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 = subdev->read(subdev, from, size, &retsize, buf);
96 /* Save information about bitflips! */
98 if (err == -EBADMSG) {
99 mtd->ecc_stats.failed++;
101 } else if (err == -EUCLEAN) {
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 = subdev->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 = subdev->writev(subdev, &vecs_copy[entry_low],
231 entry_high - entry_low + 1, to, &retsize);
233 vecs_copy[entry_high].iov_len = old_iov_len - size;
234 vecs_copy[entry_high].iov_base += size;
236 entry_low = entry_high;
256 concat_read_oob(struct mtd_info *mtd, loff_t from, struct mtd_oob_ops *ops)
258 struct mtd_concat *concat = CONCAT(mtd);
259 struct mtd_oob_ops devops = *ops;
262 ops->retlen = ops->oobretlen = 0;
264 for (i = 0; i < concat->num_subdev; i++) {
265 struct mtd_info *subdev = concat->subdev[i];
267 if (from >= subdev->size) {
268 from -= subdev->size;
273 if (from + devops.len > subdev->size)
274 devops.len = subdev->size - from;
276 err = subdev->read_oob(subdev, from, &devops);
277 ops->retlen += devops.retlen;
278 ops->oobretlen += devops.oobretlen;
280 /* Save information about bitflips! */
282 if (err == -EBADMSG) {
283 mtd->ecc_stats.failed++;
285 } else if (err == -EUCLEAN) {
286 mtd->ecc_stats.corrected++;
287 /* Do not overwrite -EBADMSG !! */
295 devops.len = ops->len - ops->retlen;
298 devops.datbuf += devops.retlen;
301 devops.ooblen = ops->ooblen - ops->oobretlen;
304 devops.oobbuf += ops->oobretlen;
313 concat_write_oob(struct mtd_info *mtd, loff_t to, struct mtd_oob_ops *ops)
315 struct mtd_concat *concat = CONCAT(mtd);
316 struct mtd_oob_ops devops = *ops;
319 if (!(mtd->flags & MTD_WRITEABLE))
322 ops->retlen = ops->oobretlen = 0;
324 for (i = 0; i < concat->num_subdev; i++) {
325 struct mtd_info *subdev = concat->subdev[i];
327 if (to >= subdev->size) {
332 /* partial write ? */
333 if (to + devops.len > subdev->size)
334 devops.len = subdev->size - to;
336 err = subdev->write_oob(subdev, to, &devops);
337 ops->retlen += devops.oobretlen;
342 devops.len = ops->len - ops->retlen;
345 devops.datbuf += devops.retlen;
348 devops.ooblen = ops->ooblen - ops->oobretlen;
351 devops.oobbuf += devops.oobretlen;
358 static void concat_erase_callback(struct erase_info *instr)
360 wake_up((wait_queue_head_t *) instr->priv);
363 static int concat_dev_erase(struct mtd_info *mtd, struct erase_info *erase)
366 wait_queue_head_t waitq;
367 DECLARE_WAITQUEUE(wait, current);
370 * This code was stol^H^H^H^Hinspired by mtdchar.c
372 init_waitqueue_head(&waitq);
375 erase->callback = concat_erase_callback;
376 erase->priv = (unsigned long) &waitq;
379 * FIXME: Allow INTERRUPTIBLE. Which means
380 * not having the wait_queue head on the stack.
382 err = mtd->erase(mtd, erase);
384 set_current_state(TASK_UNINTERRUPTIBLE);
385 add_wait_queue(&waitq, &wait);
386 if (erase->state != MTD_ERASE_DONE
387 && erase->state != MTD_ERASE_FAILED)
389 remove_wait_queue(&waitq, &wait);
390 set_current_state(TASK_RUNNING);
392 err = (erase->state == MTD_ERASE_FAILED) ? -EIO : 0;
397 static int concat_erase(struct mtd_info *mtd, struct erase_info *instr)
399 struct mtd_concat *concat = CONCAT(mtd);
400 struct mtd_info *subdev;
402 uint64_t length, offset = 0;
403 struct erase_info *erase;
405 if (!(mtd->flags & MTD_WRITEABLE))
408 if (instr->addr > concat->mtd.size)
411 if (instr->len + instr->addr > concat->mtd.size)
415 * Check for proper erase block alignment of the to-be-erased area.
416 * It is easier to do this based on the super device's erase
417 * region info rather than looking at each particular sub-device
420 if (!concat->mtd.numeraseregions) {
421 /* the easy case: device has uniform erase block size */
422 if (instr->addr & (concat->mtd.erasesize - 1))
424 if (instr->len & (concat->mtd.erasesize - 1))
427 /* device has variable erase size */
428 struct mtd_erase_region_info *erase_regions =
429 concat->mtd.eraseregions;
432 * Find the erase region where the to-be-erased area begins:
434 for (i = 0; i < concat->mtd.numeraseregions &&
435 instr->addr >= erase_regions[i].offset; i++) ;
439 * Now erase_regions[i] is the region in which the
440 * to-be-erased area begins. Verify that the starting
441 * offset is aligned to this region's erase size:
443 if (i < 0 || instr->addr & (erase_regions[i].erasesize - 1))
447 * now find the erase region where the to-be-erased area ends:
449 for (; i < concat->mtd.numeraseregions &&
450 (instr->addr + instr->len) >= erase_regions[i].offset;
454 * check if the ending offset is aligned to this region's erase size
456 if (i < 0 || ((instr->addr + instr->len) &
457 (erase_regions[i].erasesize - 1)))
461 instr->fail_addr = MTD_FAIL_ADDR_UNKNOWN;
463 /* make a local copy of instr to avoid modifying the caller's struct */
464 erase = kmalloc(sizeof (struct erase_info), GFP_KERNEL);
473 * find the subdevice where the to-be-erased area begins, adjust
474 * starting offset to be relative to the subdevice start
476 for (i = 0; i < concat->num_subdev; i++) {
477 subdev = concat->subdev[i];
478 if (subdev->size <= erase->addr) {
479 erase->addr -= subdev->size;
480 offset += subdev->size;
486 /* must never happen since size limit has been verified above */
487 BUG_ON(i >= concat->num_subdev);
489 /* now do the erase: */
491 for (; length > 0; i++) {
492 /* loop for all subdevices affected by this request */
493 subdev = concat->subdev[i]; /* get current subdevice */
495 /* limit length to subdevice's size: */
496 if (erase->addr + length > subdev->size)
497 erase->len = subdev->size - erase->addr;
501 if (!(subdev->flags & MTD_WRITEABLE)) {
505 length -= erase->len;
506 if ((err = concat_dev_erase(subdev, erase))) {
507 /* sanity check: should never happen since
508 * block alignment has been checked above */
509 BUG_ON(err == -EINVAL);
510 if (erase->fail_addr != MTD_FAIL_ADDR_UNKNOWN)
511 instr->fail_addr = erase->fail_addr + offset;
515 * erase->addr specifies the offset of the area to be
516 * erased *within the current subdevice*. It can be
517 * non-zero only the first time through this loop, i.e.
518 * for the first subdevice where blocks need to be erased.
519 * All the following erases must begin at the start of the
520 * current subdevice, i.e. at offset zero.
523 offset += subdev->size;
525 instr->state = erase->state;
531 instr->callback(instr);
535 static int concat_lock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
537 struct mtd_concat *concat = CONCAT(mtd);
538 int i, err = -EINVAL;
540 if ((len + ofs) > mtd->size)
543 for (i = 0; i < concat->num_subdev; i++) {
544 struct mtd_info *subdev = concat->subdev[i];
547 if (ofs >= subdev->size) {
552 if (ofs + len > subdev->size)
553 size = subdev->size - ofs;
558 err = subdev->lock(subdev, ofs, size);
575 static int concat_unlock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
577 struct mtd_concat *concat = CONCAT(mtd);
580 if ((len + ofs) > mtd->size)
583 for (i = 0; i < concat->num_subdev; i++) {
584 struct mtd_info *subdev = concat->subdev[i];
587 if (ofs >= subdev->size) {
592 if (ofs + len > subdev->size)
593 size = subdev->size - ofs;
597 if (subdev->unlock) {
598 err = subdev->unlock(subdev, ofs, size);
615 static void concat_sync(struct mtd_info *mtd)
617 struct mtd_concat *concat = CONCAT(mtd);
620 for (i = 0; i < concat->num_subdev; i++) {
621 struct mtd_info *subdev = concat->subdev[i];
622 subdev->sync(subdev);
626 static int concat_suspend(struct mtd_info *mtd)
628 struct mtd_concat *concat = CONCAT(mtd);
631 for (i = 0; i < concat->num_subdev; i++) {
632 struct mtd_info *subdev = concat->subdev[i];
633 if ((rc = subdev->suspend(subdev)) < 0)
639 static void concat_resume(struct mtd_info *mtd)
641 struct mtd_concat *concat = CONCAT(mtd);
644 for (i = 0; i < concat->num_subdev; i++) {
645 struct mtd_info *subdev = concat->subdev[i];
646 subdev->resume(subdev);
650 static int concat_block_isbad(struct mtd_info *mtd, loff_t ofs)
652 struct mtd_concat *concat = CONCAT(mtd);
655 if (!concat->subdev[0]->block_isbad)
661 for (i = 0; i < concat->num_subdev; i++) {
662 struct mtd_info *subdev = concat->subdev[i];
664 if (ofs >= subdev->size) {
669 res = subdev->block_isbad(subdev, ofs);
676 static int concat_block_markbad(struct mtd_info *mtd, loff_t ofs)
678 struct mtd_concat *concat = CONCAT(mtd);
679 int i, err = -EINVAL;
681 if (!concat->subdev[0]->block_markbad)
687 for (i = 0; i < concat->num_subdev; i++) {
688 struct mtd_info *subdev = concat->subdev[i];
690 if (ofs >= subdev->size) {
695 err = subdev->block_markbad(subdev, ofs);
697 mtd->ecc_stats.badblocks++;
705 * try to support NOMMU mmaps on concatenated devices
706 * - we don't support subdev spanning as we can't guarantee it'll work
708 static unsigned long concat_get_unmapped_area(struct mtd_info *mtd,
710 unsigned long offset,
713 struct mtd_concat *concat = CONCAT(mtd);
716 for (i = 0; i < concat->num_subdev; i++) {
717 struct mtd_info *subdev = concat->subdev[i];
719 if (offset >= subdev->size) {
720 offset -= subdev->size;
724 /* we've found the subdev over which the mapping will reside */
725 if (offset + len > subdev->size)
726 return (unsigned long) -EINVAL;
728 if (subdev->get_unmapped_area)
729 return subdev->get_unmapped_area(subdev, len, offset,
735 return (unsigned long) -ENOSYS;
739 * This function constructs a virtual MTD device by concatenating
740 * num_devs MTD devices. A pointer to the new device object is
741 * stored to *new_dev upon success. This function does _not_
742 * register any devices: this is the caller's responsibility.
744 struct mtd_info *mtd_concat_create(struct mtd_info *subdev[], /* subdevices to concatenate */
745 int num_devs, /* number of subdevices */
747 { /* name for the new device */
750 struct mtd_concat *concat;
751 uint32_t max_erasesize, curr_erasesize;
752 int num_erase_region;
753 int max_writebufsize = 0;
755 printk(KERN_NOTICE "Concatenating MTD devices:\n");
756 for (i = 0; i < num_devs; i++)
757 printk(KERN_NOTICE "(%d): \"%s\"\n", i, subdev[i]->name);
758 printk(KERN_NOTICE "into device \"%s\"\n", name);
760 /* allocate the device structure */
761 size = SIZEOF_STRUCT_MTD_CONCAT(num_devs);
762 concat = kzalloc(size, GFP_KERNEL);
765 ("memory allocation error while creating concatenated device \"%s\"\n",
769 concat->subdev = (struct mtd_info **) (concat + 1);
772 * Set up the new "super" device's MTD object structure, check for
773 * incompatibilites between the subdevices.
775 concat->mtd.type = subdev[0]->type;
776 concat->mtd.flags = subdev[0]->flags;
777 concat->mtd.size = subdev[0]->size;
778 concat->mtd.erasesize = subdev[0]->erasesize;
779 concat->mtd.writesize = subdev[0]->writesize;
781 for (i = 0; i < num_devs; i++)
782 if (max_writebufsize < subdev[i]->writebufsize)
783 max_writebufsize = subdev[i]->writebufsize;
784 concat->mtd.writebufsize = max_writebufsize;
786 concat->mtd.subpage_sft = subdev[0]->subpage_sft;
787 concat->mtd.oobsize = subdev[0]->oobsize;
788 concat->mtd.oobavail = subdev[0]->oobavail;
789 if (subdev[0]->writev)
790 concat->mtd.writev = concat_writev;
791 if (subdev[0]->read_oob)
792 concat->mtd.read_oob = concat_read_oob;
793 if (subdev[0]->write_oob)
794 concat->mtd.write_oob = concat_write_oob;
795 if (subdev[0]->block_isbad)
796 concat->mtd.block_isbad = concat_block_isbad;
797 if (subdev[0]->block_markbad)
798 concat->mtd.block_markbad = concat_block_markbad;
800 concat->mtd.ecc_stats.badblocks = subdev[0]->ecc_stats.badblocks;
802 concat->mtd.backing_dev_info = subdev[0]->backing_dev_info;
804 concat->subdev[0] = subdev[0];
806 for (i = 1; i < num_devs; i++) {
807 if (concat->mtd.type != subdev[i]->type) {
809 printk("Incompatible device type on \"%s\"\n",
813 if (concat->mtd.flags != subdev[i]->flags) {
815 * Expect all flags except MTD_WRITEABLE to be
816 * equal on all subdevices.
818 if ((concat->mtd.flags ^ subdev[i]->
819 flags) & ~MTD_WRITEABLE) {
821 printk("Incompatible device flags on \"%s\"\n",
825 /* if writeable attribute differs,
826 make super device writeable */
828 subdev[i]->flags & MTD_WRITEABLE;
831 /* only permit direct mapping if the BDIs are all the same
832 * - copy-mapping is still permitted
834 if (concat->mtd.backing_dev_info !=
835 subdev[i]->backing_dev_info)
836 concat->mtd.backing_dev_info =
837 &default_backing_dev_info;
839 concat->mtd.size += subdev[i]->size;
840 concat->mtd.ecc_stats.badblocks +=
841 subdev[i]->ecc_stats.badblocks;
842 if (concat->mtd.writesize != subdev[i]->writesize ||
843 concat->mtd.subpage_sft != subdev[i]->subpage_sft ||
844 concat->mtd.oobsize != subdev[i]->oobsize ||
845 !concat->mtd.read_oob != !subdev[i]->read_oob ||
846 !concat->mtd.write_oob != !subdev[i]->write_oob) {
848 printk("Incompatible OOB or ECC data on \"%s\"\n",
852 concat->subdev[i] = subdev[i];
856 concat->mtd.ecclayout = subdev[0]->ecclayout;
858 concat->num_subdev = num_devs;
859 concat->mtd.name = name;
861 concat->mtd.erase = concat_erase;
862 concat->mtd.read = concat_read;
863 concat->mtd.write = concat_write;
864 concat->mtd.sync = concat_sync;
865 concat->mtd.lock = concat_lock;
866 concat->mtd.unlock = concat_unlock;
867 concat->mtd.suspend = concat_suspend;
868 concat->mtd.resume = concat_resume;
869 concat->mtd.get_unmapped_area = concat_get_unmapped_area;
872 * Combine the erase block size info of the subdevices:
874 * first, walk the map of the new device and see how
875 * many changes in erase size we have
877 max_erasesize = curr_erasesize = subdev[0]->erasesize;
878 num_erase_region = 1;
879 for (i = 0; i < num_devs; i++) {
880 if (subdev[i]->numeraseregions == 0) {
881 /* current subdevice has uniform erase size */
882 if (subdev[i]->erasesize != curr_erasesize) {
883 /* if it differs from the last subdevice's erase size, count it */
885 curr_erasesize = subdev[i]->erasesize;
886 if (curr_erasesize > max_erasesize)
887 max_erasesize = curr_erasesize;
890 /* current subdevice has variable erase size */
892 for (j = 0; j < subdev[i]->numeraseregions; j++) {
894 /* walk the list of erase regions, count any changes */
895 if (subdev[i]->eraseregions[j].erasesize !=
899 subdev[i]->eraseregions[j].
901 if (curr_erasesize > max_erasesize)
902 max_erasesize = curr_erasesize;
908 if (num_erase_region == 1) {
910 * All subdevices have the same uniform erase size.
913 concat->mtd.erasesize = curr_erasesize;
914 concat->mtd.numeraseregions = 0;
919 * erase block size varies across the subdevices: allocate
920 * space to store the data describing the variable erase regions
922 struct mtd_erase_region_info *erase_region_p;
923 uint64_t begin, position;
925 concat->mtd.erasesize = max_erasesize;
926 concat->mtd.numeraseregions = num_erase_region;
927 concat->mtd.eraseregions = erase_region_p =
928 kmalloc(num_erase_region *
929 sizeof (struct mtd_erase_region_info), GFP_KERNEL);
930 if (!erase_region_p) {
933 ("memory allocation error while creating erase region list"
934 " for device \"%s\"\n", name);
939 * walk the map of the new device once more and fill in
940 * in erase region info:
942 curr_erasesize = subdev[0]->erasesize;
943 begin = position = 0;
944 for (i = 0; i < num_devs; i++) {
945 if (subdev[i]->numeraseregions == 0) {
946 /* current subdevice has uniform erase size */
947 if (subdev[i]->erasesize != curr_erasesize) {
949 * fill in an mtd_erase_region_info structure for the area
950 * we have walked so far:
952 erase_region_p->offset = begin;
953 erase_region_p->erasesize =
955 tmp64 = position - begin;
956 do_div(tmp64, curr_erasesize);
957 erase_region_p->numblocks = tmp64;
960 curr_erasesize = subdev[i]->erasesize;
963 position += subdev[i]->size;
965 /* current subdevice has variable erase size */
967 for (j = 0; j < subdev[i]->numeraseregions; j++) {
968 /* walk the list of erase regions, count any changes */
969 if (subdev[i]->eraseregions[j].
970 erasesize != curr_erasesize) {
971 erase_region_p->offset = begin;
972 erase_region_p->erasesize =
974 tmp64 = position - begin;
975 do_div(tmp64, curr_erasesize);
976 erase_region_p->numblocks = tmp64;
980 subdev[i]->eraseregions[j].
985 subdev[i]->eraseregions[j].
986 numblocks * (uint64_t)curr_erasesize;
990 /* Now write the final entry */
991 erase_region_p->offset = begin;
992 erase_region_p->erasesize = curr_erasesize;
993 tmp64 = position - begin;
994 do_div(tmp64, curr_erasesize);
995 erase_region_p->numblocks = tmp64;
1002 * This function destroys an MTD object obtained from concat_mtd_devs()
1005 void mtd_concat_destroy(struct mtd_info *mtd)
1007 struct mtd_concat *concat = CONCAT(mtd);
1008 if (concat->mtd.numeraseregions)
1009 kfree(concat->mtd.eraseregions);
1013 EXPORT_SYMBOL(mtd_concat_create);
1014 EXPORT_SYMBOL(mtd_concat_destroy);
1016 MODULE_LICENSE("GPL");
1017 MODULE_AUTHOR("Robert Kaiser <rkaiser@sysgo.de>");
1018 MODULE_DESCRIPTION("Generic support for concatenating of MTD devices");