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);
77 for (i = 0; i < concat->num_subdev; i++) {
78 struct mtd_info *subdev = concat->subdev[i];
81 if (from >= subdev->size) {
82 /* Not destined for this subdev */
87 if (from + len > subdev->size)
88 /* First part goes into this subdev */
89 size = subdev->size - from;
91 /* Entire transaction goes into this subdev */
94 err = mtd_read(subdev, from, size, &retsize, buf);
96 /* Save information about bitflips! */
98 if (mtd_is_eccerr(err)) {
99 mtd->ecc_stats.failed++;
101 } else if (mtd_is_bitflip(err)) {
102 mtd->ecc_stats.corrected++;
103 /* Do not overwrite -EBADMSG !! */
122 concat_write(struct mtd_info *mtd, loff_t to, size_t len,
123 size_t * retlen, const u_char * buf)
125 struct mtd_concat *concat = CONCAT(mtd);
129 if (!(mtd->flags & MTD_WRITEABLE))
134 for (i = 0; i < concat->num_subdev; i++) {
135 struct mtd_info *subdev = concat->subdev[i];
136 size_t size, retsize;
138 if (to >= subdev->size) {
143 if (to + len > subdev->size)
144 size = subdev->size - to;
148 if (!(subdev->flags & MTD_WRITEABLE))
151 err = mtd_write(subdev, to, size, &retsize, buf);
169 concat_writev(struct mtd_info *mtd, const struct kvec *vecs,
170 unsigned long count, loff_t to, size_t * retlen)
172 struct mtd_concat *concat = CONCAT(mtd);
173 struct kvec *vecs_copy;
174 unsigned long entry_low, entry_high;
175 size_t total_len = 0;
179 if (!(mtd->flags & MTD_WRITEABLE))
184 /* Calculate total length of data */
185 for (i = 0; i < count; i++)
186 total_len += vecs[i].iov_len;
188 /* Do not allow write past end of device */
189 if ((to + total_len) > mtd->size)
192 /* Check alignment */
193 if (mtd->writesize > 1) {
195 if (do_div(__to, mtd->writesize) || (total_len % mtd->writesize))
199 /* make a copy of vecs */
200 vecs_copy = kmemdup(vecs, sizeof(struct kvec) * count, GFP_KERNEL);
205 for (i = 0; i < concat->num_subdev; i++) {
206 struct mtd_info *subdev = concat->subdev[i];
207 size_t size, wsize, retsize, old_iov_len;
209 if (to >= subdev->size) {
214 size = min_t(uint64_t, total_len, subdev->size - to);
215 wsize = size; /* store for future use */
217 entry_high = entry_low;
218 while (entry_high < count) {
219 if (size <= vecs_copy[entry_high].iov_len)
221 size -= vecs_copy[entry_high++].iov_len;
224 old_iov_len = vecs_copy[entry_high].iov_len;
225 vecs_copy[entry_high].iov_len = size;
227 if (!(subdev->flags & MTD_WRITEABLE))
230 err = mtd_writev(subdev, &vecs_copy[entry_low],
231 entry_high - entry_low + 1, to,
234 vecs_copy[entry_high].iov_len = old_iov_len - size;
235 vecs_copy[entry_high].iov_base += size;
237 entry_low = entry_high;
257 concat_read_oob(struct mtd_info *mtd, loff_t from, struct mtd_oob_ops *ops)
259 struct mtd_concat *concat = CONCAT(mtd);
260 struct mtd_oob_ops devops = *ops;
263 ops->retlen = ops->oobretlen = 0;
265 for (i = 0; i < concat->num_subdev; i++) {
266 struct mtd_info *subdev = concat->subdev[i];
268 if (from >= subdev->size) {
269 from -= subdev->size;
274 if (from + devops.len > subdev->size)
275 devops.len = subdev->size - from;
277 err = mtd_read_oob(subdev, from, &devops);
278 ops->retlen += devops.retlen;
279 ops->oobretlen += devops.oobretlen;
281 /* Save information about bitflips! */
283 if (mtd_is_eccerr(err)) {
284 mtd->ecc_stats.failed++;
286 } else if (mtd_is_bitflip(err)) {
287 mtd->ecc_stats.corrected++;
288 /* Do not overwrite -EBADMSG !! */
296 devops.len = ops->len - ops->retlen;
299 devops.datbuf += devops.retlen;
302 devops.ooblen = ops->ooblen - ops->oobretlen;
305 devops.oobbuf += ops->oobretlen;
314 concat_write_oob(struct mtd_info *mtd, loff_t to, struct mtd_oob_ops *ops)
316 struct mtd_concat *concat = CONCAT(mtd);
317 struct mtd_oob_ops devops = *ops;
320 if (!(mtd->flags & MTD_WRITEABLE))
323 ops->retlen = ops->oobretlen = 0;
325 for (i = 0; i < concat->num_subdev; i++) {
326 struct mtd_info *subdev = concat->subdev[i];
328 if (to >= subdev->size) {
333 /* partial write ? */
334 if (to + devops.len > subdev->size)
335 devops.len = subdev->size - to;
337 err = mtd_write_oob(subdev, to, &devops);
338 ops->retlen += devops.oobretlen;
343 devops.len = ops->len - ops->retlen;
346 devops.datbuf += devops.retlen;
349 devops.ooblen = ops->ooblen - ops->oobretlen;
352 devops.oobbuf += devops.oobretlen;
359 static void concat_erase_callback(struct erase_info *instr)
361 wake_up((wait_queue_head_t *) instr->priv);
364 static int concat_dev_erase(struct mtd_info *mtd, struct erase_info *erase)
367 wait_queue_head_t waitq;
368 DECLARE_WAITQUEUE(wait, current);
371 * This code was stol^H^H^H^Hinspired by mtdchar.c
373 init_waitqueue_head(&waitq);
376 erase->callback = concat_erase_callback;
377 erase->priv = (unsigned long) &waitq;
380 * FIXME: Allow INTERRUPTIBLE. Which means
381 * not having the wait_queue head on the stack.
383 err = mtd_erase(mtd, erase);
385 set_current_state(TASK_UNINTERRUPTIBLE);
386 add_wait_queue(&waitq, &wait);
387 if (erase->state != MTD_ERASE_DONE
388 && erase->state != MTD_ERASE_FAILED)
390 remove_wait_queue(&waitq, &wait);
391 set_current_state(TASK_RUNNING);
393 err = (erase->state == MTD_ERASE_FAILED) ? -EIO : 0;
398 static int concat_erase(struct mtd_info *mtd, struct erase_info *instr)
400 struct mtd_concat *concat = CONCAT(mtd);
401 struct mtd_info *subdev;
403 uint64_t length, offset = 0;
404 struct erase_info *erase;
406 if (!(mtd->flags & MTD_WRITEABLE))
409 if (instr->addr > concat->mtd.size)
412 if (instr->len + instr->addr > concat->mtd.size)
416 * Check for proper erase block alignment of the to-be-erased area.
417 * It is easier to do this based on the super device's erase
418 * region info rather than looking at each particular sub-device
421 if (!concat->mtd.numeraseregions) {
422 /* the easy case: device has uniform erase block size */
423 if (instr->addr & (concat->mtd.erasesize - 1))
425 if (instr->len & (concat->mtd.erasesize - 1))
428 /* device has variable erase size */
429 struct mtd_erase_region_info *erase_regions =
430 concat->mtd.eraseregions;
433 * Find the erase region where the to-be-erased area begins:
435 for (i = 0; i < concat->mtd.numeraseregions &&
436 instr->addr >= erase_regions[i].offset; i++) ;
440 * Now erase_regions[i] is the region in which the
441 * to-be-erased area begins. Verify that the starting
442 * offset is aligned to this region's erase size:
444 if (i < 0 || instr->addr & (erase_regions[i].erasesize - 1))
448 * now find the erase region where the to-be-erased area ends:
450 for (; i < concat->mtd.numeraseregions &&
451 (instr->addr + instr->len) >= erase_regions[i].offset;
455 * check if the ending offset is aligned to this region's erase size
457 if (i < 0 || ((instr->addr + instr->len) &
458 (erase_regions[i].erasesize - 1)))
462 instr->fail_addr = MTD_FAIL_ADDR_UNKNOWN;
464 /* make a local copy of instr to avoid modifying the caller's struct */
465 erase = kmalloc(sizeof (struct erase_info), GFP_KERNEL);
474 * find the subdevice where the to-be-erased area begins, adjust
475 * starting offset to be relative to the subdevice start
477 for (i = 0; i < concat->num_subdev; i++) {
478 subdev = concat->subdev[i];
479 if (subdev->size <= erase->addr) {
480 erase->addr -= subdev->size;
481 offset += subdev->size;
487 /* must never happen since size limit has been verified above */
488 BUG_ON(i >= concat->num_subdev);
490 /* now do the erase: */
492 for (; length > 0; i++) {
493 /* loop for all subdevices affected by this request */
494 subdev = concat->subdev[i]; /* get current subdevice */
496 /* limit length to subdevice's size: */
497 if (erase->addr + length > subdev->size)
498 erase->len = subdev->size - erase->addr;
502 if (!(subdev->flags & MTD_WRITEABLE)) {
506 length -= erase->len;
507 if ((err = concat_dev_erase(subdev, erase))) {
508 /* sanity check: should never happen since
509 * block alignment has been checked above */
510 BUG_ON(err == -EINVAL);
511 if (erase->fail_addr != MTD_FAIL_ADDR_UNKNOWN)
512 instr->fail_addr = erase->fail_addr + offset;
516 * erase->addr specifies the offset of the area to be
517 * erased *within the current subdevice*. It can be
518 * non-zero only the first time through this loop, i.e.
519 * for the first subdevice where blocks need to be erased.
520 * All the following erases must begin at the start of the
521 * current subdevice, i.e. at offset zero.
524 offset += subdev->size;
526 instr->state = erase->state;
532 instr->callback(instr);
536 static int concat_lock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
538 struct mtd_concat *concat = CONCAT(mtd);
539 int i, err = -EINVAL;
541 if ((len + ofs) > mtd->size)
544 for (i = 0; i < concat->num_subdev; i++) {
545 struct mtd_info *subdev = concat->subdev[i];
548 if (ofs >= subdev->size) {
553 if (ofs + len > subdev->size)
554 size = subdev->size - ofs;
559 err = subdev->lock(subdev, ofs, size);
576 static int concat_unlock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
578 struct mtd_concat *concat = CONCAT(mtd);
581 if ((len + ofs) > mtd->size)
584 for (i = 0; i < concat->num_subdev; i++) {
585 struct mtd_info *subdev = concat->subdev[i];
588 if (ofs >= subdev->size) {
593 if (ofs + len > subdev->size)
594 size = subdev->size - ofs;
598 if (subdev->unlock) {
599 err = subdev->unlock(subdev, ofs, size);
616 static void concat_sync(struct mtd_info *mtd)
618 struct mtd_concat *concat = CONCAT(mtd);
621 for (i = 0; i < concat->num_subdev; i++) {
622 struct mtd_info *subdev = concat->subdev[i];
623 subdev->sync(subdev);
627 static int concat_suspend(struct mtd_info *mtd)
629 struct mtd_concat *concat = CONCAT(mtd);
632 for (i = 0; i < concat->num_subdev; i++) {
633 struct mtd_info *subdev = concat->subdev[i];
634 if ((rc = subdev->suspend(subdev)) < 0)
640 static void concat_resume(struct mtd_info *mtd)
642 struct mtd_concat *concat = CONCAT(mtd);
645 for (i = 0; i < concat->num_subdev; i++) {
646 struct mtd_info *subdev = concat->subdev[i];
647 subdev->resume(subdev);
651 static int concat_block_isbad(struct mtd_info *mtd, loff_t ofs)
653 struct mtd_concat *concat = CONCAT(mtd);
656 if (!concat->subdev[0]->block_isbad)
662 for (i = 0; i < concat->num_subdev; i++) {
663 struct mtd_info *subdev = concat->subdev[i];
665 if (ofs >= subdev->size) {
670 res = subdev->block_isbad(subdev, ofs);
677 static int concat_block_markbad(struct mtd_info *mtd, loff_t ofs)
679 struct mtd_concat *concat = CONCAT(mtd);
680 int i, err = -EINVAL;
682 if (!concat->subdev[0]->block_markbad)
688 for (i = 0; i < concat->num_subdev; i++) {
689 struct mtd_info *subdev = concat->subdev[i];
691 if (ofs >= subdev->size) {
696 err = subdev->block_markbad(subdev, ofs);
698 mtd->ecc_stats.badblocks++;
706 * try to support NOMMU mmaps on concatenated devices
707 * - we don't support subdev spanning as we can't guarantee it'll work
709 static unsigned long concat_get_unmapped_area(struct mtd_info *mtd,
711 unsigned long offset,
714 struct mtd_concat *concat = CONCAT(mtd);
717 for (i = 0; i < concat->num_subdev; i++) {
718 struct mtd_info *subdev = concat->subdev[i];
720 if (offset >= subdev->size) {
721 offset -= subdev->size;
725 /* we've found the subdev over which the mapping will reside */
726 if (offset + len > subdev->size)
727 return (unsigned long) -EINVAL;
729 if (subdev->get_unmapped_area)
730 return mtd_get_unmapped_area(subdev, len, offset,
736 return (unsigned long) -ENOSYS;
740 * This function constructs a virtual MTD device by concatenating
741 * num_devs MTD devices. A pointer to the new device object is
742 * stored to *new_dev upon success. This function does _not_
743 * register any devices: this is the caller's responsibility.
745 struct mtd_info *mtd_concat_create(struct mtd_info *subdev[], /* subdevices to concatenate */
746 int num_devs, /* number of subdevices */
748 { /* name for the new device */
751 struct mtd_concat *concat;
752 uint32_t max_erasesize, curr_erasesize;
753 int num_erase_region;
754 int max_writebufsize = 0;
756 printk(KERN_NOTICE "Concatenating MTD devices:\n");
757 for (i = 0; i < num_devs; i++)
758 printk(KERN_NOTICE "(%d): \"%s\"\n", i, subdev[i]->name);
759 printk(KERN_NOTICE "into device \"%s\"\n", name);
761 /* allocate the device structure */
762 size = SIZEOF_STRUCT_MTD_CONCAT(num_devs);
763 concat = kzalloc(size, GFP_KERNEL);
766 ("memory allocation error while creating concatenated device \"%s\"\n",
770 concat->subdev = (struct mtd_info **) (concat + 1);
773 * Set up the new "super" device's MTD object structure, check for
774 * incompatibilities between the subdevices.
776 concat->mtd.type = subdev[0]->type;
777 concat->mtd.flags = subdev[0]->flags;
778 concat->mtd.size = subdev[0]->size;
779 concat->mtd.erasesize = subdev[0]->erasesize;
780 concat->mtd.writesize = subdev[0]->writesize;
782 for (i = 0; i < num_devs; i++)
783 if (max_writebufsize < subdev[i]->writebufsize)
784 max_writebufsize = subdev[i]->writebufsize;
785 concat->mtd.writebufsize = max_writebufsize;
787 concat->mtd.subpage_sft = subdev[0]->subpage_sft;
788 concat->mtd.oobsize = subdev[0]->oobsize;
789 concat->mtd.oobavail = subdev[0]->oobavail;
790 if (subdev[0]->writev)
791 concat->mtd.writev = concat_writev;
792 if (subdev[0]->read_oob)
793 concat->mtd.read_oob = concat_read_oob;
794 if (subdev[0]->write_oob)
795 concat->mtd.write_oob = concat_write_oob;
796 if (subdev[0]->block_isbad)
797 concat->mtd.block_isbad = concat_block_isbad;
798 if (subdev[0]->block_markbad)
799 concat->mtd.block_markbad = concat_block_markbad;
801 concat->mtd.ecc_stats.badblocks = subdev[0]->ecc_stats.badblocks;
803 concat->mtd.backing_dev_info = subdev[0]->backing_dev_info;
805 concat->subdev[0] = subdev[0];
807 for (i = 1; i < num_devs; i++) {
808 if (concat->mtd.type != subdev[i]->type) {
810 printk("Incompatible device type on \"%s\"\n",
814 if (concat->mtd.flags != subdev[i]->flags) {
816 * Expect all flags except MTD_WRITEABLE to be
817 * equal on all subdevices.
819 if ((concat->mtd.flags ^ subdev[i]->
820 flags) & ~MTD_WRITEABLE) {
822 printk("Incompatible device flags on \"%s\"\n",
826 /* if writeable attribute differs,
827 make super device writeable */
829 subdev[i]->flags & MTD_WRITEABLE;
832 /* only permit direct mapping if the BDIs are all the same
833 * - copy-mapping is still permitted
835 if (concat->mtd.backing_dev_info !=
836 subdev[i]->backing_dev_info)
837 concat->mtd.backing_dev_info =
838 &default_backing_dev_info;
840 concat->mtd.size += subdev[i]->size;
841 concat->mtd.ecc_stats.badblocks +=
842 subdev[i]->ecc_stats.badblocks;
843 if (concat->mtd.writesize != subdev[i]->writesize ||
844 concat->mtd.subpage_sft != subdev[i]->subpage_sft ||
845 concat->mtd.oobsize != subdev[i]->oobsize ||
846 !concat->mtd.read_oob != !subdev[i]->read_oob ||
847 !concat->mtd.write_oob != !subdev[i]->write_oob) {
849 printk("Incompatible OOB or ECC data on \"%s\"\n",
853 concat->subdev[i] = subdev[i];
857 concat->mtd.ecclayout = subdev[0]->ecclayout;
859 concat->num_subdev = num_devs;
860 concat->mtd.name = name;
862 concat->mtd.erase = concat_erase;
863 concat->mtd.read = concat_read;
864 concat->mtd.write = concat_write;
865 concat->mtd.sync = concat_sync;
866 concat->mtd.lock = concat_lock;
867 concat->mtd.unlock = concat_unlock;
868 concat->mtd.suspend = concat_suspend;
869 concat->mtd.resume = concat_resume;
870 concat->mtd.get_unmapped_area = concat_get_unmapped_area;
873 * Combine the erase block size info of the subdevices:
875 * first, walk the map of the new device and see how
876 * many changes in erase size we have
878 max_erasesize = curr_erasesize = subdev[0]->erasesize;
879 num_erase_region = 1;
880 for (i = 0; i < num_devs; i++) {
881 if (subdev[i]->numeraseregions == 0) {
882 /* current subdevice has uniform erase size */
883 if (subdev[i]->erasesize != curr_erasesize) {
884 /* if it differs from the last subdevice's erase size, count it */
886 curr_erasesize = subdev[i]->erasesize;
887 if (curr_erasesize > max_erasesize)
888 max_erasesize = curr_erasesize;
891 /* current subdevice has variable erase size */
893 for (j = 0; j < subdev[i]->numeraseregions; j++) {
895 /* walk the list of erase regions, count any changes */
896 if (subdev[i]->eraseregions[j].erasesize !=
900 subdev[i]->eraseregions[j].
902 if (curr_erasesize > max_erasesize)
903 max_erasesize = curr_erasesize;
909 if (num_erase_region == 1) {
911 * All subdevices have the same uniform erase size.
914 concat->mtd.erasesize = curr_erasesize;
915 concat->mtd.numeraseregions = 0;
920 * erase block size varies across the subdevices: allocate
921 * space to store the data describing the variable erase regions
923 struct mtd_erase_region_info *erase_region_p;
924 uint64_t begin, position;
926 concat->mtd.erasesize = max_erasesize;
927 concat->mtd.numeraseregions = num_erase_region;
928 concat->mtd.eraseregions = erase_region_p =
929 kmalloc(num_erase_region *
930 sizeof (struct mtd_erase_region_info), GFP_KERNEL);
931 if (!erase_region_p) {
934 ("memory allocation error while creating erase region list"
935 " for device \"%s\"\n", name);
940 * walk the map of the new device once more and fill in
941 * in erase region info:
943 curr_erasesize = subdev[0]->erasesize;
944 begin = position = 0;
945 for (i = 0; i < num_devs; i++) {
946 if (subdev[i]->numeraseregions == 0) {
947 /* current subdevice has uniform erase size */
948 if (subdev[i]->erasesize != curr_erasesize) {
950 * fill in an mtd_erase_region_info structure for the area
951 * we have walked so far:
953 erase_region_p->offset = begin;
954 erase_region_p->erasesize =
956 tmp64 = position - begin;
957 do_div(tmp64, curr_erasesize);
958 erase_region_p->numblocks = tmp64;
961 curr_erasesize = subdev[i]->erasesize;
964 position += subdev[i]->size;
966 /* current subdevice has variable erase size */
968 for (j = 0; j < subdev[i]->numeraseregions; j++) {
969 /* walk the list of erase regions, count any changes */
970 if (subdev[i]->eraseregions[j].
971 erasesize != curr_erasesize) {
972 erase_region_p->offset = begin;
973 erase_region_p->erasesize =
975 tmp64 = position - begin;
976 do_div(tmp64, curr_erasesize);
977 erase_region_p->numblocks = tmp64;
981 subdev[i]->eraseregions[j].
986 subdev[i]->eraseregions[j].
987 numblocks * (uint64_t)curr_erasesize;
991 /* Now write the final entry */
992 erase_region_p->offset = begin;
993 erase_region_p->erasesize = curr_erasesize;
994 tmp64 = position - begin;
995 do_div(tmp64, curr_erasesize);
996 erase_region_p->numblocks = tmp64;
1003 * This function destroys an MTD object obtained from concat_mtd_devs()
1006 void mtd_concat_destroy(struct mtd_info *mtd)
1008 struct mtd_concat *concat = CONCAT(mtd);
1009 if (concat->mtd.numeraseregions)
1010 kfree(concat->mtd.eraseregions);
1014 EXPORT_SYMBOL(mtd_concat_create);
1015 EXPORT_SYMBOL(mtd_concat_destroy);
1017 MODULE_LICENSE("GPL");
1018 MODULE_AUTHOR("Robert Kaiser <rkaiser@sysgo.de>");
1019 MODULE_DESCRIPTION("Generic support for concatenating of MTD devices");