[SCSI] megaraid_sas: Add 4k FastPath DIF support
[linux-2.6.git] / drivers / mtd / rfd_ftl.c
blob233b946e5d66d95a2cfbabb19985e5f8a75efb13
1 /*
2 * rfd_ftl.c -- resident flash disk (flash translation layer)
4 * Copyright © 2005 Sean Young <sean@mess.org>
6 * This type of flash translation layer (FTL) is used by the Embedded BIOS
7 * by General Software. It is known as the Resident Flash Disk (RFD), see:
9 * http://www.gensw.com/pages/prod/bios/rfd.htm
11 * based on ftl.c
14 #include <linux/hdreg.h>
15 #include <linux/init.h>
16 #include <linux/mtd/blktrans.h>
17 #include <linux/mtd/mtd.h>
18 #include <linux/vmalloc.h>
19 #include <linux/slab.h>
20 #include <linux/jiffies.h>
21 #include <linux/module.h>
23 #include <asm/types.h>
25 static int block_size = 0;
26 module_param(block_size, int, 0);
27 MODULE_PARM_DESC(block_size, "Block size to use by RFD, defaults to erase unit size");
29 #define PREFIX "rfd_ftl: "
31 /* This major has been assigned by device@lanana.org */
32 #ifndef RFD_FTL_MAJOR
33 #define RFD_FTL_MAJOR 256
34 #endif
36 /* Maximum number of partitions in an FTL region */
37 #define PART_BITS 4
39 /* An erase unit should start with this value */
40 #define RFD_MAGIC 0x9193
42 /* the second value is 0xffff or 0xffc8; function unknown */
44 /* the third value is always 0xffff, ignored */
46 /* next is an array of mapping for each corresponding sector */
47 #define HEADER_MAP_OFFSET 3
48 #define SECTOR_DELETED 0x0000
49 #define SECTOR_ZERO 0xfffe
50 #define SECTOR_FREE 0xffff
52 #define SECTOR_SIZE 512
54 #define SECTORS_PER_TRACK 63
56 struct block {
57 enum {
58 BLOCK_OK,
59 BLOCK_ERASING,
60 BLOCK_ERASED,
61 BLOCK_UNUSED,
62 BLOCK_FAILED
63 } state;
64 int free_sectors;
65 int used_sectors;
66 int erases;
67 u_long offset;
70 struct partition {
71 struct mtd_blktrans_dev mbd;
73 u_int block_size; /* size of erase unit */
74 u_int total_blocks; /* number of erase units */
75 u_int header_sectors_per_block; /* header sectors in erase unit */
76 u_int data_sectors_per_block; /* data sectors in erase unit */
77 u_int sector_count; /* sectors in translated disk */
78 u_int header_size; /* bytes in header sector */
79 int reserved_block; /* block next up for reclaim */
80 int current_block; /* block to write to */
81 u16 *header_cache; /* cached header */
83 int is_reclaiming;
84 int cylinders;
85 int errors;
86 u_long *sector_map;
87 struct block *blocks;
90 static int rfd_ftl_writesect(struct mtd_blktrans_dev *dev, u_long sector, char *buf);
92 static int build_block_map(struct partition *part, int block_no)
94 struct block *block = &part->blocks[block_no];
95 int i;
97 block->offset = part->block_size * block_no;
99 if (le16_to_cpu(part->header_cache[0]) != RFD_MAGIC) {
100 block->state = BLOCK_UNUSED;
101 return -ENOENT;
104 block->state = BLOCK_OK;
106 for (i=0; i<part->data_sectors_per_block; i++) {
107 u16 entry;
109 entry = le16_to_cpu(part->header_cache[HEADER_MAP_OFFSET + i]);
111 if (entry == SECTOR_DELETED)
112 continue;
114 if (entry == SECTOR_FREE) {
115 block->free_sectors++;
116 continue;
119 if (entry == SECTOR_ZERO)
120 entry = 0;
122 if (entry >= part->sector_count) {
123 printk(KERN_WARNING PREFIX
124 "'%s': unit #%d: entry %d corrupt, "
125 "sector %d out of range\n",
126 part->mbd.mtd->name, block_no, i, entry);
127 continue;
130 if (part->sector_map[entry] != -1) {
131 printk(KERN_WARNING PREFIX
132 "'%s': more than one entry for sector %d\n",
133 part->mbd.mtd->name, entry);
134 part->errors = 1;
135 continue;
138 part->sector_map[entry] = block->offset +
139 (i + part->header_sectors_per_block) * SECTOR_SIZE;
141 block->used_sectors++;
144 if (block->free_sectors == part->data_sectors_per_block)
145 part->reserved_block = block_no;
147 return 0;
150 static int scan_header(struct partition *part)
152 int sectors_per_block;
153 int i, rc = -ENOMEM;
154 int blocks_found;
155 size_t retlen;
157 sectors_per_block = part->block_size / SECTOR_SIZE;
158 part->total_blocks = (u32)part->mbd.mtd->size / part->block_size;
160 if (part->total_blocks < 2)
161 return -ENOENT;
163 /* each erase block has three bytes header, followed by the map */
164 part->header_sectors_per_block =
165 ((HEADER_MAP_OFFSET + sectors_per_block) *
166 sizeof(u16) + SECTOR_SIZE - 1) / SECTOR_SIZE;
168 part->data_sectors_per_block = sectors_per_block -
169 part->header_sectors_per_block;
171 part->header_size = (HEADER_MAP_OFFSET +
172 part->data_sectors_per_block) * sizeof(u16);
174 part->cylinders = (part->data_sectors_per_block *
175 (part->total_blocks - 1) - 1) / SECTORS_PER_TRACK;
177 part->sector_count = part->cylinders * SECTORS_PER_TRACK;
179 part->current_block = -1;
180 part->reserved_block = -1;
181 part->is_reclaiming = 0;
183 part->header_cache = kmalloc(part->header_size, GFP_KERNEL);
184 if (!part->header_cache)
185 goto err;
187 part->blocks = kcalloc(part->total_blocks, sizeof(struct block),
188 GFP_KERNEL);
189 if (!part->blocks)
190 goto err;
192 part->sector_map = vmalloc(part->sector_count * sizeof(u_long));
193 if (!part->sector_map) {
194 printk(KERN_ERR PREFIX "'%s': unable to allocate memory for "
195 "sector map", part->mbd.mtd->name);
196 goto err;
199 for (i=0; i<part->sector_count; i++)
200 part->sector_map[i] = -1;
202 for (i=0, blocks_found=0; i<part->total_blocks; i++) {
203 rc = mtd_read(part->mbd.mtd, i * part->block_size,
204 part->header_size, &retlen,
205 (u_char *)part->header_cache);
207 if (!rc && retlen != part->header_size)
208 rc = -EIO;
210 if (rc)
211 goto err;
213 if (!build_block_map(part, i))
214 blocks_found++;
217 if (blocks_found == 0) {
218 printk(KERN_NOTICE PREFIX "no RFD magic found in '%s'\n",
219 part->mbd.mtd->name);
220 rc = -ENOENT;
221 goto err;
224 if (part->reserved_block == -1) {
225 printk(KERN_WARNING PREFIX "'%s': no empty erase unit found\n",
226 part->mbd.mtd->name);
228 part->errors = 1;
231 return 0;
233 err:
234 vfree(part->sector_map);
235 kfree(part->header_cache);
236 kfree(part->blocks);
238 return rc;
241 static int rfd_ftl_readsect(struct mtd_blktrans_dev *dev, u_long sector, char *buf)
243 struct partition *part = (struct partition*)dev;
244 u_long addr;
245 size_t retlen;
246 int rc;
248 if (sector >= part->sector_count)
249 return -EIO;
251 addr = part->sector_map[sector];
252 if (addr != -1) {
253 rc = mtd_read(part->mbd.mtd, addr, SECTOR_SIZE, &retlen,
254 (u_char *)buf);
255 if (!rc && retlen != SECTOR_SIZE)
256 rc = -EIO;
258 if (rc) {
259 printk(KERN_WARNING PREFIX "error reading '%s' at "
260 "0x%lx\n", part->mbd.mtd->name, addr);
261 return rc;
263 } else
264 memset(buf, 0, SECTOR_SIZE);
266 return 0;
269 static void erase_callback(struct erase_info *erase)
271 struct partition *part;
272 u16 magic;
273 int i, rc;
274 size_t retlen;
276 part = (struct partition*)erase->priv;
278 i = (u32)erase->addr / part->block_size;
279 if (i >= part->total_blocks || part->blocks[i].offset != erase->addr ||
280 erase->addr > UINT_MAX) {
281 printk(KERN_ERR PREFIX "erase callback for unknown offset %llx "
282 "on '%s'\n", (unsigned long long)erase->addr, part->mbd.mtd->name);
283 return;
286 if (erase->state != MTD_ERASE_DONE) {
287 printk(KERN_WARNING PREFIX "erase failed at 0x%llx on '%s', "
288 "state %d\n", (unsigned long long)erase->addr,
289 part->mbd.mtd->name, erase->state);
291 part->blocks[i].state = BLOCK_FAILED;
292 part->blocks[i].free_sectors = 0;
293 part->blocks[i].used_sectors = 0;
295 kfree(erase);
297 return;
300 magic = cpu_to_le16(RFD_MAGIC);
302 part->blocks[i].state = BLOCK_ERASED;
303 part->blocks[i].free_sectors = part->data_sectors_per_block;
304 part->blocks[i].used_sectors = 0;
305 part->blocks[i].erases++;
307 rc = mtd_write(part->mbd.mtd, part->blocks[i].offset, sizeof(magic),
308 &retlen, (u_char *)&magic);
310 if (!rc && retlen != sizeof(magic))
311 rc = -EIO;
313 if (rc) {
314 printk(KERN_ERR PREFIX "'%s': unable to write RFD "
315 "header at 0x%lx\n",
316 part->mbd.mtd->name,
317 part->blocks[i].offset);
318 part->blocks[i].state = BLOCK_FAILED;
320 else
321 part->blocks[i].state = BLOCK_OK;
323 kfree(erase);
326 static int erase_block(struct partition *part, int block)
328 struct erase_info *erase;
329 int rc = -ENOMEM;
331 erase = kmalloc(sizeof(struct erase_info), GFP_KERNEL);
332 if (!erase)
333 goto err;
335 erase->mtd = part->mbd.mtd;
336 erase->callback = erase_callback;
337 erase->addr = part->blocks[block].offset;
338 erase->len = part->block_size;
339 erase->priv = (u_long)part;
341 part->blocks[block].state = BLOCK_ERASING;
342 part->blocks[block].free_sectors = 0;
344 rc = mtd_erase(part->mbd.mtd, erase);
346 if (rc) {
347 printk(KERN_ERR PREFIX "erase of region %llx,%llx on '%s' "
348 "failed\n", (unsigned long long)erase->addr,
349 (unsigned long long)erase->len, part->mbd.mtd->name);
350 kfree(erase);
353 err:
354 return rc;
357 static int move_block_contents(struct partition *part, int block_no, u_long *old_sector)
359 void *sector_data;
360 u16 *map;
361 size_t retlen;
362 int i, rc = -ENOMEM;
364 part->is_reclaiming = 1;
366 sector_data = kmalloc(SECTOR_SIZE, GFP_KERNEL);
367 if (!sector_data)
368 goto err3;
370 map = kmalloc(part->header_size, GFP_KERNEL);
371 if (!map)
372 goto err2;
374 rc = mtd_read(part->mbd.mtd, part->blocks[block_no].offset,
375 part->header_size, &retlen, (u_char *)map);
377 if (!rc && retlen != part->header_size)
378 rc = -EIO;
380 if (rc) {
381 printk(KERN_ERR PREFIX "error reading '%s' at "
382 "0x%lx\n", part->mbd.mtd->name,
383 part->blocks[block_no].offset);
385 goto err;
388 for (i=0; i<part->data_sectors_per_block; i++) {
389 u16 entry = le16_to_cpu(map[HEADER_MAP_OFFSET + i]);
390 u_long addr;
393 if (entry == SECTOR_FREE || entry == SECTOR_DELETED)
394 continue;
396 if (entry == SECTOR_ZERO)
397 entry = 0;
399 /* already warned about and ignored in build_block_map() */
400 if (entry >= part->sector_count)
401 continue;
403 addr = part->blocks[block_no].offset +
404 (i + part->header_sectors_per_block) * SECTOR_SIZE;
406 if (*old_sector == addr) {
407 *old_sector = -1;
408 if (!part->blocks[block_no].used_sectors--) {
409 rc = erase_block(part, block_no);
410 break;
412 continue;
414 rc = mtd_read(part->mbd.mtd, addr, SECTOR_SIZE, &retlen,
415 sector_data);
417 if (!rc && retlen != SECTOR_SIZE)
418 rc = -EIO;
420 if (rc) {
421 printk(KERN_ERR PREFIX "'%s': Unable to "
422 "read sector for relocation\n",
423 part->mbd.mtd->name);
425 goto err;
428 rc = rfd_ftl_writesect((struct mtd_blktrans_dev*)part,
429 entry, sector_data);
431 if (rc)
432 goto err;
435 err:
436 kfree(map);
437 err2:
438 kfree(sector_data);
439 err3:
440 part->is_reclaiming = 0;
442 return rc;
445 static int reclaim_block(struct partition *part, u_long *old_sector)
447 int block, best_block, score, old_sector_block;
448 int rc;
450 /* we have a race if sync doesn't exist */
451 mtd_sync(part->mbd.mtd);
453 score = 0x7fffffff; /* MAX_INT */
454 best_block = -1;
455 if (*old_sector != -1)
456 old_sector_block = *old_sector / part->block_size;
457 else
458 old_sector_block = -1;
460 for (block=0; block<part->total_blocks; block++) {
461 int this_score;
463 if (block == part->reserved_block)
464 continue;
467 * Postpone reclaiming if there is a free sector as
468 * more removed sectors is more efficient (have to move
469 * less).
471 if (part->blocks[block].free_sectors)
472 return 0;
474 this_score = part->blocks[block].used_sectors;
476 if (block == old_sector_block)
477 this_score--;
478 else {
479 /* no point in moving a full block */
480 if (part->blocks[block].used_sectors ==
481 part->data_sectors_per_block)
482 continue;
485 this_score += part->blocks[block].erases;
487 if (this_score < score) {
488 best_block = block;
489 score = this_score;
493 if (best_block == -1)
494 return -ENOSPC;
496 part->current_block = -1;
497 part->reserved_block = best_block;
499 pr_debug("reclaim_block: reclaiming block #%d with %d used "
500 "%d free sectors\n", best_block,
501 part->blocks[best_block].used_sectors,
502 part->blocks[best_block].free_sectors);
504 if (part->blocks[best_block].used_sectors)
505 rc = move_block_contents(part, best_block, old_sector);
506 else
507 rc = erase_block(part, best_block);
509 return rc;
513 * IMPROVE: It would be best to choose the block with the most deleted sectors,
514 * because if we fill that one up first it'll have the most chance of having
515 * the least live sectors at reclaim.
517 static int find_free_block(struct partition *part)
519 int block, stop;
521 block = part->current_block == -1 ?
522 jiffies % part->total_blocks : part->current_block;
523 stop = block;
525 do {
526 if (part->blocks[block].free_sectors &&
527 block != part->reserved_block)
528 return block;
530 if (part->blocks[block].state == BLOCK_UNUSED)
531 erase_block(part, block);
533 if (++block >= part->total_blocks)
534 block = 0;
536 } while (block != stop);
538 return -1;
541 static int find_writable_block(struct partition *part, u_long *old_sector)
543 int rc, block;
544 size_t retlen;
546 block = find_free_block(part);
548 if (block == -1) {
549 if (!part->is_reclaiming) {
550 rc = reclaim_block(part, old_sector);
551 if (rc)
552 goto err;
554 block = find_free_block(part);
557 if (block == -1) {
558 rc = -ENOSPC;
559 goto err;
563 rc = mtd_read(part->mbd.mtd, part->blocks[block].offset,
564 part->header_size, &retlen,
565 (u_char *)part->header_cache);
567 if (!rc && retlen != part->header_size)
568 rc = -EIO;
570 if (rc) {
571 printk(KERN_ERR PREFIX "'%s': unable to read header at "
572 "0x%lx\n", part->mbd.mtd->name,
573 part->blocks[block].offset);
574 goto err;
577 part->current_block = block;
579 err:
580 return rc;
583 static int mark_sector_deleted(struct partition *part, u_long old_addr)
585 int block, offset, rc;
586 u_long addr;
587 size_t retlen;
588 u16 del = cpu_to_le16(SECTOR_DELETED);
590 block = old_addr / part->block_size;
591 offset = (old_addr % part->block_size) / SECTOR_SIZE -
592 part->header_sectors_per_block;
594 addr = part->blocks[block].offset +
595 (HEADER_MAP_OFFSET + offset) * sizeof(u16);
596 rc = mtd_write(part->mbd.mtd, addr, sizeof(del), &retlen,
597 (u_char *)&del);
599 if (!rc && retlen != sizeof(del))
600 rc = -EIO;
602 if (rc) {
603 printk(KERN_ERR PREFIX "error writing '%s' at "
604 "0x%lx\n", part->mbd.mtd->name, addr);
605 if (rc)
606 goto err;
608 if (block == part->current_block)
609 part->header_cache[offset + HEADER_MAP_OFFSET] = del;
611 part->blocks[block].used_sectors--;
613 if (!part->blocks[block].used_sectors &&
614 !part->blocks[block].free_sectors)
615 rc = erase_block(part, block);
617 err:
618 return rc;
621 static int find_free_sector(const struct partition *part, const struct block *block)
623 int i, stop;
625 i = stop = part->data_sectors_per_block - block->free_sectors;
627 do {
628 if (le16_to_cpu(part->header_cache[HEADER_MAP_OFFSET + i])
629 == SECTOR_FREE)
630 return i;
632 if (++i == part->data_sectors_per_block)
633 i = 0;
635 while(i != stop);
637 return -1;
640 static int do_writesect(struct mtd_blktrans_dev *dev, u_long sector, char *buf, ulong *old_addr)
642 struct partition *part = (struct partition*)dev;
643 struct block *block;
644 u_long addr;
645 int i;
646 int rc;
647 size_t retlen;
648 u16 entry;
650 if (part->current_block == -1 ||
651 !part->blocks[part->current_block].free_sectors) {
653 rc = find_writable_block(part, old_addr);
654 if (rc)
655 goto err;
658 block = &part->blocks[part->current_block];
660 i = find_free_sector(part, block);
662 if (i < 0) {
663 rc = -ENOSPC;
664 goto err;
667 addr = (i + part->header_sectors_per_block) * SECTOR_SIZE +
668 block->offset;
669 rc = mtd_write(part->mbd.mtd, addr, SECTOR_SIZE, &retlen,
670 (u_char *)buf);
672 if (!rc && retlen != SECTOR_SIZE)
673 rc = -EIO;
675 if (rc) {
676 printk(KERN_ERR PREFIX "error writing '%s' at 0x%lx\n",
677 part->mbd.mtd->name, addr);
678 if (rc)
679 goto err;
682 part->sector_map[sector] = addr;
684 entry = cpu_to_le16(sector == 0 ? SECTOR_ZERO : sector);
686 part->header_cache[i + HEADER_MAP_OFFSET] = entry;
688 addr = block->offset + (HEADER_MAP_OFFSET + i) * sizeof(u16);
689 rc = mtd_write(part->mbd.mtd, addr, sizeof(entry), &retlen,
690 (u_char *)&entry);
692 if (!rc && retlen != sizeof(entry))
693 rc = -EIO;
695 if (rc) {
696 printk(KERN_ERR PREFIX "error writing '%s' at 0x%lx\n",
697 part->mbd.mtd->name, addr);
698 if (rc)
699 goto err;
701 block->used_sectors++;
702 block->free_sectors--;
704 err:
705 return rc;
708 static int rfd_ftl_writesect(struct mtd_blktrans_dev *dev, u_long sector, char *buf)
710 struct partition *part = (struct partition*)dev;
711 u_long old_addr;
712 int i;
713 int rc = 0;
715 pr_debug("rfd_ftl_writesect(sector=0x%lx)\n", sector);
717 if (part->reserved_block == -1) {
718 rc = -EACCES;
719 goto err;
722 if (sector >= part->sector_count) {
723 rc = -EIO;
724 goto err;
727 old_addr = part->sector_map[sector];
729 for (i=0; i<SECTOR_SIZE; i++) {
730 if (!buf[i])
731 continue;
733 rc = do_writesect(dev, sector, buf, &old_addr);
734 if (rc)
735 goto err;
736 break;
739 if (i == SECTOR_SIZE)
740 part->sector_map[sector] = -1;
742 if (old_addr != -1)
743 rc = mark_sector_deleted(part, old_addr);
745 err:
746 return rc;
749 static int rfd_ftl_getgeo(struct mtd_blktrans_dev *dev, struct hd_geometry *geo)
751 struct partition *part = (struct partition*)dev;
753 geo->heads = 1;
754 geo->sectors = SECTORS_PER_TRACK;
755 geo->cylinders = part->cylinders;
757 return 0;
760 static void rfd_ftl_add_mtd(struct mtd_blktrans_ops *tr, struct mtd_info *mtd)
762 struct partition *part;
764 if (mtd->type != MTD_NORFLASH || mtd->size > UINT_MAX)
765 return;
767 part = kzalloc(sizeof(struct partition), GFP_KERNEL);
768 if (!part)
769 return;
771 part->mbd.mtd = mtd;
773 if (block_size)
774 part->block_size = block_size;
775 else {
776 if (!mtd->erasesize) {
777 printk(KERN_WARNING PREFIX "please provide block_size");
778 goto out;
779 } else
780 part->block_size = mtd->erasesize;
783 if (scan_header(part) == 0) {
784 part->mbd.size = part->sector_count;
785 part->mbd.tr = tr;
786 part->mbd.devnum = -1;
787 if (!(mtd->flags & MTD_WRITEABLE))
788 part->mbd.readonly = 1;
789 else if (part->errors) {
790 printk(KERN_WARNING PREFIX "'%s': errors found, "
791 "setting read-only\n", mtd->name);
792 part->mbd.readonly = 1;
795 printk(KERN_INFO PREFIX "name: '%s' type: %d flags %x\n",
796 mtd->name, mtd->type, mtd->flags);
798 if (!add_mtd_blktrans_dev((void*)part))
799 return;
801 out:
802 kfree(part);
805 static void rfd_ftl_remove_dev(struct mtd_blktrans_dev *dev)
807 struct partition *part = (struct partition*)dev;
808 int i;
810 for (i=0; i<part->total_blocks; i++) {
811 pr_debug("rfd_ftl_remove_dev:'%s': erase unit #%02d: %d erases\n",
812 part->mbd.mtd->name, i, part->blocks[i].erases);
815 del_mtd_blktrans_dev(dev);
816 vfree(part->sector_map);
817 kfree(part->header_cache);
818 kfree(part->blocks);
821 static struct mtd_blktrans_ops rfd_ftl_tr = {
822 .name = "rfd",
823 .major = RFD_FTL_MAJOR,
824 .part_bits = PART_BITS,
825 .blksize = SECTOR_SIZE,
827 .readsect = rfd_ftl_readsect,
828 .writesect = rfd_ftl_writesect,
829 .getgeo = rfd_ftl_getgeo,
830 .add_mtd = rfd_ftl_add_mtd,
831 .remove_dev = rfd_ftl_remove_dev,
832 .owner = THIS_MODULE,
835 static int __init init_rfd_ftl(void)
837 return register_mtd_blktrans(&rfd_ftl_tr);
840 static void __exit cleanup_rfd_ftl(void)
842 deregister_mtd_blktrans(&rfd_ftl_tr);
845 module_init(init_rfd_ftl);
846 module_exit(cleanup_rfd_ftl);
848 MODULE_LICENSE("GPL");
849 MODULE_AUTHOR("Sean Young <sean@mess.org>");
850 MODULE_DESCRIPTION("Support code for RFD Flash Translation Layer, "
851 "used by General Software's Embedded BIOS");