xref: /openbmc/linux/drivers/nvme/host/core.c (revision 680ef72a)
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 = size;
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 	put_unaligned_le64(key, &data[0]);
1453 	put_unaligned_le64(sa_key, &data[8]);
1454 
1455 	memset(&c, 0, sizeof(c));
1456 	c.common.opcode = op;
1457 	c.common.nsid = cpu_to_le32(head->ns_id);
1458 	c.common.cdw10[0] = cpu_to_le32(cdw10);
1459 
1460 	ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1461 	if (unlikely(!ns))
1462 		ret = -EWOULDBLOCK;
1463 	else
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 		blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1710 	blk_queue_virt_boundary(q, ctrl->page_size - 1);
1711 	if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1712 		vwc = true;
1713 	blk_queue_write_cache(q, vwc, vwc);
1714 }
1715 
1716 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1717 {
1718 	__le64 ts;
1719 	int ret;
1720 
1721 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1722 		return 0;
1723 
1724 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1725 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1726 			NULL);
1727 	if (ret)
1728 		dev_warn_once(ctrl->device,
1729 			"could not set timestamp (%d)\n", ret);
1730 	return ret;
1731 }
1732 
1733 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1734 {
1735 	/*
1736 	 * APST (Autonomous Power State Transition) lets us program a
1737 	 * table of power state transitions that the controller will
1738 	 * perform automatically.  We configure it with a simple
1739 	 * heuristic: we are willing to spend at most 2% of the time
1740 	 * transitioning between power states.  Therefore, when running
1741 	 * in any given state, we will enter the next lower-power
1742 	 * non-operational state after waiting 50 * (enlat + exlat)
1743 	 * microseconds, as long as that state's exit latency is under
1744 	 * the requested maximum latency.
1745 	 *
1746 	 * We will not autonomously enter any non-operational state for
1747 	 * which the total latency exceeds ps_max_latency_us.  Users
1748 	 * can set ps_max_latency_us to zero to turn off APST.
1749 	 */
1750 
1751 	unsigned apste;
1752 	struct nvme_feat_auto_pst *table;
1753 	u64 max_lat_us = 0;
1754 	int max_ps = -1;
1755 	int ret;
1756 
1757 	/*
1758 	 * If APST isn't supported or if we haven't been initialized yet,
1759 	 * then don't do anything.
1760 	 */
1761 	if (!ctrl->apsta)
1762 		return 0;
1763 
1764 	if (ctrl->npss > 31) {
1765 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1766 		return 0;
1767 	}
1768 
1769 	table = kzalloc(sizeof(*table), GFP_KERNEL);
1770 	if (!table)
1771 		return 0;
1772 
1773 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1774 		/* Turn off APST. */
1775 		apste = 0;
1776 		dev_dbg(ctrl->device, "APST disabled\n");
1777 	} else {
1778 		__le64 target = cpu_to_le64(0);
1779 		int state;
1780 
1781 		/*
1782 		 * Walk through all states from lowest- to highest-power.
1783 		 * According to the spec, lower-numbered states use more
1784 		 * power.  NPSS, despite the name, is the index of the
1785 		 * lowest-power state, not the number of states.
1786 		 */
1787 		for (state = (int)ctrl->npss; state >= 0; state--) {
1788 			u64 total_latency_us, exit_latency_us, transition_ms;
1789 
1790 			if (target)
1791 				table->entries[state] = target;
1792 
1793 			/*
1794 			 * Don't allow transitions to the deepest state
1795 			 * if it's quirked off.
1796 			 */
1797 			if (state == ctrl->npss &&
1798 			    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1799 				continue;
1800 
1801 			/*
1802 			 * Is this state a useful non-operational state for
1803 			 * higher-power states to autonomously transition to?
1804 			 */
1805 			if (!(ctrl->psd[state].flags &
1806 			      NVME_PS_FLAGS_NON_OP_STATE))
1807 				continue;
1808 
1809 			exit_latency_us =
1810 				(u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1811 			if (exit_latency_us > ctrl->ps_max_latency_us)
1812 				continue;
1813 
1814 			total_latency_us =
1815 				exit_latency_us +
1816 				le32_to_cpu(ctrl->psd[state].entry_lat);
1817 
1818 			/*
1819 			 * This state is good.  Use it as the APST idle
1820 			 * target for higher power states.
1821 			 */
1822 			transition_ms = total_latency_us + 19;
1823 			do_div(transition_ms, 20);
1824 			if (transition_ms > (1 << 24) - 1)
1825 				transition_ms = (1 << 24) - 1;
1826 
1827 			target = cpu_to_le64((state << 3) |
1828 					     (transition_ms << 8));
1829 
1830 			if (max_ps == -1)
1831 				max_ps = state;
1832 
1833 			if (total_latency_us > max_lat_us)
1834 				max_lat_us = total_latency_us;
1835 		}
1836 
1837 		apste = 1;
1838 
1839 		if (max_ps == -1) {
1840 			dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1841 		} else {
1842 			dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1843 				max_ps, max_lat_us, (int)sizeof(*table), table);
1844 		}
1845 	}
1846 
1847 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1848 				table, sizeof(*table), NULL);
1849 	if (ret)
1850 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1851 
1852 	kfree(table);
1853 	return ret;
1854 }
1855 
1856 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1857 {
1858 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1859 	u64 latency;
1860 
1861 	switch (val) {
1862 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1863 	case PM_QOS_LATENCY_ANY:
1864 		latency = U64_MAX;
1865 		break;
1866 
1867 	default:
1868 		latency = val;
1869 	}
1870 
1871 	if (ctrl->ps_max_latency_us != latency) {
1872 		ctrl->ps_max_latency_us = latency;
1873 		nvme_configure_apst(ctrl);
1874 	}
1875 }
1876 
1877 struct nvme_core_quirk_entry {
1878 	/*
1879 	 * NVMe model and firmware strings are padded with spaces.  For
1880 	 * simplicity, strings in the quirk table are padded with NULLs
1881 	 * instead.
1882 	 */
1883 	u16 vid;
1884 	const char *mn;
1885 	const char *fr;
1886 	unsigned long quirks;
1887 };
1888 
1889 static const struct nvme_core_quirk_entry core_quirks[] = {
1890 	{
1891 		/*
1892 		 * This Toshiba device seems to die using any APST states.  See:
1893 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1894 		 */
1895 		.vid = 0x1179,
1896 		.mn = "THNSF5256GPUK TOSHIBA",
1897 		.quirks = NVME_QUIRK_NO_APST,
1898 	}
1899 };
1900 
1901 /* match is null-terminated but idstr is space-padded. */
1902 static bool string_matches(const char *idstr, const char *match, size_t len)
1903 {
1904 	size_t matchlen;
1905 
1906 	if (!match)
1907 		return true;
1908 
1909 	matchlen = strlen(match);
1910 	WARN_ON_ONCE(matchlen > len);
1911 
1912 	if (memcmp(idstr, match, matchlen))
1913 		return false;
1914 
1915 	for (; matchlen < len; matchlen++)
1916 		if (idstr[matchlen] != ' ')
1917 			return false;
1918 
1919 	return true;
1920 }
1921 
1922 static bool quirk_matches(const struct nvme_id_ctrl *id,
1923 			  const struct nvme_core_quirk_entry *q)
1924 {
1925 	return q->vid == le16_to_cpu(id->vid) &&
1926 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1927 		string_matches(id->fr, q->fr, sizeof(id->fr));
1928 }
1929 
1930 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
1931 		struct nvme_id_ctrl *id)
1932 {
1933 	size_t nqnlen;
1934 	int off;
1935 
1936 	nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
1937 	if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
1938 		strncpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
1939 		return;
1940 	}
1941 
1942 	if (ctrl->vs >= NVME_VS(1, 2, 1))
1943 		dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
1944 
1945 	/* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
1946 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
1947 			"nqn.2014.08.org.nvmexpress:%4x%4x",
1948 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
1949 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
1950 	off += sizeof(id->sn);
1951 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
1952 	off += sizeof(id->mn);
1953 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
1954 }
1955 
1956 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
1957 {
1958 	ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
1959 	kfree(subsys);
1960 }
1961 
1962 static void nvme_release_subsystem(struct device *dev)
1963 {
1964 	__nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
1965 }
1966 
1967 static void nvme_destroy_subsystem(struct kref *ref)
1968 {
1969 	struct nvme_subsystem *subsys =
1970 			container_of(ref, struct nvme_subsystem, ref);
1971 
1972 	mutex_lock(&nvme_subsystems_lock);
1973 	list_del(&subsys->entry);
1974 	mutex_unlock(&nvme_subsystems_lock);
1975 
1976 	ida_destroy(&subsys->ns_ida);
1977 	device_del(&subsys->dev);
1978 	put_device(&subsys->dev);
1979 }
1980 
1981 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
1982 {
1983 	kref_put(&subsys->ref, nvme_destroy_subsystem);
1984 }
1985 
1986 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
1987 {
1988 	struct nvme_subsystem *subsys;
1989 
1990 	lockdep_assert_held(&nvme_subsystems_lock);
1991 
1992 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
1993 		if (strcmp(subsys->subnqn, subsysnqn))
1994 			continue;
1995 		if (!kref_get_unless_zero(&subsys->ref))
1996 			continue;
1997 		return subsys;
1998 	}
1999 
2000 	return NULL;
2001 }
2002 
2003 #define SUBSYS_ATTR_RO(_name, _mode, _show)			\
2004 	struct device_attribute subsys_attr_##_name = \
2005 		__ATTR(_name, _mode, _show, NULL)
2006 
2007 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2008 				    struct device_attribute *attr,
2009 				    char *buf)
2010 {
2011 	struct nvme_subsystem *subsys =
2012 		container_of(dev, struct nvme_subsystem, dev);
2013 
2014 	return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2015 }
2016 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2017 
2018 #define nvme_subsys_show_str_function(field)				\
2019 static ssize_t subsys_##field##_show(struct device *dev,		\
2020 			    struct device_attribute *attr, char *buf)	\
2021 {									\
2022 	struct nvme_subsystem *subsys =					\
2023 		container_of(dev, struct nvme_subsystem, dev);		\
2024 	return sprintf(buf, "%.*s\n",					\
2025 		       (int)sizeof(subsys->field), subsys->field);	\
2026 }									\
2027 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2028 
2029 nvme_subsys_show_str_function(model);
2030 nvme_subsys_show_str_function(serial);
2031 nvme_subsys_show_str_function(firmware_rev);
2032 
2033 static struct attribute *nvme_subsys_attrs[] = {
2034 	&subsys_attr_model.attr,
2035 	&subsys_attr_serial.attr,
2036 	&subsys_attr_firmware_rev.attr,
2037 	&subsys_attr_subsysnqn.attr,
2038 	NULL,
2039 };
2040 
2041 static struct attribute_group nvme_subsys_attrs_group = {
2042 	.attrs = nvme_subsys_attrs,
2043 };
2044 
2045 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2046 	&nvme_subsys_attrs_group,
2047 	NULL,
2048 };
2049 
2050 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2051 {
2052 	struct nvme_subsystem *subsys, *found;
2053 	int ret;
2054 
2055 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2056 	if (!subsys)
2057 		return -ENOMEM;
2058 	ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2059 	if (ret < 0) {
2060 		kfree(subsys);
2061 		return ret;
2062 	}
2063 	subsys->instance = ret;
2064 	mutex_init(&subsys->lock);
2065 	kref_init(&subsys->ref);
2066 	INIT_LIST_HEAD(&subsys->ctrls);
2067 	INIT_LIST_HEAD(&subsys->nsheads);
2068 	nvme_init_subnqn(subsys, ctrl, id);
2069 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2070 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2071 	memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2072 	subsys->vendor_id = le16_to_cpu(id->vid);
2073 	subsys->cmic = id->cmic;
2074 
2075 	subsys->dev.class = nvme_subsys_class;
2076 	subsys->dev.release = nvme_release_subsystem;
2077 	subsys->dev.groups = nvme_subsys_attrs_groups;
2078 	dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2079 	device_initialize(&subsys->dev);
2080 
2081 	mutex_lock(&nvme_subsystems_lock);
2082 	found = __nvme_find_get_subsystem(subsys->subnqn);
2083 	if (found) {
2084 		/*
2085 		 * Verify that the subsystem actually supports multiple
2086 		 * controllers, else bail out.
2087 		 */
2088 		if (!(id->cmic & (1 << 1))) {
2089 			dev_err(ctrl->device,
2090 				"ignoring ctrl due to duplicate subnqn (%s).\n",
2091 				found->subnqn);
2092 			nvme_put_subsystem(found);
2093 			ret = -EINVAL;
2094 			goto out_unlock;
2095 		}
2096 
2097 		__nvme_release_subsystem(subsys);
2098 		subsys = found;
2099 	} else {
2100 		ret = device_add(&subsys->dev);
2101 		if (ret) {
2102 			dev_err(ctrl->device,
2103 				"failed to register subsystem device.\n");
2104 			goto out_unlock;
2105 		}
2106 		ida_init(&subsys->ns_ida);
2107 		list_add_tail(&subsys->entry, &nvme_subsystems);
2108 	}
2109 
2110 	ctrl->subsys = subsys;
2111 	mutex_unlock(&nvme_subsystems_lock);
2112 
2113 	if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2114 			dev_name(ctrl->device))) {
2115 		dev_err(ctrl->device,
2116 			"failed to create sysfs link from subsystem.\n");
2117 		/* the transport driver will eventually put the subsystem */
2118 		return -EINVAL;
2119 	}
2120 
2121 	mutex_lock(&subsys->lock);
2122 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2123 	mutex_unlock(&subsys->lock);
2124 
2125 	return 0;
2126 
2127 out_unlock:
2128 	mutex_unlock(&nvme_subsystems_lock);
2129 	put_device(&subsys->dev);
2130 	return ret;
2131 }
2132 
2133 static int nvme_get_log(struct nvme_ctrl *ctrl, u8 log_page, void *log,
2134 			size_t size)
2135 {
2136 	struct nvme_command c = { };
2137 
2138 	c.common.opcode = nvme_admin_get_log_page;
2139 	c.common.nsid = cpu_to_le32(NVME_NSID_ALL);
2140 	c.common.cdw10[0] = nvme_get_log_dw10(log_page, size);
2141 
2142 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2143 }
2144 
2145 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2146 {
2147 	int ret;
2148 
2149 	if (!ctrl->effects)
2150 		ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2151 
2152 	if (!ctrl->effects)
2153 		return 0;
2154 
2155 	ret = nvme_get_log(ctrl, NVME_LOG_CMD_EFFECTS, ctrl->effects,
2156 					sizeof(*ctrl->effects));
2157 	if (ret) {
2158 		kfree(ctrl->effects);
2159 		ctrl->effects = NULL;
2160 	}
2161 	return ret;
2162 }
2163 
2164 /*
2165  * Initialize the cached copies of the Identify data and various controller
2166  * register in our nvme_ctrl structure.  This should be called as soon as
2167  * the admin queue is fully up and running.
2168  */
2169 int nvme_init_identify(struct nvme_ctrl *ctrl)
2170 {
2171 	struct nvme_id_ctrl *id;
2172 	u64 cap;
2173 	int ret, page_shift;
2174 	u32 max_hw_sectors;
2175 	bool prev_apst_enabled;
2176 
2177 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2178 	if (ret) {
2179 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2180 		return ret;
2181 	}
2182 
2183 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2184 	if (ret) {
2185 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2186 		return ret;
2187 	}
2188 	page_shift = NVME_CAP_MPSMIN(cap) + 12;
2189 
2190 	if (ctrl->vs >= NVME_VS(1, 1, 0))
2191 		ctrl->subsystem = NVME_CAP_NSSRC(cap);
2192 
2193 	ret = nvme_identify_ctrl(ctrl, &id);
2194 	if (ret) {
2195 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2196 		return -EIO;
2197 	}
2198 
2199 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2200 		ret = nvme_get_effects_log(ctrl);
2201 		if (ret < 0)
2202 			return ret;
2203 	}
2204 
2205 	if (!ctrl->identified) {
2206 		int i;
2207 
2208 		ret = nvme_init_subsystem(ctrl, id);
2209 		if (ret)
2210 			goto out_free;
2211 
2212 		/*
2213 		 * Check for quirks.  Quirk can depend on firmware version,
2214 		 * so, in principle, the set of quirks present can change
2215 		 * across a reset.  As a possible future enhancement, we
2216 		 * could re-scan for quirks every time we reinitialize
2217 		 * the device, but we'd have to make sure that the driver
2218 		 * behaves intelligently if the quirks change.
2219 		 */
2220 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2221 			if (quirk_matches(id, &core_quirks[i]))
2222 				ctrl->quirks |= core_quirks[i].quirks;
2223 		}
2224 	}
2225 
2226 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2227 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2228 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2229 	}
2230 
2231 	ctrl->oacs = le16_to_cpu(id->oacs);
2232 	ctrl->oncs = le16_to_cpup(&id->oncs);
2233 	atomic_set(&ctrl->abort_limit, id->acl + 1);
2234 	ctrl->vwc = id->vwc;
2235 	ctrl->cntlid = le16_to_cpup(&id->cntlid);
2236 	if (id->mdts)
2237 		max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2238 	else
2239 		max_hw_sectors = UINT_MAX;
2240 	ctrl->max_hw_sectors =
2241 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2242 
2243 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
2244 	ctrl->sgls = le32_to_cpu(id->sgls);
2245 	ctrl->kas = le16_to_cpu(id->kas);
2246 
2247 	if (id->rtd3e) {
2248 		/* us -> s */
2249 		u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2250 
2251 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2252 						 shutdown_timeout, 60);
2253 
2254 		if (ctrl->shutdown_timeout != shutdown_timeout)
2255 			dev_warn(ctrl->device,
2256 				 "Shutdown timeout set to %u seconds\n",
2257 				 ctrl->shutdown_timeout);
2258 	} else
2259 		ctrl->shutdown_timeout = shutdown_timeout;
2260 
2261 	ctrl->npss = id->npss;
2262 	ctrl->apsta = id->apsta;
2263 	prev_apst_enabled = ctrl->apst_enabled;
2264 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2265 		if (force_apst && id->apsta) {
2266 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2267 			ctrl->apst_enabled = true;
2268 		} else {
2269 			ctrl->apst_enabled = false;
2270 		}
2271 	} else {
2272 		ctrl->apst_enabled = id->apsta;
2273 	}
2274 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2275 
2276 	if (ctrl->ops->flags & NVME_F_FABRICS) {
2277 		ctrl->icdoff = le16_to_cpu(id->icdoff);
2278 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2279 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2280 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2281 
2282 		/*
2283 		 * In fabrics we need to verify the cntlid matches the
2284 		 * admin connect
2285 		 */
2286 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2287 			ret = -EINVAL;
2288 			goto out_free;
2289 		}
2290 
2291 		if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2292 			dev_err(ctrl->device,
2293 				"keep-alive support is mandatory for fabrics\n");
2294 			ret = -EINVAL;
2295 			goto out_free;
2296 		}
2297 	} else {
2298 		ctrl->cntlid = le16_to_cpu(id->cntlid);
2299 		ctrl->hmpre = le32_to_cpu(id->hmpre);
2300 		ctrl->hmmin = le32_to_cpu(id->hmmin);
2301 		ctrl->hmminds = le32_to_cpu(id->hmminds);
2302 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2303 	}
2304 
2305 	kfree(id);
2306 
2307 	if (ctrl->apst_enabled && !prev_apst_enabled)
2308 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
2309 	else if (!ctrl->apst_enabled && prev_apst_enabled)
2310 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
2311 
2312 	ret = nvme_configure_apst(ctrl);
2313 	if (ret < 0)
2314 		return ret;
2315 
2316 	ret = nvme_configure_timestamp(ctrl);
2317 	if (ret < 0)
2318 		return ret;
2319 
2320 	ret = nvme_configure_directives(ctrl);
2321 	if (ret < 0)
2322 		return ret;
2323 
2324 	ctrl->identified = true;
2325 
2326 	return 0;
2327 
2328 out_free:
2329 	kfree(id);
2330 	return ret;
2331 }
2332 EXPORT_SYMBOL_GPL(nvme_init_identify);
2333 
2334 static int nvme_dev_open(struct inode *inode, struct file *file)
2335 {
2336 	struct nvme_ctrl *ctrl =
2337 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2338 
2339 	if (ctrl->state != NVME_CTRL_LIVE)
2340 		return -EWOULDBLOCK;
2341 	file->private_data = ctrl;
2342 	return 0;
2343 }
2344 
2345 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2346 {
2347 	struct nvme_ns *ns;
2348 	int ret;
2349 
2350 	mutex_lock(&ctrl->namespaces_mutex);
2351 	if (list_empty(&ctrl->namespaces)) {
2352 		ret = -ENOTTY;
2353 		goto out_unlock;
2354 	}
2355 
2356 	ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2357 	if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2358 		dev_warn(ctrl->device,
2359 			"NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2360 		ret = -EINVAL;
2361 		goto out_unlock;
2362 	}
2363 
2364 	dev_warn(ctrl->device,
2365 		"using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2366 	kref_get(&ns->kref);
2367 	mutex_unlock(&ctrl->namespaces_mutex);
2368 
2369 	ret = nvme_user_cmd(ctrl, ns, argp);
2370 	nvme_put_ns(ns);
2371 	return ret;
2372 
2373 out_unlock:
2374 	mutex_unlock(&ctrl->namespaces_mutex);
2375 	return ret;
2376 }
2377 
2378 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2379 		unsigned long arg)
2380 {
2381 	struct nvme_ctrl *ctrl = file->private_data;
2382 	void __user *argp = (void __user *)arg;
2383 
2384 	switch (cmd) {
2385 	case NVME_IOCTL_ADMIN_CMD:
2386 		return nvme_user_cmd(ctrl, NULL, argp);
2387 	case NVME_IOCTL_IO_CMD:
2388 		return nvme_dev_user_cmd(ctrl, argp);
2389 	case NVME_IOCTL_RESET:
2390 		dev_warn(ctrl->device, "resetting controller\n");
2391 		return nvme_reset_ctrl_sync(ctrl);
2392 	case NVME_IOCTL_SUBSYS_RESET:
2393 		return nvme_reset_subsystem(ctrl);
2394 	case NVME_IOCTL_RESCAN:
2395 		nvme_queue_scan(ctrl);
2396 		return 0;
2397 	default:
2398 		return -ENOTTY;
2399 	}
2400 }
2401 
2402 static const struct file_operations nvme_dev_fops = {
2403 	.owner		= THIS_MODULE,
2404 	.open		= nvme_dev_open,
2405 	.unlocked_ioctl	= nvme_dev_ioctl,
2406 	.compat_ioctl	= nvme_dev_ioctl,
2407 };
2408 
2409 static ssize_t nvme_sysfs_reset(struct device *dev,
2410 				struct device_attribute *attr, const char *buf,
2411 				size_t count)
2412 {
2413 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2414 	int ret;
2415 
2416 	ret = nvme_reset_ctrl_sync(ctrl);
2417 	if (ret < 0)
2418 		return ret;
2419 	return count;
2420 }
2421 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2422 
2423 static ssize_t nvme_sysfs_rescan(struct device *dev,
2424 				struct device_attribute *attr, const char *buf,
2425 				size_t count)
2426 {
2427 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2428 
2429 	nvme_queue_scan(ctrl);
2430 	return count;
2431 }
2432 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2433 
2434 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2435 {
2436 	struct gendisk *disk = dev_to_disk(dev);
2437 
2438 	if (disk->fops == &nvme_fops)
2439 		return nvme_get_ns_from_dev(dev)->head;
2440 	else
2441 		return disk->private_data;
2442 }
2443 
2444 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2445 		char *buf)
2446 {
2447 	struct nvme_ns_head *head = dev_to_ns_head(dev);
2448 	struct nvme_ns_ids *ids = &head->ids;
2449 	struct nvme_subsystem *subsys = head->subsys;
2450 	int serial_len = sizeof(subsys->serial);
2451 	int model_len = sizeof(subsys->model);
2452 
2453 	if (!uuid_is_null(&ids->uuid))
2454 		return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2455 
2456 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2457 		return sprintf(buf, "eui.%16phN\n", ids->nguid);
2458 
2459 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2460 		return sprintf(buf, "eui.%8phN\n", ids->eui64);
2461 
2462 	while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2463 				  subsys->serial[serial_len - 1] == '\0'))
2464 		serial_len--;
2465 	while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2466 				 subsys->model[model_len - 1] == '\0'))
2467 		model_len--;
2468 
2469 	return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2470 		serial_len, subsys->serial, model_len, subsys->model,
2471 		head->ns_id);
2472 }
2473 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
2474 
2475 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2476 		char *buf)
2477 {
2478 	return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2479 }
2480 static DEVICE_ATTR(nguid, S_IRUGO, nguid_show, NULL);
2481 
2482 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2483 		char *buf)
2484 {
2485 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2486 
2487 	/* For backward compatibility expose the NGUID to userspace if
2488 	 * we have no UUID set
2489 	 */
2490 	if (uuid_is_null(&ids->uuid)) {
2491 		printk_ratelimited(KERN_WARNING
2492 				   "No UUID available providing old NGUID\n");
2493 		return sprintf(buf, "%pU\n", ids->nguid);
2494 	}
2495 	return sprintf(buf, "%pU\n", &ids->uuid);
2496 }
2497 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
2498 
2499 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2500 		char *buf)
2501 {
2502 	return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2503 }
2504 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
2505 
2506 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2507 		char *buf)
2508 {
2509 	return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2510 }
2511 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
2512 
2513 static struct attribute *nvme_ns_id_attrs[] = {
2514 	&dev_attr_wwid.attr,
2515 	&dev_attr_uuid.attr,
2516 	&dev_attr_nguid.attr,
2517 	&dev_attr_eui.attr,
2518 	&dev_attr_nsid.attr,
2519 	NULL,
2520 };
2521 
2522 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2523 		struct attribute *a, int n)
2524 {
2525 	struct device *dev = container_of(kobj, struct device, kobj);
2526 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2527 
2528 	if (a == &dev_attr_uuid.attr) {
2529 		if (uuid_is_null(&ids->uuid) &&
2530 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2531 			return 0;
2532 	}
2533 	if (a == &dev_attr_nguid.attr) {
2534 		if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2535 			return 0;
2536 	}
2537 	if (a == &dev_attr_eui.attr) {
2538 		if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2539 			return 0;
2540 	}
2541 	return a->mode;
2542 }
2543 
2544 const struct attribute_group nvme_ns_id_attr_group = {
2545 	.attrs		= nvme_ns_id_attrs,
2546 	.is_visible	= nvme_ns_id_attrs_are_visible,
2547 };
2548 
2549 #define nvme_show_str_function(field)						\
2550 static ssize_t  field##_show(struct device *dev,				\
2551 			    struct device_attribute *attr, char *buf)		\
2552 {										\
2553         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
2554         return sprintf(buf, "%.*s\n",						\
2555 		(int)sizeof(ctrl->subsys->field), ctrl->subsys->field);		\
2556 }										\
2557 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2558 
2559 nvme_show_str_function(model);
2560 nvme_show_str_function(serial);
2561 nvme_show_str_function(firmware_rev);
2562 
2563 #define nvme_show_int_function(field)						\
2564 static ssize_t  field##_show(struct device *dev,				\
2565 			    struct device_attribute *attr, char *buf)		\
2566 {										\
2567         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
2568         return sprintf(buf, "%d\n", ctrl->field);	\
2569 }										\
2570 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2571 
2572 nvme_show_int_function(cntlid);
2573 
2574 static ssize_t nvme_sysfs_delete(struct device *dev,
2575 				struct device_attribute *attr, const char *buf,
2576 				size_t count)
2577 {
2578 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2579 
2580 	if (device_remove_file_self(dev, attr))
2581 		nvme_delete_ctrl_sync(ctrl);
2582 	return count;
2583 }
2584 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2585 
2586 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2587 					 struct device_attribute *attr,
2588 					 char *buf)
2589 {
2590 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2591 
2592 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2593 }
2594 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2595 
2596 static ssize_t nvme_sysfs_show_state(struct device *dev,
2597 				     struct device_attribute *attr,
2598 				     char *buf)
2599 {
2600 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2601 	static const char *const state_name[] = {
2602 		[NVME_CTRL_NEW]		= "new",
2603 		[NVME_CTRL_LIVE]	= "live",
2604 		[NVME_CTRL_RESETTING]	= "resetting",
2605 		[NVME_CTRL_RECONNECTING]= "reconnecting",
2606 		[NVME_CTRL_DELETING]	= "deleting",
2607 		[NVME_CTRL_DEAD]	= "dead",
2608 	};
2609 
2610 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2611 	    state_name[ctrl->state])
2612 		return sprintf(buf, "%s\n", state_name[ctrl->state]);
2613 
2614 	return sprintf(buf, "unknown state\n");
2615 }
2616 
2617 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2618 
2619 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2620 					 struct device_attribute *attr,
2621 					 char *buf)
2622 {
2623 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2624 
2625 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2626 }
2627 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2628 
2629 static ssize_t nvme_sysfs_show_address(struct device *dev,
2630 					 struct device_attribute *attr,
2631 					 char *buf)
2632 {
2633 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2634 
2635 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2636 }
2637 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2638 
2639 static struct attribute *nvme_dev_attrs[] = {
2640 	&dev_attr_reset_controller.attr,
2641 	&dev_attr_rescan_controller.attr,
2642 	&dev_attr_model.attr,
2643 	&dev_attr_serial.attr,
2644 	&dev_attr_firmware_rev.attr,
2645 	&dev_attr_cntlid.attr,
2646 	&dev_attr_delete_controller.attr,
2647 	&dev_attr_transport.attr,
2648 	&dev_attr_subsysnqn.attr,
2649 	&dev_attr_address.attr,
2650 	&dev_attr_state.attr,
2651 	NULL
2652 };
2653 
2654 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2655 		struct attribute *a, int n)
2656 {
2657 	struct device *dev = container_of(kobj, struct device, kobj);
2658 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2659 
2660 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2661 		return 0;
2662 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2663 		return 0;
2664 
2665 	return a->mode;
2666 }
2667 
2668 static struct attribute_group nvme_dev_attrs_group = {
2669 	.attrs		= nvme_dev_attrs,
2670 	.is_visible	= nvme_dev_attrs_are_visible,
2671 };
2672 
2673 static const struct attribute_group *nvme_dev_attr_groups[] = {
2674 	&nvme_dev_attrs_group,
2675 	NULL,
2676 };
2677 
2678 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2679 		unsigned nsid)
2680 {
2681 	struct nvme_ns_head *h;
2682 
2683 	lockdep_assert_held(&subsys->lock);
2684 
2685 	list_for_each_entry(h, &subsys->nsheads, entry) {
2686 		if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2687 			return h;
2688 	}
2689 
2690 	return NULL;
2691 }
2692 
2693 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2694 		struct nvme_ns_head *new)
2695 {
2696 	struct nvme_ns_head *h;
2697 
2698 	lockdep_assert_held(&subsys->lock);
2699 
2700 	list_for_each_entry(h, &subsys->nsheads, entry) {
2701 		if (nvme_ns_ids_valid(&new->ids) &&
2702 		    nvme_ns_ids_equal(&new->ids, &h->ids))
2703 			return -EINVAL;
2704 	}
2705 
2706 	return 0;
2707 }
2708 
2709 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2710 		unsigned nsid, struct nvme_id_ns *id)
2711 {
2712 	struct nvme_ns_head *head;
2713 	int ret = -ENOMEM;
2714 
2715 	head = kzalloc(sizeof(*head), GFP_KERNEL);
2716 	if (!head)
2717 		goto out;
2718 	ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
2719 	if (ret < 0)
2720 		goto out_free_head;
2721 	head->instance = ret;
2722 	INIT_LIST_HEAD(&head->list);
2723 	init_srcu_struct(&head->srcu);
2724 	head->subsys = ctrl->subsys;
2725 	head->ns_id = nsid;
2726 	kref_init(&head->ref);
2727 
2728 	nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
2729 
2730 	ret = __nvme_check_ids(ctrl->subsys, head);
2731 	if (ret) {
2732 		dev_err(ctrl->device,
2733 			"duplicate IDs for nsid %d\n", nsid);
2734 		goto out_cleanup_srcu;
2735 	}
2736 
2737 	ret = nvme_mpath_alloc_disk(ctrl, head);
2738 	if (ret)
2739 		goto out_cleanup_srcu;
2740 
2741 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
2742 	return head;
2743 out_cleanup_srcu:
2744 	cleanup_srcu_struct(&head->srcu);
2745 	ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
2746 out_free_head:
2747 	kfree(head);
2748 out:
2749 	return ERR_PTR(ret);
2750 }
2751 
2752 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
2753 		struct nvme_id_ns *id, bool *new)
2754 {
2755 	struct nvme_ctrl *ctrl = ns->ctrl;
2756 	bool is_shared = id->nmic & (1 << 0);
2757 	struct nvme_ns_head *head = NULL;
2758 	int ret = 0;
2759 
2760 	mutex_lock(&ctrl->subsys->lock);
2761 	if (is_shared)
2762 		head = __nvme_find_ns_head(ctrl->subsys, nsid);
2763 	if (!head) {
2764 		head = nvme_alloc_ns_head(ctrl, nsid, id);
2765 		if (IS_ERR(head)) {
2766 			ret = PTR_ERR(head);
2767 			goto out_unlock;
2768 		}
2769 
2770 		*new = true;
2771 	} else {
2772 		struct nvme_ns_ids ids;
2773 
2774 		nvme_report_ns_ids(ctrl, nsid, id, &ids);
2775 		if (!nvme_ns_ids_equal(&head->ids, &ids)) {
2776 			dev_err(ctrl->device,
2777 				"IDs don't match for shared namespace %d\n",
2778 					nsid);
2779 			ret = -EINVAL;
2780 			goto out_unlock;
2781 		}
2782 
2783 		*new = false;
2784 	}
2785 
2786 	list_add_tail(&ns->siblings, &head->list);
2787 	ns->head = head;
2788 
2789 out_unlock:
2790 	mutex_unlock(&ctrl->subsys->lock);
2791 	return ret;
2792 }
2793 
2794 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2795 {
2796 	struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2797 	struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2798 
2799 	return nsa->head->ns_id - nsb->head->ns_id;
2800 }
2801 
2802 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2803 {
2804 	struct nvme_ns *ns, *ret = NULL;
2805 
2806 	mutex_lock(&ctrl->namespaces_mutex);
2807 	list_for_each_entry(ns, &ctrl->namespaces, list) {
2808 		if (ns->head->ns_id == nsid) {
2809 			if (!kref_get_unless_zero(&ns->kref))
2810 				continue;
2811 			ret = ns;
2812 			break;
2813 		}
2814 		if (ns->head->ns_id > nsid)
2815 			break;
2816 	}
2817 	mutex_unlock(&ctrl->namespaces_mutex);
2818 	return ret;
2819 }
2820 
2821 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
2822 {
2823 	struct streams_directive_params s;
2824 	int ret;
2825 
2826 	if (!ctrl->nr_streams)
2827 		return 0;
2828 
2829 	ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
2830 	if (ret)
2831 		return ret;
2832 
2833 	ns->sws = le32_to_cpu(s.sws);
2834 	ns->sgs = le16_to_cpu(s.sgs);
2835 
2836 	if (ns->sws) {
2837 		unsigned int bs = 1 << ns->lba_shift;
2838 
2839 		blk_queue_io_min(ns->queue, bs * ns->sws);
2840 		if (ns->sgs)
2841 			blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
2842 	}
2843 
2844 	return 0;
2845 }
2846 
2847 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2848 {
2849 	struct nvme_ns *ns;
2850 	struct gendisk *disk;
2851 	struct nvme_id_ns *id;
2852 	char disk_name[DISK_NAME_LEN];
2853 	int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
2854 	bool new = true;
2855 
2856 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2857 	if (!ns)
2858 		return;
2859 
2860 	ns->queue = blk_mq_init_queue(ctrl->tagset);
2861 	if (IS_ERR(ns->queue))
2862 		goto out_free_ns;
2863 	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2864 	ns->queue->queuedata = ns;
2865 	ns->ctrl = ctrl;
2866 
2867 	kref_init(&ns->kref);
2868 	ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2869 
2870 	blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2871 	nvme_set_queue_limits(ctrl, ns->queue);
2872 	nvme_setup_streams_ns(ctrl, ns);
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 
2884 #ifdef CONFIG_NVME_MULTIPATH
2885 	/*
2886 	 * If multipathing is enabled we need to always use the subsystem
2887 	 * instance number for numbering our devices to avoid conflicts
2888 	 * between subsystems that have multiple controllers and thus use
2889 	 * the multipath-aware subsystem node and those that have a single
2890 	 * controller and use the controller node directly.
2891 	 */
2892 	if (ns->head->disk) {
2893 		sprintf(disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
2894 				ctrl->cntlid, ns->head->instance);
2895 		flags = GENHD_FL_HIDDEN;
2896 	} else {
2897 		sprintf(disk_name, "nvme%dn%d", ctrl->subsys->instance,
2898 				ns->head->instance);
2899 	}
2900 #else
2901 	/*
2902 	 * But without the multipath code enabled, multiple controller per
2903 	 * subsystems are visible as devices and thus we cannot use the
2904 	 * subsystem instance.
2905 	 */
2906 	sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
2907 #endif
2908 
2909 	if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
2910 		if (nvme_nvm_register(ns, disk_name, node)) {
2911 			dev_warn(ctrl->device, "LightNVM init failure\n");
2912 			goto out_unlink_ns;
2913 		}
2914 	}
2915 
2916 	disk = alloc_disk_node(0, node);
2917 	if (!disk)
2918 		goto out_unlink_ns;
2919 
2920 	disk->fops = &nvme_fops;
2921 	disk->private_data = ns;
2922 	disk->queue = ns->queue;
2923 	disk->flags = flags;
2924 	memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
2925 	ns->disk = disk;
2926 
2927 	__nvme_revalidate_disk(disk, id);
2928 
2929 	mutex_lock(&ctrl->namespaces_mutex);
2930 	list_add_tail(&ns->list, &ctrl->namespaces);
2931 	mutex_unlock(&ctrl->namespaces_mutex);
2932 
2933 	nvme_get_ctrl(ctrl);
2934 
2935 	kfree(id);
2936 
2937 	device_add_disk(ctrl->device, ns->disk);
2938 	if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
2939 					&nvme_ns_id_attr_group))
2940 		pr_warn("%s: failed to create sysfs group for identification\n",
2941 			ns->disk->disk_name);
2942 	if (ns->ndev && nvme_nvm_register_sysfs(ns))
2943 		pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
2944 			ns->disk->disk_name);
2945 
2946 	if (new)
2947 		nvme_mpath_add_disk(ns->head);
2948 	nvme_mpath_add_disk_links(ns);
2949 	return;
2950  out_unlink_ns:
2951 	mutex_lock(&ctrl->subsys->lock);
2952 	list_del_rcu(&ns->siblings);
2953 	mutex_unlock(&ctrl->subsys->lock);
2954  out_free_id:
2955 	kfree(id);
2956  out_free_queue:
2957 	blk_cleanup_queue(ns->queue);
2958  out_free_ns:
2959 	kfree(ns);
2960 }
2961 
2962 static void nvme_ns_remove(struct nvme_ns *ns)
2963 {
2964 	struct nvme_ns_head *head = ns->head;
2965 
2966 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
2967 		return;
2968 
2969 	if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
2970 		if (blk_get_integrity(ns->disk))
2971 			blk_integrity_unregister(ns->disk);
2972 		nvme_mpath_remove_disk_links(ns);
2973 		sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
2974 					&nvme_ns_id_attr_group);
2975 		if (ns->ndev)
2976 			nvme_nvm_unregister_sysfs(ns);
2977 		del_gendisk(ns->disk);
2978 		blk_cleanup_queue(ns->queue);
2979 	}
2980 
2981 	mutex_lock(&ns->ctrl->subsys->lock);
2982 	nvme_mpath_clear_current_path(ns);
2983 	if (head)
2984 		list_del_rcu(&ns->siblings);
2985 	mutex_unlock(&ns->ctrl->subsys->lock);
2986 
2987 	mutex_lock(&ns->ctrl->namespaces_mutex);
2988 	list_del_init(&ns->list);
2989 	mutex_unlock(&ns->ctrl->namespaces_mutex);
2990 
2991 	synchronize_srcu(&head->srcu);
2992 	nvme_put_ns(ns);
2993 }
2994 
2995 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2996 {
2997 	struct nvme_ns *ns;
2998 
2999 	ns = nvme_find_get_ns(ctrl, nsid);
3000 	if (ns) {
3001 		if (ns->disk && revalidate_disk(ns->disk))
3002 			nvme_ns_remove(ns);
3003 		nvme_put_ns(ns);
3004 	} else
3005 		nvme_alloc_ns(ctrl, nsid);
3006 }
3007 
3008 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3009 					unsigned nsid)
3010 {
3011 	struct nvme_ns *ns, *next;
3012 
3013 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3014 		if (ns->head->ns_id > nsid)
3015 			nvme_ns_remove(ns);
3016 	}
3017 }
3018 
3019 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3020 {
3021 	struct nvme_ns *ns;
3022 	__le32 *ns_list;
3023 	unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3024 	int ret = 0;
3025 
3026 	ns_list = kzalloc(0x1000, GFP_KERNEL);
3027 	if (!ns_list)
3028 		return -ENOMEM;
3029 
3030 	for (i = 0; i < num_lists; i++) {
3031 		ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3032 		if (ret)
3033 			goto free;
3034 
3035 		for (j = 0; j < min(nn, 1024U); j++) {
3036 			nsid = le32_to_cpu(ns_list[j]);
3037 			if (!nsid)
3038 				goto out;
3039 
3040 			nvme_validate_ns(ctrl, nsid);
3041 
3042 			while (++prev < nsid) {
3043 				ns = nvme_find_get_ns(ctrl, prev);
3044 				if (ns) {
3045 					nvme_ns_remove(ns);
3046 					nvme_put_ns(ns);
3047 				}
3048 			}
3049 		}
3050 		nn -= j;
3051 	}
3052  out:
3053 	nvme_remove_invalid_namespaces(ctrl, prev);
3054  free:
3055 	kfree(ns_list);
3056 	return ret;
3057 }
3058 
3059 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3060 {
3061 	unsigned i;
3062 
3063 	for (i = 1; i <= nn; i++)
3064 		nvme_validate_ns(ctrl, i);
3065 
3066 	nvme_remove_invalid_namespaces(ctrl, nn);
3067 }
3068 
3069 static void nvme_scan_work(struct work_struct *work)
3070 {
3071 	struct nvme_ctrl *ctrl =
3072 		container_of(work, struct nvme_ctrl, scan_work);
3073 	struct nvme_id_ctrl *id;
3074 	unsigned nn;
3075 
3076 	if (ctrl->state != NVME_CTRL_LIVE)
3077 		return;
3078 
3079 	if (nvme_identify_ctrl(ctrl, &id))
3080 		return;
3081 
3082 	nn = le32_to_cpu(id->nn);
3083 	if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3084 	    !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3085 		if (!nvme_scan_ns_list(ctrl, nn))
3086 			goto done;
3087 	}
3088 	nvme_scan_ns_sequential(ctrl, nn);
3089  done:
3090 	mutex_lock(&ctrl->namespaces_mutex);
3091 	list_sort(NULL, &ctrl->namespaces, ns_cmp);
3092 	mutex_unlock(&ctrl->namespaces_mutex);
3093 	kfree(id);
3094 }
3095 
3096 void nvme_queue_scan(struct nvme_ctrl *ctrl)
3097 {
3098 	/*
3099 	 * Do not queue new scan work when a controller is reset during
3100 	 * removal.
3101 	 */
3102 	if (ctrl->state == NVME_CTRL_LIVE)
3103 		queue_work(nvme_wq, &ctrl->scan_work);
3104 }
3105 EXPORT_SYMBOL_GPL(nvme_queue_scan);
3106 
3107 /*
3108  * This function iterates the namespace list unlocked to allow recovery from
3109  * controller failure. It is up to the caller to ensure the namespace list is
3110  * not modified by scan work while this function is executing.
3111  */
3112 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3113 {
3114 	struct nvme_ns *ns, *next;
3115 
3116 	/*
3117 	 * The dead states indicates the controller was not gracefully
3118 	 * disconnected. In that case, we won't be able to flush any data while
3119 	 * removing the namespaces' disks; fail all the queues now to avoid
3120 	 * potentially having to clean up the failed sync later.
3121 	 */
3122 	if (ctrl->state == NVME_CTRL_DEAD)
3123 		nvme_kill_queues(ctrl);
3124 
3125 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
3126 		nvme_ns_remove(ns);
3127 }
3128 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3129 
3130 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3131 {
3132 	char *envp[2] = { NULL, NULL };
3133 	u32 aen_result = ctrl->aen_result;
3134 
3135 	ctrl->aen_result = 0;
3136 	if (!aen_result)
3137 		return;
3138 
3139 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3140 	if (!envp[0])
3141 		return;
3142 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3143 	kfree(envp[0]);
3144 }
3145 
3146 static void nvme_async_event_work(struct work_struct *work)
3147 {
3148 	struct nvme_ctrl *ctrl =
3149 		container_of(work, struct nvme_ctrl, async_event_work);
3150 
3151 	nvme_aen_uevent(ctrl);
3152 	ctrl->ops->submit_async_event(ctrl);
3153 }
3154 
3155 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3156 {
3157 
3158 	u32 csts;
3159 
3160 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3161 		return false;
3162 
3163 	if (csts == ~0)
3164 		return false;
3165 
3166 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3167 }
3168 
3169 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3170 {
3171 	struct nvme_fw_slot_info_log *log;
3172 
3173 	log = kmalloc(sizeof(*log), GFP_KERNEL);
3174 	if (!log)
3175 		return;
3176 
3177 	if (nvme_get_log(ctrl, NVME_LOG_FW_SLOT, log, sizeof(*log)))
3178 		dev_warn(ctrl->device,
3179 				"Get FW SLOT INFO log error\n");
3180 	kfree(log);
3181 }
3182 
3183 static void nvme_fw_act_work(struct work_struct *work)
3184 {
3185 	struct nvme_ctrl *ctrl = container_of(work,
3186 				struct nvme_ctrl, fw_act_work);
3187 	unsigned long fw_act_timeout;
3188 
3189 	if (ctrl->mtfa)
3190 		fw_act_timeout = jiffies +
3191 				msecs_to_jiffies(ctrl->mtfa * 100);
3192 	else
3193 		fw_act_timeout = jiffies +
3194 				msecs_to_jiffies(admin_timeout * 1000);
3195 
3196 	nvme_stop_queues(ctrl);
3197 	while (nvme_ctrl_pp_status(ctrl)) {
3198 		if (time_after(jiffies, fw_act_timeout)) {
3199 			dev_warn(ctrl->device,
3200 				"Fw activation timeout, reset controller\n");
3201 			nvme_reset_ctrl(ctrl);
3202 			break;
3203 		}
3204 		msleep(100);
3205 	}
3206 
3207 	if (ctrl->state != NVME_CTRL_LIVE)
3208 		return;
3209 
3210 	nvme_start_queues(ctrl);
3211 	/* read FW slot information to clear the AER */
3212 	nvme_get_fw_slot_info(ctrl);
3213 }
3214 
3215 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3216 		union nvme_result *res)
3217 {
3218 	u32 result = le32_to_cpu(res->u32);
3219 
3220 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3221 		return;
3222 
3223 	switch (result & 0x7) {
3224 	case NVME_AER_ERROR:
3225 	case NVME_AER_SMART:
3226 	case NVME_AER_CSS:
3227 	case NVME_AER_VS:
3228 		ctrl->aen_result = result;
3229 		break;
3230 	default:
3231 		break;
3232 	}
3233 
3234 	switch (result & 0xff07) {
3235 	case NVME_AER_NOTICE_NS_CHANGED:
3236 		dev_info(ctrl->device, "rescanning\n");
3237 		nvme_queue_scan(ctrl);
3238 		break;
3239 	case NVME_AER_NOTICE_FW_ACT_STARTING:
3240 		queue_work(nvme_wq, &ctrl->fw_act_work);
3241 		break;
3242 	default:
3243 		dev_warn(ctrl->device, "async event result %08x\n", result);
3244 	}
3245 	queue_work(nvme_wq, &ctrl->async_event_work);
3246 }
3247 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3248 
3249 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3250 {
3251 	nvme_stop_keep_alive(ctrl);
3252 	flush_work(&ctrl->async_event_work);
3253 	flush_work(&ctrl->scan_work);
3254 	cancel_work_sync(&ctrl->fw_act_work);
3255 }
3256 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3257 
3258 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3259 {
3260 	if (ctrl->kato)
3261 		nvme_start_keep_alive(ctrl);
3262 
3263 	if (ctrl->queue_count > 1) {
3264 		nvme_queue_scan(ctrl);
3265 		queue_work(nvme_wq, &ctrl->async_event_work);
3266 		nvme_start_queues(ctrl);
3267 	}
3268 }
3269 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3270 
3271 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3272 {
3273 	cdev_device_del(&ctrl->cdev, ctrl->device);
3274 }
3275 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3276 
3277 static void nvme_free_ctrl(struct device *dev)
3278 {
3279 	struct nvme_ctrl *ctrl =
3280 		container_of(dev, struct nvme_ctrl, ctrl_device);
3281 	struct nvme_subsystem *subsys = ctrl->subsys;
3282 
3283 	ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3284 	kfree(ctrl->effects);
3285 
3286 	if (subsys) {
3287 		mutex_lock(&subsys->lock);
3288 		list_del(&ctrl->subsys_entry);
3289 		mutex_unlock(&subsys->lock);
3290 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3291 	}
3292 
3293 	ctrl->ops->free_ctrl(ctrl);
3294 
3295 	if (subsys)
3296 		nvme_put_subsystem(subsys);
3297 }
3298 
3299 /*
3300  * Initialize a NVMe controller structures.  This needs to be called during
3301  * earliest initialization so that we have the initialized structured around
3302  * during probing.
3303  */
3304 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3305 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
3306 {
3307 	int ret;
3308 
3309 	ctrl->state = NVME_CTRL_NEW;
3310 	spin_lock_init(&ctrl->lock);
3311 	INIT_LIST_HEAD(&ctrl->namespaces);
3312 	mutex_init(&ctrl->namespaces_mutex);
3313 	ctrl->dev = dev;
3314 	ctrl->ops = ops;
3315 	ctrl->quirks = quirks;
3316 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3317 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3318 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3319 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3320 
3321 	ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3322 	if (ret < 0)
3323 		goto out;
3324 	ctrl->instance = ret;
3325 
3326 	device_initialize(&ctrl->ctrl_device);
3327 	ctrl->device = &ctrl->ctrl_device;
3328 	ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3329 	ctrl->device->class = nvme_class;
3330 	ctrl->device->parent = ctrl->dev;
3331 	ctrl->device->groups = nvme_dev_attr_groups;
3332 	ctrl->device->release = nvme_free_ctrl;
3333 	dev_set_drvdata(ctrl->device, ctrl);
3334 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3335 	if (ret)
3336 		goto out_release_instance;
3337 
3338 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
3339 	ctrl->cdev.owner = ops->module;
3340 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3341 	if (ret)
3342 		goto out_free_name;
3343 
3344 	/*
3345 	 * Initialize latency tolerance controls.  The sysfs files won't
3346 	 * be visible to userspace unless the device actually supports APST.
3347 	 */
3348 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3349 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3350 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3351 
3352 	return 0;
3353 out_free_name:
3354 	kfree_const(dev->kobj.name);
3355 out_release_instance:
3356 	ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3357 out:
3358 	return ret;
3359 }
3360 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3361 
3362 /**
3363  * nvme_kill_queues(): Ends all namespace queues
3364  * @ctrl: the dead controller that needs to end
3365  *
3366  * Call this function when the driver determines it is unable to get the
3367  * controller in a state capable of servicing IO.
3368  */
3369 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3370 {
3371 	struct nvme_ns *ns;
3372 
3373 	mutex_lock(&ctrl->namespaces_mutex);
3374 
3375 	/* Forcibly unquiesce queues to avoid blocking dispatch */
3376 	if (ctrl->admin_q)
3377 		blk_mq_unquiesce_queue(ctrl->admin_q);
3378 
3379 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3380 		/*
3381 		 * Revalidating a dead namespace sets capacity to 0. This will
3382 		 * end buffered writers dirtying pages that can't be synced.
3383 		 */
3384 		if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
3385 			continue;
3386 		revalidate_disk(ns->disk);
3387 		blk_set_queue_dying(ns->queue);
3388 
3389 		/* Forcibly unquiesce queues to avoid blocking dispatch */
3390 		blk_mq_unquiesce_queue(ns->queue);
3391 	}
3392 	mutex_unlock(&ctrl->namespaces_mutex);
3393 }
3394 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3395 
3396 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3397 {
3398 	struct nvme_ns *ns;
3399 
3400 	mutex_lock(&ctrl->namespaces_mutex);
3401 	list_for_each_entry(ns, &ctrl->namespaces, list)
3402 		blk_mq_unfreeze_queue(ns->queue);
3403 	mutex_unlock(&ctrl->namespaces_mutex);
3404 }
3405 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3406 
3407 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3408 {
3409 	struct nvme_ns *ns;
3410 
3411 	mutex_lock(&ctrl->namespaces_mutex);
3412 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3413 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3414 		if (timeout <= 0)
3415 			break;
3416 	}
3417 	mutex_unlock(&ctrl->namespaces_mutex);
3418 }
3419 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3420 
3421 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3422 {
3423 	struct nvme_ns *ns;
3424 
3425 	mutex_lock(&ctrl->namespaces_mutex);
3426 	list_for_each_entry(ns, &ctrl->namespaces, list)
3427 		blk_mq_freeze_queue_wait(ns->queue);
3428 	mutex_unlock(&ctrl->namespaces_mutex);
3429 }
3430 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3431 
3432 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3433 {
3434 	struct nvme_ns *ns;
3435 
3436 	mutex_lock(&ctrl->namespaces_mutex);
3437 	list_for_each_entry(ns, &ctrl->namespaces, list)
3438 		blk_freeze_queue_start(ns->queue);
3439 	mutex_unlock(&ctrl->namespaces_mutex);
3440 }
3441 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3442 
3443 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3444 {
3445 	struct nvme_ns *ns;
3446 
3447 	mutex_lock(&ctrl->namespaces_mutex);
3448 	list_for_each_entry(ns, &ctrl->namespaces, list)
3449 		blk_mq_quiesce_queue(ns->queue);
3450 	mutex_unlock(&ctrl->namespaces_mutex);
3451 }
3452 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3453 
3454 void nvme_start_queues(struct nvme_ctrl *ctrl)
3455 {
3456 	struct nvme_ns *ns;
3457 
3458 	mutex_lock(&ctrl->namespaces_mutex);
3459 	list_for_each_entry(ns, &ctrl->namespaces, list)
3460 		blk_mq_unquiesce_queue(ns->queue);
3461 	mutex_unlock(&ctrl->namespaces_mutex);
3462 }
3463 EXPORT_SYMBOL_GPL(nvme_start_queues);
3464 
3465 int nvme_reinit_tagset(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set)
3466 {
3467 	if (!ctrl->ops->reinit_request)
3468 		return 0;
3469 
3470 	return blk_mq_tagset_iter(set, set->driver_data,
3471 			ctrl->ops->reinit_request);
3472 }
3473 EXPORT_SYMBOL_GPL(nvme_reinit_tagset);
3474 
3475 int __init nvme_core_init(void)
3476 {
3477 	int result;
3478 
3479 	nvme_wq = alloc_workqueue("nvme-wq",
3480 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3481 	if (!nvme_wq)
3482 		return -ENOMEM;
3483 
3484 	result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3485 	if (result < 0)
3486 		goto destroy_wq;
3487 
3488 	nvme_class = class_create(THIS_MODULE, "nvme");
3489 	if (IS_ERR(nvme_class)) {
3490 		result = PTR_ERR(nvme_class);
3491 		goto unregister_chrdev;
3492 	}
3493 
3494 	nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3495 	if (IS_ERR(nvme_subsys_class)) {
3496 		result = PTR_ERR(nvme_subsys_class);
3497 		goto destroy_class;
3498 	}
3499 	return 0;
3500 
3501 destroy_class:
3502 	class_destroy(nvme_class);
3503 unregister_chrdev:
3504 	unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3505 destroy_wq:
3506 	destroy_workqueue(nvme_wq);
3507 	return result;
3508 }
3509 
3510 void nvme_core_exit(void)
3511 {
3512 	ida_destroy(&nvme_subsystems_ida);
3513 	class_destroy(nvme_subsys_class);
3514 	class_destroy(nvme_class);
3515 	unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3516 	destroy_workqueue(nvme_wq);
3517 }
3518 
3519 MODULE_LICENSE("GPL");
3520 MODULE_VERSION("1.0");
3521 module_init(nvme_core_init);
3522 module_exit(nvme_core_exit);
3523