2 * MTD device concatenation layer
4 * Copyright © 2002 Robert Kaiser <rkaiser@sysgo.de>
5 * Copyright © 2002-2010 David Woodhouse <dwmw2@infradead.org>
7 * NAND support by Christian Gan <cgan@iders.ca>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/slab.h>
28 #include <linux/sched.h>
29 #include <linux/types.h>
30 #include <linux/backing-dev.h>
32 #include <linux/mtd/mtd.h>
33 #include <linux/mtd/concat.h>
35 #include <asm/div64.h>
38 * Our storage structure:
39 * Subdev points to an array of pointers to struct mtd_info objects
40 * which is allocated along with this structure
46 struct mtd_info **subdev;
50 * how to calculate the size required for the above structure,
51 * including the pointer array subdev points to:
53 #define SIZEOF_STRUCT_MTD_CONCAT(num_subdev) \
54 ((sizeof(struct mtd_concat) + (num_subdev) * sizeof(struct mtd_info *)))
57 * Given a pointer to the MTD object in the mtd_concat structure,
58 * we can retrieve the pointer to that structure with this macro.
60 #define CONCAT(x) ((struct mtd_concat *)(x))
63 * MTD methods which look up the relevant subdevice, translate the
64 * effective address and pass through to the subdevice.
68 concat_read(struct mtd_info *mtd, loff_t from, size_t len,
69 size_t * retlen, u_char * buf)
71 struct mtd_concat *concat = CONCAT(mtd);
75 for (i = 0; i < concat->num_subdev; i++) {
76 struct mtd_info *subdev = concat->subdev[i];
79 if (from >= subdev->size) {
80 /* Not destined for this subdev */
85 if (from + len > subdev->size)
86 /* First part goes into this subdev */
87 size = subdev->size - from;
89 /* Entire transaction goes into this subdev */
92 err = mtd_read(subdev, from, size, &retsize, buf);
94 /* Save information about bitflips! */
96 if (mtd_is_eccerr(err)) {
97 mtd->ecc_stats.failed++;
99 } else if (mtd_is_bitflip(err)) {
100 mtd->ecc_stats.corrected++;
101 /* Do not overwrite -EBADMSG !! */
120 concat_write(struct mtd_info *mtd, loff_t to, size_t len,
121 size_t * retlen, const u_char * buf)
123 struct mtd_concat *concat = CONCAT(mtd);
127 for (i = 0; i < concat->num_subdev; i++) {
128 struct mtd_info *subdev = concat->subdev[i];
129 size_t size, retsize;
131 if (to >= subdev->size) {
136 if (to + len > subdev->size)
137 size = subdev->size - to;
141 err = mtd_write(subdev, to, size, &retsize, buf);
158 concat_writev(struct mtd_info *mtd, const struct kvec *vecs,
159 unsigned long count, loff_t to, size_t * retlen)
161 struct mtd_concat *concat = CONCAT(mtd);
162 struct kvec *vecs_copy;
163 unsigned long entry_low, entry_high;
164 size_t total_len = 0;
168 /* Calculate total length of data */
169 for (i = 0; i < count; i++)
170 total_len += vecs[i].iov_len;
172 /* Check alignment */
173 if (mtd->writesize > 1) {
175 if (do_div(__to, mtd->writesize) || (total_len % mtd->writesize))
179 /* make a copy of vecs */
180 vecs_copy = kmemdup(vecs, sizeof(struct kvec) * count, GFP_KERNEL);
185 for (i = 0; i < concat->num_subdev; i++) {
186 struct mtd_info *subdev = concat->subdev[i];
187 size_t size, wsize, retsize, old_iov_len;
189 if (to >= subdev->size) {
194 size = min_t(uint64_t, total_len, subdev->size - to);
195 wsize = size; /* store for future use */
197 entry_high = entry_low;
198 while (entry_high < count) {
199 if (size <= vecs_copy[entry_high].iov_len)
201 size -= vecs_copy[entry_high++].iov_len;
204 old_iov_len = vecs_copy[entry_high].iov_len;
205 vecs_copy[entry_high].iov_len = size;
207 err = mtd_writev(subdev, &vecs_copy[entry_low],
208 entry_high - entry_low + 1, to, &retsize);
210 vecs_copy[entry_high].iov_len = old_iov_len - size;
211 vecs_copy[entry_high].iov_base += size;
213 entry_low = entry_high;
233 concat_read_oob(struct mtd_info *mtd, loff_t from, struct mtd_oob_ops *ops)
235 struct mtd_concat *concat = CONCAT(mtd);
236 struct mtd_oob_ops devops = *ops;
239 ops->retlen = ops->oobretlen = 0;
241 for (i = 0; i < concat->num_subdev; i++) {
242 struct mtd_info *subdev = concat->subdev[i];
244 if (from >= subdev->size) {
245 from -= subdev->size;
250 if (from + devops.len > subdev->size)
251 devops.len = subdev->size - from;
253 err = mtd_read_oob(subdev, from, &devops);
254 ops->retlen += devops.retlen;
255 ops->oobretlen += devops.oobretlen;
257 /* Save information about bitflips! */
259 if (mtd_is_eccerr(err)) {
260 mtd->ecc_stats.failed++;
262 } else if (mtd_is_bitflip(err)) {
263 mtd->ecc_stats.corrected++;
264 /* Do not overwrite -EBADMSG !! */
272 devops.len = ops->len - ops->retlen;
275 devops.datbuf += devops.retlen;
278 devops.ooblen = ops->ooblen - ops->oobretlen;
281 devops.oobbuf += ops->oobretlen;
290 concat_write_oob(struct mtd_info *mtd, loff_t to, struct mtd_oob_ops *ops)
292 struct mtd_concat *concat = CONCAT(mtd);
293 struct mtd_oob_ops devops = *ops;
296 if (!(mtd->flags & MTD_WRITEABLE))
299 ops->retlen = ops->oobretlen = 0;
301 for (i = 0; i < concat->num_subdev; i++) {
302 struct mtd_info *subdev = concat->subdev[i];
304 if (to >= subdev->size) {
309 /* partial write ? */
310 if (to + devops.len > subdev->size)
311 devops.len = subdev->size - to;
313 err = mtd_write_oob(subdev, to, &devops);
314 ops->retlen += devops.oobretlen;
319 devops.len = ops->len - ops->retlen;
322 devops.datbuf += devops.retlen;
325 devops.ooblen = ops->ooblen - ops->oobretlen;
328 devops.oobbuf += devops.oobretlen;
335 static void concat_erase_callback(struct erase_info *instr)
337 wake_up((wait_queue_head_t *) instr->priv);
340 static int concat_dev_erase(struct mtd_info *mtd, struct erase_info *erase)
343 wait_queue_head_t waitq;
344 DECLARE_WAITQUEUE(wait, current);
347 * This code was stol^H^H^H^Hinspired by mtdchar.c
349 init_waitqueue_head(&waitq);
352 erase->callback = concat_erase_callback;
353 erase->priv = (unsigned long) &waitq;
356 * FIXME: Allow INTERRUPTIBLE. Which means
357 * not having the wait_queue head on the stack.
359 err = mtd_erase(mtd, erase);
361 set_current_state(TASK_UNINTERRUPTIBLE);
362 add_wait_queue(&waitq, &wait);
363 if (erase->state != MTD_ERASE_DONE
364 && erase->state != MTD_ERASE_FAILED)
366 remove_wait_queue(&waitq, &wait);
367 set_current_state(TASK_RUNNING);
369 err = (erase->state == MTD_ERASE_FAILED) ? -EIO : 0;
374 static int concat_erase(struct mtd_info *mtd, struct erase_info *instr)
376 struct mtd_concat *concat = CONCAT(mtd);
377 struct mtd_info *subdev;
379 uint64_t length, offset = 0;
380 struct erase_info *erase;
383 * Check for proper erase block alignment of the to-be-erased area.
384 * It is easier to do this based on the super device's erase
385 * region info rather than looking at each particular sub-device
388 if (!concat->mtd.numeraseregions) {
389 /* the easy case: device has uniform erase block size */
390 if (instr->addr & (concat->mtd.erasesize - 1))
392 if (instr->len & (concat->mtd.erasesize - 1))
395 /* device has variable erase size */
396 struct mtd_erase_region_info *erase_regions =
397 concat->mtd.eraseregions;
400 * Find the erase region where the to-be-erased area begins:
402 for (i = 0; i < concat->mtd.numeraseregions &&
403 instr->addr >= erase_regions[i].offset; i++) ;
407 * Now erase_regions[i] is the region in which the
408 * to-be-erased area begins. Verify that the starting
409 * offset is aligned to this region's erase size:
411 if (i < 0 || instr->addr & (erase_regions[i].erasesize - 1))
415 * now find the erase region where the to-be-erased area ends:
417 for (; i < concat->mtd.numeraseregions &&
418 (instr->addr + instr->len) >= erase_regions[i].offset;
422 * check if the ending offset is aligned to this region's erase size
424 if (i < 0 || ((instr->addr + instr->len) &
425 (erase_regions[i].erasesize - 1)))
429 /* make a local copy of instr to avoid modifying the caller's struct */
430 erase = kmalloc(sizeof (struct erase_info), GFP_KERNEL);
439 * find the subdevice where the to-be-erased area begins, adjust
440 * starting offset to be relative to the subdevice start
442 for (i = 0; i < concat->num_subdev; i++) {
443 subdev = concat->subdev[i];
444 if (subdev->size <= erase->addr) {
445 erase->addr -= subdev->size;
446 offset += subdev->size;
452 /* must never happen since size limit has been verified above */
453 BUG_ON(i >= concat->num_subdev);
455 /* now do the erase: */
457 for (; length > 0; i++) {
458 /* loop for all subdevices affected by this request */
459 subdev = concat->subdev[i]; /* get current subdevice */
461 /* limit length to subdevice's size: */
462 if (erase->addr + length > subdev->size)
463 erase->len = subdev->size - erase->addr;
467 length -= erase->len;
468 if ((err = concat_dev_erase(subdev, erase))) {
469 /* sanity check: should never happen since
470 * block alignment has been checked above */
471 BUG_ON(err == -EINVAL);
472 if (erase->fail_addr != MTD_FAIL_ADDR_UNKNOWN)
473 instr->fail_addr = erase->fail_addr + offset;
477 * erase->addr specifies the offset of the area to be
478 * erased *within the current subdevice*. It can be
479 * non-zero only the first time through this loop, i.e.
480 * for the first subdevice where blocks need to be erased.
481 * All the following erases must begin at the start of the
482 * current subdevice, i.e. at offset zero.
485 offset += subdev->size;
487 instr->state = erase->state;
493 instr->callback(instr);
497 static int concat_lock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
499 struct mtd_concat *concat = CONCAT(mtd);
500 int i, err = -EINVAL;
502 for (i = 0; i < concat->num_subdev; i++) {
503 struct mtd_info *subdev = concat->subdev[i];
506 if (ofs >= subdev->size) {
511 if (ofs + len > subdev->size)
512 size = subdev->size - ofs;
516 err = mtd_lock(subdev, ofs, size);
531 static int concat_unlock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
533 struct mtd_concat *concat = CONCAT(mtd);
536 for (i = 0; i < concat->num_subdev; i++) {
537 struct mtd_info *subdev = concat->subdev[i];
540 if (ofs >= subdev->size) {
545 if (ofs + len > subdev->size)
546 size = subdev->size - ofs;
550 err = mtd_unlock(subdev, ofs, size);
565 static void concat_sync(struct mtd_info *mtd)
567 struct mtd_concat *concat = CONCAT(mtd);
570 for (i = 0; i < concat->num_subdev; i++) {
571 struct mtd_info *subdev = concat->subdev[i];
576 static int concat_suspend(struct mtd_info *mtd)
578 struct mtd_concat *concat = CONCAT(mtd);
581 for (i = 0; i < concat->num_subdev; i++) {
582 struct mtd_info *subdev = concat->subdev[i];
583 if ((rc = mtd_suspend(subdev)) < 0)
589 static void concat_resume(struct mtd_info *mtd)
591 struct mtd_concat *concat = CONCAT(mtd);
594 for (i = 0; i < concat->num_subdev; i++) {
595 struct mtd_info *subdev = concat->subdev[i];
600 static int concat_block_isbad(struct mtd_info *mtd, loff_t ofs)
602 struct mtd_concat *concat = CONCAT(mtd);
605 if (!mtd_can_have_bb(concat->subdev[0]))
608 for (i = 0; i < concat->num_subdev; i++) {
609 struct mtd_info *subdev = concat->subdev[i];
611 if (ofs >= subdev->size) {
616 res = mtd_block_isbad(subdev, ofs);
623 static int concat_block_markbad(struct mtd_info *mtd, loff_t ofs)
625 struct mtd_concat *concat = CONCAT(mtd);
626 int i, err = -EINVAL;
628 for (i = 0; i < concat->num_subdev; i++) {
629 struct mtd_info *subdev = concat->subdev[i];
631 if (ofs >= subdev->size) {
636 err = mtd_block_markbad(subdev, ofs);
638 mtd->ecc_stats.badblocks++;
646 * try to support NOMMU mmaps on concatenated devices
647 * - we don't support subdev spanning as we can't guarantee it'll work
649 static unsigned long concat_get_unmapped_area(struct mtd_info *mtd,
651 unsigned long offset,
654 struct mtd_concat *concat = CONCAT(mtd);
657 for (i = 0; i < concat->num_subdev; i++) {
658 struct mtd_info *subdev = concat->subdev[i];
660 if (offset >= subdev->size) {
661 offset -= subdev->size;
665 return mtd_get_unmapped_area(subdev, len, offset, flags);
668 return (unsigned long) -ENOSYS;
672 * This function constructs a virtual MTD device by concatenating
673 * num_devs MTD devices. A pointer to the new device object is
674 * stored to *new_dev upon success. This function does _not_
675 * register any devices: this is the caller's responsibility.
677 struct mtd_info *mtd_concat_create(struct mtd_info *subdev[], /* subdevices to concatenate */
678 int num_devs, /* number of subdevices */
680 { /* name for the new device */
683 struct mtd_concat *concat;
684 uint32_t max_erasesize, curr_erasesize;
685 int num_erase_region;
686 int max_writebufsize = 0;
688 printk(KERN_NOTICE "Concatenating MTD devices:\n");
689 for (i = 0; i < num_devs; i++)
690 printk(KERN_NOTICE "(%d): \"%s\"\n", i, subdev[i]->name);
691 printk(KERN_NOTICE "into device \"%s\"\n", name);
693 /* allocate the device structure */
694 size = SIZEOF_STRUCT_MTD_CONCAT(num_devs);
695 concat = kzalloc(size, GFP_KERNEL);
698 ("memory allocation error while creating concatenated device \"%s\"\n",
702 concat->subdev = (struct mtd_info **) (concat + 1);
705 * Set up the new "super" device's MTD object structure, check for
706 * incompatibilities between the subdevices.
708 concat->mtd.type = subdev[0]->type;
709 concat->mtd.flags = subdev[0]->flags;
710 concat->mtd.size = subdev[0]->size;
711 concat->mtd.erasesize = subdev[0]->erasesize;
712 concat->mtd.writesize = subdev[0]->writesize;
714 for (i = 0; i < num_devs; i++)
715 if (max_writebufsize < subdev[i]->writebufsize)
716 max_writebufsize = subdev[i]->writebufsize;
717 concat->mtd.writebufsize = max_writebufsize;
719 concat->mtd.subpage_sft = subdev[0]->subpage_sft;
720 concat->mtd.oobsize = subdev[0]->oobsize;
721 concat->mtd.oobavail = subdev[0]->oobavail;
722 if (subdev[0]->_writev)
723 concat->mtd._writev = concat_writev;
724 if (subdev[0]->_read_oob)
725 concat->mtd._read_oob = concat_read_oob;
726 if (subdev[0]->_write_oob)
727 concat->mtd._write_oob = concat_write_oob;
728 if (subdev[0]->_block_isbad)
729 concat->mtd._block_isbad = concat_block_isbad;
730 if (subdev[0]->_block_markbad)
731 concat->mtd._block_markbad = concat_block_markbad;
733 concat->mtd.ecc_stats.badblocks = subdev[0]->ecc_stats.badblocks;
735 concat->mtd.backing_dev_info = subdev[0]->backing_dev_info;
737 concat->subdev[0] = subdev[0];
739 for (i = 1; i < num_devs; i++) {
740 if (concat->mtd.type != subdev[i]->type) {
742 printk("Incompatible device type on \"%s\"\n",
746 if (concat->mtd.flags != subdev[i]->flags) {
748 * Expect all flags except MTD_WRITEABLE to be
749 * equal on all subdevices.
751 if ((concat->mtd.flags ^ subdev[i]->
752 flags) & ~MTD_WRITEABLE) {
754 printk("Incompatible device flags on \"%s\"\n",
758 /* if writeable attribute differs,
759 make super device writeable */
761 subdev[i]->flags & MTD_WRITEABLE;
764 /* only permit direct mapping if the BDIs are all the same
765 * - copy-mapping is still permitted
767 if (concat->mtd.backing_dev_info !=
768 subdev[i]->backing_dev_info)
769 concat->mtd.backing_dev_info =
770 &default_backing_dev_info;
772 concat->mtd.size += subdev[i]->size;
773 concat->mtd.ecc_stats.badblocks +=
774 subdev[i]->ecc_stats.badblocks;
775 if (concat->mtd.writesize != subdev[i]->writesize ||
776 concat->mtd.subpage_sft != subdev[i]->subpage_sft ||
777 concat->mtd.oobsize != subdev[i]->oobsize ||
778 !concat->mtd._read_oob != !subdev[i]->_read_oob ||
779 !concat->mtd._write_oob != !subdev[i]->_write_oob) {
781 printk("Incompatible OOB or ECC data on \"%s\"\n",
785 concat->subdev[i] = subdev[i];
789 concat->mtd.ecclayout = subdev[0]->ecclayout;
791 concat->num_subdev = num_devs;
792 concat->mtd.name = name;
794 concat->mtd._erase = concat_erase;
795 concat->mtd._read = concat_read;
796 concat->mtd._write = concat_write;
797 concat->mtd._sync = concat_sync;
798 concat->mtd._lock = concat_lock;
799 concat->mtd._unlock = concat_unlock;
800 concat->mtd._suspend = concat_suspend;
801 concat->mtd._resume = concat_resume;
802 concat->mtd._get_unmapped_area = concat_get_unmapped_area;
805 * Combine the erase block size info of the subdevices:
807 * first, walk the map of the new device and see how
808 * many changes in erase size we have
810 max_erasesize = curr_erasesize = subdev[0]->erasesize;
811 num_erase_region = 1;
812 for (i = 0; i < num_devs; i++) {
813 if (subdev[i]->numeraseregions == 0) {
814 /* current subdevice has uniform erase size */
815 if (subdev[i]->erasesize != curr_erasesize) {
816 /* if it differs from the last subdevice's erase size, count it */
818 curr_erasesize = subdev[i]->erasesize;
819 if (curr_erasesize > max_erasesize)
820 max_erasesize = curr_erasesize;
823 /* current subdevice has variable erase size */
825 for (j = 0; j < subdev[i]->numeraseregions; j++) {
827 /* walk the list of erase regions, count any changes */
828 if (subdev[i]->eraseregions[j].erasesize !=
832 subdev[i]->eraseregions[j].
834 if (curr_erasesize > max_erasesize)
835 max_erasesize = curr_erasesize;
841 if (num_erase_region == 1) {
843 * All subdevices have the same uniform erase size.
846 concat->mtd.erasesize = curr_erasesize;
847 concat->mtd.numeraseregions = 0;
852 * erase block size varies across the subdevices: allocate
853 * space to store the data describing the variable erase regions
855 struct mtd_erase_region_info *erase_region_p;
856 uint64_t begin, position;
858 concat->mtd.erasesize = max_erasesize;
859 concat->mtd.numeraseregions = num_erase_region;
860 concat->mtd.eraseregions = erase_region_p =
861 kmalloc(num_erase_region *
862 sizeof (struct mtd_erase_region_info), GFP_KERNEL);
863 if (!erase_region_p) {
866 ("memory allocation error while creating erase region list"
867 " for device \"%s\"\n", name);
872 * walk the map of the new device once more and fill in
873 * in erase region info:
875 curr_erasesize = subdev[0]->erasesize;
876 begin = position = 0;
877 for (i = 0; i < num_devs; i++) {
878 if (subdev[i]->numeraseregions == 0) {
879 /* current subdevice has uniform erase size */
880 if (subdev[i]->erasesize != curr_erasesize) {
882 * fill in an mtd_erase_region_info structure for the area
883 * we have walked so far:
885 erase_region_p->offset = begin;
886 erase_region_p->erasesize =
888 tmp64 = position - begin;
889 do_div(tmp64, curr_erasesize);
890 erase_region_p->numblocks = tmp64;
893 curr_erasesize = subdev[i]->erasesize;
896 position += subdev[i]->size;
898 /* current subdevice has variable erase size */
900 for (j = 0; j < subdev[i]->numeraseregions; j++) {
901 /* walk the list of erase regions, count any changes */
902 if (subdev[i]->eraseregions[j].
903 erasesize != curr_erasesize) {
904 erase_region_p->offset = begin;
905 erase_region_p->erasesize =
907 tmp64 = position - begin;
908 do_div(tmp64, curr_erasesize);
909 erase_region_p->numblocks = tmp64;
913 subdev[i]->eraseregions[j].
918 subdev[i]->eraseregions[j].
919 numblocks * (uint64_t)curr_erasesize;
923 /* Now write the final entry */
924 erase_region_p->offset = begin;
925 erase_region_p->erasesize = curr_erasesize;
926 tmp64 = position - begin;
927 do_div(tmp64, curr_erasesize);
928 erase_region_p->numblocks = tmp64;
935 * This function destroys an MTD object obtained from concat_mtd_devs()
938 void mtd_concat_destroy(struct mtd_info *mtd)
940 struct mtd_concat *concat = CONCAT(mtd);
941 if (concat->mtd.numeraseregions)
942 kfree(concat->mtd.eraseregions);
946 EXPORT_SYMBOL(mtd_concat_create);
947 EXPORT_SYMBOL(mtd_concat_destroy);
949 MODULE_LICENSE("GPL");
950 MODULE_AUTHOR("Robert Kaiser <rkaiser@sysgo.de>");
951 MODULE_DESCRIPTION("Generic support for concatenating of MTD devices");