David Brownell <david-b@pacbell.net>:

NAND support for DaVinci-family drivers, with HW ECC support.
Declare the NAND chip on the DM355 EVM board.

Currently tested on DM355 for Linux interop using the standard
large page (2KB) chip in the EVM socket; "hwecc1" and "hwecc4"
work fine.  (Using hwecc4 relies on patches that haven't quite
made it through the Linux-MTD bottlenecks yet.)

Not yet tested:  1-bit on small-page (although it's hard to see
how that could fail); 4-bit on small page (picky layout issues);
the "hwecc_infix" mode (primarily for older boot ROMs; testing
there is blocked on having new bootloader code).


git-svn-id: svn://svn.berlios.de/openocd/trunk@1903 b42882b7-edfa-0310-969c-e2dbd0fdcd60

3 files changed, 747 insertions, 1 deletions

diff --git a/src/flash/Makefile.am b/src/flash/Makefile.am
index e5b76cb..64fc54f 100644
--- a/src/flash/Makefile.am
+++ b/src/flash/Makefile.am
@@ -6,7 +6,7 @@ AM_CPPFLAGS = \
 METASOURCES = AUTO
 noinst_LTLIBRARIES = libflash.la
 libflash_la_SOURCES = \
-	flash.c lpc2000.c cfi.c non_cfi.c at91sam7.c \
+	flash.c lpc2000.c cfi.c non_cfi.c at91sam7.c davinci_nand.c \
 	str7x.c str9x.c aduc702x.c nand.c nand_ecc.c nand_ecc_kw.c \
 	lpc3180_nand_controller.c stellaris.c str9xpec.c stm32x.c tms470.c \
 	ecos.c orion_nand.c s3c24xx_nand.c s3c2410_nand.c s3c2412_nand.c \
diff --git a/src/flash/davinci_nand.c b/src/flash/davinci_nand.c
new file mode 100644
index 0000000..cab489f
--- /dev/null
+++ b/src/flash/davinci_nand.c
@@ -0,0 +1,744 @@
+/***************************************************************************
+ *   Copyright (C) 2009 by David Brownell                                  *
+ *                                                                         *
+ *   This program 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 of the License, or     *
+ *   (at your option) any later version.                                   *
+ *                                                                         *
+ *   This program 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.                          *
+ *                                                                         *
+ *   You should have received a copy of the GNU General Public License     *
+ *   along with this program; if not, write to the                         *
+ *   Free Software Foundation, Inc.,                                       *
+ *   59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.             *
+ ***************************************************************************/
+
+/*
+ * DaVinci family NAND controller support for OpenOCD.
+ *
+ * This driver uses hardware ECC (1-bit or 4-bit) unless
+ * the chip is accessed in "raw" mode.
+ */
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include "nand.h"
+
+
+enum ecc {
+	HWECC1,		/* all controllers support 1-bit ECC */
+	HWECC4,		/* newer chips also have 4-bit ECC hardware */
+	HWECC4_INFIX,	/* avoid this layout, except maybe for boot code */
+};
+
+struct davinci_nand {
+	target_t	*target;
+
+	u8		chipsel;	/* chipselect 0..3 == CS2..CS5 */
+	u8		eccmode;
+
+	/* Async EMIF controller base */
+	u32		aemif;
+
+	/* NAND chip addresses */
+	u32		data;		/* without CLE or ALE */
+	u32		cmd;		/* with CLE */
+	u32		addr;		/* with ALE */
+
+	/* page i/o for the relevant flavor of hardware ECC */
+	int (*read_page)(struct nand_device_s *nand, u32 page,
+			u8 *data, u32 data_size, u8 *oob, u32 oob_size);
+	int (*write_page)(struct nand_device_s *nand, u32 page,
+			u8 *data, u32 data_size, u8 *oob, u32 oob_size);
+};
+
+#define NANDFCR		0x60		/* flash control register */
+#define NANDFSR		0x64		/* flash status register */
+#define NANDFECC	0x70		/* 1-bit ECC data, CS0, 1st of 4 */
+#define NAND4BITECCLOAD	0xbc		/* 4-bit ECC, load saved values */
+#define NAND4BITECC	0xc0		/* 4-bit ECC data, 1st of 4 */
+#define NANDERRADDR	0xd0		/* 4-bit ECC err addr, 1st of 2 */
+#define NANDERRVAL	0xd8		/* 4-bit ECC err value, 1st of 2 */
+
+static int halted(target_t *target, const char *label)
+{
+	if (target->state == TARGET_HALTED)
+		return true;
+
+	LOG_ERROR("Target must be halted to use NAND controller (%s)", label);
+	return false;
+}
+
+static int davinci_register_commands(struct command_context_s *cmd_ctx)
+{
+	return ERROR_OK;
+}
+
+static int davinci_init(struct nand_device_s *nand)
+{
+	struct davinci_nand *info = nand->controller_priv;
+	target_t *target = info->target;
+	u32 nandfcr;
+
+	if (!halted(target, "init"))
+		return ERROR_NAND_OPERATION_FAILED;
+
+	/* We require something else to have configured AEMIF to talk
+	 * to NAND chip in this range (including timings and width).
+	 */
+	target_read_u32(target, info->aemif + NANDFCR, &nandfcr);
+	if (!(nandfcr & (1 << info->chipsel))) {
+		LOG_ERROR("chip address %08x not NAND-enabled?", info->data);
+		return ERROR_NAND_OPERATION_FAILED;
+	}
+
+	/* REVISIT verify:  AxCR must be in 8-bit mode, since that's all we
+	 * tested.  16 bit support should work too; but not with 4-bit ECC.
+	 */
+
+	return ERROR_OK;
+}
+
+static int davinci_reset(struct nand_device_s *nand)
+{
+	return ERROR_OK;
+}
+
+static int davinci_nand_ready(struct nand_device_s *nand, int timeout)
+{
+	struct davinci_nand *info = nand->controller_priv;
+	target_t *target = info->target;
+	u32 nandfsr;
+
+	/* NOTE: return code is zero/error, else success; not ERROR_* */
+
+	if (!halted(target, "ready"))
+		return 0;
+
+	do {
+		target_read_u32(target, info->aemif + NANDFSR, &nandfsr);
+
+		if (nandfsr & 0x01)
+			return 1;
+
+		alive_sleep(1);
+	} while (timeout-- > 0);
+
+	return 0;
+}
+
+static int davinci_command(struct nand_device_s *nand, u8 command)
+{
+	struct davinci_nand *info = nand->controller_priv;
+	target_t *target = info->target;
+
+	if (!halted(target, "command"))
+		return ERROR_NAND_OPERATION_FAILED;
+
+	target_write_u8(target, info->cmd, command);
+	return ERROR_OK;
+}
+
+static int davinci_address(struct nand_device_s *nand, u8 address)
+{
+	struct davinci_nand *info = nand->controller_priv;
+	target_t *target = info->target;
+
+	if (!halted(target, "address"))
+		return ERROR_NAND_OPERATION_FAILED;
+
+	target_write_u8(target, info->addr, address);
+	return ERROR_OK;
+}
+
+static int davinci_write_data(struct nand_device_s *nand, u16 data)
+{
+	struct davinci_nand *info = nand->controller_priv;
+	target_t *target = info->target;
+
+	if (!halted(target, "write_data"))
+		return ERROR_NAND_OPERATION_FAILED;
+
+	target_write_u8(target, info->data, data);
+	return ERROR_OK;
+}
+
+static int davinci_read_data(struct nand_device_s *nand, void *data)
+{
+	struct davinci_nand *info = nand->controller_priv;
+	target_t *target = info->target;
+
+	if (!halted(target, "read_data"))
+		return ERROR_NAND_OPERATION_FAILED;
+
+	target_read_u8(target, info->data, data);
+	return ERROR_OK;
+}
+
+/* REVISIT a bit of native code should let block I/O be MUCH faster */
+
+static int davinci_read_block_data(struct nand_device_s *nand,
+		u8 *data, int data_size)
+{
+	struct davinci_nand *info = nand->controller_priv;
+	target_t *target = info->target;
+	u32 nfdata = info->data;
+	u32 tmp;
+
+	if (!halted(target, "read_block"))
+		return ERROR_NAND_OPERATION_FAILED;
+
+	while (data_size >= 4) {
+		target_read_u32(target, nfdata, &tmp);
+
+		data[0] = tmp;
+		data[1] = tmp >> 8;
+		data[2] = tmp >> 16;
+		data[3] = tmp >> 24;
+
+		data_size -= 4;
+		data += 4;
+	}
+
+	while (data_size > 0) {
+		target_read_u8(target, nfdata, data);
+
+		data_size -= 1;
+		data += 1;
+	}
+
+	return ERROR_OK;
+}
+
+static int davinci_write_block_data(struct nand_device_s *nand,
+		u8 *data, int data_size)
+{
+	struct davinci_nand *info = nand->controller_priv;
+	target_t *target = info->target;
+	u32 nfdata = info->data;
+	u32 tmp;
+
+	if (!halted(target, "write_block"))
+		return ERROR_NAND_OPERATION_FAILED;
+
+	while (data_size >= 4) {
+		tmp = le_to_h_u32(data);
+		target_write_u32(target, nfdata, tmp);
+
+		data_size -= 4;
+		data += 4;
+	}
+
+	while (data_size > 0) {
+		target_write_u8(target, nfdata, *data);
+
+		data_size -= 1;
+		data += 1;
+	}
+
+	return ERROR_OK;
+}
+
+static int davinci_write_page(struct nand_device_s *nand, u32 page,
+		u8 *data, u32 data_size, u8 *oob, u32 oob_size)
+{
+	struct davinci_nand *info = nand->controller_priv;
+	u8 *ooballoc = NULL;
+	int status;
+
+	if (!nand->device)
+		return ERROR_NAND_DEVICE_NOT_PROBED;
+	if (!halted(info->target, "write_page"))
+		return ERROR_NAND_OPERATION_FAILED;
+
+	/* Always write both data and OOB ... we are not "raw" I/O! */
+	if (!data)
+		return ERROR_NAND_OPERATION_FAILED;
+
+	/* If we're not given OOB, write 0xff where we don't write ECC codes. */
+	switch (nand->page_size) {
+	case 512:
+		oob_size = 16;
+		break;
+	case 2048:
+		oob_size = 64;
+		break;
+	case 4096:
+		oob_size = 128;
+		break;
+	default:
+		return ERROR_NAND_OPERATION_FAILED;
+	}
+	if (!oob) {
+		ooballoc = malloc(oob_size);
+		if (ooballoc)
+			return ERROR_NAND_OPERATION_FAILED;
+		oob = ooballoc;
+		memset(oob, 0x0ff, oob_size);
+	}
+
+	status = info->write_page(nand, page, data, data_size, oob, oob_size);
+	free(ooballoc);
+	return status;
+}
+
+static int davinci_read_page(struct nand_device_s *nand, u32 page,
+		u8 *data, u32 data_size, u8 *oob, u32 oob_size)
+{
+	struct davinci_nand *info = nand->controller_priv;
+
+	if (!nand->device)
+		return ERROR_NAND_DEVICE_NOT_PROBED;
+	if (!halted(info->target, "read_page"))
+		return ERROR_NAND_OPERATION_FAILED;
+
+	return info->read_page(nand, page, data, data_size, oob, oob_size);
+}
+
+static void davinci_write_pagecmd(struct nand_device_s *nand, u8 cmd, u32 page)
+{
+	struct davinci_nand *info = nand->controller_priv;
+	target_t *target = info->target;
+	int page3 = nand->address_cycles - (nand->page_size == 512);
+
+	/* write command ({page,otp}x{read,program} */
+	target_write_u8(target, info->cmd, cmd);
+
+	/* column address (beginning-of-page) */
+	target_write_u8(target, info->addr, 0);
+	if (nand->page_size > 512)
+		target_write_u8(target, info->addr, 0);
+
+	/* page address */
+	target_write_u8(target, info->addr, page);
+	target_write_u8(target, info->addr, page >> 8);
+	if (page3)
+		target_write_u8(target, info->addr, page >> 16);
+	if (page3 == 2)
+		target_write_u8(target, info->addr, page >> 24);
+}
+
+static int davinci_writepage_tail(struct nand_device_s *nand,
+		u8 *oob, u32 oob_size)
+{
+	struct davinci_nand *info = nand->controller_priv;
+	target_t *target = info->target;
+	u8 status;
+
+	if (oob_size)
+		davinci_write_block_data(nand, oob, oob_size);
+
+	/* non-cachemode page program */
+	target_write_u8(target, info->cmd, NAND_CMD_PAGEPROG);
+
+	if (!davinci_nand_ready(nand, 100))
+		return ERROR_NAND_OPERATION_TIMEOUT;
+
+	if (nand_read_status(nand, &status) != ERROR_OK) {
+		LOG_ERROR("couldn't read status");
+		return ERROR_NAND_OPERATION_FAILED;
+	}
+
+	if (status & NAND_STATUS_FAIL) {
+		LOG_ERROR("write operation failed, status: 0x%02x", status);
+		return ERROR_NAND_OPERATION_FAILED;
+	}
+
+	return ERROR_OK;
+}
+
+/*
+ * All DaVinci family chips support 1-bit ECC on a per-chipselect basis.
+ */
+static int davinci_write_page_ecc1(struct nand_device_s *nand, u32 page,
+		u8 *data, u32 data_size, u8 *oob, u32 oob_size)
+{
+	unsigned oob_offset;
+	struct davinci_nand *info = nand->controller_priv;
+	target_t *target = info->target;
+	const u32 fcr_addr = info->aemif + NANDFCR;
+	const u32 ecc1_addr = info->aemif + NANDFECC + info->chipsel;
+	u32 fcr, ecc1;
+
+	/* Write contiguous ECC bytes starting at specified offset.
+	 * NOTE: Linux reserves twice as many bytes as we need; and
+	 * for 16-bit OOB, those extra bytes are discontiguous.
+	 */
+	switch (nand->page_size) {
+	case 512:
+		oob_offset = 0;
+		break;
+	case 2048:
+		oob_offset = 40;
+		break;
+	default:
+		oob_offset = 80;
+		break;
+	}
+
+	davinci_write_pagecmd(nand, NAND_CMD_SEQIN, page);
+
+	/* scrub any old ECC state */
+	target_read_u32(target, ecc1_addr, &ecc1);
+
+	target_read_u32(target, fcr_addr, &fcr);
+	fcr |= 1 << (8 + info->chipsel);
+
+	do {
+		/* set "start csX 1bit ecc" bit */
+		target_write_u32(target, fcr_addr, fcr);
+
+		/* write 512 bytes */
+		davinci_write_block_data(nand, data, 512);
+		data += 512;
+		data_size -= 512;
+
+		/* read the ecc, pack to 3 bytes, and invert so the ecc
+		 * in an erased block is correct
+		 */
+		target_read_u32(target, ecc1_addr, &ecc1);
+		ecc1 = (ecc1 & 0x0fff) | ((ecc1 & 0x0fff0000) >> 4);
+		ecc1 = ~ecc1;
+
+		/* save correct ECC code into oob data */
+		oob[oob_offset++] = (u8)(ecc1);
+		oob[oob_offset++] = (u8)(ecc1 >> 8);
+		oob[oob_offset++] = (u8)(ecc1 >> 16);
+
+	} while (data_size);
+
+	/* write OOB into spare area */
+	return davinci_writepage_tail(nand, oob, oob_size);
+}
+
+/*
+ * Preferred "new style" ECC layout for use with 4-bit ECC.  This somewhat
+ * slows down large page reads done with error correction (since the OOB
+ * is read first, so its ECC data can be used incrementally), but the
+ * manufacturer bad block markers are safe.  Contrast:  old "infix" style.
+ */
+static int davinci_write_page_ecc4(struct nand_device_s *nand, u32 page,
+		u8 *data, u32 data_size, u8 *oob, u32 oob_size)
+{
+	static const u8 ecc512[] = {
+		0, 1, 2, 3, 4, /* 5== mfr badblock */
+		6, 7, /* 8..12 for BBT or JFFS2 */ 13, 14, 15,
+	};
+	static const u8 ecc2048[] = {
+		24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
+		34, 35, 36, 37, 38, 39, 40, 41, 42, 43,
+		44, 45, 46, 47, 48, 49, 50, 51, 52, 53,
+		54, 55, 56, 57, 58, 59, 60, 61, 62, 63,
+	};
+	static const u8 ecc4096[] = {
+		 48,  49,  50,  51,  52,  53,  54,  55,  56,  57,
+		 58,  59,  60,  61,  62,  63,  64,  65,  66,  67,
+		 68,  69,  70,  71,  72,  73,  74,  75,  76,  77,
+		 78,  79,  80,  81,  82,  83,  84,  85,  86,  87,
+		 88,  89,  90,  91,  92,  93,  94,  95,  96,  97,
+		 98,  99, 100, 101, 102, 103, 104, 105, 106, 107,
+		108, 109, 110, 111, 112, 113, 114, 115, 116, 117,
+		118, 119, 120, 121, 122, 123, 124, 125, 126, 127,
+	};
+
+	struct davinci_nand *info = nand->controller_priv;
+	const u8 *l;
+	target_t *target = info->target;
+	const u32 fcr_addr = info->aemif + NANDFCR;
+	const u32 ecc4_addr = info->aemif + NAND4BITECC;
+	u32 fcr, ecc4;
+
+	/* Use the same ECC layout Linux uses.  For small page chips
+	 * it's a bit cramped.
+	 *
+	 * NOTE:  at this writing, 4KB pages have issues in Linux
+	 * because they need more than 64 bytes of ECC data, which
+	 * the standard ECC logic can't handle.
+	 */
+	switch (nand->page_size) {
+	case 512:
+		l = ecc512;
+		break;
+	case 2048:
+		l = ecc2048;
+		break;
+	default:
+		l = ecc4096;
+		break;
+	}
+
+	davinci_write_pagecmd(nand, NAND_CMD_SEQIN, page);
+
+	/* scrub any old ECC state */
+	target_read_u32(target, info->aemif + NANDERRVAL, &ecc4);
+
+	target_read_u32(target, fcr_addr, &fcr);
+	fcr &= ~(0x03 << 4);
+	fcr |= (1 << 12) | (info->chipsel << 4);
+
+	do {
+		u32 raw_ecc[4], *p;
+		int i;
+
+		/* start 4bit ecc on csX */
+		target_write_u32(target, fcr_addr, fcr);
+
+		/* write 512 bytes */
+		davinci_write_block_data(nand, data, 512);
+		data += 512;
+		data_size -= 512;
+
+		/* read the ecc, then save it into 10 bytes in the oob */
+		for (i = 0; i < 4; i++) {
+			target_read_u32(target, ecc4_addr + 4 * i, &raw_ecc[i]);
+			raw_ecc[i] &= 0x03ff03ff;
+		}
+		for (i = 0, p = raw_ecc; i < 2; i++, p += 2) {
+			oob[*l++] =   p[0]        & 0xff;
+			oob[*l++] = ((p[0] >>  8) & 0x03) | ((p[0] >> 14) & 0xfc);
+			oob[*l++] = ((p[0] >> 22) & 0x0f) | ((p[1] <<  4) & 0xf0);
+			oob[*l++] = ((p[1] >>  4) & 0x3f) | ((p[1] >> 10) & 0xc0);
+			oob[*l++] =  (p[1] >> 18) & 0xff;
+		}
+
+	} while (data_size);
+
+	/* write OOB into spare area */
+	return davinci_writepage_tail(nand, oob, oob_size);
+}
+
+/*
+ * "Infix" OOB ... like Linux ECC_HW_SYNDROME.  Avoided because it trashes
+ * manufacturer bad block markers, except on small page chips.  Once you
+ * write to a page using this scheme, you need specialized code to update
+ * it (code which ignores now-invalid bad block markers).
+ *
+ * This is needed *only* to support older firmware.  Older ROM Boot Loaders
+ * need it to read their second stage loader (UBL) into SRAM, but from then
+ * on the whole system can use the cleaner non-infix layouts.  Systems with
+ * older second stage loaders (ABL/U-Boot, etc) or other system software
+ * (MVL 4.x/5.x kernels, filesystems, etc) may need it more generally.
+ */
+static int davinci_write_page_ecc4infix(struct nand_device_s *nand, u32 page,
+		u8 *data, u32 data_size, u8 *oob, u32 oob_size)
+{
+	struct davinci_nand *info = nand->controller_priv;
+	target_t *target = info->target;
+	const u32 fcr_addr = info->aemif + NANDFCR;
+	const u32 ecc4_addr = info->aemif + NAND4BITECC;
+	u32 fcr, ecc4;
+
+	davinci_write_pagecmd(nand, NAND_CMD_SEQIN, page);
+
+	/* scrub any old ECC state */
+	target_read_u32(target, info->aemif + NANDERRVAL, &ecc4);
+
+	target_read_u32(target, fcr_addr, &fcr);
+	fcr &= ~(0x03 << 4);
+	fcr |= (1 << 12) | (info->chipsel << 4);
+
+	do {
+		u32 raw_ecc[4], *p;
+		u8 *l;
+		int i;
+
+		/* start 4bit ecc on csX */
+		target_write_u32(target, fcr_addr, fcr);
+
+		/* write 512 bytes */
+		davinci_write_block_data(nand, data, 512);
+		data += 512;
+		data_size -= 512;
+
+		/* read the ecc */
+		for (i = 0; i < 4; i++) {
+			target_read_u32(target, ecc4_addr + 4 * i, &raw_ecc[i]);
+			raw_ecc[i] &= 0x03ff03ff;
+		}
+
+		/* skip 6 bytes of prepad, then pack 10 packed ecc bytes */
+		for (i = 0, l = oob + 6, p = raw_ecc; i < 2; i++, p += 2) {
+			*l++ =   p[0]        & 0xff;
+			*l++ = ((p[0] >>  8) & 0x03) | ((p[0] >> 14) & 0xfc);
+			*l++ = ((p[0] >> 22) & 0x0f) | ((p[1] <<  4) & 0xf0);
+			*l++ = ((p[1] >>  4) & 0x3f) | ((p[1] >> 10) & 0xc0);
+			*l++ =  (p[1] >> 18) & 0xff;
+		}
+
+		/* write this "out-of-band" data -- infix */
+		davinci_write_block_data(nand, oob, 16);
+		oob += 16;
+		oob_size -= 16;
+
+	} while (data_size);
+
+	/* the last data and OOB writes included the spare area */
+	return davinci_writepage_tail(nand, NULL, 0);
+}
+
+static int davinci_read_page_ecc4infix(struct nand_device_s *nand, u32 page,
+		u8 *data, u32 data_size, u8 *oob, u32 oob_size)
+{
+	davinci_write_pagecmd(nand, NAND_CMD_READ0, page);
+
+	/* large page devices need a start command */
+	if (nand->page_size > 512)
+		davinci_command(nand, NAND_CMD_READSTART);
+
+	if (!davinci_nand_ready(nand, 100))
+		return ERROR_NAND_OPERATION_TIMEOUT;
+
+	/* NOTE:  not bothering to compute and use ECC data for now */
+
+	do {
+		/* write 512 bytes */
+		davinci_read_block_data(nand, data, 512);
+		data += 512;
+		data_size -= 512;
+
+		/* read this "out-of-band" data -- infix */
+		davinci_read_block_data(nand, oob, 16);
+		oob += 16;
+		oob_size -= 16;
+	} while (data_size);
+
+	return ERROR_OK;
+}
+
+static int davinci_nand_device_command(struct command_context_s *cmd_ctx,
+		char *cmd, char **argv, int argc,
+		struct nand_device_s *nand)
+{
+	struct davinci_nand *info;
+	target_t *target;
+	unsigned long chip, aemif;
+	enum ecc eccmode;
+	int chipsel;
+	char *ep;
+
+	/* arguments:
+	 *  - "davinci"
+	 *  - target
+	 *  - nand chip address
+	 *  - ecc mode
+	 *  - aemif address
+	 * Plus someday, optionally, ALE and CLE masks.
+	 */
+	if (argc < 5) {
+		LOG_ERROR("parameters: %s target "
+				"chip_addr hwecc_mode aemif_addr",
+				argv[0]);
+		goto fail;
+	}
+
+	target = get_target(argv[1]);
+	if (!target) {
+		LOG_ERROR("invalid target %s", argv[1]);
+		goto fail;
+	}
+
+	chip = strtoul(argv[2], &ep, 0);
+	if (*ep || chip == 0 || chip == ULONG_MAX) {
+		LOG_ERROR("Invalid NAND chip address %s", argv[2]);
+		goto fail;
+	}
+
+	if (strcmp(argv[3], "hwecc1") == 0)
+		eccmode = HWECC1;
+	else if (strcmp(argv[3], "hwecc4") == 0)
+		eccmode = HWECC4;
+	else if (strcmp(argv[3], "hwecc4_infix") == 0)
+		eccmode = HWECC4_INFIX;
+	else {
+		LOG_ERROR("Invalid ecc mode %s", argv[3]);
+		goto fail;
+	}
+
+	aemif = strtoul(argv[4], &ep, 0);
+	if (*ep || chip == 0 || chip == ULONG_MAX) {
+		LOG_ERROR("Invalid AEMIF controller address %s", argv[4]);
+		goto fail;
+	}
+
+	/* REVISIT what we'd *like* to do is look up valid ranges using
+	 * target-specific declarations, and not even need to pass the
+	 * AEMIF controller address.
+	 */
+	if (aemif == 0x01e00000			/* dm6446, dm357 */
+			|| aemif == 0x01e10000	/* dm335, dm355 */
+			|| aemif == 0x01d10000	/* dm365 */
+			) {
+		if (chip < 0x0200000 || chip >= 0x0a000000) {
+			LOG_ERROR("NAND address %08lx out of range?", chip);
+			goto fail;
+		}
+		chipsel = (chip - 0x02000000) >> 21;
+	} else {
+		LOG_ERROR("unrecognized AEMIF controller address %08lx", aemif);
+		goto fail;
+	}
+
+	info = calloc(1, sizeof *info);
+	if (info == NULL)
+		goto fail;
+
+	info->target = target;
+	info->eccmode = eccmode;
+	info->chipsel = chipsel;
+	info->aemif = aemif;
+	info->data = chip;
+	info->cmd = chip | 0x10;
+	info->addr = chip | 0x08;
+
+	nand->controller_priv = info;
+
+	/* NOTE:  for now we don't do any error correction on read.
+	 * Nothing else in OpenOCD currently corrects read errors,
+	 * and in any case it's *writing* that we care most about.
+	 */
+	info->read_page = nand_read_page_raw;
+
+	switch (eccmode) {
+	case HWECC1:
+		/* ECC_HW, 1-bit corrections, 3 bytes ECC per 512 data bytes */
+		info->write_page = davinci_write_page_ecc1;
+		break;
+	case HWECC4:
+		/* ECC_HW, 4-bit corrections, 10 bytes ECC per 512 data bytes */
+		info->write_page = davinci_write_page_ecc4;
+		break;
+	case HWECC4_INFIX:
+		/* Same 4-bit ECC HW, with problematic page/ecc layout */
+		info->read_page = davinci_read_page_ecc4infix;
+		info->write_page = davinci_write_page_ecc4infix;
+		break;
+	}
+
+	return ERROR_OK;
+
+fail:
+	return ERROR_NAND_OPERATION_FAILED;
+}
+
+nand_flash_controller_t davinci_nand_controller = {
+	.name			= "davinci",
+	.nand_device_command	= davinci_nand_device_command,
+	.register_commands	= davinci_register_commands,
+	.init			= davinci_init,
+	.reset			= davinci_reset,
+	.command		= davinci_command,
+	.address		= davinci_address,
+	.write_data		= davinci_write_data,
+	.read_data		= davinci_read_data,
+	.write_page		= davinci_write_page,
+	.read_page		= davinci_read_page,
+	.write_block_data	= davinci_write_block_data,
+	.read_block_data	= davinci_read_block_data,
+	.nand_ready		= davinci_nand_ready,
+};
diff --git a/src/flash/nand.c b/src/flash/nand.c
index bc15f42..4046011 100644
--- a/src/flash/nand.c
+++ b/src/flash/nand.c
@@ -48,6 +48,7 @@ static int nand_write_page(struct nand_device_s *device, u32 page, u8 *data, u32
 
 /* NAND flash controller
  */
+extern nand_flash_controller_t davinci_nand_controller;
 extern nand_flash_controller_t lpc3180_nand_controller;
 extern nand_flash_controller_t orion_nand_controller;
 extern nand_flash_controller_t s3c2410_nand_controller;
@@ -59,6 +60,7 @@ extern nand_flash_controller_t s3c2443_nand_controller;
 
 static nand_flash_controller_t *nand_flash_controllers[] =
 {
+	&davinci_nand_controller,
 	&lpc3180_nand_controller,
 	&orion_nand_controller,
 	&s3c2410_nand_controller,