xref: /openbmc/linux/drivers/mtd/mtdswap.c (revision aac5987a)
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_TRESHOLD	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 	struct dentry *debugfs_root;
143 };
144 
145 struct mtdswap_oobdata {
146 	__le16 magic;
147 	__le32 count;
148 } __packed;
149 
150 #define MTDSWAP_MAGIC_CLEAN	0x2095
151 #define MTDSWAP_MAGIC_DIRTY	(MTDSWAP_MAGIC_CLEAN + 1)
152 #define MTDSWAP_TYPE_CLEAN	0
153 #define MTDSWAP_TYPE_DIRTY	1
154 #define MTDSWAP_OOBSIZE		sizeof(struct mtdswap_oobdata)
155 
156 #define MTDSWAP_ERASE_RETRIES	3 /* Before marking erase block bad */
157 #define MTDSWAP_IO_RETRIES	3
158 
159 enum {
160 	MTDSWAP_SCANNED_CLEAN,
161 	MTDSWAP_SCANNED_DIRTY,
162 	MTDSWAP_SCANNED_BITFLIP,
163 	MTDSWAP_SCANNED_BAD,
164 };
165 
166 /*
167  * In the worst case mtdswap_writesect() has allocated the last clean
168  * page from the current block and is then pre-empted by the GC
169  * thread. The thread can consume a full erase block when moving a
170  * block.
171  */
172 #define MIN_SPARE_EBLOCKS	2
173 #define MIN_ERASE_BLOCKS	(MIN_SPARE_EBLOCKS + 1)
174 
175 #define TREE_ROOT(d, name) (&d->trees[MTDSWAP_ ## name].root)
176 #define TREE_EMPTY(d, name) (TREE_ROOT(d, name)->rb_node == NULL)
177 #define TREE_NONEMPTY(d, name) (!TREE_EMPTY(d, name))
178 #define TREE_COUNT(d, name) (d->trees[MTDSWAP_ ## name].count)
179 
180 #define MTDSWAP_MBD_TO_MTDSWAP(dev) ((struct mtdswap_dev *)dev->priv)
181 
182 static char partitions[128] = "";
183 module_param_string(partitions, partitions, sizeof(partitions), 0444);
184 MODULE_PARM_DESC(partitions, "MTD partition numbers to use as swap "
185 		"partitions=\"1,3,5\"");
186 
187 static unsigned int spare_eblocks = 10;
188 module_param(spare_eblocks, uint, 0444);
189 MODULE_PARM_DESC(spare_eblocks, "Percentage of spare erase blocks for "
190 		"garbage collection (default 10%)");
191 
192 static bool header; /* false */
193 module_param(header, bool, 0444);
194 MODULE_PARM_DESC(header,
195 		"Include builtin swap header (default 0, without header)");
196 
197 static int mtdswap_gc(struct mtdswap_dev *d, unsigned int background);
198 
199 static loff_t mtdswap_eb_offset(struct mtdswap_dev *d, struct swap_eb *eb)
200 {
201 	return (loff_t)(eb - d->eb_data) * d->mtd->erasesize;
202 }
203 
204 static void mtdswap_eb_detach(struct mtdswap_dev *d, struct swap_eb *eb)
205 {
206 	unsigned int oldidx;
207 	struct mtdswap_tree *tp;
208 
209 	if (eb->root) {
210 		tp = container_of(eb->root, struct mtdswap_tree, root);
211 		oldidx = tp - &d->trees[0];
212 
213 		d->trees[oldidx].count--;
214 		rb_erase(&eb->rb, eb->root);
215 	}
216 }
217 
218 static void __mtdswap_rb_add(struct rb_root *root, struct swap_eb *eb)
219 {
220 	struct rb_node **p, *parent = NULL;
221 	struct swap_eb *cur;
222 
223 	p = &root->rb_node;
224 	while (*p) {
225 		parent = *p;
226 		cur = rb_entry(parent, struct swap_eb, rb);
227 		if (eb->erase_count > cur->erase_count)
228 			p = &(*p)->rb_right;
229 		else
230 			p = &(*p)->rb_left;
231 	}
232 
233 	rb_link_node(&eb->rb, parent, p);
234 	rb_insert_color(&eb->rb, root);
235 }
236 
237 static void mtdswap_rb_add(struct mtdswap_dev *d, struct swap_eb *eb, int idx)
238 {
239 	struct rb_root *root;
240 
241 	if (eb->root == &d->trees[idx].root)
242 		return;
243 
244 	mtdswap_eb_detach(d, eb);
245 	root = &d->trees[idx].root;
246 	__mtdswap_rb_add(root, eb);
247 	eb->root = root;
248 	d->trees[idx].count++;
249 }
250 
251 static struct rb_node *mtdswap_rb_index(struct rb_root *root, unsigned int idx)
252 {
253 	struct rb_node *p;
254 	unsigned int i;
255 
256 	p = rb_first(root);
257 	i = 0;
258 	while (i < idx && p) {
259 		p = rb_next(p);
260 		i++;
261 	}
262 
263 	return p;
264 }
265 
266 static int mtdswap_handle_badblock(struct mtdswap_dev *d, struct swap_eb *eb)
267 {
268 	int ret;
269 	loff_t offset;
270 
271 	d->spare_eblks--;
272 	eb->flags |= EBLOCK_BAD;
273 	mtdswap_eb_detach(d, eb);
274 	eb->root = NULL;
275 
276 	/* badblocks not supported */
277 	if (!mtd_can_have_bb(d->mtd))
278 		return 1;
279 
280 	offset = mtdswap_eb_offset(d, eb);
281 	dev_warn(d->dev, "Marking bad block at %08llx\n", offset);
282 	ret = mtd_block_markbad(d->mtd, offset);
283 
284 	if (ret) {
285 		dev_warn(d->dev, "Mark block bad failed for block at %08llx "
286 			"error %d\n", offset, ret);
287 		return ret;
288 	}
289 
290 	return 1;
291 
292 }
293 
294 static int mtdswap_handle_write_error(struct mtdswap_dev *d, struct swap_eb *eb)
295 {
296 	unsigned int marked = eb->flags & EBLOCK_FAILED;
297 	struct swap_eb *curr_write = d->curr_write;
298 
299 	eb->flags |= EBLOCK_FAILED;
300 	if (curr_write == eb) {
301 		d->curr_write = NULL;
302 
303 		if (!marked && d->curr_write_pos != 0) {
304 			mtdswap_rb_add(d, eb, MTDSWAP_FAILING);
305 			return 0;
306 		}
307 	}
308 
309 	return mtdswap_handle_badblock(d, eb);
310 }
311 
312 static int mtdswap_read_oob(struct mtdswap_dev *d, loff_t from,
313 			struct mtd_oob_ops *ops)
314 {
315 	int ret = mtd_read_oob(d->mtd, from, ops);
316 
317 	if (mtd_is_bitflip(ret))
318 		return ret;
319 
320 	if (ret) {
321 		dev_warn(d->dev, "Read OOB failed %d for block at %08llx\n",
322 			ret, from);
323 		return ret;
324 	}
325 
326 	if (ops->oobretlen < ops->ooblen) {
327 		dev_warn(d->dev, "Read OOB return short read (%zd bytes not "
328 			"%zd) for block at %08llx\n",
329 			ops->oobretlen, ops->ooblen, from);
330 		return -EIO;
331 	}
332 
333 	return 0;
334 }
335 
336 static int mtdswap_read_markers(struct mtdswap_dev *d, struct swap_eb *eb)
337 {
338 	struct mtdswap_oobdata *data, *data2;
339 	int ret;
340 	loff_t offset;
341 	struct mtd_oob_ops ops;
342 
343 	offset = mtdswap_eb_offset(d, eb);
344 
345 	/* Check first if the block is bad. */
346 	if (mtd_can_have_bb(d->mtd) && mtd_block_isbad(d->mtd, offset))
347 		return MTDSWAP_SCANNED_BAD;
348 
349 	ops.ooblen = 2 * d->mtd->oobavail;
350 	ops.oobbuf = d->oob_buf;
351 	ops.ooboffs = 0;
352 	ops.datbuf = NULL;
353 	ops.mode = MTD_OPS_AUTO_OOB;
354 
355 	ret = mtdswap_read_oob(d, offset, &ops);
356 
357 	if (ret && !mtd_is_bitflip(ret))
358 		return ret;
359 
360 	data = (struct mtdswap_oobdata *)d->oob_buf;
361 	data2 = (struct mtdswap_oobdata *)
362 		(d->oob_buf + d->mtd->oobavail);
363 
364 	if (le16_to_cpu(data->magic) == MTDSWAP_MAGIC_CLEAN) {
365 		eb->erase_count = le32_to_cpu(data->count);
366 		if (mtd_is_bitflip(ret))
367 			ret = MTDSWAP_SCANNED_BITFLIP;
368 		else {
369 			if (le16_to_cpu(data2->magic) == MTDSWAP_MAGIC_DIRTY)
370 				ret = MTDSWAP_SCANNED_DIRTY;
371 			else
372 				ret = MTDSWAP_SCANNED_CLEAN;
373 		}
374 	} else {
375 		eb->flags |= EBLOCK_NOMAGIC;
376 		ret = MTDSWAP_SCANNED_DIRTY;
377 	}
378 
379 	return ret;
380 }
381 
382 static int mtdswap_write_marker(struct mtdswap_dev *d, struct swap_eb *eb,
383 				u16 marker)
384 {
385 	struct mtdswap_oobdata n;
386 	int ret;
387 	loff_t offset;
388 	struct mtd_oob_ops ops;
389 
390 	ops.ooboffs = 0;
391 	ops.oobbuf = (uint8_t *)&n;
392 	ops.mode = MTD_OPS_AUTO_OOB;
393 	ops.datbuf = NULL;
394 
395 	if (marker == MTDSWAP_TYPE_CLEAN) {
396 		n.magic = cpu_to_le16(MTDSWAP_MAGIC_CLEAN);
397 		n.count = cpu_to_le32(eb->erase_count);
398 		ops.ooblen = MTDSWAP_OOBSIZE;
399 		offset = mtdswap_eb_offset(d, eb);
400 	} else {
401 		n.magic = cpu_to_le16(MTDSWAP_MAGIC_DIRTY);
402 		ops.ooblen = sizeof(n.magic);
403 		offset = mtdswap_eb_offset(d, eb) + d->mtd->writesize;
404 	}
405 
406 	ret = mtd_write_oob(d->mtd, offset, &ops);
407 
408 	if (ret) {
409 		dev_warn(d->dev, "Write OOB failed for block at %08llx "
410 			"error %d\n", offset, ret);
411 		if (ret == -EIO || mtd_is_eccerr(ret))
412 			mtdswap_handle_write_error(d, eb);
413 		return ret;
414 	}
415 
416 	if (ops.oobretlen != ops.ooblen) {
417 		dev_warn(d->dev, "Short OOB write for block at %08llx: "
418 			"%zd not %zd\n",
419 			offset, ops.oobretlen, ops.ooblen);
420 		return ret;
421 	}
422 
423 	return 0;
424 }
425 
426 /*
427  * Are there any erase blocks without MAGIC_CLEAN header, presumably
428  * because power was cut off after erase but before header write? We
429  * need to guestimate the erase count.
430  */
431 static void mtdswap_check_counts(struct mtdswap_dev *d)
432 {
433 	struct rb_root hist_root = RB_ROOT;
434 	struct rb_node *medrb;
435 	struct swap_eb *eb;
436 	unsigned int i, cnt, median;
437 
438 	cnt = 0;
439 	for (i = 0; i < d->eblks; i++) {
440 		eb = d->eb_data + i;
441 
442 		if (eb->flags & (EBLOCK_NOMAGIC | EBLOCK_BAD | EBLOCK_READERR))
443 			continue;
444 
445 		__mtdswap_rb_add(&hist_root, eb);
446 		cnt++;
447 	}
448 
449 	if (cnt == 0)
450 		return;
451 
452 	medrb = mtdswap_rb_index(&hist_root, cnt / 2);
453 	median = rb_entry(medrb, struct swap_eb, rb)->erase_count;
454 
455 	d->max_erase_count = MTDSWAP_ECNT_MAX(&hist_root);
456 
457 	for (i = 0; i < d->eblks; i++) {
458 		eb = d->eb_data + i;
459 
460 		if (eb->flags & (EBLOCK_NOMAGIC | EBLOCK_READERR))
461 			eb->erase_count = median;
462 
463 		if (eb->flags & (EBLOCK_NOMAGIC | EBLOCK_BAD | EBLOCK_READERR))
464 			continue;
465 
466 		rb_erase(&eb->rb, &hist_root);
467 	}
468 }
469 
470 static void mtdswap_scan_eblks(struct mtdswap_dev *d)
471 {
472 	int status;
473 	unsigned int i, idx;
474 	struct swap_eb *eb;
475 
476 	for (i = 0; i < d->eblks; i++) {
477 		eb = d->eb_data + i;
478 
479 		status = mtdswap_read_markers(d, eb);
480 		if (status < 0)
481 			eb->flags |= EBLOCK_READERR;
482 		else if (status == MTDSWAP_SCANNED_BAD) {
483 			eb->flags |= EBLOCK_BAD;
484 			continue;
485 		}
486 
487 		switch (status) {
488 		case MTDSWAP_SCANNED_CLEAN:
489 			idx = MTDSWAP_CLEAN;
490 			break;
491 		case MTDSWAP_SCANNED_DIRTY:
492 		case MTDSWAP_SCANNED_BITFLIP:
493 			idx = MTDSWAP_DIRTY;
494 			break;
495 		default:
496 			idx = MTDSWAP_FAILING;
497 		}
498 
499 		eb->flags |= (idx << EBLOCK_IDX_SHIFT);
500 	}
501 
502 	mtdswap_check_counts(d);
503 
504 	for (i = 0; i < d->eblks; i++) {
505 		eb = d->eb_data + i;
506 
507 		if (eb->flags & EBLOCK_BAD)
508 			continue;
509 
510 		idx = eb->flags >> EBLOCK_IDX_SHIFT;
511 		mtdswap_rb_add(d, eb, idx);
512 	}
513 }
514 
515 /*
516  * Place eblk into a tree corresponding to its number of active blocks
517  * it contains.
518  */
519 static void mtdswap_store_eb(struct mtdswap_dev *d, struct swap_eb *eb)
520 {
521 	unsigned int weight = eb->active_count;
522 	unsigned int maxweight = d->pages_per_eblk;
523 
524 	if (eb == d->curr_write)
525 		return;
526 
527 	if (eb->flags & EBLOCK_BITFLIP)
528 		mtdswap_rb_add(d, eb, MTDSWAP_BITFLIP);
529 	else if (eb->flags & (EBLOCK_READERR | EBLOCK_FAILED))
530 		mtdswap_rb_add(d, eb, MTDSWAP_FAILING);
531 	if (weight == maxweight)
532 		mtdswap_rb_add(d, eb, MTDSWAP_USED);
533 	else if (weight == 0)
534 		mtdswap_rb_add(d, eb, MTDSWAP_DIRTY);
535 	else if (weight > (maxweight/2))
536 		mtdswap_rb_add(d, eb, MTDSWAP_LOWFRAG);
537 	else
538 		mtdswap_rb_add(d, eb, MTDSWAP_HIFRAG);
539 }
540 
541 
542 static void mtdswap_erase_callback(struct erase_info *done)
543 {
544 	wait_queue_head_t *wait_q = (wait_queue_head_t *)done->priv;
545 	wake_up(wait_q);
546 }
547 
548 static int mtdswap_erase_block(struct mtdswap_dev *d, struct swap_eb *eb)
549 {
550 	struct mtd_info *mtd = d->mtd;
551 	struct erase_info erase;
552 	wait_queue_head_t wq;
553 	unsigned int retries = 0;
554 	int ret;
555 
556 	eb->erase_count++;
557 	if (eb->erase_count > d->max_erase_count)
558 		d->max_erase_count = eb->erase_count;
559 
560 retry:
561 	init_waitqueue_head(&wq);
562 	memset(&erase, 0, sizeof(struct erase_info));
563 
564 	erase.mtd	= mtd;
565 	erase.callback	= mtdswap_erase_callback;
566 	erase.addr	= mtdswap_eb_offset(d, eb);
567 	erase.len	= mtd->erasesize;
568 	erase.priv	= (u_long)&wq;
569 
570 	ret = mtd_erase(mtd, &erase);
571 	if (ret) {
572 		if (retries++ < MTDSWAP_ERASE_RETRIES) {
573 			dev_warn(d->dev,
574 				"erase of erase block %#llx on %s failed",
575 				erase.addr, mtd->name);
576 			yield();
577 			goto retry;
578 		}
579 
580 		dev_err(d->dev, "Cannot erase erase block %#llx on %s\n",
581 			erase.addr, mtd->name);
582 
583 		mtdswap_handle_badblock(d, eb);
584 		return -EIO;
585 	}
586 
587 	ret = wait_event_interruptible(wq, erase.state == MTD_ERASE_DONE ||
588 					   erase.state == MTD_ERASE_FAILED);
589 	if (ret) {
590 		dev_err(d->dev, "Interrupted erase block %#llx erasure on %s\n",
591 			erase.addr, mtd->name);
592 		return -EINTR;
593 	}
594 
595 	if (erase.state == MTD_ERASE_FAILED) {
596 		if (retries++ < MTDSWAP_ERASE_RETRIES) {
597 			dev_warn(d->dev,
598 				"erase of erase block %#llx on %s failed",
599 				erase.addr, mtd->name);
600 			yield();
601 			goto retry;
602 		}
603 
604 		mtdswap_handle_badblock(d, eb);
605 		return -EIO;
606 	}
607 
608 	return 0;
609 }
610 
611 static int mtdswap_map_free_block(struct mtdswap_dev *d, unsigned int page,
612 				unsigned int *block)
613 {
614 	int ret;
615 	struct swap_eb *old_eb = d->curr_write;
616 	struct rb_root *clean_root;
617 	struct swap_eb *eb;
618 
619 	if (old_eb == NULL || d->curr_write_pos >= d->pages_per_eblk) {
620 		do {
621 			if (TREE_EMPTY(d, CLEAN))
622 				return -ENOSPC;
623 
624 			clean_root = TREE_ROOT(d, CLEAN);
625 			eb = rb_entry(rb_first(clean_root), struct swap_eb, rb);
626 			rb_erase(&eb->rb, clean_root);
627 			eb->root = NULL;
628 			TREE_COUNT(d, CLEAN)--;
629 
630 			ret = mtdswap_write_marker(d, eb, MTDSWAP_TYPE_DIRTY);
631 		} while (ret == -EIO || mtd_is_eccerr(ret));
632 
633 		if (ret)
634 			return ret;
635 
636 		d->curr_write_pos = 0;
637 		d->curr_write = eb;
638 		if (old_eb)
639 			mtdswap_store_eb(d, old_eb);
640 	}
641 
642 	*block = (d->curr_write - d->eb_data) * d->pages_per_eblk +
643 		d->curr_write_pos;
644 
645 	d->curr_write->active_count++;
646 	d->revmap[*block] = page;
647 	d->curr_write_pos++;
648 
649 	return 0;
650 }
651 
652 static unsigned int mtdswap_free_page_cnt(struct mtdswap_dev *d)
653 {
654 	return TREE_COUNT(d, CLEAN) * d->pages_per_eblk +
655 		d->pages_per_eblk - d->curr_write_pos;
656 }
657 
658 static unsigned int mtdswap_enough_free_pages(struct mtdswap_dev *d)
659 {
660 	return mtdswap_free_page_cnt(d) > d->pages_per_eblk;
661 }
662 
663 static int mtdswap_write_block(struct mtdswap_dev *d, char *buf,
664 			unsigned int page, unsigned int *bp, int gc_context)
665 {
666 	struct mtd_info *mtd = d->mtd;
667 	struct swap_eb *eb;
668 	size_t retlen;
669 	loff_t writepos;
670 	int ret;
671 
672 retry:
673 	if (!gc_context)
674 		while (!mtdswap_enough_free_pages(d))
675 			if (mtdswap_gc(d, 0) > 0)
676 				return -ENOSPC;
677 
678 	ret = mtdswap_map_free_block(d, page, bp);
679 	eb = d->eb_data + (*bp / d->pages_per_eblk);
680 
681 	if (ret == -EIO || mtd_is_eccerr(ret)) {
682 		d->curr_write = NULL;
683 		eb->active_count--;
684 		d->revmap[*bp] = PAGE_UNDEF;
685 		goto retry;
686 	}
687 
688 	if (ret < 0)
689 		return ret;
690 
691 	writepos = (loff_t)*bp << PAGE_SHIFT;
692 	ret =  mtd_write(mtd, writepos, PAGE_SIZE, &retlen, buf);
693 	if (ret == -EIO || mtd_is_eccerr(ret)) {
694 		d->curr_write_pos--;
695 		eb->active_count--;
696 		d->revmap[*bp] = PAGE_UNDEF;
697 		mtdswap_handle_write_error(d, eb);
698 		goto retry;
699 	}
700 
701 	if (ret < 0) {
702 		dev_err(d->dev, "Write to MTD device failed: %d (%zd written)",
703 			ret, retlen);
704 		goto err;
705 	}
706 
707 	if (retlen != PAGE_SIZE) {
708 		dev_err(d->dev, "Short write to MTD device: %zd written",
709 			retlen);
710 		ret = -EIO;
711 		goto err;
712 	}
713 
714 	return ret;
715 
716 err:
717 	d->curr_write_pos--;
718 	eb->active_count--;
719 	d->revmap[*bp] = PAGE_UNDEF;
720 
721 	return ret;
722 }
723 
724 static int mtdswap_move_block(struct mtdswap_dev *d, unsigned int oldblock,
725 		unsigned int *newblock)
726 {
727 	struct mtd_info *mtd = d->mtd;
728 	struct swap_eb *eb, *oldeb;
729 	int ret;
730 	size_t retlen;
731 	unsigned int page, retries;
732 	loff_t readpos;
733 
734 	page = d->revmap[oldblock];
735 	readpos = (loff_t) oldblock << PAGE_SHIFT;
736 	retries = 0;
737 
738 retry:
739 	ret = mtd_read(mtd, readpos, PAGE_SIZE, &retlen, d->page_buf);
740 
741 	if (ret < 0 && !mtd_is_bitflip(ret)) {
742 		oldeb = d->eb_data + oldblock / d->pages_per_eblk;
743 		oldeb->flags |= EBLOCK_READERR;
744 
745 		dev_err(d->dev, "Read Error: %d (block %u)\n", ret,
746 			oldblock);
747 		retries++;
748 		if (retries < MTDSWAP_IO_RETRIES)
749 			goto retry;
750 
751 		goto read_error;
752 	}
753 
754 	if (retlen != PAGE_SIZE) {
755 		dev_err(d->dev, "Short read: %zd (block %u)\n", retlen,
756 		       oldblock);
757 		ret = -EIO;
758 		goto read_error;
759 	}
760 
761 	ret = mtdswap_write_block(d, d->page_buf, page, newblock, 1);
762 	if (ret < 0) {
763 		d->page_data[page] = BLOCK_ERROR;
764 		dev_err(d->dev, "Write error: %d\n", ret);
765 		return ret;
766 	}
767 
768 	eb = d->eb_data + *newblock / d->pages_per_eblk;
769 	d->page_data[page] = *newblock;
770 	d->revmap[oldblock] = PAGE_UNDEF;
771 	eb = d->eb_data + oldblock / d->pages_per_eblk;
772 	eb->active_count--;
773 
774 	return 0;
775 
776 read_error:
777 	d->page_data[page] = BLOCK_ERROR;
778 	d->revmap[oldblock] = PAGE_UNDEF;
779 	return ret;
780 }
781 
782 static int mtdswap_gc_eblock(struct mtdswap_dev *d, struct swap_eb *eb)
783 {
784 	unsigned int i, block, eblk_base, newblock;
785 	int ret, errcode;
786 
787 	errcode = 0;
788 	eblk_base = (eb - d->eb_data) * d->pages_per_eblk;
789 
790 	for (i = 0; i < d->pages_per_eblk; i++) {
791 		if (d->spare_eblks < MIN_SPARE_EBLOCKS)
792 			return -ENOSPC;
793 
794 		block = eblk_base + i;
795 		if (d->revmap[block] == PAGE_UNDEF)
796 			continue;
797 
798 		ret = mtdswap_move_block(d, block, &newblock);
799 		if (ret < 0 && !errcode)
800 			errcode = ret;
801 	}
802 
803 	return errcode;
804 }
805 
806 static int __mtdswap_choose_gc_tree(struct mtdswap_dev *d)
807 {
808 	int idx, stopat;
809 
810 	if (TREE_COUNT(d, CLEAN) < LOW_FRAG_GC_TRESHOLD)
811 		stopat = MTDSWAP_LOWFRAG;
812 	else
813 		stopat = MTDSWAP_HIFRAG;
814 
815 	for (idx = MTDSWAP_BITFLIP; idx >= stopat; idx--)
816 		if (d->trees[idx].root.rb_node != NULL)
817 			return idx;
818 
819 	return -1;
820 }
821 
822 static int mtdswap_wlfreq(unsigned int maxdiff)
823 {
824 	unsigned int h, x, y, dist, base;
825 
826 	/*
827 	 * Calculate linear ramp down from f1 to f2 when maxdiff goes from
828 	 * MAX_ERASE_DIFF to MAX_ERASE_DIFF + COLLECT_NONDIRTY_BASE.  Similar
829 	 * to triangle with height f1 - f1 and width COLLECT_NONDIRTY_BASE.
830 	 */
831 
832 	dist = maxdiff - MAX_ERASE_DIFF;
833 	if (dist > COLLECT_NONDIRTY_BASE)
834 		dist = COLLECT_NONDIRTY_BASE;
835 
836 	/*
837 	 * Modelling the slop as right angular triangle with base
838 	 * COLLECT_NONDIRTY_BASE and height freq1 - freq2. The ratio y/x is
839 	 * equal to the ratio h/base.
840 	 */
841 	h = COLLECT_NONDIRTY_FREQ1 - COLLECT_NONDIRTY_FREQ2;
842 	base = COLLECT_NONDIRTY_BASE;
843 
844 	x = dist - base;
845 	y = (x * h + base / 2) / base;
846 
847 	return COLLECT_NONDIRTY_FREQ2 + y;
848 }
849 
850 static int mtdswap_choose_wl_tree(struct mtdswap_dev *d)
851 {
852 	static unsigned int pick_cnt;
853 	unsigned int i, idx = -1, wear, max;
854 	struct rb_root *root;
855 
856 	max = 0;
857 	for (i = 0; i <= MTDSWAP_DIRTY; i++) {
858 		root = &d->trees[i].root;
859 		if (root->rb_node == NULL)
860 			continue;
861 
862 		wear = d->max_erase_count - MTDSWAP_ECNT_MIN(root);
863 		if (wear > max) {
864 			max = wear;
865 			idx = i;
866 		}
867 	}
868 
869 	if (max > MAX_ERASE_DIFF && pick_cnt >= mtdswap_wlfreq(max) - 1) {
870 		pick_cnt = 0;
871 		return idx;
872 	}
873 
874 	pick_cnt++;
875 	return -1;
876 }
877 
878 static int mtdswap_choose_gc_tree(struct mtdswap_dev *d,
879 				unsigned int background)
880 {
881 	int idx;
882 
883 	if (TREE_NONEMPTY(d, FAILING) &&
884 		(background || (TREE_EMPTY(d, CLEAN) && TREE_EMPTY(d, DIRTY))))
885 		return MTDSWAP_FAILING;
886 
887 	idx = mtdswap_choose_wl_tree(d);
888 	if (idx >= MTDSWAP_CLEAN)
889 		return idx;
890 
891 	return __mtdswap_choose_gc_tree(d);
892 }
893 
894 static struct swap_eb *mtdswap_pick_gc_eblk(struct mtdswap_dev *d,
895 					unsigned int background)
896 {
897 	struct rb_root *rp = NULL;
898 	struct swap_eb *eb = NULL;
899 	int idx;
900 
901 	if (background && TREE_COUNT(d, CLEAN) > CLEAN_BLOCK_THRESHOLD &&
902 		TREE_EMPTY(d, DIRTY) && TREE_EMPTY(d, FAILING))
903 		return NULL;
904 
905 	idx = mtdswap_choose_gc_tree(d, background);
906 	if (idx < 0)
907 		return NULL;
908 
909 	rp = &d->trees[idx].root;
910 	eb = rb_entry(rb_first(rp), struct swap_eb, rb);
911 
912 	rb_erase(&eb->rb, rp);
913 	eb->root = NULL;
914 	d->trees[idx].count--;
915 	return eb;
916 }
917 
918 static unsigned int mtdswap_test_patt(unsigned int i)
919 {
920 	return i % 2 ? 0x55555555 : 0xAAAAAAAA;
921 }
922 
923 static unsigned int mtdswap_eblk_passes(struct mtdswap_dev *d,
924 					struct swap_eb *eb)
925 {
926 	struct mtd_info *mtd = d->mtd;
927 	unsigned int test, i, j, patt, mtd_pages;
928 	loff_t base, pos;
929 	unsigned int *p1 = (unsigned int *)d->page_buf;
930 	unsigned char *p2 = (unsigned char *)d->oob_buf;
931 	struct mtd_oob_ops ops;
932 	int ret;
933 
934 	ops.mode = MTD_OPS_AUTO_OOB;
935 	ops.len = mtd->writesize;
936 	ops.ooblen = mtd->oobavail;
937 	ops.ooboffs = 0;
938 	ops.datbuf = d->page_buf;
939 	ops.oobbuf = d->oob_buf;
940 	base = mtdswap_eb_offset(d, eb);
941 	mtd_pages = d->pages_per_eblk * PAGE_SIZE / mtd->writesize;
942 
943 	for (test = 0; test < 2; test++) {
944 		pos = base;
945 		for (i = 0; i < mtd_pages; i++) {
946 			patt = mtdswap_test_patt(test + i);
947 			memset(d->page_buf, patt, mtd->writesize);
948 			memset(d->oob_buf, patt, mtd->oobavail);
949 			ret = mtd_write_oob(mtd, pos, &ops);
950 			if (ret)
951 				goto error;
952 
953 			pos += mtd->writesize;
954 		}
955 
956 		pos = base;
957 		for (i = 0; i < mtd_pages; i++) {
958 			ret = mtd_read_oob(mtd, pos, &ops);
959 			if (ret)
960 				goto error;
961 
962 			patt = mtdswap_test_patt(test + i);
963 			for (j = 0; j < mtd->writesize/sizeof(int); j++)
964 				if (p1[j] != patt)
965 					goto error;
966 
967 			for (j = 0; j < mtd->oobavail; j++)
968 				if (p2[j] != (unsigned char)patt)
969 					goto error;
970 
971 			pos += mtd->writesize;
972 		}
973 
974 		ret = mtdswap_erase_block(d, eb);
975 		if (ret)
976 			goto error;
977 	}
978 
979 	eb->flags &= ~EBLOCK_READERR;
980 	return 1;
981 
982 error:
983 	mtdswap_handle_badblock(d, eb);
984 	return 0;
985 }
986 
987 static int mtdswap_gc(struct mtdswap_dev *d, unsigned int background)
988 {
989 	struct swap_eb *eb;
990 	int ret;
991 
992 	if (d->spare_eblks < MIN_SPARE_EBLOCKS)
993 		return 1;
994 
995 	eb = mtdswap_pick_gc_eblk(d, background);
996 	if (!eb)
997 		return 1;
998 
999 	ret = mtdswap_gc_eblock(d, eb);
1000 	if (ret == -ENOSPC)
1001 		return 1;
1002 
1003 	if (eb->flags & EBLOCK_FAILED) {
1004 		mtdswap_handle_badblock(d, eb);
1005 		return 0;
1006 	}
1007 
1008 	eb->flags &= ~EBLOCK_BITFLIP;
1009 	ret = mtdswap_erase_block(d, eb);
1010 	if ((eb->flags & EBLOCK_READERR) &&
1011 		(ret || !mtdswap_eblk_passes(d, eb)))
1012 		return 0;
1013 
1014 	if (ret == 0)
1015 		ret = mtdswap_write_marker(d, eb, MTDSWAP_TYPE_CLEAN);
1016 
1017 	if (ret == 0)
1018 		mtdswap_rb_add(d, eb, MTDSWAP_CLEAN);
1019 	else if (ret != -EIO && !mtd_is_eccerr(ret))
1020 		mtdswap_rb_add(d, eb, MTDSWAP_DIRTY);
1021 
1022 	return 0;
1023 }
1024 
1025 static void mtdswap_background(struct mtd_blktrans_dev *dev)
1026 {
1027 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1028 	int ret;
1029 
1030 	while (1) {
1031 		ret = mtdswap_gc(d, 1);
1032 		if (ret || mtd_blktrans_cease_background(dev))
1033 			return;
1034 	}
1035 }
1036 
1037 static void mtdswap_cleanup(struct mtdswap_dev *d)
1038 {
1039 	vfree(d->eb_data);
1040 	vfree(d->revmap);
1041 	vfree(d->page_data);
1042 	kfree(d->oob_buf);
1043 	kfree(d->page_buf);
1044 }
1045 
1046 static int mtdswap_flush(struct mtd_blktrans_dev *dev)
1047 {
1048 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1049 
1050 	mtd_sync(d->mtd);
1051 	return 0;
1052 }
1053 
1054 static unsigned int mtdswap_badblocks(struct mtd_info *mtd, uint64_t size)
1055 {
1056 	loff_t offset;
1057 	unsigned int badcnt;
1058 
1059 	badcnt = 0;
1060 
1061 	if (mtd_can_have_bb(mtd))
1062 		for (offset = 0; offset < size; offset += mtd->erasesize)
1063 			if (mtd_block_isbad(mtd, offset))
1064 				badcnt++;
1065 
1066 	return badcnt;
1067 }
1068 
1069 static int mtdswap_writesect(struct mtd_blktrans_dev *dev,
1070 			unsigned long page, char *buf)
1071 {
1072 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1073 	unsigned int newblock, mapped;
1074 	struct swap_eb *eb;
1075 	int ret;
1076 
1077 	d->sect_write_count++;
1078 
1079 	if (d->spare_eblks < MIN_SPARE_EBLOCKS)
1080 		return -ENOSPC;
1081 
1082 	if (header) {
1083 		/* Ignore writes to the header page */
1084 		if (unlikely(page == 0))
1085 			return 0;
1086 
1087 		page--;
1088 	}
1089 
1090 	mapped = d->page_data[page];
1091 	if (mapped <= BLOCK_MAX) {
1092 		eb = d->eb_data + (mapped / d->pages_per_eblk);
1093 		eb->active_count--;
1094 		mtdswap_store_eb(d, eb);
1095 		d->page_data[page] = BLOCK_UNDEF;
1096 		d->revmap[mapped] = PAGE_UNDEF;
1097 	}
1098 
1099 	ret = mtdswap_write_block(d, buf, page, &newblock, 0);
1100 	d->mtd_write_count++;
1101 
1102 	if (ret < 0)
1103 		return ret;
1104 
1105 	eb = d->eb_data + (newblock / d->pages_per_eblk);
1106 	d->page_data[page] = newblock;
1107 
1108 	return 0;
1109 }
1110 
1111 /* Provide a dummy swap header for the kernel */
1112 static int mtdswap_auto_header(struct mtdswap_dev *d, char *buf)
1113 {
1114 	union swap_header *hd = (union swap_header *)(buf);
1115 
1116 	memset(buf, 0, PAGE_SIZE - 10);
1117 
1118 	hd->info.version = 1;
1119 	hd->info.last_page = d->mbd_dev->size - 1;
1120 	hd->info.nr_badpages = 0;
1121 
1122 	memcpy(buf + PAGE_SIZE - 10, "SWAPSPACE2", 10);
1123 
1124 	return 0;
1125 }
1126 
1127 static int mtdswap_readsect(struct mtd_blktrans_dev *dev,
1128 			unsigned long page, char *buf)
1129 {
1130 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1131 	struct mtd_info *mtd = d->mtd;
1132 	unsigned int realblock, retries;
1133 	loff_t readpos;
1134 	struct swap_eb *eb;
1135 	size_t retlen;
1136 	int ret;
1137 
1138 	d->sect_read_count++;
1139 
1140 	if (header) {
1141 		if (unlikely(page == 0))
1142 			return mtdswap_auto_header(d, buf);
1143 
1144 		page--;
1145 	}
1146 
1147 	realblock = d->page_data[page];
1148 	if (realblock > BLOCK_MAX) {
1149 		memset(buf, 0x0, PAGE_SIZE);
1150 		if (realblock == BLOCK_UNDEF)
1151 			return 0;
1152 		else
1153 			return -EIO;
1154 	}
1155 
1156 	eb = d->eb_data + (realblock / d->pages_per_eblk);
1157 	BUG_ON(d->revmap[realblock] == PAGE_UNDEF);
1158 
1159 	readpos = (loff_t)realblock << PAGE_SHIFT;
1160 	retries = 0;
1161 
1162 retry:
1163 	ret = mtd_read(mtd, readpos, PAGE_SIZE, &retlen, buf);
1164 
1165 	d->mtd_read_count++;
1166 	if (mtd_is_bitflip(ret)) {
1167 		eb->flags |= EBLOCK_BITFLIP;
1168 		mtdswap_rb_add(d, eb, MTDSWAP_BITFLIP);
1169 		ret = 0;
1170 	}
1171 
1172 	if (ret < 0) {
1173 		dev_err(d->dev, "Read error %d\n", ret);
1174 		eb->flags |= EBLOCK_READERR;
1175 		mtdswap_rb_add(d, eb, MTDSWAP_FAILING);
1176 		retries++;
1177 		if (retries < MTDSWAP_IO_RETRIES)
1178 			goto retry;
1179 
1180 		return ret;
1181 	}
1182 
1183 	if (retlen != PAGE_SIZE) {
1184 		dev_err(d->dev, "Short read %zd\n", retlen);
1185 		return -EIO;
1186 	}
1187 
1188 	return 0;
1189 }
1190 
1191 static int mtdswap_discard(struct mtd_blktrans_dev *dev, unsigned long first,
1192 			unsigned nr_pages)
1193 {
1194 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1195 	unsigned long page;
1196 	struct swap_eb *eb;
1197 	unsigned int mapped;
1198 
1199 	d->discard_count++;
1200 
1201 	for (page = first; page < first + nr_pages; page++) {
1202 		mapped = d->page_data[page];
1203 		if (mapped <= BLOCK_MAX) {
1204 			eb = d->eb_data + (mapped / d->pages_per_eblk);
1205 			eb->active_count--;
1206 			mtdswap_store_eb(d, eb);
1207 			d->page_data[page] = BLOCK_UNDEF;
1208 			d->revmap[mapped] = PAGE_UNDEF;
1209 			d->discard_page_count++;
1210 		} else if (mapped == BLOCK_ERROR) {
1211 			d->page_data[page] = BLOCK_UNDEF;
1212 			d->discard_page_count++;
1213 		}
1214 	}
1215 
1216 	return 0;
1217 }
1218 
1219 static int mtdswap_show(struct seq_file *s, void *data)
1220 {
1221 	struct mtdswap_dev *d = (struct mtdswap_dev *) s->private;
1222 	unsigned long sum;
1223 	unsigned int count[MTDSWAP_TREE_CNT];
1224 	unsigned int min[MTDSWAP_TREE_CNT];
1225 	unsigned int max[MTDSWAP_TREE_CNT];
1226 	unsigned int i, cw = 0, cwp = 0, cwecount = 0, bb_cnt, mapped, pages;
1227 	uint64_t use_size;
1228 	char *name[] = {"clean", "used", "low", "high", "dirty", "bitflip",
1229 			"failing"};
1230 
1231 	mutex_lock(&d->mbd_dev->lock);
1232 
1233 	for (i = 0; i < MTDSWAP_TREE_CNT; i++) {
1234 		struct rb_root *root = &d->trees[i].root;
1235 
1236 		if (root->rb_node) {
1237 			count[i] = d->trees[i].count;
1238 			min[i] = rb_entry(rb_first(root), struct swap_eb,
1239 					rb)->erase_count;
1240 			max[i] = rb_entry(rb_last(root), struct swap_eb,
1241 					rb)->erase_count;
1242 		} else
1243 			count[i] = 0;
1244 	}
1245 
1246 	if (d->curr_write) {
1247 		cw = 1;
1248 		cwp = d->curr_write_pos;
1249 		cwecount = d->curr_write->erase_count;
1250 	}
1251 
1252 	sum = 0;
1253 	for (i = 0; i < d->eblks; i++)
1254 		sum += d->eb_data[i].erase_count;
1255 
1256 	use_size = (uint64_t)d->eblks * d->mtd->erasesize;
1257 	bb_cnt = mtdswap_badblocks(d->mtd, use_size);
1258 
1259 	mapped = 0;
1260 	pages = d->mbd_dev->size;
1261 	for (i = 0; i < pages; i++)
1262 		if (d->page_data[i] != BLOCK_UNDEF)
1263 			mapped++;
1264 
1265 	mutex_unlock(&d->mbd_dev->lock);
1266 
1267 	for (i = 0; i < MTDSWAP_TREE_CNT; i++) {
1268 		if (!count[i])
1269 			continue;
1270 
1271 		if (min[i] != max[i])
1272 			seq_printf(s, "%s:\t%5d erase blocks, erased min %d, "
1273 				"max %d times\n",
1274 				name[i], count[i], min[i], max[i]);
1275 		else
1276 			seq_printf(s, "%s:\t%5d erase blocks, all erased %d "
1277 				"times\n", name[i], count[i], min[i]);
1278 	}
1279 
1280 	if (bb_cnt)
1281 		seq_printf(s, "bad:\t%5u erase blocks\n", bb_cnt);
1282 
1283 	if (cw)
1284 		seq_printf(s, "current erase block: %u pages used, %u free, "
1285 			"erased %u times\n",
1286 			cwp, d->pages_per_eblk - cwp, cwecount);
1287 
1288 	seq_printf(s, "total erasures: %lu\n", sum);
1289 
1290 	seq_puts(s, "\n");
1291 
1292 	seq_printf(s, "mtdswap_readsect count: %llu\n", d->sect_read_count);
1293 	seq_printf(s, "mtdswap_writesect count: %llu\n", d->sect_write_count);
1294 	seq_printf(s, "mtdswap_discard count: %llu\n", d->discard_count);
1295 	seq_printf(s, "mtd read count: %llu\n", d->mtd_read_count);
1296 	seq_printf(s, "mtd write count: %llu\n", d->mtd_write_count);
1297 	seq_printf(s, "discarded pages count: %llu\n", d->discard_page_count);
1298 
1299 	seq_puts(s, "\n");
1300 	seq_printf(s, "total pages: %u\n", pages);
1301 	seq_printf(s, "pages mapped: %u\n", mapped);
1302 
1303 	return 0;
1304 }
1305 
1306 static int mtdswap_open(struct inode *inode, struct file *file)
1307 {
1308 	return single_open(file, mtdswap_show, inode->i_private);
1309 }
1310 
1311 static const struct file_operations mtdswap_fops = {
1312 	.open		= mtdswap_open,
1313 	.read		= seq_read,
1314 	.llseek		= seq_lseek,
1315 	.release	= single_release,
1316 };
1317 
1318 static int mtdswap_add_debugfs(struct mtdswap_dev *d)
1319 {
1320 	struct gendisk *gd = d->mbd_dev->disk;
1321 	struct device *dev = disk_to_dev(gd);
1322 
1323 	struct dentry *root;
1324 	struct dentry *dent;
1325 
1326 	root = debugfs_create_dir(gd->disk_name, NULL);
1327 	if (IS_ERR(root))
1328 		return 0;
1329 
1330 	if (!root) {
1331 		dev_err(dev, "failed to initialize debugfs\n");
1332 		return -1;
1333 	}
1334 
1335 	d->debugfs_root = root;
1336 
1337 	dent = debugfs_create_file("stats", S_IRUSR, root, d,
1338 				&mtdswap_fops);
1339 	if (!dent) {
1340 		dev_err(d->dev, "debugfs_create_file failed\n");
1341 		debugfs_remove_recursive(root);
1342 		d->debugfs_root = NULL;
1343 		return -1;
1344 	}
1345 
1346 	return 0;
1347 }
1348 
1349 static int mtdswap_init(struct mtdswap_dev *d, unsigned int eblocks,
1350 			unsigned int spare_cnt)
1351 {
1352 	struct mtd_info *mtd = d->mbd_dev->mtd;
1353 	unsigned int i, eblk_bytes, pages, blocks;
1354 	int ret = -ENOMEM;
1355 
1356 	d->mtd = mtd;
1357 	d->eblks = eblocks;
1358 	d->spare_eblks = spare_cnt;
1359 	d->pages_per_eblk = mtd->erasesize >> PAGE_SHIFT;
1360 
1361 	pages = d->mbd_dev->size;
1362 	blocks = eblocks * d->pages_per_eblk;
1363 
1364 	for (i = 0; i < MTDSWAP_TREE_CNT; i++)
1365 		d->trees[i].root = RB_ROOT;
1366 
1367 	d->page_data = vmalloc(sizeof(int)*pages);
1368 	if (!d->page_data)
1369 		goto page_data_fail;
1370 
1371 	d->revmap = vmalloc(sizeof(int)*blocks);
1372 	if (!d->revmap)
1373 		goto revmap_fail;
1374 
1375 	eblk_bytes = sizeof(struct swap_eb)*d->eblks;
1376 	d->eb_data = vzalloc(eblk_bytes);
1377 	if (!d->eb_data)
1378 		goto eb_data_fail;
1379 
1380 	for (i = 0; i < pages; i++)
1381 		d->page_data[i] = BLOCK_UNDEF;
1382 
1383 	for (i = 0; i < blocks; i++)
1384 		d->revmap[i] = PAGE_UNDEF;
1385 
1386 	d->page_buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1387 	if (!d->page_buf)
1388 		goto page_buf_fail;
1389 
1390 	d->oob_buf = kmalloc(2 * mtd->oobavail, GFP_KERNEL);
1391 	if (!d->oob_buf)
1392 		goto oob_buf_fail;
1393 
1394 	mtdswap_scan_eblks(d);
1395 
1396 	return 0;
1397 
1398 oob_buf_fail:
1399 	kfree(d->page_buf);
1400 page_buf_fail:
1401 	vfree(d->eb_data);
1402 eb_data_fail:
1403 	vfree(d->revmap);
1404 revmap_fail:
1405 	vfree(d->page_data);
1406 page_data_fail:
1407 	printk(KERN_ERR "%s: init failed (%d)\n", MTDSWAP_PREFIX, ret);
1408 	return ret;
1409 }
1410 
1411 static void mtdswap_add_mtd(struct mtd_blktrans_ops *tr, struct mtd_info *mtd)
1412 {
1413 	struct mtdswap_dev *d;
1414 	struct mtd_blktrans_dev *mbd_dev;
1415 	char *parts;
1416 	char *this_opt;
1417 	unsigned long part;
1418 	unsigned int eblocks, eavailable, bad_blocks, spare_cnt;
1419 	uint64_t swap_size, use_size, size_limit;
1420 	int ret;
1421 
1422 	parts = &partitions[0];
1423 	if (!*parts)
1424 		return;
1425 
1426 	while ((this_opt = strsep(&parts, ",")) != NULL) {
1427 		if (kstrtoul(this_opt, 0, &part) < 0)
1428 			return;
1429 
1430 		if (mtd->index == part)
1431 			break;
1432 	}
1433 
1434 	if (mtd->index != part)
1435 		return;
1436 
1437 	if (mtd->erasesize < PAGE_SIZE || mtd->erasesize % PAGE_SIZE) {
1438 		printk(KERN_ERR "%s: Erase size %u not multiple of PAGE_SIZE "
1439 			"%lu\n", MTDSWAP_PREFIX, mtd->erasesize, PAGE_SIZE);
1440 		return;
1441 	}
1442 
1443 	if (PAGE_SIZE % mtd->writesize || mtd->writesize > PAGE_SIZE) {
1444 		printk(KERN_ERR "%s: PAGE_SIZE %lu not multiple of write size"
1445 			" %u\n", MTDSWAP_PREFIX, PAGE_SIZE, mtd->writesize);
1446 		return;
1447 	}
1448 
1449 	if (!mtd->oobsize || mtd->oobavail < MTDSWAP_OOBSIZE) {
1450 		printk(KERN_ERR "%s: Not enough free bytes in OOB, "
1451 			"%d available, %zu needed.\n",
1452 			MTDSWAP_PREFIX, mtd->oobavail, MTDSWAP_OOBSIZE);
1453 		return;
1454 	}
1455 
1456 	if (spare_eblocks > 100)
1457 		spare_eblocks = 100;
1458 
1459 	use_size = mtd->size;
1460 	size_limit = (uint64_t) BLOCK_MAX * PAGE_SIZE;
1461 
1462 	if (mtd->size > size_limit) {
1463 		printk(KERN_WARNING "%s: Device too large. Limiting size to "
1464 			"%llu bytes\n", MTDSWAP_PREFIX, size_limit);
1465 		use_size = size_limit;
1466 	}
1467 
1468 	eblocks = mtd_div_by_eb(use_size, mtd);
1469 	use_size = (uint64_t)eblocks * mtd->erasesize;
1470 	bad_blocks = mtdswap_badblocks(mtd, use_size);
1471 	eavailable = eblocks - bad_blocks;
1472 
1473 	if (eavailable < MIN_ERASE_BLOCKS) {
1474 		printk(KERN_ERR "%s: Not enough erase blocks. %u available, "
1475 			"%d needed\n", MTDSWAP_PREFIX, eavailable,
1476 			MIN_ERASE_BLOCKS);
1477 		return;
1478 	}
1479 
1480 	spare_cnt = div_u64((uint64_t)eavailable * spare_eblocks, 100);
1481 
1482 	if (spare_cnt < MIN_SPARE_EBLOCKS)
1483 		spare_cnt = MIN_SPARE_EBLOCKS;
1484 
1485 	if (spare_cnt > eavailable - 1)
1486 		spare_cnt = eavailable - 1;
1487 
1488 	swap_size = (uint64_t)(eavailable - spare_cnt) * mtd->erasesize +
1489 		(header ? PAGE_SIZE : 0);
1490 
1491 	printk(KERN_INFO "%s: Enabling MTD swap on device %lu, size %llu KB, "
1492 		"%u spare, %u bad blocks\n",
1493 		MTDSWAP_PREFIX, part, swap_size / 1024, spare_cnt, bad_blocks);
1494 
1495 	d = kzalloc(sizeof(struct mtdswap_dev), GFP_KERNEL);
1496 	if (!d)
1497 		return;
1498 
1499 	mbd_dev = kzalloc(sizeof(struct mtd_blktrans_dev), GFP_KERNEL);
1500 	if (!mbd_dev) {
1501 		kfree(d);
1502 		return;
1503 	}
1504 
1505 	d->mbd_dev = mbd_dev;
1506 	mbd_dev->priv = d;
1507 
1508 	mbd_dev->mtd = mtd;
1509 	mbd_dev->devnum = mtd->index;
1510 	mbd_dev->size = swap_size >> PAGE_SHIFT;
1511 	mbd_dev->tr = tr;
1512 
1513 	if (!(mtd->flags & MTD_WRITEABLE))
1514 		mbd_dev->readonly = 1;
1515 
1516 	if (mtdswap_init(d, eblocks, spare_cnt) < 0)
1517 		goto init_failed;
1518 
1519 	if (add_mtd_blktrans_dev(mbd_dev) < 0)
1520 		goto cleanup;
1521 
1522 	d->dev = disk_to_dev(mbd_dev->disk);
1523 
1524 	ret = mtdswap_add_debugfs(d);
1525 	if (ret < 0)
1526 		goto debugfs_failed;
1527 
1528 	return;
1529 
1530 debugfs_failed:
1531 	del_mtd_blktrans_dev(mbd_dev);
1532 
1533 cleanup:
1534 	mtdswap_cleanup(d);
1535 
1536 init_failed:
1537 	kfree(mbd_dev);
1538 	kfree(d);
1539 }
1540 
1541 static void mtdswap_remove_dev(struct mtd_blktrans_dev *dev)
1542 {
1543 	struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1544 
1545 	debugfs_remove_recursive(d->debugfs_root);
1546 	del_mtd_blktrans_dev(dev);
1547 	mtdswap_cleanup(d);
1548 	kfree(d);
1549 }
1550 
1551 static struct mtd_blktrans_ops mtdswap_ops = {
1552 	.name		= "mtdswap",
1553 	.major		= 0,
1554 	.part_bits	= 0,
1555 	.blksize	= PAGE_SIZE,
1556 	.flush		= mtdswap_flush,
1557 	.readsect	= mtdswap_readsect,
1558 	.writesect	= mtdswap_writesect,
1559 	.discard	= mtdswap_discard,
1560 	.background	= mtdswap_background,
1561 	.add_mtd	= mtdswap_add_mtd,
1562 	.remove_dev	= mtdswap_remove_dev,
1563 	.owner		= THIS_MODULE,
1564 };
1565 
1566 static int __init mtdswap_modinit(void)
1567 {
1568 	return register_mtd_blktrans(&mtdswap_ops);
1569 }
1570 
1571 static void __exit mtdswap_modexit(void)
1572 {
1573 	deregister_mtd_blktrans(&mtdswap_ops);
1574 }
1575 
1576 module_init(mtdswap_modinit);
1577 module_exit(mtdswap_modexit);
1578 
1579 
1580 MODULE_LICENSE("GPL");
1581 MODULE_AUTHOR("Jarkko Lavinen <jarkko.lavinen@nokia.com>");
1582 MODULE_DESCRIPTION("Block device access to an MTD suitable for using as "
1583 		"swap space");
1584