mtd: Add new SmartMedia/xD FTL
authorMaxim Levitsky <maximlevitsky@gmail.com>
Mon, 22 Feb 2010 18:39:41 +0000 (20:39 +0200)
committerDavid Woodhouse <David.Woodhouse@intel.com>
Fri, 26 Feb 2010 19:06:50 +0000 (19:06 +0000)
This implements new readwrite SmartMedia/xd FTL.

mtd driver must have support proper ECC and badblock verification
based on oob parts for 512 bytes nand.

Also mtd driver must define read_oob and write_oob, which are used
to read and write both data and oob together.

Signed-off-by: Maxim Levitsky <maximlevitsky@gmail.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
drivers/mtd/Kconfig
drivers/mtd/Makefile
drivers/mtd/sm_ftl.c [new file with mode: 0644]
drivers/mtd/sm_ftl.h [new file with mode: 0644]

index ecf90f5..8a91240 100644 (file)
@@ -304,6 +304,27 @@ config SSFDC
          This enables read only access to SmartMedia formatted NAND
          flash. You can mount it with FAT file system.
 
+
+config SM_FTL
+       tristate "SmartMedia/xD new translation layer"
+       depends on EXPERIMENTAL && BLOCK
+       select MTD_BLKDEVS
+       help
+         This enables new and very EXPERMENTAL support for SmartMedia/xD
+         FTL (Flash tanslation layer)
+         Write support isn't yet well tested, therefore this code IS likely to
+         eat your card, so please don't use it together with valuable data.
+         Use readonly driver (CONFIG_SSFDC) instead.
+
+config SM_FTL_MUSEUM
+       boolean "Additional Support for 1MiB and 2MiB SmartMedia cards"
+       depends on SM_FTL && MTD_NAND
+       select MTD_NAND_ECC_SMC
+       help
+         Very old SmartMedia cards need ECC to be calculated in the FTL
+         Such cards are very rare, thus enabling this option is mostly useless
+         Also this support is completely UNTESTED.
+
 config MTD_OOPS
        tristate "Log panic/oops to an MTD buffer"
        depends on MTD
index 82d1e4d..d53357b 100644 (file)
@@ -24,6 +24,7 @@ obj-$(CONFIG_NFTL)            += nftl.o
 obj-$(CONFIG_INFTL)            += inftl.o
 obj-$(CONFIG_RFD_FTL)          += rfd_ftl.o
 obj-$(CONFIG_SSFDC)            += ssfdc.o
+obj-$(CONFIG_SM_FTL)           += sm_ftl.o
 obj-$(CONFIG_MTD_OOPS)         += mtdoops.o
 
 nftl-objs              := nftlcore.o nftlmount.o
diff --git a/drivers/mtd/sm_ftl.c b/drivers/mtd/sm_ftl.c
new file mode 100644 (file)
index 0000000..a59ebb4
--- /dev/null
@@ -0,0 +1,1284 @@
+/*
+ * Copyright © 2009 - Maxim Levitsky
+ * SmartMedia/xD translation layer
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/random.h>
+#include <linux/hdreg.h>
+#include <linux/kthread.h>
+#include <linux/freezer.h>
+#include <linux/sysfs.h>
+#include <linux/bitops.h>
+#include "nand/sm_common.h"
+#include "sm_ftl.h"
+
+#ifdef CONFIG_SM_FTL_MUSEUM
+#include <linux/mtd/nand_ecc.h>
+#endif
+
+
+struct workqueue_struct *cache_flush_workqueue;
+
+static int cache_timeout = 1000;
+module_param(cache_timeout, bool, S_IRUGO);
+MODULE_PARM_DESC(cache_timeout,
+       "Timeout (in ms) for cache flush (1000 ms default");
+
+static int debug;
+module_param(debug, int, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(debug, "Debug level (0-2)");
+
+
+/* ------------------- sysfs attributtes ---------------------------------- */
+struct sm_sysfs_attribute {
+       struct device_attribute dev_attr;
+       char *data;
+       int len;
+};
+
+ssize_t sm_attr_show(struct device *dev, struct device_attribute *attr,
+                    char *buf)
+{
+       struct sm_sysfs_attribute *sm_attr =
+               container_of(attr, struct sm_sysfs_attribute, dev_attr);
+
+       strncpy(buf, sm_attr->data, sm_attr->len);
+       return sm_attr->len;
+}
+
+
+#define NUM_ATTRIBUTES 1
+#define SM_CIS_VENDOR_OFFSET 0x59
+struct attribute_group *sm_create_sysfs_attributes(struct sm_ftl *ftl)
+{
+       struct attribute_group *attr_group;
+       struct attribute **attributes;
+       struct sm_sysfs_attribute *vendor_attribute;
+
+       int vendor_len = strnlen(ftl->cis_buffer + SM_CIS_VENDOR_OFFSET,
+                                       SM_SMALL_PAGE - SM_CIS_VENDOR_OFFSET);
+
+       char *vendor = kmalloc(vendor_len, GFP_KERNEL);
+       memcpy(vendor, ftl->cis_buffer + SM_CIS_VENDOR_OFFSET, vendor_len);
+       vendor[vendor_len] = 0;
+
+       /* Initialize sysfs attributes */
+       vendor_attribute =
+               kzalloc(sizeof(struct sm_sysfs_attribute), GFP_KERNEL);
+
+       vendor_attribute->data = vendor;
+       vendor_attribute->len = vendor_len;
+       vendor_attribute->dev_attr.attr.name = "vendor";
+       vendor_attribute->dev_attr.attr.mode = S_IRUGO;
+       vendor_attribute->dev_attr.show = sm_attr_show;
+
+
+       /* Create array of pointers to the attributes */
+       attributes = kzalloc(sizeof(struct attribute *) * (NUM_ATTRIBUTES + 1),
+                                                               GFP_KERNEL);
+       attributes[0] = &vendor_attribute->dev_attr.attr;
+
+       /* Finally create the attribute group */
+       attr_group = kzalloc(sizeof(struct attribute_group), GFP_KERNEL);
+       attr_group->attrs = attributes;
+       return attr_group;
+}
+
+void sm_delete_sysfs_attributes(struct sm_ftl *ftl)
+{
+       struct attribute **attributes = ftl->disk_attributes->attrs;
+       int i;
+
+       for (i = 0; attributes[i] ; i++) {
+
+               struct device_attribute *dev_attr = container_of(attributes[i],
+                       struct device_attribute, attr);
+
+               struct sm_sysfs_attribute *sm_attr =
+                       container_of(dev_attr,
+                               struct sm_sysfs_attribute, dev_attr);
+
+               kfree(sm_attr->data);
+               kfree(sm_attr);
+       }
+
+       kfree(ftl->disk_attributes->attrs);
+       kfree(ftl->disk_attributes);
+}
+
+
+/* ----------------------- oob helpers -------------------------------------- */
+
+static int sm_get_lba(uint8_t *lba)
+{
+       /* check fixed bits */
+       if ((lba[0] & 0xF8) != 0x10)
+               return -2;
+
+       /* check parity - endianess doesn't matter */
+       if (hweight16(*(uint16_t *)lba) & 1)
+               return -2;
+
+       return (lba[1] >> 1) | ((lba[0] & 0x07) << 7);
+}
+
+
+/*
+ * Read LBA asscociated with block
+ * returns -1, if block is erased
+ * returns -2 if error happens
+ */
+static int sm_read_lba(struct sm_oob *oob)
+{
+       static const uint32_t erased_pattern[4] = {
+               0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF };
+
+       uint16_t lba_test;
+       int lba;
+
+       /* First test for erased block */
+       if (!memcmp(oob, erased_pattern, SM_OOB_SIZE))
+               return -1;
+
+       /* Now check is both copies of the LBA differ too much */
+       lba_test = *(uint16_t *)oob->lba_copy1 ^ *(uint16_t*)oob->lba_copy2;
+       if (lba_test && !is_power_of_2(lba_test))
+               return -2;
+
+       /* And read it */
+       lba = sm_get_lba(oob->lba_copy1);
+
+       if (lba == -2)
+               lba = sm_get_lba(oob->lba_copy2);
+
+       return lba;
+}
+
+static void sm_write_lba(struct sm_oob *oob, uint16_t lba)
+{
+       uint8_t tmp[2];
+
+       WARN_ON(lba >= 1000);
+
+       tmp[0] = 0x10 | ((lba >> 7) & 0x07);
+       tmp[1] = (lba << 1) & 0xFF;
+
+       if (hweight16(*(uint16_t *)tmp) & 0x01)
+               tmp[1] |= 1;
+
+       oob->lba_copy1[0] = oob->lba_copy2[0] = tmp[0];
+       oob->lba_copy1[1] = oob->lba_copy2[1] = tmp[1];
+}
+
+
+/* Make offset from parts */
+static loff_t sm_mkoffset(struct sm_ftl *ftl, int zone, int block, int boffset)
+{
+       WARN_ON(boffset & (SM_SECTOR_SIZE - 1));
+       WARN_ON(zone < 0 || zone >= ftl->zone_count);
+       WARN_ON(block >= ftl->zone_size);
+       WARN_ON(boffset >= ftl->block_size);
+
+       if (block == -1)
+               return -1;
+
+       return (zone * SM_MAX_ZONE_SIZE + block) * ftl->block_size + boffset;
+}
+
+/* Breaks offset into parts */
+static void sm_break_offset(struct sm_ftl *ftl, loff_t offset,
+                           int *zone, int *block, int *boffset)
+{
+       *boffset = do_div(offset, ftl->block_size);
+       *block = do_div(offset, ftl->max_lba);
+       *zone = offset >= ftl->zone_count ? -1 : offset;
+}
+
+/* ---------------------- low level IO ------------------------------------- */
+
+static int sm_correct_sector(uint8_t *buffer, struct sm_oob *oob)
+{
+#ifdef CONFIG_SM_FTL_MUSEUM
+       uint8_t ecc[3];
+
+       __nand_calculate_ecc(buffer, SM_SMALL_PAGE, ecc);
+       if (__nand_correct_data(buffer, ecc, oob->ecc1, SM_SMALL_PAGE) < 0)
+               return -EIO;
+
+       buffer += SM_SMALL_PAGE;
+
+       __nand_calculate_ecc(buffer, SM_SMALL_PAGE, ecc);
+       if (__nand_correct_data(buffer, ecc, oob->ecc2, SM_SMALL_PAGE) < 0)
+               return -EIO;
+#endif
+       return 0;
+}
+
+/* Reads a sector + oob*/
+static int sm_read_sector(struct sm_ftl *ftl,
+                         int zone, int block, int boffset,
+                         uint8_t *buffer, struct sm_oob *oob)
+{
+       struct mtd_info *mtd = ftl->trans->mtd;
+       struct mtd_oob_ops ops;
+       struct sm_oob tmp_oob;
+       int ret;
+       int try = 0;
+
+       /* FTL can contain -1 entries that are by default filled with bits */
+       if (block == -1) {
+               memset(buffer, 0xFF, SM_SECTOR_SIZE);
+               return 0;
+       }
+
+       /* User might not need the oob, but we do for data vertification */
+       if (!oob)
+               oob = &tmp_oob;
+
+       ops.mode = ftl->smallpagenand ? MTD_OOB_RAW : MTD_OOB_PLACE;
+       ops.ooboffs = 0;
+       ops.ooblen = SM_OOB_SIZE;
+       ops.oobbuf = (void *)oob;
+       ops.len = SM_SECTOR_SIZE;
+       ops.datbuf = buffer;
+
+again:
+       if (try++) {
+               /* Avoid infinite recursion on CIS reads, sm_recheck_media
+                       won't help anyway */
+               if (zone == 0 && block == ftl->cis_block && boffset ==
+                       ftl->cis_boffset)
+                       return ret;
+
+               /* Test if media is stable */
+               if (try == 3 || sm_recheck_media(ftl))
+                       return ret;
+       }
+
+       /* Unfortunelly, oob read will _always_ succeed,
+               despite card removal..... */
+       ret = mtd->read_oob(mtd, sm_mkoffset(ftl, zone, block, boffset), &ops);
+
+       /* Test for unknown errors */
+       if (ret != 0 && ret != -EUCLEAN && ret != -EBADMSG) {
+               dbg("read of block %d at zone %d, failed due to error (%d)",
+                       block, zone, ret);
+               goto again;
+       }
+
+       /* Do a basic test on the oob, to guard against returned garbage */
+       if (oob->reserved != 0xFFFFFFFF && !is_power_of_2(~oob->reserved))
+               goto again;
+
+       /* This should never happen, unless there is a bug in the mtd driver */
+       WARN_ON(ops.oobretlen != SM_OOB_SIZE);
+       WARN_ON(buffer && ops.retlen != SM_SECTOR_SIZE);
+
+       if (!buffer)
+               return 0;
+
+       /* Test if sector marked as bad */
+       if (!sm_sector_valid(oob)) {
+               dbg("read of block %d at zone %d, failed because it is marked"
+                       " as bad" , block, zone);
+               goto again;
+       }
+
+       /* Test ECC*/
+       if (ret == -EBADMSG ||
+               (ftl->smallpagenand && sm_correct_sector(buffer, oob))) {
+
+               dbg("read of block %d at zone %d, failed due to ECC error",
+                       block, zone);
+               goto again;
+       }
+
+       return 0;
+}
+
+/* Writes a sector to media */
+static int sm_write_sector(struct sm_ftl *ftl,
+                          int zone, int block, int boffset,
+                          uint8_t *buffer, struct sm_oob *oob)
+{
+       struct mtd_oob_ops ops;
+       struct mtd_info *mtd = ftl->trans->mtd;
+       int ret;
+
+       BUG_ON(ftl->readonly);
+
+       if (zone == 0 && (block == ftl->cis_block || block == 0)) {
+               dbg("attempted to write the CIS!");
+               return -EIO;
+       }
+
+       if (ftl->unstable)
+               return -EIO;
+
+       ops.mode = ftl->smallpagenand ? MTD_OOB_RAW : MTD_OOB_PLACE;
+       ops.len = SM_SECTOR_SIZE;
+       ops.datbuf = buffer;
+       ops.ooboffs = 0;
+       ops.ooblen = SM_OOB_SIZE;
+       ops.oobbuf = (void *)oob;
+
+       ret = mtd->write_oob(mtd, sm_mkoffset(ftl, zone, block, boffset), &ops);
+
+       /* Now we assume that hardware will catch write bitflip errors */
+       /* If you are paranoid, use CONFIG_MTD_NAND_VERIFY_WRITE */
+
+       if (ret) {
+               dbg("write to block %d at zone %d, failed with error %d",
+                       block, zone, ret);
+
+               sm_recheck_media(ftl);
+               return ret;
+       }
+
+       /* This should never happen, unless there is a bug in the driver */
+       WARN_ON(ops.oobretlen != SM_OOB_SIZE);
+       WARN_ON(buffer && ops.retlen != SM_SECTOR_SIZE);
+
+       return 0;
+}
+
+/* ------------------------ block IO ------------------------------------- */
+
+/* Write a block using data and lba, and invalid sector bitmap */
+static int sm_write_block(struct sm_ftl *ftl, uint8_t *buf,
+                         int zone, int block, int lba,
+                         unsigned long invalid_bitmap)
+{
+       struct sm_oob oob;
+       int boffset;
+       int retry = 0;
+
+       /* Initialize the oob with requested values */
+       memset(&oob, 0xFF, SM_OOB_SIZE);
+       sm_write_lba(&oob, lba);
+restart:
+       if (ftl->unstable)
+               return -EIO;
+
+       for (boffset = 0; boffset < ftl->block_size;
+                               boffset += SM_SECTOR_SIZE) {
+
+               oob.data_status = 0xFF;
+
+               if (test_bit(boffset / SM_SECTOR_SIZE, &invalid_bitmap)) {
+
+                       sm_printk("sector %d of block at LBA %d of zone %d"
+                               " coudn't be read, marking it as invalid",
+                               boffset / SM_SECTOR_SIZE, lba, zone);
+
+                       oob.data_status = 0;
+               }
+
+#ifdef CONFIG_SM_FTL_MUSEUM
+               if (ftl->smallpagenand) {
+                       __nand_calculate_ecc(buf + boffset,
+                                               SM_SMALL_PAGE, oob.ecc1);
+
+                       __nand_calculate_ecc(buf + boffset + SM_SMALL_PAGE,
+                                               SM_SMALL_PAGE, oob.ecc2);
+               }
+#endif
+               if (!sm_write_sector(ftl, zone, block, boffset,
+                                                       buf + boffset, &oob))
+                       continue;
+
+               if (!retry) {
+
+                       /* If write fails. try to erase the block */
+                       /* This is safe, because we never write in blocks
+                               that contain valuable data.
+                       This is intended to repair block that are marked
+                       as erased, but that isn't fully erased*/
+
+                       if (sm_erase_block(ftl, zone, block, 0))
+                               return -EIO;
+
+                       retry = 1;
+                       goto restart;
+               } else {
+                       sm_mark_block_bad(ftl, zone, block);
+                       return -EIO;
+               }
+       }
+       return 0;
+}
+
+
+/* Mark whole block at offset 'offs' as bad. */
+static void sm_mark_block_bad(struct sm_ftl *ftl, int zone, int block)
+{
+       struct sm_oob oob;
+       int boffset;
+
+       memset(&oob, 0xFF, SM_OOB_SIZE);
+       oob.block_status = 0xF0;
+
+       if (ftl->unstable)
+               return;
+
+       if (sm_recheck_media(ftl))
+               return;
+
+       sm_printk("marking block %d of zone %d as bad", block, zone);
+
+       /* We aren't checking the return value, because we don't care */
+       /* This also fails on fake xD cards, but I guess these won't expose
+               any bad blocks till fail completly */
+       for (boffset = 0; boffset < ftl->block_size; boffset += SM_SECTOR_SIZE)
+               sm_write_sector(ftl, zone, block, boffset, NULL, &oob);
+}
+
+/*
+ * Erase a block within a zone
+ * If erase succedes, it updates free block fifo, otherwise marks block as bad
+ */
+static int sm_erase_block(struct sm_ftl *ftl, int zone_num, uint16_t block,
+                         int put_free)
+{
+       struct ftl_zone *zone = &ftl->zones[zone_num];
+       struct mtd_info *mtd = ftl->trans->mtd;
+       struct erase_info erase;
+
+       erase.mtd = mtd;
+       erase.callback = sm_erase_callback;
+       erase.addr = sm_mkoffset(ftl, zone_num, block, 0);
+       erase.len = ftl->block_size;
+       erase.priv = (u_long)ftl;
+
+       if (ftl->unstable)
+               return -EIO;
+
+       BUG_ON(ftl->readonly);
+
+       if (zone_num == 0 && (block == ftl->cis_block || block == 0)) {
+               sm_printk("attempted to erase the CIS!");
+               return -EIO;
+       }
+
+       if (mtd->erase(mtd, &erase)) {
+               sm_printk("erase of block %d in zone %d failed",
+                                                       block, zone_num);
+               goto error;
+       }
+
+       if (erase.state == MTD_ERASE_PENDING)
+               wait_for_completion(&ftl->erase_completion);
+
+       if (erase.state != MTD_ERASE_DONE) {
+               sm_printk("erase of block %d in zone %d failed after wait",
+                       block, zone_num);
+               goto error;
+       }
+
+       if (put_free)
+               kfifo_in(&zone->free_sectors,
+                       (const unsigned char *)&block, sizeof(block));
+
+       return 0;
+error:
+       sm_mark_block_bad(ftl, zone_num, block);
+       return -EIO;
+}
+
+static void sm_erase_callback(struct erase_info *self)
+{
+       struct sm_ftl *ftl = (struct sm_ftl *)self->priv;
+       complete(&ftl->erase_completion);
+}
+
+/* Throughtly test that block is valid. */
+static int sm_check_block(struct sm_ftl *ftl, int zone, int block)
+{
+       int boffset;
+       struct sm_oob oob;
+       int lbas[] = { -3, 0, 0, 0 };
+       int i = 0;
+       int test_lba;
+
+
+       /* First just check that block doesn't look fishy */
+       /* Only blocks that are valid or are sliced in two parts, are
+               accepted */
+       for (boffset = 0; boffset < ftl->block_size;
+                                       boffset += SM_SECTOR_SIZE) {
+
+               /* This shoudn't happen anyway */
+               if (sm_read_sector(ftl, zone, block, boffset, NULL, &oob))
+                       return -2;
+
+               test_lba = sm_read_lba(&oob);
+
+               if (lbas[i] != test_lba)
+                       lbas[++i] = test_lba;
+
+               /* If we found three different LBAs, something is fishy */
+               if (i == 3)
+                       return -EIO;
+       }
+
+       /* If the block is sliced (partialy erased usually) erase it */
+       if (i == 2) {
+               sm_erase_block(ftl, zone, block, 1);
+               return 1;
+       }
+
+       return 0;
+}
+
+/* ----------------- media scanning --------------------------------- */
+static const struct chs_entry chs_table[] = {
+       { 1,    125,  4,  4  },
+       { 2,    125,  4,  8  },
+       { 4,    250,  4,  8  },
+       { 8,    250,  4,  16 },
+       { 16,   500,  4,  16 },
+       { 32,   500,  8,  16 },
+       { 64,   500,  8,  32 },
+       { 128,  500,  16, 32 },
+       { 256,  1000, 16, 32 },
+       { 512,  1015, 32, 63 },
+       { 1024, 985,  33, 63 },
+       { 2048, 985,  33, 63 },
+       { 0 },
+};
+
+
+static const uint8_t cis_signature[] = {
+       0x01, 0x03, 0xD9, 0x01, 0xFF, 0x18, 0x02, 0xDF, 0x01, 0x20
+};
+/* Find out media parameters.
+ * This ideally has to be based on nand id, but for now device size is enough */
+int sm_get_media_info(struct sm_ftl *ftl, struct mtd_info *mtd)
+{
+       int i;
+       int size_in_megs = mtd->size / (1024 * 1024);
+
+       ftl->readonly = mtd->type == MTD_ROM;
+
+       /* Manual settings for very old devices */
+       ftl->zone_count = 1;
+       ftl->smallpagenand = 0;
+
+       switch (size_in_megs) {
+       case 1:
+               /* 1 MiB flash/rom SmartMedia card (256 byte pages)*/
+               ftl->zone_size = 256;
+               ftl->max_lba = 250;
+               ftl->block_size = 8 * SM_SECTOR_SIZE;
+               ftl->smallpagenand = 1;
+
+               break;
+       case 2:
+               /* 2 MiB flash SmartMedia (256 byte pages)*/
+               if (mtd->writesize == SM_SMALL_PAGE) {
+                       ftl->zone_size = 512;
+                       ftl->max_lba = 500;
+                       ftl->block_size = 8 * SM_SECTOR_SIZE;
+                       ftl->smallpagenand = 1;
+               /* 2 MiB rom SmartMedia */
+               } else {
+
+                       if (!ftl->readonly)
+                               return -ENODEV;
+
+                       ftl->zone_size = 256;
+                       ftl->max_lba = 250;
+                       ftl->block_size = 16 * SM_SECTOR_SIZE;
+               }
+               break;
+       case 4:
+               /* 4 MiB flash/rom SmartMedia device */
+               ftl->zone_size = 512;
+               ftl->max_lba = 500;
+               ftl->block_size = 16 * SM_SECTOR_SIZE;
+               break;
+       case 8:
+               /* 8 MiB flash/rom SmartMedia device */
+               ftl->zone_size = 1024;
+               ftl->max_lba = 1000;
+               ftl->block_size = 16 * SM_SECTOR_SIZE;
+       }
+
+       /* Minimum xD size is 16MiB. Also, all xD cards have standard zone
+          sizes. SmartMedia cards exist up to 128 MiB and have same layout*/
+       if (size_in_megs >= 16) {
+               ftl->zone_count = size_in_megs / 16;
+               ftl->zone_size = 1024;
+               ftl->max_lba = 1000;
+               ftl->block_size = 32 * SM_SECTOR_SIZE;
+       }
+
+       /* Test for proper write,erase and oob sizes */
+       if (mtd->erasesize > ftl->block_size)
+               return -ENODEV;
+
+       if (mtd->writesize > SM_SECTOR_SIZE)
+               return -ENODEV;
+
+       if (ftl->smallpagenand && mtd->oobsize < SM_SMALL_OOB_SIZE)
+               return -ENODEV;
+
+       if (!ftl->smallpagenand && mtd->oobsize < SM_OOB_SIZE)
+               return -ENODEV;
+
+       /* We use these functions for IO */
+       if (!mtd->read_oob || !mtd->write_oob)
+               return -ENODEV;
+
+       /* Find geometry information */
+       for (i = 0 ; i < ARRAY_SIZE(chs_table) ; i++) {
+               if (chs_table[i].size == size_in_megs) {
+                       ftl->cylinders = chs_table[i].cyl;
+                       ftl->heads = chs_table[i].head;
+                       ftl->sectors = chs_table[i].sec;
+                       return 0;
+               }
+       }
+
+       sm_printk("media has unknown size : %dMiB", size_in_megs);
+       ftl->cylinders = 985;
+       ftl->heads =  33;
+       ftl->sectors = 63;
+       return 0;
+}
+
+/* Validate the CIS */
+static int sm_read_cis(struct sm_ftl *ftl)
+{
+       struct sm_oob oob;
+
+       if (sm_read_sector(ftl,
+               0, ftl->cis_block, ftl->cis_boffset, ftl->cis_buffer, &oob))
+                       return -EIO;
+
+       if (!sm_sector_valid(&oob) || !sm_block_valid(&oob))
+               return -EIO;
+
+       if (!memcmp(ftl->cis_buffer + ftl->cis_page_offset,
+                       cis_signature, sizeof(cis_signature))) {
+               return 0;
+       }
+
+       return -EIO;
+}
+
+/* Scan the media for the CIS */
+static int sm_find_cis(struct sm_ftl *ftl)
+{
+       struct sm_oob oob;
+       int block, boffset;
+       int block_found = 0;
+       int cis_found = 0;
+
+       /* Search for first valid block */
+       for (block = 0 ; block < ftl->zone_size - ftl->max_lba ; block++) {
+
+               if (sm_read_sector(ftl, 0, block, 0, NULL, &oob))
+                       continue;
+
+               if (!sm_block_valid(&oob))
+                       continue;
+               block_found = 1;
+               break;
+       }
+
+       if (!block_found)
+               return -EIO;
+
+       /* Search for first valid sector in this block */
+       for (boffset = 0 ; boffset < ftl->block_size;
+                                               boffset += SM_SECTOR_SIZE) {
+
+               if (sm_read_sector(ftl, 0, block, boffset, NULL, &oob))
+                       continue;
+
+               if (!sm_sector_valid(&oob))
+                       continue;
+               break;
+       }
+
+       if (boffset == ftl->block_size)
+               return -EIO;
+
+       ftl->cis_block = block;
+       ftl->cis_boffset = boffset;
+       ftl->cis_page_offset = 0;
+
+       cis_found = !sm_read_cis(ftl);
+
+       if (!cis_found) {
+               ftl->cis_page_offset = SM_SMALL_PAGE;
+               cis_found = !sm_read_cis(ftl);
+       }
+
+       if (cis_found) {
+               dbg("CIS block found at offset %x",
+                       block * ftl->block_size +
+                               boffset + ftl->cis_page_offset);
+               return 0;
+       }
+       return -EIO;
+}
+
+/* Basic test to determine if underlying mtd device if functional */
+static int sm_recheck_media(struct sm_ftl *ftl)
+{
+       if (sm_read_cis(ftl)) {
+
+               if (!ftl->unstable) {
+                       sm_printk("media unstable, not allowing writes");
+                       ftl->unstable = 1;
+               }
+               return -EIO;
+       }
+       return 0;
+}
+
+/* Initialize a FTL zone */
+static int sm_init_zone(struct sm_ftl *ftl, int zone_num)
+{
+       struct ftl_zone *zone = &ftl->zones[zone_num];
+       struct sm_oob oob;
+       uint16_t block;
+       int lba;
+       int i = 0;
+
+       dbg("initializing zone %d", zone_num);
+
+       /* Allocate memory for FTL table */
+       zone->lba_to_phys_table = kmalloc(ftl->max_lba * 2, GFP_KERNEL);
+
+       if (!zone->lba_to_phys_table)
+               return -ENOMEM;
+       memset(zone->lba_to_phys_table, -1, ftl->max_lba * 2);
+
+
+       /* Allocate memory for free sectors FIFO */
+       if (kfifo_alloc(&zone->free_sectors, ftl->zone_size * 2, GFP_KERNEL)) {
+               kfree(zone->lba_to_phys_table);
+               return -ENOMEM;
+       }
+
+       /* Now scan the zone */
+       for (block = 0 ; block < ftl->zone_size ; block++) {
+
+               /* Skip blocks till the CIS (including) */
+               if (zone_num == 0 && block <= ftl->cis_block)
+                       continue;
+
+               /* Read the oob of first sector */
+               if (sm_read_sector(ftl, zone_num, block, 0, NULL, &oob))
+                       return -EIO;
+
+               /* Test to see if block is erased. It is enough to test
+                       first sector, because erase happens in one shot */
+               if (sm_block_erased(&oob)) {
+                       kfifo_in(&zone->free_sectors,
+                               (unsigned char *)&block, 2);
+                       continue;
+               }
+
+               /* If block is marked as bad, skip it */
+               /* This assumes we can trust first sector*/
+               /* However the way the block valid status is defined, ensures
+                       very low probability of failure here */
+               if (!sm_block_valid(&oob)) {
+                       dbg("PH %04d <-> <marked bad>", block);
+                       continue;
+               }
+
+
+               lba = sm_read_lba(&oob);
+
+               /* Invalid LBA means that block is damaged. */
+               /* We can try to erase it, or mark it as bad, but
+                       lets leave that to recovery application */
+               if (lba == -2 || lba >= ftl->max_lba) {
+                       dbg("PH %04d <-> LBA %04d(bad)", block, lba);
+                       continue;
+               }
+
+
+               /* If there is no collision,
+                       just put the sector in the FTL table */
+               if (zone->lba_to_phys_table[lba] < 0) {
+                       dbg_verbose("PH %04d <-> LBA %04d", block, lba);
+                       zone->lba_to_phys_table[lba] = block;
+                       continue;
+               }
+
+               sm_printk("collision"
+                       " of LBA %d between blocks %d and %d in zone %d",
+                       lba, zone->lba_to_phys_table[lba], block, zone_num);
+
+               /* Test that this block is valid*/
+               if (sm_check_block(ftl, zone_num, block))
+                       continue;
+
+               /* Test now the old block */
+               if (sm_check_block(ftl, zone_num,
+                                       zone->lba_to_phys_table[lba])) {
+                       zone->lba_to_phys_table[lba] = block;
+                       continue;
+               }
+
+               /* If both blocks are valid and share same LBA, it means that
+                       they hold different versions of same data. It not
+                       known which is more recent, thus just erase one of them
+               */
+               sm_printk("both blocks are valid, erasing the later");
+               sm_erase_block(ftl, zone_num, block, 1);
+       }
+
+       dbg("zone initialized");
+       zone->initialized = 1;
+
+       /* No free sectors, means that the zone is heavily damaged, write won't
+               work, but it can still can be (partially) read */
+       if (!kfifo_len(&zone->free_sectors)) {
+               sm_printk("no free blocks in zone %d", zone_num);
+               return 0;
+       }
+
+       /* Randomize first block we write to */
+       get_random_bytes(&i, 2);
+       i %= (kfifo_len(&zone->free_sectors) / 2);
+
+       while (i--) {
+               kfifo_out(&zone->free_sectors, (unsigned char *)&block, 2);
+               kfifo_in(&zone->free_sectors, (const unsigned char *)&block, 2);
+       }
+       return 0;
+}
+
+/* Get and automaticly initialize an FTL mapping for one zone */
+struct ftl_zone *sm_get_zone(struct sm_ftl *ftl, int zone_num)
+{
+       struct ftl_zone *zone;
+       int error;
+
+       BUG_ON(zone_num >= ftl->zone_count);
+       zone = &ftl->zones[zone_num];
+
+       if (!zone->initialized) {
+               error = sm_init_zone(ftl, zone_num);
+
+               if (error)
+                       return ERR_PTR(error);
+       }
+       return zone;
+}
+
+
+/* ----------------- cache handling ------------------------------------------*/
+
+/* Initialize the one block cache */
+void sm_cache_init(struct sm_ftl *ftl)
+{
+       ftl->cache_data_invalid_bitmap = 0xFFFFFFFF;
+       ftl->cache_clean = 1;
+       ftl->cache_zone = -1;
+       ftl->cache_block = -1;
+       /*memset(ftl->cache_data, 0xAA, ftl->block_size);*/
+}
+
+/* Put sector in one block cache */
+void sm_cache_put(struct sm_ftl *ftl, char *buffer, int boffset)
+{
+       memcpy(ftl->cache_data + boffset, buffer, SM_SECTOR_SIZE);
+       clear_bit(boffset / SM_SECTOR_SIZE, &ftl->cache_data_invalid_bitmap);
+       ftl->cache_clean = 0;
+}
+
+/* Read a sector from the cache */
+int sm_cache_get(struct sm_ftl *ftl, char *buffer, int boffset)
+{
+       if (test_bit(boffset / SM_SECTOR_SIZE,
+               &ftl->cache_data_invalid_bitmap))
+                       return -1;
+
+       memcpy(buffer, ftl->cache_data + boffset, SM_SECTOR_SIZE);
+       return 0;
+}
+
+/* Write the cache to hardware */
+int sm_cache_flush(struct sm_ftl *ftl)
+{
+       struct ftl_zone *zone;
+
+       int sector_num;
+       uint16_t write_sector;
+       int zone_num = ftl->cache_zone;
+       int block_num;
+
+       if (ftl->cache_clean)
+               return 0;
+
+       if (ftl->unstable)
+               return -EIO;
+
+       BUG_ON(zone_num < 0);
+       zone = &ftl->zones[zone_num];
+       block_num = zone->lba_to_phys_table[ftl->cache_block];
+
+
+       /* Try to read all unread areas of the cache block*/
+       for_each_bit(sector_num, &ftl->cache_data_invalid_bitmap,
+               ftl->block_size / SM_SECTOR_SIZE) {
+
+               if (!sm_read_sector(ftl,
+                       zone_num, block_num, sector_num * SM_SECTOR_SIZE,
+                       ftl->cache_data + sector_num * SM_SECTOR_SIZE, NULL))
+                               clear_bit(sector_num,
+                                       &ftl->cache_data_invalid_bitmap);
+       }
+restart:
+
+       if (ftl->unstable)
+               return -EIO;
+       /* No spare blocks */
+       /* We could still continue by erasing the current block,
+               but for such worn out media it doesn't worth the trouble,
+                       and the dangers */
+
+       if (!kfifo_len(&zone->free_sectors)) {
+               dbg("no free sectors for write!");
+               return -EIO;
+       }
+
+       kfifo_out(&zone->free_sectors, (unsigned char *)&write_sector, 2);
+
+       if (sm_write_block(ftl, ftl->cache_data, zone_num, write_sector,
+               ftl->cache_block, ftl->cache_data_invalid_bitmap))
+                       goto restart;
+
+       /* Update the FTL table */
+       zone->lba_to_phys_table[ftl->cache_block] = write_sector;
+
+       /* Write succesfull, so erase and free the old block */
+       if (block_num > 0)
+               sm_erase_block(ftl, zone_num, block_num, 1);
+
+       sm_cache_init(ftl);
+       return 0;
+}
+
+
+/* flush timer, runs a second after last write */
+static void sm_cache_flush_timer(unsigned long data)
+{
+       struct sm_ftl *ftl = (struct sm_ftl *)data;
+       queue_work(cache_flush_workqueue, &ftl->flush_work);
+}
+
+/* cache flush work, kicked by timer */
+static void sm_cache_flush_work(struct work_struct *work)
+{
+       struct sm_ftl *ftl = container_of(work, struct sm_ftl, flush_work);
+       mutex_lock(&ftl->mutex);
+       sm_cache_flush(ftl);
+       mutex_unlock(&ftl->mutex);
+       return;
+}
+
+/* ---------------- outside interface -------------------------------------- */
+
+/* outside interface: read a sector */
+static int sm_read(struct mtd_blktrans_dev *dev,
+                  unsigned long sect_no, char *buf)
+{
+       struct sm_ftl *ftl = dev->priv;
+       struct ftl_zone *zone;
+       int error = 0, in_cache = 0;
+       int zone_num, block, boffset;
+
+       sm_break_offset(ftl, sect_no << 9, &zone_num, &block, &boffset);
+       mutex_lock(&ftl->mutex);
+
+
+       zone = sm_get_zone(ftl, zone_num);
+       if (IS_ERR(zone)) {
+               error = PTR_ERR(zone);
+               goto unlock;
+       }
+
+       /* Have to look at cache first */
+       if (ftl->cache_zone == zone_num && ftl->cache_block == block) {
+               in_cache = 1;
+               if (!sm_cache_get(ftl, buf, boffset))
+                       goto unlock;
+       }
+
+       /* Translate the block and return if doesn't exist in the table */
+       block = zone->lba_to_phys_table[block];
+
+       if (block == -1) {
+               memset(buf, 0xFF, SM_SECTOR_SIZE);
+               goto unlock;
+       }
+
+       if (sm_read_sector(ftl, zone_num, block, boffset, buf, NULL)) {
+               error = -EIO;
+               goto unlock;
+       }
+
+       if (in_cache)
+               sm_cache_put(ftl, buf, boffset);
+unlock:
+       mutex_unlock(&ftl->mutex);
+       return error;
+}
+
+/* outside interface: write a sector */
+static int sm_write(struct mtd_blktrans_dev *dev,
+                               unsigned long sec_no, char *buf)
+{
+       struct sm_ftl *ftl = dev->priv;
+       struct ftl_zone *zone;
+       int error, zone_num, block, boffset;
+
+       BUG_ON(ftl->readonly);
+       sm_break_offset(ftl, sec_no << 9, &zone_num, &block, &boffset);
+
+       /* No need in flush thread running now */
+       del_timer(&ftl->timer);
+       mutex_lock(&ftl->mutex);
+
+       zone = sm_get_zone(ftl, zone_num);
+       if (IS_ERR(zone)) {
+               error = PTR_ERR(zone);
+               goto unlock;
+       }
+
+       /* If entry is not in cache, flush it */
+       if (ftl->cache_block != block || ftl->cache_zone != zone_num) {
+
+               error = sm_cache_flush(ftl);
+               if (error)
+                       goto unlock;
+
+               ftl->cache_block = block;
+               ftl->cache_zone = zone_num;
+       }
+
+       sm_cache_put(ftl, buf, boffset);
+unlock:
+       mod_timer(&ftl->timer, jiffies + msecs_to_jiffies(cache_timeout));
+       mutex_unlock(&ftl->mutex);
+       return error;
+}
+
+/* outside interface: flush everything */
+static int sm_flush(struct mtd_blktrans_dev *dev)
+{
+       struct sm_ftl *ftl = dev->priv;
+       int retval;
+
+       mutex_lock(&ftl->mutex);
+       retval =  sm_cache_flush(ftl);
+       mutex_unlock(&ftl->mutex);
+       return retval;
+}
+
+/* outside interface: device is released */
+static int sm_release(struct mtd_blktrans_dev *dev)
+{
+       struct sm_ftl *ftl = dev->priv;
+
+       mutex_lock(&ftl->mutex);
+       del_timer_sync(&ftl->timer);
+       cancel_work_sync(&ftl->flush_work);
+       sm_cache_flush(ftl);
+       mutex_unlock(&ftl->mutex);
+       return 0;
+}
+
+/* outside interface: get geometry */
+static int sm_getgeo(struct mtd_blktrans_dev *dev, struct hd_geometry *geo)
+{
+       struct sm_ftl *ftl = dev->priv;
+       geo->heads = ftl->heads;
+       geo->sectors = ftl->sectors;
+       geo->cylinders = ftl->cylinders;
+       return 0;
+}
+
+/* external interface: main initialization function */
+static void sm_add_mtd(struct mtd_blktrans_ops *tr, struct mtd_info *mtd)
+{
+       struct mtd_blktrans_dev *trans;
+       struct sm_ftl *ftl;
+
+       /* Allocate & initialize our private structure */
+       ftl = kzalloc(sizeof(struct sm_ftl), GFP_KERNEL);
+       if (!ftl)
+               goto error1;
+
+
+       mutex_init(&ftl->mutex);
+       setup_timer(&ftl->timer, sm_cache_flush_timer, (unsigned long)ftl);
+       INIT_WORK(&ftl->flush_work, sm_cache_flush_work);
+       init_completion(&ftl->erase_completion);
+
+       /* Read media information */
+       if (sm_get_media_info(ftl, mtd)) {
+               dbg("found unsupported mtd device, aborting");
+               goto error2;
+       }
+
+
+       /* Allocate temporary CIS buffer for read retry support */
+       ftl->cis_buffer = kzalloc(SM_SECTOR_SIZE, GFP_KERNEL);
+       if (!ftl->cis_buffer)
+               goto error2;
+
+       /* Allocate zone array, it will be initialized on demand */
+       ftl->zones = kzalloc(sizeof(struct ftl_zone) * ftl->zone_count,
+                                                               GFP_KERNEL);
+       if (!ftl->zones)
+               goto error3;
+
+       /* Allocate the cache*/
+       ftl->cache_data = kzalloc(ftl->block_size, GFP_KERNEL);
+
+       if (!ftl->cache_data)
+               goto error4;
+
+       sm_cache_init(ftl);
+
+
+       /* Allocate upper layer structure and initialize it */
+       trans = kzalloc(sizeof(struct mtd_blktrans_dev), GFP_KERNEL);
+       if (!trans)
+               goto error5;
+
+       ftl->trans = trans;
+       trans->priv = ftl;
+
+       trans->tr = tr;
+       trans->mtd = mtd;
+       trans->devnum = -1;
+       trans->size = (ftl->block_size * ftl->max_lba * ftl->zone_count) >> 9;
+       trans->readonly = ftl->readonly;
+
+       if (sm_find_cis(ftl)) {
+               dbg("CIS not found on mtd device, aborting");
+               goto error6;
+       }
+
+       ftl->disk_attributes = sm_create_sysfs_attributes(ftl);
+       trans->disk_attributes = ftl->disk_attributes;
+
+       sm_printk("Found %d MiB xD/SmartMedia FTL on mtd%d",
+               (int)(mtd->size / (1024 * 1024)), mtd->index);
+
+       dbg("FTL layout:");
+       dbg("%d zone(s), each consists of %d blocks (+%d spares)",
+               ftl->zone_count, ftl->max_lba,
+               ftl->zone_size - ftl->max_lba);
+       dbg("each block consists of %d bytes",
+               ftl->block_size);
+
+
+       /* Register device*/
+       if (add_mtd_blktrans_dev(trans)) {
+               dbg("error in mtdblktrans layer");
+               goto error6;
+       }
+       return;
+error6:
+       kfree(trans);
+error5:
+       kfree(ftl->cache_data);
+error4:
+       kfree(ftl->zones);
+error3:
+       kfree(ftl->cis_buffer);
+error2:
+       kfree(ftl);
+error1:
+       return;
+}
+
+/* main interface: device {surprise,} removal */
+static void sm_remove_dev(struct mtd_blktrans_dev *dev)
+{
+       struct sm_ftl *ftl = dev->priv;
+       int i;
+
+       del_mtd_blktrans_dev(dev);
+       ftl->trans = NULL;
+
+       for (i = 0 ; i < ftl->zone_count; i++) {
+
+               if (!ftl->zones[i].initialized)
+                       continue;
+
+               kfree(ftl->zones[i].lba_to_phys_table);
+               kfifo_free(&ftl->zones[i].free_sectors);
+       }
+
+       sm_delete_sysfs_attributes(ftl);
+       kfree(ftl->cis_buffer);
+       kfree(ftl->zones);
+       kfree(ftl->cache_data);
+       kfree(ftl);
+}
+
+static struct mtd_blktrans_ops sm_ftl_ops = {
+       .name           = "smblk",
+       .major          = -1,
+       .part_bits      = SM_FTL_PARTN_BITS,
+       .blksize        = SM_SECTOR_SIZE,
+       .getgeo         = sm_getgeo,
+
+       .add_mtd        = sm_add_mtd,
+       .remove_dev     = sm_remove_dev,
+
+       .readsect       = sm_read,
+       .writesect      = sm_write,
+
+       .flush          = sm_flush,
+       .release        = sm_release,
+
+       .owner          = THIS_MODULE,
+};
+
+static __init int sm_module_init(void)
+{
+       int error = 0;
+       cache_flush_workqueue = create_freezeable_workqueue("smflush");
+
+       if (IS_ERR(cache_flush_workqueue))
+               return PTR_ERR(cache_flush_workqueue);
+
+       error = register_mtd_blktrans(&sm_ftl_ops);
+       if (error)
+               destroy_workqueue(cache_flush_workqueue);
+       return error;
+
+}
+
+static void __exit sm_module_exit(void)
+{
+       destroy_workqueue(cache_flush_workqueue);
+       deregister_mtd_blktrans(&sm_ftl_ops);
+}
+
+module_init(sm_module_init);
+module_exit(sm_module_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Maxim Levitsky <maximlevitsky@gmail.com>");
+MODULE_DESCRIPTION("Smartmedia/xD mtd translation layer");
diff --git a/drivers/mtd/sm_ftl.h b/drivers/mtd/sm_ftl.h
new file mode 100644 (file)
index 0000000..e30e48e
--- /dev/null
@@ -0,0 +1,94 @@
+/*
+ * Copyright © 2009 - Maxim Levitsky
+ * SmartMedia/xD translation layer
+ *
+ * Based loosly on ssfdc.c which is
+ *  © 2005 Eptar srl
+ *  Author: Claudio Lanconelli <lanconelli.claudio@eptar.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/mtd/blktrans.h>
+#include <linux/kfifo.h>
+#include <linux/sched.h>
+#include <linux/completion.h>
+#include <linux/mtd/mtd.h>
+
+
+
+struct ftl_zone {
+       int initialized;
+       int16_t *lba_to_phys_table;             /* LBA to physical table */
+       struct kfifo free_sectors;      /* queue of free sectors */
+};
+
+struct sm_ftl {
+       struct mtd_blktrans_dev *trans;
+
+       struct mutex mutex;             /* protects the structure */
+       struct ftl_zone *zones;         /* FTL tables for each zone */
+
+       /* Media information */
+       int block_size;                 /* block size in bytes */
+       int zone_size;                  /* zone size in blocks */
+       int zone_count;                 /* number of zones */
+       int max_lba;                    /* maximum lba in a zone */
+       int smallpagenand;              /* 256 bytes/page nand */
+       int readonly;                   /* is FS readonly */
+       int unstable;
+       int cis_block;                  /* CIS block location */
+       int cis_boffset;                /* CIS offset in the block */
+       int cis_page_offset;            /* CIS offset in the page */
+       void *cis_buffer;               /* tmp buffer for cis reads */
+
+       /* Cache */
+       int cache_block;                /* block number of cached block */
+       int cache_zone;                 /* zone of cached block */
+       unsigned char *cache_data;      /* cached block data */
+       long unsigned int cache_data_invalid_bitmap;
+       int cache_clean;
+       struct work_struct flush_work;
+       struct timer_list timer;
+
+       /* Async erase stuff */
+       struct completion erase_completion;
+
+       /* Geometry stuff */
+       int heads;
+       int sectors;
+       int cylinders;
+
+       struct attribute_group *disk_attributes;
+};
+
+struct chs_entry {
+       unsigned long size;
+       unsigned short cyl;
+       unsigned char head;
+       unsigned char sec;
+};
+
+
+#define SM_FTL_PARTN_BITS      3
+
+#define sm_printk(format, ...) \
+       printk(KERN_WARNING "sm_ftl" ": " format "\n", ## __VA_ARGS__)
+
+#define dbg(format, ...) \
+       if (debug) \
+               printk(KERN_DEBUG "sm_ftl" ": " format "\n", ## __VA_ARGS__)
+
+#define dbg_verbose(format, ...) \
+       if (debug > 1) \
+               printk(KERN_DEBUG "sm_ftl" ": " format "\n", ## __VA_ARGS__)
+
+
+static void sm_erase_callback(struct erase_info *self);
+static int sm_erase_block(struct sm_ftl *ftl, int zone_num, uint16_t block,
+                                                               int put_free);
+static void sm_mark_block_bad(struct sm_ftl *ftl, int zone_num, int block);
+
+static int sm_recheck_media(struct sm_ftl *ftl);