enabled ?
1146 						"initialisation" : "reset");
1147 			return -ENODEV;
1148 		}
1149 	}
1150 
1151 	return ret;
1152 }
1153 
1154 /*
1155  * If the device has been passed off to us in an enabled state, just clear
1156  * the enabled bit.  The spec says we should set the 'shutdown notification
1157  * bits', but doing so may cause the device to complete commands to the
1158  * admin queue ... and we don't know what memory that might be pointing at!
1159  */
1160 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1161 {
1162 	int ret;
1163 
1164 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1165 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1166 
1167 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1168 	if (ret)
1169 		return ret;
1170 
1171 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1172 		msleep(NVME_QUIRK_DELAY_AMOUNT);
1173 
1174 	return nvme_wait_ready(ctrl, cap, false);
1175 }
1176 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1177 
1178 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1179 {
1180 	/*
1181 	 * Default to a 4K page size, with the intention to update this
1182 	 * path in the future to accomodate architectures with differing
1183 	 * kernel and IO page sizes.
1184 	 */
1185 	unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1186 	int ret;
1187 
1188 	if (page_shift < dev_page_min) {
1189 		dev_err(ctrl->device,
1190 			"Minimum device page size %u too large for host (%u)\n",
1191 			1 << dev_page_min, 1 << page_shift);
1192 		return -ENODEV;
1193 	}
1194 
1195 	ctrl->page_size = 1 << page_shift;
1196 
1197 	ctrl->ctrl_config = NVME_CC_CSS_NVM;
1198 	ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1199 	ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1200 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1201 	ctrl->ctrl_config |= NVME_CC_ENABLE;
1202 
1203 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1204 	if (ret)
1205 		return ret;
1206 	return nvme_wait_ready(ctrl, cap, true);
1207 }
1208 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1209 
1210 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1211 {
1212 	unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1213 	u32 csts;
1214 	int ret;
1215 
1216 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1217 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1218 
1219 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1220 	if (ret)
1221 		return ret;
1222 
1223 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1224 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1225 			break;
1226 
1227 		msleep(100);
1228 		if (fatal_signal_pending(current))
1229 			return -EINTR;
1230 		if (time_after(jiffies, timeout)) {
1231 			dev_err(ctrl->device,
1232 				"Device shutdown incomplete; abort shutdown\n");
1233 			return -ENODEV;
1234 		}
1235 	}
1236 
1237 	return ret;
1238 }
1239 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1240 
1241 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1242 		struct request_queue *q)
1243 {
1244 	bool vwc = false;
1245 
1246 	if (ctrl->max_hw_sectors) {
1247 		u32 max_segments =
1248 			(ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1249 
1250 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1251 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1252 	}
1253 	if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1254 		blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1255 	blk_queue_virt_boundary(q, ctrl->page_size - 1);
1256 	if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1257 		vwc = true;
1258 	blk_queue_write_cache(q, vwc, vwc);
1259 }
1260 
1261 static void nvme_configure_apst(struct nvme_ctrl *ctrl)
1262 {
1263 	/*
1264 	 * APST (Autonomous Power State Transition) lets us program a
1265 	 * table of power state transitions that the controller will
1266 	 * perform automatically.  We configure it with a simple
1267 	 * heuristic: we are willing to spend at most 2% of the time
1268 	 * transitioning between power states.  Therefore, when running
1269 	 * in any given state, we will enter the next lower-power
1270 	 * non-operational state after waiting 100 * (enlat + exlat)
1271 	 * microseconds, as long as that state's total latency is under
1272 	 * the requested maximum latency.
1273 	 *
1274 	 * We will not autonomously enter any non-operational state for
1275 	 * which the total latency exceeds ps_max_latency_us.  Users
1276 	 * can set ps_max_latency_us to zero to turn off APST.
1277 	 */
1278 
1279 	unsigned apste;
1280 	struct nvme_feat_auto_pst *table;
1281 	int ret;
1282 
1283 	/*
1284 	 * If APST isn't supported or if we haven't been initialized yet,
1285 	 * then don't do anything.
1286 	 */
1287 	if (!ctrl->apsta)
1288 		return;
1289 
1290 	if (ctrl->npss > 31) {
1291 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1292 		return;
1293 	}
1294 
1295 	table = kzalloc(sizeof(*table), GFP_KERNEL);
1296 	if (!table)
1297 		return;
1298 
1299 	if (ctrl->ps_max_latency_us == 0) {
1300 		/* Turn off APST. */
1301 		apste = 0;
1302 	} else {
1303 		__le64 target = cpu_to_le64(0);
1304 		int state;
1305 
1306 		/*
1307 		 * Walk through all states from lowest- to highest-power.
1308 		 * According to the spec, lower-numbered states use more
1309 		 * power.  NPSS, despite the name, is the index of the
1310 		 * lowest-power state, not the number of states.
1311 		 */
1312 		for (state = (int)ctrl->npss; state >= 0; state--) {
1313 			u64 total_latency_us, transition_ms;
1314 
1315 			if (target)
1316 				table->entries[state] = target;
1317 
1318 			/*
1319 			 * Is this state a useful non-operational state for
1320 			 * higher-power states to autonomously transition to?
1321 			 */
1322 			if (!(ctrl->psd[state].flags &
1323 			      NVME_PS_FLAGS_NON_OP_STATE))
1324 				continue;
1325 
1326 			total_latency_us =
1327 				(u64)le32_to_cpu(ctrl->psd[state].entry_lat) +
1328 				+ le32_to_cpu(ctrl->psd[state].exit_lat);
1329 			if (total_latency_us > ctrl->ps_max_latency_us)
1330 				continue;
1331 
1332 			/*
1333 			 * This state is good.  Use it as the APST idle
1334 			 * target for higher power states.
1335 			 */
1336 			transition_ms = total_latency_us + 19;
1337 			do_div(transition_ms, 20);
1338 			if (transition_ms > (1 << 24) - 1)
1339 				transition_ms = (1 << 24) - 1;
1340 
1341 			target = cpu_to_le64((state << 3) |
1342 					     (transition_ms << 8));
1343 		}
1344 
1345 		apste = 1;
1346 	}
1347 
1348 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1349 				table, sizeof(*table), NULL);
1350 	if (ret)
1351 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1352 
1353 	kfree(table);
1354 }
1355 
1356 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1357 {
1358 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1359 	u64 latency;
1360 
1361 	switch (val) {
1362 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1363 	case PM_QOS_LATENCY_ANY:
1364 		latency = U64_MAX;
1365 		break;
1366 
1367 	default:
1368 		latency = val;
1369 	}
1370 
1371 	if (ctrl->ps_max_latency_us != latency) {
1372 		ctrl->ps_max_latency_us = latency;
1373 		nvme_configure_apst(ctrl);
1374 	}
1375 }
1376 
1377 struct nvme_core_quirk_entry {
1378 	/*
1379 	 * NVMe model and firmware strings are padded with spaces.  For
1380 	 * simplicity, strings in the quirk table are padded with NULLs
1381 	 * instead.
1382 	 */
1383 	u16 vid;
1384 	const char *mn;
1385 	const char *fr;
1386 	unsigned long quirks;
1387 };
1388 
1389 static const struct nvme_core_quirk_entry core_quirks[] = {
1390 	/*
1391 	 * Seen on a Samsung "SM951 NVMe SAMSUNG 256GB": using APST causes
1392 	 * the controller to go out to lunch.  It dies when the watchdog
1393 	 * timer reads CSTS and gets 0xffffffff.
1394 	 */
1395 	{
1396 		.vid = 0x144d,
1397 		.fr = "BXW75D0Q",
1398 		.quirks = NVME_QUIRK_NO_APST,
1399 	},
1400 };
1401 
1402 /* match is null-terminated but idstr is space-padded. */
1403 static bool string_matches(const char *idstr, const char *match, size_t len)
1404 {
1405 	size_t matchlen;
1406 
1407 	if (!match)
1408 		return true;
1409 
1410 	matchlen = strlen(match);
1411 	WARN_ON_ONCE(matchlen > len);
1412 
1413 	if (memcmp(idstr, match, matchlen))
1414 		return false;
1415 
1416 	for (; matchlen < len; matchlen++)
1417 		if (idstr[matchlen] != ' ')
1418 			return false;
1419 
1420 	return true;
1421 }
1422 
1423 static bool quirk_matches(const struct nvme_id_ctrl *id,
1424 			  const struct nvme_core_quirk_entry *q)
1425 {
1426 	return q->vid == le16_to_cpu(id->vid) &&
1427 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1428 		string_matches(id->fr, q->fr, sizeof(id->fr));
1429 }
1430 
1431 /*
1432  * Initialize the cached copies of the Identify data and various controller
1433  * register in our nvme_ctrl structure.  This should be called as soon as
1434  * the admin queue is fully up and running.
1435  */
1436 int nvme_init_identify(struct nvme_ctrl *ctrl)
1437 {
1438 	struct nvme_id_ctrl *id;
1439 	u64 cap;
1440 	int ret, page_shift;
1441 	u32 max_hw_sectors;
1442 	u8 prev_apsta;
1443 
1444 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1445 	if (ret) {
1446 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1447 		return ret;
1448 	}
1449 
1450 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1451 	if (ret) {
1452 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1453 		return ret;
1454 	}
1455 	page_shift = NVME_CAP_MPSMIN(cap) + 12;
1456 
1457 	if (ctrl->vs >= NVME_VS(1, 1, 0))
1458 		ctrl->subsystem = NVME_CAP_NSSRC(cap);
1459 
1460 	ret = nvme_identify_ctrl(ctrl, &id);
1461 	if (ret) {
1462 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1463 		return -EIO;
1464 	}
1465 
1466 	if (!ctrl->identified) {
1467 		/*
1468 		 * Check for quirks.  Quirk can depend on firmware version,
1469 		 * so, in principle, the set of quirks present can change
1470 		 * across a reset.  As a possible future enhancement, we
1471 		 * could re-scan for quirks every time we reinitialize
1472 		 * the device, but we'd have to make sure that the driver
1473 		 * behaves intelligently if the quirks change.
1474 		 */
1475 
1476 		int i;
1477 
1478 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
1479 			if (quirk_matches(id, &core_quirks[i]))
1480 				ctrl->quirks |= core_quirks[i].quirks;
1481 		}
1482 	}
1483 
1484 	ctrl->oacs = le16_to_cpu(id->oacs);
1485 	ctrl->vid = le16_to_cpu(id->vid);
1486 	ctrl->oncs = le16_to_cpup(&id->oncs);
1487 	atomic_set(&ctrl->abort_limit, id->acl + 1);
1488 	ctrl->vwc = id->vwc;
1489 	ctrl->cntlid = le16_to_cpup(&id->cntlid);
1490 	memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1491 	memcpy(ctrl->model, id->mn, sizeof(id->mn));
1492 	memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1493 	if (id->mdts)
1494 		max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1495 	else
1496 		max_hw_sectors = UINT_MAX;
1497 	ctrl->max_hw_sectors =
1498 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1499 
1500 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
1501 	ctrl->sgls = le32_to_cpu(id->sgls);
1502 	ctrl->kas = le16_to_cpu(id->kas);
1503 
1504 	ctrl->npss = id->npss;
1505 	prev_apsta = ctrl->apsta;
1506 	ctrl->apsta = (ctrl->quirks & NVME_QUIRK_NO_APST) ? 0 : id->apsta;
1507 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
1508 
1509 	if (ctrl->ops->is_fabrics) {
1510 		ctrl->icdoff = le16_to_cpu(id->icdoff);
1511 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1512 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1513 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1514 
1515 		/*
1516 		 * In fabrics we need to verify the cntlid matches the
1517 		 * admin connect
1518 		 */
1519 		if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1520 			ret = -EINVAL;
1521 
1522 		if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1523 			dev_err(ctrl->dev,
1524 				"keep-alive support is mandatory for fabrics\n");
1525 			ret = -EINVAL;
1526 		}
1527 	} else {
1528 		ctrl->cntlid = le16_to_cpu(id->cntlid);
1529 	}
1530 
1531 	kfree(id);
1532 
1533 	if (ctrl->apsta && !prev_apsta)
1534 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
1535 	else if (!ctrl->apsta && prev_apsta)
1536 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
1537 
1538 	nvme_configure_apst(ctrl);
1539 
1540 	ctrl->identified = true;
1541 
1542 	return ret;
1543 }
1544 EXPORT_SYMBOL_GPL(nvme_init_identify);
1545 
1546 static int nvme_dev_open(struct inode *inode, struct file *file)
1547 {
1548 	struct nvme_ctrl *ctrl;
1549 	int instance = iminor(inode);
1550 	int ret = -ENODEV;
1551 
1552 	spin_lock(&dev_list_lock);
1553 	list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1554 		if (ctrl->instance != instance)
1555 			continue;
1556 
1557 		if (!ctrl->admin_q) {
1558 			ret = -EWOULDBLOCK;
1559 			break;
1560 		}
1561 		if (!kref_get_unless_zero(&ctrl->kref))
1562 			break;
1563 		file->private_data = ctrl;
1564 		ret = 0;
1565 		break;
1566 	}
1567 	spin_unlock(&dev_list_lock);
1568 
1569 	return ret;
1570 }
1571 
1572 static int nvme_dev_release(struct inode *inode, struct file *file)
1573 {
1574 	nvme_put_ctrl(file->private_data);
1575 	return 0;
1576 }
1577 
1578 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1579 {
1580 	struct nvme_ns *ns;
1581 	int ret;
1582 
1583 	mutex_lock(&ctrl->namespaces_mutex);
1584 	if (list_empty(&ctrl->namespaces)) {
1585 		ret = -ENOTTY;
1586 		goto out_unlock;
1587 	}
1588 
1589 	ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1590 	if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1591 		dev_warn(ctrl->device,
1592 			"NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1593 		ret = -EINVAL;
1594 		goto out_unlock;
1595 	}
1596 
1597 	dev_warn(ctrl->device,
1598 		"using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1599 	kref_get(&ns->kref);
1600 	mutex_unlock(&ctrl->namespaces_mutex);
1601 
1602 	ret = nvme_user_cmd(ctrl, ns, argp);
1603 	nvme_put_ns(ns);
1604 	return ret;
1605 
1606 out_unlock:
1607 	mutex_unlock(&ctrl->namespaces_mutex);
1608 	return ret;
1609 }
1610 
1611 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1612 		unsigned long arg)
1613 {
1614 	struct nvme_ctrl *ctrl = file->private_data;
1615 	void __user *argp = (void __user *)arg;
1616 
1617 	switch (cmd) {
1618 	case NVME_IOCTL_ADMIN_CMD:
1619 		return nvme_user_cmd(ctrl, NULL, argp);
1620 	case NVME_IOCTL_IO_CMD:
1621 		return nvme_dev_user_cmd(ctrl, argp);
1622 	case NVME_IOCTL_RESET:
1623 		dev_warn(ctrl->device, "resetting controller\n");
1624 		return ctrl->ops->reset_ctrl(ctrl);
1625 	case NVME_IOCTL_SUBSYS_RESET:
1626 		return nvme_reset_subsystem(ctrl);
1627 	case NVME_IOCTL_RESCAN:
1628 		nvme_queue_scan(ctrl);
1629 		return 0;
1630 	default:
1631 		return -ENOTTY;
1632 	}
1633 }
1634 
1635 static const struct file_operations nvme_dev_fops = {
1636 	.owner		= THIS_MODULE,
1637 	.open		= nvme_dev_open,
1638 	.release	= nvme_dev_release,
1639 	.unlocked_ioctl	= nvme_dev_ioctl,
1640 	.compat_ioctl	= nvme_dev_ioctl,
1641 };
1642 
1643 static ssize_t nvme_sysfs_reset(struct device *dev,
1644 				struct device_attribute *attr, const char *buf,
1645 				size_t count)
1646 {
1647 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1648 	int ret;
1649 
1650 	ret = ctrl->ops->reset_ctrl(ctrl);
1651 	if (ret < 0)
1652 		return ret;
1653 	return count;
1654 }
1655 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1656 
1657 static ssize_t nvme_sysfs_rescan(struct device *dev,
1658 				struct device_attribute *attr, const char *buf,
1659 				size_t count)
1660 {
1661 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1662 
1663 	nvme_queue_scan(ctrl);
1664 	return count;
1665 }
1666 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1667 
1668 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1669 								char *buf)
1670 {
1671 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1672 	struct nvme_ctrl *ctrl = ns->ctrl;
1673 	int serial_len = sizeof(ctrl->serial);
1674 	int model_len = sizeof(ctrl->model);
1675 
1676 	if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1677 		return sprintf(buf, "eui.%16phN\n", ns->uuid);
1678 
1679 	if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1680 		return sprintf(buf, "eui.%8phN\n", ns->eui);
1681 
1682 	while (ctrl->serial[serial_len - 1] == ' ')
1683 		serial_len--;
1684 	while (ctrl->model[model_len - 1] == ' ')
1685 		model_len--;
1686 
1687 	return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1688 		serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1689 }
1690 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1691 
1692 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1693 								char *buf)
1694 {
1695 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1696 	return sprintf(buf, "%pU\n", ns->uuid);
1697 }
1698 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1699 
1700 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1701 								char *buf)
1702 {
1703 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1704 	return sprintf(buf, "%8phd\n", ns->eui);
1705 }
1706 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1707 
1708 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1709 								char *buf)
1710 {
1711 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1712 	return sprintf(buf, "%d\n", ns->ns_id);
1713 }
1714 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1715 
1716 static struct attribute *nvme_ns_attrs[] = {
1717 	&dev_attr_wwid.attr,
1718 	&dev_attr_uuid.attr,
1719 	&dev_attr_eui.attr,
1720 	&dev_attr_nsid.attr,
1721 	NULL,
1722 };
1723 
1724 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1725 		struct attribute *a, int n)
1726 {
1727 	struct device *dev = container_of(kobj, struct device, kobj);
1728 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1729 
1730 	if (a == &dev_attr_uuid.attr) {
1731 		if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1732 			return 0;
1733 	}
1734 	if (a == &dev_attr_eui.attr) {
1735 		if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1736 			return 0;
1737 	}
1738 	return a->mode;
1739 }
1740 
1741 static const struct attribute_group nvme_ns_attr_group = {
1742 	.attrs		= nvme_ns_attrs,
1743 	.is_visible	= nvme_ns_attrs_are_visible,
1744 };
1745 
1746 #define nvme_show_str_function(field)						\
1747 static ssize_t  field##_show(struct device *dev,				\
1748 			    struct device_attribute *attr, char *buf)		\
1749 {										\
1750         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
1751         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);	\
1752 }										\
1753 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1754 
1755 #define nvme_show_int_function(field)						\
1756 static ssize_t  field##_show(struct device *dev,				\
1757 			    struct device_attribute *attr, char *buf)		\
1758 {										\
1759         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
1760         return sprintf(buf, "%d\n", ctrl->field);	\
1761 }										\
1762 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1763 
1764 nvme_show_str_function(model);
1765 nvme_show_str_function(serial);
1766 nvme_show_str_function(firmware_rev);
1767 nvme_show_int_function(cntlid);
1768 
1769 static ssize_t nvme_sysfs_delete(struct device *dev,
1770 				struct device_attribute *attr, const char *buf,
1771 				size_t count)
1772 {
1773 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1774 
1775 	if (device_remove_file_self(dev, attr))
1776 		ctrl->ops->delete_ctrl(ctrl);
1777 	return count;
1778 }
1779 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1780 
1781 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1782 					 struct device_attribute *attr,
1783 					 char *buf)
1784 {
1785 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1786 
1787 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1788 }
1789 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1790 
1791 static ssize_t nvme_sysfs_show_state(struct device *dev,
1792 				     struct device_attribute *attr,
1793 				     char *buf)
1794 {
1795 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1796 	static const char *const state_name[] = {
1797 		[NVME_CTRL_NEW]		= "new",
1798 		[NVME_CTRL_LIVE]	= "live",
1799 		[NVME_CTRL_RESETTING]	= "resetting",
1800 		[NVME_CTRL_RECONNECTING]= "reconnecting",
1801 		[NVME_CTRL_DELETING]	= "deleting",
1802 		[NVME_CTRL_DEAD]	= "dead",
1803 	};
1804 
1805 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
1806 	    state_name[ctrl->state])
1807 		return sprintf(buf, "%s\n", state_name[ctrl->state]);
1808 
1809 	return sprintf(buf, "unknown state\n");
1810 }
1811 
1812 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
1813 
1814 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1815 					 struct device_attribute *attr,
1816 					 char *buf)
1817 {
1818 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1819 
1820 	return snprintf(buf, PAGE_SIZE, "%s\n",
1821 			ctrl->ops->get_subsysnqn(ctrl));
1822 }
1823 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1824 
1825 static ssize_t nvme_sysfs_show_address(struct device *dev,
1826 					 struct device_attribute *attr,
1827 					 char *buf)
1828 {
1829 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1830 
1831 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1832 }
1833 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1834 
1835 static struct attribute *nvme_dev_attrs[] = {
1836 	&dev_attr_reset_controller.attr,
1837 	&dev_attr_rescan_controller.attr,
1838 	&dev_attr_model.attr,
1839 	&dev_attr_serial.attr,
1840 	&dev_attr_firmware_rev.attr,
1841 	&dev_attr_cntlid.attr,
1842 	&dev_attr_delete_controller.attr,
1843 	&dev_attr_transport.attr,
1844 	&dev_attr_subsysnqn.attr,
1845 	&dev_attr_address.attr,
1846 	&dev_attr_state.attr,
1847 	NULL
1848 };
1849 
1850 #define CHECK_ATTR(ctrl, a, name)		\
1851 	if ((a) == &dev_attr_##name.attr &&	\
1852 	    !(ctrl)->ops->get_##name)		\
1853 		return 0
1854 
1855 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1856 		struct attribute *a, int n)
1857 {
1858 	struct device *dev = container_of(kobj, struct device, kobj);
1859 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1860 
1861 	if (a == &dev_attr_delete_controller.attr) {
1862 		if (!ctrl->ops->delete_ctrl)
1863 			return 0;
1864 	}
1865 
1866 	CHECK_ATTR(ctrl, a, subsysnqn);
1867 	CHECK_ATTR(ctrl, a, address);
1868 
1869 	return a->mode;
1870 }
1871 
1872 static struct attribute_group nvme_dev_attrs_group = {
1873 	.attrs		= nvme_dev_attrs,
1874 	.is_visible	= nvme_dev_attrs_are_visible,
1875 };
1876 
1877 static const struct attribute_group *nvme_dev_attr_groups[] = {
1878 	&nvme_dev_attrs_group,
1879 	NULL,
1880 };
1881 
1882 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1883 {
1884 	struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1885 	struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1886 
1887 	return nsa->ns_id - nsb->ns_id;
1888 }
1889 
1890 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1891 {
1892 	struct nvme_ns *ns, *ret = NULL;
1893 
1894 	mutex_lock(&ctrl->namespaces_mutex);
1895 	list_for_each_entry(ns, &ctrl->namespaces, list) {
1896 		if (ns->ns_id == nsid) {
1897 			kref_get(&ns->kref);
1898 			ret = ns;
1899 			break;
1900 		}
1901 		if (ns->ns_id > nsid)
1902 			break;
1903 	}
1904 	mutex_unlock(&ctrl->namespaces_mutex);
1905 	return ret;
1906 }
1907 
1908 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1909 {
1910 	struct nvme_ns *ns;
1911 	struct gendisk *disk;
1912 	struct nvme_id_ns *id;
1913 	char disk_name[DISK_NAME_LEN];
1914 	int node = dev_to_node(ctrl->dev);
1915 
1916 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1917 	if (!ns)
1918 		return;
1919 
1920 	ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1921 	if (ns->instance < 0)
1922 		goto out_free_ns;
1923 
1924 	ns->queue = blk_mq_init_queue(ctrl->tagset);
1925 	if (IS_ERR(ns->queue))
1926 		goto out_release_instance;
1927 	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1928 	ns->queue->queuedata = ns;
1929 	ns->ctrl = ctrl;
1930 
1931 	kref_init(&ns->kref);
1932 	ns->ns_id = nsid;
1933 	ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1934 
1935 	blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1936 	nvme_set_queue_limits(ctrl, ns->queue);
1937 
1938 	sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1939 
1940 	if (nvme_revalidate_ns(ns, &id))
1941 		goto out_free_queue;
1942 
1943 	if (nvme_nvm_ns_supported(ns, id) &&
1944 				nvme_nvm_register(ns, disk_name, node)) {
1945 		dev_warn(ctrl->dev, "%s: LightNVM init failure\n", __func__);
1946 		goto out_free_id;
1947 	}
1948 
1949 	disk = alloc_disk_node(0, node);
1950 	if (!disk)
1951 		goto out_free_id;
1952 
1953 	disk->fops = &nvme_fops;
1954 	disk->private_data = ns;
1955 	disk->queue = ns->queue;
1956 	disk->flags = GENHD_FL_EXT_DEVT;
1957 	memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
1958 	ns->disk = disk;
1959 
1960 	__nvme_revalidate_disk(disk, id);
1961 
1962 	mutex_lock(&ctrl->namespaces_mutex);
1963 	list_add_tail(&ns->list, &ctrl->namespaces);
1964 	mutex_unlock(&ctrl->namespaces_mutex);
1965 
1966 	kref_get(&ctrl->kref);
1967 
1968 	kfree(id);
1969 
1970 	device_add_disk(ctrl->device, ns->disk);
1971 	if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1972 					&nvme_ns_attr_group))
1973 		pr_warn("%s: failed to create sysfs group for identification\n",
1974 			ns->disk->disk_name);
1975 	if (ns->ndev && nvme_nvm_register_sysfs(ns))
1976 		pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
1977 			ns->disk->disk_name);
1978 	return;
1979  out_free_id:
1980 	kfree(id);
1981  out_free_queue:
1982 	blk_cleanup_queue(ns->queue);
1983  out_release_instance:
1984 	ida_simple_remove(&ctrl->ns_ida, ns->instance);
1985  out_free_ns:
1986 	kfree(ns);
1987 }
1988 
1989 static void nvme_ns_remove(struct nvme_ns *ns)
1990 {
1991 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1992 		return;
1993 
1994 	if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
1995 		if (blk_get_integrity(ns->disk))
1996 			blk_integrity_unregister(ns->disk);
1997 		sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1998 					&nvme_ns_attr_group);
1999 		if (ns->ndev)
2000 			nvme_nvm_unregister_sysfs(ns);
2001 		del_gendisk(ns->disk);
2002 		blk_mq_abort_requeue_list(ns->queue);
2003 		blk_cleanup_queue(ns->queue);
2004 	}
2005 
2006 	mutex_lock(&ns->ctrl->namespaces_mutex);
2007 	list_del_init(&ns->list);
2008 	mutex_unlock(&ns->ctrl->namespaces_mutex);
2009 
2010 	nvme_put_ns(ns);
2011 }
2012 
2013 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2014 {
2015 	struct nvme_ns *ns;
2016 
2017 	ns = nvme_find_get_ns(ctrl, nsid);
2018 	if (ns) {
2019 		if (ns->disk && revalidate_disk(ns->disk))
2020 			nvme_ns_remove(ns);
2021 		nvme_put_ns(ns);
2022 	} else
2023 		nvme_alloc_ns(ctrl, nsid);
2024 }
2025 
2026 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
2027 					unsigned nsid)
2028 {
2029 	struct nvme_ns *ns, *next;
2030 
2031 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
2032 		if (ns->ns_id > nsid)
2033 			nvme_ns_remove(ns);
2034 	}
2035 }
2036 
2037 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
2038 {
2039 	struct nvme_ns *ns;
2040 	__le32 *ns_list;
2041 	unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
2042 	int ret = 0;
2043 
2044 	ns_list = kzalloc(0x1000, GFP_KERNEL);
2045 	if (!ns_list)
2046 		return -ENOMEM;
2047 
2048 	for (i = 0; i < num_lists; i++) {
2049 		ret = nvme_identify_ns_list(ctrl, prev, ns_list);
2050 		if (ret)
2051 			goto free;
2052 
2053 		for (j = 0; j < min(nn, 1024U); j++) {
2054 			nsid = le32_to_cpu(ns_list[j]);
2055 			if (!nsid)
2056 				goto out;
2057 
2058 			nvme_validate_ns(ctrl, nsid);
2059 
2060 			while (++prev < nsid) {
2061 				ns = nvme_find_get_ns(ctrl, prev);
2062 				if (ns) {
2063 					nvme_ns_remove(ns);
2064 					nvme_put_ns(ns);
2065 				}
2066 			}
2067 		}
2068 		nn -= j;
2069 	}
2070  out:
2071 	nvme_remove_invalid_namespaces(ctrl, prev);
2072  free:
2073 	kfree(ns_list);
2074 	return ret;
2075 }
2076 
2077 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
2078 {
2079 	unsigned i;
2080 
2081 	for (i = 1; i <= nn; i++)
2082 		nvme_validate_ns(ctrl, i);
2083 
2084 	nvme_remove_invalid_namespaces(ctrl, nn);
2085 }
2086 
2087 static void nvme_scan_work(struct work_struct *work)
2088 {
2089 	struct nvme_ctrl *ctrl =
2090 		container_of(work, struct nvme_ctrl, scan_work);
2091 	struct nvme_id_ctrl *id;
2092 	unsigned nn;
2093 
2094 	if (ctrl->state != NVME_CTRL_LIVE)
2095 		return;
2096 
2097 	if (nvme_identify_ctrl(ctrl, &id))
2098 		return;
2099 
2100 	nn = le32_to_cpu(id->nn);
2101 	if (ctrl->vs >= NVME_VS(1, 1, 0) &&
2102 	    !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
2103 		if (!nvme_scan_ns_list(ctrl, nn))
2104 			goto done;
2105 	}
2106 	nvme_scan_ns_sequential(ctrl, nn);
2107  done:
2108 	mutex_lock(&ctrl->namespaces_mutex);
2109 	list_sort(NULL, &ctrl->namespaces, ns_cmp);
2110 	mutex_unlock(&ctrl->namespaces_mutex);
2111 	kfree(id);
2112 }
2113 
2114 void nvme_queue_scan(struct nvme_ctrl *ctrl)
2115 {
2116 	/*
2117 	 * Do not queue new scan work when a controller is reset during
2118 	 * removal.
2119 	 */
2120 	if (ctrl->state == NVME_CTRL_LIVE)
2121 		schedule_work(&ctrl->scan_work);
2122 }
2123 EXPORT_SYMBOL_GPL(nvme_queue_scan);
2124 
2125 /*
2126  * This function iterates the namespace list unlocked to allow recovery from
2127  * controller failure. It is up to the caller to ensure the namespace list is
2128  * not modified by scan work while this function is executing.
2129  */
2130 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
2131 {
2132 	struct nvme_ns *ns, *next;
2133 
2134 	/*
2135 	 * The dead states indicates the controller was not gracefully
2136 	 * disconnected. In that case, we won't be able to flush any data while
2137 	 * removing the namespaces' disks; fail all the queues now to avoid
2138 	 * potentially having to clean up the failed sync later.
2139 	 */
2140 	if (ctrl->state == NVME_CTRL_DEAD)
2141 		nvme_kill_queues(ctrl);
2142 
2143 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
2144 		nvme_ns_remove(ns);
2145 }
2146 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
2147 
2148 static void nvme_async_event_work(struct work_struct *work)
2149 {
2150 	struct nvme_ctrl *ctrl =
2151 		container_of(work, struct nvme_ctrl, async_event_work);
2152 
2153 	spin_lock_irq(&ctrl->lock);
2154 	while (ctrl->event_limit > 0) {
2155 		int aer_idx = --ctrl->event_limit;
2156 
2157 		spin_unlock_irq(&ctrl->lock);
2158 		ctrl->ops->submit_async_event(ctrl, aer_idx);
2159 		spin_lock_irq(&ctrl->lock);
2160 	}
2161 	spin_unlock_irq(&ctrl->lock);
2162 }
2163 
2164 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
2165 		union nvme_result *res)
2166 {
2167 	u32 result = le32_to_cpu(res->u32);
2168 	bool done = true;
2169 
2170 	switch (le16_to_cpu(status) >> 1) {
2171 	case NVME_SC_SUCCESS:
2172 		done = false;
2173 		/*FALLTHRU*/
2174 	case NVME_SC_ABORT_REQ:
2175 		++ctrl->event_limit;
2176 		schedule_work(&ctrl->async_event_work);
2177 		break;
2178 	default:
2179 		break;
2180 	}
2181 
2182 	if (done)
2183 		return;
2184 
2185 	switch (result & 0xff07) {
2186 	case NVME_AER_NOTICE_NS_CHANGED:
2187 		dev_info(ctrl->device, "rescanning\n");
2188 		nvme_queue_scan(ctrl);
2189 		break;
2190 	default:
2191 		dev_warn(ctrl->device, "async event result %08x\n", result);
2192 	}
2193 }
2194 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
2195 
2196 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
2197 {
2198 	ctrl->event_limit = NVME_NR_AERS;
2199 	schedule_work(&ctrl->async_event_work);
2200 }
2201 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
2202 
2203 static DEFINE_IDA(nvme_instance_ida);
2204 
2205 static int nvme_set_instance(struct nvme_ctrl *ctrl)
2206 {
2207 	int instance, error;
2208 
2209 	do {
2210 		if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2211 			return -ENODEV;
2212 
2213 		spin_lock(&dev_list_lock);
2214 		error = ida_get_new(&nvme_instance_ida, &instance);
2215 		spin_unlock(&dev_list_lock);
2216 	} while (error == -EAGAIN);
2217 
2218 	if (error)
2219 		return -ENODEV;
2220 
2221 	ctrl->instance = instance;
2222 	return 0;
2223 }
2224 
2225 static void nvme_release_instance(struct nvme_ctrl *ctrl)
2226 {
2227 	spin_lock(&dev_list_lock);
2228 	ida_remove(&nvme_instance_ida, ctrl->instance);
2229 	spin_unlock(&dev_list_lock);
2230 }
2231 
2232 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
2233 {
2234 	flush_work(&ctrl->async_event_work);
2235 	flush_work(&ctrl->scan_work);
2236 	nvme_remove_namespaces(ctrl);
2237 
2238 	device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
2239 
2240 	spin_lock(&dev_list_lock);
2241 	list_del(&ctrl->node);
2242 	spin_unlock(&dev_list_lock);
2243 }
2244 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
2245 
2246 static void nvme_free_ctrl(struct kref *kref)
2247 {
2248 	struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
2249 
2250 	put_device(ctrl->device);
2251 	nvme_release_instance(ctrl);
2252 	ida_destroy(&ctrl->ns_ida);
2253 
2254 	ctrl->ops->free_ctrl(ctrl);
2255 }
2256 
2257 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
2258 {
2259 	kref_put(&ctrl->kref, nvme_free_ctrl);
2260 }
2261 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2262 
2263 /*
2264  * Initialize a NVMe controller structures.  This needs to be called during
2265  * earliest initialization so that we have the initialized structured around
2266  * during probing.
2267  */
2268 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2269 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
2270 {
2271 	int ret;
2272 
2273 	ctrl->state = NVME_CTRL_NEW;
2274 	spin_lock_init(&ctrl->lock);
2275 	INIT_LIST_HEAD(&ctrl->namespaces);
2276 	mutex_init(&ctrl->namespaces_mutex);
2277 	kref_init(&ctrl->kref);
2278 	ctrl->dev = dev;
2279 	ctrl->ops = ops;
2280 	ctrl->quirks = quirks;
2281 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2282 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2283 
2284 	ret = nvme_set_instance(ctrl);
2285 	if (ret)
2286 		goto out;
2287 
2288 	ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2289 				MKDEV(nvme_char_major, ctrl->instance),
2290 				ctrl, nvme_dev_attr_groups,
2291 				"nvme%d", ctrl->instance);
2292 	if (IS_ERR(ctrl->device)) {
2293 		ret = PTR_ERR(ctrl->device);
2294 		goto out_release_instance;
2295 	}
2296 	get_device(ctrl->device);
2297 	ida_init(&ctrl->ns_ida);
2298 
2299 	spin_lock(&dev_list_lock);
2300 	list_add_tail(&ctrl->node, &nvme_ctrl_list);
2301 	spin_unlock(&dev_list_lock);
2302 
2303 	/*
2304 	 * Initialize latency tolerance controls.  The sysfs files won't
2305 	 * be visible to userspace unless the device actually supports APST.
2306 	 */
2307 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
2308 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
2309 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
2310 
2311 	return 0;
2312 out_release_instance:
2313 	nvme_release_instance(ctrl);
2314 out:
2315 	return ret;
2316 }
2317 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2318 
2319 /**
2320  * nvme_kill_queues(): Ends all namespace queues
2321  * @ctrl: the dead controller that needs to end
2322  *
2323  * Call this function when the driver determines it is unable to get the
2324  * controller in a state capable of servicing IO.
2325  */
2326 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2327 {
2328 	struct nvme_ns *ns;
2329 
2330 	mutex_lock(&ctrl->namespaces_mutex);
2331 	list_for_each_entry(ns, &ctrl->namespaces, list) {
2332 		/*
2333 		 * Revalidating a dead namespace sets capacity to 0. This will
2334 		 * end buffered writers dirtying pages that can't be synced.
2335 		 */
2336 		if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2337 			continue;
2338 		revalidate_disk(ns->disk);
2339 		blk_set_queue_dying(ns->queue);
2340 		blk_mq_abort_requeue_list(ns->queue);
2341 		blk_mq_start_stopped_hw_queues(ns->queue, true);
2342 	}
2343 	mutex_unlock(&ctrl->namespaces_mutex);
2344 }
2345 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2346 
2347 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2348 {
2349 	struct nvme_ns *ns;
2350 
2351 	mutex_lock(&ctrl->namespaces_mutex);
2352 	list_for_each_entry(ns, &ctrl->namespaces, list)
2353 		blk_mq_quiesce_queue(ns->queue);
2354 	mutex_unlock(&ctrl->namespaces_mutex);
2355 }
2356 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2357 
2358 void nvme_start_queues(struct nvme_ctrl *ctrl)
2359 {
2360 	struct nvme_ns *ns;
2361 
2362 	mutex_lock(&ctrl->namespaces_mutex);
2363 	list_for_each_entry(ns, &ctrl->namespaces, list) {
2364 		blk_mq_start_stopped_hw_queues(ns->queue, true);
2365 		blk_mq_kick_requeue_list(ns->queue);
2366 	}
2367 	mutex_unlock(&ctrl->namespaces_mutex);
2368 }
2369 EXPORT_SYMBOL_GPL(nvme_start_queues);
2370 
2371 int __init nvme_core_init(void)
2372 {
2373 	int result;
2374 
2375 	result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2376 							&nvme_dev_fops);
2377 	if (result < 0)
2378 		return result;
2379 	else if (result > 0)
2380 		nvme_char_major = result;
2381 
2382 	nvme_class = class_create(THIS_MODULE, "nvme");
2383 	if (IS_ERR(nvme_class)) {
2384 		result = PTR_ERR(nvme_class);
2385 		goto unregister_chrdev;
2386 	}
2387 
2388 	return 0;
2389 
2390  unregister_chrdev:
2391 	__unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2392 	return result;
2393 }
2394 
2395 void nvme_core_exit(void)
2396 {
2397 	class_destroy(nvme_class);
2398 	__unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2399 }
2400 
2401 MODULE_LICENSE("GPL");
2402 MODULE_VERSION("1.0");
2403 module_init(nvme_core_init);
2404 module_exit(nvme_core_exit);
2405