2 * MTD device concatenation layer
4 * (C) 2002 Robert Kaiser <rkaiser@sysgo.de>
6 * NAND support by Christian Gan <cgan@iders.ca>
11 #include <linux/kernel.h>
12 #include <linux/module.h>
13 #include <linux/slab.h>
14 #include <linux/sched.h>
15 #include <linux/types.h>
16 #include <linux/backing-dev.h>
18 #include <linux/mtd/mtd.h>
19 #include <linux/mtd/concat.h>
21 #include <asm/div64.h>
24 * Our storage structure:
25 * Subdev points to an array of pointers to struct mtd_info objects
26 * which is allocated along with this structure
32 struct mtd_info **subdev;
36 * how to calculate the size required for the above structure,
37 * including the pointer array subdev points to:
39 #define SIZEOF_STRUCT_MTD_CONCAT(num_subdev) \
40 ((sizeof(struct mtd_concat) + (num_subdev) * sizeof(struct mtd_info *)))
43 * Given a pointer to the MTD object in the mtd_concat structure,
44 * we can retrieve the pointer to that structure with this macro.
46 #define CONCAT(x) ((struct mtd_concat *)(x))
49 * MTD methods which look up the relevant subdevice, translate the
50 * effective address and pass through to the subdevice.
54 concat_read(struct mtd_info *mtd, loff_t from, size_t len,
55 size_t * retlen, u_char * buf)
57 struct mtd_concat *concat = CONCAT(mtd);
63 for (i = 0; i < concat->num_subdev; i++) {
64 struct mtd_info *subdev = concat->subdev[i];
67 if (from >= subdev->size) {
68 /* Not destined for this subdev */
73 if (from + len > subdev->size)
74 /* First part goes into this subdev */
75 size = subdev->size - from;
77 /* Entire transaction goes into this subdev */
80 err = subdev->read(subdev, from, size, &retsize, buf);
82 /* Save information about bitflips! */
84 if (err == -EBADMSG) {
85 mtd->ecc_stats.failed++;
87 } else if (err == -EUCLEAN) {
88 mtd->ecc_stats.corrected++;
89 /* Do not overwrite -EBADMSG !! */
108 concat_write(struct mtd_info *mtd, loff_t to, size_t len,
109 size_t * retlen, const u_char * buf)
111 struct mtd_concat *concat = CONCAT(mtd);
115 if (!(mtd->flags & MTD_WRITEABLE))
120 for (i = 0; i < concat->num_subdev; i++) {
121 struct mtd_info *subdev = concat->subdev[i];
122 size_t size, retsize;
124 if (to >= subdev->size) {
129 if (to + len > subdev->size)
130 size = subdev->size - to;
134 if (!(subdev->flags & MTD_WRITEABLE))
137 err = subdev->write(subdev, to, size, &retsize, buf);
155 concat_writev(struct mtd_info *mtd, const struct kvec *vecs,
156 unsigned long count, loff_t to, size_t * retlen)
158 struct mtd_concat *concat = CONCAT(mtd);
159 struct kvec *vecs_copy;
160 unsigned long entry_low, entry_high;
161 size_t total_len = 0;
165 if (!(mtd->flags & MTD_WRITEABLE))
170 /* Calculate total length of data */
171 for (i = 0; i < count; i++)
172 total_len += vecs[i].iov_len;
174 /* Do not allow write past end of device */
175 if ((to + total_len) > mtd->size)
178 /* Check alignment */
179 if (mtd->writesize > 1) {
181 if (do_div(__to, mtd->writesize) || (total_len % mtd->writesize))
185 /* make a copy of vecs */
186 vecs_copy = kmemdup(vecs, sizeof(struct kvec) * count, GFP_KERNEL);
191 for (i = 0; i < concat->num_subdev; i++) {
192 struct mtd_info *subdev = concat->subdev[i];
193 size_t size, wsize, retsize, old_iov_len;
195 if (to >= subdev->size) {
200 size = min_t(uint64_t, total_len, subdev->size - to);
201 wsize = size; /* store for future use */
203 entry_high = entry_low;
204 while (entry_high < count) {
205 if (size <= vecs_copy[entry_high].iov_len)
207 size -= vecs_copy[entry_high++].iov_len;
210 old_iov_len = vecs_copy[entry_high].iov_len;
211 vecs_copy[entry_high].iov_len = size;
213 if (!(subdev->flags & MTD_WRITEABLE))
216 err = subdev->writev(subdev, &vecs_copy[entry_low],
217 entry_high - entry_low + 1, to, &retsize);
219 vecs_copy[entry_high].iov_len = old_iov_len - size;
220 vecs_copy[entry_high].iov_base += size;
222 entry_low = entry_high;
242 concat_read_oob(struct mtd_info *mtd, loff_t from, struct mtd_oob_ops *ops)
244 struct mtd_concat *concat = CONCAT(mtd);
245 struct mtd_oob_ops devops = *ops;
248 ops->retlen = ops->oobretlen = 0;
250 for (i = 0; i < concat->num_subdev; i++) {
251 struct mtd_info *subdev = concat->subdev[i];
253 if (from >= subdev->size) {
254 from -= subdev->size;
259 if (from + devops.len > subdev->size)
260 devops.len = subdev->size - from;
262 err = subdev->read_oob(subdev, from, &devops);
263 ops->retlen += devops.retlen;
264 ops->oobretlen += devops.oobretlen;
266 /* Save information about bitflips! */
268 if (err == -EBADMSG) {
269 mtd->ecc_stats.failed++;
271 } else if (err == -EUCLEAN) {
272 mtd->ecc_stats.corrected++;
273 /* Do not overwrite -EBADMSG !! */
281 devops.len = ops->len - ops->retlen;
284 devops.datbuf += devops.retlen;
287 devops.ooblen = ops->ooblen - ops->oobretlen;
290 devops.oobbuf += ops->oobretlen;
299 concat_write_oob(struct mtd_info *mtd, loff_t to, struct mtd_oob_ops *ops)
301 struct mtd_concat *concat = CONCAT(mtd);
302 struct mtd_oob_ops devops = *ops;
305 if (!(mtd->flags & MTD_WRITEABLE))
310 for (i = 0; i < concat->num_subdev; i++) {
311 struct mtd_info *subdev = concat->subdev[i];
313 if (to >= subdev->size) {
318 /* partial write ? */
319 if (to + devops.len > subdev->size)
320 devops.len = subdev->size - to;
322 err = subdev->write_oob(subdev, to, &devops);
323 ops->retlen += devops.retlen;
328 devops.len = ops->len - ops->retlen;
331 devops.datbuf += devops.retlen;
334 devops.ooblen = ops->ooblen - ops->oobretlen;
337 devops.oobbuf += devops.oobretlen;
344 static void concat_erase_callback(struct erase_info *instr)
346 wake_up((wait_queue_head_t *) instr->priv);
349 static int concat_dev_erase(struct mtd_info *mtd, struct erase_info *erase)
352 wait_queue_head_t waitq;
353 DECLARE_WAITQUEUE(wait, current);
356 * This code was stol^H^H^H^Hinspired by mtdchar.c
358 init_waitqueue_head(&waitq);
361 erase->callback = concat_erase_callback;
362 erase->priv = (unsigned long) &waitq;
365 * FIXME: Allow INTERRUPTIBLE. Which means
366 * not having the wait_queue head on the stack.
368 err = mtd->erase(mtd, erase);
370 set_current_state(TASK_UNINTERRUPTIBLE);
371 add_wait_queue(&waitq, &wait);
372 if (erase->state != MTD_ERASE_DONE
373 && erase->state != MTD_ERASE_FAILED)
375 remove_wait_queue(&waitq, &wait);
376 set_current_state(TASK_RUNNING);
378 err = (erase->state == MTD_ERASE_FAILED) ? -EIO : 0;
383 static int concat_erase(struct mtd_info *mtd, struct erase_info *instr)
385 struct mtd_concat *concat = CONCAT(mtd);
386 struct mtd_info *subdev;
388 uint64_t length, offset = 0;
389 struct erase_info *erase;
391 if (!(mtd->flags & MTD_WRITEABLE))
394 if (instr->addr > concat->mtd.size)
397 if (instr->len + instr->addr > concat->mtd.size)
401 * Check for proper erase block alignment of the to-be-erased area.
402 * It is easier to do this based on the super device's erase
403 * region info rather than looking at each particular sub-device
406 if (!concat->mtd.numeraseregions) {
407 /* the easy case: device has uniform erase block size */
408 if (instr->addr & (concat->mtd.erasesize - 1))
410 if (instr->len & (concat->mtd.erasesize - 1))
413 /* device has variable erase size */
414 struct mtd_erase_region_info *erase_regions =
415 concat->mtd.eraseregions;
418 * Find the erase region where the to-be-erased area begins:
420 for (i = 0; i < concat->mtd.numeraseregions &&
421 instr->addr >= erase_regions[i].offset; i++) ;
425 * Now erase_regions[i] is the region in which the
426 * to-be-erased area begins. Verify that the starting
427 * offset is aligned to this region's erase size:
429 if (i < 0 || instr->addr & (erase_regions[i].erasesize - 1))
433 * now find the erase region where the to-be-erased area ends:
435 for (; i < concat->mtd.numeraseregions &&
436 (instr->addr + instr->len) >= erase_regions[i].offset;
440 * check if the ending offset is aligned to this region's erase size
442 if (i < 0 || ((instr->addr + instr->len) &
443 (erase_regions[i].erasesize - 1)))
447 instr->fail_addr = MTD_FAIL_ADDR_UNKNOWN;
449 /* make a local copy of instr to avoid modifying the caller's struct */
450 erase = kmalloc(sizeof (struct erase_info), GFP_KERNEL);
459 * find the subdevice where the to-be-erased area begins, adjust
460 * starting offset to be relative to the subdevice start
462 for (i = 0; i < concat->num_subdev; i++) {
463 subdev = concat->subdev[i];
464 if (subdev->size <= erase->addr) {
465 erase->addr -= subdev->size;
466 offset += subdev->size;
472 /* must never happen since size limit has been verified above */
473 BUG_ON(i >= concat->num_subdev);
475 /* now do the erase: */
477 for (; length > 0; i++) {
478 /* loop for all subdevices affected by this request */
479 subdev = concat->subdev[i]; /* get current subdevice */
481 /* limit length to subdevice's size: */
482 if (erase->addr + length > subdev->size)
483 erase->len = subdev->size - erase->addr;
487 if (!(subdev->flags & MTD_WRITEABLE)) {
491 length -= erase->len;
492 if ((err = concat_dev_erase(subdev, erase))) {
493 /* sanity check: should never happen since
494 * block alignment has been checked above */
495 BUG_ON(err == -EINVAL);
496 if (erase->fail_addr != MTD_FAIL_ADDR_UNKNOWN)
497 instr->fail_addr = erase->fail_addr + offset;
501 * erase->addr specifies the offset of the area to be
502 * erased *within the current subdevice*. It can be
503 * non-zero only the first time through this loop, i.e.
504 * for the first subdevice where blocks need to be erased.
505 * All the following erases must begin at the start of the
506 * current subdevice, i.e. at offset zero.
509 offset += subdev->size;
511 instr->state = erase->state;
517 instr->callback(instr);
521 static int concat_lock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
523 struct mtd_concat *concat = CONCAT(mtd);
524 int i, err = -EINVAL;
526 if ((len + ofs) > mtd->size)
529 for (i = 0; i < concat->num_subdev; i++) {
530 struct mtd_info *subdev = concat->subdev[i];
533 if (ofs >= subdev->size) {
538 if (ofs + len > subdev->size)
539 size = subdev->size - ofs;
544 err = subdev->lock(subdev, ofs, size);
561 static int concat_unlock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
563 struct mtd_concat *concat = CONCAT(mtd);
566 if ((len + ofs) > mtd->size)
569 for (i = 0; i < concat->num_subdev; i++) {
570 struct mtd_info *subdev = concat->subdev[i];
573 if (ofs >= subdev->size) {
578 if (ofs + len > subdev->size)
579 size = subdev->size - ofs;
583 if (subdev->unlock) {
584 err = subdev->unlock(subdev, ofs, size);
601 static void concat_sync(struct mtd_info *mtd)
603 struct mtd_concat *concat = CONCAT(mtd);
606 for (i = 0; i < concat->num_subdev; i++) {
607 struct mtd_info *subdev = concat->subdev[i];
608 subdev->sync(subdev);
612 static int concat_suspend(struct mtd_info *mtd)
614 struct mtd_concat *concat = CONCAT(mtd);
617 for (i = 0; i < concat->num_subdev; i++) {
618 struct mtd_info *subdev = concat->subdev[i];
619 if ((rc = subdev->suspend(subdev)) < 0)
625 static void concat_resume(struct mtd_info *mtd)
627 struct mtd_concat *concat = CONCAT(mtd);
630 for (i = 0; i < concat->num_subdev; i++) {
631 struct mtd_info *subdev = concat->subdev[i];
632 subdev->resume(subdev);
636 static int concat_block_isbad(struct mtd_info *mtd, loff_t ofs)
638 struct mtd_concat *concat = CONCAT(mtd);
641 if (!concat->subdev[0]->block_isbad)
647 for (i = 0; i < concat->num_subdev; i++) {
648 struct mtd_info *subdev = concat->subdev[i];
650 if (ofs >= subdev->size) {
655 res = subdev->block_isbad(subdev, ofs);
662 static int concat_block_markbad(struct mtd_info *mtd, loff_t ofs)
664 struct mtd_concat *concat = CONCAT(mtd);
665 int i, err = -EINVAL;
667 if (!concat->subdev[0]->block_markbad)
673 for (i = 0; i < concat->num_subdev; i++) {
674 struct mtd_info *subdev = concat->subdev[i];
676 if (ofs >= subdev->size) {
681 err = subdev->block_markbad(subdev, ofs);
683 mtd->ecc_stats.badblocks++;
691 * try to support NOMMU mmaps on concatenated devices
692 * - we don't support subdev spanning as we can't guarantee it'll work
694 static unsigned long concat_get_unmapped_area(struct mtd_info *mtd,
696 unsigned long offset,
699 struct mtd_concat *concat = CONCAT(mtd);
702 for (i = 0; i < concat->num_subdev; i++) {
703 struct mtd_info *subdev = concat->subdev[i];
705 if (offset >= subdev->size) {
706 offset -= subdev->size;
710 /* we've found the subdev over which the mapping will reside */
711 if (offset + len > subdev->size)
712 return (unsigned long) -EINVAL;
714 if (subdev->get_unmapped_area)
715 return subdev->get_unmapped_area(subdev, len, offset,
721 return (unsigned long) -ENOSYS;
725 * This function constructs a virtual MTD device by concatenating
726 * num_devs MTD devices. A pointer to the new device object is
727 * stored to *new_dev upon success. This function does _not_
728 * register any devices: this is the caller's responsibility.
730 struct mtd_info *mtd_concat_create(struct mtd_info *subdev[], /* subdevices to concatenate */
731 int num_devs, /* number of subdevices */
733 { /* name for the new device */
736 struct mtd_concat *concat;
737 uint32_t max_erasesize, curr_erasesize;
738 int num_erase_region;
740 printk(KERN_NOTICE "Concatenating MTD devices:\n");
741 for (i = 0; i < num_devs; i++)
742 printk(KERN_NOTICE "(%d): \"%s\"\n", i, subdev[i]->name);
743 printk(KERN_NOTICE "into device \"%s\"\n", name);
745 /* allocate the device structure */
746 size = SIZEOF_STRUCT_MTD_CONCAT(num_devs);
747 concat = kzalloc(size, GFP_KERNEL);
750 ("memory allocation error while creating concatenated device \"%s\"\n",
754 concat->subdev = (struct mtd_info **) (concat + 1);
757 * Set up the new "super" device's MTD object structure, check for
758 * incompatibilites between the subdevices.
760 concat->mtd.type = subdev[0]->type;
761 concat->mtd.flags = subdev[0]->flags;
762 concat->mtd.size = subdev[0]->size;
763 concat->mtd.erasesize = subdev[0]->erasesize;
764 concat->mtd.writesize = subdev[0]->writesize;
765 concat->mtd.subpage_sft = subdev[0]->subpage_sft;
766 concat->mtd.oobsize = subdev[0]->oobsize;
767 concat->mtd.oobavail = subdev[0]->oobavail;
768 if (subdev[0]->writev)
769 concat->mtd.writev = concat_writev;
770 if (subdev[0]->read_oob)
771 concat->mtd.read_oob = concat_read_oob;
772 if (subdev[0]->write_oob)
773 concat->mtd.write_oob = concat_write_oob;
774 if (subdev[0]->block_isbad)
775 concat->mtd.block_isbad = concat_block_isbad;
776 if (subdev[0]->block_markbad)
777 concat->mtd.block_markbad = concat_block_markbad;
779 concat->mtd.ecc_stats.badblocks = subdev[0]->ecc_stats.badblocks;
781 concat->mtd.backing_dev_info = subdev[0]->backing_dev_info;
783 concat->subdev[0] = subdev[0];
785 for (i = 1; i < num_devs; i++) {
786 if (concat->mtd.type != subdev[i]->type) {
788 printk("Incompatible device type on \"%s\"\n",
792 if (concat->mtd.flags != subdev[i]->flags) {
794 * Expect all flags except MTD_WRITEABLE to be
795 * equal on all subdevices.
797 if ((concat->mtd.flags ^ subdev[i]->
798 flags) & ~MTD_WRITEABLE) {
800 printk("Incompatible device flags on \"%s\"\n",
804 /* if writeable attribute differs,
805 make super device writeable */
807 subdev[i]->flags & MTD_WRITEABLE;
810 /* only permit direct mapping if the BDIs are all the same
811 * - copy-mapping is still permitted
813 if (concat->mtd.backing_dev_info !=
814 subdev[i]->backing_dev_info)
815 concat->mtd.backing_dev_info =
816 &default_backing_dev_info;
818 concat->mtd.size += subdev[i]->size;
819 concat->mtd.ecc_stats.badblocks +=
820 subdev[i]->ecc_stats.badblocks;
821 if (concat->mtd.writesize != subdev[i]->writesize ||
822 concat->mtd.subpage_sft != subdev[i]->subpage_sft ||
823 concat->mtd.oobsize != subdev[i]->oobsize ||
824 !concat->mtd.read_oob != !subdev[i]->read_oob ||
825 !concat->mtd.write_oob != !subdev[i]->write_oob) {
827 printk("Incompatible OOB or ECC data on \"%s\"\n",
831 concat->subdev[i] = subdev[i];
835 concat->mtd.ecclayout = subdev[0]->ecclayout;
837 concat->num_subdev = num_devs;
838 concat->mtd.name = name;
840 concat->mtd.erase = concat_erase;
841 concat->mtd.read = concat_read;
842 concat->mtd.write = concat_write;
843 concat->mtd.sync = concat_sync;
844 concat->mtd.lock = concat_lock;
845 concat->mtd.unlock = concat_unlock;
846 concat->mtd.suspend = concat_suspend;
847 concat->mtd.resume = concat_resume;
848 concat->mtd.get_unmapped_area = concat_get_unmapped_area;
851 * Combine the erase block size info of the subdevices:
853 * first, walk the map of the new device and see how
854 * many changes in erase size we have
856 max_erasesize = curr_erasesize = subdev[0]->erasesize;
857 num_erase_region = 1;
858 for (i = 0; i < num_devs; i++) {
859 if (subdev[i]->numeraseregions == 0) {
860 /* current subdevice has uniform erase size */
861 if (subdev[i]->erasesize != curr_erasesize) {
862 /* if it differs from the last subdevice's erase size, count it */
864 curr_erasesize = subdev[i]->erasesize;
865 if (curr_erasesize > max_erasesize)
866 max_erasesize = curr_erasesize;
869 /* current subdevice has variable erase size */
871 for (j = 0; j < subdev[i]->numeraseregions; j++) {
873 /* walk the list of erase regions, count any changes */
874 if (subdev[i]->eraseregions[j].erasesize !=
878 subdev[i]->eraseregions[j].
880 if (curr_erasesize > max_erasesize)
881 max_erasesize = curr_erasesize;
887 if (num_erase_region == 1) {
889 * All subdevices have the same uniform erase size.
892 concat->mtd.erasesize = curr_erasesize;
893 concat->mtd.numeraseregions = 0;
898 * erase block size varies across the subdevices: allocate
899 * space to store the data describing the variable erase regions
901 struct mtd_erase_region_info *erase_region_p;
902 uint64_t begin, position;
904 concat->mtd.erasesize = max_erasesize;
905 concat->mtd.numeraseregions = num_erase_region;
906 concat->mtd.eraseregions = erase_region_p =
907 kmalloc(num_erase_region *
908 sizeof (struct mtd_erase_region_info), GFP_KERNEL);
909 if (!erase_region_p) {
912 ("memory allocation error while creating erase region list"
913 " for device \"%s\"\n", name);
918 * walk the map of the new device once more and fill in
919 * in erase region info:
921 curr_erasesize = subdev[0]->erasesize;
922 begin = position = 0;
923 for (i = 0; i < num_devs; i++) {
924 if (subdev[i]->numeraseregions == 0) {
925 /* current subdevice has uniform erase size */
926 if (subdev[i]->erasesize != curr_erasesize) {
928 * fill in an mtd_erase_region_info structure for the area
929 * we have walked so far:
931 erase_region_p->offset = begin;
932 erase_region_p->erasesize =
934 tmp64 = position - begin;
935 do_div(tmp64, curr_erasesize);
936 erase_region_p->numblocks = tmp64;
939 curr_erasesize = subdev[i]->erasesize;
942 position += subdev[i]->size;
944 /* current subdevice has variable erase size */
946 for (j = 0; j < subdev[i]->numeraseregions; j++) {
947 /* walk the list of erase regions, count any changes */
948 if (subdev[i]->eraseregions[j].
949 erasesize != curr_erasesize) {
950 erase_region_p->offset = begin;
951 erase_region_p->erasesize =
953 tmp64 = position - begin;
954 do_div(tmp64, curr_erasesize);
955 erase_region_p->numblocks = tmp64;
959 subdev[i]->eraseregions[j].
964 subdev[i]->eraseregions[j].
965 numblocks * (uint64_t)curr_erasesize;
969 /* Now write the final entry */
970 erase_region_p->offset = begin;
971 erase_region_p->erasesize = curr_erasesize;
972 tmp64 = position - begin;
973 do_div(tmp64, curr_erasesize);
974 erase_region_p->numblocks = tmp64;
981 * This function destroys an MTD object obtained from concat_mtd_devs()
984 void mtd_concat_destroy(struct mtd_info *mtd)
986 struct mtd_concat *concat = CONCAT(mtd);
987 if (concat->mtd.numeraseregions)
988 kfree(concat->mtd.eraseregions);
992 EXPORT_SYMBOL(mtd_concat_create);
993 EXPORT_SYMBOL(mtd_concat_destroy);
995 MODULE_LICENSE("GPL");
996 MODULE_AUTHOR("Robert Kaiser <rkaiser@sysgo.de>");
997 MODULE_DESCRIPTION("Generic support for concatenating of MTD devices");