ftl.c

/* This version ported to the Linux-MTD system by dwmw2@infradead.org
 *
 * Fixes: Arnaldo Carvalho de Melo <acme@conectiva.com.br>
 * - fixes some leaks on failure in build_maps and ftl_notify_add, cleanups
 *
 * Based on:
 */
/*======================================================================

    A Flash Translation Layer memory card driver

    This driver implements a disk-like block device driver with an
    apparent block size of 512 bytes for flash memory cards.

    ftl_cs.c 1.62 2000/02/01 00:59:04

    The contents of this file are subject to the Mozilla Public
    License Version 1.1 (the "License"); you may not use this file
    except in compliance with the License. You may obtain a copy of
    the License at http://www.mozilla.org/MPL/

    Software distributed under the License is distributed on an "AS
    IS" basis, WITHOUT WARRANTY OF ANY KIND, either express or
    implied. See the License for the specific language governing
    rights and limitations under the License.

    The initial developer of the original code is David A. Hinds
    <dahinds@users.sourceforge.net>.  Portions created by David A. Hinds
    are Copyright (C) 1999 David A. Hinds.  All Rights Reserved.

    Alternatively, the contents of this file may be used under the
    terms of the GNU General Public License version 2 (the "GPL"), in
    which case the provisions of the GPL are applicable instead of the
    above.  If you wish to allow the use of your version of this file
    only under the terms of the GPL and not to allow others to use
    your version of this file under the MPL, indicate your decision
    by deleting the provisions above and replace them with the notice
    and other provisions required by the GPL.  If you do not delete
    the provisions above, a recipient may use your version of this
    file under either the MPL or the GPL.

    LEGAL NOTE: The FTL format is patented by M-Systems.  They have
    granted a license for its use with PCMCIA devices:

     "M-Systems grants a royalty-free, non-exclusive license under
      any presently existing M-Systems intellectual property rights
      necessary for the design and development of FTL-compatible
      drivers, file systems and utilities using the data formats with
      PCMCIA PC Cards as described in the PCMCIA Flash Translation
      Layer (FTL) Specification."

    Use of the FTL format for non-PCMCIA applications may be an
    infringement of these patents.  For additional information,
    contact M-Systems (http://www.m-sys.com) directly.

======================================================================*/
#include <linux/mtd/blktrans.h>
#include <linux/module.h>
#include <linux/mtd/mtd.h>
/*#define PSYCHO_DEBUG */

#include <linux/kernel.h>
#include <linux/ptrace.h>
#include <linux/slab.h>
#include <linux/string.h>
#include <linux/timer.h>
#include <linux/major.h>
#include <linux/fs.h>
#include <linux/init.h>
#include <linux/hdreg.h>
#include <linux/vmalloc.h>
#include <linux/blkpg.h>
#include <asm/uaccess.h>

#include <linux/mtd/ftl.h>

/*====================================================================*/

/* Parameters that can be set with 'insmod' */
static int shuffle_freq = 50;
module_param(shuffle_freq, int, 0);

/*====================================================================*/

/* Major device # for FTL device */
#ifndef FTL_MAJOR
#define FTL_MAJOR	44
#endif


/*====================================================================*/

/* Maximum number of separate memory devices we'll allow */
#define MAX_DEV		4

/* Maximum number of regions per device */
#define MAX_REGION	4

/* Maximum number of partitions in an FTL region */
#define PART_BITS	4

/* Maximum number of outstanding erase requests per socket */
#define MAX_ERASE	8

/* Sector size -- shouldn't need to change */
#define SECTOR_SIZE	512


/* Each memory region corresponds to a minor device */
typedef struct partition_t {
    struct mtd_blktrans_dev mbd;
    u_int32_t		state;
    u_int32_t		*VirtualBlockMap;
    u_int32_t		*VirtualPageMap;
    u_int32_t		FreeTotal;
    struct eun_info_t {
	u_int32_t		Offset;
	u_int32_t		EraseCount;
	u_int32_t		Free;
	u_int32_t		Deleted;
    } *EUNInfo;
    struct xfer_info_t {
	u_int32_t		Offset;
	u_int32_t		EraseCount;
	u_int16_t		state;
    } *XferInfo;
    u_int16_t		bam_index;
    u_int32_t		*bam_cache;
    u_int16_t		DataUnits;
    u_int32_t		BlocksPerUnit;
    erase_unit_header_t	header;
} partition_t;

/* Partition state flags */
#define FTL_FORMATTED	0x01

/* Transfer unit states */
#define XFER_UNKNOWN	0x00
#define XFER_ERASING	0x01
#define XFER_ERASED	0x02
#define XFER_PREPARED	0x03
#define XFER_FAILED	0x04

/*====================================================================*/


static void ftl_erase_callback(struct erase_info *done);


/*======================================================================

    Scan_header() checks to see if a memory region contains an FTL
    partition.  build_maps() reads all the erase unit headers, builds
    the erase unit map, and then builds the virtual page map.

======================================================================*/

static int scan_header(partition_t *part)
{
    erase_unit_header_t header;
    loff_t offset, max_offset;
    size_t ret;
    int err;
    part->header.FormattedSize = 0;
    max_offset = (0x100000<part->mbd.mtd->size)?0x100000:part->mbd.mtd->size;
    /* Search first megabyte for a valid FTL header */
    for (offset = 0;
	 (offset + sizeof(header)) < max_offset;
	 offset += part->mbd.mtd->erasesize ? : 0x2000) {

	err = part->mbd.mtd->read(part->mbd.mtd, offset, sizeof(header), &ret,
			      (unsigned char *)&header);

	if (err)
	    return err;

	if (strcmp(header.DataOrgTuple+3, "FTL100") == 0) break;
    }

    if (offset == max_offset) {
	printk(KERN_NOTICE "ftl_cs: FTL header not found.\n");
	return -ENOENT;
    }
    if (header.BlockSize != 9 ||
	(header.EraseUnitSize < 10) || (header.EraseUnitSize > 31) ||
	(header.NumTransferUnits >= le16_to_cpu(header.NumEraseUnits))) {
	printk(KERN_NOTICE "ftl_cs: FTL header corrupt!\n");
	return -1;
    }
    if ((1 << header.EraseUnitSize) != part->mbd.mtd->erasesize) {
	printk(KERN_NOTICE "ftl: FTL EraseUnitSize %x != MTD erasesize %x\n",
	       1 << header.EraseUnitSize,part->mbd.mtd->erasesize);
	return -1;
    }
    part->header = header;
    return 0;
}

static int build_maps(partition_t *part)
{
    erase_unit_header_t header;
    u_int16_t xvalid, xtrans, i;
    u_int blocks, j;
    int hdr_ok, ret = -1;
    ssize_t retval;
    loff_t offset;

    /* Set up erase unit maps */
    part->DataUnits = le16_to_cpu(part->header.NumEraseUnits) -
	part->header.NumTransferUnits;
    part->EUNInfo = kmalloc(part->DataUnits * sizeof(struct eun_info_t),
			    GFP_KERNEL);
    if (!part->EUNInfo)
	    goto out;
    for (i = 0; i < part->DataUnits; i++)
	part->EUNInfo[i].Offset = 0xffffffff;
    part->XferInfo =
	kmalloc(part->header.NumTransferUnits * sizeof(struct xfer_info_t),
		GFP_KERNEL);
    if (!part->XferInfo)
	    goto out_EUNInfo;

    xvalid = xtrans = 0;
    for (i = 0; i < le16_to_cpu(part->header.NumEraseUnits); i++) {
	offset = ((i + le16_to_cpu(part->header.FirstPhysicalEUN))
		      << part->header.EraseUnitSize);
	ret = part->mbd.mtd->read(part->mbd.mtd, offset, sizeof(header), &retval,
			      (unsigned char *)&header);

	if (ret)
	    goto out_XferInfo;

	ret = -1;
	/* Is this a transfer partition? */
	hdr_ok = (strcmp(header.DataOrgTuple+3, "FTL100") == 0);
	if (hdr_ok && (le16_to_cpu(header.LogicalEUN) < part->DataUnits) &&
	    (part->EUNInfo[le16_to_cpu(header.LogicalEUN)].Offset == 0xffffffff)) {
	    part->EUNInfo[le16_to_cpu(header.LogicalEUN)].Offset = offset;
	    part->EUNInfo[le16_to_cpu(header.LogicalEUN)].EraseCount =
		le32_to_cpu(header.EraseCount);
	    xvalid++;
	} else {
	    if (xtrans == part->header.NumTransferUnits) {
		printk(KERN_NOTICE "ftl_cs: format error: too many "
		       "transfer units!\n");
		goto out_XferInfo;
	    }
	    if (hdr_ok && (le16_to_cpu(header.LogicalEUN) == 0xffff)) {
		part->XferInfo[xtrans].state = XFER_PREPARED;
		part->XferInfo[xtrans].EraseCount = le32_to_cpu(header.EraseCount);
	    } else {
		part->XferInfo[xtrans].state = XFER_UNKNOWN;
		/* Pick anything reasonable for the erase count */
		part->XferInfo[xtrans].EraseCount =
		    le32_to_cpu(part->header.EraseCount);
	    }
	    part->XferInfo[xtrans].Offset = offset;
	    xtrans++;
	}
    }
    /* Check for format trouble */
    header = part->header;
    if ((xtrans != header.NumTransferUnits) ||
	(xvalid+xtrans != le16_to_cpu(header.NumEraseUnits))) {
	printk(KERN_NOTICE "ftl_cs: format error: erase units "
	       "don't add up!\n");
	goto out_XferInfo;
    }

    /* Set up virtual page map */
    blocks = le32_to_cpu(header.FormattedSize) >> header.BlockSize;
    part->VirtualBlockMap = vmalloc(blocks * sizeof(u_int32_t));
    if (!part->VirtualBlockMap)
	    goto out_XferInfo;

    memset(part->VirtualBlockMap, 0xff, blocks * sizeof(u_int32_t));
    part->BlocksPerUnit = (1 << header.EraseUnitSize) >> header.BlockSize;

    part->bam_cache = kmalloc(part->BlocksPerUnit * sizeof(u_int32_t),
			      GFP_KERNEL);
    if (!part->bam_cache)
	    goto out_VirtualBlockMap;

    part->bam_index = 0xffff;
    part->FreeTotal = 0;

    for (i = 0; i < part->DataUnits; i++) {
	part->EUNInfo[i].Free = 0;
	part->EUNInfo[i].Deleted = 0;
	offset = part->EUNInfo[i].Offset + le32_to_cpu(header.BAMOffset);

	ret = part->mbd.mtd->read(part->mbd.mtd, offset,
			      part->BlocksPerUnit * sizeof(u_int32_t), &retval,
			      (unsigned char *)part->bam_cache);

	if (ret)
		goto out_bam_cache;

	for (j = 0; j < part->BlocksPerUnit; j++) {
	    if (BLOCK_FREE(le32_to_cpu(part->bam_cache[j]))) {
		part->EUNInfo[i].Free++;
		part->FreeTotal++;
	    } else if ((BLOCK_TYPE(le32_to_cpu(part->bam_cache[j])) == BLOCK_DATA) &&
		     (BLOCK_NUMBER(le32_to_cpu(part->bam_cache[j])) < blocks))
		part->VirtualBlockMap[BLOCK_NUMBER(le32_to_cpu(part->bam_cache[j]))] =
		    (i << header.EraseUnitSize) + (j << header.BlockSize);
	    else if (BLOCK_DELETED(le32_to_cpu(part->bam_cache[j])))
		part->EUNInfo[i].Deleted++;
	}
    }

    ret = 0;
    goto out;

out_bam_cache:
    kfree(part->bam_cache);
out_VirtualBlockMap:
    vfree(part->VirtualBlockMap);
out_XferInfo:
    kfree(part->XferInfo);
out_EUNInfo:
    kfree(part->EUNInfo);
out:
    return ret;
} /* build_maps */

/*======================================================================

    Erase_xfer() schedules an asynchronous erase operation for a
    transfer unit.

======================================================================*/

static int erase_xfer(partition_t *part,
		      u_int16_t xfernum)
{
    int ret;
    struct xfer_info_t *xfer;
    struct erase_info *erase;

    xfer = &part->XferInfo[xfernum];
    DEBUG(1, "ftl_cs: erasing xfer unit at 0x%x\n", xfer->Offset);
    xfer->state = XFER_ERASING;

    /* Is there a free erase slot? Always in MTD. */


    erase=kmalloc(sizeof(struct erase_info), GFP_KERNEL);
    if (!erase)
            return -ENOMEM;

    erase->mtd = part->mbd.mtd;
    erase->callback = ftl_erase_callback;
    erase->addr = xfer->Offset;
    erase->len = 1 << part->header.EraseUnitSize;
    erase->priv = (u_long)part;

    ret = part->mbd.mtd->erase(part->mbd.mtd, erase);

    if (!ret)
	    xfer->EraseCount++;
    else
	    kfree(erase);

    return ret;
} /* erase_xfer */

/*======================================================================

    Prepare_xfer() takes a freshly erased transfer unit and gives
    it an appropriate header.

======================================================================*/

static void ftl_erase_callback(struct erase_info *erase)
{
    partition_t *part;
    struct xfer_info_t *xfer;
    int i;

    /* Look up the transfer unit */
    part = (partition_t *)(erase->priv);

    for (i = 0; i < part->header.NumTransferUnits; i++)
	if (part->XferInfo[i].Offset == erase->addr) break;

    if (i == part->header.NumTransferUnits) {
	printk(KERN_NOTICE "ftl_cs: internal error: "
	       "erase lookup failed!\n");
	return;
    }

    xfer = &part->XferInfo[i];
    if (erase->state == MTD_ERASE_DONE)
	xfer->state = XFER_ERASED;
    else {
	xfer->state = XFER_FAILED;
	printk(KERN_NOTICE "ftl_cs: erase failed: state = %d\n",
	       erase->state);
    }

    kfree(erase);

} /* ftl_erase_callback */

static int prepare_xfer(partition_t *part, int i)
{
    erase_unit_header_t header;
    struct xfer_info_t *xfer;
    int nbam, ret;
    u_int32_t ctl;
    ssize_t retlen;
    loff_t offset;

    xfer = &part->XferInfo[i];
    xfer->state = XFER_FAILED;

    DEBUG(1, "ftl_cs: preparing xfer unit at 0x%x\n", xfer->Offset);

    /* Write the transfer unit header */
    header = part->header;
    header.LogicalEUN = cpu_to_le16(0xffff);
    header.EraseCount = cpu_to_le32(xfer->EraseCount);

    ret = part->mbd.mtd->write(part->mbd.mtd, xfer->Offset, sizeof(header),
			   &retlen, (u_char *)&header);

    if (ret) {
	return ret;
    }

    /* Write the BAM stub */
    nbam = (part->BlocksPerUnit * sizeof(u_int32_t) +
	    le32_to_cpu(part->header.BAMOffset) + SECTOR_SIZE - 1) / SECTOR_SIZE;

    offset = xfer->Offset + le32_to_cpu(part->header.BAMOffset);
    ctl = cpu_to_le32(BLOCK_CONTROL);

    for (i = 0; i < nbam; i++, offset += sizeof(u_int32_t)) {

	ret = part->mbd.mtd->write(part->mbd.mtd, offset, sizeof(u_int32_t),
			       &retlen, (u_char *)&ctl);

	if (ret)
	    return ret;
    }
    xfer->state = XFER_PREPARED;
    return 0;

} /* prepare_xfer */

/*======================================================================

    Copy_erase_unit() takes a full erase block and a transfer unit,
    copies everything to the transfer unit, then swaps the block
    pointers.

    All data blocks are copied to the corresponding blocks in the
    target unit, so the virtual block map does not need to be
    updated.

======================================================================*/

static int copy_erase_unit(partition_t *part, u_int16_t srcunit,
			   u_int16_t xferunit)
{
    u_char buf[SECTOR_SIZE];
    struct eun_info_t *eun;
    struct xfer_info_t *xfer;
    u_int32_t src, dest, free, i;
    u_int16_t unit;
    int ret;
    ssize_t retlen;
    loff_t offset;
    u_int16_t srcunitswap = cpu_to_le16(srcunit);

    eun = &part->EUNInfo[srcunit];
    xfer = &part->XferInfo[xferunit];
    DEBUG(2, "ftl_cs: copying block 0x%x to 0x%x\n",
	  eun->Offset, xfer->Offset);


    /* Read current BAM */
    if (part->bam_index != srcunit) {

	offset = eun->Offset + le32_to_cpu(part->header.BAMOffset);

	ret = part->mbd.mtd->read(part->mbd.mtd, offset,
			      part->BlocksPerUnit * sizeof(u_int32_t),
			      &retlen, (u_char *) (part->bam_cache));

	/* mark the cache bad, in case we get an error later */
	part->bam_index = 0xffff;

	if (ret) {
	    printk( KERN_WARNING "ftl: Failed to read BAM cache in copy_erase_unit()!\n");
	    return ret;
	}
    }

    /* Write the LogicalEUN for the transfer unit */
    xfer->state = XFER_UNKNOWN;
    offset = xfer->Offset + 20; /* Bad! */
    unit = cpu_to_le16(0x7fff);

    ret = part->mbd.mtd->write(part->mbd.mtd, offset, sizeof(u_int16_t),
			   &retlen, (u_char *) &unit);

    if (ret) {
	printk( KERN_WARNING "ftl: Failed to write back to BAM cache in copy_erase_unit()!\n");
	return ret;
    }

    /* Copy all data blocks from source unit to transfer unit */
    src = eun->Offset; dest = xfer->Offset;

    free = 0;
    ret = 0;
    for (i = 0; i < part->BlocksPerUnit; i++) {
	switch (BLOCK_TYPE(le32_to_cpu(part->bam_cache[i]))) {
	case BLOCK_CONTROL:
	    /* This gets updated later */
	    break;
	case BLOCK_DATA:
	case BLOCK_REPLACEMENT:
	    ret = part->mbd.mtd->read(part->mbd.mtd, src, SECTOR_SIZE,
                        &retlen, (u_char *) buf);
	    if (ret) {
		printk(KERN_WARNING "ftl: Error reading old xfer unit in copy_erase_unit\n");
		return ret;
            }


	    ret = part->mbd.mtd->write(part->mbd.mtd, dest, SECTOR_SIZE,
                        &retlen, (u_char *) buf);
	    if (ret)  {
		printk(KERN_WARNING "ftl: Error writing new xfer unit in copy_erase_unit\n");
		return ret;
            }

	    break;
	default:
	    /* All other blocks must be free */
	    part->bam_cache[i] = cpu_to_le32(0xffffffff);
	    free++;
	    break;
	}
	src += SECTOR_SIZE;
	dest += SECTOR_SIZE;
    }

    /* Write the BAM to the transfer unit */
    ret = part->mbd.mtd->write(part->mbd.mtd, xfer->Offset + le32_to_cpu(part->header.BAMOffset),
                    part->BlocksPerUnit * sizeof(int32_t), &retlen,
		    (u_char *)part->bam_cache);
    if (ret) {
	printk( KERN_WARNING "ftl: Error writing BAM in copy_erase_unit\n");
	return ret;
    }


    /* All clear? Then update the LogicalEUN again */
    ret = part->mbd.mtd->write(part->mbd.mtd, xfer->Offset + 20, sizeof(u_int16_t),
			   &retlen, (u_char *)&srcunitswap);

    if (ret) {
	printk(KERN_WARNING "ftl: Error writing new LogicalEUN in copy_erase_unit\n");
	return ret;
    }


    /* Update the maps and usage stats*/
    i = xfer->EraseCount;
    xfer->EraseCount = eun->EraseCount;
    eun->EraseCount = i;
    i = xfer->Offset;
    xfer->Offset = eun->Offset;
    eun->Offset = i;
    part->FreeTotal -= eun->Free;
    part->FreeTotal += free;
    eun->Free = free;
    eun->Deleted = 0;

    /* Now, the cache should be valid for the new block */
    part->bam_index = srcunit;

    return 0;
} /* copy_erase_unit */

/*======================================================================

    reclaim_block() picks a full erase unit and a transfer unit and
    then calls copy_erase_unit() to copy one to the other.  Then, it
    schedules an erase on the expired block.

    What's a good way to decide which transfer unit and which erase
    unit to use?  Beats me.  My way is to always pick the transfer
    unit with the fewest erases, and usually pick the data unit with
    the most deleted blocks.  But with a small probability, pick the
    oldest data unit instead.  This means that we generally postpone
    the next reclaimation as long as possible, but shuffle static
    stuff around a bit for wear leveling.

======================================================================*/

static int reclaim_block(partition_t *part)
{
    u_int16_t i, eun, xfer;
    u_int32_t best;
    int queued, ret;

    DEBUG(0, "ftl_cs: reclaiming space...\n");
    DEBUG(3, "NumTransferUnits == %x\n", part->header.NumTransferUnits);
    /* Pick the least erased transfer unit */
    best = 0xffffffff; xfer = 0xffff;
    do {
	queued = 0;
	for (i = 0; i < part->header.NumTransferUnits; i++) {
	    int n=0;
	    if (part->XferInfo[i].state == XFER_UNKNOWN) {
		DEBUG(3,"XferInfo[%d].state == XFER_UNKNOWN\n",i);
		n=1;
		erase_xfer(part, i);
	    }
	    if (part->XferInfo[i].state == XFER_ERASING) {
		DEBUG(3,"XferInfo[%d].state == XFER_ERASING\n",i);
		n=1;
		queued = 1;
	    }
	    else if (part->XferInfo[i].state == XFER_ERASED) {
		DEBUG(3,"XferInfo[%d].state == XFER_ERASED\n",i);
		n=1;
		prepare_xfer(part, i);
	    }
	    if (part->XferInfo[i].state == XFER_PREPARED) {
		DEBUG(3,"XferInfo[%d].state == XFER_PREPARED\n",i);
		n=1;
		if (part->XferInfo[i].EraseCount <= best) {
		    best = part->XferInfo[i].EraseCount;
		    xfer = i;
		}
	    }
		if (!n)
		    DEBUG(3,"XferInfo[%d].state == %x\n",i, part->XferInfo[i].state);

	}
	if (xfer == 0xffff) {
	    if (queued) {
		DEBUG(1, "ftl_cs: waiting for transfer "
		      "unit to be prepared...\n");
		if (part->mbd.mtd->sync)
			part->mbd.mtd->sync(part->mbd.mtd);
	    } else {
		static int ne = 0;
		if (++ne < 5)
		    printk(KERN_NOTICE "ftl_cs: reclaim failed: no "
			   "suitable transfer units!\n");
		else
		    DEBUG(1, "ftl_cs: reclaim failed: no "
			  "suitable transfer units!\n");

		return -EIO;
	    }
	}
    } while (xfer == 0xffff);

    eun = 0;
    if ((jiffies % shuffle_freq) == 0) {
	DEBUG(1, "ftl_cs: recycling freshest block...\n");
	best = 0xffffffff;
	for (i = 0; i < part->DataUnits; i++)
	    if (part->EUNInfo[i].EraseCount <= best) {
		best = part->EUNInfo[i].EraseCount;
		eun = i;
	    }
    } else {
	best = 0;
	for (i = 0; i < part->DataUnits; i++)
	    if (part->EUNInfo[i].Deleted >= best) {
		best = part->EUNInfo[i].Deleted;
		eun = i;
	    }
	if (best == 0) {
	    static int ne = 0;
	    if (++ne < 5)
		printk(KERN_NOTICE "ftl_cs: reclaim failed: "
		       "no free blocks!\n");
	    else
		DEBUG(1,"ftl_cs: reclaim failed: "
		       "no free blocks!\n");

	    return -EIO;
	}
    }
    ret = copy_erase_unit(part, eun, xfer);
    if (!ret)
	erase_xfer(part, xfer);
    else
	printk(KERN_NOTICE "ftl_cs: copy_erase_unit failed!\n");
    return ret;
} /* reclaim_block */

/*======================================================================

    Find_free() searches for a free block.  If necessary, it updates
    the BAM cache for the erase unit containing the free block.  It
    returns the block index -- the erase unit is just the currently
    cached unit.  If there are no free blocks, it returns 0 -- this
    is never a valid data block because it contains the header.

======================================================================*/

#ifdef PSYCHO_DEBUG
static void dump_lists(partition_t *part)
{
    int i;
    printk(KERN_DEBUG "ftl_cs: Free total = %d\n", part->FreeTotal);
    for (i = 0; i < part->DataUnits; i++)
	printk(KERN_DEBUG "ftl_cs:   unit %d: %d phys, %d free, "
	       "%d deleted\n", i,
	       part->EUNInfo[i].Offset >> part->header.EraseUnitSize,
	       part->EUNInfo[i].Free, part->EUNInfo[i].Deleted);
}
#endif

static u_int32_t find_free(partition_t *part)
{
    u_int16_t stop, eun;
    u_int32_t blk;
    size_t retlen;
    int ret;

    /* Find an erase unit with some free space */
    stop = (part->bam_index == 0xffff) ? 0 : part->bam_index;
    eun = stop;
    do {
	if (part->EUNInfo[eun].Free != 0) break;
	/* Wrap around at end of table */
	if (++eun == part->DataUnits) eun = 0;
    } while (eun != stop);

    if (part->EUNInfo[eun].Free == 0)
	return 0;

    /* Is this unit's BAM cached? */
    if (eun != part->bam_index) {
	/* Invalidate cache */
	part->bam_index = 0xffff;

	ret = part->mbd.mtd->read(part->mbd.mtd,
		       part->EUNInfo[eun].Offset + le32_to_cpu(part->header.BAMOffset),
		       part->BlocksPerUnit * sizeof(u_int32_t),
		       &retlen, (u_char *) (part->bam_cache));

	if (ret) {
	    printk(KERN_WARNING"ftl: Error reading BAM in find_free\n");
	    return 0;
	}
	part->bam_index = eun;
    }

    /* Find a free block */
    for (blk = 0; blk < part->BlocksPerUnit; blk++)
	if (BLOCK_FREE(le32_to_cpu(part->bam_cache[blk]))) break;
    if (blk == part->BlocksPerUnit) {
#ifdef PSYCHO_DEBUG
	static int ne = 0;
	if (++ne == 1)
	    dump_lists(part);
#endif
	printk(KERN_NOTICE "ftl_cs: bad free list!\n");
	return 0;
    }
    DEBUG(2, "ftl_cs: found free block at %d in %d\n", blk, eun);
    return blk;

} /* find_free */


/*======================================================================

    Read a series of sectors from an FTL partition.

======================================================================*/

static int ftl_read(partition_t *part, caddr_t buffer,
		    u_long sector, u_long nblocks)
{
    u_int32_t log_addr, bsize;
    u_long i;
    int ret;
    size_t offset, retlen;

    DEBUG(2, "ftl_cs: ftl_read(0x%p, 0x%lx, %ld)\n",
	  part, sector, nblocks);
    if (!(part->state & FTL_FORMATTED)) {
	printk(KERN_NOTICE "ftl_cs: bad partition\n");
	return -EIO;
    }
    bsize = 1 << part->header.EraseUnitSize;

    for (i = 0; i < nblocks; i++) {
	if (((sector+i) * SECTOR_SIZE) >= le32_to_cpu(part->header.FormattedSize)) {
	    printk(KERN_NOTICE "ftl_cs: bad read offset\n");
	    return -EIO;
	}
	log_addr = part->VirtualBlockMap[sector+i];
	if (log_addr == 0xffffffff)
	    memset(buffer, 0, SECTOR_SIZE);
	else {
	    offset = (part->EUNInfo[log_addr / bsize].Offset
			  + (log_addr % bsize));
	    ret = part->mbd.mtd->read(part->mbd.mtd, offset, SECTOR_SIZE,
			   &retlen, (u_char *) buffer);

	    if (ret) {
		printk(KERN_WARNING "Error reading MTD device in ftl_read()\n");
		return ret;
	    }
	}
	buffer += SECTOR_SIZE;
    }
    return 0;
} /* ftl_read */

/*======================================================================

    Write a series of sectors to an FTL partition

======================================================================*/

static int set_bam_entry(partition_t *part, u_int32_t log_addr,
			 u_int32_t virt_addr)
{
    u_int32_t bsize, blk, le_virt_addr;
#ifdef PSYCHO_DEBUG
    u_int32_t old_addr;
#endif
    u_int16_t eun;
    int ret;
    size_t retlen, offset;

    DEBUG(2, "ftl_cs: set_bam_entry(0x%p, 0x%x, 0x%x)\n",
	  part, log_addr, virt_addr);
    bsize = 1 << part->header.EraseUnitSize;
    eun = log_addr / bsize;
    blk = (log_addr % bsize) / SECTOR_SIZE;
    offset = (part->EUNInfo[eun].Offset + blk * sizeof(u_int32_t) +
		  le32_to_cpu(part->header.BAMOffset));

#ifdef PSYCHO_DEBUG
    ret = part->mbd.mtd->read(part->mbd.mtd, offset, sizeof(u_int32_t),
                        &retlen, (u_char *)&old_addr);
    if (ret) {
	printk(KERN_WARNING"ftl: Error reading old_addr in set_bam_entry: %d\n",ret);
	return ret;
    }
    old_addr = le32_to_cpu(old_addr);

    if (((virt_addr == 0xfffffffe) && !BLOCK_FREE(old_addr)) ||
	((virt_addr == 0) && (BLOCK_TYPE(old_addr) != BLOCK_DATA)) ||
	(!BLOCK_DELETED(virt_addr) && (old_addr != 0xfffffffe))) {
	static int ne = 0;
	if (++ne < 5) {
	    printk(KERN_NOTICE "ftl_cs: set_bam_entry() inconsistency!\n");
	    printk(KERN_NOTICE "ftl_cs:   log_addr = 0x%x, old = 0x%x"
		   ", new = 0x%x\n", log_addr, old_addr, virt_addr);
	}
	return -EIO;
    }
#endif
    le_virt_addr = cpu_to_le32(virt_addr);
    if (part->bam_index == eun) {
#ifdef PSYCHO_DEBUG
	if (le32_to_cpu(part->bam_cache[blk]) != old_addr) {
	    static int ne = 0;
	    if (++ne < 5) {
		printk(KERN_NOTICE "ftl_cs: set_bam_entry() "
		       "inconsistency!\n");
		printk(KERN_NOTICE "ftl_cs:   log_addr = 0x%x, cache"
		       " = 0x%x\n",
		       le32_to_cpu(part->bam_cache[blk]), old_addr);
	    }
	    return -EIO;
	}
#endif
	part->bam_cache[blk] = le_virt_addr;
    }
    ret = part->mbd.mtd->write(part->mbd.mtd, offset, sizeof(u_int32_t),
                            &retlen, (u_char *)&le_virt_addr);

    if (ret) {
	printk(KERN_NOTICE "ftl_cs: set_bam_entry() failed!\n");
	printk(KERN_NOTICE "ftl_cs:   log_addr = 0x%x, new = 0x%x\n",
	       log_addr, virt_addr);
    }
    return ret;
} /* set_bam_entry */

static int ftl_write(partition_t *part, caddr_t buffer,
		     u_long sector, u_long nblocks)
{
    u_int32_t bsize, log_addr, virt_addr, old_addr, blk;
    u_long i;
    int ret;
    size_t retlen, offset;

    DEBUG(2, "ftl_cs: ftl_write(0x%p, %ld, %ld)\n",
	  part, sector, nblocks);
    if (!(part->state & FTL_FORMATTED)) {
	printk(KERN_NOTICE "ftl_cs: bad partition\n");
	return -EIO;
    }
    /* See if we need to reclaim space, before we start */
    while (part->FreeTotal < nblocks) {
	ret = reclaim_block(part);
	if (ret)
	    return ret;
    }

    bsize = 1 << part->header.EraseUnitSize;

    virt_addr = sector * SECTOR_SIZE | BLOCK_DATA;
    for (i = 0; i < nblocks; i++) {
	if (virt_addr >= le32_to_cpu(part->header.FormattedSize)) {
	    printk(KERN_NOTICE "ftl_cs: bad write offset\n");
	    return -EIO;
	}

	/* Grab a free block */
	blk = find_free(part);
	if (blk == 0) {
	    static int ne = 0;
	    if (++ne < 5)
		printk(KERN_NOTICE "ftl_cs: internal error: "
		       "no free blocks!\n");
	    return -ENOSPC;
	}

	/* Tag the BAM entry, and write the new block */
	log_addr = part->bam_index * bsize + blk * SECTOR_SIZE;
	part->EUNInfo[part->bam_index].Free--;
	part->FreeTotal--;
	if (set_bam_entry(part, log_addr, 0xfffffffe))
	    return -EIO;
	part->EUNInfo[part->bam_index].Deleted++;
	offset = (part->EUNInfo[part->bam_index].Offset +
		      blk * SECTOR_SIZE);
	ret = part->mbd.mtd->write(part->mbd.mtd, offset, SECTOR_SIZE, &retlen,
                                     buffer);

	if (ret) {
	    printk(KERN_NOTICE "ftl_cs: block write failed!\n");
	    printk(KERN_NOTICE "ftl_cs:   log_addr = 0x%x, virt_addr"
		   " = 0x%x, Offset = 0x%zx\n", log_addr, virt_addr,
		   offset);
	    return -EIO;
	}

	/* Only delete the old entry when the new entry is ready */
	old_addr = part->VirtualBlockMap[sector+i];
	if (old_addr != 0xffffffff) {
	    part->VirtualBlockMap[sector+i] = 0xffffffff;
	    part->EUNInfo[old_addr/bsize].Deleted++;
	    if (set_bam_entry(part, old_addr, 0))
		return -EIO;
	}

	/* Finally, set up the new pointers */
	if (set_bam_entry(part, log_addr, virt_addr))
	    return -EIO;
	part->VirtualBlockMap[sector+i] = log_addr;
	part->EUNInfo[part->bam_index].Deleted--;

	buffer += SECTOR_SIZE;
	virt_addr += SECTOR_SIZE;
    }
    return 0;
} /* ftl_write */

static int ftl_getgeo(struct mtd_blktrans_dev *dev, struct hd_geometry *geo)
{
	partition_t *part = (void *)dev;
	u_long sect;

	/* Sort of arbitrary: round size down to 4KiB boundary */
	sect = le32_to_cpu(part->header.FormattedSize)/SECTOR_SIZE;

	geo->heads = 1;
	geo->sectors = 8;
	geo->cylinders = sect >> 3;

	return 0;
}

static int ftl_readsect(struct mtd_blktrans_dev *dev,
			      unsigned long block, char *buf)
{
	return ftl_read((void *)dev, buf, block, 1);
}

static int ftl_writesect(struct mtd_blktrans_dev *dev,
			      unsigned long block, char *buf)
{
	return ftl_write((void *)dev, buf, block, 1);
}

static int ftl_discardsect(struct mtd_blktrans_dev *dev,
			   unsigned long sector, unsigned nr_sects)
{
	partition_t *part = (void *)dev;
	uint32_t bsize = 1 << part->header.EraseUnitSize;

	DEBUG(1, "FTL erase sector %ld for %d sectors\n",
	      sector, nr_sects);

	while (nr_sects) {
		uint32_t old_addr = part->VirtualBlockMap[sector];
		if (old_addr != 0xffffffff) {
			part->VirtualBlockMap[sector] = 0xffffffff;
			part->EUNInfo[old_addr/bsize].Deleted++;
			if (set_bam_entry(part, old_addr, 0))
				return -EIO;
		}
		nr_sects--;
		sector++;
	}

	return 0;
}
/*====================================================================*/

static void ftl_freepart(partition_t *part)
{
	vfree(part->VirtualBlockMap);
	part->VirtualBlockMap = NULL;
	kfree(part->VirtualPageMap);
	part->VirtualPageMap = NULL;
	kfree(part->EUNInfo);
	part->EUNInfo = NULL;
	kfree(part->XferInfo);
	part->XferInfo = NULL;
	kfree(part->bam_cache);
	part->bam_cache = NULL;
} /* ftl_freepart */

static void ftl_add_mtd(struct mtd_blktrans_ops *tr, struct mtd_info *mtd)
{
	partition_t *partition;

	partition = kzalloc(sizeof(partition_t), GFP_KERNEL);

	if (!partition) {
		printk(KERN_WARNING "No memory to scan for FTL on %s\n",
		       mtd->name);
		return;
	}

	partition->mbd.mtd = mtd;

	if ((scan_header(partition) == 0) &&
	    (build_maps(partition) == 0)) {

		partition->state = FTL_FORMATTED;
#ifdef PCMCIA_DEBUG
		printk(KERN_INFO "ftl_cs: opening %d KiB FTL partition\n",
		       le32_to_cpu(partition->header.FormattedSize) >> 10);
#endif
		partition->mbd.size = le32_to_cpu(partition->header.FormattedSize) >> 9;

		partition->mbd.tr = tr;
		partition->mbd.devnum = -1;
		if (!add_mtd_blktrans_dev((void *)partition))
			return;
	}

	ftl_freepart(partition);
	kfree(partition);
}

static void ftl_remove_dev(struct mtd_blktrans_dev *dev)
{
	del_mtd_blktrans_dev(dev);
	ftl_freepart((partition_t *)dev);
	kfree(dev);
}

static struct mtd_blktrans_ops ftl_tr = {
	.name		= "ftl",
	.major		= FTL_MAJOR,
	.part_bits	= PART_BITS,
	.blksize 	= SECTOR_SIZE,
	.readsect	= ftl_readsect,
	.writesect	= ftl_writesect,
	.discard	= ftl_discardsect,
	.getgeo		= ftl_getgeo,
	.add_mtd	= ftl_add_mtd,
	.remove_dev	= ftl_remove_dev,
	.owner		= THIS_MODULE,
};

static int init_ftl(void)
{
	return register_mtd_blktrans(&ftl_tr);
}

static void __exit cleanup_ftl(void)
{
	deregister_mtd_blktrans(&ftl_tr);
}

module_init(init_ftl);
module_exit(cleanup_ftl);


MODULE_LICENSE("Dual MPL/GPL");
MODULE_AUTHOR("David Hinds <dahinds@users.sourceforge.net>");
MODULE_DESCRIPTION("Support code for Flash Translation Layer, used on PCMCIA devices");