Merge branch 'for-next/mtd-nand'

1 files changed, 431 insertions, 142 deletions

diff --git a/drivers/mtd/nand/nand_ecc.c b/drivers/mtd/nand/nand_ecc.c
index fd6ad7edc8..58fb335bb4 100644
--- a/drivers/mtd/nand/nand_ecc.c
+++ b/drivers/mtd/nand/nand_ecc.c
@@ -1,194 +1,483 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
 /*
- * This file contains an ECC algorithm from Toshiba that detects and
- * corrects 1 bit errors in a 256 byte block of data.
+ * This file contains an ECC algorithm that detects and corrects 1 bit
+ * errors in a 256 byte block of data.
  *
- * drivers/mtd/nand/nand_ecc.c
+ * Copyright © 2008 Koninklijke Philips Electronics NV.
+ *                  Author: Frans Meulenbroeks
  *
- * Copyright (C) 2000-2004 Steven J. Hill (sjhill@realitydiluted.com)
- *                         Toshiba America Electronics Components, Inc.
+ * Completely replaces the previous ECC implementation which was written by:
+ *   Steven J. Hill (sjhill@realitydiluted.com)
+ *   Thomas Gleixner (tglx@linutronix.de)
  *
- * Copyright (C) 2006 Thomas Gleixner <tglx@linutronix.de>
- *
- * $Id: nand_ecc.c,v 1.15 2005/11/07 11:14:30 gleixner Exp $
- *
- * This file is free software; you can redistribute it and/or modify it
- * under the terms of the GNU General Public License as published by the
- * Free Software Foundation; either version 2 or (at your option) any
- * later version.
- *
- * This file is distributed in the hope that it will be useful, but WITHOUT
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
- * for more details.
- *
- * As a special exception, if other files instantiate templates or use
- * macros or inline functions from these files, or you compile these
- * files and link them with other works to produce a work based on these
- * files, these files do not by themselves cause the resulting work to be
- * covered by the GNU General Public License. However the source code for
- * these files must still be made available in accordance with section (3)
- * of the GNU General Public License.
- *
- * This exception does not invalidate any other reasons why a work based on
- * this file might be covered by the GNU General Public License.
+ * Information on how this algorithm works and how it was developed
+ * can be found in Documentation/driver-api/mtd/nand_ecc.rst
  */
 
 #include <linux/types.h>
-#include <common.h>
-#include <errno.h>
+#include <linux/kernel.h>
+#include <linux/mtd/mtd.h>
+#include <linux/mtd/rawnand.h>
 #include <linux/mtd/nand_ecc.h>
+#include <asm/byteorder.h>
+
+/*
+ * invparity is a 256 byte table that contains the odd parity
+ * for each byte. So if the number of bits in a byte is even,
+ * the array element is 1, and when the number of bits is odd
+ * the array eleemnt is 0.
+ */
+static const char invparity[256] = {
+	1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 1,
+	0, 1, 1, 0, 1, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0,
+	0, 1, 1, 0, 1, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0,
+	1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 1,
+	0, 1, 1, 0, 1, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0,
+	1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 1,
+	1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 1,
+	0, 1, 1, 0, 1, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0,
+	0, 1, 1, 0, 1, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0,
+	1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 1,
+	1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 1,
+	0, 1, 1, 0, 1, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0,
+	1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 1,
+	0, 1, 1, 0, 1, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0,
+	0, 1, 1, 0, 1, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0,
+	1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 1
+};
+
+/*
+ * bitsperbyte contains the number of bits per byte
+ * this is only used for testing and repairing parity
+ * (a precalculated value slightly improves performance)
+ */
+static const char bitsperbyte[256] = {
+	0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4,
+	1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
+	1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
+	2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
+	1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
+	2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
+	2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
+	3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
+	1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
+	2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
+	2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
+	3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
+	2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
+	3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
+	3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
+	4, 5, 5, 6, 5, 6, 6, 7, 5, 6, 6, 7, 6, 7, 7, 8,
+};
 
 /*
- * Pre-calculated 256-way 1 byte column parity
+ * addressbits is a lookup table to filter out the bits from the xor-ed
+ * ECC data that identify the faulty location.
+ * this is only used for repairing parity
+ * see the comments in nand_correct_data for more details
  */
-static const u_char nand_ecc_precalc_table[] = {
-	0x00, 0x55, 0x56, 0x03, 0x59, 0x0c, 0x0f, 0x5a, 0x5a, 0x0f, 0x0c, 0x59, 0x03, 0x56, 0x55, 0x00,
-	0x65, 0x30, 0x33, 0x66, 0x3c, 0x69, 0x6a, 0x3f, 0x3f, 0x6a, 0x69, 0x3c, 0x66, 0x33, 0x30, 0x65,
-	0x66, 0x33, 0x30, 0x65, 0x3f, 0x6a, 0x69, 0x3c, 0x3c, 0x69, 0x6a, 0x3f, 0x65, 0x30, 0x33, 0x66,
-	0x03, 0x56, 0x55, 0x00, 0x5a, 0x0f, 0x0c, 0x59, 0x59, 0x0c, 0x0f, 0x5a, 0x00, 0x55, 0x56, 0x03,
-	0x69, 0x3c, 0x3f, 0x6a, 0x30, 0x65, 0x66, 0x33, 0x33, 0x66, 0x65, 0x30, 0x6a, 0x3f, 0x3c, 0x69,
-	0x0c, 0x59, 0x5a, 0x0f, 0x55, 0x00, 0x03, 0x56, 0x56, 0x03, 0x00, 0x55, 0x0f, 0x5a, 0x59, 0x0c,
-	0x0f, 0x5a, 0x59, 0x0c, 0x56, 0x03, 0x00, 0x55, 0x55, 0x00, 0x03, 0x56, 0x0c, 0x59, 0x5a, 0x0f,
-	0x6a, 0x3f, 0x3c, 0x69, 0x33, 0x66, 0x65, 0x30, 0x30, 0x65, 0x66, 0x33, 0x69, 0x3c, 0x3f, 0x6a,
-	0x6a, 0x3f, 0x3c, 0x69, 0x33, 0x66, 0x65, 0x30, 0x30, 0x65, 0x66, 0x33, 0x69, 0x3c, 0x3f, 0x6a,
-	0x0f, 0x5a, 0x59, 0x0c, 0x56, 0x03, 0x00, 0x55, 0x55, 0x00, 0x03, 0x56, 0x0c, 0x59, 0x5a, 0x0f,
-	0x0c, 0x59, 0x5a, 0x0f, 0x55, 0x00, 0x03, 0x56, 0x56, 0x03, 0x00, 0x55, 0x0f, 0x5a, 0x59, 0x0c,
-	0x69, 0x3c, 0x3f, 0x6a, 0x30, 0x65, 0x66, 0x33, 0x33, 0x66, 0x65, 0x30, 0x6a, 0x3f, 0x3c, 0x69,
-	0x03, 0x56, 0x55, 0x00, 0x5a, 0x0f, 0x0c, 0x59, 0x59, 0x0c, 0x0f, 0x5a, 0x00, 0x55, 0x56, 0x03,
-	0x66, 0x33, 0x30, 0x65, 0x3f, 0x6a, 0x69, 0x3c, 0x3c, 0x69, 0x6a, 0x3f, 0x65, 0x30, 0x33, 0x66,
-	0x65, 0x30, 0x33, 0x66, 0x3c, 0x69, 0x6a, 0x3f, 0x3f, 0x6a, 0x69, 0x3c, 0x66, 0x33, 0x30, 0x65,
-	0x00, 0x55, 0x56, 0x03, 0x59, 0x0c, 0x0f, 0x5a, 0x5a, 0x0f, 0x0c, 0x59, 0x03, 0x56, 0x55, 0x00
+static const char addressbits[256] = {
+	0x00, 0x00, 0x01, 0x01, 0x00, 0x00, 0x01, 0x01,
+	0x02, 0x02, 0x03, 0x03, 0x02, 0x02, 0x03, 0x03,
+	0x00, 0x00, 0x01, 0x01, 0x00, 0x00, 0x01, 0x01,
+	0x02, 0x02, 0x03, 0x03, 0x02, 0x02, 0x03, 0x03,
+	0x04, 0x04, 0x05, 0x05, 0x04, 0x04, 0x05, 0x05,
+	0x06, 0x06, 0x07, 0x07, 0x06, 0x06, 0x07, 0x07,
+	0x04, 0x04, 0x05, 0x05, 0x04, 0x04, 0x05, 0x05,
+	0x06, 0x06, 0x07, 0x07, 0x06, 0x06, 0x07, 0x07,
+	0x00, 0x00, 0x01, 0x01, 0x00, 0x00, 0x01, 0x01,
+	0x02, 0x02, 0x03, 0x03, 0x02, 0x02, 0x03, 0x03,
+	0x00, 0x00, 0x01, 0x01, 0x00, 0x00, 0x01, 0x01,
+	0x02, 0x02, 0x03, 0x03, 0x02, 0x02, 0x03, 0x03,
+	0x04, 0x04, 0x05, 0x05, 0x04, 0x04, 0x05, 0x05,
+	0x06, 0x06, 0x07, 0x07, 0x06, 0x06, 0x07, 0x07,
+	0x04, 0x04, 0x05, 0x05, 0x04, 0x04, 0x05, 0x05,
+	0x06, 0x06, 0x07, 0x07, 0x06, 0x06, 0x07, 0x07,
+	0x08, 0x08, 0x09, 0x09, 0x08, 0x08, 0x09, 0x09,
+	0x0a, 0x0a, 0x0b, 0x0b, 0x0a, 0x0a, 0x0b, 0x0b,
+	0x08, 0x08, 0x09, 0x09, 0x08, 0x08, 0x09, 0x09,
+	0x0a, 0x0a, 0x0b, 0x0b, 0x0a, 0x0a, 0x0b, 0x0b,
+	0x0c, 0x0c, 0x0d, 0x0d, 0x0c, 0x0c, 0x0d, 0x0d,
+	0x0e, 0x0e, 0x0f, 0x0f, 0x0e, 0x0e, 0x0f, 0x0f,
+	0x0c, 0x0c, 0x0d, 0x0d, 0x0c, 0x0c, 0x0d, 0x0d,
+	0x0e, 0x0e, 0x0f, 0x0f, 0x0e, 0x0e, 0x0f, 0x0f,
+	0x08, 0x08, 0x09, 0x09, 0x08, 0x08, 0x09, 0x09,
+	0x0a, 0x0a, 0x0b, 0x0b, 0x0a, 0x0a, 0x0b, 0x0b,
+	0x08, 0x08, 0x09, 0x09, 0x08, 0x08, 0x09, 0x09,
+	0x0a, 0x0a, 0x0b, 0x0b, 0x0a, 0x0a, 0x0b, 0x0b,
+	0x0c, 0x0c, 0x0d, 0x0d, 0x0c, 0x0c, 0x0d, 0x0d,
+	0x0e, 0x0e, 0x0f, 0x0f, 0x0e, 0x0e, 0x0f, 0x0f,
+	0x0c, 0x0c, 0x0d, 0x0d, 0x0c, 0x0c, 0x0d, 0x0d,
+	0x0e, 0x0e, 0x0f, 0x0f, 0x0e, 0x0e, 0x0f, 0x0f
 };
 
 /**
- * nand_calculate_ecc - [NAND Interface] Calculate 3-byte ECC for 256-byte block
- * @mtd:	MTD block structure
- * @dat:	raw data
- * @ecc_code:	buffer for ECC
+ * __nand_calculate_ecc - [NAND Interface] Calculate 3-byte ECC for 256/512-byte
+ *			 block
+ * @buf:	input buffer with raw data
+ * @eccsize:	data bytes per ECC step (256 or 512)
+ * @code:	output buffer with ECC
+ * @sm_order:	Smart Media byte ordering
  */
-int nand_calculate_ecc(struct mtd_info *mtd, const u_char *dat,
-		       u_char *ecc_code)
+void __nand_calculate_ecc(const unsigned char *buf, unsigned int eccsize,
+			  unsigned char *code, bool sm_order)
 {
-	uint8_t idx, reg1, reg2, reg3, tmp1, tmp2;
 	int i;
+	const uint32_t *bp = (uint32_t *)buf;
+	/* 256 or 512 bytes/ecc  */
+	const uint32_t eccsize_mult = eccsize >> 8;
+	uint32_t cur;		/* current value in buffer */
+	/* rp0..rp15..rp17 are the various accumulated parities (per byte) */
+	uint32_t rp0, rp1, rp2, rp3, rp4, rp5, rp6, rp7;
+	uint32_t rp8, rp9, rp10, rp11, rp12, rp13, rp14, rp15, rp16;
+	uint32_t rp17 = 0;
+	uint32_t par;		/* the cumulative parity for all data */
+	uint32_t tmppar;	/* the cumulative parity for this iteration;
+				   for rp12, rp14 and rp16 at the end of the
+				   loop */
 
-	/* Initialize variables */
-	reg1 = reg2 = reg3 = 0;
+	par = 0;
+	rp4 = 0;
+	rp6 = 0;
+	rp8 = 0;
+	rp10 = 0;
+	rp12 = 0;
+	rp14 = 0;
+	rp16 = 0;
 
-	/* Build up column parity */
-	for(i = 0; i < 256; i++) {
-		/* Get CP0 - CP5 from table */
-		idx = nand_ecc_precalc_table[*dat++];
-		reg1 ^= (idx & 0x3f);
+	/*
+	 * The loop is unrolled a number of times;
+	 * This avoids if statements to decide on which rp value to update
+	 * Also we process the data by longwords.
+	 * Note: passing unaligned data might give a performance penalty.
+	 * It is assumed that the buffers are aligned.
+	 * tmppar is the cumulative sum of this iteration.
+	 * needed for calculating rp12, rp14, rp16 and par
+	 * also used as a performance improvement for rp6, rp8 and rp10
+	 */
+	for (i = 0; i < eccsize_mult << 2; i++) {
+		cur = *bp++;
+		tmppar = cur;
+		rp4 ^= cur;
+		cur = *bp++;
+		tmppar ^= cur;
+		rp6 ^= tmppar;
+		cur = *bp++;
+		tmppar ^= cur;
+		rp4 ^= cur;
+		cur = *bp++;
+		tmppar ^= cur;
+		rp8 ^= tmppar;
 
-		/* All bit XOR = 1 ? */
-		if (idx & 0x40) {
-			reg3 ^= (uint8_t) i;
-			reg2 ^= ~((uint8_t) i);
-		}
+		cur = *bp++;
+		tmppar ^= cur;
+		rp4 ^= cur;
+		rp6 ^= cur;
+		cur = *bp++;
+		tmppar ^= cur;
+		rp6 ^= cur;
+		cur = *bp++;
+		tmppar ^= cur;
+		rp4 ^= cur;
+		cur = *bp++;
+		tmppar ^= cur;
+		rp10 ^= tmppar;
+
+		cur = *bp++;
+		tmppar ^= cur;
+		rp4 ^= cur;
+		rp6 ^= cur;
+		rp8 ^= cur;
+		cur = *bp++;
+		tmppar ^= cur;
+		rp6 ^= cur;
+		rp8 ^= cur;
+		cur = *bp++;
+		tmppar ^= cur;
+		rp4 ^= cur;
+		rp8 ^= cur;
+		cur = *bp++;
+		tmppar ^= cur;
+		rp8 ^= cur;
+
+		cur = *bp++;
+		tmppar ^= cur;
+		rp4 ^= cur;
+		rp6 ^= cur;
+		cur = *bp++;
+		tmppar ^= cur;
+		rp6 ^= cur;
+		cur = *bp++;
+		tmppar ^= cur;
+		rp4 ^= cur;
+		cur = *bp++;
+		tmppar ^= cur;
+
+		par ^= tmppar;
+		if ((i & 0x1) == 0)
+			rp12 ^= tmppar;
+		if ((i & 0x2) == 0)
+			rp14 ^= tmppar;
+		if (eccsize_mult == 2 && (i & 0x4) == 0)
+			rp16 ^= tmppar;
 	}
 
-	/* Create non-inverted ECC code from line parity */
-	tmp1  = (reg3 & 0x80) >> 0; /* B7 -> B7 */
-	tmp1 |= (reg2 & 0x80) >> 1; /* B7 -> B6 */
-	tmp1 |= (reg3 & 0x40) >> 1; /* B6 -> B5 */
-	tmp1 |= (reg2 & 0x40) >> 2; /* B6 -> B4 */
-	tmp1 |= (reg3 & 0x20) >> 2; /* B5 -> B3 */
-	tmp1 |= (reg2 & 0x20) >> 3; /* B5 -> B2 */
-	tmp1 |= (reg3 & 0x10) >> 3; /* B4 -> B1 */
-	tmp1 |= (reg2 & 0x10) >> 4; /* B4 -> B0 */
-
-	tmp2  = (reg3 & 0x08) << 4; /* B3 -> B7 */
-	tmp2 |= (reg2 & 0x08) << 3; /* B3 -> B6 */
-	tmp2 |= (reg3 & 0x04) << 3; /* B2 -> B5 */
-	tmp2 |= (reg2 & 0x04) << 2; /* B2 -> B4 */
-	tmp2 |= (reg3 & 0x02) << 2; /* B1 -> B3 */
-	tmp2 |= (reg2 & 0x02) << 1; /* B1 -> B2 */
-	tmp2 |= (reg3 & 0x01) << 1; /* B0 -> B1 */
-	tmp2 |= (reg2 & 0x01) << 0; /* B7 -> B0 */
-
-	/* Calculate final ECC code */
-#ifdef CONFIG_MTD_NAND_ECC_SMC
-	ecc_code[0] = ~tmp2;
-	ecc_code[1] = ~tmp1;
+	/*
+	 * handle the fact that we use longword operations
+	 * we'll bring rp4..rp14..rp16 back to single byte entities by
+	 * shifting and xoring first fold the upper and lower 16 bits,
+	 * then the upper and lower 8 bits.
+	 */
+	rp4 ^= (rp4 >> 16);
+	rp4 ^= (rp4 >> 8);
+	rp4 &= 0xff;
+	rp6 ^= (rp6 >> 16);
+	rp6 ^= (rp6 >> 8);
+	rp6 &= 0xff;
+	rp8 ^= (rp8 >> 16);
+	rp8 ^= (rp8 >> 8);
+	rp8 &= 0xff;
+	rp10 ^= (rp10 >> 16);
+	rp10 ^= (rp10 >> 8);
+	rp10 &= 0xff;
+	rp12 ^= (rp12 >> 16);
+	rp12 ^= (rp12 >> 8);
+	rp12 &= 0xff;
+	rp14 ^= (rp14 >> 16);
+	rp14 ^= (rp14 >> 8);
+	rp14 &= 0xff;
+	if (eccsize_mult == 2) {
+		rp16 ^= (rp16 >> 16);
+		rp16 ^= (rp16 >> 8);
+		rp16 &= 0xff;
+	}
+
+	/*
+	 * we also need to calculate the row parity for rp0..rp3
+	 * This is present in par, because par is now
+	 * rp3 rp3 rp2 rp2 in little endian and
+	 * rp2 rp2 rp3 rp3 in big endian
+	 * as well as
+	 * rp1 rp0 rp1 rp0 in little endian and
+	 * rp0 rp1 rp0 rp1 in big endian
+	 * First calculate rp2 and rp3
+	 */
+#ifdef __BIG_ENDIAN
+	rp2 = (par >> 16);
+	rp2 ^= (rp2 >> 8);
+	rp2 &= 0xff;
+	rp3 = par & 0xffff;
+	rp3 ^= (rp3 >> 8);
+	rp3 &= 0xff;
 #else
-	ecc_code[0] = ~tmp1;
-	ecc_code[1] = ~tmp2;
+	rp3 = (par >> 16);
+	rp3 ^= (rp3 >> 8);
+	rp3 &= 0xff;
+	rp2 = par & 0xffff;
+	rp2 ^= (rp2 >> 8);
+	rp2 &= 0xff;
 #endif
-	ecc_code[2] = ((~reg1) << 2) | 0x03;
 
-	return 0;
+	/* reduce par to 16 bits then calculate rp1 and rp0 */
+	par ^= (par >> 16);
+#ifdef __BIG_ENDIAN
+	rp0 = (par >> 8) & 0xff;
+	rp1 = (par & 0xff);
+#else
+	rp1 = (par >> 8) & 0xff;
+	rp0 = (par & 0xff);
+#endif
+
+	/* finally reduce par to 8 bits */
+	par ^= (par >> 8);
+	par &= 0xff;
+
+	/*
+	 * and calculate rp5..rp15..rp17
+	 * note that par = rp4 ^ rp5 and due to the commutative property
+	 * of the ^ operator we can say:
+	 * rp5 = (par ^ rp4);
+	 * The & 0xff seems superfluous, but benchmarking learned that
+	 * leaving it out gives slightly worse results. No idea why, probably
+	 * it has to do with the way the pipeline in pentium is organized.
+	 */
+	rp5 = (par ^ rp4) & 0xff;
+	rp7 = (par ^ rp6) & 0xff;
+	rp9 = (par ^ rp8) & 0xff;
+	rp11 = (par ^ rp10) & 0xff;
+	rp13 = (par ^ rp12) & 0xff;
+	rp15 = (par ^ rp14) & 0xff;
+	if (eccsize_mult == 2)
+		rp17 = (par ^ rp16) & 0xff;
+
+	/*
+	 * Finally calculate the ECC bits.
+	 * Again here it might seem that there are performance optimisations
+	 * possible, but benchmarks showed that on the system this is developed
+	 * the code below is the fastest
+	 */
+	if (sm_order) {
+		code[0] = (invparity[rp7] << 7) | (invparity[rp6] << 6) |
+			  (invparity[rp5] << 5) | (invparity[rp4] << 4) |
+			  (invparity[rp3] << 3) | (invparity[rp2] << 2) |
+			  (invparity[rp1] << 1) | (invparity[rp0]);
+		code[1] = (invparity[rp15] << 7) | (invparity[rp14] << 6) |
+			  (invparity[rp13] << 5) | (invparity[rp12] << 4) |
+			  (invparity[rp11] << 3) | (invparity[rp10] << 2) |
+			  (invparity[rp9] << 1) | (invparity[rp8]);
+	} else {
+		code[1] = (invparity[rp7] << 7) | (invparity[rp6] << 6) |
+			  (invparity[rp5] << 5) | (invparity[rp4] << 4) |
+			  (invparity[rp3] << 3) | (invparity[rp2] << 2) |
+			  (invparity[rp1] << 1) | (invparity[rp0]);
+		code[0] = (invparity[rp15] << 7) | (invparity[rp14] << 6) |
+			  (invparity[rp13] << 5) | (invparity[rp12] << 4) |
+			  (invparity[rp11] << 3) | (invparity[rp10] << 2) |
+			  (invparity[rp9] << 1) | (invparity[rp8]);
+	}
+
+	if (eccsize_mult == 1)
+		code[2] =
+		    (invparity[par & 0xf0] << 7) |
+		    (invparity[par & 0x0f] << 6) |
+		    (invparity[par & 0xcc] << 5) |
+		    (invparity[par & 0x33] << 4) |
+		    (invparity[par & 0xaa] << 3) |
+		    (invparity[par & 0x55] << 2) |
+		    3;
+	else
+		code[2] =
+		    (invparity[par & 0xf0] << 7) |
+		    (invparity[par & 0x0f] << 6) |
+		    (invparity[par & 0xcc] << 5) |
+		    (invparity[par & 0x33] << 4) |
+		    (invparity[par & 0xaa] << 3) |
+		    (invparity[par & 0x55] << 2) |
+		    (invparity[rp17] << 1) |
+		    (invparity[rp16] << 0);
 }
-EXPORT_SYMBOL(nand_calculate_ecc);
+EXPORT_SYMBOL(__nand_calculate_ecc);
 
-static inline int countbits(uint32_t byte)
+/**
+ * nand_calculate_ecc - [NAND Interface] Calculate 3-byte ECC for 256/512-byte
+ *			 block
+ * @chip:	NAND chip object
+ * @buf:	input buffer with raw data
+ * @code:	output buffer with ECC
+ */
+int nand_calculate_ecc(struct nand_chip *chip, const unsigned char *buf,
+		       unsigned char *code)
 {
-	int res = 0;
+	bool sm_order = chip->ecc.options & NAND_ECC_SOFT_HAMMING_SM_ORDER;
+
+	__nand_calculate_ecc(buf, chip->ecc.size, code, sm_order);
 
-	for (;byte; byte >>= 1)
-		res += byte & 0x01;
-	return res;
+	return 0;
 }
+EXPORT_SYMBOL(nand_calculate_ecc);
 
 /**
- * nand_correct_data - [NAND Interface] Detect and correct bit error(s)
- * @mtd:	MTD block structure
- * @dat:	raw data read from the chip
+ * __nand_correct_data - [NAND Interface] Detect and correct bit error(s)
+ * @buf:	raw data read from the chip
  * @read_ecc:	ECC from the chip
  * @calc_ecc:	the ECC calculated from raw data
+ * @eccsize:	data bytes per ECC step (256 or 512)
+ * @sm_order:	Smart Media byte order
  *
- * Detect and correct a 1 bit error for 256 byte block
+ * Detect and correct a 1 bit error for eccsize byte block
  */
-int nand_correct_data(struct mtd_info *mtd, u_char *dat,
-		      u_char *read_ecc, u_char *calc_ecc)
+int __nand_correct_data(unsigned char *buf,
+			unsigned char *read_ecc, unsigned char *calc_ecc,
+			unsigned int eccsize, bool sm_order)
 {
-	uint8_t s0, s1, s2;
+	unsigned char b0, b1, b2, bit_addr;
+	unsigned int byte_addr;
+	/* 256 or 512 bytes/ecc  */
+	const uint32_t eccsize_mult = eccsize >> 8;
 
-#ifdef CONFIG_MTD_NAND_ECC_SMC
-	s0 = calc_ecc[0] ^ read_ecc[0];
-	s1 = calc_ecc[1] ^ read_ecc[1];
-	s2 = calc_ecc[2] ^ read_ecc[2];
-#else
-	s1 = calc_ecc[0] ^ read_ecc[0];
-	s0 = calc_ecc[1] ^ read_ecc[1];
-	s2 = calc_ecc[2] ^ read_ecc[2];
-#endif
-	if ((s0 | s1 | s2) == 0)
-		return 0;
-
-	/* Check for a single bit error */
-	if( ((s0 ^ (s0 >> 1)) & 0x55) == 0x55 &&
-	    ((s1 ^ (s1 >> 1)) & 0x55) == 0x55 &&
-	    ((s2 ^ (s2 >> 1)) & 0x54) == 0x54) {
-
-		uint32_t byteoffs, bitnum;
+	/*
+	 * b0 to b2 indicate which bit is faulty (if any)
+	 * we might need the xor result  more than once,
+	 * so keep them in a local var
+	*/
+	if (sm_order) {
+		b0 = read_ecc[0] ^ calc_ecc[0];
+		b1 = read_ecc[1] ^ calc_ecc[1];
+	} else {
+		b0 = read_ecc[1] ^ calc_ecc[1];
+		b1 = read_ecc[0] ^ calc_ecc[0];
+	}
 
-		byteoffs = (s1 << 0) & 0x80;
-		byteoffs |= (s1 << 1) & 0x40;
-		byteoffs |= (s1 << 2) & 0x20;
-		byteoffs |= (s1 << 3) & 0x10;
+	b2 = read_ecc[2] ^ calc_ecc[2];
 
-		byteoffs |= (s0 >> 4) & 0x08;
-		byteoffs |= (s0 >> 3) & 0x04;
-		byteoffs |= (s0 >> 2) & 0x02;
-		byteoffs |= (s0 >> 1) & 0x01;
+	/* check if there are any bitfaults */
 
-		bitnum = (s2 >> 5) & 0x04;
-		bitnum |= (s2 >> 4) & 0x02;
-		bitnum |= (s2 >> 3) & 0x01;
+	/* repeated if statements are slightly more efficient than switch ... */
+	/* ordered in order of likelihood */
 
-		dat[byteoffs] ^= (1 << bitnum);
+	if ((b0 | b1 | b2) == 0)
+		return 0;	/* no error */
 
+	if ((((b0 ^ (b0 >> 1)) & 0x55) == 0x55) &&
+	    (((b1 ^ (b1 >> 1)) & 0x55) == 0x55) &&
+	    ((eccsize_mult == 1 && ((b2 ^ (b2 >> 1)) & 0x54) == 0x54) ||
+	     (eccsize_mult == 2 && ((b2 ^ (b2 >> 1)) & 0x55) == 0x55))) {
+	/* single bit error */
+		/*
+		 * rp17/rp15/13/11/9/7/5/3/1 indicate which byte is the faulty
+		 * byte, cp 5/3/1 indicate the faulty bit.
+		 * A lookup table (called addressbits) is used to filter
+		 * the bits from the byte they are in.
+		 * A marginal optimisation is possible by having three
+		 * different lookup tables.
+		 * One as we have now (for b0), one for b2
+		 * (that would avoid the >> 1), and one for b1 (with all values
+		 * << 4). However it was felt that introducing two more tables
+		 * hardly justify the gain.
+		 *
+		 * The b2 shift is there to get rid of the lowest two bits.
+		 * We could also do addressbits[b2] >> 1 but for the
+		 * performance it does not make any difference
+		 */
+		if (eccsize_mult == 1)
+			byte_addr = (addressbits[b1] << 4) + addressbits[b0];
+		else
+			byte_addr = (addressbits[b2 & 0x3] << 8) +
+				    (addressbits[b1] << 4) + addressbits[b0];
+		bit_addr = addressbits[b2 >> 2];
+		/* flip the bit */
+		buf[byte_addr] ^= (1 << bit_addr);
 		return 1;
-	}
 
-	if(countbits(s0 | ((uint32_t)s1 << 8) | ((uint32_t)s2 <<16)) == 1)
-		return 1;
+	}
+	/* count nr of bits; use table lookup, faster than calculating it */
+	if ((bitsperbyte[b0] + bitsperbyte[b1] + bitsperbyte[b2]) == 1)
+		return 1;	/* error in ECC data; no action needed */
 
+	pr_err("%s: uncorrectable ECC error\n", __func__);
 	return -EBADMSG;
 }
+EXPORT_SYMBOL(__nand_correct_data);
+
+/**
+ * nand_correct_data - [NAND Interface] Detect and correct bit error(s)
+ * @chip:	NAND chip object
+ * @buf:	raw data read from the chip
+ * @read_ecc:	ECC from the chip
+ * @calc_ecc:	the ECC calculated from raw data
+ *
+ * Detect and correct a 1 bit error for 256/512 byte block
+ */
+int nand_correct_data(struct nand_chip *chip, unsigned char *buf,
+		      unsigned char *read_ecc, unsigned char *calc_ecc)
+{
+	bool sm_order = chip->ecc.options & NAND_ECC_SOFT_HAMMING_SM_ORDER;
+
+	return __nand_correct_data(buf, read_ecc, calc_ecc, chip->ecc.size,
+				   sm_order);
+}
 EXPORT_SYMBOL(nand_correct_data);
 
 MODULE_LICENSE("GPL");
-MODULE_AUTHOR("Steven J. Hill <sjhill@realitydiluted.com>");
+MODULE_AUTHOR("Frans Meulenbroeks <fransmeulenbroeks@gmail.com>");
 MODULE_DESCRIPTION("Generic NAND ECC support");