diff options
author | David A. Marlin <dmarlin@redhat.com> | 2005-01-17 18:35:25 +0000 |
---|---|---|
committer | Thomas Gleixner <tglx@mtd.linutronix.de> | 2005-05-23 11:26:45 +0200 |
commit | 30f464b74b51127b9b9a170157b75c7e8e80d2c4 (patch) | |
tree | 95a9e2f352612a4bfb94a276c1eaf21b0aec8f25 /drivers | |
parent | 28a48de72b876af794853593cc1412119ada9efc (diff) |
[MTD] NAND workaround for AG-AND disturb issue. AG-AND recovery
Added workaround for Renesas AG-AND chips "disturb" issue
for Bad Block Table.
Added support for the device recovery command sequence
for Renesas AG-AND chips.
Signed-off-by: David A. Marlin <dmarlin@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Diffstat (limited to 'drivers')
-rw-r--r-- | drivers/mtd/nand/nand_base.c | 74 |
1 files changed, 69 insertions, 5 deletions
diff --git a/drivers/mtd/nand/nand_base.c b/drivers/mtd/nand/nand_base.c index 44d5b128911..2ac452e3ad6 100644 --- a/drivers/mtd/nand/nand_base.c +++ b/drivers/mtd/nand/nand_base.c @@ -28,6 +28,20 @@ * among multiple independend devices. Suggestions and initial patch * from Ben Dooks <ben-mtd@fluff.org> * + * 12-05-2004 dmarlin: add workaround for Renesas AG-AND chips "disturb" issue. + * Basically, any block not rewritten may lose data when surrounding blocks + * are rewritten many times. JFFS2 ensures this doesn't happen for blocks + * it uses, but the Bad Block Table(s) may not be rewritten. To ensure they + * do not lose data, force them to be rewritten when some of the surrounding + * blocks are erased. Rather than tracking a specific nearby block (which + * could itself go bad), use a page address 'mask' to select several blocks + * in the same area, and rewrite the BBT when any of them are erased. + * + * 01-03-2005 dmarlin: added support for the device recovery command sequence for Renesas + * AG-AND chips. If there was a sudden loss of power during an erase operation, + * a "device recovery" operation must be performed when power is restored + * to ensure correct operation. + * * Credits: * David Woodhouse for adding multichip support * @@ -41,7 +55,7 @@ * The AG-AND chips have nice features for speed improvement, * which are not supported yet. Read / program 4 pages in one go. * - * $Id: nand_base.c,v 1.126 2004/12/13 11:22:25 lavinen Exp $ + * $Id: nand_base.c,v 1.127 2005/01/17 18:35:22 dmarlin Exp $ * * 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 @@ -619,7 +633,7 @@ static void nand_command_lp (struct mtd_info *mtd, unsigned command, int column, /* Begin command latch cycle */ this->hwcontrol(mtd, NAND_CTL_SETCLE); /* Write out the command to the device. */ - this->write_byte(mtd, command); + this->write_byte(mtd, (command & 0xff)); /* End command latch cycle */ this->hwcontrol(mtd, NAND_CTL_CLRCLE); @@ -647,8 +661,8 @@ static void nand_command_lp (struct mtd_info *mtd, unsigned command, int column, /* * program and erase have their own busy handlers - * status and sequential in needs no delay - */ + * status, sequential in, and deplete1 need no delay + */ switch (command) { case NAND_CMD_CACHEDPROG: @@ -657,8 +671,19 @@ static void nand_command_lp (struct mtd_info *mtd, unsigned command, int column, case NAND_CMD_ERASE2: case NAND_CMD_SEQIN: case NAND_CMD_STATUS: + case NAND_CMD_DEPLETE1: return; + /* + * read error status commands require only a short delay + */ + case NAND_CMD_STATUS_ERROR: + case NAND_CMD_STATUS_ERROR0: + case NAND_CMD_STATUS_ERROR1: + case NAND_CMD_STATUS_ERROR2: + case NAND_CMD_STATUS_ERROR3: + udelay(this->chip_delay); + return; case NAND_CMD_RESET: if (this->dev_ready) @@ -1051,7 +1076,7 @@ static int nand_read_ecc (struct mtd_info *mtd, loff_t from, size_t len, } /* Grab the lock and see if the device is available */ - nand_get_device (this, mtd ,FL_READING); + nand_get_device (this, mtd, FL_READING); /* use userspace supplied oobinfo, if zero */ if (oobsel == NULL) @@ -1987,6 +2012,7 @@ static int nand_erase (struct mtd_info *mtd, struct erase_info *instr) return nand_erase_nand (mtd, instr, 0); } +#define BBT_PAGE_MASK 0xffffff3f /** * nand_erase_intern - [NAND Interface] erase block(s) * @mtd: MTD device structure @@ -1999,6 +2025,10 @@ int nand_erase_nand (struct mtd_info *mtd, struct erase_info *instr, int allowbb { int page, len, status, pages_per_block, ret, chipnr; struct nand_chip *this = mtd->priv; + int rewrite_bbt[NAND_MAX_CHIPS]={0}; /* flags to indicate the page, if bbt needs to be rewritten. */ + unsigned int bbt_masked_page; /* bbt mask to compare to page being erased. */ + /* It is used to see if the current page is in the same */ + /* 256 block group and the same bank as the bbt. */ DEBUG (MTD_DEBUG_LEVEL3, "nand_erase: start = 0x%08x, len = %i\n", (unsigned int) instr->addr, (unsigned int) instr->len); @@ -2044,6 +2074,13 @@ int nand_erase_nand (struct mtd_info *mtd, struct erase_info *instr, int allowbb goto erase_exit; } + /* if BBT requires refresh, set the BBT page mask to see if the BBT should be rewritten */ + if (this->options & BBT_AUTO_REFRESH) { + bbt_masked_page = this->bbt_td->pages[chipnr] & BBT_PAGE_MASK; + } else { + bbt_masked_page = 0xffffffff; /* should not match anything */ + } + /* Loop through the pages */ len = instr->len; @@ -2073,6 +2110,14 @@ int nand_erase_nand (struct mtd_info *mtd, struct erase_info *instr, int allowbb instr->fail_addr = (page << this->page_shift); goto erase_exit; } + + /* if BBT requires refresh, set the BBT rewrite flag to the page being erased */ + if (this->options & BBT_AUTO_REFRESH) { + if (((page & BBT_PAGE_MASK) == bbt_masked_page) && + (page != this->bbt_td->pages[chipnr])) { + rewrite_bbt[chipnr] = (page << this->page_shift); + } + } /* Increment page address and decrement length */ len -= (1 << this->phys_erase_shift); @@ -2083,6 +2128,13 @@ int nand_erase_nand (struct mtd_info *mtd, struct erase_info *instr, int allowbb chipnr++; this->select_chip(mtd, -1); this->select_chip(mtd, chipnr); + + /* if BBT requires refresh and BBT-PERCHIP, + * set the BBT page mask to see if this BBT should be rewritten */ + if ((this->options & BBT_AUTO_REFRESH) && (this->bbt_td->options & NAND_BBT_PERCHIP)) { + bbt_masked_page = this->bbt_td->pages[chipnr] & BBT_PAGE_MASK; + } + } } instr->state = MTD_ERASE_DONE; @@ -2097,6 +2149,18 @@ erase_exit: /* Deselect and wake up anyone waiting on the device */ nand_release_device(mtd); + /* if BBT requires refresh and erase was successful, rewrite any selected bad block tables */ + if ((this->options & BBT_AUTO_REFRESH) && (!ret)) { + for (chipnr = 0; chipnr < this->numchips; chipnr++) { + if (rewrite_bbt[chipnr]) { + /* update the BBT for chip */ + DEBUG (MTD_DEBUG_LEVEL0, "nand_erase_nand: nand_update_bbt (%d:0x%0x 0x%0x)\n", + chipnr, rewrite_bbt[chipnr], this->bbt_td->pages[chipnr]); + nand_update_bbt (mtd, rewrite_bbt[chipnr]); + } + } + } + /* Return more or less happy */ return ret; } |