xref: /openbmc/linux/drivers/nvme/host/core.c (revision c127f98ba9aba1818a6ca3a1da5a24653a10d966)
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 int admin_timeout = 60;
38 module_param(admin_timeout, uint, 0644);
39 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
40 EXPORT_SYMBOL_GPL(admin_timeout);
41 
42 unsigned int nvme_io_timeout = 30;
43 module_param_named(io_timeout, nvme_io_timeout, uint, 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 unsigned long default_ps_max_latency_us = 100000;
56 module_param(default_ps_max_latency_us, ulong, 0644);
57 MODULE_PARM_DESC(default_ps_max_latency_us,
58 		 "max power saving latency for new devices; use PM QOS to change per device");
59 
60 static bool force_apst;
61 module_param(force_apst, bool, 0644);
62 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
63 
64 static bool streams;
65 module_param(streams, bool, 0644);
66 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
67 
68 struct workqueue_struct *nvme_wq;
69 EXPORT_SYMBOL_GPL(nvme_wq);
70 
71 static DEFINE_IDA(nvme_subsystems_ida);
72 static LIST_HEAD(nvme_subsystems);
73 static DEFINE_MUTEX(nvme_subsystems_lock);
74 
75 static DEFINE_IDA(nvme_instance_ida);
76 static dev_t nvme_chr_devt;
77 static struct class *nvme_class;
78 static struct class *nvme_subsys_class;
79 
80 static void nvme_ns_remove(struct nvme_ns *ns);
81 static int nvme_revalidate_disk(struct gendisk *disk);
82 
83 static __le32 nvme_get_log_dw10(u8 lid, size_t size)
84 {
85 	return cpu_to_le32((((size / 4) - 1) << 16) | lid);
86 }
87 
88 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
89 {
90 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
91 		return -EBUSY;
92 	if (!queue_work(nvme_wq, &ctrl->reset_work))
93 		return -EBUSY;
94 	return 0;
95 }
96 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
97 
98 static int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
99 {
100 	int ret;
101 
102 	ret = nvme_reset_ctrl(ctrl);
103 	if (!ret)
104 		flush_work(&ctrl->reset_work);
105 	return ret;
106 }
107 
108 static void nvme_delete_ctrl_work(struct work_struct *work)
109 {
110 	struct nvme_ctrl *ctrl =
111 		container_of(work, struct nvme_ctrl, delete_work);
112 
113 	flush_work(&ctrl->reset_work);
114 	nvme_stop_ctrl(ctrl);
115 	nvme_remove_namespaces(ctrl);
116 	ctrl->ops->delete_ctrl(ctrl);
117 	nvme_uninit_ctrl(ctrl);
118 	nvme_put_ctrl(ctrl);
119 }
120 
121 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
122 {
123 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
124 		return -EBUSY;
125 	if (!queue_work(nvme_wq, &ctrl->delete_work))
126 		return -EBUSY;
127 	return 0;
128 }
129 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
130 
131 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
132 {
133 	int ret = 0;
134 
135 	/*
136 	 * Keep a reference until the work is flushed since ->delete_ctrl
137 	 * can free the controller.
138 	 */
139 	nvme_get_ctrl(ctrl);
140 	ret = nvme_delete_ctrl(ctrl);
141 	if (!ret)
142 		flush_work(&ctrl->delete_work);
143 	nvme_put_ctrl(ctrl);
144 	return ret;
145 }
146 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
147 
148 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
149 {
150 	return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
151 }
152 
153 static blk_status_t nvme_error_status(struct request *req)
154 {
155 	switch (nvme_req(req)->status & 0x7ff) {
156 	case NVME_SC_SUCCESS:
157 		return BLK_STS_OK;
158 	case NVME_SC_CAP_EXCEEDED:
159 		return BLK_STS_NOSPC;
160 	case NVME_SC_ONCS_NOT_SUPPORTED:
161 		return BLK_STS_NOTSUPP;
162 	case NVME_SC_WRITE_FAULT:
163 	case NVME_SC_READ_ERROR:
164 	case NVME_SC_UNWRITTEN_BLOCK:
165 	case NVME_SC_ACCESS_DENIED:
166 	case NVME_SC_READ_ONLY:
167 		return BLK_STS_MEDIUM;
168 	case NVME_SC_GUARD_CHECK:
169 	case NVME_SC_APPTAG_CHECK:
170 	case NVME_SC_REFTAG_CHECK:
171 	case NVME_SC_INVALID_PI:
172 		return BLK_STS_PROTECTION;
173 	case NVME_SC_RESERVATION_CONFLICT:
174 		return BLK_STS_NEXUS;
175 	default:
176 		return BLK_STS_IOERR;
177 	}
178 }
179 
180 static inline bool nvme_req_needs_retry(struct request *req)
181 {
182 	if (blk_noretry_request(req))
183 		return false;
184 	if (nvme_req(req)->status & NVME_SC_DNR)
185 		return false;
186 	if (nvme_req(req)->retries >= nvme_max_retries)
187 		return false;
188 	return true;
189 }
190 
191 void nvme_complete_rq(struct request *req)
192 {
193 	if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) {
194 		if (nvme_req_needs_failover(req)) {
195 			nvme_failover_req(req);
196 			return;
197 		}
198 
199 		if (!blk_queue_dying(req->q)) {
200 			nvme_req(req)->retries++;
201 			blk_mq_requeue_request(req, true);
202 			return;
203 		}
204 	}
205 
206 	blk_mq_end_request(req, nvme_error_status(req));
207 }
208 EXPORT_SYMBOL_GPL(nvme_complete_rq);
209 
210 void nvme_cancel_request(struct request *req, void *data, bool reserved)
211 {
212 	if (!blk_mq_request_started(req))
213 		return;
214 
215 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
216 				"Cancelling I/O %d", req->tag);
217 
218 	nvme_req(req)->status = NVME_SC_ABORT_REQ;
219 	blk_mq_complete_request(req);
220 
221 }
222 EXPORT_SYMBOL_GPL(nvme_cancel_request);
223 
224 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
225 		enum nvme_ctrl_state new_state)
226 {
227 	enum nvme_ctrl_state old_state;
228 	unsigned long flags;
229 	bool changed = false;
230 
231 	spin_lock_irqsave(&ctrl->lock, flags);
232 
233 	old_state = ctrl->state;
234 	switch (new_state) {
235 	case NVME_CTRL_LIVE:
236 		switch (old_state) {
237 		case NVME_CTRL_NEW:
238 		case NVME_CTRL_RESETTING:
239 		case NVME_CTRL_RECONNECTING:
240 			changed = true;
241 			/* FALLTHRU */
242 		default:
243 			break;
244 		}
245 		break;
246 	case NVME_CTRL_RESETTING:
247 		switch (old_state) {
248 		case NVME_CTRL_NEW:
249 		case NVME_CTRL_LIVE:
250 			changed = true;
251 			/* FALLTHRU */
252 		default:
253 			break;
254 		}
255 		break;
256 	case NVME_CTRL_RECONNECTING:
257 		switch (old_state) {
258 		case NVME_CTRL_LIVE:
259 		case NVME_CTRL_RESETTING:
260 			changed = true;
261 			/* FALLTHRU */
262 		default:
263 			break;
264 		}
265 		break;
266 	case NVME_CTRL_DELETING:
267 		switch (old_state) {
268 		case NVME_CTRL_LIVE:
269 		case NVME_CTRL_RESETTING:
270 		case NVME_CTRL_RECONNECTING:
271 			changed = true;
272 			/* FALLTHRU */
273 		default:
274 			break;
275 		}
276 		break;
277 	case NVME_CTRL_DEAD:
278 		switch (old_state) {
279 		case NVME_CTRL_DELETING:
280 			changed = true;
281 			/* FALLTHRU */
282 		default:
283 			break;
284 		}
285 		break;
286 	default:
287 		break;
288 	}
289 
290 	if (changed)
291 		ctrl->state = new_state;
292 
293 	spin_unlock_irqrestore(&ctrl->lock, flags);
294 	if (changed && ctrl->state == NVME_CTRL_LIVE)
295 		nvme_kick_requeue_lists(ctrl);
296 	return changed;
297 }
298 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
299 
300 static void nvme_free_ns_head(struct kref *ref)
301 {
302 	struct nvme_ns_head *head =
303 		container_of(ref, struct nvme_ns_head, ref);
304 
305 	nvme_mpath_remove_disk(head);
306 	ida_simple_remove(&head->subsys->ns_ida, head->instance);
307 	list_del_init(&head->entry);
308 	cleanup_srcu_struct(&head->srcu);
309 	kfree(head);
310 }
311 
312 static void nvme_put_ns_head(struct nvme_ns_head *head)
313 {
314 	kref_put(&head->ref, nvme_free_ns_head);
315 }
316 
317 static void nvme_free_ns(struct kref *kref)
318 {
319 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
320 
321 	if (ns->ndev)
322 		nvme_nvm_unregister(ns);
323 
324 	put_disk(ns->disk);
325 	nvme_put_ns_head(ns->head);
326 	nvme_put_ctrl(ns->ctrl);
327 	kfree(ns);
328 }
329 
330 static void nvme_put_ns(struct nvme_ns *ns)
331 {
332 	kref_put(&ns->kref, nvme_free_ns);
333 }
334 
335 struct request *nvme_alloc_request(struct request_queue *q,
336 		struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
337 {
338 	unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
339 	struct request *req;
340 
341 	if (qid == NVME_QID_ANY) {
342 		req = blk_mq_alloc_request(q, op, flags);
343 	} else {
344 		req = blk_mq_alloc_request_hctx(q, op, flags,
345 				qid ? qid - 1 : 0);
346 	}
347 	if (IS_ERR(req))
348 		return req;
349 
350 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
351 	nvme_req(req)->cmd = cmd;
352 
353 	return req;
354 }
355 EXPORT_SYMBOL_GPL(nvme_alloc_request);
356 
357 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
358 {
359 	struct nvme_command c;
360 
361 	memset(&c, 0, sizeof(c));
362 
363 	c.directive.opcode = nvme_admin_directive_send;
364 	c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
365 	c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
366 	c.directive.dtype = NVME_DIR_IDENTIFY;
367 	c.directive.tdtype = NVME_DIR_STREAMS;
368 	c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
369 
370 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
371 }
372 
373 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
374 {
375 	return nvme_toggle_streams(ctrl, false);
376 }
377 
378 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
379 {
380 	return nvme_toggle_streams(ctrl, true);
381 }
382 
383 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
384 				  struct streams_directive_params *s, u32 nsid)
385 {
386 	struct nvme_command c;
387 
388 	memset(&c, 0, sizeof(c));
389 	memset(s, 0, sizeof(*s));
390 
391 	c.directive.opcode = nvme_admin_directive_recv;
392 	c.directive.nsid = cpu_to_le32(nsid);
393 	c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
394 	c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
395 	c.directive.dtype = NVME_DIR_STREAMS;
396 
397 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
398 }
399 
400 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
401 {
402 	struct streams_directive_params s;
403 	int ret;
404 
405 	if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
406 		return 0;
407 	if (!streams)
408 		return 0;
409 
410 	ret = nvme_enable_streams(ctrl);
411 	if (ret)
412 		return ret;
413 
414 	ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
415 	if (ret)
416 		return ret;
417 
418 	ctrl->nssa = le16_to_cpu(s.nssa);
419 	if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
420 		dev_info(ctrl->device, "too few streams (%u) available\n",
421 					ctrl->nssa);
422 		nvme_disable_streams(ctrl);
423 		return 0;
424 	}
425 
426 	ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
427 	dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
428 	return 0;
429 }
430 
431 /*
432  * Check if 'req' has a write hint associated with it. If it does, assign
433  * a valid namespace stream to the write.
434  */
435 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
436 				     struct request *req, u16 *control,
437 				     u32 *dsmgmt)
438 {
439 	enum rw_hint streamid = req->write_hint;
440 
441 	if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
442 		streamid = 0;
443 	else {
444 		streamid--;
445 		if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
446 			return;
447 
448 		*control |= NVME_RW_DTYPE_STREAMS;
449 		*dsmgmt |= streamid << 16;
450 	}
451 
452 	if (streamid < ARRAY_SIZE(req->q->write_hints))
453 		req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
454 }
455 
456 static inline void nvme_setup_flush(struct nvme_ns *ns,
457 		struct nvme_command *cmnd)
458 {
459 	memset(cmnd, 0, sizeof(*cmnd));
460 	cmnd->common.opcode = nvme_cmd_flush;
461 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
462 }
463 
464 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
465 		struct nvme_command *cmnd)
466 {
467 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
468 	struct nvme_dsm_range *range;
469 	struct bio *bio;
470 
471 	range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
472 	if (!range)
473 		return BLK_STS_RESOURCE;
474 
475 	__rq_for_each_bio(bio, req) {
476 		u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
477 		u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
478 
479 		range[n].cattr = cpu_to_le32(0);
480 		range[n].nlb = cpu_to_le32(nlb);
481 		range[n].slba = cpu_to_le64(slba);
482 		n++;
483 	}
484 
485 	if (WARN_ON_ONCE(n != segments)) {
486 		kfree(range);
487 		return BLK_STS_IOERR;
488 	}
489 
490 	memset(cmnd, 0, sizeof(*cmnd));
491 	cmnd->dsm.opcode = nvme_cmd_dsm;
492 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
493 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
494 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
495 
496 	req->special_vec.bv_page = virt_to_page(range);
497 	req->special_vec.bv_offset = offset_in_page(range);
498 	req->special_vec.bv_len = sizeof(*range) * segments;
499 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
500 
501 	return BLK_STS_OK;
502 }
503 
504 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
505 		struct request *req, struct nvme_command *cmnd)
506 {
507 	struct nvme_ctrl *ctrl = ns->ctrl;
508 	u16 control = 0;
509 	u32 dsmgmt = 0;
510 
511 	if (req->cmd_flags & REQ_FUA)
512 		control |= NVME_RW_FUA;
513 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
514 		control |= NVME_RW_LR;
515 
516 	if (req->cmd_flags & REQ_RAHEAD)
517 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
518 
519 	memset(cmnd, 0, sizeof(*cmnd));
520 	cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
521 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
522 	cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
523 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
524 
525 	if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
526 		nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
527 
528 	if (ns->ms) {
529 		/*
530 		 * If formated with metadata, the block layer always provides a
531 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
532 		 * we enable the PRACT bit for protection information or set the
533 		 * namespace capacity to zero to prevent any I/O.
534 		 */
535 		if (!blk_integrity_rq(req)) {
536 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
537 				return BLK_STS_NOTSUPP;
538 			control |= NVME_RW_PRINFO_PRACT;
539 		}
540 
541 		switch (ns->pi_type) {
542 		case NVME_NS_DPS_PI_TYPE3:
543 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
544 			break;
545 		case NVME_NS_DPS_PI_TYPE1:
546 		case NVME_NS_DPS_PI_TYPE2:
547 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
548 					NVME_RW_PRINFO_PRCHK_REF;
549 			cmnd->rw.reftag = cpu_to_le32(
550 					nvme_block_nr(ns, blk_rq_pos(req)));
551 			break;
552 		}
553 	}
554 
555 	cmnd->rw.control = cpu_to_le16(control);
556 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
557 	return 0;
558 }
559 
560 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
561 		struct nvme_command *cmd)
562 {
563 	blk_status_t ret = BLK_STS_OK;
564 
565 	if (!(req->rq_flags & RQF_DONTPREP)) {
566 		nvme_req(req)->retries = 0;
567 		nvme_req(req)->flags = 0;
568 		req->rq_flags |= RQF_DONTPREP;
569 	}
570 
571 	switch (req_op(req)) {
572 	case REQ_OP_DRV_IN:
573 	case REQ_OP_DRV_OUT:
574 		memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
575 		break;
576 	case REQ_OP_FLUSH:
577 		nvme_setup_flush(ns, cmd);
578 		break;
579 	case REQ_OP_WRITE_ZEROES:
580 		/* currently only aliased to deallocate for a few ctrls: */
581 	case REQ_OP_DISCARD:
582 		ret = nvme_setup_discard(ns, req, cmd);
583 		break;
584 	case REQ_OP_READ:
585 	case REQ_OP_WRITE:
586 		ret = nvme_setup_rw(ns, req, cmd);
587 		break;
588 	default:
589 		WARN_ON_ONCE(1);
590 		return BLK_STS_IOERR;
591 	}
592 
593 	cmd->common.command_id = req->tag;
594 	return ret;
595 }
596 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
597 
598 /*
599  * Returns 0 on success.  If the result is negative, it's a Linux error code;
600  * if the result is positive, it's an NVM Express status code
601  */
602 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
603 		union nvme_result *result, void *buffer, unsigned bufflen,
604 		unsigned timeout, int qid, int at_head,
605 		blk_mq_req_flags_t flags)
606 {
607 	struct request *req;
608 	int ret;
609 
610 	req = nvme_alloc_request(q, cmd, flags, qid);
611 	if (IS_ERR(req))
612 		return PTR_ERR(req);
613 
614 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
615 
616 	if (buffer && bufflen) {
617 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
618 		if (ret)
619 			goto out;
620 	}
621 
622 	blk_execute_rq(req->q, NULL, req, at_head);
623 	if (result)
624 		*result = nvme_req(req)->result;
625 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
626 		ret = -EINTR;
627 	else
628 		ret = nvme_req(req)->status;
629  out:
630 	blk_mq_free_request(req);
631 	return ret;
632 }
633 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
634 
635 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
636 		void *buffer, unsigned bufflen)
637 {
638 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
639 			NVME_QID_ANY, 0, 0);
640 }
641 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
642 
643 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
644 		unsigned len, u32 seed, bool write)
645 {
646 	struct bio_integrity_payload *bip;
647 	int ret = -ENOMEM;
648 	void *buf;
649 
650 	buf = kmalloc(len, GFP_KERNEL);
651 	if (!buf)
652 		goto out;
653 
654 	ret = -EFAULT;
655 	if (write && copy_from_user(buf, ubuf, len))
656 		goto out_free_meta;
657 
658 	bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
659 	if (IS_ERR(bip)) {
660 		ret = PTR_ERR(bip);
661 		goto out_free_meta;
662 	}
663 
664 	bip->bip_iter.bi_size = len;
665 	bip->bip_iter.bi_sector = seed;
666 	ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
667 			offset_in_page(buf));
668 	if (ret == len)
669 		return buf;
670 	ret = -ENOMEM;
671 out_free_meta:
672 	kfree(buf);
673 out:
674 	return ERR_PTR(ret);
675 }
676 
677 static int nvme_submit_user_cmd(struct request_queue *q,
678 		struct nvme_command *cmd, void __user *ubuffer,
679 		unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
680 		u32 meta_seed, u32 *result, unsigned timeout)
681 {
682 	bool write = nvme_is_write(cmd);
683 	struct nvme_ns *ns = q->queuedata;
684 	struct gendisk *disk = ns ? ns->disk : NULL;
685 	struct request *req;
686 	struct bio *bio = NULL;
687 	void *meta = NULL;
688 	int ret;
689 
690 	req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
691 	if (IS_ERR(req))
692 		return PTR_ERR(req);
693 
694 	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
695 
696 	if (ubuffer && bufflen) {
697 		ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
698 				GFP_KERNEL);
699 		if (ret)
700 			goto out;
701 		bio = req->bio;
702 		bio->bi_disk = disk;
703 		if (disk && meta_buffer && meta_len) {
704 			meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
705 					meta_seed, write);
706 			if (IS_ERR(meta)) {
707 				ret = PTR_ERR(meta);
708 				goto out_unmap;
709 			}
710 		}
711 	}
712 
713 	blk_execute_rq(req->q, disk, req, 0);
714 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
715 		ret = -EINTR;
716 	else
717 		ret = nvme_req(req)->status;
718 	if (result)
719 		*result = le32_to_cpu(nvme_req(req)->result.u32);
720 	if (meta && !ret && !write) {
721 		if (copy_to_user(meta_buffer, meta, meta_len))
722 			ret = -EFAULT;
723 	}
724 	kfree(meta);
725  out_unmap:
726 	if (bio)
727 		blk_rq_unmap_user(bio);
728  out:
729 	blk_mq_free_request(req);
730 	return ret;
731 }
732 
733 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
734 {
735 	struct nvme_ctrl *ctrl = rq->end_io_data;
736 
737 	blk_mq_free_request(rq);
738 
739 	if (status) {
740 		dev_err(ctrl->device,
741 			"failed nvme_keep_alive_end_io error=%d\n",
742 				status);
743 		return;
744 	}
745 
746 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
747 }
748 
749 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
750 {
751 	struct nvme_command c;
752 	struct request *rq;
753 
754 	memset(&c, 0, sizeof(c));
755 	c.common.opcode = nvme_admin_keep_alive;
756 
757 	rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
758 			NVME_QID_ANY);
759 	if (IS_ERR(rq))
760 		return PTR_ERR(rq);
761 
762 	rq->timeout = ctrl->kato * HZ;
763 	rq->end_io_data = ctrl;
764 
765 	blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
766 
767 	return 0;
768 }
769 
770 static void nvme_keep_alive_work(struct work_struct *work)
771 {
772 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
773 			struct nvme_ctrl, ka_work);
774 
775 	if (nvme_keep_alive(ctrl)) {
776 		/* allocation failure, reset the controller */
777 		dev_err(ctrl->device, "keep-alive failed\n");
778 		nvme_reset_ctrl(ctrl);
779 		return;
780 	}
781 }
782 
783 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
784 {
785 	if (unlikely(ctrl->kato == 0))
786 		return;
787 
788 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
789 	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
790 }
791 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
792 
793 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
794 {
795 	if (unlikely(ctrl->kato == 0))
796 		return;
797 
798 	cancel_delayed_work_sync(&ctrl->ka_work);
799 }
800 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
801 
802 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
803 {
804 	struct nvme_command c = { };
805 	int error;
806 
807 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
808 	c.identify.opcode = nvme_admin_identify;
809 	c.identify.cns = NVME_ID_CNS_CTRL;
810 
811 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
812 	if (!*id)
813 		return -ENOMEM;
814 
815 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
816 			sizeof(struct nvme_id_ctrl));
817 	if (error)
818 		kfree(*id);
819 	return error;
820 }
821 
822 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
823 		struct nvme_ns_ids *ids)
824 {
825 	struct nvme_command c = { };
826 	int status;
827 	void *data;
828 	int pos;
829 	int len;
830 
831 	c.identify.opcode = nvme_admin_identify;
832 	c.identify.nsid = cpu_to_le32(nsid);
833 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
834 
835 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
836 	if (!data)
837 		return -ENOMEM;
838 
839 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
840 				      NVME_IDENTIFY_DATA_SIZE);
841 	if (status)
842 		goto free_data;
843 
844 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
845 		struct nvme_ns_id_desc *cur = data + pos;
846 
847 		if (cur->nidl == 0)
848 			break;
849 
850 		switch (cur->nidt) {
851 		case NVME_NIDT_EUI64:
852 			if (cur->nidl != NVME_NIDT_EUI64_LEN) {
853 				dev_warn(ctrl->device,
854 					 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
855 					 cur->nidl);
856 				goto free_data;
857 			}
858 			len = NVME_NIDT_EUI64_LEN;
859 			memcpy(ids->eui64, data + pos + sizeof(*cur), len);
860 			break;
861 		case NVME_NIDT_NGUID:
862 			if (cur->nidl != NVME_NIDT_NGUID_LEN) {
863 				dev_warn(ctrl->device,
864 					 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
865 					 cur->nidl);
866 				goto free_data;
867 			}
868 			len = NVME_NIDT_NGUID_LEN;
869 			memcpy(ids->nguid, data + pos + sizeof(*cur), len);
870 			break;
871 		case NVME_NIDT_UUID:
872 			if (cur->nidl != NVME_NIDT_UUID_LEN) {
873 				dev_warn(ctrl->device,
874 					 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
875 					 cur->nidl);
876 				goto free_data;
877 			}
878 			len = NVME_NIDT_UUID_LEN;
879 			uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
880 			break;
881 		default:
882 			/* Skip unnkown types */
883 			len = cur->nidl;
884 			break;
885 		}
886 
887 		len += sizeof(*cur);
888 	}
889 free_data:
890 	kfree(data);
891 	return status;
892 }
893 
894 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
895 {
896 	struct nvme_command c = { };
897 
898 	c.identify.opcode = nvme_admin_identify;
899 	c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
900 	c.identify.nsid = cpu_to_le32(nsid);
901 	return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
902 }
903 
904 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
905 		unsigned nsid)
906 {
907 	struct nvme_id_ns *id;
908 	struct nvme_command c = { };
909 	int error;
910 
911 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
912 	c.identify.opcode = nvme_admin_identify;
913 	c.identify.nsid = cpu_to_le32(nsid);
914 	c.identify.cns = NVME_ID_CNS_NS;
915 
916 	id = kmalloc(sizeof(*id), GFP_KERNEL);
917 	if (!id)
918 		return NULL;
919 
920 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
921 	if (error) {
922 		dev_warn(ctrl->device, "Identify namespace failed\n");
923 		kfree(id);
924 		return NULL;
925 	}
926 
927 	return id;
928 }
929 
930 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
931 		      void *buffer, size_t buflen, u32 *result)
932 {
933 	struct nvme_command c;
934 	union nvme_result res;
935 	int ret;
936 
937 	memset(&c, 0, sizeof(c));
938 	c.features.opcode = nvme_admin_set_features;
939 	c.features.fid = cpu_to_le32(fid);
940 	c.features.dword11 = cpu_to_le32(dword11);
941 
942 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
943 			buffer, buflen, 0, NVME_QID_ANY, 0, 0);
944 	if (ret >= 0 && result)
945 		*result = le32_to_cpu(res.u32);
946 	return ret;
947 }
948 
949 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
950 {
951 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
952 	u32 result;
953 	int status, nr_io_queues;
954 
955 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
956 			&result);
957 	if (status < 0)
958 		return status;
959 
960 	/*
961 	 * Degraded controllers might return an error when setting the queue
962 	 * count.  We still want to be able to bring them online and offer
963 	 * access to the admin queue, as that might be only way to fix them up.
964 	 */
965 	if (status > 0) {
966 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
967 		*count = 0;
968 	} else {
969 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
970 		*count = min(*count, nr_io_queues);
971 	}
972 
973 	return 0;
974 }
975 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
976 
977 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
978 {
979 	struct nvme_user_io io;
980 	struct nvme_command c;
981 	unsigned length, meta_len;
982 	void __user *metadata;
983 
984 	if (copy_from_user(&io, uio, sizeof(io)))
985 		return -EFAULT;
986 	if (io.flags)
987 		return -EINVAL;
988 
989 	switch (io.opcode) {
990 	case nvme_cmd_write:
991 	case nvme_cmd_read:
992 	case nvme_cmd_compare:
993 		break;
994 	default:
995 		return -EINVAL;
996 	}
997 
998 	length = (io.nblocks + 1) << ns->lba_shift;
999 	meta_len = (io.nblocks + 1) * ns->ms;
1000 	metadata = (void __user *)(uintptr_t)io.metadata;
1001 
1002 	if (ns->ext) {
1003 		length += meta_len;
1004 		meta_len = 0;
1005 	} else if (meta_len) {
1006 		if ((io.metadata & 3) || !io.metadata)
1007 			return -EINVAL;
1008 	}
1009 
1010 	memset(&c, 0, sizeof(c));
1011 	c.rw.opcode = io.opcode;
1012 	c.rw.flags = io.flags;
1013 	c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1014 	c.rw.slba = cpu_to_le64(io.slba);
1015 	c.rw.length = cpu_to_le16(io.nblocks);
1016 	c.rw.control = cpu_to_le16(io.control);
1017 	c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1018 	c.rw.reftag = cpu_to_le32(io.reftag);
1019 	c.rw.apptag = cpu_to_le16(io.apptag);
1020 	c.rw.appmask = cpu_to_le16(io.appmask);
1021 
1022 	return nvme_submit_user_cmd(ns->queue, &c,
1023 			(void __user *)(uintptr_t)io.addr, length,
1024 			metadata, meta_len, io.slba, NULL, 0);
1025 }
1026 
1027 static u32 nvme_known_admin_effects(u8 opcode)
1028 {
1029 	switch (opcode) {
1030 	case nvme_admin_format_nvm:
1031 		return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1032 					NVME_CMD_EFFECTS_CSE_MASK;
1033 	case nvme_admin_sanitize_nvm:
1034 		return NVME_CMD_EFFECTS_CSE_MASK;
1035 	default:
1036 		break;
1037 	}
1038 	return 0;
1039 }
1040 
1041 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1042 								u8 opcode)
1043 {
1044 	u32 effects = 0;
1045 
1046 	if (ns) {
1047 		if (ctrl->effects)
1048 			effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1049 		if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1050 			dev_warn(ctrl->device,
1051 				 "IO command:%02x has unhandled effects:%08x\n",
1052 				 opcode, effects);
1053 		return 0;
1054 	}
1055 
1056 	if (ctrl->effects)
1057 		effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1058 	else
1059 		effects = nvme_known_admin_effects(opcode);
1060 
1061 	/*
1062 	 * For simplicity, IO to all namespaces is quiesced even if the command
1063 	 * effects say only one namespace is affected.
1064 	 */
1065 	if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1066 		nvme_start_freeze(ctrl);
1067 		nvme_wait_freeze(ctrl);
1068 	}
1069 	return effects;
1070 }
1071 
1072 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1073 {
1074 	struct nvme_ns *ns;
1075 
1076 	mutex_lock(&ctrl->namespaces_mutex);
1077 	list_for_each_entry(ns, &ctrl->namespaces, list) {
1078 		if (ns->disk && nvme_revalidate_disk(ns->disk))
1079 			nvme_ns_remove(ns);
1080 	}
1081 	mutex_unlock(&ctrl->namespaces_mutex);
1082 }
1083 
1084 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1085 {
1086 	/*
1087 	 * Revalidate LBA changes prior to unfreezing. This is necessary to
1088 	 * prevent memory corruption if a logical block size was changed by
1089 	 * this command.
1090 	 */
1091 	if (effects & NVME_CMD_EFFECTS_LBCC)
1092 		nvme_update_formats(ctrl);
1093 	if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK))
1094 		nvme_unfreeze(ctrl);
1095 	if (effects & NVME_CMD_EFFECTS_CCC)
1096 		nvme_init_identify(ctrl);
1097 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1098 		nvme_queue_scan(ctrl);
1099 }
1100 
1101 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1102 			struct nvme_passthru_cmd __user *ucmd)
1103 {
1104 	struct nvme_passthru_cmd cmd;
1105 	struct nvme_command c;
1106 	unsigned timeout = 0;
1107 	u32 effects;
1108 	int status;
1109 
1110 	if (!capable(CAP_SYS_ADMIN))
1111 		return -EACCES;
1112 	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1113 		return -EFAULT;
1114 	if (cmd.flags)
1115 		return -EINVAL;
1116 
1117 	memset(&c, 0, sizeof(c));
1118 	c.common.opcode = cmd.opcode;
1119 	c.common.flags = cmd.flags;
1120 	c.common.nsid = cpu_to_le32(cmd.nsid);
1121 	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1122 	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1123 	c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1124 	c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1125 	c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1126 	c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1127 	c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1128 	c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1129 
1130 	if (cmd.timeout_ms)
1131 		timeout = msecs_to_jiffies(cmd.timeout_ms);
1132 
1133 	effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1134 	status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1135 			(void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1136 			(void __user *)(uintptr_t)cmd.metadata, cmd.metadata,
1137 			0, &cmd.result, timeout);
1138 	nvme_passthru_end(ctrl, effects);
1139 
1140 	if (status >= 0) {
1141 		if (put_user(cmd.result, &ucmd->result))
1142 			return -EFAULT;
1143 	}
1144 
1145 	return status;
1146 }
1147 
1148 /*
1149  * Issue ioctl requests on the first available path.  Note that unlike normal
1150  * block layer requests we will not retry failed request on another controller.
1151  */
1152 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1153 		struct nvme_ns_head **head, int *srcu_idx)
1154 {
1155 #ifdef CONFIG_NVME_MULTIPATH
1156 	if (disk->fops == &nvme_ns_head_ops) {
1157 		*head = disk->private_data;
1158 		*srcu_idx = srcu_read_lock(&(*head)->srcu);
1159 		return nvme_find_path(*head);
1160 	}
1161 #endif
1162 	*head = NULL;
1163 	*srcu_idx = -1;
1164 	return disk->private_data;
1165 }
1166 
1167 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1168 {
1169 	if (head)
1170 		srcu_read_unlock(&head->srcu, idx);
1171 }
1172 
1173 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1174 {
1175 	switch (cmd) {
1176 	case NVME_IOCTL_ID:
1177 		force_successful_syscall_return();
1178 		return ns->head->ns_id;
1179 	case NVME_IOCTL_ADMIN_CMD:
1180 		return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1181 	case NVME_IOCTL_IO_CMD:
1182 		return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1183 	case NVME_IOCTL_SUBMIT_IO:
1184 		return nvme_submit_io(ns, (void __user *)arg);
1185 	default:
1186 #ifdef CONFIG_NVM
1187 		if (ns->ndev)
1188 			return nvme_nvm_ioctl(ns, cmd, arg);
1189 #endif
1190 		if (is_sed_ioctl(cmd))
1191 			return sed_ioctl(ns->ctrl->opal_dev, cmd,
1192 					 (void __user *) arg);
1193 		return -ENOTTY;
1194 	}
1195 }
1196 
1197 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1198 		unsigned int cmd, unsigned long arg)
1199 {
1200 	struct nvme_ns_head *head = NULL;
1201 	struct nvme_ns *ns;
1202 	int srcu_idx, ret;
1203 
1204 	ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1205 	if (unlikely(!ns))
1206 		ret = -EWOULDBLOCK;
1207 	else
1208 		ret = nvme_ns_ioctl(ns, cmd, arg);
1209 	nvme_put_ns_from_disk(head, srcu_idx);
1210 	return ret;
1211 }
1212 
1213 static int nvme_open(struct block_device *bdev, fmode_t mode)
1214 {
1215 	struct nvme_ns *ns = bdev->bd_disk->private_data;
1216 
1217 #ifdef CONFIG_NVME_MULTIPATH
1218 	/* should never be called due to GENHD_FL_HIDDEN */
1219 	if (WARN_ON_ONCE(ns->head->disk))
1220 		return -ENXIO;
1221 #endif
1222 	if (!kref_get_unless_zero(&ns->kref))
1223 		return -ENXIO;
1224 	return 0;
1225 }
1226 
1227 static void nvme_release(struct gendisk *disk, fmode_t mode)
1228 {
1229 	nvme_put_ns(disk->private_data);
1230 }
1231 
1232 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1233 {
1234 	/* some standard values */
1235 	geo->heads = 1 << 6;
1236 	geo->sectors = 1 << 5;
1237 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1238 	return 0;
1239 }
1240 
1241 #ifdef CONFIG_BLK_DEV_INTEGRITY
1242 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1243 {
1244 	struct blk_integrity integrity;
1245 
1246 	memset(&integrity, 0, sizeof(integrity));
1247 	switch (pi_type) {
1248 	case NVME_NS_DPS_PI_TYPE3:
1249 		integrity.profile = &t10_pi_type3_crc;
1250 		integrity.tag_size = sizeof(u16) + sizeof(u32);
1251 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1252 		break;
1253 	case NVME_NS_DPS_PI_TYPE1:
1254 	case NVME_NS_DPS_PI_TYPE2:
1255 		integrity.profile = &t10_pi_type1_crc;
1256 		integrity.tag_size = sizeof(u16);
1257 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1258 		break;
1259 	default:
1260 		integrity.profile = NULL;
1261 		break;
1262 	}
1263 	integrity.tuple_size = ms;
1264 	blk_integrity_register(disk, &integrity);
1265 	blk_queue_max_integrity_segments(disk->queue, 1);
1266 }
1267 #else
1268 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1269 {
1270 }
1271 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1272 
1273 static void nvme_set_chunk_size(struct nvme_ns *ns)
1274 {
1275 	u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1276 	blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1277 }
1278 
1279 static void nvme_config_discard(struct nvme_ctrl *ctrl,
1280 		unsigned stream_alignment, struct request_queue *queue)
1281 {
1282 	u32 size = queue_logical_block_size(queue);
1283 
1284 	if (stream_alignment)
1285 		size *= stream_alignment;
1286 
1287 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1288 			NVME_DSM_MAX_RANGES);
1289 
1290 	queue->limits.discard_alignment = 0;
1291 	queue->limits.discard_granularity = size;
1292 
1293 	blk_queue_max_discard_sectors(queue, UINT_MAX);
1294 	blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1295 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, queue);
1296 
1297 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1298 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1299 }
1300 
1301 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1302 		struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1303 {
1304 	memset(ids, 0, sizeof(*ids));
1305 
1306 	if (ctrl->vs >= NVME_VS(1, 1, 0))
1307 		memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1308 	if (ctrl->vs >= NVME_VS(1, 2, 0))
1309 		memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1310 	if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1311 		 /* Don't treat error as fatal we potentially
1312 		  * already have a NGUID or EUI-64
1313 		  */
1314 		if (nvme_identify_ns_descs(ctrl, nsid, ids))
1315 			dev_warn(ctrl->device,
1316 				 "%s: Identify Descriptors failed\n", __func__);
1317 	}
1318 }
1319 
1320 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1321 {
1322 	return !uuid_is_null(&ids->uuid) ||
1323 		memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1324 		memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1325 }
1326 
1327 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1328 {
1329 	return uuid_equal(&a->uuid, &b->uuid) &&
1330 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1331 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1332 }
1333 
1334 static void nvme_update_disk_info(struct gendisk *disk,
1335 		struct nvme_ns *ns, struct nvme_id_ns *id)
1336 {
1337 	sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1338 	unsigned stream_alignment = 0;
1339 
1340 	if (ns->ctrl->nr_streams && ns->sws && ns->sgs)
1341 		stream_alignment = ns->sws * ns->sgs;
1342 
1343 	blk_mq_freeze_queue(disk->queue);
1344 	blk_integrity_unregister(disk);
1345 
1346 	blk_queue_logical_block_size(disk->queue, 1 << ns->lba_shift);
1347 	if (ns->ms && !ns->ext &&
1348 	    (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1349 		nvme_init_integrity(disk, ns->ms, ns->pi_type);
1350 	if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1351 		capacity = 0;
1352 	set_capacity(disk, capacity);
1353 
1354 	if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1355 		nvme_config_discard(ns->ctrl, stream_alignment, disk->queue);
1356 	blk_mq_unfreeze_queue(disk->queue);
1357 }
1358 
1359 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1360 {
1361 	struct nvme_ns *ns = disk->private_data;
1362 
1363 	/*
1364 	 * If identify namespace failed, use default 512 byte block size so
1365 	 * block layer can use before failing read/write for 0 capacity.
1366 	 */
1367 	ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1368 	if (ns->lba_shift == 0)
1369 		ns->lba_shift = 9;
1370 	ns->noiob = le16_to_cpu(id->noiob);
1371 	ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1372 	ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1373 	/* the PI implementation requires metadata equal t10 pi tuple size */
1374 	if (ns->ms == sizeof(struct t10_pi_tuple))
1375 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1376 	else
1377 		ns->pi_type = 0;
1378 
1379 	if (ns->noiob)
1380 		nvme_set_chunk_size(ns);
1381 	nvme_update_disk_info(disk, ns, id);
1382 #ifdef CONFIG_NVME_MULTIPATH
1383 	if (ns->head->disk)
1384 		nvme_update_disk_info(ns->head->disk, ns, id);
1385 #endif
1386 }
1387 
1388 static int nvme_revalidate_disk(struct gendisk *disk)
1389 {
1390 	struct nvme_ns *ns = disk->private_data;
1391 	struct nvme_ctrl *ctrl = ns->ctrl;
1392 	struct nvme_id_ns *id;
1393 	struct nvme_ns_ids ids;
1394 	int ret = 0;
1395 
1396 	if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1397 		set_capacity(disk, 0);
1398 		return -ENODEV;
1399 	}
1400 
1401 	id = nvme_identify_ns(ctrl, ns->head->ns_id);
1402 	if (!id)
1403 		return -ENODEV;
1404 
1405 	if (id->ncap == 0) {
1406 		ret = -ENODEV;
1407 		goto out;
1408 	}
1409 
1410 	__nvme_revalidate_disk(disk, id);
1411 	nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1412 	if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1413 		dev_err(ctrl->device,
1414 			"identifiers changed for nsid %d\n", ns->head->ns_id);
1415 		ret = -ENODEV;
1416 	}
1417 
1418 out:
1419 	kfree(id);
1420 	return ret;
1421 }
1422 
1423 static char nvme_pr_type(enum pr_type type)
1424 {
1425 	switch (type) {
1426 	case PR_WRITE_EXCLUSIVE:
1427 		return 1;
1428 	case PR_EXCLUSIVE_ACCESS:
1429 		return 2;
1430 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
1431 		return 3;
1432 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1433 		return 4;
1434 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
1435 		return 5;
1436 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1437 		return 6;
1438 	default:
1439 		return 0;
1440 	}
1441 };
1442 
1443 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1444 				u64 key, u64 sa_key, u8 op)
1445 {
1446 	struct nvme_ns_head *head = NULL;
1447 	struct nvme_ns *ns;
1448 	struct nvme_command c;
1449 	int srcu_idx, ret;
1450 	u8 data[16] = { 0, };
1451 
1452 	ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1453 	if (unlikely(!ns))
1454 		return -EWOULDBLOCK;
1455 
1456 	put_unaligned_le64(key, &data[0]);
1457 	put_unaligned_le64(sa_key, &data[8]);
1458 
1459 	memset(&c, 0, sizeof(c));
1460 	c.common.opcode = op;
1461 	c.common.nsid = cpu_to_le32(ns->head->ns_id);
1462 	c.common.cdw10[0] = cpu_to_le32(cdw10);
1463 
1464 	ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1465 	nvme_put_ns_from_disk(head, srcu_idx);
1466 	return ret;
1467 }
1468 
1469 static int nvme_pr_register(struct block_device *bdev, u64 old,
1470 		u64 new, unsigned flags)
1471 {
1472 	u32 cdw10;
1473 
1474 	if (flags & ~PR_FL_IGNORE_KEY)
1475 		return -EOPNOTSUPP;
1476 
1477 	cdw10 = old ? 2 : 0;
1478 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1479 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1480 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1481 }
1482 
1483 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1484 		enum pr_type type, unsigned flags)
1485 {
1486 	u32 cdw10;
1487 
1488 	if (flags & ~PR_FL_IGNORE_KEY)
1489 		return -EOPNOTSUPP;
1490 
1491 	cdw10 = nvme_pr_type(type) << 8;
1492 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1493 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1494 }
1495 
1496 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1497 		enum pr_type type, bool abort)
1498 {
1499 	u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1500 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1501 }
1502 
1503 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1504 {
1505 	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1506 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1507 }
1508 
1509 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1510 {
1511 	u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1512 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1513 }
1514 
1515 static const struct pr_ops nvme_pr_ops = {
1516 	.pr_register	= nvme_pr_register,
1517 	.pr_reserve	= nvme_pr_reserve,
1518 	.pr_release	= nvme_pr_release,
1519 	.pr_preempt	= nvme_pr_preempt,
1520 	.pr_clear	= nvme_pr_clear,
1521 };
1522 
1523 #ifdef CONFIG_BLK_SED_OPAL
1524 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1525 		bool send)
1526 {
1527 	struct nvme_ctrl *ctrl = data;
1528 	struct nvme_command cmd;
1529 
1530 	memset(&cmd, 0, sizeof(cmd));
1531 	if (send)
1532 		cmd.common.opcode = nvme_admin_security_send;
1533 	else
1534 		cmd.common.opcode = nvme_admin_security_recv;
1535 	cmd.common.nsid = 0;
1536 	cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1537 	cmd.common.cdw10[1] = cpu_to_le32(len);
1538 
1539 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1540 				      ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1541 }
1542 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1543 #endif /* CONFIG_BLK_SED_OPAL */
1544 
1545 static const struct block_device_operations nvme_fops = {
1546 	.owner		= THIS_MODULE,
1547 	.ioctl		= nvme_ioctl,
1548 	.compat_ioctl	= nvme_ioctl,
1549 	.open		= nvme_open,
1550 	.release	= nvme_release,
1551 	.getgeo		= nvme_getgeo,
1552 	.revalidate_disk= nvme_revalidate_disk,
1553 	.pr_ops		= &nvme_pr_ops,
1554 };
1555 
1556 #ifdef CONFIG_NVME_MULTIPATH
1557 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1558 {
1559 	struct nvme_ns_head *head = bdev->bd_disk->private_data;
1560 
1561 	if (!kref_get_unless_zero(&head->ref))
1562 		return -ENXIO;
1563 	return 0;
1564 }
1565 
1566 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1567 {
1568 	nvme_put_ns_head(disk->private_data);
1569 }
1570 
1571 const struct block_device_operations nvme_ns_head_ops = {
1572 	.owner		= THIS_MODULE,
1573 	.open		= nvme_ns_head_open,
1574 	.release	= nvme_ns_head_release,
1575 	.ioctl		= nvme_ioctl,
1576 	.compat_ioctl	= nvme_ioctl,
1577 	.getgeo		= nvme_getgeo,
1578 	.pr_ops		= &nvme_pr_ops,
1579 };
1580 #endif /* CONFIG_NVME_MULTIPATH */
1581 
1582 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1583 {
1584 	unsigned long timeout =
1585 		((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1586 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1587 	int ret;
1588 
1589 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1590 		if (csts == ~0)
1591 			return -ENODEV;
1592 		if ((csts & NVME_CSTS_RDY) == bit)
1593 			break;
1594 
1595 		msleep(100);
1596 		if (fatal_signal_pending(current))
1597 			return -EINTR;
1598 		if (time_after(jiffies, timeout)) {
1599 			dev_err(ctrl->device,
1600 				"Device not ready; aborting %s\n", enabled ?
1601 						"initialisation" : "reset");
1602 			return -ENODEV;
1603 		}
1604 	}
1605 
1606 	return ret;
1607 }
1608 
1609 /*
1610  * If the device has been passed off to us in an enabled state, just clear
1611  * the enabled bit.  The spec says we should set the 'shutdown notification
1612  * bits', but doing so may cause the device to complete commands to the
1613  * admin queue ... and we don't know what memory that might be pointing at!
1614  */
1615 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1616 {
1617 	int ret;
1618 
1619 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1620 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1621 
1622 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1623 	if (ret)
1624 		return ret;
1625 
1626 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1627 		msleep(NVME_QUIRK_DELAY_AMOUNT);
1628 
1629 	return nvme_wait_ready(ctrl, cap, false);
1630 }
1631 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1632 
1633 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1634 {
1635 	/*
1636 	 * Default to a 4K page size, with the intention to update this
1637 	 * path in the future to accomodate architectures with differing
1638 	 * kernel and IO page sizes.
1639 	 */
1640 	unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1641 	int ret;
1642 
1643 	if (page_shift < dev_page_min) {
1644 		dev_err(ctrl->device,
1645 			"Minimum device page size %u too large for host (%u)\n",
1646 			1 << dev_page_min, 1 << page_shift);
1647 		return -ENODEV;
1648 	}
1649 
1650 	ctrl->page_size = 1 << page_shift;
1651 
1652 	ctrl->ctrl_config = NVME_CC_CSS_NVM;
1653 	ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1654 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1655 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1656 	ctrl->ctrl_config |= NVME_CC_ENABLE;
1657 
1658 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1659 	if (ret)
1660 		return ret;
1661 	return nvme_wait_ready(ctrl, cap, true);
1662 }
1663 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1664 
1665 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1666 {
1667 	unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1668 	u32 csts;
1669 	int ret;
1670 
1671 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1672 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1673 
1674 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1675 	if (ret)
1676 		return ret;
1677 
1678 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1679 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1680 			break;
1681 
1682 		msleep(100);
1683 		if (fatal_signal_pending(current))
1684 			return -EINTR;
1685 		if (time_after(jiffies, timeout)) {
1686 			dev_err(ctrl->device,
1687 				"Device shutdown incomplete; abort shutdown\n");
1688 			return -ENODEV;
1689 		}
1690 	}
1691 
1692 	return ret;
1693 }
1694 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1695 
1696 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1697 		struct request_queue *q)
1698 {
1699 	bool vwc = false;
1700 
1701 	if (ctrl->max_hw_sectors) {
1702 		u32 max_segments =
1703 			(ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1704 
1705 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1706 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1707 	}
1708 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1709 	    is_power_of_2(ctrl->max_hw_sectors))
1710 		blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1711 	blk_queue_virt_boundary(q, ctrl->page_size - 1);
1712 	if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1713 		vwc = true;
1714 	blk_queue_write_cache(q, vwc, vwc);
1715 }
1716 
1717 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1718 {
1719 	__le64 ts;
1720 	int ret;
1721 
1722 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1723 		return 0;
1724 
1725 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1726 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1727 			NULL);
1728 	if (ret)
1729 		dev_warn_once(ctrl->device,
1730 			"could not set timestamp (%d)\n", ret);
1731 	return ret;
1732 }
1733 
1734 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1735 {
1736 	/*
1737 	 * APST (Autonomous Power State Transition) lets us program a
1738 	 * table of power state transitions that the controller will
1739 	 * perform automatically.  We configure it with a simple
1740 	 * heuristic: we are willing to spend at most 2% of the time
1741 	 * transitioning between power states.  Therefore, when running
1742 	 * in any given state, we will enter the next lower-power
1743 	 * non-operational state after waiting 50 * (enlat + exlat)
1744 	 * microseconds, as long as that state's exit latency is under
1745 	 * the requested maximum latency.
1746 	 *
1747 	 * We will not autonomously enter any non-operational state for
1748 	 * which the total latency exceeds ps_max_latency_us.  Users
1749 	 * can set ps_max_latency_us to zero to turn off APST.
1750 	 */
1751 
1752 	unsigned apste;
1753 	struct nvme_feat_auto_pst *table;
1754 	u64 max_lat_us = 0;
1755 	int max_ps = -1;
1756 	int ret;
1757 
1758 	/*
1759 	 * If APST isn't supported or if we haven't been initialized yet,
1760 	 * then don't do anything.
1761 	 */
1762 	if (!ctrl->apsta)
1763 		return 0;
1764 
1765 	if (ctrl->npss > 31) {
1766 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1767 		return 0;
1768 	}
1769 
1770 	table = kzalloc(sizeof(*table), GFP_KERNEL);
1771 	if (!table)
1772 		return 0;
1773 
1774 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1775 		/* Turn off APST. */
1776 		apste = 0;
1777 		dev_dbg(ctrl->device, "APST disabled\n");
1778 	} else {
1779 		__le64 target = cpu_to_le64(0);
1780 		int state;
1781 
1782 		/*
1783 		 * Walk through all states from lowest- to highest-power.
1784 		 * According to the spec, lower-numbered states use more
1785 		 * power.  NPSS, despite the name, is the index of the
1786 		 * lowest-power state, not the number of states.
1787 		 */
1788 		for (state = (int)ctrl->npss; state >= 0; state--) {
1789 			u64 total_latency_us, exit_latency_us, transition_ms;
1790 
1791 			if (target)
1792 				table->entries[state] = target;
1793 
1794 			/*
1795 			 * Don't allow transitions to the deepest state
1796 			 * if it's quirked off.
1797 			 */
1798 			if (state == ctrl->npss &&
1799 			    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1800 				continue;
1801 
1802 			/*
1803 			 * Is this state a useful non-operational state for
1804 			 * higher-power states to autonomously transition to?
1805 			 */
1806 			if (!(ctrl->psd[state].flags &
1807 			      NVME_PS_FLAGS_NON_OP_STATE))
1808 				continue;
1809 
1810 			exit_latency_us =
1811 				(u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1812 			if (exit_latency_us > ctrl->ps_max_latency_us)
1813 				continue;
1814 
1815 			total_latency_us =
1816 				exit_latency_us +
1817 				le32_to_cpu(ctrl->psd[state].entry_lat);
1818 
1819 			/*
1820 			 * This state is good.  Use it as the APST idle
1821 			 * target for higher power states.
1822 			 */
1823 			transition_ms = total_latency_us + 19;
1824 			do_div(transition_ms, 20);
1825 			if (transition_ms > (1 << 24) - 1)
1826 				transition_ms = (1 << 24) - 1;
1827 
1828 			target = cpu_to_le64((state << 3) |
1829 					     (transition_ms << 8));
1830 
1831 			if (max_ps == -1)
1832 				max_ps = state;
1833 
1834 			if (total_latency_us > max_lat_us)
1835 				max_lat_us = total_latency_us;
1836 		}
1837 
1838 		apste = 1;
1839 
1840 		if (max_ps == -1) {
1841 			dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1842 		} else {
1843 			dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1844 				max_ps, max_lat_us, (int)sizeof(*table), table);
1845 		}
1846 	}
1847 
1848 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1849 				table, sizeof(*table), NULL);
1850 	if (ret)
1851 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1852 
1853 	kfree(table);
1854 	return ret;
1855 }
1856 
1857 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1858 {
1859 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1860 	u64 latency;
1861 
1862 	switch (val) {
1863 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1864 	case PM_QOS_LATENCY_ANY:
1865 		latency = U64_MAX;
1866 		break;
1867 
1868 	default:
1869 		latency = val;
1870 	}
1871 
1872 	if (ctrl->ps_max_latency_us != latency) {
1873 		ctrl->ps_max_latency_us = latency;
1874 		nvme_configure_apst(ctrl);
1875 	}
1876 }
1877 
1878 struct nvme_core_quirk_entry {
1879 	/*
1880 	 * NVMe model and firmware strings are padded with spaces.  For
1881 	 * simplicity, strings in the quirk table are padded with NULLs
1882 	 * instead.
1883 	 */
1884 	u16 vid;
1885 	const char *mn;
1886 	const char *fr;
1887 	unsigned long quirks;
1888 };
1889 
1890 static const struct nvme_core_quirk_entry core_quirks[] = {
1891 	{
1892 		/*
1893 		 * This Toshiba device seems to die using any APST states.  See:
1894 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1895 		 */
1896 		.vid = 0x1179,
1897 		.mn = "THNSF5256GPUK TOSHIBA",
1898 		.quirks = NVME_QUIRK_NO_APST,
1899 	}
1900 };
1901 
1902 /* match is null-terminated but idstr is space-padded. */
1903 static bool string_matches(const char *idstr, const char *match, size_t len)
1904 {
1905 	size_t matchlen;
1906 
1907 	if (!match)
1908 		return true;
1909 
1910 	matchlen = strlen(match);
1911 	WARN_ON_ONCE(matchlen > len);
1912 
1913 	if (memcmp(idstr, match, matchlen))
1914 		return false;
1915 
1916 	for (; matchlen < len; matchlen++)
1917 		if (idstr[matchlen] != ' ')
1918 			return false;
1919 
1920 	return true;
1921 }
1922 
1923 static bool quirk_matches(const struct nvme_id_ctrl *id,
1924 			  const struct nvme_core_quirk_entry *q)
1925 {
1926 	return q->vid == le16_to_cpu(id->vid) &&
1927 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1928 		string_matches(id->fr, q->fr, sizeof(id->fr));
1929 }
1930 
1931 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
1932 		struct nvme_id_ctrl *id)
1933 {
1934 	size_t nqnlen;
1935 	int off;
1936 
1937 	nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
1938 	if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
1939 		strncpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
1940 		return;
1941 	}
1942 
1943 	if (ctrl->vs >= NVME_VS(1, 2, 1))
1944 		dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
1945 
1946 	/* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
1947 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
1948 			"nqn.2014.08.org.nvmexpress:%4x%4x",
1949 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
1950 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
1951 	off += sizeof(id->sn);
1952 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
1953 	off += sizeof(id->mn);
1954 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
1955 }
1956 
1957 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
1958 {
1959 	ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
1960 	kfree(subsys);
1961 }
1962 
1963 static void nvme_release_subsystem(struct device *dev)
1964 {
1965 	__nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
1966 }
1967 
1968 static void nvme_destroy_subsystem(struct kref *ref)
1969 {
1970 	struct nvme_subsystem *subsys =
1971 			container_of(ref, struct nvme_subsystem, ref);
1972 
1973 	mutex_lock(&nvme_subsystems_lock);
1974 	list_del(&subsys->entry);
1975 	mutex_unlock(&nvme_subsystems_lock);
1976 
1977 	ida_destroy(&subsys->ns_ida);
1978 	device_del(&subsys->dev);
1979 	put_device(&subsys->dev);
1980 }
1981 
1982 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
1983 {
1984 	kref_put(&subsys->ref, nvme_destroy_subsystem);
1985 }
1986 
1987 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
1988 {
1989 	struct nvme_subsystem *subsys;
1990 
1991 	lockdep_assert_held(&nvme_subsystems_lock);
1992 
1993 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
1994 		if (strcmp(subsys->subnqn, subsysnqn))
1995 			continue;
1996 		if (!kref_get_unless_zero(&subsys->ref))
1997 			continue;
1998 		return subsys;
1999 	}
2000 
2001 	return NULL;
2002 }
2003 
2004 #define SUBSYS_ATTR_RO(_name, _mode, _show)			\
2005 	struct device_attribute subsys_attr_##_name = \
2006 		__ATTR(_name, _mode, _show, NULL)
2007 
2008 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2009 				    struct device_attribute *attr,
2010 				    char *buf)
2011 {
2012 	struct nvme_subsystem *subsys =
2013 		container_of(dev, struct nvme_subsystem, dev);
2014 
2015 	return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2016 }
2017 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2018 
2019 #define nvme_subsys_show_str_function(field)				\
2020 static ssize_t subsys_##field##_show(struct device *dev,		\
2021 			    struct device_attribute *attr, char *buf)	\
2022 {									\
2023 	struct nvme_subsystem *subsys =					\
2024 		container_of(dev, struct nvme_subsystem, dev);		\
2025 	return sprintf(buf, "%.*s\n",					\
2026 		       (int)sizeof(subsys->field), subsys->field);	\
2027 }									\
2028 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2029 
2030 nvme_subsys_show_str_function(model);
2031 nvme_subsys_show_str_function(serial);
2032 nvme_subsys_show_str_function(firmware_rev);
2033 
2034 static struct attribute *nvme_subsys_attrs[] = {
2035 	&subsys_attr_model.attr,
2036 	&subsys_attr_serial.attr,
2037 	&subsys_attr_firmware_rev.attr,
2038 	&subsys_attr_subsysnqn.attr,
2039 	NULL,
2040 };
2041 
2042 static struct attribute_group nvme_subsys_attrs_group = {
2043 	.attrs = nvme_subsys_attrs,
2044 };
2045 
2046 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2047 	&nvme_subsys_attrs_group,
2048 	NULL,
2049 };
2050 
2051 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2052 {
2053 	struct nvme_subsystem *subsys, *found;
2054 	int ret;
2055 
2056 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2057 	if (!subsys)
2058 		return -ENOMEM;
2059 	ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2060 	if (ret < 0) {
2061 		kfree(subsys);
2062 		return ret;
2063 	}
2064 	subsys->instance = ret;
2065 	mutex_init(&subsys->lock);
2066 	kref_init(&subsys->ref);
2067 	INIT_LIST_HEAD(&subsys->ctrls);
2068 	INIT_LIST_HEAD(&subsys->nsheads);
2069 	nvme_init_subnqn(subsys, ctrl, id);
2070 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2071 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2072 	memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2073 	subsys->vendor_id = le16_to_cpu(id->vid);
2074 	subsys->cmic = id->cmic;
2075 
2076 	subsys->dev.class = nvme_subsys_class;
2077 	subsys->dev.release = nvme_release_subsystem;
2078 	subsys->dev.groups = nvme_subsys_attrs_groups;
2079 	dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2080 	device_initialize(&subsys->dev);
2081 
2082 	mutex_lock(&nvme_subsystems_lock);
2083 	found = __nvme_find_get_subsystem(subsys->subnqn);
2084 	if (found) {
2085 		/*
2086 		 * Verify that the subsystem actually supports multiple
2087 		 * controllers, else bail out.
2088 		 */
2089 		if (!(id->cmic & (1 << 1))) {
2090 			dev_err(ctrl->device,
2091 				"ignoring ctrl due to duplicate subnqn (%s).\n",
2092 				found->subnqn);
2093 			nvme_put_subsystem(found);
2094 			ret = -EINVAL;
2095 			goto out_unlock;
2096 		}
2097 
2098 		__nvme_release_subsystem(subsys);
2099 		subsys = found;
2100 	} else {
2101 		ret = device_add(&subsys->dev);
2102 		if (ret) {
2103 			dev_err(ctrl->device,
2104 				"failed to register subsystem device.\n");
2105 			goto out_unlock;
2106 		}
2107 		ida_init(&subsys->ns_ida);
2108 		list_add_tail(&subsys->entry, &nvme_subsystems);
2109 	}
2110 
2111 	ctrl->subsys = subsys;
2112 	mutex_unlock(&nvme_subsystems_lock);
2113 
2114 	if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2115 			dev_name(ctrl->device))) {
2116 		dev_err(ctrl->device,
2117 			"failed to create sysfs link from subsystem.\n");
2118 		/* the transport driver will eventually put the subsystem */
2119 		return -EINVAL;
2120 	}
2121 
2122 	mutex_lock(&subsys->lock);
2123 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2124 	mutex_unlock(&subsys->lock);
2125 
2126 	return 0;
2127 
2128 out_unlock:
2129 	mutex_unlock(&nvme_subsystems_lock);
2130 	put_device(&subsys->dev);
2131 	return ret;
2132 }
2133 
2134 static int nvme_get_log(struct nvme_ctrl *ctrl, u8 log_page, void *log,
2135 			size_t size)
2136 {
2137 	struct nvme_command c = { };
2138 
2139 	c.common.opcode = nvme_admin_get_log_page;
2140 	c.common.nsid = cpu_to_le32(NVME_NSID_ALL);
2141 	c.common.cdw10[0] = nvme_get_log_dw10(log_page, size);
2142 
2143 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2144 }
2145 
2146 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2147 {
2148 	int ret;
2149 
2150 	if (!ctrl->effects)
2151 		ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2152 
2153 	if (!ctrl->effects)
2154 		return 0;
2155 
2156 	ret = nvme_get_log(ctrl, NVME_LOG_CMD_EFFECTS, ctrl->effects,
2157 					sizeof(*ctrl->effects));
2158 	if (ret) {
2159 		kfree(ctrl->effects);
2160 		ctrl->effects = NULL;
2161 	}
2162 	return ret;
2163 }
2164 
2165 /*
2166  * Initialize the cached copies of the Identify data and various controller
2167  * register in our nvme_ctrl structure.  This should be called as soon as
2168  * the admin queue is fully up and running.
2169  */
2170 int nvme_init_identify(struct nvme_ctrl *ctrl)
2171 {
2172 	struct nvme_id_ctrl *id;
2173 	u64 cap;
2174 	int ret, page_shift;
2175 	u32 max_hw_sectors;
2176 	bool prev_apst_enabled;
2177 
2178 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2179 	if (ret) {
2180 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2181 		return ret;
2182 	}
2183 
2184 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2185 	if (ret) {
2186 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2187 		return ret;
2188 	}
2189 	page_shift = NVME_CAP_MPSMIN(cap) + 12;
2190 
2191 	if (ctrl->vs >= NVME_VS(1, 1, 0))
2192 		ctrl->subsystem = NVME_CAP_NSSRC(cap);
2193 
2194 	ret = nvme_identify_ctrl(ctrl, &id);
2195 	if (ret) {
2196 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2197 		return -EIO;
2198 	}
2199 
2200 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2201 		ret = nvme_get_effects_log(ctrl);
2202 		if (ret < 0)
2203 			return ret;
2204 	}
2205 
2206 	if (!ctrl->identified) {
2207 		int i;
2208 
2209 		ret = nvme_init_subsystem(ctrl, id);
2210 		if (ret)
2211 			goto out_free;
2212 
2213 		/*
2214 		 * Check for quirks.  Quirk can depend on firmware version,
2215 		 * so, in principle, the set of quirks present can change
2216 		 * across a reset.  As a possible future enhancement, we
2217 		 * could re-scan for quirks every time we reinitialize
2218 		 * the device, but we'd have to make sure that the driver
2219 		 * behaves intelligently if the quirks change.
2220 		 */
2221 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2222 			if (quirk_matches(id, &core_quirks[i]))
2223 				ctrl->quirks |= core_quirks[i].quirks;
2224 		}
2225 	}
2226 
2227 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2228 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2229 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2230 	}
2231 
2232 	ctrl->oacs = le16_to_cpu(id->oacs);
2233 	ctrl->oncs = le16_to_cpup(&id->oncs);
2234 	atomic_set(&ctrl->abort_limit, id->acl + 1);
2235 	ctrl->vwc = id->vwc;
2236 	ctrl->cntlid = le16_to_cpup(&id->cntlid);
2237 	if (id->mdts)
2238 		max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2239 	else
2240 		max_hw_sectors = UINT_MAX;
2241 	ctrl->max_hw_sectors =
2242 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2243 
2244 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
2245 	ctrl->sgls = le32_to_cpu(id->sgls);
2246 	ctrl->kas = le16_to_cpu(id->kas);
2247 
2248 	if (id->rtd3e) {
2249 		/* us -> s */
2250 		u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2251 
2252 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2253 						 shutdown_timeout, 60);
2254 
2255 		if (ctrl->shutdown_timeout != shutdown_timeout)
2256 			dev_warn(ctrl->device,
2257 				 "Shutdown timeout set to %u seconds\n",
2258 				 ctrl->shutdown_timeout);
2259 	} else
2260 		ctrl->shutdown_timeout = shutdown_timeout;
2261 
2262 	ctrl->npss = id->npss;
2263 	ctrl->apsta = id->apsta;
2264 	prev_apst_enabled = ctrl->apst_enabled;
2265 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2266 		if (force_apst && id->apsta) {
2267 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2268 			ctrl->apst_enabled = true;
2269 		} else {
2270 			ctrl->apst_enabled = false;
2271 		}
2272 	} else {
2273 		ctrl->apst_enabled = id->apsta;
2274 	}
2275 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2276 
2277 	if (ctrl->ops->flags & NVME_F_FABRICS) {
2278 		ctrl->icdoff = le16_to_cpu(id->icdoff);
2279 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2280 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2281 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2282 
2283 		/*
2284 		 * In fabrics we need to verify the cntlid matches the
2285 		 * admin connect
2286 		 */
2287 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2288 			ret = -EINVAL;
2289 			goto out_free;
2290 		}
2291 
2292 		if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2293 			dev_err(ctrl->device,
2294 				"keep-alive support is mandatory for fabrics\n");
2295 			ret = -EINVAL;
2296 			goto out_free;
2297 		}
2298 	} else {
2299 		ctrl->cntlid = le16_to_cpu(id->cntlid);
2300 		ctrl->hmpre = le32_to_cpu(id->hmpre);
2301 		ctrl->hmmin = le32_to_cpu(id->hmmin);
2302 		ctrl->hmminds = le32_to_cpu(id->hmminds);
2303 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2304 	}
2305 
2306 	kfree(id);
2307 
2308 	if (ctrl->apst_enabled && !prev_apst_enabled)
2309 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
2310 	else if (!ctrl->apst_enabled && prev_apst_enabled)
2311 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
2312 
2313 	ret = nvme_configure_apst(ctrl);
2314 	if (ret < 0)
2315 		return ret;
2316 
2317 	ret = nvme_configure_timestamp(ctrl);
2318 	if (ret < 0)
2319 		return ret;
2320 
2321 	ret = nvme_configure_directives(ctrl);
2322 	if (ret < 0)
2323 		return ret;
2324 
2325 	ctrl->identified = true;
2326 
2327 	return 0;
2328 
2329 out_free:
2330 	kfree(id);
2331 	return ret;
2332 }
2333 EXPORT_SYMBOL_GPL(nvme_init_identify);
2334 
2335 static int nvme_dev_open(struct inode *inode, struct file *file)
2336 {
2337 	struct nvme_ctrl *ctrl =
2338 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2339 
2340 	if (ctrl->state != NVME_CTRL_LIVE)
2341 		return -EWOULDBLOCK;
2342 	file->private_data = ctrl;
2343 	return 0;
2344 }
2345 
2346 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2347 {
2348 	struct nvme_ns *ns;
2349 	int ret;
2350 
2351 	mutex_lock(&ctrl->namespaces_mutex);
2352 	if (list_empty(&ctrl->namespaces)) {
2353 		ret = -ENOTTY;
2354 		goto out_unlock;
2355 	}
2356 
2357 	ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2358 	if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2359 		dev_warn(ctrl->device,
2360 			"NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2361 		ret = -EINVAL;
2362 		goto out_unlock;
2363 	}
2364 
2365 	dev_warn(ctrl->device,
2366 		"using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2367 	kref_get(&ns->kref);
2368 	mutex_unlock(&ctrl->namespaces_mutex);
2369 
2370 	ret = nvme_user_cmd(ctrl, ns, argp);
2371 	nvme_put_ns(ns);
2372 	return ret;
2373 
2374 out_unlock:
2375 	mutex_unlock(&ctrl->namespaces_mutex);
2376 	return ret;
2377 }
2378 
2379 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2380 		unsigned long arg)
2381 {
2382 	struct nvme_ctrl *ctrl = file->private_data;
2383 	void __user *argp = (void __user *)arg;
2384 
2385 	switch (cmd) {
2386 	case NVME_IOCTL_ADMIN_CMD:
2387 		return nvme_user_cmd(ctrl, NULL, argp);
2388 	case NVME_IOCTL_IO_CMD:
2389 		return nvme_dev_user_cmd(ctrl, argp);
2390 	case NVME_IOCTL_RESET:
2391 		dev_warn(ctrl->device, "resetting controller\n");
2392 		return nvme_reset_ctrl_sync(ctrl);
2393 	case NVME_IOCTL_SUBSYS_RESET:
2394 		return nvme_reset_subsystem(ctrl);
2395 	case NVME_IOCTL_RESCAN:
2396 		nvme_queue_scan(ctrl);
2397 		return 0;
2398 	default:
2399 		return -ENOTTY;
2400 	}
2401 }
2402 
2403 static const struct file_operations nvme_dev_fops = {
2404 	.owner		= THIS_MODULE,
2405 	.open		= nvme_dev_open,
2406 	.unlocked_ioctl	= nvme_dev_ioctl,
2407 	.compat_ioctl	= nvme_dev_ioctl,
2408 };
2409 
2410 static ssize_t nvme_sysfs_reset(struct device *dev,
2411 				struct device_attribute *attr, const char *buf,
2412 				size_t count)
2413 {
2414 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2415 	int ret;
2416 
2417 	ret = nvme_reset_ctrl_sync(ctrl);
2418 	if (ret < 0)
2419 		return ret;
2420 	return count;
2421 }
2422 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2423 
2424 static ssize_t nvme_sysfs_rescan(struct device *dev,
2425 				struct device_attribute *attr, const char *buf,
2426 				size_t count)
2427 {
2428 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2429 
2430 	nvme_queue_scan(ctrl);
2431 	return count;
2432 }
2433 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2434 
2435 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2436 {
2437 	struct gendisk *disk = dev_to_disk(dev);
2438 
2439 	if (disk->fops == &nvme_fops)
2440 		return nvme_get_ns_from_dev(dev)->head;
2441 	else
2442 		return disk->private_data;
2443 }
2444 
2445 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2446 		char *buf)
2447 {
2448 	struct nvme_ns_head *head = dev_to_ns_head(dev);
2449 	struct nvme_ns_ids *ids = &head->ids;
2450 	struct nvme_subsystem *subsys = head->subsys;
2451 	int serial_len = sizeof(subsys->serial);
2452 	int model_len = sizeof(subsys->model);
2453 
2454 	if (!uuid_is_null(&ids->uuid))
2455 		return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2456 
2457 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2458 		return sprintf(buf, "eui.%16phN\n", ids->nguid);
2459 
2460 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2461 		return sprintf(buf, "eui.%8phN\n", ids->eui64);
2462 
2463 	while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2464 				  subsys->serial[serial_len - 1] == '\0'))
2465 		serial_len--;
2466 	while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2467 				 subsys->model[model_len - 1] == '\0'))
2468 		model_len--;
2469 
2470 	return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2471 		serial_len, subsys->serial, model_len, subsys->model,
2472 		head->ns_id);
2473 }
2474 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
2475 
2476 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2477 		char *buf)
2478 {
2479 	return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2480 }
2481 static DEVICE_ATTR(nguid, S_IRUGO, nguid_show, NULL);
2482 
2483 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2484 		char *buf)
2485 {
2486 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2487 
2488 	/* For backward compatibility expose the NGUID to userspace if
2489 	 * we have no UUID set
2490 	 */
2491 	if (uuid_is_null(&ids->uuid)) {
2492 		printk_ratelimited(KERN_WARNING
2493 				   "No UUID available providing old NGUID\n");
2494 		return sprintf(buf, "%pU\n", ids->nguid);
2495 	}
2496 	return sprintf(buf, "%pU\n", &ids->uuid);
2497 }
2498 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
2499 
2500 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2501 		char *buf)
2502 {
2503 	return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2504 }
2505 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
2506 
2507 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2508 		char *buf)
2509 {
2510 	return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2511 }
2512 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
2513 
2514 static struct attribute *nvme_ns_id_attrs[] = {
2515 	&dev_attr_wwid.attr,
2516 	&dev_attr_uuid.attr,
2517 	&dev_attr_nguid.attr,
2518 	&dev_attr_eui.attr,
2519 	&dev_attr_nsid.attr,
2520 	NULL,
2521 };
2522 
2523 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2524 		struct attribute *a, int n)
2525 {
2526 	struct device *dev = container_of(kobj, struct device, kobj);
2527 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2528 
2529 	if (a == &dev_attr_uuid.attr) {
2530 		if (uuid_is_null(&ids->uuid) &&
2531 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2532 			return 0;
2533 	}
2534 	if (a == &dev_attr_nguid.attr) {
2535 		if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2536 			return 0;
2537 	}
2538 	if (a == &dev_attr_eui.attr) {
2539 		if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2540 			return 0;
2541 	}
2542 	return a->mode;
2543 }
2544 
2545 const struct attribute_group nvme_ns_id_attr_group = {
2546 	.attrs		= nvme_ns_id_attrs,
2547 	.is_visible	= nvme_ns_id_attrs_are_visible,
2548 };
2549 
2550 #define nvme_show_str_function(field)						\
2551 static ssize_t  field##_show(struct device *dev,				\
2552 			    struct device_attribute *attr, char *buf)		\
2553 {										\
2554         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
2555         return sprintf(buf, "%.*s\n",						\
2556 		(int)sizeof(ctrl->subsys->field), ctrl->subsys->field);		\
2557 }										\
2558 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2559 
2560 nvme_show_str_function(model);
2561 nvme_show_str_function(serial);
2562 nvme_show_str_function(firmware_rev);
2563 
2564 #define nvme_show_int_function(field)						\
2565 static ssize_t  field##_show(struct device *dev,				\
2566 			    struct device_attribute *attr, char *buf)		\
2567 {										\
2568         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
2569         return sprintf(buf, "%d\n", ctrl->field);	\
2570 }										\
2571 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2572 
2573 nvme_show_int_function(cntlid);
2574 
2575 static ssize_t nvme_sysfs_delete(struct device *dev,
2576 				struct device_attribute *attr, const char *buf,
2577 				size_t count)
2578 {
2579 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2580 
2581 	if (device_remove_file_self(dev, attr))
2582 		nvme_delete_ctrl_sync(ctrl);
2583 	return count;
2584 }
2585 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2586 
2587 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2588 					 struct device_attribute *attr,
2589 					 char *buf)
2590 {
2591 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2592 
2593 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2594 }
2595 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2596 
2597 static ssize_t nvme_sysfs_show_state(struct device *dev,
2598 				     struct device_attribute *attr,
2599 				     char *buf)
2600 {
2601 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2602 	static const char *const state_name[] = {
2603 		[NVME_CTRL_NEW]		= "new",
2604 		[NVME_CTRL_LIVE]	= "live",
2605 		[NVME_CTRL_RESETTING]	= "resetting",
2606 		[NVME_CTRL_RECONNECTING]= "reconnecting",
2607 		[NVME_CTRL_DELETING]	= "deleting",
2608 		[NVME_CTRL_DEAD]	= "dead",
2609 	};
2610 
2611 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2612 	    state_name[ctrl->state])
2613 		return sprintf(buf, "%s\n", state_name[ctrl->state]);
2614 
2615 	return sprintf(buf, "unknown state\n");
2616 }
2617 
2618 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2619 
2620 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2621 					 struct device_attribute *attr,
2622 					 char *buf)
2623 {
2624 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2625 
2626 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2627 }
2628 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2629 
2630 static ssize_t nvme_sysfs_show_address(struct device *dev,
2631 					 struct device_attribute *attr,
2632 					 char *buf)
2633 {
2634 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2635 
2636 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2637 }
2638 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2639 
2640 static struct attribute *nvme_dev_attrs[] = {
2641 	&dev_attr_reset_controller.attr,
2642 	&dev_attr_rescan_controller.attr,
2643 	&dev_attr_model.attr,
2644 	&dev_attr_serial.attr,
2645 	&dev_attr_firmware_rev.attr,
2646 	&dev_attr_cntlid.attr,
2647 	&dev_attr_delete_controller.attr,
2648 	&dev_attr_transport.attr,
2649 	&dev_attr_subsysnqn.attr,
2650 	&dev_attr_address.attr,
2651 	&dev_attr_state.attr,
2652 	NULL
2653 };
2654 
2655 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2656 		struct attribute *a, int n)
2657 {
2658 	struct device *dev = container_of(kobj, struct device, kobj);
2659 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2660 
2661 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2662 		return 0;
2663 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2664 		return 0;
2665 
2666 	return a->mode;
2667 }
2668 
2669 static struct attribute_group nvme_dev_attrs_group = {
2670 	.attrs		= nvme_dev_attrs,
2671 	.is_visible	= nvme_dev_attrs_are_visible,
2672 };
2673 
2674 static const struct attribute_group *nvme_dev_attr_groups[] = {
2675 	&nvme_dev_attrs_group,
2676 	NULL,
2677 };
2678 
2679 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2680 		unsigned nsid)
2681 {
2682 	struct nvme_ns_head *h;
2683 
2684 	lockdep_assert_held(&subsys->lock);
2685 
2686 	list_for_each_entry(h, &subsys->nsheads, entry) {
2687 		if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2688 			return h;
2689 	}
2690 
2691 	return NULL;
2692 }
2693 
2694 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2695 		struct nvme_ns_head *new)
2696 {
2697 	struct nvme_ns_head *h;
2698 
2699 	lockdep_assert_held(&subsys->lock);
2700 
2701 	list_for_each_entry(h, &subsys->nsheads, entry) {
2702 		if (nvme_ns_ids_valid(&new->ids) &&
2703 		    nvme_ns_ids_equal(&new->ids, &h->ids))
2704 			return -EINVAL;
2705 	}
2706 
2707 	return 0;
2708 }
2709 
2710 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2711 		unsigned nsid, struct nvme_id_ns *id)
2712 {
2713 	struct nvme_ns_head *head;
2714 	int ret = -ENOMEM;
2715 
2716 	head = kzalloc(sizeof(*head), GFP_KERNEL);
2717 	if (!head)
2718 		goto out;
2719 	ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
2720 	if (ret < 0)
2721 		goto out_free_head;
2722 	head->instance = ret;
2723 	INIT_LIST_HEAD(&head->list);
2724 	init_srcu_struct(&head->srcu);
2725 	head->subsys = ctrl->subsys;
2726 	head->ns_id = nsid;
2727 	kref_init(&head->ref);
2728 
2729 	nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
2730 
2731 	ret = __nvme_check_ids(ctrl->subsys, head);
2732 	if (ret) {
2733 		dev_err(ctrl->device,
2734 			"duplicate IDs for nsid %d\n", nsid);
2735 		goto out_cleanup_srcu;
2736 	}
2737 
2738 	ret = nvme_mpath_alloc_disk(ctrl, head);
2739 	if (ret)
2740 		goto out_cleanup_srcu;
2741 
2742 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
2743 	return head;
2744 out_cleanup_srcu:
2745 	cleanup_srcu_struct(&head->srcu);
2746 	ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
2747 out_free_head:
2748 	kfree(head);
2749 out:
2750 	return ERR_PTR(ret);
2751 }
2752 
2753 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
2754 		struct nvme_id_ns *id, bool *new)
2755 {
2756 	struct nvme_ctrl *ctrl = ns->ctrl;
2757 	bool is_shared = id->nmic & (1 << 0);
2758 	struct nvme_ns_head *head = NULL;
2759 	int ret = 0;
2760 
2761 	mutex_lock(&ctrl->subsys->lock);
2762 	if (is_shared)
2763 		head = __nvme_find_ns_head(ctrl->subsys, nsid);
2764 	if (!head) {
2765 		head = nvme_alloc_ns_head(ctrl, nsid, id);
2766 		if (IS_ERR(head)) {
2767 			ret = PTR_ERR(head);
2768 			goto out_unlock;
2769 		}
2770 
2771 		*new = true;
2772 	} else {
2773 		struct nvme_ns_ids ids;
2774 
2775 		nvme_report_ns_ids(ctrl, nsid, id, &ids);
2776 		if (!nvme_ns_ids_equal(&head->ids, &ids)) {
2777 			dev_err(ctrl->device,
2778 				"IDs don't match for shared namespace %d\n",
2779 					nsid);
2780 			ret = -EINVAL;
2781 			goto out_unlock;
2782 		}
2783 
2784 		*new = false;
2785 	}
2786 
2787 	list_add_tail(&ns->siblings, &head->list);
2788 	ns->head = head;
2789 
2790 out_unlock:
2791 	mutex_unlock(&ctrl->subsys->lock);
2792 	return ret;
2793 }
2794 
2795 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2796 {
2797 	struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2798 	struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2799 
2800 	return nsa->head->ns_id - nsb->head->ns_id;
2801 }
2802 
2803 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2804 {
2805 	struct nvme_ns *ns, *ret = NULL;
2806 
2807 	mutex_lock(&ctrl->namespaces_mutex);
2808 	list_for_each_entry(ns, &ctrl->namespaces, list) {
2809 		if (ns->head->ns_id == nsid) {
2810 			if (!kref_get_unless_zero(&ns->kref))
2811 				continue;
2812 			ret = ns;
2813 			break;
2814 		}
2815 		if (ns->head->ns_id > nsid)
2816 			break;
2817 	}
2818 	mutex_unlock(&ctrl->namespaces_mutex);
2819 	return ret;
2820 }
2821 
2822 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
2823 {
2824 	struct streams_directive_params s;
2825 	int ret;
2826 
2827 	if (!ctrl->nr_streams)
2828 		return 0;
2829 
2830 	ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
2831 	if (ret)
2832 		return ret;
2833 
2834 	ns->sws = le32_to_cpu(s.sws);
2835 	ns->sgs = le16_to_cpu(s.sgs);
2836 
2837 	if (ns->sws) {
2838 		unsigned int bs = 1 << ns->lba_shift;
2839 
2840 		blk_queue_io_min(ns->queue, bs * ns->sws);
2841 		if (ns->sgs)
2842 			blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
2843 	}
2844 
2845 	return 0;
2846 }
2847 
2848 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2849 {
2850 	struct nvme_ns *ns;
2851 	struct gendisk *disk;
2852 	struct nvme_id_ns *id;
2853 	char disk_name[DISK_NAME_LEN];
2854 	int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
2855 	bool new = true;
2856 
2857 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2858 	if (!ns)
2859 		return;
2860 
2861 	ns->queue = blk_mq_init_queue(ctrl->tagset);
2862 	if (IS_ERR(ns->queue))
2863 		goto out_free_ns;
2864 	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2865 	ns->queue->queuedata = ns;
2866 	ns->ctrl = ctrl;
2867 
2868 	kref_init(&ns->kref);
2869 	ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2870 
2871 	blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2872 	nvme_set_queue_limits(ctrl, ns->queue);
2873 
2874 	id = nvme_identify_ns(ctrl, nsid);
2875 	if (!id)
2876 		goto out_free_queue;
2877 
2878 	if (id->ncap == 0)
2879 		goto out_free_id;
2880 
2881 	if (nvme_init_ns_head(ns, nsid, id, &new))
2882 		goto out_free_id;
2883 	nvme_setup_streams_ns(ctrl, ns);
2884 
2885 #ifdef CONFIG_NVME_MULTIPATH
2886 	/*
2887 	 * If multipathing is enabled we need to always use the subsystem
2888 	 * instance number for numbering our devices to avoid conflicts
2889 	 * between subsystems that have multiple controllers and thus use
2890 	 * the multipath-aware subsystem node and those that have a single
2891 	 * controller and use the controller node directly.
2892 	 */
2893 	if (ns->head->disk) {
2894 		sprintf(disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
2895 				ctrl->cntlid, ns->head->instance);
2896 		flags = GENHD_FL_HIDDEN;
2897 	} else {
2898 		sprintf(disk_name, "nvme%dn%d", ctrl->subsys->instance,
2899 				ns->head->instance);
2900 	}
2901 #else
2902 	/*
2903 	 * But without the multipath code enabled, multiple controller per
2904 	 * subsystems are visible as devices and thus we cannot use the
2905 	 * subsystem instance.
2906 	 */
2907 	sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
2908 #endif
2909 
2910 	if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
2911 		if (nvme_nvm_register(ns, disk_name, node)) {
2912 			dev_warn(ctrl->device, "LightNVM init failure\n");
2913 			goto out_unlink_ns;
2914 		}
2915 	}
2916 
2917 	disk = alloc_disk_node(0, node);
2918 	if (!disk)
2919 		goto out_unlink_ns;
2920 
2921 	disk->fops = &nvme_fops;
2922 	disk->private_data = ns;
2923 	disk->queue = ns->queue;
2924 	disk->flags = flags;
2925 	memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
2926 	ns->disk = disk;
2927 
2928 	__nvme_revalidate_disk(disk, id);
2929 
2930 	mutex_lock(&ctrl->namespaces_mutex);
2931 	list_add_tail(&ns->list, &ctrl->namespaces);
2932 	mutex_unlock(&ctrl->namespaces_mutex);
2933 
2934 	nvme_get_ctrl(ctrl);
2935 
2936 	kfree(id);
2937 
2938 	device_add_disk(ctrl->device, ns->disk);
2939 	if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
2940 					&nvme_ns_id_attr_group))
2941 		pr_warn("%s: failed to create sysfs group for identification\n",
2942 			ns->disk->disk_name);
2943 	if (ns->ndev && nvme_nvm_register_sysfs(ns))
2944 		pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
2945 			ns->disk->disk_name);
2946 
2947 	if (new)
2948 		nvme_mpath_add_disk(ns->head);
2949 	nvme_mpath_add_disk_links(ns);
2950 	return;
2951  out_unlink_ns:
2952 	mutex_lock(&ctrl->subsys->lock);
2953 	list_del_rcu(&ns->siblings);
2954 	mutex_unlock(&ctrl->subsys->lock);
2955  out_free_id:
2956 	kfree(id);
2957  out_free_queue:
2958 	blk_cleanup_queue(ns->queue);
2959  out_free_ns:
2960 	kfree(ns);
2961 }
2962 
2963 static void nvme_ns_remove(struct nvme_ns *ns)
2964 {
2965 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
2966 		return;
2967 
2968 	if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
2969 		nvme_mpath_remove_disk_links(ns);
2970 		sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
2971 					&nvme_ns_id_attr_group);
2972 		if (ns->ndev)
2973 			nvme_nvm_unregister_sysfs(ns);
2974 		del_gendisk(ns->disk);
2975 		blk_cleanup_queue(ns->queue);
2976 		if (blk_get_integrity(ns->disk))
2977 			blk_integrity_unregister(ns->disk);
2978 	}
2979 
2980 	mutex_lock(&ns->ctrl->subsys->lock);
2981 	nvme_mpath_clear_current_path(ns);
2982 	list_del_rcu(&ns->siblings);
2983 	mutex_unlock(&ns->ctrl->subsys->lock);
2984 
2985 	mutex_lock(&ns->ctrl->namespaces_mutex);
2986 	list_del_init(&ns->list);
2987 	mutex_unlock(&ns->ctrl->namespaces_mutex);
2988 
2989 	synchronize_srcu(&ns->head->srcu);
2990 	nvme_put_ns(ns);
2991 }
2992 
2993 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2994 {
2995 	struct nvme_ns *ns;
2996 
2997 	ns = nvme_find_get_ns(ctrl, nsid);
2998 	if (ns) {
2999 		if (ns->disk && revalidate_disk(ns->disk))
3000 			nvme_ns_remove(ns);
3001 		nvme_put_ns(ns);
3002 	} else
3003 		nvme_alloc_ns(ctrl, nsid);
3004 }
3005 
3006 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3007 					unsigned nsid)
3008 {
3009 	struct nvme_ns *ns, *next;
3010 
3011 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3012 		if (ns->head->ns_id > nsid)
3013 			nvme_ns_remove(ns);
3014 	}
3015 }
3016 
3017 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3018 {
3019 	struct nvme_ns *ns;
3020 	__le32 *ns_list;
3021 	unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3022 	int ret = 0;
3023 
3024 	ns_list = kzalloc(0x1000, GFP_KERNEL);
3025 	if (!ns_list)
3026 		return -ENOMEM;
3027 
3028 	for (i = 0; i < num_lists; i++) {
3029 		ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3030 		if (ret)
3031 			goto free;
3032 
3033 		for (j = 0; j < min(nn, 1024U); j++) {
3034 			nsid = le32_to_cpu(ns_list[j]);
3035 			if (!nsid)
3036 				goto out;
3037 
3038 			nvme_validate_ns(ctrl, nsid);
3039 
3040 			while (++prev < nsid) {
3041 				ns = nvme_find_get_ns(ctrl, prev);
3042 				if (ns) {
3043 					nvme_ns_remove(ns);
3044 					nvme_put_ns(ns);
3045 				}
3046 			}
3047 		}
3048 		nn -= j;
3049 	}
3050  out:
3051 	nvme_remove_invalid_namespaces(ctrl, prev);
3052  free:
3053 	kfree(ns_list);
3054 	return ret;
3055 }
3056 
3057 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3058 {
3059 	unsigned i;
3060 
3061 	for (i = 1; i <= nn; i++)
3062 		nvme_validate_ns(ctrl, i);
3063 
3064 	nvme_remove_invalid_namespaces(ctrl, nn);
3065 }
3066 
3067 static void nvme_scan_work(struct work_struct *work)
3068 {
3069 	struct nvme_ctrl *ctrl =
3070 		container_of(work, struct nvme_ctrl, scan_work);
3071 	struct nvme_id_ctrl *id;
3072 	unsigned nn;
3073 
3074 	if (ctrl->state != NVME_CTRL_LIVE)
3075 		return;
3076 
3077 	if (nvme_identify_ctrl(ctrl, &id))
3078 		return;
3079 
3080 	nn = le32_to_cpu(id->nn);
3081 	if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3082 	    !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3083 		if (!nvme_scan_ns_list(ctrl, nn))
3084 			goto done;
3085 	}
3086 	nvme_scan_ns_sequential(ctrl, nn);
3087  done:
3088 	mutex_lock(&ctrl->namespaces_mutex);
3089 	list_sort(NULL, &ctrl->namespaces, ns_cmp);
3090 	mutex_unlock(&ctrl->namespaces_mutex);
3091 	kfree(id);
3092 }
3093 
3094 void nvme_queue_scan(struct nvme_ctrl *ctrl)
3095 {
3096 	/*
3097 	 * Do not queue new scan work when a controller is reset during
3098 	 * removal.
3099 	 */
3100 	if (ctrl->state == NVME_CTRL_LIVE)
3101 		queue_work(nvme_wq, &ctrl->scan_work);
3102 }
3103 EXPORT_SYMBOL_GPL(nvme_queue_scan);
3104 
3105 /*
3106  * This function iterates the namespace list unlocked to allow recovery from
3107  * controller failure. It is up to the caller to ensure the namespace list is
3108  * not modified by scan work while this function is executing.
3109  */
3110 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3111 {
3112 	struct nvme_ns *ns, *next;
3113 
3114 	/*
3115 	 * The dead states indicates the controller was not gracefully
3116 	 * disconnected. In that case, we won't be able to flush any data while
3117 	 * removing the namespaces' disks; fail all the queues now to avoid
3118 	 * potentially having to clean up the failed sync later.
3119 	 */
3120 	if (ctrl->state == NVME_CTRL_DEAD)
3121 		nvme_kill_queues(ctrl);
3122 
3123 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
3124 		nvme_ns_remove(ns);
3125 }
3126 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3127 
3128 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3129 {
3130 	char *envp[2] = { NULL, NULL };
3131 	u32 aen_result = ctrl->aen_result;
3132 
3133 	ctrl->aen_result = 0;
3134 	if (!aen_result)
3135 		return;
3136 
3137 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3138 	if (!envp[0])
3139 		return;
3140 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3141 	kfree(envp[0]);
3142 }
3143 
3144 static void nvme_async_event_work(struct work_struct *work)
3145 {
3146 	struct nvme_ctrl *ctrl =
3147 		container_of(work, struct nvme_ctrl, async_event_work);
3148 
3149 	nvme_aen_uevent(ctrl);
3150 	ctrl->ops->submit_async_event(ctrl);
3151 }
3152 
3153 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3154 {
3155 
3156 	u32 csts;
3157 
3158 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3159 		return false;
3160 
3161 	if (csts == ~0)
3162 		return false;
3163 
3164 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3165 }
3166 
3167 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3168 {
3169 	struct nvme_fw_slot_info_log *log;
3170 
3171 	log = kmalloc(sizeof(*log), GFP_KERNEL);
3172 	if (!log)
3173 		return;
3174 
3175 	if (nvme_get_log(ctrl, NVME_LOG_FW_SLOT, log, sizeof(*log)))
3176 		dev_warn(ctrl->device,
3177 				"Get FW SLOT INFO log error\n");
3178 	kfree(log);
3179 }
3180 
3181 static void nvme_fw_act_work(struct work_struct *work)
3182 {
3183 	struct nvme_ctrl *ctrl = container_of(work,
3184 				struct nvme_ctrl, fw_act_work);
3185 	unsigned long fw_act_timeout;
3186 
3187 	if (ctrl->mtfa)
3188 		fw_act_timeout = jiffies +
3189 				msecs_to_jiffies(ctrl->mtfa * 100);
3190 	else
3191 		fw_act_timeout = jiffies +
3192 				msecs_to_jiffies(admin_timeout * 1000);
3193 
3194 	nvme_stop_queues(ctrl);
3195 	while (nvme_ctrl_pp_status(ctrl)) {
3196 		if (time_after(jiffies, fw_act_timeout)) {
3197 			dev_warn(ctrl->device,
3198 				"Fw activation timeout, reset controller\n");
3199 			nvme_reset_ctrl(ctrl);
3200 			break;
3201 		}
3202 		msleep(100);
3203 	}
3204 
3205 	if (ctrl->state != NVME_CTRL_LIVE)
3206 		return;
3207 
3208 	nvme_start_queues(ctrl);
3209 	/* read FW slot information to clear the AER */
3210 	nvme_get_fw_slot_info(ctrl);
3211 }
3212 
3213 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3214 		union nvme_result *res)
3215 {
3216 	u32 result = le32_to_cpu(res->u32);
3217 
3218 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3219 		return;
3220 
3221 	switch (result & 0x7) {
3222 	case NVME_AER_ERROR:
3223 	case NVME_AER_SMART:
3224 	case NVME_AER_CSS:
3225 	case NVME_AER_VS:
3226 		ctrl->aen_result = result;
3227 		break;
3228 	default:
3229 		break;
3230 	}
3231 
3232 	switch (result & 0xff07) {
3233 	case NVME_AER_NOTICE_NS_CHANGED:
3234 		dev_info(ctrl->device, "rescanning\n");
3235 		nvme_queue_scan(ctrl);
3236 		break;
3237 	case NVME_AER_NOTICE_FW_ACT_STARTING:
3238 		queue_work(nvme_wq, &ctrl->fw_act_work);
3239 		break;
3240 	default:
3241 		dev_warn(ctrl->device, "async event result %08x\n", result);
3242 	}
3243 	queue_work(nvme_wq, &ctrl->async_event_work);
3244 }
3245 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3246 
3247 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3248 {
3249 	nvme_stop_keep_alive(ctrl);
3250 	flush_work(&ctrl->async_event_work);
3251 	flush_work(&ctrl->scan_work);
3252 	cancel_work_sync(&ctrl->fw_act_work);
3253 }
3254 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3255 
3256 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3257 {
3258 	if (ctrl->kato)
3259 		nvme_start_keep_alive(ctrl);
3260 
3261 	if (ctrl->queue_count > 1) {
3262 		nvme_queue_scan(ctrl);
3263 		queue_work(nvme_wq, &ctrl->async_event_work);
3264 		nvme_start_queues(ctrl);
3265 	}
3266 }
3267 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3268 
3269 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3270 {
3271 	cdev_device_del(&ctrl->cdev, ctrl->device);
3272 }
3273 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3274 
3275 static void nvme_free_ctrl(struct device *dev)
3276 {
3277 	struct nvme_ctrl *ctrl =
3278 		container_of(dev, struct nvme_ctrl, ctrl_device);
3279 	struct nvme_subsystem *subsys = ctrl->subsys;
3280 
3281 	ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3282 	kfree(ctrl->effects);
3283 
3284 	if (subsys) {
3285 		mutex_lock(&subsys->lock);
3286 		list_del(&ctrl->subsys_entry);
3287 		mutex_unlock(&subsys->lock);
3288 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3289 	}
3290 
3291 	ctrl->ops->free_ctrl(ctrl);
3292 
3293 	if (subsys)
3294 		nvme_put_subsystem(subsys);
3295 }
3296 
3297 /*
3298  * Initialize a NVMe controller structures.  This needs to be called during
3299  * earliest initialization so that we have the initialized structured around
3300  * during probing.
3301  */
3302 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3303 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
3304 {
3305 	int ret;
3306 
3307 	ctrl->state = NVME_CTRL_NEW;
3308 	spin_lock_init(&ctrl->lock);
3309 	INIT_LIST_HEAD(&ctrl->namespaces);
3310 	mutex_init(&ctrl->namespaces_mutex);
3311 	ctrl->dev = dev;
3312 	ctrl->ops = ops;
3313 	ctrl->quirks = quirks;
3314 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3315 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3316 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3317 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3318 
3319 	ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3320 	if (ret < 0)
3321 		goto out;
3322 	ctrl->instance = ret;
3323 
3324 	device_initialize(&ctrl->ctrl_device);
3325 	ctrl->device = &ctrl->ctrl_device;
3326 	ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3327 	ctrl->device->class = nvme_class;
3328 	ctrl->device->parent = ctrl->dev;
3329 	ctrl->device->groups = nvme_dev_attr_groups;
3330 	ctrl->device->release = nvme_free_ctrl;
3331 	dev_set_drvdata(ctrl->device, ctrl);
3332 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3333 	if (ret)
3334 		goto out_release_instance;
3335 
3336 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
3337 	ctrl->cdev.owner = ops->module;
3338 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3339 	if (ret)
3340 		goto out_free_name;
3341 
3342 	/*
3343 	 * Initialize latency tolerance controls.  The sysfs files won't
3344 	 * be visible to userspace unless the device actually supports APST.
3345 	 */
3346 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3347 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3348 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3349 
3350 	return 0;
3351 out_free_name:
3352 	kfree_const(dev->kobj.name);
3353 out_release_instance:
3354 	ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3355 out:
3356 	return ret;
3357 }
3358 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3359 
3360 /**
3361  * nvme_kill_queues(): Ends all namespace queues
3362  * @ctrl: the dead controller that needs to end
3363  *
3364  * Call this function when the driver determines it is unable to get the
3365  * controller in a state capable of servicing IO.
3366  */
3367 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3368 {
3369 	struct nvme_ns *ns;
3370 
3371 	mutex_lock(&ctrl->namespaces_mutex);
3372 
3373 	/* Forcibly unquiesce queues to avoid blocking dispatch */
3374 	if (ctrl->admin_q)
3375 		blk_mq_unquiesce_queue(ctrl->admin_q);
3376 
3377 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3378 		/*
3379 		 * Revalidating a dead namespace sets capacity to 0. This will
3380 		 * end buffered writers dirtying pages that can't be synced.
3381 		 */
3382 		if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
3383 			continue;
3384 		revalidate_disk(ns->disk);
3385 		blk_set_queue_dying(ns->queue);
3386 
3387 		/* Forcibly unquiesce queues to avoid blocking dispatch */
3388 		blk_mq_unquiesce_queue(ns->queue);
3389 	}
3390 	mutex_unlock(&ctrl->namespaces_mutex);
3391 }
3392 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3393 
3394 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3395 {
3396 	struct nvme_ns *ns;
3397 
3398 	mutex_lock(&ctrl->namespaces_mutex);
3399 	list_for_each_entry(ns, &ctrl->namespaces, list)
3400 		blk_mq_unfreeze_queue(ns->queue);
3401 	mutex_unlock(&ctrl->namespaces_mutex);
3402 }
3403 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3404 
3405 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3406 {
3407 	struct nvme_ns *ns;
3408 
3409 	mutex_lock(&ctrl->namespaces_mutex);
3410 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3411 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3412 		if (timeout <= 0)
3413 			break;
3414 	}
3415 	mutex_unlock(&ctrl->namespaces_mutex);
3416 }
3417 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3418 
3419 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3420 {
3421 	struct nvme_ns *ns;
3422 
3423 	mutex_lock(&ctrl->namespaces_mutex);
3424 	list_for_each_entry(ns, &ctrl->namespaces, list)
3425 		blk_mq_freeze_queue_wait(ns->queue);
3426 	mutex_unlock(&ctrl->namespaces_mutex);
3427 }
3428 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3429 
3430 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3431 {
3432 	struct nvme_ns *ns;
3433 
3434 	mutex_lock(&ctrl->namespaces_mutex);
3435 	list_for_each_entry(ns, &ctrl->namespaces, list)
3436 		blk_freeze_queue_start(ns->queue);
3437 	mutex_unlock(&ctrl->namespaces_mutex);
3438 }
3439 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3440 
3441 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3442 {
3443 	struct nvme_ns *ns;
3444 
3445 	mutex_lock(&ctrl->namespaces_mutex);
3446 	list_for_each_entry(ns, &ctrl->namespaces, list)
3447 		blk_mq_quiesce_queue(ns->queue);
3448 	mutex_unlock(&ctrl->namespaces_mutex);
3449 }
3450 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3451 
3452 void nvme_start_queues(struct nvme_ctrl *ctrl)
3453 {
3454 	struct nvme_ns *ns;
3455 
3456 	mutex_lock(&ctrl->namespaces_mutex);
3457 	list_for_each_entry(ns, &ctrl->namespaces, list)
3458 		blk_mq_unquiesce_queue(ns->queue);
3459 	mutex_unlock(&ctrl->namespaces_mutex);
3460 }
3461 EXPORT_SYMBOL_GPL(nvme_start_queues);
3462 
3463 int nvme_reinit_tagset(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set)
3464 {
3465 	if (!ctrl->ops->reinit_request)
3466 		return 0;
3467 
3468 	return blk_mq_tagset_iter(set, set->driver_data,
3469 			ctrl->ops->reinit_request);
3470 }
3471 EXPORT_SYMBOL_GPL(nvme_reinit_tagset);
3472 
3473 int __init nvme_core_init(void)
3474 {
3475 	int result;
3476 
3477 	nvme_wq = alloc_workqueue("nvme-wq",
3478 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3479 	if (!nvme_wq)
3480 		return -ENOMEM;
3481 
3482 	result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3483 	if (result < 0)
3484 		goto destroy_wq;
3485 
3486 	nvme_class = class_create(THIS_MODULE, "nvme");
3487 	if (IS_ERR(nvme_class)) {
3488 		result = PTR_ERR(nvme_class);
3489 		goto unregister_chrdev;
3490 	}
3491 
3492 	nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3493 	if (IS_ERR(nvme_subsys_class)) {
3494 		result = PTR_ERR(nvme_subsys_class);
3495 		goto destroy_class;
3496 	}
3497 	return 0;
3498 
3499 destroy_class:
3500 	class_destroy(nvme_class);
3501 unregister_chrdev:
3502 	unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3503 destroy_wq:
3504 	destroy_workqueue(nvme_wq);
3505 	return result;
3506 }
3507 
3508 void nvme_core_exit(void)
3509 {
3510 	ida_destroy(&nvme_subsystems_ida);
3511 	class_destroy(nvme_subsys_class);
3512 	class_destroy(nvme_class);
3513 	unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3514 	destroy_workqueue(nvme_wq);
3515 }
3516 
3517 MODULE_LICENSE("GPL");
3518 MODULE_VERSION("1.0");
3519 module_init(nvme_core_init);
3520 module_exit(nvme_core_exit);
3521