Patchwork [U-Boot,v5] palmtreo680: add utility that writes u-boot to flash

login
register
mail settings
Submitter Mike Dunn
Date April 19, 2013, 7:28 p.m.
Message ID <1366399701-2446-1-git-send-email-mikedunn@newsguy.com>
Download mbox | patch
Permalink /patch/238095/
State Superseded
Delegated to: Marek Vasut
Headers show

Comments

Mike Dunn - April 19, 2013, 7:28 p.m.
This adds a userspace linux utility that writes the u-boot image to an mtd
partition on the docg4 nand flash.

A special utility is required to do this because u-boot is partially loaded by
an initial program loader (IPL) that is permanently programmed to the boot
region of the flash.  This IPL expects the image to be written in a unique
format. The characteristics of this format can be summarized as follows:
  - Flash blocks to be loaded must have a magic number in the oob bytes of the
    first page of the block.
  - Each page must be written redundantly in the subsequent page.
  - The integrated flash controller's "reliable mode" is used, requiring that
    alternate 2k regions (4 pages) are skipped when writing.
For these reasons, a u-boot image can not be written using nandwrite from
mtd-utils.

Signed-off-by: Mike Dunn <mikedunn@newsguy.com>
---
Changelog:
  v5: on error, return sensible code from errno.h
  v4:
    - use return instead of exit
    - use calloc() instead of malloc()
    - remove call to read() from within while loop test 
  v3: new patch; split off from patch 7 in v2 of patchset

 tools/palmtreo680/flash_u-boot.c |  177 ++++++++++++++++++++++++++++++++++++++
 1 files changed, 177 insertions(+), 0 deletions(-)
 create mode 100644 tools/palmtreo680/flash_u-boot.c
Marek Vasut - April 20, 2013, 10:25 a.m.
Dear Mike Dunn,

> This adds a userspace linux utility that writes the u-boot image to an mtd
> partition on the docg4 nand flash.
> 
> A special utility is required to do this because u-boot is partially loaded
> by an initial program loader (IPL) that is permanently programmed to the
> boot region of the flash.  This IPL expects the image to be written in a
> unique format. The characteristics of this format can be summarized as
> follows: - Flash blocks to be loaded must have a magic number in the oob
> bytes of the first page of the block.
>   - Each page must be written redundantly in the subsequent page.
>   - The integrated flash controller's "reliable mode" is used, requiring
> that alternate 2k regions (4 pages) are skipped when writing.
> For these reasons, a u-boot image can not be written using nandwrite from
> mtd-utils.

I fixed the calloc() semantics and applied, thanks. Please repost whatever is 
still missing in u-boot-pxa/master and I'll then roll pullRQ for mainline.

Best regards,
Marek Vasut
Mike Dunn - April 20, 2013, 4:37 p.m.
On 04/20/2013 03:25 AM, Marek Vasut wrote:
> Dear Mike Dunn,

[...]

> 
> I fixed the calloc() semantics and applied, thanks. Please repost whatever is 
> still missing in u-boot-pxa/master and I'll then roll pullRQ for mainline.

Thanks Marek.  Now I see your point regarding calloc() semantics.  Only

    arm: bootm: call udc_disable() before booting linux

is not in your tree; I think you were worried about testing on musb boards.
I'll repost, but no reason to take a risk; I hope to get a patch for the Linux
udc driver upstream soon.

Thanks,
Mike
Marek Vasut - April 20, 2013, 5:12 p.m.
Dear Mike Dunn,

> On 04/20/2013 03:25 AM, Marek Vasut wrote:
> > Dear Mike Dunn,
> 
> [...]
> 
> > I fixed the calloc() semantics and applied, thanks. Please repost
> > whatever is still missing in u-boot-pxa/master and I'll then roll pullRQ
> > for mainline.
> 
> Thanks Marek.  Now I see your point regarding calloc() semantics.  Only
> 
>     arm: bootm: call udc_disable() before booting linux
> 
> is not in your tree; I think you were worried about testing on musb boards.
> I'll repost, but no reason to take a risk; I hope to get a patch for the
> Linux udc driver upstream soon.

Ah yes, sorry about that. I wonder if Tom tested it or maybe Albert.

Best regards,
Marek Vasut
Mike Dunn - April 20, 2013, 5:53 p.m.
On 04/20/2013 03:25 AM, Marek Vasut wrote:
> Dear Mike Dunn,
> 
> 
> I fixed the calloc() semantics and applied, thanks. Please repost whatever is 
> still missing in u-boot-pxa/master and I'll then roll pullRQ for mainline.


Oops, the board support patch is not in your tree either.  Will repost.

Thanks,
Mike
Marek Vasut - April 20, 2013, 6:01 p.m.
Dear Mike Dunn,

> On 04/20/2013 03:25 AM, Marek Vasut wrote:
> > Dear Mike Dunn,
> > 
> > 
> > I fixed the calloc() semantics and applied, thanks. Please repost
> > whatever is still missing in u-boot-pxa/master and I'll then roll pullRQ
> > for mainline.
> 
> Oops, the board support patch is not in your tree either.  Will repost.

Did I not ramble about it? Check that first please, I'm not sure now ;-)

Best regards,
Marek Vasut
Mike Dunn - April 20, 2013, 7:07 p.m.
On 04/20/2013 11:01 AM, Marek Vasut wrote:
> Dear Mike Dunn,
> 
>> On 04/20/2013 03:25 AM, Marek Vasut wrote:
>>> Dear Mike Dunn,
>>>
>>>
>>> I fixed the calloc() semantics and applied, thanks. Please repost
>>> whatever is still missing in u-boot-pxa/master and I'll then roll pullRQ
>>> for mainline.
>>
>> Oops, the board support patch is not in your tree either.  Will repost.
> 
> Did I not ramble about it? Check that first please, I'm not sure now ;-)
> 


Yes, we went back-and-forth about it, and the latest version was v4.  But I need
to rework it because the spl build now fails due to a recent change to
arch/arm/cpu/pxa/start.S in commit 7086e91b0e9930139e703c8f9f857b52baf2bef4

  arm: relocate_code(): Use __image_copy_end for end of relocation

I was using my own linker script for the spl, which does not define the
__image_copy_end symbol that is now referenced in start.S.  I don't exactly
remember why I felt the need to avoid using the script in
arch/arm/cpu/u-boot-spl.lds, except that I didn't like all the sections created
by the --ffunction-sections and --fdata-sections compiler flags and wanted them
combined with a '*' wildcard.  But I removed my linker script and the
CONFIG_SPL_LDSCRIPT line from my config file and it is testing fine.  So I will
have to post a v5 of the patch.

Thanks again Marek,
Mike
Smithlife - April 26, 2013, 8:34 a.m.
Dies ist, was ich interessiert bin in.Welcome mich contace.

Patch

diff --git a/tools/palmtreo680/flash_u-boot.c b/tools/palmtreo680/flash_u-boot.c
new file mode 100644
index 0000000..3d8296f
--- /dev/null
+++ b/tools/palmtreo680/flash_u-boot.c
@@ -0,0 +1,177 @@ 
+/*
+ * Copyright (C) 2013 Mike Dunn <mikedunn@newsguy.com>
+ *
+ * This file is released under the terms of GPL v2 and any later version.
+ * See the file COPYING in the root directory of the source tree for details.
+ *
+ *
+ * This is a userspace Linux utility that, when run on the Treo 680, will
+ * program u-boot to flash.  The docg4 driver *must* be loaded with the
+ * reliable_mode and ignore_badblocks parameters enabled:
+ *
+ *        modprobe docg4 ignore_badblocks=1 reliable_mode=1
+ *
+ * This utility writes the concatenated spl + u-boot image to the start of the
+ * mtd device in the format expected by the IPL/SPL.  The image file and mtd
+ * device node are passed to the utility as arguments.  The blocks must have
+ * been erased beforehand.
+ *
+ * When you compile this, note that it links to libmtd from mtd-utils, so ensure
+ * that your include and lib paths include this.
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <string.h>
+#include <sys/types.h>
+#include <unistd.h>
+#include <errno.h>
+#include <mtd/mtd-user.h>
+#include "libmtd.h"
+
+#define RELIABLE_BLOCKSIZE  0x10000 /* block capacity in reliable mode */
+#define STANDARD_BLOCKSIZE  0x40000 /* block capacity in normal mode */
+#define PAGESIZE 512
+#define PAGES_PER_BLOCK 512
+#define OOBSIZE 7		/* available to user (16 total) */
+
+uint8_t ff_oob[OOBSIZE] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
+
+/* this is the magic number the IPL looks for (ASCII "BIPO") */
+uint8_t page0_oob[OOBSIZE] = {'B', 'I', 'P', 'O', 0xff, 0xff, 0xff};
+
+int main(int argc, char * const argv[])
+{
+	int devfd, datafd, num_blocks, block;
+	off_t file_size;
+	libmtd_t mtd_desc;
+	struct mtd_dev_info devinfo;
+	uint8_t *blockbuf;
+	char response[8];
+
+	if (argc != 3) {
+		printf("usage: %s <image file> <mtd dev node>\n", argv[0]);
+		return -EINVAL;
+	}
+
+	mtd_desc = libmtd_open();
+	if (mtd_desc == NULL) {
+		int errsv = errno;
+		fprintf(stderr, "can't initialize libmtd\n");
+		return -errsv;
+	}
+
+	/* open the spl image file and mtd device */
+	datafd = open(argv[1], O_RDONLY);
+	if (datafd == -1) {
+		int errsv = errno;
+		perror(argv[1]);
+		return -errsv;
+	}
+	devfd = open(argv[2], O_WRONLY);
+	if (devfd == -1) {
+		int errsv = errno;
+		perror(argv[2]);
+		return -errsv;
+	}
+	if (mtd_get_dev_info(mtd_desc, argv[2], &devinfo) < 0) {
+		int errsv = errno;
+		perror(argv[2]);
+		return -errsv;
+	}
+
+	/* determine the number of blocks needed by the image */
+	file_size = lseek(datafd, 0, SEEK_END);
+	if (file_size == (off_t)-1) {
+		int errsv = errno;
+		perror("lseek");
+		return -errsv;
+	}
+	num_blocks = (file_size + RELIABLE_BLOCKSIZE - 1) / RELIABLE_BLOCKSIZE;
+	file_size = lseek(datafd, 0, SEEK_SET);
+	if (file_size == (off_t)-1) {
+		int errsv = errno;
+		perror("lseek");
+		return -errsv;
+	}
+	printf("The mtd partition contains %d blocks\n", devinfo.eb_cnt);
+	printf("U-boot will occupy %d blocks\n", num_blocks);
+	if (num_blocks > devinfo.eb_cnt) {
+		fprintf(stderr, "Insufficient blocks on partition\n");
+		return -EINVAL;
+	}
+
+	printf("IMPORTANT: These blocks must be in an erased state!\n");
+	printf("Do you want to proceed?\n");
+	scanf("%s", response);
+	if ((response[0] != 'y') && (response[0] != 'Y')) {
+		printf("Exiting\n");
+		close(devfd);
+		close(datafd);
+		return 0;
+	}
+
+	blockbuf = calloc(RELIABLE_BLOCKSIZE, 1);
+	if (blockbuf == NULL) {
+		int errsv = errno;
+		perror("calloc");
+		return -errsv;
+	}
+
+	for (block = 0; block < num_blocks; block++) {
+		int ofs, page;
+		uint8_t *pagebuf = blockbuf, *buf = blockbuf;
+		uint8_t *oobbuf = page0_oob; /* magic num in oob of 1st page */
+		size_t len = RELIABLE_BLOCKSIZE;
+		int ret;
+
+		/* read data for one block from file */
+		while (len) {
+			ssize_t read_ret = read(datafd, buf, len);
+			if (read_ret == -1) {
+				int errsv = errno;
+				if (errno == EINTR)
+					continue;
+				perror("read");
+				return -errsv;
+			} else if (read_ret == 0) {
+				break; /* EOF */
+			}
+			len -= read_ret;
+			buf += read_ret;
+		}
+
+		printf("Block %d: writing\r", block + 1);
+		fflush(stdout);
+
+		for (page = 0, ofs = 0;
+		     page < PAGES_PER_BLOCK;
+		     page++, ofs += PAGESIZE) {
+			if (page & 0x04)  /* Odd-numbered 2k page */
+				continue; /* skipped in reliable mode */
+
+			ret = mtd_write(mtd_desc, &devinfo, devfd, block, ofs,
+					pagebuf, PAGESIZE, oobbuf, OOBSIZE,
+					MTD_OPS_PLACE_OOB);
+			if (ret) {
+				fprintf(stderr,
+					"\nmtd_write returned %d on block %d, ofs %x\n",
+					ret, block + 1, ofs);
+				return -EIO;
+			}
+			oobbuf = ff_oob;  /* oob for subsequent pages */
+
+			if (page & 0x01)  /* odd-numbered subpage */
+				pagebuf += PAGESIZE;
+		}
+	}
+
+	printf("\nDone\n");
+
+	close(devfd);
+	close(datafd);
+	free(blockbuf);
+	return 0;
+}