xref: /openbmc/linux/drivers/mtd/mtdswap.c (revision f9a82c48)
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 static int mtdswap_erase_block(struct mtdswap_dev *d, struct swap_eb *eb)
540 {
541 	struct mtd_info *mtd = d->mtd;
542 	struct erase_info erase;
543 	unsigned int retries = 0;
544 	int ret;
545 
546 	eb->erase_count++;
547 	if (eb->erase_count > d->max_erase_count)
548 		d->max_erase_count = eb->erase_count;
549 
550 retry:
551 	memset(&erase, 0, sizeof(struct erase_info));
552 	erase.addr	= mtdswap_eb_offset(d, eb);
553 	erase.len	= mtd->erasesize;
554 
555 	ret = mtd_erase(mtd, &erase);
556 	if (ret) {
557 		if (retries++ < MTDSWAP_ERASE_RETRIES) {
558 			dev_warn(d->dev,
559 				"erase of erase block %#llx on %s failed",
560 				erase.addr, mtd->name);
561 			yield();
562 			goto retry;
563 		}
564 
565 		dev_err(d->dev, "Cannot erase erase block %#llx on %s\n",
566 			erase.addr, mtd->name);
567 
568 		mtdswap_handle_badblock(d, eb);
569 		return -EIO;
570 	}
571 
572 	return 0;
573 }
574 
575 static int mtdswap_map_free_block(struct mtdswap_dev *d, unsigned int page,
576 				unsigned int *block)
577 {
578 	int ret;
579 	struct swap_eb *old_eb = d->curr_write;
580 	struct rb_root *clean_root;
581 	struct swap_eb *eb;
582 
583 	if (old_eb == NULL || d->curr_write_pos >= d->pages_per_eblk) {
584 		do {
585 			if (TREE_EMPTY(d, CLEAN))
586 				return -ENOSPC;
587 
588 			clean_root = TREE_ROOT(d, CLEAN);
589 			eb = rb_entry(rb_first(clean_root), struct swap_eb, rb);
590 			rb_erase(&eb->rb, clean_root);
591 			eb->root = NULL;
592 			TREE_COUNT(d, CLEAN)--;
593 
594 			ret = mtdswap_write_marker(d, eb, MTDSWAP_TYPE_DIRTY);
595 		} while (ret == -EIO || mtd_is_eccerr(ret));
596 
597 		if (ret)
598 			return ret;
599 
600 		d->curr_write_pos = 0;
601 		d->curr_write = eb;
602 		if (old_eb)
603 			mtdswap_store_eb(d, old_eb);
604 	}
605 
606 	*block = (d->curr_write - d->eb_data) * d->pages_per_eblk +
607 		d->curr_write_pos;
608 
609 	d->curr_write->active_count++;
610 	d->revmap[*block] = page;
611 	d->curr_write_pos++;
612 
613 	return 0;
614 }
615 
616 static unsigned int mtdswap_free_page_cnt(struct mtdswap_dev *d)
617 {
618 	return TREE_COUNT(d, CLEAN) * d->pages_per_eblk +
619 		d->pages_per_eblk - d->curr_write_pos;
620 }
621 
622 static unsigned int mtdswap_enough_free_pages(struct mtdswap_dev *d)
623 {
624 	return mtdswap_free_page_cnt(d) > d->pages_per_eblk;
625 }
626 
627 static int mtdswap_write_block(struct mtdswap_dev *d, char *buf,
628 			unsigned int page, unsigned int *bp, int gc_context)
629 {
630 	struct mtd_info *mtd = d->mtd;
631 	struct swap_eb *eb;
632 	size_t retlen;
633 	loff_t writepos;
634 	int ret;
635 
636 retry:
637 	if (!gc_context)
638 		while (!mtdswap_enough_free_pages(d))
639 			if (mtdswap_gc(d, 0) > 0)
640 				return -ENOSPC;
641 
642 	ret = mtdswap_map_free_block(d, page, bp);
643 	eb = d->eb_data + (*bp / d->pages_per_eblk);
644 
645 	if (ret == -EIO || mtd_is_eccerr(ret)) {
646 		d->curr_write = NULL;
647 		eb->active_count--;
648 		d->revmap[*bp] = PAGE_UNDEF;
649 		goto retry;
650 	}
651 
652 	if (ret < 0)
653 		return ret;
654 
655 	writepos = (loff_t)*bp << PAGE_SHIFT;
656 	ret =  mtd_write(mtd, writepos, PAGE_SIZE, &retlen, buf);
657 	if (ret == -EIO || mtd_is_eccerr(ret)) {
658 		d->curr_write_pos--;
659 		eb->active_count--;
660 		d->revmap[*bp] = PAGE_UNDEF;
661 		mtdswap_handle_write_error(d, eb);
662 		goto retry;
663 	}
664 
665 	if (ret < 0) {
666 		dev_err(d->dev, "Write to MTD device failed: %d (%zd written)",
667 			ret, retlen);
668 		goto err;
669 	}
670 
671 	if (retlen != PAGE_SIZE) {
672 		dev_err(d->dev, "Short write to MTD device: %zd written",
673 			retlen);
674 		ret = -EIO;
675 		goto err;
676 	}
677 
678 	return ret;
679 
680 err:
681 	d->curr_write_pos--;
682 	eb->active_count--;
683 	d->revmap[*bp] = PAGE_UNDEF;
684 
685 	return ret;
686 }
687 
688 static int mtdswap_move_block(struct mtdswap_dev *d, unsigned int oldblock,
689 		unsigned int *newblock)
690 {
691 	struct mtd_info *mtd = d->mtd;
692 	struct swap_eb *eb, *oldeb;
693 	int ret;
694 	size_t retlen;
695 	unsigned int page, retries;
696 	loff_t readpos;
697 
698 	page = d->revmap[oldblock];
699 	readpos = (loff_t) oldblock << PAGE_SHIFT;
700 	retries = 0;
701 
702 retry:
703 	ret = mtd_read(mtd, readpos, PAGE_SIZE, &retlen, d->page_buf);
704 
705 	if (ret < 0 && !mtd_is_bitflip(ret)) {
706 		oldeb = d->eb_data + oldblock / d->pages_per_eblk;
707 		oldeb->flags |= EBLOCK_READERR;
708 
709 		dev_err(d->dev, "Read Error: %d (block %u)\n", ret,
710 			oldblock);
711 		retries++;
712 		if (retries < MTDSWAP_IO_RETRIES)
713 			goto retry;
714 
715 		goto read_error;
716 	}
717 
718 	if (retlen != PAGE_SIZE) {
719 		dev_err(d->dev, "Short read: %zd (block %u)\n", retlen,
720 		       oldblock);
721 		ret = -EIO;
722 		goto read_error;
723 	}
724 
725 	ret = mtdswap_write_block(d, d->page_buf, page, newblock, 1);
726 	if (ret < 0) {
727 		d->page_data[page] = BLOCK_ERROR;
728 		dev_err(d->dev, "Write error: %d\n", ret);
729 		return ret;
730 	}
731 
732 	eb = d->eb_data + *newblock / d->pages_per_eblk;
733 	d->page_data[page] = *newblock;
734 	d->revmap[oldblock] = PAGE_UNDEF;
735 	eb = d->eb_data + oldblock / d->pages_per_eblk;
736 	eb->active_count--;
737 
738 	return 0;
739 
740 read_error:
741 	d->page_data[page] = BLOCK_ERROR;
742 	d->revmap[oldblock] = PAGE_UNDEF;
743 	return ret;
744 }
745 
746 static int mtdswap_gc_eblock(struct mtdswap_dev *d, struct swap_eb *eb)
747 {
748 	unsigned int i, block, eblk_base, newblock;
749 	int ret, errcode;
750 
751 	errcode = 0;
752 	eblk_base = (eb - d->eb_data) * d->pages_per_eblk;
753 
754 	for (i = 0; i < d->pages_per_eblk; i++) {
755 		if (d->spare_eblks < MIN_SPARE_EBLOCKS)
756 			return -ENOSPC;
757 
758 		block = eblk_base + i;
759 		if (d->revmap[block] == PAGE_UNDEF)
760 			continue;
761 
762 		ret = mtdswap_move_block(d, block, &newblock);
763 		if (ret < 0 && !errcode)
764 			errcode = ret;
765 	}
766 
767 	return errcode;
768 }
769 
770 static int __mtdswap_choose_gc_tree(struct mtdswap_dev *d)
771 {
772 	int idx, stopat;
773 
774 	if (TREE_COUNT(d, CLEAN) < LOW_FRAG_GC_THRESHOLD)
775 		stopat = MTDSWAP_LOWFRAG;
776 	else
777 		stopat = MTDSWAP_HIFRAG;
778 
779 	for (idx = MTDSWAP_BITFLIP; idx >= stopat; idx--)
780 		if (d->trees[idx].root.rb_node != NULL)
781 			return idx;
782 
783 	return -1;
784 }
785 
786 static int mtdswap_wlfreq(unsigned int maxdiff)
787 {
788 	unsigned int h, x, y, dist, base;
789 
790 	/*
791 	 * Calculate linear ramp down from f1 to f2 when maxdiff goes from
792 	 * MAX_ERASE_DIFF to MAX_ERASE_DIFF + COLLECT_NONDIRTY_BASE.  Similar
793 	 * to triangle with height f1 - f1 and width COLLECT_NONDIRTY_BASE.
794 	 */
795 
796 	dist = maxdiff - MAX_ERASE_DIFF;
797 	if (dist > COLLECT_NONDIRTY_BASE)
798 		dist = COLLECT_NONDIRTY_BASE;
799 
800 	/*
801 	 * Modelling the slop as right angular triangle with base
802 	 * COLLECT_NONDIRTY_BASE and height freq1 - freq2. The ratio y/x is
803 	 * equal to the ratio h/base.
804 	 */
805 	h = COLLECT_NONDIRTY_FREQ1 - COLLECT_NONDIRTY_FREQ2;
806 	base = COLLECT_NONDIRTY_BASE;
807 
808 	x = dist - base;
809 	y = (x * h + base / 2) / base;
810 
811 	return COLLECT_NONDIRTY_FREQ2 + y;
812 }
813 
814 static int mtdswap_choose_wl_tree(struct mtdswap_dev *d)
815 {
816 	static unsigned int pick_cnt;
817 	unsigned int i, idx = -1, wear, max;
818 	struct rb_root *root;
819 
820 	max = 0;
821 	for (i = 0; i <= MTDSWAP_DIRTY; i++) {
822 		root = &d->trees[i].root;
823 		if (root->rb_node == NULL)
824 			continue;
825 
826 		wear = d->max_erase_count - MTDSWAP_ECNT_MIN(root);
827 		if (wear > max) {
828 			max = wear;
829 			idx = i;
830 		}
831 	}
832 
833 	if (max > MAX_ERASE_DIFF && pick_cnt >= mtdswap_wlfreq(max) - 1) {
834 		pick_cnt = 0;
835 		return idx;
836 	}
837 
838 	pick_cnt++;
839 	return -1;
840 }
841 
842 static int mtdswap_choose_gc_tree(struct mtdswap_dev *d,
843 				unsigned int background)
844 {
845 	int idx;
846 
847 	if (TREE_NONEMPTY(d, FAILING) &&
848 		(background || (TREE_EMPTY(d, CLEAN) && TREE_EMPTY(d, DIRTY))))
849 		return MTDSWAP_FAILING;
850 
851 	idx = mtdswap_choose_wl_tree(d);
852 	if (idx >= MTDSWAP_CLEAN)
853 		return idx;
854 
855 	return __mtdswap_choose_gc_tree(d);
856 }
857 
858 static struct swap_eb *mtdswap_pick_gc_eblk(struct mtdswap_dev *d,
859 					unsigned int background)
860 {
861 	struct rb_root *rp = NULL;
862 	struct swap_eb *eb = NULL;
863 	int idx;
864 
865 	if (background && TREE_COUNT(d, CLEAN) > CLEAN_BLOCK_THRESHOLD &&
866 		TREE_EMPTY(d, DIRTY) && TREE_EMPTY(d, FAILING))
867 		return NULL;
868 
869 	idx = mtdswap_choose_gc_tree(d, background);
870 	if (idx < 0)
871 		return NULL;
872 
873 	rp = &d->trees[idx].root;
874 	eb = rb_entry(rb_first(rp), struct swap_eb, rb);
875 
876 	rb_erase(&eb->rb, rp);
877 	eb->root = NULL;
878 	d->trees[idx].count--;
879 	return eb;
880 }
881 
882 static unsigned int mtdswap_test_patt(unsigned int i)
883 {
884 	return i % 2 ? 0x55555555 : 0xAAAAAAAA;
885 }
886 
887 static unsigned int mtdswap_eblk_passes(struct mtdswap_dev *d,
888 					struct swap_eb *eb)
889 {
890 	struct mtd_info *mtd = d->mtd;
891 	unsigned int test, i, j, patt, mtd_pages;
892 	loff_t base, pos;
893 	unsigned int *p1 = (unsigned int *)d->page_buf;
894 	unsigned char *p2 = (unsigned char *)d->oob_buf;
895 	struct mtd_oob_ops ops;
896 	int ret;
897 
898 	ops.mode = MTD_OPS_AUTO_OOB;
899 	ops.len = mtd->writesize;
900 	ops.ooblen = mtd->oobavail;
901 	ops.ooboffs = 0;
902 	ops.datbuf = d->page_buf;
903 	ops.oobbuf = d->oob_buf;
904 	base = mtdswap_eb_offset(d, eb);
905 	mtd_pages = d->pages_per_eblk * PAGE_SIZE / mtd->writesize;
906 
907 	for (test = 0; test < 2; test++) {
908 		pos = base;
909 		for (i = 0; i < mtd_pages; i++) {
910 			patt = mtdswap_test_patt(test + i);
911 			memset(d->page_buf, patt, mtd->writesize);
912 			memset(d->oob_buf, patt, mtd->oobavail);
913 			ret = mtd_write_oob(mtd, pos, &ops);
914 			if (ret)
915 				goto error;
916 
917 			pos += mtd->writesize;
918 		}
919 
920 		pos = base;
921 		for (i = 0; i < mtd_pages; i++) {
922 			ret = mtd_read_oob(mtd, pos, &ops);
923 			if (ret)
924 				goto error;
925 
926 			patt = mtdswap_test_patt(test + i);
927 			for (j = 0; j < mtd->writesize/sizeof(int); j++)
928 				if (p1[j] != patt)
929 					goto error;
930 
931 			for (j = 0; j < mtd->oobavail; j++)
932 				if (p2[j] != (unsigned char)patt)
933 					goto error;
934 
935 			pos += mtd->writesize;
936 		}
937 
938 		ret = mtdswap_erase_block(d, eb);
939 		if (ret)
940 			goto error;
941 	}
942 
943 	eb->flags &= ~EBLOCK_READERR;
944 	return 1;
945 
946 error:
947 	mtdswap_handle_badblock(d, eb);
948 	return 0;
949 }
950 
951 static int mtdswap_gc(struct mtdswap_dev *d, unsigned int background)
952 {
953 	struct swap_eb *eb;
954 	int ret;
955 
956 	if (d->spare_eblks < MIN_SPARE_EBLOCKS)
957 		return 1;
958 
959 	eb = mtdswap_pick_gc_eblk(d, background);
960 	if (!eb)
961 		return 1;
962 
963 	ret = mtdswap_gc_eblock(d, eb);
964 	if (ret == -ENOSPC)
965 		return 1;
966 
967 	if (eb->flags & EBLOCK_FAILED) {
968 		mtdswap_handle_badblock(d, eb);
969 		return 0;
970 	}
971 
972 	eb->flags &= ~EBLOCK_BITFLIP;
973 	ret = mtdswap_erase_block(d, eb);
974 	if ((eb->flags & EBLOCK_READERR) &&
975 		(ret || !mtdswap_eblk_passes(d, eb)))
976 		return 0;
977 
978 	if (ret == 0)
979 		ret = mtdswap_write_marker(d, eb, MTDSWAP_TYPE_CLEAN);
980 
981 	if (ret == 0)
982 		mtdswap_rb_add(d, eb, MTDSWAP_CLEAN);
983 	else if (ret != -EIO && !mtd_is_eccerr(ret))
984 		mtdswap_rb_add(d, eb, MTDSWAP_DIRTY);
985 
986 	return 0;
987 }
988 
989 static void mtdswap_background(struct mtd_blktrans_dev *dev)
990 {
991 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
992 	int ret;
993 
994 	while (1) {
995 		ret = mtdswap_gc(d, 1);
996 		if (ret || mtd_blktrans_cease_background(dev))
997 			return;
998 	}
999 }
1000 
1001 static void mtdswap_cleanup(struct mtdswap_dev *d)
1002 {
1003 	vfree(d->eb_data);
1004 	vfree(d->revmap);
1005 	vfree(d->page_data);
1006 	kfree(d->oob_buf);
1007 	kfree(d->page_buf);
1008 }
1009 
1010 static int mtdswap_flush(struct mtd_blktrans_dev *dev)
1011 {
1012 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1013 
1014 	mtd_sync(d->mtd);
1015 	return 0;
1016 }
1017 
1018 static unsigned int mtdswap_badblocks(struct mtd_info *mtd, uint64_t size)
1019 {
1020 	loff_t offset;
1021 	unsigned int badcnt;
1022 
1023 	badcnt = 0;
1024 
1025 	if (mtd_can_have_bb(mtd))
1026 		for (offset = 0; offset < size; offset += mtd->erasesize)
1027 			if (mtd_block_isbad(mtd, offset))
1028 				badcnt++;
1029 
1030 	return badcnt;
1031 }
1032 
1033 static int mtdswap_writesect(struct mtd_blktrans_dev *dev,
1034 			unsigned long page, char *buf)
1035 {
1036 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1037 	unsigned int newblock, mapped;
1038 	struct swap_eb *eb;
1039 	int ret;
1040 
1041 	d->sect_write_count++;
1042 
1043 	if (d->spare_eblks < MIN_SPARE_EBLOCKS)
1044 		return -ENOSPC;
1045 
1046 	if (header) {
1047 		/* Ignore writes to the header page */
1048 		if (unlikely(page == 0))
1049 			return 0;
1050 
1051 		page--;
1052 	}
1053 
1054 	mapped = d->page_data[page];
1055 	if (mapped <= BLOCK_MAX) {
1056 		eb = d->eb_data + (mapped / d->pages_per_eblk);
1057 		eb->active_count--;
1058 		mtdswap_store_eb(d, eb);
1059 		d->page_data[page] = BLOCK_UNDEF;
1060 		d->revmap[mapped] = PAGE_UNDEF;
1061 	}
1062 
1063 	ret = mtdswap_write_block(d, buf, page, &newblock, 0);
1064 	d->mtd_write_count++;
1065 
1066 	if (ret < 0)
1067 		return ret;
1068 
1069 	eb = d->eb_data + (newblock / d->pages_per_eblk);
1070 	d->page_data[page] = newblock;
1071 
1072 	return 0;
1073 }
1074 
1075 /* Provide a dummy swap header for the kernel */
1076 static int mtdswap_auto_header(struct mtdswap_dev *d, char *buf)
1077 {
1078 	union swap_header *hd = (union swap_header *)(buf);
1079 
1080 	memset(buf, 0, PAGE_SIZE - 10);
1081 
1082 	hd->info.version = 1;
1083 	hd->info.last_page = d->mbd_dev->size - 1;
1084 	hd->info.nr_badpages = 0;
1085 
1086 	memcpy(buf + PAGE_SIZE - 10, "SWAPSPACE2", 10);
1087 
1088 	return 0;
1089 }
1090 
1091 static int mtdswap_readsect(struct mtd_blktrans_dev *dev,
1092 			unsigned long page, char *buf)
1093 {
1094 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1095 	struct mtd_info *mtd = d->mtd;
1096 	unsigned int realblock, retries;
1097 	loff_t readpos;
1098 	struct swap_eb *eb;
1099 	size_t retlen;
1100 	int ret;
1101 
1102 	d->sect_read_count++;
1103 
1104 	if (header) {
1105 		if (unlikely(page == 0))
1106 			return mtdswap_auto_header(d, buf);
1107 
1108 		page--;
1109 	}
1110 
1111 	realblock = d->page_data[page];
1112 	if (realblock > BLOCK_MAX) {
1113 		memset(buf, 0x0, PAGE_SIZE);
1114 		if (realblock == BLOCK_UNDEF)
1115 			return 0;
1116 		else
1117 			return -EIO;
1118 	}
1119 
1120 	eb = d->eb_data + (realblock / d->pages_per_eblk);
1121 	BUG_ON(d->revmap[realblock] == PAGE_UNDEF);
1122 
1123 	readpos = (loff_t)realblock << PAGE_SHIFT;
1124 	retries = 0;
1125 
1126 retry:
1127 	ret = mtd_read(mtd, readpos, PAGE_SIZE, &retlen, buf);
1128 
1129 	d->mtd_read_count++;
1130 	if (mtd_is_bitflip(ret)) {
1131 		eb->flags |= EBLOCK_BITFLIP;
1132 		mtdswap_rb_add(d, eb, MTDSWAP_BITFLIP);
1133 		ret = 0;
1134 	}
1135 
1136 	if (ret < 0) {
1137 		dev_err(d->dev, "Read error %d\n", ret);
1138 		eb->flags |= EBLOCK_READERR;
1139 		mtdswap_rb_add(d, eb, MTDSWAP_FAILING);
1140 		retries++;
1141 		if (retries < MTDSWAP_IO_RETRIES)
1142 			goto retry;
1143 
1144 		return ret;
1145 	}
1146 
1147 	if (retlen != PAGE_SIZE) {
1148 		dev_err(d->dev, "Short read %zd\n", retlen);
1149 		return -EIO;
1150 	}
1151 
1152 	return 0;
1153 }
1154 
1155 static int mtdswap_discard(struct mtd_blktrans_dev *dev, unsigned long first,
1156 			unsigned nr_pages)
1157 {
1158 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1159 	unsigned long page;
1160 	struct swap_eb *eb;
1161 	unsigned int mapped;
1162 
1163 	d->discard_count++;
1164 
1165 	for (page = first; page < first + nr_pages; page++) {
1166 		mapped = d->page_data[page];
1167 		if (mapped <= BLOCK_MAX) {
1168 			eb = d->eb_data + (mapped / d->pages_per_eblk);
1169 			eb->active_count--;
1170 			mtdswap_store_eb(d, eb);
1171 			d->page_data[page] = BLOCK_UNDEF;
1172 			d->revmap[mapped] = PAGE_UNDEF;
1173 			d->discard_page_count++;
1174 		} else if (mapped == BLOCK_ERROR) {
1175 			d->page_data[page] = BLOCK_UNDEF;
1176 			d->discard_page_count++;
1177 		}
1178 	}
1179 
1180 	return 0;
1181 }
1182 
1183 static int mtdswap_show(struct seq_file *s, void *data)
1184 {
1185 	struct mtdswap_dev *d = (struct mtdswap_dev *) s->private;
1186 	unsigned long sum;
1187 	unsigned int count[MTDSWAP_TREE_CNT];
1188 	unsigned int min[MTDSWAP_TREE_CNT];
1189 	unsigned int max[MTDSWAP_TREE_CNT];
1190 	unsigned int i, cw = 0, cwp = 0, cwecount = 0, bb_cnt, mapped, pages;
1191 	uint64_t use_size;
1192 	static const char * const name[] = {
1193 		"clean", "used", "low", "high", "dirty", "bitflip", "failing"
1194 	};
1195 
1196 	mutex_lock(&d->mbd_dev->lock);
1197 
1198 	for (i = 0; i < MTDSWAP_TREE_CNT; i++) {
1199 		struct rb_root *root = &d->trees[i].root;
1200 
1201 		if (root->rb_node) {
1202 			count[i] = d->trees[i].count;
1203 			min[i] = MTDSWAP_ECNT_MIN(root);
1204 			max[i] = MTDSWAP_ECNT_MAX(root);
1205 		} else
1206 			count[i] = 0;
1207 	}
1208 
1209 	if (d->curr_write) {
1210 		cw = 1;
1211 		cwp = d->curr_write_pos;
1212 		cwecount = d->curr_write->erase_count;
1213 	}
1214 
1215 	sum = 0;
1216 	for (i = 0; i < d->eblks; i++)
1217 		sum += d->eb_data[i].erase_count;
1218 
1219 	use_size = (uint64_t)d->eblks * d->mtd->erasesize;
1220 	bb_cnt = mtdswap_badblocks(d->mtd, use_size);
1221 
1222 	mapped = 0;
1223 	pages = d->mbd_dev->size;
1224 	for (i = 0; i < pages; i++)
1225 		if (d->page_data[i] != BLOCK_UNDEF)
1226 			mapped++;
1227 
1228 	mutex_unlock(&d->mbd_dev->lock);
1229 
1230 	for (i = 0; i < MTDSWAP_TREE_CNT; i++) {
1231 		if (!count[i])
1232 			continue;
1233 
1234 		if (min[i] != max[i])
1235 			seq_printf(s, "%s:\t%5d erase blocks, erased min %d, "
1236 				"max %d times\n",
1237 				name[i], count[i], min[i], max[i]);
1238 		else
1239 			seq_printf(s, "%s:\t%5d erase blocks, all erased %d "
1240 				"times\n", name[i], count[i], min[i]);
1241 	}
1242 
1243 	if (bb_cnt)
1244 		seq_printf(s, "bad:\t%5u erase blocks\n", bb_cnt);
1245 
1246 	if (cw)
1247 		seq_printf(s, "current erase block: %u pages used, %u free, "
1248 			"erased %u times\n",
1249 			cwp, d->pages_per_eblk - cwp, cwecount);
1250 
1251 	seq_printf(s, "total erasures: %lu\n", sum);
1252 
1253 	seq_puts(s, "\n");
1254 
1255 	seq_printf(s, "mtdswap_readsect count: %llu\n", d->sect_read_count);
1256 	seq_printf(s, "mtdswap_writesect count: %llu\n", d->sect_write_count);
1257 	seq_printf(s, "mtdswap_discard count: %llu\n", d->discard_count);
1258 	seq_printf(s, "mtd read count: %llu\n", d->mtd_read_count);
1259 	seq_printf(s, "mtd write count: %llu\n", d->mtd_write_count);
1260 	seq_printf(s, "discarded pages count: %llu\n", d->discard_page_count);
1261 
1262 	seq_puts(s, "\n");
1263 	seq_printf(s, "total pages: %u\n", pages);
1264 	seq_printf(s, "pages mapped: %u\n", mapped);
1265 
1266 	return 0;
1267 }
1268 DEFINE_SHOW_ATTRIBUTE(mtdswap);
1269 
1270 static int mtdswap_add_debugfs(struct mtdswap_dev *d)
1271 {
1272 	struct dentry *root = d->mtd->dbg.dfs_dir;
1273 	struct dentry *dent;
1274 
1275 	if (!IS_ENABLED(CONFIG_DEBUG_FS))
1276 		return 0;
1277 
1278 	if (IS_ERR_OR_NULL(root))
1279 		return -1;
1280 
1281 	dent = debugfs_create_file("mtdswap_stats", S_IRUSR, root, d,
1282 				&mtdswap_fops);
1283 	if (!dent) {
1284 		dev_err(d->dev, "debugfs_create_file failed\n");
1285 		return -1;
1286 	}
1287 
1288 	return 0;
1289 }
1290 
1291 static int mtdswap_init(struct mtdswap_dev *d, unsigned int eblocks,
1292 			unsigned int spare_cnt)
1293 {
1294 	struct mtd_info *mtd = d->mbd_dev->mtd;
1295 	unsigned int i, eblk_bytes, pages, blocks;
1296 	int ret = -ENOMEM;
1297 
1298 	d->mtd = mtd;
1299 	d->eblks = eblocks;
1300 	d->spare_eblks = spare_cnt;
1301 	d->pages_per_eblk = mtd->erasesize >> PAGE_SHIFT;
1302 
1303 	pages = d->mbd_dev->size;
1304 	blocks = eblocks * d->pages_per_eblk;
1305 
1306 	for (i = 0; i < MTDSWAP_TREE_CNT; i++)
1307 		d->trees[i].root = RB_ROOT;
1308 
1309 	d->page_data = vmalloc(array_size(pages, sizeof(int)));
1310 	if (!d->page_data)
1311 		goto page_data_fail;
1312 
1313 	d->revmap = vmalloc(array_size(blocks, sizeof(int)));
1314 	if (!d->revmap)
1315 		goto revmap_fail;
1316 
1317 	eblk_bytes = sizeof(struct swap_eb)*d->eblks;
1318 	d->eb_data = vzalloc(eblk_bytes);
1319 	if (!d->eb_data)
1320 		goto eb_data_fail;
1321 
1322 	for (i = 0; i < pages; i++)
1323 		d->page_data[i] = BLOCK_UNDEF;
1324 
1325 	for (i = 0; i < blocks; i++)
1326 		d->revmap[i] = PAGE_UNDEF;
1327 
1328 	d->page_buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1329 	if (!d->page_buf)
1330 		goto page_buf_fail;
1331 
1332 	d->oob_buf = kmalloc_array(2, mtd->oobavail, GFP_KERNEL);
1333 	if (!d->oob_buf)
1334 		goto oob_buf_fail;
1335 
1336 	mtdswap_scan_eblks(d);
1337 
1338 	return 0;
1339 
1340 oob_buf_fail:
1341 	kfree(d->page_buf);
1342 page_buf_fail:
1343 	vfree(d->eb_data);
1344 eb_data_fail:
1345 	vfree(d->revmap);
1346 revmap_fail:
1347 	vfree(d->page_data);
1348 page_data_fail:
1349 	printk(KERN_ERR "%s: init failed (%d)\n", MTDSWAP_PREFIX, ret);
1350 	return ret;
1351 }
1352 
1353 static void mtdswap_add_mtd(struct mtd_blktrans_ops *tr, struct mtd_info *mtd)
1354 {
1355 	struct mtdswap_dev *d;
1356 	struct mtd_blktrans_dev *mbd_dev;
1357 	char *parts;
1358 	char *this_opt;
1359 	unsigned long part;
1360 	unsigned int eblocks, eavailable, bad_blocks, spare_cnt;
1361 	uint64_t swap_size, use_size, size_limit;
1362 	int ret;
1363 
1364 	parts = &partitions[0];
1365 	if (!*parts)
1366 		return;
1367 
1368 	while ((this_opt = strsep(&parts, ",")) != NULL) {
1369 		if (kstrtoul(this_opt, 0, &part) < 0)
1370 			return;
1371 
1372 		if (mtd->index == part)
1373 			break;
1374 	}
1375 
1376 	if (mtd->index != part)
1377 		return;
1378 
1379 	if (mtd->erasesize < PAGE_SIZE || mtd->erasesize % PAGE_SIZE) {
1380 		printk(KERN_ERR "%s: Erase size %u not multiple of PAGE_SIZE "
1381 			"%lu\n", MTDSWAP_PREFIX, mtd->erasesize, PAGE_SIZE);
1382 		return;
1383 	}
1384 
1385 	if (PAGE_SIZE % mtd->writesize || mtd->writesize > PAGE_SIZE) {
1386 		printk(KERN_ERR "%s: PAGE_SIZE %lu not multiple of write size"
1387 			" %u\n", MTDSWAP_PREFIX, PAGE_SIZE, mtd->writesize);
1388 		return;
1389 	}
1390 
1391 	if (!mtd->oobsize || mtd->oobavail < MTDSWAP_OOBSIZE) {
1392 		printk(KERN_ERR "%s: Not enough free bytes in OOB, "
1393 			"%d available, %zu needed.\n",
1394 			MTDSWAP_PREFIX, mtd->oobavail, MTDSWAP_OOBSIZE);
1395 		return;
1396 	}
1397 
1398 	if (spare_eblocks > 100)
1399 		spare_eblocks = 100;
1400 
1401 	use_size = mtd->size;
1402 	size_limit = (uint64_t) BLOCK_MAX * PAGE_SIZE;
1403 
1404 	if (mtd->size > size_limit) {
1405 		printk(KERN_WARNING "%s: Device too large. Limiting size to "
1406 			"%llu bytes\n", MTDSWAP_PREFIX, size_limit);
1407 		use_size = size_limit;
1408 	}
1409 
1410 	eblocks = mtd_div_by_eb(use_size, mtd);
1411 	use_size = (uint64_t)eblocks * mtd->erasesize;
1412 	bad_blocks = mtdswap_badblocks(mtd, use_size);
1413 	eavailable = eblocks - bad_blocks;
1414 
1415 	if (eavailable < MIN_ERASE_BLOCKS) {
1416 		printk(KERN_ERR "%s: Not enough erase blocks. %u available, "
1417 			"%d needed\n", MTDSWAP_PREFIX, eavailable,
1418 			MIN_ERASE_BLOCKS);
1419 		return;
1420 	}
1421 
1422 	spare_cnt = div_u64((uint64_t)eavailable * spare_eblocks, 100);
1423 
1424 	if (spare_cnt < MIN_SPARE_EBLOCKS)
1425 		spare_cnt = MIN_SPARE_EBLOCKS;
1426 
1427 	if (spare_cnt > eavailable - 1)
1428 		spare_cnt = eavailable - 1;
1429 
1430 	swap_size = (uint64_t)(eavailable - spare_cnt) * mtd->erasesize +
1431 		(header ? PAGE_SIZE : 0);
1432 
1433 	printk(KERN_INFO "%s: Enabling MTD swap on device %lu, size %llu KB, "
1434 		"%u spare, %u bad blocks\n",
1435 		MTDSWAP_PREFIX, part, swap_size / 1024, spare_cnt, bad_blocks);
1436 
1437 	d = kzalloc(sizeof(struct mtdswap_dev), GFP_KERNEL);
1438 	if (!d)
1439 		return;
1440 
1441 	mbd_dev = kzalloc(sizeof(struct mtd_blktrans_dev), GFP_KERNEL);
1442 	if (!mbd_dev) {
1443 		kfree(d);
1444 		return;
1445 	}
1446 
1447 	d->mbd_dev = mbd_dev;
1448 	mbd_dev->priv = d;
1449 
1450 	mbd_dev->mtd = mtd;
1451 	mbd_dev->devnum = mtd->index;
1452 	mbd_dev->size = swap_size >> PAGE_SHIFT;
1453 	mbd_dev->tr = tr;
1454 
1455 	if (!(mtd->flags & MTD_WRITEABLE))
1456 		mbd_dev->readonly = 1;
1457 
1458 	if (mtdswap_init(d, eblocks, spare_cnt) < 0)
1459 		goto init_failed;
1460 
1461 	if (add_mtd_blktrans_dev(mbd_dev) < 0)
1462 		goto cleanup;
1463 
1464 	d->dev = disk_to_dev(mbd_dev->disk);
1465 
1466 	ret = mtdswap_add_debugfs(d);
1467 	if (ret < 0)
1468 		goto debugfs_failed;
1469 
1470 	return;
1471 
1472 debugfs_failed:
1473 	del_mtd_blktrans_dev(mbd_dev);
1474 
1475 cleanup:
1476 	mtdswap_cleanup(d);
1477 
1478 init_failed:
1479 	kfree(mbd_dev);
1480 	kfree(d);
1481 }
1482 
1483 static void mtdswap_remove_dev(struct mtd_blktrans_dev *dev)
1484 {
1485 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1486 
1487 	del_mtd_blktrans_dev(dev);
1488 	mtdswap_cleanup(d);
1489 	kfree(d);
1490 }
1491 
1492 static struct mtd_blktrans_ops mtdswap_ops = {
1493 	.name		= "mtdswap",
1494 	.major		= 0,
1495 	.part_bits	= 0,
1496 	.blksize	= PAGE_SIZE,
1497 	.flush		= mtdswap_flush,
1498 	.readsect	= mtdswap_readsect,
1499 	.writesect	= mtdswap_writesect,
1500 	.discard	= mtdswap_discard,
1501 	.background	= mtdswap_background,
1502 	.add_mtd	= mtdswap_add_mtd,
1503 	.remove_dev	= mtdswap_remove_dev,
1504 	.owner		= THIS_MODULE,
1505 };
1506 
1507 static int __init mtdswap_modinit(void)
1508 {
1509 	return register_mtd_blktrans(&mtdswap_ops);
1510 }
1511 
1512 static void __exit mtdswap_modexit(void)
1513 {
1514 	deregister_mtd_blktrans(&mtdswap_ops);
1515 }
1516 
1517 module_init(mtdswap_modinit);
1518 module_exit(mtdswap_modexit);
1519 
1520 
1521 MODULE_LICENSE("GPL");
1522 MODULE_AUTHOR("Jarkko Lavinen <jarkko.lavinen@nokia.com>");
1523 MODULE_DESCRIPTION("Block device access to an MTD suitable for using as "
1524 		"swap space");
1525