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.retlen;
315 ops->oobretlen += devops.oobretlen;
320 devops.len = ops->len - ops->retlen;
323 devops.datbuf += devops.retlen;
326 devops.ooblen = ops->ooblen - ops->oobretlen;
329 devops.oobbuf += devops.oobretlen;
336 static void concat_erase_callback(struct erase_info *instr)
338 wake_up((wait_queue_head_t *) instr->priv);
341 static int concat_dev_erase(struct mtd_info *mtd, struct erase_info *erase)
344 wait_queue_head_t waitq;
345 DECLARE_WAITQUEUE(wait, current);
348 * This code was stol^H^H^H^Hinspired by mtdchar.c
350 init_waitqueue_head(&waitq);
353 erase->callback = concat_erase_callback;
354 erase->priv = (unsigned long) &waitq;
357 * FIXME: Allow INTERRUPTIBLE. Which means
358 * not having the wait_queue head on the stack.
360 err = mtd_erase(mtd, erase);
362 set_current_state(TASK_UNINTERRUPTIBLE);
363 add_wait_queue(&waitq, &wait);
364 if (erase->state != MTD_ERASE_DONE
365 && erase->state != MTD_ERASE_FAILED)
367 remove_wait_queue(&waitq, &wait);
368 set_current_state(TASK_RUNNING);
370 err = (erase->state == MTD_ERASE_FAILED) ? -EIO : 0;
375 static int concat_erase(struct mtd_info *mtd, struct erase_info *instr)
377 struct mtd_concat *concat = CONCAT(mtd);
378 struct mtd_info *subdev;
380 uint64_t length, offset = 0;
381 struct erase_info *erase;
384 * Check for proper erase block alignment of the to-be-erased area.
385 * It is easier to do this based on the super device's erase
386 * region info rather than looking at each particular sub-device
389 if (!concat->mtd.numeraseregions) {
390 /* the easy case: device has uniform erase block size */
391 if (instr->addr & (concat->mtd.erasesize - 1))
393 if (instr->len & (concat->mtd.erasesize - 1))
396 /* device has variable erase size */
397 struct mtd_erase_region_info *erase_regions =
398 concat->mtd.eraseregions;
401 * Find the erase region where the to-be-erased area begins:
403 for (i = 0; i < concat->mtd.numeraseregions &&
404 instr->addr >= erase_regions[i].offset; i++) ;
408 * Now erase_regions[i] is the region in which the
409 * to-be-erased area begins. Verify that the starting
410 * offset is aligned to this region's erase size:
412 if (i < 0 || instr->addr & (erase_regions[i].erasesize - 1))
416 * now find the erase region where the to-be-erased area ends:
418 for (; i < concat->mtd.numeraseregions &&
419 (instr->addr + instr->len) >= erase_regions[i].offset;
423 * check if the ending offset is aligned to this region's erase size
425 if (i < 0 || ((instr->addr + instr->len) &
426 (erase_regions[i].erasesize - 1)))
430 /* make a local copy of instr to avoid modifying the caller's struct */
431 erase = kmalloc(sizeof (struct erase_info), GFP_KERNEL);
440 * find the subdevice where the to-be-erased area begins, adjust
441 * starting offset to be relative to the subdevice start
443 for (i = 0; i < concat->num_subdev; i++) {
444 subdev = concat->subdev[i];
445 if (subdev->size <= erase->addr) {
446 erase->addr -= subdev->size;
447 offset += subdev->size;
453 /* must never happen since size limit has been verified above */
454 BUG_ON(i >= concat->num_subdev);
456 /* now do the erase: */
458 for (; length > 0; i++) {
459 /* loop for all subdevices affected by this request */
460 subdev = concat->subdev[i]; /* get current subdevice */
462 /* limit length to subdevice's size: */
463 if (erase->addr + length > subdev->size)
464 erase->len = subdev->size - erase->addr;
468 length -= erase->len;
469 if ((err = concat_dev_erase(subdev, erase))) {
470 /* sanity check: should never happen since
471 * block alignment has been checked above */
472 BUG_ON(err == -EINVAL);
473 if (erase->fail_addr != MTD_FAIL_ADDR_UNKNOWN)
474 instr->fail_addr = erase->fail_addr + offset;
478 * erase->addr specifies the offset of the area to be
479 * erased *within the current subdevice*. It can be
480 * non-zero only the first time through this loop, i.e.
481 * for the first subdevice where blocks need to be erased.
482 * All the following erases must begin at the start of the
483 * current subdevice, i.e. at offset zero.
486 offset += subdev->size;
488 instr->state = erase->state;
494 instr->callback(instr);
498 static int concat_lock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
500 struct mtd_concat *concat = CONCAT(mtd);
501 int i, err = -EINVAL;
503 for (i = 0; i < concat->num_subdev; i++) {
504 struct mtd_info *subdev = concat->subdev[i];
507 if (ofs >= subdev->size) {
512 if (ofs + len > subdev->size)
513 size = subdev->size - ofs;
517 err = mtd_lock(subdev, ofs, size);
532 static int concat_unlock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
534 struct mtd_concat *concat = CONCAT(mtd);
537 for (i = 0; i < concat->num_subdev; i++) {
538 struct mtd_info *subdev = concat->subdev[i];
541 if (ofs >= subdev->size) {
546 if (ofs + len > subdev->size)
547 size = subdev->size - ofs;
551 err = mtd_unlock(subdev, ofs, size);
566 static void concat_sync(struct mtd_info *mtd)
568 struct mtd_concat *concat = CONCAT(mtd);
571 for (i = 0; i < concat->num_subdev; i++) {
572 struct mtd_info *subdev = concat->subdev[i];
577 static int concat_suspend(struct mtd_info *mtd)
579 struct mtd_concat *concat = CONCAT(mtd);
582 for (i = 0; i < concat->num_subdev; i++) {
583 struct mtd_info *subdev = concat->subdev[i];
584 if ((rc = mtd_suspend(subdev)) < 0)
590 static void concat_resume(struct mtd_info *mtd)
592 struct mtd_concat *concat = CONCAT(mtd);
595 for (i = 0; i < concat->num_subdev; i++) {
596 struct mtd_info *subdev = concat->subdev[i];
601 static int concat_block_isbad(struct mtd_info *mtd, loff_t ofs)
603 struct mtd_concat *concat = CONCAT(mtd);
606 if (!mtd_can_have_bb(concat->subdev[0]))
609 for (i = 0; i < concat->num_subdev; i++) {
610 struct mtd_info *subdev = concat->subdev[i];
612 if (ofs >= subdev->size) {
617 res = mtd_block_isbad(subdev, ofs);
624 static int concat_block_markbad(struct mtd_info *mtd, loff_t ofs)
626 struct mtd_concat *concat = CONCAT(mtd);
627 int i, err = -EINVAL;
629 for (i = 0; i < concat->num_subdev; i++) {
630 struct mtd_info *subdev = concat->subdev[i];
632 if (ofs >= subdev->size) {
637 err = mtd_block_markbad(subdev, ofs);
639 mtd->ecc_stats.badblocks++;
647 * This function constructs a virtual MTD device by concatenating
648 * num_devs MTD devices. A pointer to the new device object is
649 * stored to *new_dev upon success. This function does _not_
650 * register any devices: this is the caller's responsibility.
652 struct mtd_info *mtd_concat_create(struct mtd_info *subdev[], /* subdevices to concatenate */
653 int num_devs, /* number of subdevices */
655 { /* name for the new device */
658 struct mtd_concat *concat;
659 uint32_t max_erasesize, curr_erasesize;
660 int num_erase_region;
661 int max_writebufsize = 0;
663 printk(KERN_NOTICE "Concatenating MTD devices:\n");
664 for (i = 0; i < num_devs; i++)
665 printk(KERN_NOTICE "(%d): \"%s\"\n", i, subdev[i]->name);
666 printk(KERN_NOTICE "into device \"%s\"\n", name);
668 /* allocate the device structure */
669 size = SIZEOF_STRUCT_MTD_CONCAT(num_devs);
670 concat = kzalloc(size, GFP_KERNEL);
673 ("memory allocation error while creating concatenated device \"%s\"\n",
677 concat->subdev = (struct mtd_info **) (concat + 1);
680 * Set up the new "super" device's MTD object structure, check for
681 * incompatibilities between the subdevices.
683 concat->mtd.type = subdev[0]->type;
684 concat->mtd.flags = subdev[0]->flags;
685 concat->mtd.size = subdev[0]->size;
686 concat->mtd.erasesize = subdev[0]->erasesize;
687 concat->mtd.writesize = subdev[0]->writesize;
689 for (i = 0; i < num_devs; i++)
690 if (max_writebufsize < subdev[i]->writebufsize)
691 max_writebufsize = subdev[i]->writebufsize;
692 concat->mtd.writebufsize = max_writebufsize;
694 concat->mtd.subpage_sft = subdev[0]->subpage_sft;
695 concat->mtd.oobsize = subdev[0]->oobsize;
696 concat->mtd.oobavail = subdev[0]->oobavail;
697 if (subdev[0]->_writev)
698 concat->mtd._writev = concat_writev;
699 if (subdev[0]->_read_oob)
700 concat->mtd._read_oob = concat_read_oob;
701 if (subdev[0]->_write_oob)
702 concat->mtd._write_oob = concat_write_oob;
703 if (subdev[0]->_block_isbad)
704 concat->mtd._block_isbad = concat_block_isbad;
705 if (subdev[0]->_block_markbad)
706 concat->mtd._block_markbad = concat_block_markbad;
708 concat->mtd.ecc_stats.badblocks = subdev[0]->ecc_stats.badblocks;
710 concat->subdev[0] = subdev[0];
712 for (i = 1; i < num_devs; i++) {
713 if (concat->mtd.type != subdev[i]->type) {
715 printk("Incompatible device type on \"%s\"\n",
719 if (concat->mtd.flags != subdev[i]->flags) {
721 * Expect all flags except MTD_WRITEABLE to be
722 * equal on all subdevices.
724 if ((concat->mtd.flags ^ subdev[i]->
725 flags) & ~MTD_WRITEABLE) {
727 printk("Incompatible device flags on \"%s\"\n",
731 /* if writeable attribute differs,
732 make super device writeable */
734 subdev[i]->flags & MTD_WRITEABLE;
737 concat->mtd.size += subdev[i]->size;
738 concat->mtd.ecc_stats.badblocks +=
739 subdev[i]->ecc_stats.badblocks;
740 if (concat->mtd.writesize != subdev[i]->writesize ||
741 concat->mtd.subpage_sft != subdev[i]->subpage_sft ||
742 concat->mtd.oobsize != subdev[i]->oobsize ||
743 !concat->mtd._read_oob != !subdev[i]->_read_oob ||
744 !concat->mtd._write_oob != !subdev[i]->_write_oob) {
746 printk("Incompatible OOB or ECC data on \"%s\"\n",
750 concat->subdev[i] = subdev[i];
754 mtd_set_ooblayout(&concat->mtd, subdev[0]->ooblayout);
756 concat->num_subdev = num_devs;
757 concat->mtd.name = name;
759 concat->mtd._erase = concat_erase;
760 concat->mtd._read = concat_read;
761 concat->mtd._write = concat_write;
762 concat->mtd._sync = concat_sync;
763 concat->mtd._lock = concat_lock;
764 concat->mtd._unlock = concat_unlock;
765 concat->mtd._suspend = concat_suspend;
766 concat->mtd._resume = concat_resume;
769 * Combine the erase block size info of the subdevices:
771 * first, walk the map of the new device and see how
772 * many changes in erase size we have
774 max_erasesize = curr_erasesize = subdev[0]->erasesize;
775 num_erase_region = 1;
776 for (i = 0; i < num_devs; i++) {
777 if (subdev[i]->numeraseregions == 0) {
778 /* current subdevice has uniform erase size */
779 if (subdev[i]->erasesize != curr_erasesize) {
780 /* if it differs from the last subdevice's erase size, count it */
782 curr_erasesize = subdev[i]->erasesize;
783 if (curr_erasesize > max_erasesize)
784 max_erasesize = curr_erasesize;
787 /* current subdevice has variable erase size */
789 for (j = 0; j < subdev[i]->numeraseregions; j++) {
791 /* walk the list of erase regions, count any changes */
792 if (subdev[i]->eraseregions[j].erasesize !=
796 subdev[i]->eraseregions[j].
798 if (curr_erasesize > max_erasesize)
799 max_erasesize = curr_erasesize;
805 if (num_erase_region == 1) {
807 * All subdevices have the same uniform erase size.
810 concat->mtd.erasesize = curr_erasesize;
811 concat->mtd.numeraseregions = 0;
816 * erase block size varies across the subdevices: allocate
817 * space to store the data describing the variable erase regions
819 struct mtd_erase_region_info *erase_region_p;
820 uint64_t begin, position;
822 concat->mtd.erasesize = max_erasesize;
823 concat->mtd.numeraseregions = num_erase_region;
824 concat->mtd.eraseregions = erase_region_p =
825 kmalloc(num_erase_region *
826 sizeof (struct mtd_erase_region_info), GFP_KERNEL);
827 if (!erase_region_p) {
830 ("memory allocation error while creating erase region list"
831 " for device \"%s\"\n", name);
836 * walk the map of the new device once more and fill in
837 * in erase region info:
839 curr_erasesize = subdev[0]->erasesize;
840 begin = position = 0;
841 for (i = 0; i < num_devs; i++) {
842 if (subdev[i]->numeraseregions == 0) {
843 /* current subdevice has uniform erase size */
844 if (subdev[i]->erasesize != curr_erasesize) {
846 * fill in an mtd_erase_region_info structure for the area
847 * we have walked so far:
849 erase_region_p->offset = begin;
850 erase_region_p->erasesize =
852 tmp64 = position - begin;
853 do_div(tmp64, curr_erasesize);
854 erase_region_p->numblocks = tmp64;
857 curr_erasesize = subdev[i]->erasesize;
860 position += subdev[i]->size;
862 /* current subdevice has variable erase size */
864 for (j = 0; j < subdev[i]->numeraseregions; j++) {
865 /* walk the list of erase regions, count any changes */
866 if (subdev[i]->eraseregions[j].
867 erasesize != curr_erasesize) {
868 erase_region_p->offset = begin;
869 erase_region_p->erasesize =
871 tmp64 = position - begin;
872 do_div(tmp64, curr_erasesize);
873 erase_region_p->numblocks = tmp64;
877 subdev[i]->eraseregions[j].
882 subdev[i]->eraseregions[j].
883 numblocks * (uint64_t)curr_erasesize;
887 /* Now write the final entry */
888 erase_region_p->offset = begin;
889 erase_region_p->erasesize = curr_erasesize;
890 tmp64 = position - begin;
891 do_div(tmp64, curr_erasesize);
892 erase_region_p->numblocks = tmp64;
899 * This function destroys an MTD object obtained from concat_mtd_devs()
902 void mtd_concat_destroy(struct mtd_info *mtd)
904 struct mtd_concat *concat = CONCAT(mtd);
905 if (concat->mtd.numeraseregions)
906 kfree(concat->mtd.eraseregions);
910 EXPORT_SYMBOL(mtd_concat_create);
911 EXPORT_SYMBOL(mtd_concat_destroy);
913 MODULE_LICENSE("GPL");
914 MODULE_AUTHOR("Robert Kaiser <rkaiser@sysgo.de>");
915 MODULE_DESCRIPTION("Generic support for concatenating of MTD devices");