xref: /openbmc/linux/drivers/mtd/mtdswap.c (revision f87deada)
1 /*
2  * Swap block device support for MTDs
3  * Turns an MTD device into a swap device with block wear leveling
4  *
5  * Copyright © 2007,2011 Nokia Corporation. All rights reserved.
6  *
7  * Authors: Jarkko Lavinen <jarkko.lavinen@nokia.com>
8  *
9  * Based on Richard Purdie's earlier implementation in 2007. Background
10  * support and lock-less operation written by Adrian Hunter.
11  *
12  * This program is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU General Public License
14  * version 2 as published by the Free Software Foundation.
15  *
16  * This program is distributed in the hope that it will be useful, but
17  * WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
19  * General Public License for more details.
20  *
21  * You should have received a copy of the GNU General Public License
22  * along with this program; if not, write to the Free Software
23  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA
24  * 02110-1301 USA
25  */
26 
27 #include <linux/kernel.h>
28 #include <linux/module.h>
29 #include <linux/mtd/mtd.h>
30 #include <linux/mtd/blktrans.h>
31 #include <linux/rbtree.h>
32 #include <linux/sched.h>
33 #include <linux/slab.h>
34 #include <linux/vmalloc.h>
35 #include <linux/genhd.h>
36 #include <linux/swap.h>
37 #include <linux/debugfs.h>
38 #include <linux/seq_file.h>
39 #include <linux/device.h>
40 #include <linux/math64.h>
41 
42 #define MTDSWAP_PREFIX "mtdswap"
43 
44 /*
45  * The number of free eraseblocks when GC should stop
46  */
47 #define CLEAN_BLOCK_THRESHOLD	20
48 
49 /*
50  * Number of free eraseblocks below which GC can also collect low frag
51  * blocks.
52  */
53 #define LOW_FRAG_GC_THRESHOLD	5
54 
55 /*
56  * Wear level cost amortization. We want to do wear leveling on the background
57  * without disturbing gc too much. This is made by defining max GC frequency.
58  * Frequency value 6 means 1/6 of the GC passes will pick an erase block based
59  * on the biggest wear difference rather than the biggest dirtiness.
60  *
61  * The lower freq2 should be chosen so that it makes sure the maximum erase
62  * difference will decrease even if a malicious application is deliberately
63  * trying to make erase differences large.
64  */
65 #define MAX_ERASE_DIFF		4000
66 #define COLLECT_NONDIRTY_BASE	MAX_ERASE_DIFF
67 #define COLLECT_NONDIRTY_FREQ1	6
68 #define COLLECT_NONDIRTY_FREQ2	4
69 
70 #define PAGE_UNDEF		UINT_MAX
71 #define BLOCK_UNDEF		UINT_MAX
72 #define BLOCK_ERROR		(UINT_MAX - 1)
73 #define BLOCK_MAX		(UINT_MAX - 2)
74 
75 #define EBLOCK_BAD		(1 << 0)
76 #define EBLOCK_NOMAGIC		(1 << 1)
77 #define EBLOCK_BITFLIP		(1 << 2)
78 #define EBLOCK_FAILED		(1 << 3)
79 #define EBLOCK_READERR		(1 << 4)
80 #define EBLOCK_IDX_SHIFT	5
81 
82 struct swap_eb {
83 	struct rb_node rb;
84 	struct rb_root *root;
85 
86 	unsigned int flags;
87 	unsigned int active_count;
88 	unsigned int erase_count;
89 	unsigned int pad;		/* speeds up pointer decrement */
90 };
91 
92 #define MTDSWAP_ECNT_MIN(rbroot) (rb_entry(rb_first(rbroot), struct swap_eb, \
93 				rb)->erase_count)
94 #define MTDSWAP_ECNT_MAX(rbroot) (rb_entry(rb_last(rbroot), struct swap_eb, \
95 				rb)->erase_count)
96 
97 struct mtdswap_tree {
98 	struct rb_root root;
99 	unsigned int count;
100 };
101 
102 enum {
103 	MTDSWAP_CLEAN,
104 	MTDSWAP_USED,
105 	MTDSWAP_LOWFRAG,
106 	MTDSWAP_HIFRAG,
107 	MTDSWAP_DIRTY,
108 	MTDSWAP_BITFLIP,
109 	MTDSWAP_FAILING,
110 	MTDSWAP_TREE_CNT,
111 };
112 
113 struct mtdswap_dev {
114 	struct mtd_blktrans_dev *mbd_dev;
115 	struct mtd_info *mtd;
116 	struct device *dev;
117 
118 	unsigned int *page_data;
119 	unsigned int *revmap;
120 
121 	unsigned int eblks;
122 	unsigned int spare_eblks;
123 	unsigned int pages_per_eblk;
124 	unsigned int max_erase_count;
125 	struct swap_eb *eb_data;
126 
127 	struct mtdswap_tree trees[MTDSWAP_TREE_CNT];
128 
129 	unsigned long long sect_read_count;
130 	unsigned long long sect_write_count;
131 	unsigned long long mtd_write_count;
132 	unsigned long long mtd_read_count;
133 	unsigned long long discard_count;
134 	unsigned long long discard_page_count;
135 
136 	unsigned int curr_write_pos;
137 	struct swap_eb *curr_write;
138 
139 	char *page_buf;
140 	char *oob_buf;
141 };
142 
143 struct mtdswap_oobdata {
144 	__le16 magic;
145 	__le32 count;
146 } __packed;
147 
148 #define MTDSWAP_MAGIC_CLEAN	0x2095
149 #define MTDSWAP_MAGIC_DIRTY	(MTDSWAP_MAGIC_CLEAN + 1)
150 #define MTDSWAP_TYPE_CLEAN	0
151 #define MTDSWAP_TYPE_DIRTY	1
152 #define MTDSWAP_OOBSIZE		sizeof(struct mtdswap_oobdata)
153 
154 #define MTDSWAP_ERASE_RETRIES	3 /* Before marking erase block bad */
155 #define MTDSWAP_IO_RETRIES	3
156 
157 enum {
158 	MTDSWAP_SCANNED_CLEAN,
159 	MTDSWAP_SCANNED_DIRTY,
160 	MTDSWAP_SCANNED_BITFLIP,
161 	MTDSWAP_SCANNED_BAD,
162 };
163 
164 /*
165  * In the worst case mtdswap_writesect() has allocated the last clean
166  * page from the current block and is then pre-empted by the GC
167  * thread. The thread can consume a full erase block when moving a
168  * block.
169  */
170 #define MIN_SPARE_EBLOCKS	2
171 #define MIN_ERASE_BLOCKS	(MIN_SPARE_EBLOCKS + 1)
172 
173 #define TREE_ROOT(d, name) (&d->trees[MTDSWAP_ ## name].root)
174 #define TREE_EMPTY(d, name) (TREE_ROOT(d, name)->rb_node == NULL)
175 #define TREE_NONEMPTY(d, name) (!TREE_EMPTY(d, name))
176 #define TREE_COUNT(d, name) (d->trees[MTDSWAP_ ## name].count)
177 
178 #define MTDSWAP_MBD_TO_MTDSWAP(dev) ((struct mtdswap_dev *)dev->priv)
179 
180 static char partitions[128] = "";
181 module_param_string(partitions, partitions, sizeof(partitions), 0444);
182 MODULE_PARM_DESC(partitions, "MTD partition numbers to use as swap "
183 		"partitions=\"1,3,5\"");
184 
185 static unsigned int spare_eblocks = 10;
186 module_param(spare_eblocks, uint, 0444);
187 MODULE_PARM_DESC(spare_eblocks, "Percentage of spare erase blocks for "
188 		"garbage collection (default 10%)");
189 
190 static bool header; /* false */
191 module_param(header, bool, 0444);
192 MODULE_PARM_DESC(header,
193 		"Include builtin swap header (default 0, without header)");
194 
195 static int mtdswap_gc(struct mtdswap_dev *d, unsigned int background);
196 
197 static loff_t mtdswap_eb_offset(struct mtdswap_dev *d, struct swap_eb *eb)
198 {
199 	return (loff_t)(eb - d->eb_data) * d->mtd->erasesize;
200 }
201 
202 static void mtdswap_eb_detach(struct mtdswap_dev *d, struct swap_eb *eb)
203 {
204 	unsigned int oldidx;
205 	struct mtdswap_tree *tp;
206 
207 	if (eb->root) {
208 		tp = container_of(eb->root, struct mtdswap_tree, root);
209 		oldidx = tp - &d->trees[0];
210 
211 		d->trees[oldidx].count--;
212 		rb_erase(&eb->rb, eb->root);
213 	}
214 }
215 
216 static void __mtdswap_rb_add(struct rb_root *root, struct swap_eb *eb)
217 {
218 	struct rb_node **p, *parent = NULL;
219 	struct swap_eb *cur;
220 
221 	p = &root->rb_node;
222 	while (*p) {
223 		parent = *p;
224 		cur = rb_entry(parent, struct swap_eb, rb);
225 		if (eb->erase_count > cur->erase_count)
226 			p = &(*p)->rb_right;
227 		else
228 			p = &(*p)->rb_left;
229 	}
230 
231 	rb_link_node(&eb->rb, parent, p);
232 	rb_insert_color(&eb->rb, root);
233 }
234 
235 static void mtdswap_rb_add(struct mtdswap_dev *d, struct swap_eb *eb, int idx)
236 {
237 	struct rb_root *root;
238 
239 	if (eb->root == &d->trees[idx].root)
240 		return;
241 
242 	mtdswap_eb_detach(d, eb);
243 	root = &d->trees[idx].root;
244 	__mtdswap_rb_add(root, eb);
245 	eb->root = root;
246 	d->trees[idx].count++;
247 }
248 
249 static struct rb_node *mtdswap_rb_index(struct rb_root *root, unsigned int idx)
250 {
251 	struct rb_node *p;
252 	unsigned int i;
253 
254 	p = rb_first(root);
255 	i = 0;
256 	while (i < idx && p) {
257 		p = rb_next(p);
258 		i++;
259 	}
260 
261 	return p;
262 }
263 
264 static int mtdswap_handle_badblock(struct mtdswap_dev *d, struct swap_eb *eb)
265 {
266 	int ret;
267 	loff_t offset;
268 
269 	d->spare_eblks--;
270 	eb->flags |= EBLOCK_BAD;
271 	mtdswap_eb_detach(d, eb);
272 	eb->root = NULL;
273 
274 	/* badblocks not supported */
275 	if (!mtd_can_have_bb(d->mtd))
276 		return 1;
277 
278 	offset = mtdswap_eb_offset(d, eb);
279 	dev_warn(d->dev, "Marking bad block at %08llx\n", offset);
280 	ret = mtd_block_markbad(d->mtd, offset);
281 
282 	if (ret) {
283 		dev_warn(d->dev, "Mark block bad failed for block at %08llx "
284 			"error %d\n", offset, ret);
285 		return ret;
286 	}
287 
288 	return 1;
289 
290 }
291 
292 static int mtdswap_handle_write_error(struct mtdswap_dev *d, struct swap_eb *eb)
293 {
294 	unsigned int marked = eb->flags & EBLOCK_FAILED;
295 	struct swap_eb *curr_write = d->curr_write;
296 
297 	eb->flags |= EBLOCK_FAILED;
298 	if (curr_write == eb) {
299 		d->curr_write = NULL;
300 
301 		if (!marked && d->curr_write_pos != 0) {
302 			mtdswap_rb_add(d, eb, MTDSWAP_FAILING);
303 			return 0;
304 		}
305 	}
306 
307 	return mtdswap_handle_badblock(d, eb);
308 }
309 
310 static int mtdswap_read_oob(struct mtdswap_dev *d, loff_t from,
311 			struct mtd_oob_ops *ops)
312 {
313 	int ret = mtd_read_oob(d->mtd, from, ops);
314 
315 	if (mtd_is_bitflip(ret))
316 		return ret;
317 
318 	if (ret) {
319 		dev_warn(d->dev, "Read OOB failed %d for block at %08llx\n",
320 			ret, from);
321 		return ret;
322 	}
323 
324 	if (ops->oobretlen < ops->ooblen) {
325 		dev_warn(d->dev, "Read OOB return short read (%zd bytes not "
326 			"%zd) for block at %08llx\n",
327 			ops->oobretlen, ops->ooblen, from);
328 		return -EIO;
329 	}
330 
331 	return 0;
332 }
333 
334 static int mtdswap_read_markers(struct mtdswap_dev *d, struct swap_eb *eb)
335 {
336 	struct mtdswap_oobdata *data, *data2;
337 	int ret;
338 	loff_t offset;
339 	struct mtd_oob_ops ops;
340 
341 	offset = mtdswap_eb_offset(d, eb);
342 
343 	/* Check first if the block is bad. */
344 	if (mtd_can_have_bb(d->mtd) && mtd_block_isbad(d->mtd, offset))
345 		return MTDSWAP_SCANNED_BAD;
346 
347 	ops.ooblen = 2 * d->mtd->oobavail;
348 	ops.oobbuf = d->oob_buf;
349 	ops.ooboffs = 0;
350 	ops.datbuf = NULL;
351 	ops.mode = MTD_OPS_AUTO_OOB;
352 
353 	ret = mtdswap_read_oob(d, offset, &ops);
354 
355 	if (ret && !mtd_is_bitflip(ret))
356 		return ret;
357 
358 	data = (struct mtdswap_oobdata *)d->oob_buf;
359 	data2 = (struct mtdswap_oobdata *)
360 		(d->oob_buf + d->mtd->oobavail);
361 
362 	if (le16_to_cpu(data->magic) == MTDSWAP_MAGIC_CLEAN) {
363 		eb->erase_count = le32_to_cpu(data->count);
364 		if (mtd_is_bitflip(ret))
365 			ret = MTDSWAP_SCANNED_BITFLIP;
366 		else {
367 			if (le16_to_cpu(data2->magic) == MTDSWAP_MAGIC_DIRTY)
368 				ret = MTDSWAP_SCANNED_DIRTY;
369 			else
370 				ret = MTDSWAP_SCANNED_CLEAN;
371 		}
372 	} else {
373 		eb->flags |= EBLOCK_NOMAGIC;
374 		ret = MTDSWAP_SCANNED_DIRTY;
375 	}
376 
377 	return ret;
378 }
379 
380 static int mtdswap_write_marker(struct mtdswap_dev *d, struct swap_eb *eb,
381 				u16 marker)
382 {
383 	struct mtdswap_oobdata n;
384 	int ret;
385 	loff_t offset;
386 	struct mtd_oob_ops ops;
387 
388 	ops.ooboffs = 0;
389 	ops.oobbuf = (uint8_t *)&n;
390 	ops.mode = MTD_OPS_AUTO_OOB;
391 	ops.datbuf = NULL;
392 
393 	if (marker == MTDSWAP_TYPE_CLEAN) {
394 		n.magic = cpu_to_le16(MTDSWAP_MAGIC_CLEAN);
395 		n.count = cpu_to_le32(eb->erase_count);
396 		ops.ooblen = MTDSWAP_OOBSIZE;
397 		offset = mtdswap_eb_offset(d, eb);
398 	} else {
399 		n.magic = cpu_to_le16(MTDSWAP_MAGIC_DIRTY);
400 		ops.ooblen = sizeof(n.magic);
401 		offset = mtdswap_eb_offset(d, eb) + d->mtd->writesize;
402 	}
403 
404 	ret = mtd_write_oob(d->mtd, offset, &ops);
405 
406 	if (ret) {
407 		dev_warn(d->dev, "Write OOB failed for block at %08llx "
408 			"error %d\n", offset, ret);
409 		if (ret == -EIO || mtd_is_eccerr(ret))
410 			mtdswap_handle_write_error(d, eb);
411 		return ret;
412 	}
413 
414 	if (ops.oobretlen != ops.ooblen) {
415 		dev_warn(d->dev, "Short OOB write for block at %08llx: "
416 			"%zd not %zd\n",
417 			offset, ops.oobretlen, ops.ooblen);
418 		return ret;
419 	}
420 
421 	return 0;
422 }
423 
424 /*
425  * Are there any erase blocks without MAGIC_CLEAN header, presumably
426  * because power was cut off after erase but before header write? We
427  * need to guestimate the erase count.
428  */
429 static void mtdswap_check_counts(struct mtdswap_dev *d)
430 {
431 	struct rb_root hist_root = RB_ROOT;
432 	struct rb_node *medrb;
433 	struct swap_eb *eb;
434 	unsigned int i, cnt, median;
435 
436 	cnt = 0;
437 	for (i = 0; i < d->eblks; i++) {
438 		eb = d->eb_data + i;
439 
440 		if (eb->flags & (EBLOCK_NOMAGIC | EBLOCK_BAD | EBLOCK_READERR))
441 			continue;
442 
443 		__mtdswap_rb_add(&hist_root, eb);
444 		cnt++;
445 	}
446 
447 	if (cnt == 0)
448 		return;
449 
450 	medrb = mtdswap_rb_index(&hist_root, cnt / 2);
451 	median = rb_entry(medrb, struct swap_eb, rb)->erase_count;
452 
453 	d->max_erase_count = MTDSWAP_ECNT_MAX(&hist_root);
454 
455 	for (i = 0; i < d->eblks; i++) {
456 		eb = d->eb_data + i;
457 
458 		if (eb->flags & (EBLOCK_NOMAGIC | EBLOCK_READERR))
459 			eb->erase_count = median;
460 
461 		if (eb->flags & (EBLOCK_NOMAGIC | EBLOCK_BAD | EBLOCK_READERR))
462 			continue;
463 
464 		rb_erase(&eb->rb, &hist_root);
465 	}
466 }
467 
468 static void mtdswap_scan_eblks(struct mtdswap_dev *d)
469 {
470 	int status;
471 	unsigned int i, idx;
472 	struct swap_eb *eb;
473 
474 	for (i = 0; i < d->eblks; i++) {
475 		eb = d->eb_data + i;
476 
477 		status = mtdswap_read_markers(d, eb);
478 		if (status < 0)
479 			eb->flags |= EBLOCK_READERR;
480 		else if (status == MTDSWAP_SCANNED_BAD) {
481 			eb->flags |= EBLOCK_BAD;
482 			continue;
483 		}
484 
485 		switch (status) {
486 		case MTDSWAP_SCANNED_CLEAN:
487 			idx = MTDSWAP_CLEAN;
488 			break;
489 		case MTDSWAP_SCANNED_DIRTY:
490 		case MTDSWAP_SCANNED_BITFLIP:
491 			idx = MTDSWAP_DIRTY;
492 			break;
493 		default:
494 			idx = MTDSWAP_FAILING;
495 		}
496 
497 		eb->flags |= (idx << EBLOCK_IDX_SHIFT);
498 	}
499 
500 	mtdswap_check_counts(d);
501 
502 	for (i = 0; i < d->eblks; i++) {
503 		eb = d->eb_data + i;
504 
505 		if (eb->flags & EBLOCK_BAD)
506 			continue;
507 
508 		idx = eb->flags >> EBLOCK_IDX_SHIFT;
509 		mtdswap_rb_add(d, eb, idx);
510 	}
511 }
512 
513 /*
514  * Place eblk into a tree corresponding to its number of active blocks
515  * it contains.
516  */
517 static void mtdswap_store_eb(struct mtdswap_dev *d, struct swap_eb *eb)
518 {
519 	unsigned int weight = eb->active_count;
520 	unsigned int maxweight = d->pages_per_eblk;
521 
522 	if (eb == d->curr_write)
523 		return;
524 
525 	if (eb->flags & EBLOCK_BITFLIP)
526 		mtdswap_rb_add(d, eb, MTDSWAP_BITFLIP);
527 	else if (eb->flags & (EBLOCK_READERR | EBLOCK_FAILED))
528 		mtdswap_rb_add(d, eb, MTDSWAP_FAILING);
529 	if (weight == maxweight)
530 		mtdswap_rb_add(d, eb, MTDSWAP_USED);
531 	else if (weight == 0)
532 		mtdswap_rb_add(d, eb, MTDSWAP_DIRTY);
533 	else if (weight > (maxweight/2))
534 		mtdswap_rb_add(d, eb, MTDSWAP_LOWFRAG);
535 	else
536 		mtdswap_rb_add(d, eb, MTDSWAP_HIFRAG);
537 }
538 
539 
540 static void mtdswap_erase_callback(struct erase_info *done)
541 {
542 	wait_queue_head_t *wait_q = (wait_queue_head_t *)done->priv;
543 	wake_up(wait_q);
544 }
545 
546 static int mtdswap_erase_block(struct mtdswap_dev *d, struct swap_eb *eb)
547 {
548 	struct mtd_info *mtd = d->mtd;
549 	struct erase_info erase;
550 	wait_queue_head_t wq;
551 	unsigned int retries = 0;
552 	int ret;
553 
554 	eb->erase_count++;
555 	if (eb->erase_count > d->max_erase_count)
556 		d->max_erase_count = eb->erase_count;
557 
558 retry:
559 	init_waitqueue_head(&wq);
560 	memset(&erase, 0, sizeof(struct erase_info));
561 
562 	erase.mtd	= mtd;
563 	erase.callback	= mtdswap_erase_callback;
564 	erase.addr	= mtdswap_eb_offset(d, eb);
565 	erase.len	= mtd->erasesize;
566 	erase.priv	= (u_long)&wq;
567 
568 	ret = mtd_erase(mtd, &erase);
569 	if (ret) {
570 		if (retries++ < MTDSWAP_ERASE_RETRIES) {
571 			dev_warn(d->dev,
572 				"erase of erase block %#llx on %s failed",
573 				erase.addr, mtd->name);
574 			yield();
575 			goto retry;
576 		}
577 
578 		dev_err(d->dev, "Cannot erase erase block %#llx on %s\n",
579 			erase.addr, mtd->name);
580 
581 		mtdswap_handle_badblock(d, eb);
582 		return -EIO;
583 	}
584 
585 	ret = wait_event_interruptible(wq, erase.state == MTD_ERASE_DONE ||
586 					   erase.state == MTD_ERASE_FAILED);
587 	if (ret) {
588 		dev_err(d->dev, "Interrupted erase block %#llx erasure on %s\n",
589 			erase.addr, mtd->name);
590 		return -EINTR;
591 	}
592 
593 	if (erase.state == MTD_ERASE_FAILED) {
594 		if (retries++ < MTDSWAP_ERASE_RETRIES) {
595 			dev_warn(d->dev,
596 				"erase of erase block %#llx on %s failed",
597 				erase.addr, mtd->name);
598 			yield();
599 			goto retry;
600 		}
601 
602 		mtdswap_handle_badblock(d, eb);
603 		return -EIO;
604 	}
605 
606 	return 0;
607 }
608 
609 static int mtdswap_map_free_block(struct mtdswap_dev *d, unsigned int page,
610 				unsigned int *block)
611 {
612 	int ret;
613 	struct swap_eb *old_eb = d->curr_write;
614 	struct rb_root *clean_root;
615 	struct swap_eb *eb;
616 
617 	if (old_eb == NULL || d->curr_write_pos >= d->pages_per_eblk) {
618 		do {
619 			if (TREE_EMPTY(d, CLEAN))
620 				return -ENOSPC;
621 
622 			clean_root = TREE_ROOT(d, CLEAN);
623 			eb = rb_entry(rb_first(clean_root), struct swap_eb, rb);
624 			rb_erase(&eb->rb, clean_root);
625 			eb->root = NULL;
626 			TREE_COUNT(d, CLEAN)--;
627 
628 			ret = mtdswap_write_marker(d, eb, MTDSWAP_TYPE_DIRTY);
629 		} while (ret == -EIO || mtd_is_eccerr(ret));
630 
631 		if (ret)
632 			return ret;
633 
634 		d->curr_write_pos = 0;
635 		d->curr_write = eb;
636 		if (old_eb)
637 			mtdswap_store_eb(d, old_eb);
638 	}
639 
640 	*block = (d->curr_write - d->eb_data) * d->pages_per_eblk +
641 		d->curr_write_pos;
642 
643 	d->curr_write->active_count++;
644 	d->revmap[*block] = page;
645 	d->curr_write_pos++;
646 
647 	return 0;
648 }
649 
650 static unsigned int mtdswap_free_page_cnt(struct mtdswap_dev *d)
651 {
652 	return TREE_COUNT(d, CLEAN) * d->pages_per_eblk +
653 		d->pages_per_eblk - d->curr_write_pos;
654 }
655 
656 static unsigned int mtdswap_enough_free_pages(struct mtdswap_dev *d)
657 {
658 	return mtdswap_free_page_cnt(d) > d->pages_per_eblk;
659 }
660 
661 static int mtdswap_write_block(struct mtdswap_dev *d, char *buf,
662 			unsigned int page, unsigned int *bp, int gc_context)
663 {
664 	struct mtd_info *mtd = d->mtd;
665 	struct swap_eb *eb;
666 	size_t retlen;
667 	loff_t writepos;
668 	int ret;
669 
670 retry:
671 	if (!gc_context)
672 		while (!mtdswap_enough_free_pages(d))
673 			if (mtdswap_gc(d, 0) > 0)
674 				return -ENOSPC;
675 
676 	ret = mtdswap_map_free_block(d, page, bp);
677 	eb = d->eb_data + (*bp / d->pages_per_eblk);
678 
679 	if (ret == -EIO || mtd_is_eccerr(ret)) {
680 		d->curr_write = NULL;
681 		eb->active_count--;
682 		d->revmap[*bp] = PAGE_UNDEF;
683 		goto retry;
684 	}
685 
686 	if (ret < 0)
687 		return ret;
688 
689 	writepos = (loff_t)*bp << PAGE_SHIFT;
690 	ret =  mtd_write(mtd, writepos, PAGE_SIZE, &retlen, buf);
691 	if (ret == -EIO || mtd_is_eccerr(ret)) {
692 		d->curr_write_pos--;
693 		eb->active_count--;
694 		d->revmap[*bp] = PAGE_UNDEF;
695 		mtdswap_handle_write_error(d, eb);
696 		goto retry;
697 	}
698 
699 	if (ret < 0) {
700 		dev_err(d->dev, "Write to MTD device failed: %d (%zd written)",
701 			ret, retlen);
702 		goto err;
703 	}
704 
705 	if (retlen != PAGE_SIZE) {
706 		dev_err(d->dev, "Short write to MTD device: %zd written",
707 			retlen);
708 		ret = -EIO;
709 		goto err;
710 	}
711 
712 	return ret;
713 
714 err:
715 	d->curr_write_pos--;
716 	eb->active_count--;
717 	d->revmap[*bp] = PAGE_UNDEF;
718 
719 	return ret;
720 }
721 
722 static int mtdswap_move_block(struct mtdswap_dev *d, unsigned int oldblock,
723 		unsigned int *newblock)
724 {
725 	struct mtd_info *mtd = d->mtd;
726 	struct swap_eb *eb, *oldeb;
727 	int ret;
728 	size_t retlen;
729 	unsigned int page, retries;
730 	loff_t readpos;
731 
732 	page = d->revmap[oldblock];
733 	readpos = (loff_t) oldblock << PAGE_SHIFT;
734 	retries = 0;
735 
736 retry:
737 	ret = mtd_read(mtd, readpos, PAGE_SIZE, &retlen, d->page_buf);
738 
739 	if (ret < 0 && !mtd_is_bitflip(ret)) {
740 		oldeb = d->eb_data + oldblock / d->pages_per_eblk;
741 		oldeb->flags |= EBLOCK_READERR;
742 
743 		dev_err(d->dev, "Read Error: %d (block %u)\n", ret,
744 			oldblock);
745 		retries++;
746 		if (retries < MTDSWAP_IO_RETRIES)
747 			goto retry;
748 
749 		goto read_error;
750 	}
751 
752 	if (retlen != PAGE_SIZE) {
753 		dev_err(d->dev, "Short read: %zd (block %u)\n", retlen,
754 		       oldblock);
755 		ret = -EIO;
756 		goto read_error;
757 	}
758 
759 	ret = mtdswap_write_block(d, d->page_buf, page, newblock, 1);
760 	if (ret < 0) {
761 		d->page_data[page] = BLOCK_ERROR;
762 		dev_err(d->dev, "Write error: %d\n", ret);
763 		return ret;
764 	}
765 
766 	eb = d->eb_data + *newblock / d->pages_per_eblk;
767 	d->page_data[page] = *newblock;
768 	d->revmap[oldblock] = PAGE_UNDEF;
769 	eb = d->eb_data + oldblock / d->pages_per_eblk;
770 	eb->active_count--;
771 
772 	return 0;
773 
774 read_error:
775 	d->page_data[page] = BLOCK_ERROR;
776 	d->revmap[oldblock] = PAGE_UNDEF;
777 	return ret;
778 }
779 
780 static int mtdswap_gc_eblock(struct mtdswap_dev *d, struct swap_eb *eb)
781 {
782 	unsigned int i, block, eblk_base, newblock;
783 	int ret, errcode;
784 
785 	errcode = 0;
786 	eblk_base = (eb - d->eb_data) * d->pages_per_eblk;
787 
788 	for (i = 0; i < d->pages_per_eblk; i++) {
789 		if (d->spare_eblks < MIN_SPARE_EBLOCKS)
790 			return -ENOSPC;
791 
792 		block = eblk_base + i;
793 		if (d->revmap[block] == PAGE_UNDEF)
794 			continue;
795 
796 		ret = mtdswap_move_block(d, block, &newblock);
797 		if (ret < 0 && !errcode)
798 			errcode = ret;
799 	}
800 
801 	return errcode;
802 }
803 
804 static int __mtdswap_choose_gc_tree(struct mtdswap_dev *d)
805 {
806 	int idx, stopat;
807 
808 	if (TREE_COUNT(d, CLEAN) < LOW_FRAG_GC_THRESHOLD)
809 		stopat = MTDSWAP_LOWFRAG;
810 	else
811 		stopat = MTDSWAP_HIFRAG;
812 
813 	for (idx = MTDSWAP_BITFLIP; idx >= stopat; idx--)
814 		if (d->trees[idx].root.rb_node != NULL)
815 			return idx;
816 
817 	return -1;
818 }
819 
820 static int mtdswap_wlfreq(unsigned int maxdiff)
821 {
822 	unsigned int h, x, y, dist, base;
823 
824 	/*
825 	 * Calculate linear ramp down from f1 to f2 when maxdiff goes from
826 	 * MAX_ERASE_DIFF to MAX_ERASE_DIFF + COLLECT_NONDIRTY_BASE.  Similar
827 	 * to triangle with height f1 - f1 and width COLLECT_NONDIRTY_BASE.
828 	 */
829 
830 	dist = maxdiff - MAX_ERASE_DIFF;
831 	if (dist > COLLECT_NONDIRTY_BASE)
832 		dist = COLLECT_NONDIRTY_BASE;
833 
834 	/*
835 	 * Modelling the slop as right angular triangle with base
836 	 * COLLECT_NONDIRTY_BASE and height freq1 - freq2. The ratio y/x is
837 	 * equal to the ratio h/base.
838 	 */
839 	h = COLLECT_NONDIRTY_FREQ1 - COLLECT_NONDIRTY_FREQ2;
840 	base = COLLECT_NONDIRTY_BASE;
841 
842 	x = dist - base;
843 	y = (x * h + base / 2) / base;
844 
845 	return COLLECT_NONDIRTY_FREQ2 + y;
846 }
847 
848 static int mtdswap_choose_wl_tree(struct mtdswap_dev *d)
849 {
850 	static unsigned int pick_cnt;
851 	unsigned int i, idx = -1, wear, max;
852 	struct rb_root *root;
853 
854 	max = 0;
855 	for (i = 0; i <= MTDSWAP_DIRTY; i++) {
856 		root = &d->trees[i].root;
857 		if (root->rb_node == NULL)
858 			continue;
859 
860 		wear = d->max_erase_count - MTDSWAP_ECNT_MIN(root);
861 		if (wear > max) {
862 			max = wear;
863 			idx = i;
864 		}
865 	}
866 
867 	if (max > MAX_ERASE_DIFF && pick_cnt >= mtdswap_wlfreq(max) - 1) {
868 		pick_cnt = 0;
869 		return idx;
870 	}
871 
872 	pick_cnt++;
873 	return -1;
874 }
875 
876 static int mtdswap_choose_gc_tree(struct mtdswap_dev *d,
877 				unsigned int background)
878 {
879 	int idx;
880 
881 	if (TREE_NONEMPTY(d, FAILING) &&
882 		(background || (TREE_EMPTY(d, CLEAN) && TREE_EMPTY(d, DIRTY))))
883 		return MTDSWAP_FAILING;
884 
885 	idx = mtdswap_choose_wl_tree(d);
886 	if (idx >= MTDSWAP_CLEAN)
887 		return idx;
888 
889 	return __mtdswap_choose_gc_tree(d);
890 }
891 
892 static struct swap_eb *mtdswap_pick_gc_eblk(struct mtdswap_dev *d,
893 					unsigned int background)
894 {
895 	struct rb_root *rp = NULL;
896 	struct swap_eb *eb = NULL;
897 	int idx;
898 
899 	if (background && TREE_COUNT(d, CLEAN) > CLEAN_BLOCK_THRESHOLD &&
900 		TREE_EMPTY(d, DIRTY) && TREE_EMPTY(d, FAILING))
901 		return NULL;
902 
903 	idx = mtdswap_choose_gc_tree(d, background);
904 	if (idx < 0)
905 		return NULL;
906 
907 	rp = &d->trees[idx].root;
908 	eb = rb_entry(rb_first(rp), struct swap_eb, rb);
909 
910 	rb_erase(&eb->rb, rp);
911 	eb->root = NULL;
912 	d->trees[idx].count--;
913 	return eb;
914 }
915 
916 static unsigned int mtdswap_test_patt(unsigned int i)
917 {
918 	return i % 2 ? 0x55555555 : 0xAAAAAAAA;
919 }
920 
921 static unsigned int mtdswap_eblk_passes(struct mtdswap_dev *d,
922 					struct swap_eb *eb)
923 {
924 	struct mtd_info *mtd = d->mtd;
925 	unsigned int test, i, j, patt, mtd_pages;
926 	loff_t base, pos;
927 	unsigned int *p1 = (unsigned int *)d->page_buf;
928 	unsigned char *p2 = (unsigned char *)d->oob_buf;
929 	struct mtd_oob_ops ops;
930 	int ret;
931 
932 	ops.mode = MTD_OPS_AUTO_OOB;
933 	ops.len = mtd->writesize;
934 	ops.ooblen = mtd->oobavail;
935 	ops.ooboffs = 0;
936 	ops.datbuf = d->page_buf;
937 	ops.oobbuf = d->oob_buf;
938 	base = mtdswap_eb_offset(d, eb);
939 	mtd_pages = d->pages_per_eblk * PAGE_SIZE / mtd->writesize;
940 
941 	for (test = 0; test < 2; test++) {
942 		pos = base;
943 		for (i = 0; i < mtd_pages; i++) {
944 			patt = mtdswap_test_patt(test + i);
945 			memset(d->page_buf, patt, mtd->writesize);
946 			memset(d->oob_buf, patt, mtd->oobavail);
947 			ret = mtd_write_oob(mtd, pos, &ops);
948 			if (ret)
949 				goto error;
950 
951 			pos += mtd->writesize;
952 		}
953 
954 		pos = base;
955 		for (i = 0; i < mtd_pages; i++) {
956 			ret = mtd_read_oob(mtd, pos, &ops);
957 			if (ret)
958 				goto error;
959 
960 			patt = mtdswap_test_patt(test + i);
961 			for (j = 0; j < mtd->writesize/sizeof(int); j++)
962 				if (p1[j] != patt)
963 					goto error;
964 
965 			for (j = 0; j < mtd->oobavail; j++)
966 				if (p2[j] != (unsigned char)patt)
967 					goto error;
968 
969 			pos += mtd->writesize;
970 		}
971 
972 		ret = mtdswap_erase_block(d, eb);
973 		if (ret)
974 			goto error;
975 	}
976 
977 	eb->flags &= ~EBLOCK_READERR;
978 	return 1;
979 
980 error:
981 	mtdswap_handle_badblock(d, eb);
982 	return 0;
983 }
984 
985 static int mtdswap_gc(struct mtdswap_dev *d, unsigned int background)
986 {
987 	struct swap_eb *eb;
988 	int ret;
989 
990 	if (d->spare_eblks < MIN_SPARE_EBLOCKS)
991 		return 1;
992 
993 	eb = mtdswap_pick_gc_eblk(d, background);
994 	if (!eb)
995 		return 1;
996 
997 	ret = mtdswap_gc_eblock(d, eb);
998 	if (ret == -ENOSPC)
999 		return 1;
1000 
1001 	if (eb->flags & EBLOCK_FAILED) {
1002 		mtdswap_handle_badblock(d, eb);
1003 		return 0;
1004 	}
1005 
1006 	eb->flags &= ~EBLOCK_BITFLIP;
1007 	ret = mtdswap_erase_block(d, eb);
1008 	if ((eb->flags & EBLOCK_READERR) &&
1009 		(ret || !mtdswap_eblk_passes(d, eb)))
1010 		return 0;
1011 
1012 	if (ret == 0)
1013 		ret = mtdswap_write_marker(d, eb, MTDSWAP_TYPE_CLEAN);
1014 
1015 	if (ret == 0)
1016 		mtdswap_rb_add(d, eb, MTDSWAP_CLEAN);
1017 	else if (ret != -EIO && !mtd_is_eccerr(ret))
1018 		mtdswap_rb_add(d, eb, MTDSWAP_DIRTY);
1019 
1020 	return 0;
1021 }
1022 
1023 static void mtdswap_background(struct mtd_blktrans_dev *dev)
1024 {
1025 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1026 	int ret;
1027 
1028 	while (1) {
1029 		ret = mtdswap_gc(d, 1);
1030 		if (ret || mtd_blktrans_cease_background(dev))
1031 			return;
1032 	}
1033 }
1034 
1035 static void mtdswap_cleanup(struct mtdswap_dev *d)
1036 {
1037 	vfree(d->eb_data);
1038 	vfree(d->revmap);
1039 	vfree(d->page_data);
1040 	kfree(d->oob_buf);
1041 	kfree(d->page_buf);
1042 }
1043 
1044 static int mtdswap_flush(struct mtd_blktrans_dev *dev)
1045 {
1046 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1047 
1048 	mtd_sync(d->mtd);
1049 	return 0;
1050 }
1051 
1052 static unsigned int mtdswap_badblocks(struct mtd_info *mtd, uint64_t size)
1053 {
1054 	loff_t offset;
1055 	unsigned int badcnt;
1056 
1057 	badcnt = 0;
1058 
1059 	if (mtd_can_have_bb(mtd))
1060 		for (offset = 0; offset < size; offset += mtd->erasesize)
1061 			if (mtd_block_isbad(mtd, offset))
1062 				badcnt++;
1063 
1064 	return badcnt;
1065 }
1066 
1067 static int mtdswap_writesect(struct mtd_blktrans_dev *dev,
1068 			unsigned long page, char *buf)
1069 {
1070 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1071 	unsigned int newblock, mapped;
1072 	struct swap_eb *eb;
1073 	int ret;
1074 
1075 	d->sect_write_count++;
1076 
1077 	if (d->spare_eblks < MIN_SPARE_EBLOCKS)
1078 		return -ENOSPC;
1079 
1080 	if (header) {
1081 		/* Ignore writes to the header page */
1082 		if (unlikely(page == 0))
1083 			return 0;
1084 
1085 		page--;
1086 	}
1087 
1088 	mapped = d->page_data[page];
1089 	if (mapped <= BLOCK_MAX) {
1090 		eb = d->eb_data + (mapped / d->pages_per_eblk);
1091 		eb->active_count--;
1092 		mtdswap_store_eb(d, eb);
1093 		d->page_data[page] = BLOCK_UNDEF;
1094 		d->revmap[mapped] = PAGE_UNDEF;
1095 	}
1096 
1097 	ret = mtdswap_write_block(d, buf, page, &newblock, 0);
1098 	d->mtd_write_count++;
1099 
1100 	if (ret < 0)
1101 		return ret;
1102 
1103 	eb = d->eb_data + (newblock / d->pages_per_eblk);
1104 	d->page_data[page] = newblock;
1105 
1106 	return 0;
1107 }
1108 
1109 /* Provide a dummy swap header for the kernel */
1110 static int mtdswap_auto_header(struct mtdswap_dev *d, char *buf)
1111 {
1112 	union swap_header *hd = (union swap_header *)(buf);
1113 
1114 	memset(buf, 0, PAGE_SIZE - 10);
1115 
1116 	hd->info.version = 1;
1117 	hd->info.last_page = d->mbd_dev->size - 1;
1118 	hd->info.nr_badpages = 0;
1119 
1120 	memcpy(buf + PAGE_SIZE - 10, "SWAPSPACE2", 10);
1121 
1122 	return 0;
1123 }
1124 
1125 static int mtdswap_readsect(struct mtd_blktrans_dev *dev,
1126 			unsigned long page, char *buf)
1127 {
1128 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1129 	struct mtd_info *mtd = d->mtd;
1130 	unsigned int realblock, retries;
1131 	loff_t readpos;
1132 	struct swap_eb *eb;
1133 	size_t retlen;
1134 	int ret;
1135 
1136 	d->sect_read_count++;
1137 
1138 	if (header) {
1139 		if (unlikely(page == 0))
1140 			return mtdswap_auto_header(d, buf);
1141 
1142 		page--;
1143 	}
1144 
1145 	realblock = d->page_data[page];
1146 	if (realblock > BLOCK_MAX) {
1147 		memset(buf, 0x0, PAGE_SIZE);
1148 		if (realblock == BLOCK_UNDEF)
1149 			return 0;
1150 		else
1151 			return -EIO;
1152 	}
1153 
1154 	eb = d->eb_data + (realblock / d->pages_per_eblk);
1155 	BUG_ON(d->revmap[realblock] == PAGE_UNDEF);
1156 
1157 	readpos = (loff_t)realblock << PAGE_SHIFT;
1158 	retries = 0;
1159 
1160 retry:
1161 	ret = mtd_read(mtd, readpos, PAGE_SIZE, &retlen, buf);
1162 
1163 	d->mtd_read_count++;
1164 	if (mtd_is_bitflip(ret)) {
1165 		eb->flags |= EBLOCK_BITFLIP;
1166 		mtdswap_rb_add(d, eb, MTDSWAP_BITFLIP);
1167 		ret = 0;
1168 	}
1169 
1170 	if (ret < 0) {
1171 		dev_err(d->dev, "Read error %d\n", ret);
1172 		eb->flags |= EBLOCK_READERR;
1173 		mtdswap_rb_add(d, eb, MTDSWAP_FAILING);
1174 		retries++;
1175 		if (retries < MTDSWAP_IO_RETRIES)
1176 			goto retry;
1177 
1178 		return ret;
1179 	}
1180 
1181 	if (retlen != PAGE_SIZE) {
1182 		dev_err(d->dev, "Short read %zd\n", retlen);
1183 		return -EIO;
1184 	}
1185 
1186 	return 0;
1187 }
1188 
1189 static int mtdswap_discard(struct mtd_blktrans_dev *dev, unsigned long first,
1190 			unsigned nr_pages)
1191 {
1192 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1193 	unsigned long page;
1194 	struct swap_eb *eb;
1195 	unsigned int mapped;
1196 
1197 	d->discard_count++;
1198 
1199 	for (page = first; page < first + nr_pages; page++) {
1200 		mapped = d->page_data[page];
1201 		if (mapped <= BLOCK_MAX) {
1202 			eb = d->eb_data + (mapped / d->pages_per_eblk);
1203 			eb->active_count--;
1204 			mtdswap_store_eb(d, eb);
1205 			d->page_data[page] = BLOCK_UNDEF;
1206 			d->revmap[mapped] = PAGE_UNDEF;
1207 			d->discard_page_count++;
1208 		} else if (mapped == BLOCK_ERROR) {
1209 			d->page_data[page] = BLOCK_UNDEF;
1210 			d->discard_page_count++;
1211 		}
1212 	}
1213 
1214 	return 0;
1215 }
1216 
1217 static int mtdswap_show(struct seq_file *s, void *data)
1218 {
1219 	struct mtdswap_dev *d = (struct mtdswap_dev *) s->private;
1220 	unsigned long sum;
1221 	unsigned int count[MTDSWAP_TREE_CNT];
1222 	unsigned int min[MTDSWAP_TREE_CNT];
1223 	unsigned int max[MTDSWAP_TREE_CNT];
1224 	unsigned int i, cw = 0, cwp = 0, cwecount = 0, bb_cnt, mapped, pages;
1225 	uint64_t use_size;
1226 	static const char * const name[] = {
1227 		"clean", "used", "low", "high", "dirty", "bitflip", "failing"
1228 	};
1229 
1230 	mutex_lock(&d->mbd_dev->lock);
1231 
1232 	for (i = 0; i < MTDSWAP_TREE_CNT; i++) {
1233 		struct rb_root *root = &d->trees[i].root;
1234 
1235 		if (root->rb_node) {
1236 			count[i] = d->trees[i].count;
1237 			min[i] = MTDSWAP_ECNT_MIN(root);
1238 			max[i] = MTDSWAP_ECNT_MAX(root);
1239 		} else
1240 			count[i] = 0;
1241 	}
1242 
1243 	if (d->curr_write) {
1244 		cw = 1;
1245 		cwp = d->curr_write_pos;
1246 		cwecount = d->curr_write->erase_count;
1247 	}
1248 
1249 	sum = 0;
1250 	for (i = 0; i < d->eblks; i++)
1251 		sum += d->eb_data[i].erase_count;
1252 
1253 	use_size = (uint64_t)d->eblks * d->mtd->erasesize;
1254 	bb_cnt = mtdswap_badblocks(d->mtd, use_size);
1255 
1256 	mapped = 0;
1257 	pages = d->mbd_dev->size;
1258 	for (i = 0; i < pages; i++)
1259 		if (d->page_data[i] != BLOCK_UNDEF)
1260 			mapped++;
1261 
1262 	mutex_unlock(&d->mbd_dev->lock);
1263 
1264 	for (i = 0; i < MTDSWAP_TREE_CNT; i++) {
1265 		if (!count[i])
1266 			continue;
1267 
1268 		if (min[i] != max[i])
1269 			seq_printf(s, "%s:\t%5d erase blocks, erased min %d, "
1270 				"max %d times\n",
1271 				name[i], count[i], min[i], max[i]);
1272 		else
1273 			seq_printf(s, "%s:\t%5d erase blocks, all erased %d "
1274 				"times\n", name[i], count[i], min[i]);
1275 	}
1276 
1277 	if (bb_cnt)
1278 		seq_printf(s, "bad:\t%5u erase blocks\n", bb_cnt);
1279 
1280 	if (cw)
1281 		seq_printf(s, "current erase block: %u pages used, %u free, "
1282 			"erased %u times\n",
1283 			cwp, d->pages_per_eblk - cwp, cwecount);
1284 
1285 	seq_printf(s, "total erasures: %lu\n", sum);
1286 
1287 	seq_puts(s, "\n");
1288 
1289 	seq_printf(s, "mtdswap_readsect count: %llu\n", d->sect_read_count);
1290 	seq_printf(s, "mtdswap_writesect count: %llu\n", d->sect_write_count);
1291 	seq_printf(s, "mtdswap_discard count: %llu\n", d->discard_count);
1292 	seq_printf(s, "mtd read count: %llu\n", d->mtd_read_count);
1293 	seq_printf(s, "mtd write count: %llu\n", d->mtd_write_count);
1294 	seq_printf(s, "discarded pages count: %llu\n", d->discard_page_count);
1295 
1296 	seq_puts(s, "\n");
1297 	seq_printf(s, "total pages: %u\n", pages);
1298 	seq_printf(s, "pages mapped: %u\n", mapped);
1299 
1300 	return 0;
1301 }
1302 
1303 static int mtdswap_open(struct inode *inode, struct file *file)
1304 {
1305 	return single_open(file, mtdswap_show, inode->i_private);
1306 }
1307 
1308 static const struct file_operations mtdswap_fops = {
1309 	.open		= mtdswap_open,
1310 	.read		= seq_read,
1311 	.llseek		= seq_lseek,
1312 	.release	= single_release,
1313 };
1314 
1315 static int mtdswap_add_debugfs(struct mtdswap_dev *d)
1316 {
1317 	struct dentry *root = d->mtd->dbg.dfs_dir;
1318 	struct dentry *dent;
1319 
1320 	if (!IS_ENABLED(CONFIG_DEBUG_FS))
1321 		return 0;
1322 
1323 	if (IS_ERR_OR_NULL(root))
1324 		return -1;
1325 
1326 	dent = debugfs_create_file("mtdswap_stats", S_IRUSR, root, d,
1327 				&mtdswap_fops);
1328 	if (!dent) {
1329 		dev_err(d->dev, "debugfs_create_file failed\n");
1330 		return -1;
1331 	}
1332 
1333 	return 0;
1334 }
1335 
1336 static int mtdswap_init(struct mtdswap_dev *d, unsigned int eblocks,
1337 			unsigned int spare_cnt)
1338 {
1339 	struct mtd_info *mtd = d->mbd_dev->mtd;
1340 	unsigned int i, eblk_bytes, pages, blocks;
1341 	int ret = -ENOMEM;
1342 
1343 	d->mtd = mtd;
1344 	d->eblks = eblocks;
1345 	d->spare_eblks = spare_cnt;
1346 	d->pages_per_eblk = mtd->erasesize >> PAGE_SHIFT;
1347 
1348 	pages = d->mbd_dev->size;
1349 	blocks = eblocks * d->pages_per_eblk;
1350 
1351 	for (i = 0; i < MTDSWAP_TREE_CNT; i++)
1352 		d->trees[i].root = RB_ROOT;
1353 
1354 	d->page_data = vmalloc(sizeof(int)*pages);
1355 	if (!d->page_data)
1356 		goto page_data_fail;
1357 
1358 	d->revmap = vmalloc(sizeof(int)*blocks);
1359 	if (!d->revmap)
1360 		goto revmap_fail;
1361 
1362 	eblk_bytes = sizeof(struct swap_eb)*d->eblks;
1363 	d->eb_data = vzalloc(eblk_bytes);
1364 	if (!d->eb_data)
1365 		goto eb_data_fail;
1366 
1367 	for (i = 0; i < pages; i++)
1368 		d->page_data[i] = BLOCK_UNDEF;
1369 
1370 	for (i = 0; i < blocks; i++)
1371 		d->revmap[i] = PAGE_UNDEF;
1372 
1373 	d->page_buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1374 	if (!d->page_buf)
1375 		goto page_buf_fail;
1376 
1377 	d->oob_buf = kmalloc(2 * mtd->oobavail, GFP_KERNEL);
1378 	if (!d->oob_buf)
1379 		goto oob_buf_fail;
1380 
1381 	mtdswap_scan_eblks(d);
1382 
1383 	return 0;
1384 
1385 oob_buf_fail:
1386 	kfree(d->page_buf);
1387 page_buf_fail:
1388 	vfree(d->eb_data);
1389 eb_data_fail:
1390 	vfree(d->revmap);
1391 revmap_fail:
1392 	vfree(d->page_data);
1393 page_data_fail:
1394 	printk(KERN_ERR "%s: init failed (%d)\n", MTDSWAP_PREFIX, ret);
1395 	return ret;
1396 }
1397 
1398 static void mtdswap_add_mtd(struct mtd_blktrans_ops *tr, struct mtd_info *mtd)
1399 {
1400 	struct mtdswap_dev *d;
1401 	struct mtd_blktrans_dev *mbd_dev;
1402 	char *parts;
1403 	char *this_opt;
1404 	unsigned long part;
1405 	unsigned int eblocks, eavailable, bad_blocks, spare_cnt;
1406 	uint64_t swap_size, use_size, size_limit;
1407 	int ret;
1408 
1409 	parts = &partitions[0];
1410 	if (!*parts)
1411 		return;
1412 
1413 	while ((this_opt = strsep(&parts, ",")) != NULL) {
1414 		if (kstrtoul(this_opt, 0, &part) < 0)
1415 			return;
1416 
1417 		if (mtd->index == part)
1418 			break;
1419 	}
1420 
1421 	if (mtd->index != part)
1422 		return;
1423 
1424 	if (mtd->erasesize < PAGE_SIZE || mtd->erasesize % PAGE_SIZE) {
1425 		printk(KERN_ERR "%s: Erase size %u not multiple of PAGE_SIZE "
1426 			"%lu\n", MTDSWAP_PREFIX, mtd->erasesize, PAGE_SIZE);
1427 		return;
1428 	}
1429 
1430 	if (PAGE_SIZE % mtd->writesize || mtd->writesize > PAGE_SIZE) {
1431 		printk(KERN_ERR "%s: PAGE_SIZE %lu not multiple of write size"
1432 			" %u\n", MTDSWAP_PREFIX, PAGE_SIZE, mtd->writesize);
1433 		return;
1434 	}
1435 
1436 	if (!mtd->oobsize || mtd->oobavail < MTDSWAP_OOBSIZE) {
1437 		printk(KERN_ERR "%s: Not enough free bytes in OOB, "
1438 			"%d available, %zu needed.\n",
1439 			MTDSWAP_PREFIX, mtd->oobavail, MTDSWAP_OOBSIZE);
1440 		return;
1441 	}
1442 
1443 	if (spare_eblocks > 100)
1444 		spare_eblocks = 100;
1445 
1446 	use_size = mtd->size;
1447 	size_limit = (uint64_t) BLOCK_MAX * PAGE_SIZE;
1448 
1449 	if (mtd->size > size_limit) {
1450 		printk(KERN_WARNING "%s: Device too large. Limiting size to "
1451 			"%llu bytes\n", MTDSWAP_PREFIX, size_limit);
1452 		use_size = size_limit;
1453 	}
1454 
1455 	eblocks = mtd_div_by_eb(use_size, mtd);
1456 	use_size = (uint64_t)eblocks * mtd->erasesize;
1457 	bad_blocks = mtdswap_badblocks(mtd, use_size);
1458 	eavailable = eblocks - bad_blocks;
1459 
1460 	if (eavailable < MIN_ERASE_BLOCKS) {
1461 		printk(KERN_ERR "%s: Not enough erase blocks. %u available, "
1462 			"%d needed\n", MTDSWAP_PREFIX, eavailable,
1463 			MIN_ERASE_BLOCKS);
1464 		return;
1465 	}
1466 
1467 	spare_cnt = div_u64((uint64_t)eavailable * spare_eblocks, 100);
1468 
1469 	if (spare_cnt < MIN_SPARE_EBLOCKS)
1470 		spare_cnt = MIN_SPARE_EBLOCKS;
1471 
1472 	if (spare_cnt > eavailable - 1)
1473 		spare_cnt = eavailable - 1;
1474 
1475 	swap_size = (uint64_t)(eavailable - spare_cnt) * mtd->erasesize +
1476 		(header ? PAGE_SIZE : 0);
1477 
1478 	printk(KERN_INFO "%s: Enabling MTD swap on device %lu, size %llu KB, "
1479 		"%u spare, %u bad blocks\n",
1480 		MTDSWAP_PREFIX, part, swap_size / 1024, spare_cnt, bad_blocks);
1481 
1482 	d = kzalloc(sizeof(struct mtdswap_dev), GFP_KERNEL);
1483 	if (!d)
1484 		return;
1485 
1486 	mbd_dev = kzalloc(sizeof(struct mtd_blktrans_dev), GFP_KERNEL);
1487 	if (!mbd_dev) {
1488 		kfree(d);
1489 		return;
1490 	}
1491 
1492 	d->mbd_dev = mbd_dev;
1493 	mbd_dev->priv = d;
1494 
1495 	mbd_dev->mtd = mtd;
1496 	mbd_dev->devnum = mtd->index;
1497 	mbd_dev->size = swap_size >> PAGE_SHIFT;
1498 	mbd_dev->tr = tr;
1499 
1500 	if (!(mtd->flags & MTD_WRITEABLE))
1501 		mbd_dev->readonly = 1;
1502 
1503 	if (mtdswap_init(d, eblocks, spare_cnt) < 0)
1504 		goto init_failed;
1505 
1506 	if (add_mtd_blktrans_dev(mbd_dev) < 0)
1507 		goto cleanup;
1508 
1509 	d->dev = disk_to_dev(mbd_dev->disk);
1510 
1511 	ret = mtdswap_add_debugfs(d);
1512 	if (ret < 0)
1513 		goto debugfs_failed;
1514 
1515 	return;
1516 
1517 debugfs_failed:
1518 	del_mtd_blktrans_dev(mbd_dev);
1519 
1520 cleanup:
1521 	mtdswap_cleanup(d);
1522 
1523 init_failed:
1524 	kfree(mbd_dev);
1525 	kfree(d);
1526 }
1527 
1528 static void mtdswap_remove_dev(struct mtd_blktrans_dev *dev)
1529 {
1530 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1531 
1532 	del_mtd_blktrans_dev(dev);
1533 	mtdswap_cleanup(d);
1534 	kfree(d);
1535 }
1536 
1537 static struct mtd_blktrans_ops mtdswap_ops = {
1538 	.name		= "mtdswap",
1539 	.major		= 0,
1540 	.part_bits	= 0,
1541 	.blksize	= PAGE_SIZE,
1542 	.flush		= mtdswap_flush,
1543 	.readsect	= mtdswap_readsect,
1544 	.writesect	= mtdswap_writesect,
1545 	.discard	= mtdswap_discard,
1546 	.background	= mtdswap_background,
1547 	.add_mtd	= mtdswap_add_mtd,
1548 	.remove_dev	= mtdswap_remove_dev,
1549 	.owner		= THIS_MODULE,
1550 };
1551 
1552 static int __init mtdswap_modinit(void)
1553 {
1554 	return register_mtd_blktrans(&mtdswap_ops);
1555 }
1556 
1557 static void __exit mtdswap_modexit(void)
1558 {
1559 	deregister_mtd_blktrans(&mtdswap_ops);
1560 }
1561 
1562 module_init(mtdswap_modinit);
1563 module_exit(mtdswap_modexit);
1564 
1565 
1566 MODULE_LICENSE("GPL");
1567 MODULE_AUTHOR("Jarkko Lavinen <jarkko.lavinen@nokia.com>");
1568 MODULE_DESCRIPTION("Block device access to an MTD suitable for using as "
1569 		"swap space");
1570