xref: /linux/drivers/nvme/host/core.c (revision 92ce4c3ea7c44e61ca2b6ef3e5682bfcea851d87)
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 <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 static 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 static u8 nvme_max_retries = 5;
52 module_param_named(max_retries, nvme_max_retries, byte, 0644);
53 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
54 
55 static int nvme_char_major;
56 module_param(nvme_char_major, int, 0);
57 
58 static unsigned long default_ps_max_latency_us = 100000;
59 module_param(default_ps_max_latency_us, ulong, 0644);
60 MODULE_PARM_DESC(default_ps_max_latency_us,
61 		 "max power saving latency for new devices; use PM QOS to change per device");
62 
63 static bool force_apst;
64 module_param(force_apst, bool, 0644);
65 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
66 
67 static bool streams;
68 module_param(streams, bool, 0644);
69 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
70 
71 struct workqueue_struct *nvme_wq;
72 EXPORT_SYMBOL_GPL(nvme_wq);
73 
74 static LIST_HEAD(nvme_ctrl_list);
75 static DEFINE_SPINLOCK(dev_list_lock);
76 
77 static struct class *nvme_class;
78 
79 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
80 {
81 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
82 		return -EBUSY;
83 	if (!queue_work(nvme_wq, &ctrl->reset_work))
84 		return -EBUSY;
85 	return 0;
86 }
87 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
88 
89 static int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
90 {
91 	int ret;
92 
93 	ret = nvme_reset_ctrl(ctrl);
94 	if (!ret)
95 		flush_work(&ctrl->reset_work);
96 	return ret;
97 }
98 
99 static blk_status_t nvme_error_status(struct request *req)
100 {
101 	switch (nvme_req(req)->status & 0x7ff) {
102 	case NVME_SC_SUCCESS:
103 		return BLK_STS_OK;
104 	case NVME_SC_CAP_EXCEEDED:
105 		return BLK_STS_NOSPC;
106 	case NVME_SC_ONCS_NOT_SUPPORTED:
107 		return BLK_STS_NOTSUPP;
108 	case NVME_SC_WRITE_FAULT:
109 	case NVME_SC_READ_ERROR:
110 	case NVME_SC_UNWRITTEN_BLOCK:
111 		return BLK_STS_MEDIUM;
112 	default:
113 		return BLK_STS_IOERR;
114 	}
115 }
116 
117 static inline bool nvme_req_needs_retry(struct request *req)
118 {
119 	if (blk_noretry_request(req))
120 		return false;
121 	if (nvme_req(req)->status & NVME_SC_DNR)
122 		return false;
123 	if (jiffies - req->start_time >= req->timeout)
124 		return false;
125 	if (nvme_req(req)->retries >= nvme_max_retries)
126 		return false;
127 	return true;
128 }
129 
130 void nvme_complete_rq(struct request *req)
131 {
132 	if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) {
133 		nvme_req(req)->retries++;
134 		blk_mq_requeue_request(req, true);
135 		return;
136 	}
137 
138 	blk_mq_end_request(req, nvme_error_status(req));
139 }
140 EXPORT_SYMBOL_GPL(nvme_complete_rq);
141 
142 void nvme_cancel_request(struct request *req, void *data, bool reserved)
143 {
144 	int status;
145 
146 	if (!blk_mq_request_started(req))
147 		return;
148 
149 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
150 				"Cancelling I/O %d", req->tag);
151 
152 	status = NVME_SC_ABORT_REQ;
153 	if (blk_queue_dying(req->q))
154 		status |= NVME_SC_DNR;
155 	nvme_req(req)->status = status;
156 	blk_mq_complete_request(req);
157 
158 }
159 EXPORT_SYMBOL_GPL(nvme_cancel_request);
160 
161 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
162 		enum nvme_ctrl_state new_state)
163 {
164 	enum nvme_ctrl_state old_state;
165 	bool changed = false;
166 
167 	spin_lock_irq(&ctrl->lock);
168 
169 	old_state = ctrl->state;
170 	switch (new_state) {
171 	case NVME_CTRL_LIVE:
172 		switch (old_state) {
173 		case NVME_CTRL_NEW:
174 		case NVME_CTRL_RESETTING:
175 		case NVME_CTRL_RECONNECTING:
176 			changed = true;
177 			/* FALLTHRU */
178 		default:
179 			break;
180 		}
181 		break;
182 	case NVME_CTRL_RESETTING:
183 		switch (old_state) {
184 		case NVME_CTRL_NEW:
185 		case NVME_CTRL_LIVE:
186 			changed = true;
187 			/* FALLTHRU */
188 		default:
189 			break;
190 		}
191 		break;
192 	case NVME_CTRL_RECONNECTING:
193 		switch (old_state) {
194 		case NVME_CTRL_LIVE:
195 			changed = true;
196 			/* FALLTHRU */
197 		default:
198 			break;
199 		}
200 		break;
201 	case NVME_CTRL_DELETING:
202 		switch (old_state) {
203 		case NVME_CTRL_LIVE:
204 		case NVME_CTRL_RESETTING:
205 		case NVME_CTRL_RECONNECTING:
206 			changed = true;
207 			/* FALLTHRU */
208 		default:
209 			break;
210 		}
211 		break;
212 	case NVME_CTRL_DEAD:
213 		switch (old_state) {
214 		case NVME_CTRL_DELETING:
215 			changed = true;
216 			/* FALLTHRU */
217 		default:
218 			break;
219 		}
220 		break;
221 	default:
222 		break;
223 	}
224 
225 	if (changed)
226 		ctrl->state = new_state;
227 
228 	spin_unlock_irq(&ctrl->lock);
229 
230 	return changed;
231 }
232 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
233 
234 static void nvme_free_ns(struct kref *kref)
235 {
236 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
237 
238 	if (ns->ndev)
239 		nvme_nvm_unregister(ns);
240 
241 	if (ns->disk) {
242 		spin_lock(&dev_list_lock);
243 		ns->disk->private_data = NULL;
244 		spin_unlock(&dev_list_lock);
245 	}
246 
247 	put_disk(ns->disk);
248 	ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
249 	nvme_put_ctrl(ns->ctrl);
250 	kfree(ns);
251 }
252 
253 static void nvme_put_ns(struct nvme_ns *ns)
254 {
255 	kref_put(&ns->kref, nvme_free_ns);
256 }
257 
258 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
259 {
260 	struct nvme_ns *ns;
261 
262 	spin_lock(&dev_list_lock);
263 	ns = disk->private_data;
264 	if (ns) {
265 		if (!kref_get_unless_zero(&ns->kref))
266 			goto fail;
267 		if (!try_module_get(ns->ctrl->ops->module))
268 			goto fail_put_ns;
269 	}
270 	spin_unlock(&dev_list_lock);
271 
272 	return ns;
273 
274 fail_put_ns:
275 	kref_put(&ns->kref, nvme_free_ns);
276 fail:
277 	spin_unlock(&dev_list_lock);
278 	return NULL;
279 }
280 
281 struct request *nvme_alloc_request(struct request_queue *q,
282 		struct nvme_command *cmd, unsigned int flags, int qid)
283 {
284 	unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
285 	struct request *req;
286 
287 	if (qid == NVME_QID_ANY) {
288 		req = blk_mq_alloc_request(q, op, flags);
289 	} else {
290 		req = blk_mq_alloc_request_hctx(q, op, flags,
291 				qid ? qid - 1 : 0);
292 	}
293 	if (IS_ERR(req))
294 		return req;
295 
296 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
297 	nvme_req(req)->cmd = cmd;
298 
299 	return req;
300 }
301 EXPORT_SYMBOL_GPL(nvme_alloc_request);
302 
303 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
304 {
305 	struct nvme_command c;
306 
307 	memset(&c, 0, sizeof(c));
308 
309 	c.directive.opcode = nvme_admin_directive_send;
310 	c.directive.nsid = cpu_to_le32(0xffffffff);
311 	c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
312 	c.directive.dtype = NVME_DIR_IDENTIFY;
313 	c.directive.tdtype = NVME_DIR_STREAMS;
314 	c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
315 
316 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
317 }
318 
319 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
320 {
321 	return nvme_toggle_streams(ctrl, false);
322 }
323 
324 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
325 {
326 	return nvme_toggle_streams(ctrl, true);
327 }
328 
329 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
330 				  struct streams_directive_params *s, u32 nsid)
331 {
332 	struct nvme_command c;
333 
334 	memset(&c, 0, sizeof(c));
335 	memset(s, 0, sizeof(*s));
336 
337 	c.directive.opcode = nvme_admin_directive_recv;
338 	c.directive.nsid = cpu_to_le32(nsid);
339 	c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
340 	c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
341 	c.directive.dtype = NVME_DIR_STREAMS;
342 
343 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
344 }
345 
346 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
347 {
348 	struct streams_directive_params s;
349 	int ret;
350 
351 	if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
352 		return 0;
353 	if (!streams)
354 		return 0;
355 
356 	ret = nvme_enable_streams(ctrl);
357 	if (ret)
358 		return ret;
359 
360 	ret = nvme_get_stream_params(ctrl, &s, 0xffffffff);
361 	if (ret)
362 		return ret;
363 
364 	ctrl->nssa = le16_to_cpu(s.nssa);
365 	if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
366 		dev_info(ctrl->device, "too few streams (%u) available\n",
367 					ctrl->nssa);
368 		nvme_disable_streams(ctrl);
369 		return 0;
370 	}
371 
372 	ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
373 	dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
374 	return 0;
375 }
376 
377 /*
378  * Check if 'req' has a write hint associated with it. If it does, assign
379  * a valid namespace stream to the write.
380  */
381 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
382 				     struct request *req, u16 *control,
383 				     u32 *dsmgmt)
384 {
385 	enum rw_hint streamid = req->write_hint;
386 
387 	if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
388 		streamid = 0;
389 	else {
390 		streamid--;
391 		if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
392 			return;
393 
394 		*control |= NVME_RW_DTYPE_STREAMS;
395 		*dsmgmt |= streamid << 16;
396 	}
397 
398 	if (streamid < ARRAY_SIZE(req->q->write_hints))
399 		req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
400 }
401 
402 static inline void nvme_setup_flush(struct nvme_ns *ns,
403 		struct nvme_command *cmnd)
404 {
405 	memset(cmnd, 0, sizeof(*cmnd));
406 	cmnd->common.opcode = nvme_cmd_flush;
407 	cmnd->common.nsid = cpu_to_le32(ns->ns_id);
408 }
409 
410 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
411 		struct nvme_command *cmnd)
412 {
413 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
414 	struct nvme_dsm_range *range;
415 	struct bio *bio;
416 
417 	range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
418 	if (!range)
419 		return BLK_STS_RESOURCE;
420 
421 	__rq_for_each_bio(bio, req) {
422 		u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
423 		u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
424 
425 		range[n].cattr = cpu_to_le32(0);
426 		range[n].nlb = cpu_to_le32(nlb);
427 		range[n].slba = cpu_to_le64(slba);
428 		n++;
429 	}
430 
431 	if (WARN_ON_ONCE(n != segments)) {
432 		kfree(range);
433 		return BLK_STS_IOERR;
434 	}
435 
436 	memset(cmnd, 0, sizeof(*cmnd));
437 	cmnd->dsm.opcode = nvme_cmd_dsm;
438 	cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
439 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
440 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
441 
442 	req->special_vec.bv_page = virt_to_page(range);
443 	req->special_vec.bv_offset = offset_in_page(range);
444 	req->special_vec.bv_len = sizeof(*range) * segments;
445 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
446 
447 	return BLK_STS_OK;
448 }
449 
450 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
451 		struct request *req, struct nvme_command *cmnd)
452 {
453 	struct nvme_ctrl *ctrl = ns->ctrl;
454 	u16 control = 0;
455 	u32 dsmgmt = 0;
456 
457 	/*
458 	 * If formated with metadata, require the block layer provide a buffer
459 	 * unless this namespace is formated such that the metadata can be
460 	 * stripped/generated by the controller with PRACT=1.
461 	 */
462 	if (ns && ns->ms &&
463 	    (!ns->pi_type || ns->ms != sizeof(struct t10_pi_tuple)) &&
464 	    !blk_integrity_rq(req) && !blk_rq_is_passthrough(req))
465 		return BLK_STS_NOTSUPP;
466 
467 	if (req->cmd_flags & REQ_FUA)
468 		control |= NVME_RW_FUA;
469 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
470 		control |= NVME_RW_LR;
471 
472 	if (req->cmd_flags & REQ_RAHEAD)
473 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
474 
475 	memset(cmnd, 0, sizeof(*cmnd));
476 	cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
477 	cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
478 	cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
479 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
480 
481 	if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
482 		nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
483 
484 	if (ns->ms) {
485 		switch (ns->pi_type) {
486 		case NVME_NS_DPS_PI_TYPE3:
487 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
488 			break;
489 		case NVME_NS_DPS_PI_TYPE1:
490 		case NVME_NS_DPS_PI_TYPE2:
491 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
492 					NVME_RW_PRINFO_PRCHK_REF;
493 			cmnd->rw.reftag = cpu_to_le32(
494 					nvme_block_nr(ns, blk_rq_pos(req)));
495 			break;
496 		}
497 		if (!blk_integrity_rq(req))
498 			control |= NVME_RW_PRINFO_PRACT;
499 	}
500 
501 	cmnd->rw.control = cpu_to_le16(control);
502 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
503 	return 0;
504 }
505 
506 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
507 		struct nvme_command *cmd)
508 {
509 	blk_status_t ret = BLK_STS_OK;
510 
511 	if (!(req->rq_flags & RQF_DONTPREP)) {
512 		nvme_req(req)->retries = 0;
513 		nvme_req(req)->flags = 0;
514 		req->rq_flags |= RQF_DONTPREP;
515 	}
516 
517 	switch (req_op(req)) {
518 	case REQ_OP_DRV_IN:
519 	case REQ_OP_DRV_OUT:
520 		memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
521 		break;
522 	case REQ_OP_FLUSH:
523 		nvme_setup_flush(ns, cmd);
524 		break;
525 	case REQ_OP_WRITE_ZEROES:
526 		/* currently only aliased to deallocate for a few ctrls: */
527 	case REQ_OP_DISCARD:
528 		ret = nvme_setup_discard(ns, req, cmd);
529 		break;
530 	case REQ_OP_READ:
531 	case REQ_OP_WRITE:
532 		ret = nvme_setup_rw(ns, req, cmd);
533 		break;
534 	default:
535 		WARN_ON_ONCE(1);
536 		return BLK_STS_IOERR;
537 	}
538 
539 	cmd->common.command_id = req->tag;
540 	return ret;
541 }
542 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
543 
544 /*
545  * Returns 0 on success.  If the result is negative, it's a Linux error code;
546  * if the result is positive, it's an NVM Express status code
547  */
548 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
549 		union nvme_result *result, void *buffer, unsigned bufflen,
550 		unsigned timeout, int qid, int at_head, int flags)
551 {
552 	struct request *req;
553 	int ret;
554 
555 	req = nvme_alloc_request(q, cmd, flags, qid);
556 	if (IS_ERR(req))
557 		return PTR_ERR(req);
558 
559 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
560 
561 	if (buffer && bufflen) {
562 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
563 		if (ret)
564 			goto out;
565 	}
566 
567 	blk_execute_rq(req->q, NULL, req, at_head);
568 	if (result)
569 		*result = nvme_req(req)->result;
570 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
571 		ret = -EINTR;
572 	else
573 		ret = nvme_req(req)->status;
574  out:
575 	blk_mq_free_request(req);
576 	return ret;
577 }
578 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
579 
580 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
581 		void *buffer, unsigned bufflen)
582 {
583 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
584 			NVME_QID_ANY, 0, 0);
585 }
586 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
587 
588 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
589 		void __user *ubuffer, unsigned bufflen,
590 		void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
591 		u32 *result, unsigned timeout)
592 {
593 	bool write = nvme_is_write(cmd);
594 	struct nvme_ns *ns = q->queuedata;
595 	struct gendisk *disk = ns ? ns->disk : NULL;
596 	struct request *req;
597 	struct bio *bio = NULL;
598 	void *meta = NULL;
599 	int ret;
600 
601 	req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
602 	if (IS_ERR(req))
603 		return PTR_ERR(req);
604 
605 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
606 
607 	if (ubuffer && bufflen) {
608 		ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
609 				GFP_KERNEL);
610 		if (ret)
611 			goto out;
612 		bio = req->bio;
613 
614 		if (!disk)
615 			goto submit;
616 		bio->bi_disk = disk;
617 
618 		if (meta_buffer && meta_len) {
619 			struct bio_integrity_payload *bip;
620 
621 			meta = kmalloc(meta_len, GFP_KERNEL);
622 			if (!meta) {
623 				ret = -ENOMEM;
624 				goto out_unmap;
625 			}
626 
627 			if (write) {
628 				if (copy_from_user(meta, meta_buffer,
629 						meta_len)) {
630 					ret = -EFAULT;
631 					goto out_free_meta;
632 				}
633 			}
634 
635 			bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
636 			if (IS_ERR(bip)) {
637 				ret = PTR_ERR(bip);
638 				goto out_free_meta;
639 			}
640 
641 			bip->bip_iter.bi_size = meta_len;
642 			bip->bip_iter.bi_sector = meta_seed;
643 
644 			ret = bio_integrity_add_page(bio, virt_to_page(meta),
645 					meta_len, offset_in_page(meta));
646 			if (ret != meta_len) {
647 				ret = -ENOMEM;
648 				goto out_free_meta;
649 			}
650 		}
651 	}
652  submit:
653 	blk_execute_rq(req->q, disk, req, 0);
654 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
655 		ret = -EINTR;
656 	else
657 		ret = nvme_req(req)->status;
658 	if (result)
659 		*result = le32_to_cpu(nvme_req(req)->result.u32);
660 	if (meta && !ret && !write) {
661 		if (copy_to_user(meta_buffer, meta, meta_len))
662 			ret = -EFAULT;
663 	}
664  out_free_meta:
665 	kfree(meta);
666  out_unmap:
667 	if (bio)
668 		blk_rq_unmap_user(bio);
669  out:
670 	blk_mq_free_request(req);
671 	return ret;
672 }
673 
674 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
675 		void __user *ubuffer, unsigned bufflen, u32 *result,
676 		unsigned timeout)
677 {
678 	return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
679 			result, timeout);
680 }
681 
682 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
683 {
684 	struct nvme_ctrl *ctrl = rq->end_io_data;
685 
686 	blk_mq_free_request(rq);
687 
688 	if (status) {
689 		dev_err(ctrl->device,
690 			"failed nvme_keep_alive_end_io error=%d\n",
691 				status);
692 		return;
693 	}
694 
695 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
696 }
697 
698 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
699 {
700 	struct nvme_command c;
701 	struct request *rq;
702 
703 	memset(&c, 0, sizeof(c));
704 	c.common.opcode = nvme_admin_keep_alive;
705 
706 	rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
707 			NVME_QID_ANY);
708 	if (IS_ERR(rq))
709 		return PTR_ERR(rq);
710 
711 	rq->timeout = ctrl->kato * HZ;
712 	rq->end_io_data = ctrl;
713 
714 	blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
715 
716 	return 0;
717 }
718 
719 static void nvme_keep_alive_work(struct work_struct *work)
720 {
721 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
722 			struct nvme_ctrl, ka_work);
723 
724 	if (nvme_keep_alive(ctrl)) {
725 		/* allocation failure, reset the controller */
726 		dev_err(ctrl->device, "keep-alive failed\n");
727 		nvme_reset_ctrl(ctrl);
728 		return;
729 	}
730 }
731 
732 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
733 {
734 	if (unlikely(ctrl->kato == 0))
735 		return;
736 
737 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
738 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
739 }
740 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
741 
742 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
743 {
744 	if (unlikely(ctrl->kato == 0))
745 		return;
746 
747 	cancel_delayed_work_sync(&ctrl->ka_work);
748 }
749 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
750 
751 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
752 {
753 	struct nvme_command c = { };
754 	int error;
755 
756 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
757 	c.identify.opcode = nvme_admin_identify;
758 	c.identify.cns = NVME_ID_CNS_CTRL;
759 
760 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
761 	if (!*id)
762 		return -ENOMEM;
763 
764 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
765 			sizeof(struct nvme_id_ctrl));
766 	if (error)
767 		kfree(*id);
768 	return error;
769 }
770 
771 static int nvme_identify_ns_descs(struct nvme_ns *ns, unsigned nsid)
772 {
773 	struct nvme_command c = { };
774 	int status;
775 	void *data;
776 	int pos;
777 	int len;
778 
779 	c.identify.opcode = nvme_admin_identify;
780 	c.identify.nsid = cpu_to_le32(nsid);
781 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
782 
783 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
784 	if (!data)
785 		return -ENOMEM;
786 
787 	status = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, data,
788 				      NVME_IDENTIFY_DATA_SIZE);
789 	if (status)
790 		goto free_data;
791 
792 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
793 		struct nvme_ns_id_desc *cur = data + pos;
794 
795 		if (cur->nidl == 0)
796 			break;
797 
798 		switch (cur->nidt) {
799 		case NVME_NIDT_EUI64:
800 			if (cur->nidl != NVME_NIDT_EUI64_LEN) {
801 				dev_warn(ns->ctrl->device,
802 					 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
803 					 cur->nidl);
804 				goto free_data;
805 			}
806 			len = NVME_NIDT_EUI64_LEN;
807 			memcpy(ns->eui, data + pos + sizeof(*cur), len);
808 			break;
809 		case NVME_NIDT_NGUID:
810 			if (cur->nidl != NVME_NIDT_NGUID_LEN) {
811 				dev_warn(ns->ctrl->device,
812 					 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
813 					 cur->nidl);
814 				goto free_data;
815 			}
816 			len = NVME_NIDT_NGUID_LEN;
817 			memcpy(ns->nguid, data + pos + sizeof(*cur), len);
818 			break;
819 		case NVME_NIDT_UUID:
820 			if (cur->nidl != NVME_NIDT_UUID_LEN) {
821 				dev_warn(ns->ctrl->device,
822 					 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
823 					 cur->nidl);
824 				goto free_data;
825 			}
826 			len = NVME_NIDT_UUID_LEN;
827 			uuid_copy(&ns->uuid, data + pos + sizeof(*cur));
828 			break;
829 		default:
830 			/* Skip unnkown types */
831 			len = cur->nidl;
832 			break;
833 		}
834 
835 		len += sizeof(*cur);
836 	}
837 free_data:
838 	kfree(data);
839 	return status;
840 }
841 
842 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
843 {
844 	struct nvme_command c = { };
845 
846 	c.identify.opcode = nvme_admin_identify;
847 	c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
848 	c.identify.nsid = cpu_to_le32(nsid);
849 	return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
850 }
851 
852 static int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
853 		struct nvme_id_ns **id)
854 {
855 	struct nvme_command c = { };
856 	int error;
857 
858 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
859 	c.identify.opcode = nvme_admin_identify;
860 	c.identify.nsid = cpu_to_le32(nsid);
861 	c.identify.cns = NVME_ID_CNS_NS;
862 
863 	*id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
864 	if (!*id)
865 		return -ENOMEM;
866 
867 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
868 			sizeof(struct nvme_id_ns));
869 	if (error)
870 		kfree(*id);
871 	return error;
872 }
873 
874 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
875 		      void *buffer, size_t buflen, u32 *result)
876 {
877 	struct nvme_command c;
878 	union nvme_result res;
879 	int ret;
880 
881 	memset(&c, 0, sizeof(c));
882 	c.features.opcode = nvme_admin_set_features;
883 	c.features.fid = cpu_to_le32(fid);
884 	c.features.dword11 = cpu_to_le32(dword11);
885 
886 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
887 			buffer, buflen, 0, NVME_QID_ANY, 0, 0);
888 	if (ret >= 0 && result)
889 		*result = le32_to_cpu(res.u32);
890 	return ret;
891 }
892 
893 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
894 {
895 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
896 	u32 result;
897 	int status, nr_io_queues;
898 
899 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
900 			&result);
901 	if (status < 0)
902 		return status;
903 
904 	/*
905 	 * Degraded controllers might return an error when setting the queue
906 	 * count.  We still want to be able to bring them online and offer
907 	 * access to the admin queue, as that might be only way to fix them up.
908 	 */
909 	if (status > 0) {
910 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
911 		*count = 0;
912 	} else {
913 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
914 		*count = min(*count, nr_io_queues);
915 	}
916 
917 	return 0;
918 }
919 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
920 
921 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
922 {
923 	struct nvme_user_io io;
924 	struct nvme_command c;
925 	unsigned length, meta_len;
926 	void __user *metadata;
927 
928 	if (copy_from_user(&io, uio, sizeof(io)))
929 		return -EFAULT;
930 	if (io.flags)
931 		return -EINVAL;
932 
933 	switch (io.opcode) {
934 	case nvme_cmd_write:
935 	case nvme_cmd_read:
936 	case nvme_cmd_compare:
937 		break;
938 	default:
939 		return -EINVAL;
940 	}
941 
942 	length = (io.nblocks + 1) << ns->lba_shift;
943 	meta_len = (io.nblocks + 1) * ns->ms;
944 	metadata = (void __user *)(uintptr_t)io.metadata;
945 
946 	if (ns->ext) {
947 		length += meta_len;
948 		meta_len = 0;
949 	} else if (meta_len) {
950 		if ((io.metadata & 3) || !io.metadata)
951 			return -EINVAL;
952 	}
953 
954 	memset(&c, 0, sizeof(c));
955 	c.rw.opcode = io.opcode;
956 	c.rw.flags = io.flags;
957 	c.rw.nsid = cpu_to_le32(ns->ns_id);
958 	c.rw.slba = cpu_to_le64(io.slba);
959 	c.rw.length = cpu_to_le16(io.nblocks);
960 	c.rw.control = cpu_to_le16(io.control);
961 	c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
962 	c.rw.reftag = cpu_to_le32(io.reftag);
963 	c.rw.apptag = cpu_to_le16(io.apptag);
964 	c.rw.appmask = cpu_to_le16(io.appmask);
965 
966 	return __nvme_submit_user_cmd(ns->queue, &c,
967 			(void __user *)(uintptr_t)io.addr, length,
968 			metadata, meta_len, io.slba, NULL, 0);
969 }
970 
971 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
972 			struct nvme_passthru_cmd __user *ucmd)
973 {
974 	struct nvme_passthru_cmd cmd;
975 	struct nvme_command c;
976 	unsigned timeout = 0;
977 	int status;
978 
979 	if (!capable(CAP_SYS_ADMIN))
980 		return -EACCES;
981 	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
982 		return -EFAULT;
983 	if (cmd.flags)
984 		return -EINVAL;
985 
986 	memset(&c, 0, sizeof(c));
987 	c.common.opcode = cmd.opcode;
988 	c.common.flags = cmd.flags;
989 	c.common.nsid = cpu_to_le32(cmd.nsid);
990 	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
991 	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
992 	c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
993 	c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
994 	c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
995 	c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
996 	c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
997 	c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
998 
999 	if (cmd.timeout_ms)
1000 		timeout = msecs_to_jiffies(cmd.timeout_ms);
1001 
1002 	status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1003 			(void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1004 			&cmd.result, timeout);
1005 	if (status >= 0) {
1006 		if (put_user(cmd.result, &ucmd->result))
1007 			return -EFAULT;
1008 	}
1009 
1010 	return status;
1011 }
1012 
1013 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1014 		unsigned int cmd, unsigned long arg)
1015 {
1016 	struct nvme_ns *ns = bdev->bd_disk->private_data;
1017 
1018 	switch (cmd) {
1019 	case NVME_IOCTL_ID:
1020 		force_successful_syscall_return();
1021 		return ns->ns_id;
1022 	case NVME_IOCTL_ADMIN_CMD:
1023 		return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1024 	case NVME_IOCTL_IO_CMD:
1025 		return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1026 	case NVME_IOCTL_SUBMIT_IO:
1027 		return nvme_submit_io(ns, (void __user *)arg);
1028 	default:
1029 #ifdef CONFIG_NVM
1030 		if (ns->ndev)
1031 			return nvme_nvm_ioctl(ns, cmd, arg);
1032 #endif
1033 		if (is_sed_ioctl(cmd))
1034 			return sed_ioctl(ns->ctrl->opal_dev, cmd,
1035 					 (void __user *) arg);
1036 		return -ENOTTY;
1037 	}
1038 }
1039 
1040 #ifdef CONFIG_COMPAT
1041 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1042 			unsigned int cmd, unsigned long arg)
1043 {
1044 	return nvme_ioctl(bdev, mode, cmd, arg);
1045 }
1046 #else
1047 #define nvme_compat_ioctl	NULL
1048 #endif
1049 
1050 static int nvme_open(struct block_device *bdev, fmode_t mode)
1051 {
1052 	return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
1053 }
1054 
1055 static void nvme_release(struct gendisk *disk, fmode_t mode)
1056 {
1057 	struct nvme_ns *ns = disk->private_data;
1058 
1059 	module_put(ns->ctrl->ops->module);
1060 	nvme_put_ns(ns);
1061 }
1062 
1063 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1064 {
1065 	/* some standard values */
1066 	geo->heads = 1 << 6;
1067 	geo->sectors = 1 << 5;
1068 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1069 	return 0;
1070 }
1071 
1072 #ifdef CONFIG_BLK_DEV_INTEGRITY
1073 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
1074 		u16 bs)
1075 {
1076 	struct nvme_ns *ns = disk->private_data;
1077 	u16 old_ms = ns->ms;
1078 	u8 pi_type = 0;
1079 
1080 	ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1081 	ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1082 
1083 	/* PI implementation requires metadata equal t10 pi tuple size */
1084 	if (ns->ms == sizeof(struct t10_pi_tuple))
1085 		pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1086 
1087 	if (blk_get_integrity(disk) &&
1088 	    (ns->pi_type != pi_type || ns->ms != old_ms ||
1089 	     bs != queue_logical_block_size(disk->queue) ||
1090 	     (ns->ms && ns->ext)))
1091 		blk_integrity_unregister(disk);
1092 
1093 	ns->pi_type = pi_type;
1094 }
1095 
1096 static void nvme_init_integrity(struct nvme_ns *ns)
1097 {
1098 	struct blk_integrity integrity;
1099 
1100 	memset(&integrity, 0, sizeof(integrity));
1101 	switch (ns->pi_type) {
1102 	case NVME_NS_DPS_PI_TYPE3:
1103 		integrity.profile = &t10_pi_type3_crc;
1104 		integrity.tag_size = sizeof(u16) + sizeof(u32);
1105 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1106 		break;
1107 	case NVME_NS_DPS_PI_TYPE1:
1108 	case NVME_NS_DPS_PI_TYPE2:
1109 		integrity.profile = &t10_pi_type1_crc;
1110 		integrity.tag_size = sizeof(u16);
1111 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1112 		break;
1113 	default:
1114 		integrity.profile = NULL;
1115 		break;
1116 	}
1117 	integrity.tuple_size = ns->ms;
1118 	blk_integrity_register(ns->disk, &integrity);
1119 	blk_queue_max_integrity_segments(ns->queue, 1);
1120 }
1121 #else
1122 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
1123 		u16 bs)
1124 {
1125 }
1126 static void nvme_init_integrity(struct nvme_ns *ns)
1127 {
1128 }
1129 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1130 
1131 static void nvme_set_chunk_size(struct nvme_ns *ns)
1132 {
1133 	u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1134 	blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1135 }
1136 
1137 static void nvme_config_discard(struct nvme_ns *ns)
1138 {
1139 	struct nvme_ctrl *ctrl = ns->ctrl;
1140 	u32 logical_block_size = queue_logical_block_size(ns->queue);
1141 
1142 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1143 			NVME_DSM_MAX_RANGES);
1144 
1145 	if (ctrl->nr_streams && ns->sws && ns->sgs) {
1146 		unsigned int sz = logical_block_size * ns->sws * ns->sgs;
1147 
1148 		ns->queue->limits.discard_alignment = sz;
1149 		ns->queue->limits.discard_granularity = sz;
1150 	} else {
1151 		ns->queue->limits.discard_alignment = logical_block_size;
1152 		ns->queue->limits.discard_granularity = logical_block_size;
1153 	}
1154 	blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
1155 	blk_queue_max_discard_segments(ns->queue, NVME_DSM_MAX_RANGES);
1156 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
1157 
1158 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1159 		blk_queue_max_write_zeroes_sectors(ns->queue, UINT_MAX);
1160 }
1161 
1162 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
1163 {
1164 	if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
1165 		dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
1166 		return -ENODEV;
1167 	}
1168 
1169 	if ((*id)->ncap == 0) {
1170 		kfree(*id);
1171 		return -ENODEV;
1172 	}
1173 
1174 	if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
1175 		memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
1176 	if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
1177 		memcpy(ns->nguid, (*id)->nguid, sizeof(ns->nguid));
1178 	if (ns->ctrl->vs >= NVME_VS(1, 3, 0)) {
1179 		 /* Don't treat error as fatal we potentially
1180 		  * already have a NGUID or EUI-64
1181 		  */
1182 		if (nvme_identify_ns_descs(ns, ns->ns_id))
1183 			dev_warn(ns->ctrl->device,
1184 				 "%s: Identify Descriptors failed\n", __func__);
1185 	}
1186 
1187 	return 0;
1188 }
1189 
1190 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1191 {
1192 	struct nvme_ns *ns = disk->private_data;
1193 	struct nvme_ctrl *ctrl = ns->ctrl;
1194 	u16 bs;
1195 
1196 	/*
1197 	 * If identify namespace failed, use default 512 byte block size so
1198 	 * block layer can use before failing read/write for 0 capacity.
1199 	 */
1200 	ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1201 	if (ns->lba_shift == 0)
1202 		ns->lba_shift = 9;
1203 	bs = 1 << ns->lba_shift;
1204 	ns->noiob = le16_to_cpu(id->noiob);
1205 
1206 	blk_mq_freeze_queue(disk->queue);
1207 
1208 	if (ctrl->ops->flags & NVME_F_METADATA_SUPPORTED)
1209 		nvme_prep_integrity(disk, id, bs);
1210 	blk_queue_logical_block_size(ns->queue, bs);
1211 	if (ns->noiob)
1212 		nvme_set_chunk_size(ns);
1213 	if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
1214 		nvme_init_integrity(ns);
1215 	if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
1216 		set_capacity(disk, 0);
1217 	else
1218 		set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
1219 
1220 	if (ctrl->oncs & NVME_CTRL_ONCS_DSM)
1221 		nvme_config_discard(ns);
1222 	blk_mq_unfreeze_queue(disk->queue);
1223 }
1224 
1225 static int nvme_revalidate_disk(struct gendisk *disk)
1226 {
1227 	struct nvme_ns *ns = disk->private_data;
1228 	struct nvme_id_ns *id = NULL;
1229 	int ret;
1230 
1231 	if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1232 		set_capacity(disk, 0);
1233 		return -ENODEV;
1234 	}
1235 
1236 	ret = nvme_revalidate_ns(ns, &id);
1237 	if (ret)
1238 		return ret;
1239 
1240 	__nvme_revalidate_disk(disk, id);
1241 	kfree(id);
1242 
1243 	return 0;
1244 }
1245 
1246 static char nvme_pr_type(enum pr_type type)
1247 {
1248 	switch (type) {
1249 	case PR_WRITE_EXCLUSIVE:
1250 		return 1;
1251 	case PR_EXCLUSIVE_ACCESS:
1252 		return 2;
1253 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1254 		return 3;
1255 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1256 		return 4;
1257 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1258 		return 5;
1259 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1260 		return 6;
1261 	default:
1262 		return 0;
1263 	}
1264 };
1265 
1266 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1267 				u64 key, u64 sa_key, u8 op)
1268 {
1269 	struct nvme_ns *ns = bdev->bd_disk->private_data;
1270 	struct nvme_command c;
1271 	u8 data[16] = { 0, };
1272 
1273 	put_unaligned_le64(key, &data[0]);
1274 	put_unaligned_le64(sa_key, &data[8]);
1275 
1276 	memset(&c, 0, sizeof(c));
1277 	c.common.opcode = op;
1278 	c.common.nsid = cpu_to_le32(ns->ns_id);
1279 	c.common.cdw10[0] = cpu_to_le32(cdw10);
1280 
1281 	return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1282 }
1283 
1284 static int nvme_pr_register(struct block_device *bdev, u64 old,
1285 		u64 new, unsigned flags)
1286 {
1287 	u32 cdw10;
1288 
1289 	if (flags & ~PR_FL_IGNORE_KEY)
1290 		return -EOPNOTSUPP;
1291 
1292 	cdw10 = old ? 2 : 0;
1293 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1294 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1295 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1296 }
1297 
1298 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1299 		enum pr_type type, unsigned flags)
1300 {
1301 	u32 cdw10;
1302 
1303 	if (flags & ~PR_FL_IGNORE_KEY)
1304 		return -EOPNOTSUPP;
1305 
1306 	cdw10 = nvme_pr_type(type) << 8;
1307 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1308 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1309 }
1310 
1311 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1312 		enum pr_type type, bool abort)
1313 {
1314 	u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1315 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1316 }
1317 
1318 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1319 {
1320 	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1321 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1322 }
1323 
1324 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1325 {
1326 	u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1327 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1328 }
1329 
1330 static const struct pr_ops nvme_pr_ops = {
1331 	.pr_register	= nvme_pr_register,
1332 	.pr_reserve	= nvme_pr_reserve,
1333 	.pr_release	= nvme_pr_release,
1334 	.pr_preempt	= nvme_pr_preempt,
1335 	.pr_clear	= nvme_pr_clear,
1336 };
1337 
1338 #ifdef CONFIG_BLK_SED_OPAL
1339 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1340 		bool send)
1341 {
1342 	struct nvme_ctrl *ctrl = data;
1343 	struct nvme_command cmd;
1344 
1345 	memset(&cmd, 0, sizeof(cmd));
1346 	if (send)
1347 		cmd.common.opcode = nvme_admin_security_send;
1348 	else
1349 		cmd.common.opcode = nvme_admin_security_recv;
1350 	cmd.common.nsid = 0;
1351 	cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1352 	cmd.common.cdw10[1] = cpu_to_le32(len);
1353 
1354 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1355 				      ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1356 }
1357 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1358 #endif /* CONFIG_BLK_SED_OPAL */
1359 
1360 static const struct block_device_operations nvme_fops = {
1361 	.owner		= THIS_MODULE,
1362 	.ioctl		= nvme_ioctl,
1363 	.compat_ioctl	= nvme_compat_ioctl,
1364 	.open		= nvme_open,
1365 	.release	= nvme_release,
1366 	.getgeo		= nvme_getgeo,
1367 	.revalidate_disk= nvme_revalidate_disk,
1368 	.pr_ops		= &nvme_pr_ops,
1369 };
1370 
1371 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1372 {
1373 	unsigned long timeout =
1374 		((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1375 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1376 	int ret;
1377 
1378 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1379 		if (csts == ~0)
1380 			return -ENODEV;
1381 		if ((csts & NVME_CSTS_RDY) == bit)
1382 			break;
1383 
1384 		msleep(100);
1385 		if (fatal_signal_pending(current))
1386 			return -EINTR;
1387 		if (time_after(jiffies, timeout)) {
1388 			dev_err(ctrl->device,
1389 				"Device not ready; aborting %s\n", enabled ?
1390 						"initialisation" : "reset");
1391 			return -ENODEV;
1392 		}
1393 	}
1394 
1395 	return ret;
1396 }
1397 
1398 /*
1399  * If the device has been passed off to us in an enabled state, just clear
1400  * the enabled bit.  The spec says we should set the 'shutdown notification
1401  * bits', but doing so may cause the device to complete commands to the
1402  * admin queue ... and we don't know what memory that might be pointing at!
1403  */
1404 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1405 {
1406 	int ret;
1407 
1408 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1409 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1410 
1411 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1412 	if (ret)
1413 		return ret;
1414 
1415 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1416 		msleep(NVME_QUIRK_DELAY_AMOUNT);
1417 
1418 	return nvme_wait_ready(ctrl, cap, false);
1419 }
1420 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1421 
1422 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1423 {
1424 	/*
1425 	 * Default to a 4K page size, with the intention to update this
1426 	 * path in the future to accomodate architectures with differing
1427 	 * kernel and IO page sizes.
1428 	 */
1429 	unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1430 	int ret;
1431 
1432 	if (page_shift < dev_page_min) {
1433 		dev_err(ctrl->device,
1434 			"Minimum device page size %u too large for host (%u)\n",
1435 			1 << dev_page_min, 1 << page_shift);
1436 		return -ENODEV;
1437 	}
1438 
1439 	ctrl->page_size = 1 << page_shift;
1440 
1441 	ctrl->ctrl_config = NVME_CC_CSS_NVM;
1442 	ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1443 	ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1444 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1445 	ctrl->ctrl_config |= NVME_CC_ENABLE;
1446 
1447 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1448 	if (ret)
1449 		return ret;
1450 	return nvme_wait_ready(ctrl, cap, true);
1451 }
1452 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1453 
1454 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1455 {
1456 	unsigned long timeout = jiffies + (shutdown_timeout * HZ);
1457 	u32 csts;
1458 	int ret;
1459 
1460 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1461 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1462 
1463 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1464 	if (ret)
1465 		return ret;
1466 
1467 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1468 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1469 			break;
1470 
1471 		msleep(100);
1472 		if (fatal_signal_pending(current))
1473 			return -EINTR;
1474 		if (time_after(jiffies, timeout)) {
1475 			dev_err(ctrl->device,
1476 				"Device shutdown incomplete; abort shutdown\n");
1477 			return -ENODEV;
1478 		}
1479 	}
1480 
1481 	return ret;
1482 }
1483 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1484 
1485 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1486 		struct request_queue *q)
1487 {
1488 	bool vwc = false;
1489 
1490 	if (ctrl->max_hw_sectors) {
1491 		u32 max_segments =
1492 			(ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1493 
1494 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1495 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1496 	}
1497 	if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1498 		blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1499 	blk_queue_virt_boundary(q, ctrl->page_size - 1);
1500 	if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1501 		vwc = true;
1502 	blk_queue_write_cache(q, vwc, vwc);
1503 }
1504 
1505 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1506 {
1507 	/*
1508 	 * APST (Autonomous Power State Transition) lets us program a
1509 	 * table of power state transitions that the controller will
1510 	 * perform automatically.  We configure it with a simple
1511 	 * heuristic: we are willing to spend at most 2% of the time
1512 	 * transitioning between power states.  Therefore, when running
1513 	 * in any given state, we will enter the next lower-power
1514 	 * non-operational state after waiting 50 * (enlat + exlat)
1515 	 * microseconds, as long as that state's exit latency is under
1516 	 * the requested maximum latency.
1517 	 *
1518 	 * We will not autonomously enter any non-operational state for
1519 	 * which the total latency exceeds ps_max_latency_us.  Users
1520 	 * can set ps_max_latency_us to zero to turn off APST.
1521 	 */
1522 
1523 	unsigned apste;
1524 	struct nvme_feat_auto_pst *table;
1525 	u64 max_lat_us = 0;
1526 	int max_ps = -1;
1527 	int ret;
1528 
1529 	/*
1530 	 * If APST isn't supported or if we haven't been initialized yet,
1531 	 * then don't do anything.
1532 	 */
1533 	if (!ctrl->apsta)
1534 		return 0;
1535 
1536 	if (ctrl->npss > 31) {
1537 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1538 		return 0;
1539 	}
1540 
1541 	table = kzalloc(sizeof(*table), GFP_KERNEL);
1542 	if (!table)
1543 		return 0;
1544 
1545 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1546 		/* Turn off APST. */
1547 		apste = 0;
1548 		dev_dbg(ctrl->device, "APST disabled\n");
1549 	} else {
1550 		__le64 target = cpu_to_le64(0);
1551 		int state;
1552 
1553 		/*
1554 		 * Walk through all states from lowest- to highest-power.
1555 		 * According to the spec, lower-numbered states use more
1556 		 * power.  NPSS, despite the name, is the index of the
1557 		 * lowest-power state, not the number of states.
1558 		 */
1559 		for (state = (int)ctrl->npss; state >= 0; state--) {
1560 			u64 total_latency_us, exit_latency_us, transition_ms;
1561 
1562 			if (target)
1563 				table->entries[state] = target;
1564 
1565 			/*
1566 			 * Don't allow transitions to the deepest state
1567 			 * if it's quirked off.
1568 			 */
1569 			if (state == ctrl->npss &&
1570 			    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1571 				continue;
1572 
1573 			/*
1574 			 * Is this state a useful non-operational state for
1575 			 * higher-power states to autonomously transition to?
1576 			 */
1577 			if (!(ctrl->psd[state].flags &
1578 			      NVME_PS_FLAGS_NON_OP_STATE))
1579 				continue;
1580 
1581 			exit_latency_us =
1582 				(u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1583 			if (exit_latency_us > ctrl->ps_max_latency_us)
1584 				continue;
1585 
1586 			total_latency_us =
1587 				exit_latency_us +
1588 				le32_to_cpu(ctrl->psd[state].entry_lat);
1589 
1590 			/*
1591 			 * This state is good.  Use it as the APST idle
1592 			 * target for higher power states.
1593 			 */
1594 			transition_ms = total_latency_us + 19;
1595 			do_div(transition_ms, 20);
1596 			if (transition_ms > (1 << 24) - 1)
1597 				transition_ms = (1 << 24) - 1;
1598 
1599 			target = cpu_to_le64((state << 3) |
1600 					     (transition_ms << 8));
1601 
1602 			if (max_ps == -1)
1603 				max_ps = state;
1604 
1605 			if (total_latency_us > max_lat_us)
1606 				max_lat_us = total_latency_us;
1607 		}
1608 
1609 		apste = 1;
1610 
1611 		if (max_ps == -1) {
1612 			dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1613 		} else {
1614 			dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1615 				max_ps, max_lat_us, (int)sizeof(*table), table);
1616 		}
1617 	}
1618 
1619 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1620 				table, sizeof(*table), NULL);
1621 	if (ret)
1622 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1623 
1624 	kfree(table);
1625 	return ret;
1626 }
1627 
1628 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1629 {
1630 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1631 	u64 latency;
1632 
1633 	switch (val) {
1634 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1635 	case PM_QOS_LATENCY_ANY:
1636 		latency = U64_MAX;
1637 		break;
1638 
1639 	default:
1640 		latency = val;
1641 	}
1642 
1643 	if (ctrl->ps_max_latency_us != latency) {
1644 		ctrl->ps_max_latency_us = latency;
1645 		nvme_configure_apst(ctrl);
1646 	}
1647 }
1648 
1649 struct nvme_core_quirk_entry {
1650 	/*
1651 	 * NVMe model and firmware strings are padded with spaces.  For
1652 	 * simplicity, strings in the quirk table are padded with NULLs
1653 	 * instead.
1654 	 */
1655 	u16 vid;
1656 	const char *mn;
1657 	const char *fr;
1658 	unsigned long quirks;
1659 };
1660 
1661 static const struct nvme_core_quirk_entry core_quirks[] = {
1662 	{
1663 		/*
1664 		 * This Toshiba device seems to die using any APST states.  See:
1665 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1666 		 */
1667 		.vid = 0x1179,
1668 		.mn = "THNSF5256GPUK TOSHIBA",
1669 		.quirks = NVME_QUIRK_NO_APST,
1670 	}
1671 };
1672 
1673 /* match is null-terminated but idstr is space-padded. */
1674 static bool string_matches(const char *idstr, const char *match, size_t len)
1675 {
1676 	size_t matchlen;
1677 
1678 	if (!match)
1679 		return true;
1680 
1681 	matchlen = strlen(match);
1682 	WARN_ON_ONCE(matchlen > len);
1683 
1684 	if (memcmp(idstr, match, matchlen))
1685 		return false;
1686 
1687 	for (; matchlen < len; matchlen++)
1688 		if (idstr[matchlen] != ' ')
1689 			return false;
1690 
1691 	return true;
1692 }
1693 
1694 static bool quirk_matches(const struct nvme_id_ctrl *id,
1695 			  const struct nvme_core_quirk_entry *q)
1696 {
1697 	return q->vid == le16_to_cpu(id->vid) &&
1698 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1699 		string_matches(id->fr, q->fr, sizeof(id->fr));
1700 }
1701 
1702 static void nvme_init_subnqn(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
1703 {
1704 	size_t nqnlen;
1705 	int off;
1706 
1707 	nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
1708 	if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
1709 		strcpy(ctrl->subnqn, id->subnqn);
1710 		return;
1711 	}
1712 
1713 	if (ctrl->vs >= NVME_VS(1, 2, 1))
1714 		dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
1715 
1716 	/* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
1717 	off = snprintf(ctrl->subnqn, NVMF_NQN_SIZE,
1718 			"nqn.2014.08.org.nvmexpress:%4x%4x",
1719 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
1720 	memcpy(ctrl->subnqn + off, id->sn, sizeof(id->sn));
1721 	off += sizeof(id->sn);
1722 	memcpy(ctrl->subnqn + off, id->mn, sizeof(id->mn));
1723 	off += sizeof(id->mn);
1724 	memset(ctrl->subnqn + off, 0, sizeof(ctrl->subnqn) - off);
1725 }
1726 
1727 /*
1728  * Initialize the cached copies of the Identify data and various controller
1729  * register in our nvme_ctrl structure.  This should be called as soon as
1730  * the admin queue is fully up and running.
1731  */
1732 int nvme_init_identify(struct nvme_ctrl *ctrl)
1733 {
1734 	struct nvme_id_ctrl *id;
1735 	u64 cap;
1736 	int ret, page_shift;
1737 	u32 max_hw_sectors;
1738 	bool prev_apst_enabled;
1739 
1740 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1741 	if (ret) {
1742 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1743 		return ret;
1744 	}
1745 
1746 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1747 	if (ret) {
1748 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1749 		return ret;
1750 	}
1751 	page_shift = NVME_CAP_MPSMIN(cap) + 12;
1752 
1753 	if (ctrl->vs >= NVME_VS(1, 1, 0))
1754 		ctrl->subsystem = NVME_CAP_NSSRC(cap);
1755 
1756 	ret = nvme_identify_ctrl(ctrl, &id);
1757 	if (ret) {
1758 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1759 		return -EIO;
1760 	}
1761 
1762 	nvme_init_subnqn(ctrl, id);
1763 
1764 	if (!ctrl->identified) {
1765 		/*
1766 		 * Check for quirks.  Quirk can depend on firmware version,
1767 		 * so, in principle, the set of quirks present can change
1768 		 * across a reset.  As a possible future enhancement, we
1769 		 * could re-scan for quirks every time we reinitialize
1770 		 * the device, but we'd have to make sure that the driver
1771 		 * behaves intelligently if the quirks change.
1772 		 */
1773 
1774 		int i;
1775 
1776 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
1777 			if (quirk_matches(id, &core_quirks[i]))
1778 				ctrl->quirks |= core_quirks[i].quirks;
1779 		}
1780 	}
1781 
1782 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
1783 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
1784 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
1785 	}
1786 
1787 	ctrl->oacs = le16_to_cpu(id->oacs);
1788 	ctrl->vid = le16_to_cpu(id->vid);
1789 	ctrl->oncs = le16_to_cpup(&id->oncs);
1790 	atomic_set(&ctrl->abort_limit, id->acl + 1);
1791 	ctrl->vwc = id->vwc;
1792 	ctrl->cntlid = le16_to_cpup(&id->cntlid);
1793 	memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1794 	memcpy(ctrl->model, id->mn, sizeof(id->mn));
1795 	memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1796 	if (id->mdts)
1797 		max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1798 	else
1799 		max_hw_sectors = UINT_MAX;
1800 	ctrl->max_hw_sectors =
1801 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1802 
1803 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
1804 	ctrl->sgls = le32_to_cpu(id->sgls);
1805 	ctrl->kas = le16_to_cpu(id->kas);
1806 
1807 	ctrl->npss = id->npss;
1808 	ctrl->apsta = id->apsta;
1809 	prev_apst_enabled = ctrl->apst_enabled;
1810 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
1811 		if (force_apst && id->apsta) {
1812 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
1813 			ctrl->apst_enabled = true;
1814 		} else {
1815 			ctrl->apst_enabled = false;
1816 		}
1817 	} else {
1818 		ctrl->apst_enabled = id->apsta;
1819 	}
1820 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
1821 
1822 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1823 		ctrl->icdoff = le16_to_cpu(id->icdoff);
1824 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1825 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1826 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1827 
1828 		/*
1829 		 * In fabrics we need to verify the cntlid matches the
1830 		 * admin connect
1831 		 */
1832 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
1833 			ret = -EINVAL;
1834 			goto out_free;
1835 		}
1836 
1837 		if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1838 			dev_err(ctrl->device,
1839 				"keep-alive support is mandatory for fabrics\n");
1840 			ret = -EINVAL;
1841 			goto out_free;
1842 		}
1843 	} else {
1844 		ctrl->cntlid = le16_to_cpu(id->cntlid);
1845 		ctrl->hmpre = le32_to_cpu(id->hmpre);
1846 		ctrl->hmmin = le32_to_cpu(id->hmmin);
1847 	}
1848 
1849 	kfree(id);
1850 
1851 	if (ctrl->apst_enabled && !prev_apst_enabled)
1852 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
1853 	else if (!ctrl->apst_enabled && prev_apst_enabled)
1854 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
1855 
1856 	ret = nvme_configure_apst(ctrl);
1857 	if (ret < 0)
1858 		return ret;
1859 
1860 	ret = nvme_configure_directives(ctrl);
1861 	if (ret < 0)
1862 		return ret;
1863 
1864 	ctrl->identified = true;
1865 
1866 	return 0;
1867 
1868 out_free:
1869 	kfree(id);
1870 	return ret;
1871 }
1872 EXPORT_SYMBOL_GPL(nvme_init_identify);
1873 
1874 static int nvme_dev_open(struct inode *inode, struct file *file)
1875 {
1876 	struct nvme_ctrl *ctrl;
1877 	int instance = iminor(inode);
1878 	int ret = -ENODEV;
1879 
1880 	spin_lock(&dev_list_lock);
1881 	list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1882 		if (ctrl->instance != instance)
1883 			continue;
1884 
1885 		if (!ctrl->admin_q) {
1886 			ret = -EWOULDBLOCK;
1887 			break;
1888 		}
1889 		if (!kref_get_unless_zero(&ctrl->kref))
1890 			break;
1891 		file->private_data = ctrl;
1892 		ret = 0;
1893 		break;
1894 	}
1895 	spin_unlock(&dev_list_lock);
1896 
1897 	return ret;
1898 }
1899 
1900 static int nvme_dev_release(struct inode *inode, struct file *file)
1901 {
1902 	nvme_put_ctrl(file->private_data);
1903 	return 0;
1904 }
1905 
1906 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1907 {
1908 	struct nvme_ns *ns;
1909 	int ret;
1910 
1911 	mutex_lock(&ctrl->namespaces_mutex);
1912 	if (list_empty(&ctrl->namespaces)) {
1913 		ret = -ENOTTY;
1914 		goto out_unlock;
1915 	}
1916 
1917 	ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1918 	if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1919 		dev_warn(ctrl->device,
1920 			"NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1921 		ret = -EINVAL;
1922 		goto out_unlock;
1923 	}
1924 
1925 	dev_warn(ctrl->device,
1926 		"using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1927 	kref_get(&ns->kref);
1928 	mutex_unlock(&ctrl->namespaces_mutex);
1929 
1930 	ret = nvme_user_cmd(ctrl, ns, argp);
1931 	nvme_put_ns(ns);
1932 	return ret;
1933 
1934 out_unlock:
1935 	mutex_unlock(&ctrl->namespaces_mutex);
1936 	return ret;
1937 }
1938 
1939 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1940 		unsigned long arg)
1941 {
1942 	struct nvme_ctrl *ctrl = file->private_data;
1943 	void __user *argp = (void __user *)arg;
1944 
1945 	switch (cmd) {
1946 	case NVME_IOCTL_ADMIN_CMD:
1947 		return nvme_user_cmd(ctrl, NULL, argp);
1948 	case NVME_IOCTL_IO_CMD:
1949 		return nvme_dev_user_cmd(ctrl, argp);
1950 	case NVME_IOCTL_RESET:
1951 		dev_warn(ctrl->device, "resetting controller\n");
1952 		return nvme_reset_ctrl_sync(ctrl);
1953 	case NVME_IOCTL_SUBSYS_RESET:
1954 		return nvme_reset_subsystem(ctrl);
1955 	case NVME_IOCTL_RESCAN:
1956 		nvme_queue_scan(ctrl);
1957 		return 0;
1958 	default:
1959 		return -ENOTTY;
1960 	}
1961 }
1962 
1963 static const struct file_operations nvme_dev_fops = {
1964 	.owner		= THIS_MODULE,
1965 	.open		= nvme_dev_open,
1966 	.release	= nvme_dev_release,
1967 	.unlocked_ioctl	= nvme_dev_ioctl,
1968 	.compat_ioctl	= nvme_dev_ioctl,
1969 };
1970 
1971 static ssize_t nvme_sysfs_reset(struct device *dev,
1972 				struct device_attribute *attr, const char *buf,
1973 				size_t count)
1974 {
1975 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1976 	int ret;
1977 
1978 	ret = nvme_reset_ctrl_sync(ctrl);
1979 	if (ret < 0)
1980 		return ret;
1981 	return count;
1982 }
1983 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1984 
1985 static ssize_t nvme_sysfs_rescan(struct device *dev,
1986 				struct device_attribute *attr, const char *buf,
1987 				size_t count)
1988 {
1989 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1990 
1991 	nvme_queue_scan(ctrl);
1992 	return count;
1993 }
1994 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1995 
1996 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1997 								char *buf)
1998 {
1999 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2000 	struct nvme_ctrl *ctrl = ns->ctrl;
2001 	int serial_len = sizeof(ctrl->serial);
2002 	int model_len = sizeof(ctrl->model);
2003 
2004 	if (!uuid_is_null(&ns->uuid))
2005 		return sprintf(buf, "uuid.%pU\n", &ns->uuid);
2006 
2007 	if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
2008 		return sprintf(buf, "eui.%16phN\n", ns->nguid);
2009 
2010 	if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
2011 		return sprintf(buf, "eui.%8phN\n", ns->eui);
2012 
2013 	while (serial_len > 0 && (ctrl->serial[serial_len - 1] == ' ' ||
2014 				  ctrl->serial[serial_len - 1] == '\0'))
2015 		serial_len--;
2016 	while (model_len > 0 && (ctrl->model[model_len - 1] == ' ' ||
2017 				 ctrl->model[model_len - 1] == '\0'))
2018 		model_len--;
2019 
2020 	return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
2021 		serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
2022 }
2023 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
2024 
2025 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2026 			  char *buf)
2027 {
2028 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2029 	return sprintf(buf, "%pU\n", ns->nguid);
2030 }
2031 static DEVICE_ATTR(nguid, S_IRUGO, nguid_show, NULL);
2032 
2033 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2034 								char *buf)
2035 {
2036 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2037 
2038 	/* For backward compatibility expose the NGUID to userspace if
2039 	 * we have no UUID set
2040 	 */
2041 	if (uuid_is_null(&ns->uuid)) {
2042 		printk_ratelimited(KERN_WARNING
2043 				   "No UUID available providing old NGUID\n");
2044 		return sprintf(buf, "%pU\n", ns->nguid);
2045 	}
2046 	return sprintf(buf, "%pU\n", &ns->uuid);
2047 }
2048 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
2049 
2050 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2051 								char *buf)
2052 {
2053 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2054 	return sprintf(buf, "%8phd\n", ns->eui);
2055 }
2056 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
2057 
2058 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2059 								char *buf)
2060 {
2061 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2062 	return sprintf(buf, "%d\n", ns->ns_id);
2063 }
2064 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
2065 
2066 static struct attribute *nvme_ns_attrs[] = {
2067 	&dev_attr_wwid.attr,
2068 	&dev_attr_uuid.attr,
2069 	&dev_attr_nguid.attr,
2070 	&dev_attr_eui.attr,
2071 	&dev_attr_nsid.attr,
2072 	NULL,
2073 };
2074 
2075 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
2076 		struct attribute *a, int n)
2077 {
2078 	struct device *dev = container_of(kobj, struct device, kobj);
2079 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2080 
2081 	if (a == &dev_attr_uuid.attr) {
2082 		if (uuid_is_null(&ns->uuid) ||
2083 		    !memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
2084 			return 0;
2085 	}
2086 	if (a == &dev_attr_nguid.attr) {
2087 		if (!memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
2088 			return 0;
2089 	}
2090 	if (a == &dev_attr_eui.attr) {
2091 		if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
2092 			return 0;
2093 	}
2094 	return a->mode;
2095 }
2096 
2097 static const struct attribute_group nvme_ns_attr_group = {
2098 	.attrs		= nvme_ns_attrs,
2099 	.is_visible	= nvme_ns_attrs_are_visible,
2100 };
2101 
2102 #define nvme_show_str_function(field)						\
2103 static ssize_t  field##_show(struct device *dev,				\
2104 			    struct device_attribute *attr, char *buf)		\
2105 {										\
2106         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
2107         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);	\
2108 }										\
2109 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2110 
2111 #define nvme_show_int_function(field)						\
2112 static ssize_t  field##_show(struct device *dev,				\
2113 			    struct device_attribute *attr, char *buf)		\
2114 {										\
2115         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
2116         return sprintf(buf, "%d\n", ctrl->field);	\
2117 }										\
2118 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2119 
2120 nvme_show_str_function(model);
2121 nvme_show_str_function(serial);
2122 nvme_show_str_function(firmware_rev);
2123 nvme_show_int_function(cntlid);
2124 
2125 static ssize_t nvme_sysfs_delete(struct device *dev,
2126 				struct device_attribute *attr, const char *buf,
2127 				size_t count)
2128 {
2129 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2130 
2131 	if (device_remove_file_self(dev, attr))
2132 		ctrl->ops->delete_ctrl(ctrl);
2133 	return count;
2134 }
2135 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2136 
2137 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2138 					 struct device_attribute *attr,
2139 					 char *buf)
2140 {
2141 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2142 
2143 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2144 }
2145 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2146 
2147 static ssize_t nvme_sysfs_show_state(struct device *dev,
2148 				     struct device_attribute *attr,
2149 				     char *buf)
2150 {
2151 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2152 	static const char *const state_name[] = {
2153 		[NVME_CTRL_NEW]		= "new",
2154 		[NVME_CTRL_LIVE]	= "live",
2155 		[NVME_CTRL_RESETTING]	= "resetting",
2156 		[NVME_CTRL_RECONNECTING]= "reconnecting",
2157 		[NVME_CTRL_DELETING]	= "deleting",
2158 		[NVME_CTRL_DEAD]	= "dead",
2159 	};
2160 
2161 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2162 	    state_name[ctrl->state])
2163 		return sprintf(buf, "%s\n", state_name[ctrl->state]);
2164 
2165 	return sprintf(buf, "unknown state\n");
2166 }
2167 
2168 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2169 
2170 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2171 					 struct device_attribute *attr,
2172 					 char *buf)
2173 {
2174 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2175 
2176 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subnqn);
2177 }
2178 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2179 
2180 static ssize_t nvme_sysfs_show_address(struct device *dev,
2181 					 struct device_attribute *attr,
2182 					 char *buf)
2183 {
2184 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2185 
2186 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2187 }
2188 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2189 
2190 static struct attribute *nvme_dev_attrs[] = {
2191 	&dev_attr_reset_controller.attr,
2192 	&dev_attr_rescan_controller.attr,
2193 	&dev_attr_model.attr,
2194 	&dev_attr_serial.attr,
2195 	&dev_attr_firmware_rev.attr,
2196 	&dev_attr_cntlid.attr,
2197 	&dev_attr_delete_controller.attr,
2198 	&dev_attr_transport.attr,
2199 	&dev_attr_subsysnqn.attr,
2200 	&dev_attr_address.attr,
2201 	&dev_attr_state.attr,
2202 	NULL
2203 };
2204 
2205 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2206 		struct attribute *a, int n)
2207 {
2208 	struct device *dev = container_of(kobj, struct device, kobj);
2209 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2210 
2211 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2212 		return 0;
2213 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2214 		return 0;
2215 
2216 	return a->mode;
2217 }
2218 
2219 static struct attribute_group nvme_dev_attrs_group = {
2220 	.attrs		= nvme_dev_attrs,
2221 	.is_visible	= nvme_dev_attrs_are_visible,
2222 };
2223 
2224 static const struct attribute_group *nvme_dev_attr_groups[] = {
2225 	&nvme_dev_attrs_group,
2226 	NULL,
2227 };
2228 
2229 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2230 {
2231 	struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2232 	struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2233 
2234 	return nsa->ns_id - nsb->ns_id;
2235 }
2236 
2237 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2238 {
2239 	struct nvme_ns *ns, *ret = NULL;
2240 
2241 	mutex_lock(&ctrl->namespaces_mutex);
2242 	list_for_each_entry(ns, &ctrl->namespaces, list) {
2243 		if (ns->ns_id == nsid) {
2244 			kref_get(&ns->kref);
2245 			ret = ns;
2246 			break;
2247 		}
2248 		if (ns->ns_id > nsid)
2249 			break;
2250 	}
2251 	mutex_unlock(&ctrl->namespaces_mutex);
2252 	return ret;
2253 }
2254 
2255 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
2256 {
2257 	struct streams_directive_params s;
2258 	int ret;
2259 
2260 	if (!ctrl->nr_streams)
2261 		return 0;
2262 
2263 	ret = nvme_get_stream_params(ctrl, &s, ns->ns_id);
2264 	if (ret)
2265 		return ret;
2266 
2267 	ns->sws = le32_to_cpu(s.sws);
2268 	ns->sgs = le16_to_cpu(s.sgs);
2269 
2270 	if (ns->sws) {
2271 		unsigned int bs = 1 << ns->lba_shift;
2272 
2273 		blk_queue_io_min(ns->queue, bs * ns->sws);
2274 		if (ns->sgs)
2275 			blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
2276 	}
2277 
2278 	return 0;
2279 }
2280 
2281 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2282 {
2283 	struct nvme_ns *ns;
2284 	struct gendisk *disk;
2285 	struct nvme_id_ns *id;
2286 	char disk_name[DISK_NAME_LEN];
2287 	int node = dev_to_node(ctrl->dev);
2288 
2289 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2290 	if (!ns)
2291 		return;
2292 
2293 	ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
2294 	if (ns->instance < 0)
2295 		goto out_free_ns;
2296 
2297 	ns->queue = blk_mq_init_queue(ctrl->tagset);
2298 	if (IS_ERR(ns->queue))
2299 		goto out_release_instance;
2300 	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2301 	ns->queue->queuedata = ns;
2302 	ns->ctrl = ctrl;
2303 
2304 	kref_init(&ns->kref);
2305 	ns->ns_id = nsid;
2306 	ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2307 
2308 	blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2309 	nvme_set_queue_limits(ctrl, ns->queue);
2310 	nvme_setup_streams_ns(ctrl, ns);
2311 
2312 	sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
2313 
2314 	if (nvme_revalidate_ns(ns, &id))
2315 		goto out_free_queue;
2316 
2317 	if (nvme_nvm_ns_supported(ns, id) &&
2318 				nvme_nvm_register(ns, disk_name, node)) {
2319 		dev_warn(ctrl->device, "%s: LightNVM init failure\n", __func__);
2320 		goto out_free_id;
2321 	}
2322 
2323 	disk = alloc_disk_node(0, node);
2324 	if (!disk)
2325 		goto out_free_id;
2326 
2327 	disk->fops = &nvme_fops;
2328 	disk->private_data = ns;
2329 	disk->queue = ns->queue;
2330 	disk->flags = GENHD_FL_EXT_DEVT;
2331 	memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
2332 	ns->disk = disk;
2333 
2334 	__nvme_revalidate_disk(disk, id);
2335 
2336 	mutex_lock(&ctrl->namespaces_mutex);
2337 	list_add_tail(&ns->list, &ctrl->namespaces);
2338 	mutex_unlock(&ctrl->namespaces_mutex);
2339 
2340 	kref_get(&ctrl->kref);
2341 
2342 	kfree(id);
2343 
2344 	device_add_disk(ctrl->device, ns->disk);
2345 	if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
2346 					&nvme_ns_attr_group))
2347 		pr_warn("%s: failed to create sysfs group for identification\n",
2348 			ns->disk->disk_name);
2349 	if (ns->ndev && nvme_nvm_register_sysfs(ns))
2350 		pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
2351 			ns->disk->disk_name);
2352 	return;
2353  out_free_id:
2354 	kfree(id);
2355  out_free_queue:
2356 	blk_cleanup_queue(ns->queue);
2357  out_release_instance:
2358 	ida_simple_remove(&ctrl->ns_ida, ns->instance);
2359  out_free_ns:
2360 	kfree(ns);
2361 }
2362 
2363 static void nvme_ns_remove(struct nvme_ns *ns)
2364 {
2365 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
2366 		return;
2367 
2368 	if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
2369 		if (blk_get_integrity(ns->disk))
2370 			blk_integrity_unregister(ns->disk);
2371 		sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
2372 					&nvme_ns_attr_group);
2373 		if (ns->ndev)
2374 			nvme_nvm_unregister_sysfs(ns);
2375 		del_gendisk(ns->disk);
2376 		blk_cleanup_queue(ns->queue);
2377 	}
2378 
2379 	mutex_lock(&ns->ctrl->namespaces_mutex);
2380 	list_del_init(&ns->list);
2381 	mutex_unlock(&ns->ctrl->namespaces_mutex);
2382 
2383 	nvme_put_ns(ns);
2384 }
2385 
2386 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2387 {
2388 	struct nvme_ns *ns;
2389 
2390 	ns = nvme_find_get_ns(ctrl, nsid);
2391 	if (ns) {
2392 		if (ns->disk && revalidate_disk(ns->disk))
2393 			nvme_ns_remove(ns);
2394 		nvme_put_ns(ns);
2395 	} else
2396 		nvme_alloc_ns(ctrl, nsid);
2397 }
2398 
2399 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
2400 					unsigned nsid)
2401 {
2402 	struct nvme_ns *ns, *next;
2403 
2404 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
2405 		if (ns->ns_id > nsid)
2406 			nvme_ns_remove(ns);
2407 	}
2408 }
2409 
2410 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
2411 {
2412 	struct nvme_ns *ns;
2413 	__le32 *ns_list;
2414 	unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
2415 	int ret = 0;
2416 
2417 	ns_list = kzalloc(0x1000, GFP_KERNEL);
2418 	if (!ns_list)
2419 		return -ENOMEM;
2420 
2421 	for (i = 0; i < num_lists; i++) {
2422 		ret = nvme_identify_ns_list(ctrl, prev, ns_list);
2423 		if (ret)
2424 			goto free;
2425 
2426 		for (j = 0; j < min(nn, 1024U); j++) {
2427 			nsid = le32_to_cpu(ns_list[j]);
2428 			if (!nsid)
2429 				goto out;
2430 
2431 			nvme_validate_ns(ctrl, nsid);
2432 
2433 			while (++prev < nsid) {
2434 				ns = nvme_find_get_ns(ctrl, prev);
2435 				if (ns) {
2436 					nvme_ns_remove(ns);
2437 					nvme_put_ns(ns);
2438 				}
2439 			}
2440 		}
2441 		nn -= j;
2442 	}
2443  out:
2444 	nvme_remove_invalid_namespaces(ctrl, prev);
2445  free:
2446 	kfree(ns_list);
2447 	return ret;
2448 }
2449 
2450 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
2451 {
2452 	unsigned i;
2453 
2454 	for (i = 1; i <= nn; i++)
2455 		nvme_validate_ns(ctrl, i);
2456 
2457 	nvme_remove_invalid_namespaces(ctrl, nn);
2458 }
2459 
2460 static void nvme_scan_work(struct work_struct *work)
2461 {
2462 	struct nvme_ctrl *ctrl =
2463 		container_of(work, struct nvme_ctrl, scan_work);
2464 	struct nvme_id_ctrl *id;
2465 	unsigned nn;
2466 
2467 	if (ctrl->state != NVME_CTRL_LIVE)
2468 		return;
2469 
2470 	if (nvme_identify_ctrl(ctrl, &id))
2471 		return;
2472 
2473 	nn = le32_to_cpu(id->nn);
2474 	if (ctrl->vs >= NVME_VS(1, 1, 0) &&
2475 	    !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
2476 		if (!nvme_scan_ns_list(ctrl, nn))
2477 			goto done;
2478 	}
2479 	nvme_scan_ns_sequential(ctrl, nn);
2480  done:
2481 	mutex_lock(&ctrl->namespaces_mutex);
2482 	list_sort(NULL, &ctrl->namespaces, ns_cmp);
2483 	mutex_unlock(&ctrl->namespaces_mutex);
2484 	kfree(id);
2485 }
2486 
2487 void nvme_queue_scan(struct nvme_ctrl *ctrl)
2488 {
2489 	/*
2490 	 * Do not queue new scan work when a controller is reset during
2491 	 * removal.
2492 	 */
2493 	if (ctrl->state == NVME_CTRL_LIVE)
2494 		queue_work(nvme_wq, &ctrl->scan_work);
2495 }
2496 EXPORT_SYMBOL_GPL(nvme_queue_scan);
2497 
2498 /*
2499  * This function iterates the namespace list unlocked to allow recovery from
2500  * controller failure. It is up to the caller to ensure the namespace list is
2501  * not modified by scan work while this function is executing.
2502  */
2503 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
2504 {
2505 	struct nvme_ns *ns, *next;
2506 
2507 	/*
2508 	 * The dead states indicates the controller was not gracefully
2509 	 * disconnected. In that case, we won't be able to flush any data while
2510 	 * removing the namespaces' disks; fail all the queues now to avoid
2511 	 * potentially having to clean up the failed sync later.
2512 	 */
2513 	if (ctrl->state == NVME_CTRL_DEAD)
2514 		nvme_kill_queues(ctrl);
2515 
2516 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
2517 		nvme_ns_remove(ns);
2518 }
2519 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
2520 
2521 static void nvme_async_event_work(struct work_struct *work)
2522 {
2523 	struct nvme_ctrl *ctrl =
2524 		container_of(work, struct nvme_ctrl, async_event_work);
2525 
2526 	spin_lock_irq(&ctrl->lock);
2527 	while (ctrl->event_limit > 0) {
2528 		int aer_idx = --ctrl->event_limit;
2529 
2530 		spin_unlock_irq(&ctrl->lock);
2531 		ctrl->ops->submit_async_event(ctrl, aer_idx);
2532 		spin_lock_irq(&ctrl->lock);
2533 	}
2534 	spin_unlock_irq(&ctrl->lock);
2535 }
2536 
2537 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
2538 		union nvme_result *res)
2539 {
2540 	u32 result = le32_to_cpu(res->u32);
2541 	bool done = true;
2542 
2543 	switch (le16_to_cpu(status) >> 1) {
2544 	case NVME_SC_SUCCESS:
2545 		done = false;
2546 		/*FALLTHRU*/
2547 	case NVME_SC_ABORT_REQ:
2548 		++ctrl->event_limit;
2549 		queue_work(nvme_wq, &ctrl->async_event_work);
2550 		break;
2551 	default:
2552 		break;
2553 	}
2554 
2555 	if (done)
2556 		return;
2557 
2558 	switch (result & 0xff07) {
2559 	case NVME_AER_NOTICE_NS_CHANGED:
2560 		dev_info(ctrl->device, "rescanning\n");
2561 		nvme_queue_scan(ctrl);
2562 		break;
2563 	default:
2564 		dev_warn(ctrl->device, "async event result %08x\n", result);
2565 	}
2566 }
2567 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
2568 
2569 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
2570 {
2571 	ctrl->event_limit = NVME_NR_AERS;
2572 	queue_work(nvme_wq, &ctrl->async_event_work);
2573 }
2574 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
2575 
2576 static DEFINE_IDA(nvme_instance_ida);
2577 
2578 static int nvme_set_instance(struct nvme_ctrl *ctrl)
2579 {
2580 	int instance, error;
2581 
2582 	do {
2583 		if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2584 			return -ENODEV;
2585 
2586 		spin_lock(&dev_list_lock);
2587 		error = ida_get_new(&nvme_instance_ida, &instance);
2588 		spin_unlock(&dev_list_lock);
2589 	} while (error == -EAGAIN);
2590 
2591 	if (error)
2592 		return -ENODEV;
2593 
2594 	ctrl->instance = instance;
2595 	return 0;
2596 }
2597 
2598 static void nvme_release_instance(struct nvme_ctrl *ctrl)
2599 {
2600 	spin_lock(&dev_list_lock);
2601 	ida_remove(&nvme_instance_ida, ctrl->instance);
2602 	spin_unlock(&dev_list_lock);
2603 }
2604 
2605 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
2606 {
2607 	nvme_stop_keep_alive(ctrl);
2608 	flush_work(&ctrl->async_event_work);
2609 	flush_work(&ctrl->scan_work);
2610 }
2611 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
2612 
2613 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
2614 {
2615 	if (ctrl->kato)
2616 		nvme_start_keep_alive(ctrl);
2617 
2618 	if (ctrl->queue_count > 1) {
2619 		nvme_queue_scan(ctrl);
2620 		nvme_queue_async_events(ctrl);
2621 		nvme_start_queues(ctrl);
2622 	}
2623 }
2624 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
2625 
2626 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
2627 {
2628 	device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
2629 
2630 	spin_lock(&dev_list_lock);
2631 	list_del(&ctrl->node);
2632 	spin_unlock(&dev_list_lock);
2633 }
2634 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
2635 
2636 static void nvme_free_ctrl(struct kref *kref)
2637 {
2638 	struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
2639 
2640 	put_device(ctrl->device);
2641 	nvme_release_instance(ctrl);
2642 	ida_destroy(&ctrl->ns_ida);
2643 
2644 	ctrl->ops->free_ctrl(ctrl);
2645 }
2646 
2647 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
2648 {
2649 	kref_put(&ctrl->kref, nvme_free_ctrl);
2650 }
2651 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2652 
2653 /*
2654  * Initialize a NVMe controller structures.  This needs to be called during
2655  * earliest initialization so that we have the initialized structured around
2656  * during probing.
2657  */
2658 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2659 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
2660 {
2661 	int ret;
2662 
2663 	ctrl->state = NVME_CTRL_NEW;
2664 	spin_lock_init(&ctrl->lock);
2665 	INIT_LIST_HEAD(&ctrl->namespaces);
2666 	mutex_init(&ctrl->namespaces_mutex);
2667 	kref_init(&ctrl->kref);
2668 	ctrl->dev = dev;
2669 	ctrl->ops = ops;
2670 	ctrl->quirks = quirks;
2671 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2672 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2673 
2674 	ret = nvme_set_instance(ctrl);
2675 	if (ret)
2676 		goto out;
2677 
2678 	ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2679 				MKDEV(nvme_char_major, ctrl->instance),
2680 				ctrl, nvme_dev_attr_groups,
2681 				"nvme%d", ctrl->instance);
2682 	if (IS_ERR(ctrl->device)) {
2683 		ret = PTR_ERR(ctrl->device);
2684 		goto out_release_instance;
2685 	}
2686 	get_device(ctrl->device);
2687 	ida_init(&ctrl->ns_ida);
2688 
2689 	spin_lock(&dev_list_lock);
2690 	list_add_tail(&ctrl->node, &nvme_ctrl_list);
2691 	spin_unlock(&dev_list_lock);
2692 
2693 	/*
2694 	 * Initialize latency tolerance controls.  The sysfs files won't
2695 	 * be visible to userspace unless the device actually supports APST.
2696 	 */
2697 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
2698 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
2699 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
2700 
2701 	return 0;
2702 out_release_instance:
2703 	nvme_release_instance(ctrl);
2704 out:
2705 	return ret;
2706 }
2707 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2708 
2709 /**
2710  * nvme_kill_queues(): Ends all namespace queues
2711  * @ctrl: the dead controller that needs to end
2712  *
2713  * Call this function when the driver determines it is unable to get the
2714  * controller in a state capable of servicing IO.
2715  */
2716 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2717 {
2718 	struct nvme_ns *ns;
2719 
2720 	mutex_lock(&ctrl->namespaces_mutex);
2721 
2722 	/* Forcibly unquiesce queues to avoid blocking dispatch */
2723 	if (ctrl->admin_q)
2724 		blk_mq_unquiesce_queue(ctrl->admin_q);
2725 
2726 	list_for_each_entry(ns, &ctrl->namespaces, list) {
2727 		/*
2728 		 * Revalidating a dead namespace sets capacity to 0. This will
2729 		 * end buffered writers dirtying pages that can't be synced.
2730 		 */
2731 		if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2732 			continue;
2733 		revalidate_disk(ns->disk);
2734 		blk_set_queue_dying(ns->queue);
2735 
2736 		/* Forcibly unquiesce queues to avoid blocking dispatch */
2737 		blk_mq_unquiesce_queue(ns->queue);
2738 	}
2739 	mutex_unlock(&ctrl->namespaces_mutex);
2740 }
2741 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2742 
2743 void nvme_unfreeze(struct nvme_ctrl *ctrl)
2744 {
2745 	struct nvme_ns *ns;
2746 
2747 	mutex_lock(&ctrl->namespaces_mutex);
2748 	list_for_each_entry(ns, &ctrl->namespaces, list)
2749 		blk_mq_unfreeze_queue(ns->queue);
2750 	mutex_unlock(&ctrl->namespaces_mutex);
2751 }
2752 EXPORT_SYMBOL_GPL(nvme_unfreeze);
2753 
2754 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
2755 {
2756 	struct nvme_ns *ns;
2757 
2758 	mutex_lock(&ctrl->namespaces_mutex);
2759 	list_for_each_entry(ns, &ctrl->namespaces, list) {
2760 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
2761 		if (timeout <= 0)
2762 			break;
2763 	}
2764 	mutex_unlock(&ctrl->namespaces_mutex);
2765 }
2766 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
2767 
2768 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
2769 {
2770 	struct nvme_ns *ns;
2771 
2772 	mutex_lock(&ctrl->namespaces_mutex);
2773 	list_for_each_entry(ns, &ctrl->namespaces, list)
2774 		blk_mq_freeze_queue_wait(ns->queue);
2775 	mutex_unlock(&ctrl->namespaces_mutex);
2776 }
2777 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
2778 
2779 void nvme_start_freeze(struct nvme_ctrl *ctrl)
2780 {
2781 	struct nvme_ns *ns;
2782 
2783 	mutex_lock(&ctrl->namespaces_mutex);
2784 	list_for_each_entry(ns, &ctrl->namespaces, list)
2785 		blk_freeze_queue_start(ns->queue);
2786 	mutex_unlock(&ctrl->namespaces_mutex);
2787 }
2788 EXPORT_SYMBOL_GPL(nvme_start_freeze);
2789 
2790 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2791 {
2792 	struct nvme_ns *ns;
2793 
2794 	mutex_lock(&ctrl->namespaces_mutex);
2795 	list_for_each_entry(ns, &ctrl->namespaces, list)
2796 		blk_mq_quiesce_queue(ns->queue);
2797 	mutex_unlock(&ctrl->namespaces_mutex);
2798 }
2799 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2800 
2801 void nvme_start_queues(struct nvme_ctrl *ctrl)
2802 {
2803 	struct nvme_ns *ns;
2804 
2805 	mutex_lock(&ctrl->namespaces_mutex);
2806 	list_for_each_entry(ns, &ctrl->namespaces, list)
2807 		blk_mq_unquiesce_queue(ns->queue);
2808 	mutex_unlock(&ctrl->namespaces_mutex);
2809 }
2810 EXPORT_SYMBOL_GPL(nvme_start_queues);
2811 
2812 int __init nvme_core_init(void)
2813 {
2814 	int result;
2815 
2816 	nvme_wq = alloc_workqueue("nvme-wq",
2817 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
2818 	if (!nvme_wq)
2819 		return -ENOMEM;
2820 
2821 	result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2822 							&nvme_dev_fops);
2823 	if (result < 0)
2824 		goto destroy_wq;
2825 	else if (result > 0)
2826 		nvme_char_major = result;
2827 
2828 	nvme_class = class_create(THIS_MODULE, "nvme");
2829 	if (IS_ERR(nvme_class)) {
2830 		result = PTR_ERR(nvme_class);
2831 		goto unregister_chrdev;
2832 	}
2833 
2834 	return 0;
2835 
2836 unregister_chrdev:
2837 	__unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2838 destroy_wq:
2839 	destroy_workqueue(nvme_wq);
2840 	return result;
2841 }
2842 
2843 void nvme_core_exit(void)
2844 {
2845 	class_destroy(nvme_class);
2846 	__unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2847 	destroy_workqueue(nvme_wq);
2848 }
2849 
2850 MODULE_LICENSE("GPL");
2851 MODULE_VERSION("1.0");
2852 module_init(nvme_core_init);
2853 module_exit(nvme_core_exit);
2854