1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Swap block device support for MTDs
4 * Turns an MTD device into a swap device with block wear leveling
5 *
6 * Copyright © 2007,2011 Nokia Corporation. All rights reserved.
7 *
8 * Authors: Jarkko Lavinen <jarkko.lavinen@nokia.com>
9 *
10 * Based on Richard Purdie's earlier implementation in 2007. Background
11 * support and lock-less operation written by Adrian Hunter.
12 */
13
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/mtd/mtd.h>
17 #include <linux/mtd/blktrans.h>
18 #include <linux/rbtree.h>
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/vmalloc.h>
22 #include <linux/genhd.h>
23 #include <linux/swap.h>
24 #include <linux/debugfs.h>
25 #include <linux/seq_file.h>
26 #include <linux/device.h>
27 #include <linux/math64.h>
28
29 #define MTDSWAP_PREFIX "mtdswap"
30
31 /*
32 * The number of free eraseblocks when GC should stop
33 */
34 #define CLEAN_BLOCK_THRESHOLD 20
35
36 /*
37 * Number of free eraseblocks below which GC can also collect low frag
38 * blocks.
39 */
40 #define LOW_FRAG_GC_THRESHOLD 5
41
42 /*
43 * Wear level cost amortization. We want to do wear leveling on the background
44 * without disturbing gc too much. This is made by defining max GC frequency.
45 * Frequency value 6 means 1/6 of the GC passes will pick an erase block based
46 * on the biggest wear difference rather than the biggest dirtiness.
47 *
48 * The lower freq2 should be chosen so that it makes sure the maximum erase
49 * difference will decrease even if a malicious application is deliberately
50 * trying to make erase differences large.
51 */
52 #define MAX_ERASE_DIFF 4000
53 #define COLLECT_NONDIRTY_BASE MAX_ERASE_DIFF
54 #define COLLECT_NONDIRTY_FREQ1 6
55 #define COLLECT_NONDIRTY_FREQ2 4
56
57 #define PAGE_UNDEF UINT_MAX
58 #define BLOCK_UNDEF UINT_MAX
59 #define BLOCK_ERROR (UINT_MAX - 1)
60 #define BLOCK_MAX (UINT_MAX - 2)
61
62 #define EBLOCK_BAD (1 << 0)
63 #define EBLOCK_NOMAGIC (1 << 1)
64 #define EBLOCK_BITFLIP (1 << 2)
65 #define EBLOCK_FAILED (1 << 3)
66 #define EBLOCK_READERR (1 << 4)
67 #define EBLOCK_IDX_SHIFT 5
68
69 struct swap_eb {
70 struct rb_node rb;
71 struct rb_root *root;
72
73 unsigned int flags;
74 unsigned int active_count;
75 unsigned int erase_count;
76 unsigned int pad; /* speeds up pointer decrement */
77 };
78
79 #define MTDSWAP_ECNT_MIN(rbroot) (rb_entry(rb_first(rbroot), struct swap_eb, \
80 rb)->erase_count)
81 #define MTDSWAP_ECNT_MAX(rbroot) (rb_entry(rb_last(rbroot), struct swap_eb, \
82 rb)->erase_count)
83
84 struct mtdswap_tree {
85 struct rb_root root;
86 unsigned int count;
87 };
88
89 enum {
90 MTDSWAP_CLEAN,
91 MTDSWAP_USED,
92 MTDSWAP_LOWFRAG,
93 MTDSWAP_HIFRAG,
94 MTDSWAP_DIRTY,
95 MTDSWAP_BITFLIP,
96 MTDSWAP_FAILING,
97 MTDSWAP_TREE_CNT,
98 };
99
100 struct mtdswap_dev {
101 struct mtd_blktrans_dev *mbd_dev;
102 struct mtd_info *mtd;
103 struct device *dev;
104
105 unsigned int *page_data;
106 unsigned int *revmap;
107
108 unsigned int eblks;
109 unsigned int spare_eblks;
110 unsigned int pages_per_eblk;
111 unsigned int max_erase_count;
112 struct swap_eb *eb_data;
113
114 struct mtdswap_tree trees[MTDSWAP_TREE_CNT];
115
116 unsigned long long sect_read_count;
117 unsigned long long sect_write_count;
118 unsigned long long mtd_write_count;
119 unsigned long long mtd_read_count;
120 unsigned long long discard_count;
121 unsigned long long discard_page_count;
122
123 unsigned int curr_write_pos;
124 struct swap_eb *curr_write;
125
126 char *page_buf;
127 char *oob_buf;
128 };
129
130 struct mtdswap_oobdata {
131 __le16 magic;
132 __le32 count;
133 } __packed;
134
135 #define MTDSWAP_MAGIC_CLEAN 0x2095
136 #define MTDSWAP_MAGIC_DIRTY (MTDSWAP_MAGIC_CLEAN + 1)
137 #define MTDSWAP_TYPE_CLEAN 0
138 #define MTDSWAP_TYPE_DIRTY 1
139 #define MTDSWAP_OOBSIZE sizeof(struct mtdswap_oobdata)
140
141 #define MTDSWAP_ERASE_RETRIES 3 /* Before marking erase block bad */
142 #define MTDSWAP_IO_RETRIES 3
143
144 enum {
145 MTDSWAP_SCANNED_CLEAN,
146 MTDSWAP_SCANNED_DIRTY,
147 MTDSWAP_SCANNED_BITFLIP,
148 MTDSWAP_SCANNED_BAD,
149 };
150
151 /*
152 * In the worst case mtdswap_writesect() has allocated the last clean
153 * page from the current block and is then pre-empted by the GC
154 * thread. The thread can consume a full erase block when moving a
155 * block.
156 */
157 #define MIN_SPARE_EBLOCKS 2
158 #define MIN_ERASE_BLOCKS (MIN_SPARE_EBLOCKS + 1)
159
160 #define TREE_ROOT(d, name) (&d->trees[MTDSWAP_ ## name].root)
161 #define TREE_EMPTY(d, name) (TREE_ROOT(d, name)->rb_node == NULL)
162 #define TREE_NONEMPTY(d, name) (!TREE_EMPTY(d, name))
163 #define TREE_COUNT(d, name) (d->trees[MTDSWAP_ ## name].count)
164
165 #define MTDSWAP_MBD_TO_MTDSWAP(dev) ((struct mtdswap_dev *)dev->priv)
166
167 static char partitions[128] = "";
168 module_param_string(partitions, partitions, sizeof(partitions), 0444);
169 MODULE_PARM_DESC(partitions, "MTD partition numbers to use as swap "
170 "partitions=\"1,3,5\"");
171
172 static unsigned int spare_eblocks = 10;
173 module_param(spare_eblocks, uint, 0444);
174 MODULE_PARM_DESC(spare_eblocks, "Percentage of spare erase blocks for "
175 "garbage collection (default 10%)");
176
177 static bool header; /* false */
178 module_param(header, bool, 0444);
179 MODULE_PARM_DESC(header,
180 "Include builtin swap header (default 0, without header)");
181
182 static int mtdswap_gc(struct mtdswap_dev *d, unsigned int background);
183
mtdswap_eb_offset(struct mtdswap_dev * d,struct swap_eb * eb)184 static loff_t mtdswap_eb_offset(struct mtdswap_dev *d, struct swap_eb *eb)
185 {
186 return (loff_t)(eb - d->eb_data) * d->mtd->erasesize;
187 }
188
mtdswap_eb_detach(struct mtdswap_dev * d,struct swap_eb * eb)189 static void mtdswap_eb_detach(struct mtdswap_dev *d, struct swap_eb *eb)
190 {
191 unsigned int oldidx;
192 struct mtdswap_tree *tp;
193
194 if (eb->root) {
195 tp = container_of(eb->root, struct mtdswap_tree, root);
196 oldidx = tp - &d->trees[0];
197
198 d->trees[oldidx].count--;
199 rb_erase(&eb->rb, eb->root);
200 }
201 }
202
__mtdswap_rb_add(struct rb_root * root,struct swap_eb * eb)203 static void __mtdswap_rb_add(struct rb_root *root, struct swap_eb *eb)
204 {
205 struct rb_node **p, *parent = NULL;
206 struct swap_eb *cur;
207
208 p = &root->rb_node;
209 while (*p) {
210 parent = *p;
211 cur = rb_entry(parent, struct swap_eb, rb);
212 if (eb->erase_count > cur->erase_count)
213 p = &(*p)->rb_right;
214 else
215 p = &(*p)->rb_left;
216 }
217
218 rb_link_node(&eb->rb, parent, p);
219 rb_insert_color(&eb->rb, root);
220 }
221
mtdswap_rb_add(struct mtdswap_dev * d,struct swap_eb * eb,int idx)222 static void mtdswap_rb_add(struct mtdswap_dev *d, struct swap_eb *eb, int idx)
223 {
224 struct rb_root *root;
225
226 if (eb->root == &d->trees[idx].root)
227 return;
228
229 mtdswap_eb_detach(d, eb);
230 root = &d->trees[idx].root;
231 __mtdswap_rb_add(root, eb);
232 eb->root = root;
233 d->trees[idx].count++;
234 }
235
mtdswap_rb_index(struct rb_root * root,unsigned int idx)236 static struct rb_node *mtdswap_rb_index(struct rb_root *root, unsigned int idx)
237 {
238 struct rb_node *p;
239 unsigned int i;
240
241 p = rb_first(root);
242 i = 0;
243 while (i < idx && p) {
244 p = rb_next(p);
245 i++;
246 }
247
248 return p;
249 }
250
mtdswap_handle_badblock(struct mtdswap_dev * d,struct swap_eb * eb)251 static int mtdswap_handle_badblock(struct mtdswap_dev *d, struct swap_eb *eb)
252 {
253 int ret;
254 loff_t offset;
255
256 d->spare_eblks--;
257 eb->flags |= EBLOCK_BAD;
258 mtdswap_eb_detach(d, eb);
259 eb->root = NULL;
260
261 /* badblocks not supported */
262 if (!mtd_can_have_bb(d->mtd))
263 return 1;
264
265 offset = mtdswap_eb_offset(d, eb);
266 dev_warn(d->dev, "Marking bad block at %08llx\n", offset);
267 ret = mtd_block_markbad(d->mtd, offset);
268
269 if (ret) {
270 dev_warn(d->dev, "Mark block bad failed for block at %08llx "
271 "error %d\n", offset, ret);
272 return ret;
273 }
274
275 return 1;
276
277 }
278
mtdswap_handle_write_error(struct mtdswap_dev * d,struct swap_eb * eb)279 static int mtdswap_handle_write_error(struct mtdswap_dev *d, struct swap_eb *eb)
280 {
281 unsigned int marked = eb->flags & EBLOCK_FAILED;
282 struct swap_eb *curr_write = d->curr_write;
283
284 eb->flags |= EBLOCK_FAILED;
285 if (curr_write == eb) {
286 d->curr_write = NULL;
287
288 if (!marked && d->curr_write_pos != 0) {
289 mtdswap_rb_add(d, eb, MTDSWAP_FAILING);
290 return 0;
291 }
292 }
293
294 return mtdswap_handle_badblock(d, eb);
295 }
296
mtdswap_read_oob(struct mtdswap_dev * d,loff_t from,struct mtd_oob_ops * ops)297 static int mtdswap_read_oob(struct mtdswap_dev *d, loff_t from,
298 struct mtd_oob_ops *ops)
299 {
300 int ret = mtd_read_oob(d->mtd, from, ops);
301
302 if (mtd_is_bitflip(ret))
303 return ret;
304
305 if (ret) {
306 dev_warn(d->dev, "Read OOB failed %d for block at %08llx\n",
307 ret, from);
308 return ret;
309 }
310
311 if (ops->oobretlen < ops->ooblen) {
312 dev_warn(d->dev, "Read OOB return short read (%zd bytes not "
313 "%zd) for block at %08llx\n",
314 ops->oobretlen, ops->ooblen, from);
315 return -EIO;
316 }
317
318 return 0;
319 }
320
mtdswap_read_markers(struct mtdswap_dev * d,struct swap_eb * eb)321 static int mtdswap_read_markers(struct mtdswap_dev *d, struct swap_eb *eb)
322 {
323 struct mtdswap_oobdata *data, *data2;
324 int ret;
325 loff_t offset;
326 struct mtd_oob_ops ops;
327
328 offset = mtdswap_eb_offset(d, eb);
329
330 /* Check first if the block is bad. */
331 if (mtd_can_have_bb(d->mtd) && mtd_block_isbad(d->mtd, offset))
332 return MTDSWAP_SCANNED_BAD;
333
334 ops.ooblen = 2 * d->mtd->oobavail;
335 ops.oobbuf = d->oob_buf;
336 ops.ooboffs = 0;
337 ops.datbuf = NULL;
338 ops.mode = MTD_OPS_AUTO_OOB;
339
340 ret = mtdswap_read_oob(d, offset, &ops);
341
342 if (ret && !mtd_is_bitflip(ret))
343 return ret;
344
345 data = (struct mtdswap_oobdata *)d->oob_buf;
346 data2 = (struct mtdswap_oobdata *)
347 (d->oob_buf + d->mtd->oobavail);
348
349 if (le16_to_cpu(data->magic) == MTDSWAP_MAGIC_CLEAN) {
350 eb->erase_count = le32_to_cpu(data->count);
351 if (mtd_is_bitflip(ret))
352 ret = MTDSWAP_SCANNED_BITFLIP;
353 else {
354 if (le16_to_cpu(data2->magic) == MTDSWAP_MAGIC_DIRTY)
355 ret = MTDSWAP_SCANNED_DIRTY;
356 else
357 ret = MTDSWAP_SCANNED_CLEAN;
358 }
359 } else {
360 eb->flags |= EBLOCK_NOMAGIC;
361 ret = MTDSWAP_SCANNED_DIRTY;
362 }
363
364 return ret;
365 }
366
mtdswap_write_marker(struct mtdswap_dev * d,struct swap_eb * eb,u16 marker)367 static int mtdswap_write_marker(struct mtdswap_dev *d, struct swap_eb *eb,
368 u16 marker)
369 {
370 struct mtdswap_oobdata n;
371 int ret;
372 loff_t offset;
373 struct mtd_oob_ops ops;
374
375 ops.ooboffs = 0;
376 ops.oobbuf = (uint8_t *)&n;
377 ops.mode = MTD_OPS_AUTO_OOB;
378 ops.datbuf = NULL;
379
380 if (marker == MTDSWAP_TYPE_CLEAN) {
381 n.magic = cpu_to_le16(MTDSWAP_MAGIC_CLEAN);
382 n.count = cpu_to_le32(eb->erase_count);
383 ops.ooblen = MTDSWAP_OOBSIZE;
384 offset = mtdswap_eb_offset(d, eb);
385 } else {
386 n.magic = cpu_to_le16(MTDSWAP_MAGIC_DIRTY);
387 ops.ooblen = sizeof(n.magic);
388 offset = mtdswap_eb_offset(d, eb) + d->mtd->writesize;
389 }
390
391 ret = mtd_write_oob(d->mtd, offset, &ops);
392
393 if (ret) {
394 dev_warn(d->dev, "Write OOB failed for block at %08llx "
395 "error %d\n", offset, ret);
396 if (ret == -EIO || mtd_is_eccerr(ret))
397 mtdswap_handle_write_error(d, eb);
398 return ret;
399 }
400
401 if (ops.oobretlen != ops.ooblen) {
402 dev_warn(d->dev, "Short OOB write for block at %08llx: "
403 "%zd not %zd\n",
404 offset, ops.oobretlen, ops.ooblen);
405 return ret;
406 }
407
408 return 0;
409 }
410
411 /*
412 * Are there any erase blocks without MAGIC_CLEAN header, presumably
413 * because power was cut off after erase but before header write? We
414 * need to guestimate the erase count.
415 */
mtdswap_check_counts(struct mtdswap_dev * d)416 static void mtdswap_check_counts(struct mtdswap_dev *d)
417 {
418 struct rb_root hist_root = RB_ROOT;
419 struct rb_node *medrb;
420 struct swap_eb *eb;
421 unsigned int i, cnt, median;
422
423 cnt = 0;
424 for (i = 0; i < d->eblks; i++) {
425 eb = d->eb_data + i;
426
427 if (eb->flags & (EBLOCK_NOMAGIC | EBLOCK_BAD | EBLOCK_READERR))
428 continue;
429
430 __mtdswap_rb_add(&hist_root, eb);
431 cnt++;
432 }
433
434 if (cnt == 0)
435 return;
436
437 medrb = mtdswap_rb_index(&hist_root, cnt / 2);
438 median = rb_entry(medrb, struct swap_eb, rb)->erase_count;
439
440 d->max_erase_count = MTDSWAP_ECNT_MAX(&hist_root);
441
442 for (i = 0; i < d->eblks; i++) {
443 eb = d->eb_data + i;
444
445 if (eb->flags & (EBLOCK_NOMAGIC | EBLOCK_READERR))
446 eb->erase_count = median;
447
448 if (eb->flags & (EBLOCK_NOMAGIC | EBLOCK_BAD | EBLOCK_READERR))
449 continue;
450
451 rb_erase(&eb->rb, &hist_root);
452 }
453 }
454
mtdswap_scan_eblks(struct mtdswap_dev * d)455 static void mtdswap_scan_eblks(struct mtdswap_dev *d)
456 {
457 int status;
458 unsigned int i, idx;
459 struct swap_eb *eb;
460
461 for (i = 0; i < d->eblks; i++) {
462 eb = d->eb_data + i;
463
464 status = mtdswap_read_markers(d, eb);
465 if (status < 0)
466 eb->flags |= EBLOCK_READERR;
467 else if (status == MTDSWAP_SCANNED_BAD) {
468 eb->flags |= EBLOCK_BAD;
469 continue;
470 }
471
472 switch (status) {
473 case MTDSWAP_SCANNED_CLEAN:
474 idx = MTDSWAP_CLEAN;
475 break;
476 case MTDSWAP_SCANNED_DIRTY:
477 case MTDSWAP_SCANNED_BITFLIP:
478 idx = MTDSWAP_DIRTY;
479 break;
480 default:
481 idx = MTDSWAP_FAILING;
482 }
483
484 eb->flags |= (idx << EBLOCK_IDX_SHIFT);
485 }
486
487 mtdswap_check_counts(d);
488
489 for (i = 0; i < d->eblks; i++) {
490 eb = d->eb_data + i;
491
492 if (eb->flags & EBLOCK_BAD)
493 continue;
494
495 idx = eb->flags >> EBLOCK_IDX_SHIFT;
496 mtdswap_rb_add(d, eb, idx);
497 }
498 }
499
500 /*
501 * Place eblk into a tree corresponding to its number of active blocks
502 * it contains.
503 */
mtdswap_store_eb(struct mtdswap_dev * d,struct swap_eb * eb)504 static void mtdswap_store_eb(struct mtdswap_dev *d, struct swap_eb *eb)
505 {
506 unsigned int weight = eb->active_count;
507 unsigned int maxweight = d->pages_per_eblk;
508
509 if (eb == d->curr_write)
510 return;
511
512 if (eb->flags & EBLOCK_BITFLIP)
513 mtdswap_rb_add(d, eb, MTDSWAP_BITFLIP);
514 else if (eb->flags & (EBLOCK_READERR | EBLOCK_FAILED))
515 mtdswap_rb_add(d, eb, MTDSWAP_FAILING);
516 if (weight == maxweight)
517 mtdswap_rb_add(d, eb, MTDSWAP_USED);
518 else if (weight == 0)
519 mtdswap_rb_add(d, eb, MTDSWAP_DIRTY);
520 else if (weight > (maxweight/2))
521 mtdswap_rb_add(d, eb, MTDSWAP_LOWFRAG);
522 else
523 mtdswap_rb_add(d, eb, MTDSWAP_HIFRAG);
524 }
525
mtdswap_erase_block(struct mtdswap_dev * d,struct swap_eb * eb)526 static int mtdswap_erase_block(struct mtdswap_dev *d, struct swap_eb *eb)
527 {
528 struct mtd_info *mtd = d->mtd;
529 struct erase_info erase;
530 unsigned int retries = 0;
531 int ret;
532
533 eb->erase_count++;
534 if (eb->erase_count > d->max_erase_count)
535 d->max_erase_count = eb->erase_count;
536
537 retry:
538 memset(&erase, 0, sizeof(struct erase_info));
539 erase.addr = mtdswap_eb_offset(d, eb);
540 erase.len = mtd->erasesize;
541
542 ret = mtd_erase(mtd, &erase);
543 if (ret) {
544 if (retries++ < MTDSWAP_ERASE_RETRIES) {
545 dev_warn(d->dev,
546 "erase of erase block %#llx on %s failed",
547 erase.addr, mtd->name);
548 yield();
549 goto retry;
550 }
551
552 dev_err(d->dev, "Cannot erase erase block %#llx on %s\n",
553 erase.addr, mtd->name);
554
555 mtdswap_handle_badblock(d, eb);
556 return -EIO;
557 }
558
559 return 0;
560 }
561
mtdswap_map_free_block(struct mtdswap_dev * d,unsigned int page,unsigned int * block)562 static int mtdswap_map_free_block(struct mtdswap_dev *d, unsigned int page,
563 unsigned int *block)
564 {
565 int ret;
566 struct swap_eb *old_eb = d->curr_write;
567 struct rb_root *clean_root;
568 struct swap_eb *eb;
569
570 if (old_eb == NULL || d->curr_write_pos >= d->pages_per_eblk) {
571 do {
572 if (TREE_EMPTY(d, CLEAN))
573 return -ENOSPC;
574
575 clean_root = TREE_ROOT(d, CLEAN);
576 eb = rb_entry(rb_first(clean_root), struct swap_eb, rb);
577 rb_erase(&eb->rb, clean_root);
578 eb->root = NULL;
579 TREE_COUNT(d, CLEAN)--;
580
581 ret = mtdswap_write_marker(d, eb, MTDSWAP_TYPE_DIRTY);
582 } while (ret == -EIO || mtd_is_eccerr(ret));
583
584 if (ret)
585 return ret;
586
587 d->curr_write_pos = 0;
588 d->curr_write = eb;
589 if (old_eb)
590 mtdswap_store_eb(d, old_eb);
591 }
592
593 *block = (d->curr_write - d->eb_data) * d->pages_per_eblk +
594 d->curr_write_pos;
595
596 d->curr_write->active_count++;
597 d->revmap[*block] = page;
598 d->curr_write_pos++;
599
600 return 0;
601 }
602
mtdswap_free_page_cnt(struct mtdswap_dev * d)603 static unsigned int mtdswap_free_page_cnt(struct mtdswap_dev *d)
604 {
605 return TREE_COUNT(d, CLEAN) * d->pages_per_eblk +
606 d->pages_per_eblk - d->curr_write_pos;
607 }
608
mtdswap_enough_free_pages(struct mtdswap_dev * d)609 static unsigned int mtdswap_enough_free_pages(struct mtdswap_dev *d)
610 {
611 return mtdswap_free_page_cnt(d) > d->pages_per_eblk;
612 }
613
mtdswap_write_block(struct mtdswap_dev * d,char * buf,unsigned int page,unsigned int * bp,int gc_context)614 static int mtdswap_write_block(struct mtdswap_dev *d, char *buf,
615 unsigned int page, unsigned int *bp, int gc_context)
616 {
617 struct mtd_info *mtd = d->mtd;
618 struct swap_eb *eb;
619 size_t retlen;
620 loff_t writepos;
621 int ret;
622
623 retry:
624 if (!gc_context)
625 while (!mtdswap_enough_free_pages(d))
626 if (mtdswap_gc(d, 0) > 0)
627 return -ENOSPC;
628
629 ret = mtdswap_map_free_block(d, page, bp);
630 eb = d->eb_data + (*bp / d->pages_per_eblk);
631
632 if (ret == -EIO || mtd_is_eccerr(ret)) {
633 d->curr_write = NULL;
634 eb->active_count--;
635 d->revmap[*bp] = PAGE_UNDEF;
636 goto retry;
637 }
638
639 if (ret < 0)
640 return ret;
641
642 writepos = (loff_t)*bp << PAGE_SHIFT;
643 ret = mtd_write(mtd, writepos, PAGE_SIZE, &retlen, buf);
644 if (ret == -EIO || mtd_is_eccerr(ret)) {
645 d->curr_write_pos--;
646 eb->active_count--;
647 d->revmap[*bp] = PAGE_UNDEF;
648 mtdswap_handle_write_error(d, eb);
649 goto retry;
650 }
651
652 if (ret < 0) {
653 dev_err(d->dev, "Write to MTD device failed: %d (%zd written)",
654 ret, retlen);
655 goto err;
656 }
657
658 if (retlen != PAGE_SIZE) {
659 dev_err(d->dev, "Short write to MTD device: %zd written",
660 retlen);
661 ret = -EIO;
662 goto err;
663 }
664
665 return ret;
666
667 err:
668 d->curr_write_pos--;
669 eb->active_count--;
670 d->revmap[*bp] = PAGE_UNDEF;
671
672 return ret;
673 }
674
mtdswap_move_block(struct mtdswap_dev * d,unsigned int oldblock,unsigned int * newblock)675 static int mtdswap_move_block(struct mtdswap_dev *d, unsigned int oldblock,
676 unsigned int *newblock)
677 {
678 struct mtd_info *mtd = d->mtd;
679 struct swap_eb *eb, *oldeb;
680 int ret;
681 size_t retlen;
682 unsigned int page, retries;
683 loff_t readpos;
684
685 page = d->revmap[oldblock];
686 readpos = (loff_t) oldblock << PAGE_SHIFT;
687 retries = 0;
688
689 retry:
690 ret = mtd_read(mtd, readpos, PAGE_SIZE, &retlen, d->page_buf);
691
692 if (ret < 0 && !mtd_is_bitflip(ret)) {
693 oldeb = d->eb_data + oldblock / d->pages_per_eblk;
694 oldeb->flags |= EBLOCK_READERR;
695
696 dev_err(d->dev, "Read Error: %d (block %u)\n", ret,
697 oldblock);
698 retries++;
699 if (retries < MTDSWAP_IO_RETRIES)
700 goto retry;
701
702 goto read_error;
703 }
704
705 if (retlen != PAGE_SIZE) {
706 dev_err(d->dev, "Short read: %zd (block %u)\n", retlen,
707 oldblock);
708 ret = -EIO;
709 goto read_error;
710 }
711
712 ret = mtdswap_write_block(d, d->page_buf, page, newblock, 1);
713 if (ret < 0) {
714 d->page_data[page] = BLOCK_ERROR;
715 dev_err(d->dev, "Write error: %d\n", ret);
716 return ret;
717 }
718
719 eb = d->eb_data + *newblock / d->pages_per_eblk;
720 d->page_data[page] = *newblock;
721 d->revmap[oldblock] = PAGE_UNDEF;
722 eb = d->eb_data + oldblock / d->pages_per_eblk;
723 eb->active_count--;
724
725 return 0;
726
727 read_error:
728 d->page_data[page] = BLOCK_ERROR;
729 d->revmap[oldblock] = PAGE_UNDEF;
730 return ret;
731 }
732
mtdswap_gc_eblock(struct mtdswap_dev * d,struct swap_eb * eb)733 static int mtdswap_gc_eblock(struct mtdswap_dev *d, struct swap_eb *eb)
734 {
735 unsigned int i, block, eblk_base, newblock;
736 int ret, errcode;
737
738 errcode = 0;
739 eblk_base = (eb - d->eb_data) * d->pages_per_eblk;
740
741 for (i = 0; i < d->pages_per_eblk; i++) {
742 if (d->spare_eblks < MIN_SPARE_EBLOCKS)
743 return -ENOSPC;
744
745 block = eblk_base + i;
746 if (d->revmap[block] == PAGE_UNDEF)
747 continue;
748
749 ret = mtdswap_move_block(d, block, &newblock);
750 if (ret < 0 && !errcode)
751 errcode = ret;
752 }
753
754 return errcode;
755 }
756
__mtdswap_choose_gc_tree(struct mtdswap_dev * d)757 static int __mtdswap_choose_gc_tree(struct mtdswap_dev *d)
758 {
759 int idx, stopat;
760
761 if (TREE_COUNT(d, CLEAN) < LOW_FRAG_GC_THRESHOLD)
762 stopat = MTDSWAP_LOWFRAG;
763 else
764 stopat = MTDSWAP_HIFRAG;
765
766 for (idx = MTDSWAP_BITFLIP; idx >= stopat; idx--)
767 if (d->trees[idx].root.rb_node != NULL)
768 return idx;
769
770 return -1;
771 }
772
mtdswap_wlfreq(unsigned int maxdiff)773 static int mtdswap_wlfreq(unsigned int maxdiff)
774 {
775 unsigned int h, x, y, dist, base;
776
777 /*
778 * Calculate linear ramp down from f1 to f2 when maxdiff goes from
779 * MAX_ERASE_DIFF to MAX_ERASE_DIFF + COLLECT_NONDIRTY_BASE. Similar
780 * to triangle with height f1 - f1 and width COLLECT_NONDIRTY_BASE.
781 */
782
783 dist = maxdiff - MAX_ERASE_DIFF;
784 if (dist > COLLECT_NONDIRTY_BASE)
785 dist = COLLECT_NONDIRTY_BASE;
786
787 /*
788 * Modelling the slop as right angular triangle with base
789 * COLLECT_NONDIRTY_BASE and height freq1 - freq2. The ratio y/x is
790 * equal to the ratio h/base.
791 */
792 h = COLLECT_NONDIRTY_FREQ1 - COLLECT_NONDIRTY_FREQ2;
793 base = COLLECT_NONDIRTY_BASE;
794
795 x = dist - base;
796 y = (x * h + base / 2) / base;
797
798 return COLLECT_NONDIRTY_FREQ2 + y;
799 }
800
mtdswap_choose_wl_tree(struct mtdswap_dev * d)801 static int mtdswap_choose_wl_tree(struct mtdswap_dev *d)
802 {
803 static unsigned int pick_cnt;
804 unsigned int i, idx = -1, wear, max;
805 struct rb_root *root;
806
807 max = 0;
808 for (i = 0; i <= MTDSWAP_DIRTY; i++) {
809 root = &d->trees[i].root;
810 if (root->rb_node == NULL)
811 continue;
812
813 wear = d->max_erase_count - MTDSWAP_ECNT_MIN(root);
814 if (wear > max) {
815 max = wear;
816 idx = i;
817 }
818 }
819
820 if (max > MAX_ERASE_DIFF && pick_cnt >= mtdswap_wlfreq(max) - 1) {
821 pick_cnt = 0;
822 return idx;
823 }
824
825 pick_cnt++;
826 return -1;
827 }
828
mtdswap_choose_gc_tree(struct mtdswap_dev * d,unsigned int background)829 static int mtdswap_choose_gc_tree(struct mtdswap_dev *d,
830 unsigned int background)
831 {
832 int idx;
833
834 if (TREE_NONEMPTY(d, FAILING) &&
835 (background || (TREE_EMPTY(d, CLEAN) && TREE_EMPTY(d, DIRTY))))
836 return MTDSWAP_FAILING;
837
838 idx = mtdswap_choose_wl_tree(d);
839 if (idx >= MTDSWAP_CLEAN)
840 return idx;
841
842 return __mtdswap_choose_gc_tree(d);
843 }
844
mtdswap_pick_gc_eblk(struct mtdswap_dev * d,unsigned int background)845 static struct swap_eb *mtdswap_pick_gc_eblk(struct mtdswap_dev *d,
846 unsigned int background)
847 {
848 struct rb_root *rp = NULL;
849 struct swap_eb *eb = NULL;
850 int idx;
851
852 if (background && TREE_COUNT(d, CLEAN) > CLEAN_BLOCK_THRESHOLD &&
853 TREE_EMPTY(d, DIRTY) && TREE_EMPTY(d, FAILING))
854 return NULL;
855
856 idx = mtdswap_choose_gc_tree(d, background);
857 if (idx < 0)
858 return NULL;
859
860 rp = &d->trees[idx].root;
861 eb = rb_entry(rb_first(rp), struct swap_eb, rb);
862
863 rb_erase(&eb->rb, rp);
864 eb->root = NULL;
865 d->trees[idx].count--;
866 return eb;
867 }
868
mtdswap_test_patt(unsigned int i)869 static unsigned int mtdswap_test_patt(unsigned int i)
870 {
871 return i % 2 ? 0x55555555 : 0xAAAAAAAA;
872 }
873
mtdswap_eblk_passes(struct mtdswap_dev * d,struct swap_eb * eb)874 static unsigned int mtdswap_eblk_passes(struct mtdswap_dev *d,
875 struct swap_eb *eb)
876 {
877 struct mtd_info *mtd = d->mtd;
878 unsigned int test, i, j, patt, mtd_pages;
879 loff_t base, pos;
880 unsigned int *p1 = (unsigned int *)d->page_buf;
881 unsigned char *p2 = (unsigned char *)d->oob_buf;
882 struct mtd_oob_ops ops;
883 int ret;
884
885 ops.mode = MTD_OPS_AUTO_OOB;
886 ops.len = mtd->writesize;
887 ops.ooblen = mtd->oobavail;
888 ops.ooboffs = 0;
889 ops.datbuf = d->page_buf;
890 ops.oobbuf = d->oob_buf;
891 base = mtdswap_eb_offset(d, eb);
892 mtd_pages = d->pages_per_eblk * PAGE_SIZE / mtd->writesize;
893
894 for (test = 0; test < 2; test++) {
895 pos = base;
896 for (i = 0; i < mtd_pages; i++) {
897 patt = mtdswap_test_patt(test + i);
898 memset(d->page_buf, patt, mtd->writesize);
899 memset(d->oob_buf, patt, mtd->oobavail);
900 ret = mtd_write_oob(mtd, pos, &ops);
901 if (ret)
902 goto error;
903
904 pos += mtd->writesize;
905 }
906
907 pos = base;
908 for (i = 0; i < mtd_pages; i++) {
909 ret = mtd_read_oob(mtd, pos, &ops);
910 if (ret)
911 goto error;
912
913 patt = mtdswap_test_patt(test + i);
914 for (j = 0; j < mtd->writesize/sizeof(int); j++)
915 if (p1[j] != patt)
916 goto error;
917
918 for (j = 0; j < mtd->oobavail; j++)
919 if (p2[j] != (unsigned char)patt)
920 goto error;
921
922 pos += mtd->writesize;
923 }
924
925 ret = mtdswap_erase_block(d, eb);
926 if (ret)
927 goto error;
928 }
929
930 eb->flags &= ~EBLOCK_READERR;
931 return 1;
932
933 error:
934 mtdswap_handle_badblock(d, eb);
935 return 0;
936 }
937
mtdswap_gc(struct mtdswap_dev * d,unsigned int background)938 static int mtdswap_gc(struct mtdswap_dev *d, unsigned int background)
939 {
940 struct swap_eb *eb;
941 int ret;
942
943 if (d->spare_eblks < MIN_SPARE_EBLOCKS)
944 return 1;
945
946 eb = mtdswap_pick_gc_eblk(d, background);
947 if (!eb)
948 return 1;
949
950 ret = mtdswap_gc_eblock(d, eb);
951 if (ret == -ENOSPC)
952 return 1;
953
954 if (eb->flags & EBLOCK_FAILED) {
955 mtdswap_handle_badblock(d, eb);
956 return 0;
957 }
958
959 eb->flags &= ~EBLOCK_BITFLIP;
960 ret = mtdswap_erase_block(d, eb);
961 if ((eb->flags & EBLOCK_READERR) &&
962 (ret || !mtdswap_eblk_passes(d, eb)))
963 return 0;
964
965 if (ret == 0)
966 ret = mtdswap_write_marker(d, eb, MTDSWAP_TYPE_CLEAN);
967
968 if (ret == 0)
969 mtdswap_rb_add(d, eb, MTDSWAP_CLEAN);
970 else if (ret != -EIO && !mtd_is_eccerr(ret))
971 mtdswap_rb_add(d, eb, MTDSWAP_DIRTY);
972
973 return 0;
974 }
975
mtdswap_background(struct mtd_blktrans_dev * dev)976 static void mtdswap_background(struct mtd_blktrans_dev *dev)
977 {
978 struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
979 int ret;
980
981 while (1) {
982 ret = mtdswap_gc(d, 1);
983 if (ret || mtd_blktrans_cease_background(dev))
984 return;
985 }
986 }
987
mtdswap_cleanup(struct mtdswap_dev * d)988 static void mtdswap_cleanup(struct mtdswap_dev *d)
989 {
990 vfree(d->eb_data);
991 vfree(d->revmap);
992 vfree(d->page_data);
993 kfree(d->oob_buf);
994 kfree(d->page_buf);
995 }
996
mtdswap_flush(struct mtd_blktrans_dev * dev)997 static int mtdswap_flush(struct mtd_blktrans_dev *dev)
998 {
999 struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1000
1001 mtd_sync(d->mtd);
1002 return 0;
1003 }
1004
mtdswap_badblocks(struct mtd_info * mtd,uint64_t size)1005 static unsigned int mtdswap_badblocks(struct mtd_info *mtd, uint64_t size)
1006 {
1007 loff_t offset;
1008 unsigned int badcnt;
1009
1010 badcnt = 0;
1011
1012 if (mtd_can_have_bb(mtd))
1013 for (offset = 0; offset < size; offset += mtd->erasesize)
1014 if (mtd_block_isbad(mtd, offset))
1015 badcnt++;
1016
1017 return badcnt;
1018 }
1019
mtdswap_writesect(struct mtd_blktrans_dev * dev,unsigned long page,char * buf)1020 static int mtdswap_writesect(struct mtd_blktrans_dev *dev,
1021 unsigned long page, char *buf)
1022 {
1023 struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1024 unsigned int newblock, mapped;
1025 struct swap_eb *eb;
1026 int ret;
1027
1028 d->sect_write_count++;
1029
1030 if (d->spare_eblks < MIN_SPARE_EBLOCKS)
1031 return -ENOSPC;
1032
1033 if (header) {
1034 /* Ignore writes to the header page */
1035 if (unlikely(page == 0))
1036 return 0;
1037
1038 page--;
1039 }
1040
1041 mapped = d->page_data[page];
1042 if (mapped <= BLOCK_MAX) {
1043 eb = d->eb_data + (mapped / d->pages_per_eblk);
1044 eb->active_count--;
1045 mtdswap_store_eb(d, eb);
1046 d->page_data[page] = BLOCK_UNDEF;
1047 d->revmap[mapped] = PAGE_UNDEF;
1048 }
1049
1050 ret = mtdswap_write_block(d, buf, page, &newblock, 0);
1051 d->mtd_write_count++;
1052
1053 if (ret < 0)
1054 return ret;
1055
1056 eb = d->eb_data + (newblock / d->pages_per_eblk);
1057 d->page_data[page] = newblock;
1058
1059 return 0;
1060 }
1061
1062 /* Provide a dummy swap header for the kernel */
mtdswap_auto_header(struct mtdswap_dev * d,char * buf)1063 static int mtdswap_auto_header(struct mtdswap_dev *d, char *buf)
1064 {
1065 union swap_header *hd = (union swap_header *)(buf);
1066
1067 memset(buf, 0, PAGE_SIZE - 10);
1068
1069 hd->info.version = 1;
1070 hd->info.last_page = d->mbd_dev->size - 1;
1071 hd->info.nr_badpages = 0;
1072
1073 memcpy(buf + PAGE_SIZE - 10, "SWAPSPACE2", 10);
1074
1075 return 0;
1076 }
1077
mtdswap_readsect(struct mtd_blktrans_dev * dev,unsigned long page,char * buf)1078 static int mtdswap_readsect(struct mtd_blktrans_dev *dev,
1079 unsigned long page, char *buf)
1080 {
1081 struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1082 struct mtd_info *mtd = d->mtd;
1083 unsigned int realblock, retries;
1084 loff_t readpos;
1085 struct swap_eb *eb;
1086 size_t retlen;
1087 int ret;
1088
1089 d->sect_read_count++;
1090
1091 if (header) {
1092 if (unlikely(page == 0))
1093 return mtdswap_auto_header(d, buf);
1094
1095 page--;
1096 }
1097
1098 realblock = d->page_data[page];
1099 if (realblock > BLOCK_MAX) {
1100 memset(buf, 0x0, PAGE_SIZE);
1101 if (realblock == BLOCK_UNDEF)
1102 return 0;
1103 else
1104 return -EIO;
1105 }
1106
1107 eb = d->eb_data + (realblock / d->pages_per_eblk);
1108 BUG_ON(d->revmap[realblock] == PAGE_UNDEF);
1109
1110 readpos = (loff_t)realblock << PAGE_SHIFT;
1111 retries = 0;
1112
1113 retry:
1114 ret = mtd_read(mtd, readpos, PAGE_SIZE, &retlen, buf);
1115
1116 d->mtd_read_count++;
1117 if (mtd_is_bitflip(ret)) {
1118 eb->flags |= EBLOCK_BITFLIP;
1119 mtdswap_rb_add(d, eb, MTDSWAP_BITFLIP);
1120 ret = 0;
1121 }
1122
1123 if (ret < 0) {
1124 dev_err(d->dev, "Read error %d\n", ret);
1125 eb->flags |= EBLOCK_READERR;
1126 mtdswap_rb_add(d, eb, MTDSWAP_FAILING);
1127 retries++;
1128 if (retries < MTDSWAP_IO_RETRIES)
1129 goto retry;
1130
1131 return ret;
1132 }
1133
1134 if (retlen != PAGE_SIZE) {
1135 dev_err(d->dev, "Short read %zd\n", retlen);
1136 return -EIO;
1137 }
1138
1139 return 0;
1140 }
1141
mtdswap_discard(struct mtd_blktrans_dev * dev,unsigned long first,unsigned nr_pages)1142 static int mtdswap_discard(struct mtd_blktrans_dev *dev, unsigned long first,
1143 unsigned nr_pages)
1144 {
1145 struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1146 unsigned long page;
1147 struct swap_eb *eb;
1148 unsigned int mapped;
1149
1150 d->discard_count++;
1151
1152 for (page = first; page < first + nr_pages; page++) {
1153 mapped = d->page_data[page];
1154 if (mapped <= BLOCK_MAX) {
1155 eb = d->eb_data + (mapped / d->pages_per_eblk);
1156 eb->active_count--;
1157 mtdswap_store_eb(d, eb);
1158 d->page_data[page] = BLOCK_UNDEF;
1159 d->revmap[mapped] = PAGE_UNDEF;
1160 d->discard_page_count++;
1161 } else if (mapped == BLOCK_ERROR) {
1162 d->page_data[page] = BLOCK_UNDEF;
1163 d->discard_page_count++;
1164 }
1165 }
1166
1167 return 0;
1168 }
1169
mtdswap_show(struct seq_file * s,void * data)1170 static int mtdswap_show(struct seq_file *s, void *data)
1171 {
1172 struct mtdswap_dev *d = (struct mtdswap_dev *) s->private;
1173 unsigned long sum;
1174 unsigned int count[MTDSWAP_TREE_CNT];
1175 unsigned int min[MTDSWAP_TREE_CNT];
1176 unsigned int max[MTDSWAP_TREE_CNT];
1177 unsigned int i, cw = 0, cwp = 0, cwecount = 0, bb_cnt, mapped, pages;
1178 uint64_t use_size;
1179 static const char * const name[] = {
1180 "clean", "used", "low", "high", "dirty", "bitflip", "failing"
1181 };
1182
1183 mutex_lock(&d->mbd_dev->lock);
1184
1185 for (i = 0; i < MTDSWAP_TREE_CNT; i++) {
1186 struct rb_root *root = &d->trees[i].root;
1187
1188 if (root->rb_node) {
1189 count[i] = d->trees[i].count;
1190 min[i] = MTDSWAP_ECNT_MIN(root);
1191 max[i] = MTDSWAP_ECNT_MAX(root);
1192 } else
1193 count[i] = 0;
1194 }
1195
1196 if (d->curr_write) {
1197 cw = 1;
1198 cwp = d->curr_write_pos;
1199 cwecount = d->curr_write->erase_count;
1200 }
1201
1202 sum = 0;
1203 for (i = 0; i < d->eblks; i++)
1204 sum += d->eb_data[i].erase_count;
1205
1206 use_size = (uint64_t)d->eblks * d->mtd->erasesize;
1207 bb_cnt = mtdswap_badblocks(d->mtd, use_size);
1208
1209 mapped = 0;
1210 pages = d->mbd_dev->size;
1211 for (i = 0; i < pages; i++)
1212 if (d->page_data[i] != BLOCK_UNDEF)
1213 mapped++;
1214
1215 mutex_unlock(&d->mbd_dev->lock);
1216
1217 for (i = 0; i < MTDSWAP_TREE_CNT; i++) {
1218 if (!count[i])
1219 continue;
1220
1221 if (min[i] != max[i])
1222 seq_printf(s, "%s:\t%5d erase blocks, erased min %d, "
1223 "max %d times\n",
1224 name[i], count[i], min[i], max[i]);
1225 else
1226 seq_printf(s, "%s:\t%5d erase blocks, all erased %d "
1227 "times\n", name[i], count[i], min[i]);
1228 }
1229
1230 if (bb_cnt)
1231 seq_printf(s, "bad:\t%5u erase blocks\n", bb_cnt);
1232
1233 if (cw)
1234 seq_printf(s, "current erase block: %u pages used, %u free, "
1235 "erased %u times\n",
1236 cwp, d->pages_per_eblk - cwp, cwecount);
1237
1238 seq_printf(s, "total erasures: %lu\n", sum);
1239
1240 seq_puts(s, "\n");
1241
1242 seq_printf(s, "mtdswap_readsect count: %llu\n", d->sect_read_count);
1243 seq_printf(s, "mtdswap_writesect count: %llu\n", d->sect_write_count);
1244 seq_printf(s, "mtdswap_discard count: %llu\n", d->discard_count);
1245 seq_printf(s, "mtd read count: %llu\n", d->mtd_read_count);
1246 seq_printf(s, "mtd write count: %llu\n", d->mtd_write_count);
1247 seq_printf(s, "discarded pages count: %llu\n", d->discard_page_count);
1248
1249 seq_puts(s, "\n");
1250 seq_printf(s, "total pages: %u\n", pages);
1251 seq_printf(s, "pages mapped: %u\n", mapped);
1252
1253 return 0;
1254 }
1255 DEFINE_SHOW_ATTRIBUTE(mtdswap);
1256
mtdswap_add_debugfs(struct mtdswap_dev * d)1257 static int mtdswap_add_debugfs(struct mtdswap_dev *d)
1258 {
1259 struct dentry *root = d->mtd->dbg.dfs_dir;
1260 struct dentry *dent;
1261
1262 if (!IS_ENABLED(CONFIG_DEBUG_FS))
1263 return 0;
1264
1265 if (IS_ERR_OR_NULL(root))
1266 return -1;
1267
1268 dent = debugfs_create_file("mtdswap_stats", S_IRUSR, root, d,
1269 &mtdswap_fops);
1270 if (!dent) {
1271 dev_err(d->dev, "debugfs_create_file failed\n");
1272 return -1;
1273 }
1274
1275 return 0;
1276 }
1277
mtdswap_init(struct mtdswap_dev * d,unsigned int eblocks,unsigned int spare_cnt)1278 static int mtdswap_init(struct mtdswap_dev *d, unsigned int eblocks,
1279 unsigned int spare_cnt)
1280 {
1281 struct mtd_info *mtd = d->mbd_dev->mtd;
1282 unsigned int i, eblk_bytes, pages, blocks;
1283 int ret = -ENOMEM;
1284
1285 d->mtd = mtd;
1286 d->eblks = eblocks;
1287 d->spare_eblks = spare_cnt;
1288 d->pages_per_eblk = mtd->erasesize >> PAGE_SHIFT;
1289
1290 pages = d->mbd_dev->size;
1291 blocks = eblocks * d->pages_per_eblk;
1292
1293 for (i = 0; i < MTDSWAP_TREE_CNT; i++)
1294 d->trees[i].root = RB_ROOT;
1295
1296 d->page_data = vmalloc(array_size(pages, sizeof(int)));
1297 if (!d->page_data)
1298 goto page_data_fail;
1299
1300 d->revmap = vmalloc(array_size(blocks, sizeof(int)));
1301 if (!d->revmap)
1302 goto revmap_fail;
1303
1304 eblk_bytes = sizeof(struct swap_eb)*d->eblks;
1305 d->eb_data = vzalloc(eblk_bytes);
1306 if (!d->eb_data)
1307 goto eb_data_fail;
1308
1309 for (i = 0; i < pages; i++)
1310 d->page_data[i] = BLOCK_UNDEF;
1311
1312 for (i = 0; i < blocks; i++)
1313 d->revmap[i] = PAGE_UNDEF;
1314
1315 d->page_buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1316 if (!d->page_buf)
1317 goto page_buf_fail;
1318
1319 d->oob_buf = kmalloc_array(2, mtd->oobavail, GFP_KERNEL);
1320 if (!d->oob_buf)
1321 goto oob_buf_fail;
1322
1323 mtdswap_scan_eblks(d);
1324
1325 return 0;
1326
1327 oob_buf_fail:
1328 kfree(d->page_buf);
1329 page_buf_fail:
1330 vfree(d->eb_data);
1331 eb_data_fail:
1332 vfree(d->revmap);
1333 revmap_fail:
1334 vfree(d->page_data);
1335 page_data_fail:
1336 printk(KERN_ERR "%s: init failed (%d)\n", MTDSWAP_PREFIX, ret);
1337 return ret;
1338 }
1339
mtdswap_add_mtd(struct mtd_blktrans_ops * tr,struct mtd_info * mtd)1340 static void mtdswap_add_mtd(struct mtd_blktrans_ops *tr, struct mtd_info *mtd)
1341 {
1342 struct mtdswap_dev *d;
1343 struct mtd_blktrans_dev *mbd_dev;
1344 char *parts;
1345 char *this_opt;
1346 unsigned long part;
1347 unsigned int eblocks, eavailable, bad_blocks, spare_cnt;
1348 uint64_t swap_size, use_size, size_limit;
1349 int ret;
1350
1351 parts = &partitions[0];
1352 if (!*parts)
1353 return;
1354
1355 while ((this_opt = strsep(&parts, ",")) != NULL) {
1356 if (kstrtoul(this_opt, 0, &part) < 0)
1357 return;
1358
1359 if (mtd->index == part)
1360 break;
1361 }
1362
1363 if (mtd->index != part)
1364 return;
1365
1366 if (mtd->erasesize < PAGE_SIZE || mtd->erasesize % PAGE_SIZE) {
1367 printk(KERN_ERR "%s: Erase size %u not multiple of PAGE_SIZE "
1368 "%lu\n", MTDSWAP_PREFIX, mtd->erasesize, PAGE_SIZE);
1369 return;
1370 }
1371
1372 if (PAGE_SIZE % mtd->writesize || mtd->writesize > PAGE_SIZE) {
1373 printk(KERN_ERR "%s: PAGE_SIZE %lu not multiple of write size"
1374 " %u\n", MTDSWAP_PREFIX, PAGE_SIZE, mtd->writesize);
1375 return;
1376 }
1377
1378 if (!mtd->oobsize || mtd->oobavail < MTDSWAP_OOBSIZE) {
1379 printk(KERN_ERR "%s: Not enough free bytes in OOB, "
1380 "%d available, %zu needed.\n",
1381 MTDSWAP_PREFIX, mtd->oobavail, MTDSWAP_OOBSIZE);
1382 return;
1383 }
1384
1385 if (spare_eblocks > 100)
1386 spare_eblocks = 100;
1387
1388 use_size = mtd->size;
1389 size_limit = (uint64_t) BLOCK_MAX * PAGE_SIZE;
1390
1391 if (mtd->size > size_limit) {
1392 printk(KERN_WARNING "%s: Device too large. Limiting size to "
1393 "%llu bytes\n", MTDSWAP_PREFIX, size_limit);
1394 use_size = size_limit;
1395 }
1396
1397 eblocks = mtd_div_by_eb(use_size, mtd);
1398 use_size = (uint64_t)eblocks * mtd->erasesize;
1399 bad_blocks = mtdswap_badblocks(mtd, use_size);
1400 eavailable = eblocks - bad_blocks;
1401
1402 if (eavailable < MIN_ERASE_BLOCKS) {
1403 printk(KERN_ERR "%s: Not enough erase blocks. %u available, "
1404 "%d needed\n", MTDSWAP_PREFIX, eavailable,
1405 MIN_ERASE_BLOCKS);
1406 return;
1407 }
1408
1409 spare_cnt = div_u64((uint64_t)eavailable * spare_eblocks, 100);
1410
1411 if (spare_cnt < MIN_SPARE_EBLOCKS)
1412 spare_cnt = MIN_SPARE_EBLOCKS;
1413
1414 if (spare_cnt > eavailable - 1)
1415 spare_cnt = eavailable - 1;
1416
1417 swap_size = (uint64_t)(eavailable - spare_cnt) * mtd->erasesize +
1418 (header ? PAGE_SIZE : 0);
1419
1420 printk(KERN_INFO "%s: Enabling MTD swap on device %lu, size %llu KB, "
1421 "%u spare, %u bad blocks\n",
1422 MTDSWAP_PREFIX, part, swap_size / 1024, spare_cnt, bad_blocks);
1423
1424 d = kzalloc(sizeof(struct mtdswap_dev), GFP_KERNEL);
1425 if (!d)
1426 return;
1427
1428 mbd_dev = kzalloc(sizeof(struct mtd_blktrans_dev), GFP_KERNEL);
1429 if (!mbd_dev) {
1430 kfree(d);
1431 return;
1432 }
1433
1434 d->mbd_dev = mbd_dev;
1435 mbd_dev->priv = d;
1436
1437 mbd_dev->mtd = mtd;
1438 mbd_dev->devnum = mtd->index;
1439 mbd_dev->size = swap_size >> PAGE_SHIFT;
1440 mbd_dev->tr = tr;
1441
1442 if (!(mtd->flags & MTD_WRITEABLE))
1443 mbd_dev->readonly = 1;
1444
1445 if (mtdswap_init(d, eblocks, spare_cnt) < 0)
1446 goto init_failed;
1447
1448 if (add_mtd_blktrans_dev(mbd_dev) < 0)
1449 goto cleanup;
1450
1451 d->dev = disk_to_dev(mbd_dev->disk);
1452
1453 ret = mtdswap_add_debugfs(d);
1454 if (ret < 0)
1455 goto debugfs_failed;
1456
1457 return;
1458
1459 debugfs_failed:
1460 del_mtd_blktrans_dev(mbd_dev);
1461
1462 cleanup:
1463 mtdswap_cleanup(d);
1464
1465 init_failed:
1466 kfree(mbd_dev);
1467 kfree(d);
1468 }
1469
mtdswap_remove_dev(struct mtd_blktrans_dev * dev)1470 static void mtdswap_remove_dev(struct mtd_blktrans_dev *dev)
1471 {
1472 struct mtdswap_dev *d = MTDSWAP_MBD_TO_MTDSWAP(dev);
1473
1474 del_mtd_blktrans_dev(dev);
1475 mtdswap_cleanup(d);
1476 kfree(d);
1477 }
1478
1479 static struct mtd_blktrans_ops mtdswap_ops = {
1480 .name = "mtdswap",
1481 .major = 0,
1482 .part_bits = 0,
1483 .blksize = PAGE_SIZE,
1484 .flush = mtdswap_flush,
1485 .readsect = mtdswap_readsect,
1486 .writesect = mtdswap_writesect,
1487 .discard = mtdswap_discard,
1488 .background = mtdswap_background,
1489 .add_mtd = mtdswap_add_mtd,
1490 .remove_dev = mtdswap_remove_dev,
1491 .owner = THIS_MODULE,
1492 };
1493
mtdswap_modinit(void)1494 static int __init mtdswap_modinit(void)
1495 {
1496 return register_mtd_blktrans(&mtdswap_ops);
1497 }
1498
mtdswap_modexit(void)1499 static void __exit mtdswap_modexit(void)
1500 {
1501 deregister_mtd_blktrans(&mtdswap_ops);
1502 }
1503
1504 module_init(mtdswap_modinit);
1505 module_exit(mtdswap_modexit);
1506
1507
1508 MODULE_LICENSE("GPL");
1509 MODULE_AUTHOR("Jarkko Lavinen <jarkko.lavinen@nokia.com>");
1510 MODULE_DESCRIPTION("Block device access to an MTD suitable for using as "
1511 "swap space");
1512