Skip to contentSkip to conversation

fw_env: add NAND support

PATCH New Guennadi Liakhovetski 3 participants 12 messages Sep 2, 2008 – Sep 3, 2008

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fw_env: add NAND support

Add support for environment in NAND with automatic NOR / NAND recognition, including unaligned environment, bad-block skipping, redundant environment copy.

---

This is now a single patch. No union any more, instead a struct with pointers into a flat image buffer is used. Still, MTD_VERSION=old doesn't work, it looks it is also broken in the current version.

tools/env/README +6 -0
index f8a644e..f32f872 100644
@@ -22,9 +22,11 @@ following lines are relevant:
 #define DEVICE1_OFFSET    0x0000
 #define ENV1_SIZE         0x4000
 #define DEVICE1_ESIZE     0x4000
+#define DEVICE1_ENVSECTORS     2
 #define DEVICE2_OFFSET    0x0000
 #define ENV2_SIZE         0x4000
 #define DEVICE2_ESIZE     0x4000
+#define DEVICE2_ENVSECTORS     2
 
 Current configuration matches the environment layout of the TRAB
 board.
@@ -46,3 +48,7 @@ then 1 sector.
 
 DEVICEx_ESIZE defines the size of the first sector in the flash
 partition where the environment resides.
+
+DEVICEx_ENVSECTORS defines the number of sectors that may be used for
+this environment instance. On NAND this is used to limit the range
+within which bad blocks are skipped, on NOR it is unused.
tools/env/fw_env.c +531 -262

1,048 lines not on this page · raw message

tools/env/fw_env.config +5 -1
index 2432bd8..0fe37c9 100644
@@ -1,7 +1,11 @@
 # Configuration file for fw_(printenv/saveenv) utility.
 # Up to two entries are valid, in this case the redundand
 # environment sector is assumed present.
+# Notice, that the "Number of sectors" is ignored on NOR.
 
-# MTD device name	Device offset	Env. size	Flash sector size
+# MTD device name	Device offset	Env. size	Flash sector size	Number of sectors
 /dev/mtd1		0x0000		0x4000		0x4000
 /dev/mtd2		0x0000		0x4000		0x4000
+
+# NAND example
+#/dev/mtd0		0x4000		0x4000		0x20000			2
#1 lore raw

PATCH 1/1 fw_env: add NAND support

Dear Guennadi Liakhovetski,

Quoted 10 lines · In message <[email protected]> you wrote:

Add support for environment in NAND with automatic NOR / NAND recognition, including unaligned environment, bad-block skipping, redundant environment copy.

---

This is now a single patch. No union any more, instead a struct with pointers into a flat image buffer is used. Still, MTD_VERSION=old doesn't work, it looks it is also broken in the current version.

I wonder why you didn't even try to fix it.

See [PATCH] fw_env.c: fix build problems with MTD_VERSION=old

Please rebase your patch and resubmit.

Quoted 17 lines
tools/env/README +2 -0
index f8a644e..f32f872 100644
@@ -22,9 +22,11 @@ following lines are relevant:
 #define DEVICE1_OFFSET    0x0000
 #define ENV1_SIZE         0x4000
 #define DEVICE1_ESIZE     0x4000
+#define DEVICE1_ENVSECTORS     2
 #define DEVICE2_OFFSET    0x0000
 #define ENV2_SIZE         0x4000
 #define DEVICE2_ESIZE     0x4000
+#define DEVICE2_ENVSECTORS     2
 
 Current configuration matches the environment layout of the TRAB
 board.

tools/env/README then 1 sector. · line 54

Quoted 7 lines
tools/env/README +4 -0
@@ -46,3 +48,7 @@ then 1 sector.
 
 DEVICEx_ESIZE defines the size of the first sector in the flash
 partition where the environment resides.
+
+DEVICEx_ENVSECTORS defines the number of sectors that may be used for
+this environment instance. On NAND this is used to limit the range
+within which bad blocks are skipped, on NOR it is unused.

I think "...on NOR it is not used" would be clearer ?

Quoted 7 lines
tools/env/README +7 -0
+static int flash_read_buf (int dev, int fd, void *buf, size_t count,
+			   off_t offset)
+{
+	struct mtd_info_user mtdinfo;
+	/* erase / write length - one block on NAND, 0 on NOR */
+	size_t blocklen;
+	/* progress counter */

tools/env/README

Quoted 10 lines
tools/env/README +10 -0
+	size_t processed = 0;
+	/* current read length */
+	size_t readlen = count;
+	/* end of the last block we may use */
+	off_t top_of_range;
+	/* offset inside the current block to the start of the data */
+	off_t block_seek;
+	/* running start of the current block - MEMGETBADBLOCK needs 64 bits */
+	loff_t blockstart;
+	int rc;

That's difficult to read. Please reformat (comments on smae line with code, if necessary continued on next line).

tools/env/README

Quoted 5 lines
tools/env/README +5 -0
+	/*
+	 * Start of the first block to be read, relies on the fact, that
+	 * erase sector size is always a power of 2
+	 */
+	blockstart = offset & ~(DEVESIZE (dev) - 1);

Insert empty line here.

Quoted 2 lines
tools/env/README +2 -0
+	/* Offset inside a block */
+	block_seek = offset - blockstart;

...

tools/env/README

Quoted 1 line
tools/env/README +1 -0
+		/* Limit to one block for the first read */

Why?

Quoted 11 lines
tools/env/README +11 -0
+		if (readlen > blocklen - block_seek)
+			readlen = blocklen - block_seek;
+	} else {
+		blocklen = 0;
+		top_of_range = offset + count;
+	}
+
+	/* This only runs once on NOR flash */
+	while (processed < count) {
+		rc = flash_bad_block (dev, fd, &mtdinfo, &blockstart);
+		if (rc < 0) {

tools/env/README

Quoted 10 lines
tools/env/README +8 -0
+			return -1;
+		} else if (blockstart + block_seek + readlen > top_of_range) {
+			/* End of range is reached */
 			fprintf (stderr,
+				 "Too few good blocks within range\n");
+			return -1;
+		} else if (rc) {
+			blockstart += blocklen;
+			continue;
 		}

Please rewrite as:

		if (rc < 0)		/* block test failed */
			return -1;

		if (blockstart + block_seek + readlen > top_of_range) {
			/* End of range is reached */
			fprintf (stderr,
				"Too few good blocks within range\n");
			return -1;
		}

		if (rc) {		/* block is bad */
			blockstart += blocklen;
			continue;
		}

I can much easier follow the flow when written that way.

tools/env/README

Quoted 6 lines
tools/env/README +6 -0
+
+		/*
+		 * If a block is bad, we retry in the next block at the same
+		 * offset - see common/env_nand.c::writeenv()
+		 */
+		lseek (fd, blockstart + block_seek, SEEK_SET);

Hm... there was a "continue" for the badblock case just above, so do we really try the _next_ block here?

Quoted 14 lines
tools/env/README +14 -0
+static int __flash_write_buf (int dev, int fd, void *buf, size_t count,
+			      off_t offset, struct mtd_info_user *mtdinfo)
+{
+	char *data = NULL;
+	erase_info_t erase;
+	/* length of NAND block / NOR erase sector */
+	size_t blocklen;
+	/* whole area that can be erased - may include bad blocks */
+	size_t erase_len;
+	/* erase / write length - one block on NAND, whole area on NOR */
+	size_t erasesize;
+	/* progress counter */
+	size_t processed = 0;
+	/* total size to actually write - excludinig bad blocks */

tools/env/README

Quoted 10 lines
tools/env/README +10 -0
+	size_t write_total;
+	/* offset to the first erase block (aligned) below offset */
+	off_t erase_offset;
+	/* offset inside the erase block to the start of the data */
+	off_t block_seek;
+	/* end of the last block we may use */
+	off_t top_of_range;
+	/* running start of the current block - MEMGETBADBLOCK needs 64 bits */
+	loff_t blockstart;
+	int rc;

Not readable. Please reformat.

tools/env/README

Quoted 3 lines
tools/env/README +3 -0
+	/*
+	 * Data size we actually have to write: from the start of the block
+	 * to the start of the data, then count bytes ob data, and to the

-------------------------------------------------------^ f

tools/env/README

Quoted 2 lines
tools/env/README +2 -0
+	 * end of the block
+	 */

Hm... sounds as if this was always exactly one full block, then?

tools/env/README

Quoted 9 lines
tools/env/README +9 -0
+	write_total = (block_seek + count + blocklen - 1) & ~(blocklen - 1);
+
+	/*
+	 * Support data anywhere within erase sectors: read out the complete
+	 * area to be erased, replace the environment image, write the whole
+	 * block back again.
+	 */
+	if (write_total > count) {
+		data = malloc (erase_len);

My understanding is, that erase_len can be > block size. Is this correct?

I don't see where the actual size of the environment data is taken into considration?

tools/env/README

Quoted 2 lines
tools/env/README +2 -0
+	} else {
+		/* Offset is block-aligned by construction here */

What does "block-aligned by construction" mean?

tools/env/README

Quoted 1 line
tools/env/README +1 -0
+	if (mtdinfo->type == MTD_NANDFLASH)

Add ----------------------------------------> {

tools/env/README

Quoted 6 lines
tools/env/README +6 -0
+		/*
+		 * NAND: calculate which blocks we are writing. We have
+		 * to write one block at a time to skip bad blocks.
+		 */
+		erasesize = blocklen;
+	else
	} else {

tools/env/README

Quoted 1 line
tools/env/README +1 -0
+		erasesize = erase_len;
	}

tools/env/README

Quoted 3 lines
tools/env/README +3 -0
+	erase.length = erasesize;
+
+	/* This only runs once on NOR flash */

How comes that?

tools/env/README

Quoted 11 lines
tools/env/README +10 -0
+	while (processed < write_total) {
+		rc = flash_bad_block (dev, fd, mtdinfo, &blockstart);
+		if (rc < 0) {
+			return rc;
+		} else if (blockstart + erasesize > top_of_range) {
+			fprintf (stderr, "End of range reached, aborting\n");
+			return -1;
+		} else if (rc) {
+			blockstart += blocklen;
+			continue;
 		}

Please rewrite - see above.

tools/env/README

Quoted 3 lines
tools/env/README +3 -0
+		processed += blocklen;
+		block_seek = 0;
+		blockstart += blocklen;
		processed  += blocklen;
		blockstart += blocklen;
		block_seek = 0;

tools/env/README

Quoted 1 line
tools/env/README +1 -0
+	printf ("Writing new environment at 0x%lx\n", DEVOFFSET (dev_target));

If you really want to keep this printf(), thenplease also print the device name (i. e try to stick to the format of the output of the old version and at least provide the same information).

You may also consider the comments posted before, that such informa- tion should be printed only when the verbose option (argument -v) was selected.

Quoted 11 lines

static int flash_write (void)

{
        int fd_current, fd_target, rc, dev_target;
 
        /* dev_current: fd_current, erase_current */
        if ((fd_current = open (DEVNAME (dev_current), O_RDWR)) < 0) { 
                fprintf (stderr,
                         "Can't open %s: %s\n",
                         DEVNAME (dev_current), strerror (errno));
                return -1;
        }

Please factor out the open()/close() as discussed several times before.

tools/env/README

Quoted 8 lines
tools/env/README +8 -0
+	if (rc < 0)
+		return rc;
+
+	if (HaveRedundEnv) {
+		off_t offset = DEVOFFSET (dev_current) +
+			offsetof(struct env_image_redundant, flags);
+		printf ("Setting obsolete flag for environment at 0x%lx\n",
+			DEVOFFSET (dev_current));

Please make this a debug print only.

In general, please check printed output.

tools/env/README

Quoted 9 lines
tools/env/README +7 -0
+		flash_flag_obsolete(dev_current, fd_current, offset);
+	}
+
+	if (close (fd_target)) {
 		fprintf (stderr,
+			 "I/O error on %s: %s\n",
+			 DEVNAME (dev_target), strerror (errno));
+		return -1;
 	}

Please factor out the open()/close() as discussed several times before.

Quoted 10 lines
tools/env/README +10 -0
+static int flash_read (void)
+{
+	int fd, rc;
+
+	if ((fd = open (DEVNAME (dev_current), O_RDONLY)) < 0) {
+		fprintf (stderr,
+			 "Can't open %s: %s\n",
+			 DEVNAME (dev_current), strerror (errno));
+		return -1;
+	}

...

tools/env/README

Quoted 6 lines
tools/env/README +6 -0
+	if (close (fd)) {
+		fprintf (stderr,
+			 "I/O error on %s: %s\n",
+			 DEVNAME (dev_current), strerror (errno));
+		return -1;
+	}

Please factor out the open()/close() as discussed several times before.

tools/env/README

Quoted 1 line
tools/env/README +1 -0
+	return rc < 0 ? rc : 0;
	return (rc < 0) ? rc : 0;

tools/env/README static char *envmatch (char * s1, char * s2) · line 850

Quoted 8 lines
tools/env/README +2 -0
@@ -596,107 +844,120 @@ static char *envmatch (char * s1, char * s2)
 static int env_init (void)
 {
 	int crc1, crc1_ok;
+	char flag1, *addr1;
 
 	int crc2, crc2_ok;
+	char flag2, *addr2;

I think you should number these 0 an 1, respectively here.

Because then you can read this code much better:

tools/env/README static char *envmatch (char * s1, char * s2) · line 854

Quoted 4 lines
tools/env/README +2 -0
 		if (crc1_ok && !crc2_ok) {
+			dev_current = 0;
 		} else if (!crc1_ok && crc2_ok) {
+			dev_current = 1;

would become:

		if (crc0_ok && !crc1_ok) { 
			dev_current = 0;
		} else if (!crc0_ok && crc1_ok) {
			dev_current = 1;

If "0" is OK, then use "0"; if "1" is ok then use "1".

Your version: if "1" is OK then use "0", if "2" is OK then use "1" is more difficult to read.

[Or stick with the old identifiers, i. e. use 1 and 2 consistently, but don't mix 0/1 here and 1/2 there.]

...

tools/env/README static char *envmatch (char * s1, char * s2) · line 857

Quoted 3 lines
tools/env/README +3 -0
+			/* prepare for device 2 */
+			dev_current = 0;
+		/* From here: both CRCs correct */

The comment doesn't fit (indentation is wrong), and it's obvious anyway - omit it.

tools/env/README static char *envmatch (char * s1, char * s2) · line 858

Quoted 1 line
tools/env/README +1 -0
+		/* From here: invalid flag configuration */

Ditto.

tools/env/README static char *envmatch (char * s1, char * s2) · line 862

Quoted 4 lines
tools/env/README +4 -0
+		/*
+		 * If we are reading, we don't need the flag and the crc any
+		 * more, if we are writing, we will set them before writing out
+		 */

Note that this is a serious impairment compared to the old version.

The whole logic of writing the envrionment (and especially so when redundancy is used) is based on separate logic steps that are per- formed strictly sequentially, and only when the previous one was successfully completed. This essential to maintain consistent data. For the redundant case that means:

1) write the environment data to flash.
*After* this has completed, make the data valid by
2) writing the CRC checksum.
*After* this has completed, make this copy of the environment valid by
3) writing the valid flag to this copy.
*After* this has completed, make the other copy of the environment obsolete by
4) writing the obsolete flag to the other copy.

Your new implementation breaks this by writing out the whole buffer at once. This is not a good ideas, because you may end up ith situations that were impossible before, like having a valid flag byte in flash even though the environment data were not completely written.

I am aware that you have little options with NAND flash, but for NOR this is IMHO a serious change for the worse.

Can we please avoid this, and re-establish the old mode of operation, which was designed for reliability?

tools/env/README static char *envmatch (char * s1, char * s2) · line 869

Quoted 7 lines
tools/env/README +4 -0
+	if ((fp = fopen (fname, "r")) == NULL)
+		return -1;
 
+	while (i < 2 && fgets (dump, sizeof (dump), fp)) {
 		/* Skip incomplete conversions and comment strings */
+		if (dump[0] == '#')
 			continue;

I think you must initialize ENVSECTORS(i) here...

tools/env/README static char *envmatch (char * s1, char * s2) · line 882

Quoted 13 lines
tools/env/README +13 -0
+		rc = sscanf (dump, "%s %lx %lx %lx %lx",
+			     DEVNAME (i),
+			     &DEVOFFSET (i),
+			     &ENVSIZE (i),
+			     &DEVESIZE (i),
+			     &ENVSECTORS (i));
+
+		if (rc < 4)
+			continue;
+
+		if (rc < 5)
+			/* Default - 1 sector */
+			ENVSECTORS (i) = 1;

Because if rc < 4, you already continued and left ENVSECTORS uninitialized.

Best regards,

Wolfgang Denk

Signature · 6 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: [email protected]
It became apparent that one reason why the Ice Giants were  known  as
the  Ice  Giants  was  because they were, well, giants. The other was
that they were made of ice.              -Terry Pratchett, _Sourcery_
#2 lore raw

PATCH 1/1 fw_env: add NAND support

Quoted 6 lines · Wolfgang Denk
Quoted 2 lines
tools/env/fw_env.c +2 -0
+	/* Offset inside a block */
+	block_seek = offset - blockstart;

...

tools/env/fw_env.c int fw_setenv (int argc, char *argv[]) · line 504

Quoted 1 line
tools/env/fw_env.c +1 -0
+		/* Limit to one block for the first read */

Why?

Because this is on NAND, there we have to perform all IO block at a time.

Quoted 9 lines

tools/env/fw_env.c int fw_setenv (int argc, char *argv[]) · line 531

Quoted 6 lines
tools/env/fw_env.c +6 -0
+
+		/*
+		 * If a block is bad, we retry in the next block at the same
+		 * offset - see common/env_nand.c::writeenv()
+		 */
+		lseek (fd, blockstart + block_seek, SEEK_SET);

Hm... there was a "continue" for the badblock case just above, so do we really try the _next_ block here?

I meant, if the first block with non-zero offset (block_seek) was bad, then in the _next_ after it block we try at the same offset.

Quoted 8 lines

tools/env/fw_env.c int fw_setenv (int argc, char *argv[]) · line 599

Quoted 3 lines
tools/env/fw_env.c +3 -0
+	/*
+	 * Data size we actually have to write: from the start of the block
+	 * to the start of the data, then count bytes ob data, and to the

-------------------------------------------------------^ f

Review

Quoted 2 lines
(quoted diff) +2 -0
+	 * end of the block
+	 */

Hm... sounds as if this was always exactly one full block, then?

or two, or three...

Quoted 4 lines
Quoted 4 lines
tools/env/fw_env.c +4 -0
+	write_total = (block_seek + count + blocklen - 1) & ~(blocklen - 1);
+
+	/*
+	 * Support data anywhere within erase sectors: read out the complete
Quoted 8 lines

tools/env/fw_env.c int fw_setenv (int argc, char *argv[]) · line 610

Quoted 5 lines
tools/env/fw_env.c +5 -0
+	 * area to be erased, replace the environment image, write the whole
+	 * block back again.
+	 */
+	if (write_total > count) {
+		data = malloc (erase_len);

My understanding is, that erase_len can be > block size. Is this correct?

Yes.

Quoted 2 lines

I don't see where the actual size of the environment data is taken into considration?

in the "count" function parameter, which is then used to calculate write_total.

Quoted 4 lines

tools/env/fw_env.c int fw_setenv (int argc, char *argv[]) · line 625

Quoted 2 lines
tools/env/fw_env.c +2 -0
+	} else {
+		/* Offset is block-aligned by construction here */

What does "block-aligned by construction" mean?

It means, by the way I constructed (calculated) lengths above this place, we land here only if offset is block-aligned.

Quoted 5 lines

tools/env/fw_env.c int fw_setenv (int argc, char *argv[]) · line 641

Quoted 3 lines
tools/env/fw_env.c +3 -0
+	erase.length = erasesize;
+
+	/* This only runs once on NOR flash */

How comes that?

Because on NOR I set "erasesize" to the total length of the data plus alignment.

Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.

DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-0 Fax: +49-8142-66989-80 Email: office at denx.de

#3 lore raw

PATCH 1/1 fw_env: add NAND support

In reply to Wolfgang Denk
Quoted 21 lines · Wolfgang Denk

tools/env/fw_env.c static char *envmatch (char * s1, char * s2) · line 947

Quoted 4 lines
tools/env/fw_env.c +4 -0
+		/*
+		 * If we are reading, we don't need the flag and the crc any
+		 * more, if we are writing, we will set them before writing out
+		 */

Note that this is a serious impairment compared to the old version.

The whole logic of writing the envrionment (and especially so when redundancy is used) is based on separate logic steps that are per- formed strictly sequentially, and only when the previous one was successfully completed. This essential to maintain consistent data. For the redundant case that means:

1) write the environment data to flash.
*After* this has completed, make the data valid by
2) writing the CRC checksum.
*After* this has completed, make this copy of the environment valid by
3) writing the valid flag to this copy.
*After* this has completed, make the other copy of the environment obsolete by
4) writing the obsolete flag to the other copy.

Quoted 5 lines

Your new implementation breaks this by writing out the whole buffer at once. This is not a good ideas, because you may end up ith situations that were impossible before, like having a valid flag byte in flash even though the environment data were not completely written.

Sorry, I still do not understand what bad can happen with this whole-image-at-once implementation. Yes, it may happen, that crc has been written, flag has been written, but the data has not completely been written. Then, you get a CRC error on reading. Just as if you were writing in portions and an error occurred when writing the data. Yes, the flag would have been different, with the previous approach you would still have the "redundant" flag set in the corrupted copy that you were trying to update and "active" in the other one. Writing the whole image at once you end up with two "active" flags. But this seems unimportant, because CRCs are checked before - at least in fw_env - so, in both cases you end up using the non-damaged copy.

If you were first setting the flag to "invalid, write in progress", then wrote the environment, then reset the flag to "valid, write completed successfully", then yes, writing per one write would be essentially different.

In short, I think, CRC provides sufficient protection of data integrity.

Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.

DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-0 Fax: +49-8142-66989-80 Email: office at denx.de

#4 lore raw

PATCH 1/1 fw_env: add NAND support

Dear Guennadi Liakhovetski,

Quoted 5 lines · In message <[email protected]> you wrote:

If you were first setting the flag to "invalid, write in progress", then wrote the environment, then reset the flag to "valid, write completed successfully", then yes, writing per one write would be essentially different.

But that's exactly what we are doing. The flag value for "invalid, write in progress" is 0xFF and is set as soon as we start to destroy the valid data by erasing the flash.

Quoted 1 line

In short, I think, CRC provides sufficient protection of data integrity.

CRC32 is not a perfect test. Yes, it's good enough for most ordinary cases, but when we can design in additional robustness we should do this. And we should not give up what we already have without very good reasons.

Best regards,

Wolfgang Denk

Signature · 5 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
The Wright Bothers weren't the first to fly. They were just the first
not to crash.
#5 lore raw

PATCH 1/1 fw_env: add NAND support

In reply to Wolfgang Denk
Quoted 26 lines · Wolfgang Denk

tools/env/fw_env.c static char *envmatch (char * s1, char * s2) · line 850

Quoted 8 lines
tools/env/fw_env.c +2 -0
@@ -596,107 +844,120 @@ static char *envmatch (char * s1, char * s2)
 static int env_init (void)
 {
 	int crc1, crc1_ok;
+	char flag1, *addr1;
 
 	int crc2, crc2_ok;
+	char flag2, *addr2;

I think you should number these 0 an 1, respectively here.

Because then you can read this code much better:

tools/env/fw_env.c static char *envmatch (char * s1, char * s2) · line 920

Quoted 4 lines
tools/env/fw_env.c +2 -0
 		if (crc1_ok && !crc2_ok) {
+			dev_current = 0;
 		} else if (!crc1_ok && crc2_ok) {
+			dev_current = 1;

would become:

		if (crc0_ok && !crc1_ok) { 
			dev_current = 0;
		} else if (!crc0_ok && crc1_ok) {
			dev_current = 1;
Quoted 7 lines

If "0" is OK, then use "0"; if "1" is ok then use "1".

Your version: if "1" is OK then use "0", if "2" is OK then use "1" is more difficult to read.

[Or stick with the old identifiers, i. e. use 1 and 2 consistently, but don't mix 0/1 here and 1/2 there.]

Sorry, don't understand. This is the original code:

		if (crc1_ok && !crc2_ok) {
			environment.data = addr1;
			environment.flags = flag1;
			environment.crc = crc1;
			curdev = 0;
			free (addr2);

Which looks to me like "if (crc1_ok...) {... curdev = 0;...}", which I tried to preserve. So, you would prefer me to _change_ it?

Quoted 10 lines

tools/env/fw_env.c static int get_config (char *fname) · line 1020

Quoted 7 lines
tools/env/fw_env.c +4 -0
+	if ((fp = fopen (fname, "r")) == NULL)
+		return -1;
 
+	while (i < 2 && fgets (dump, sizeof (dump), fp)) {
 		/* Skip incomplete conversions and comment strings */
+		if (dump[0] == '#')
 			continue;

I think you must initialize ENVSECTORS(i) here...

Quoted 15 lines

tools/env/fw_env.c static int get_config (char *fname) · line 1034

Quoted 13 lines
tools/env/fw_env.c +13 -0
+		rc = sscanf (dump, "%s %lx %lx %lx %lx",
+			     DEVNAME (i),
+			     &DEVOFFSET (i),
+			     &ENVSIZE (i),
+			     &DEVESIZE (i),
+			     &ENVSECTORS (i));
+
+		if (rc < 4)
+			continue;
+
+		if (rc < 5)
+			/* Default - 1 sector */
+			ENVSECTORS (i) = 1;

Because if rc < 4, you already continued and left ENVSECTORS uninitialized.

As far as I understand, with rc == 3 there is no DEVESIZE in the line, which doesn't look like a valid configuration line to me. The orginal code however accepted such lines and only dropped lines with rc < 3. I do not understand how the original code managed to work with rc == 3. As I thought this was a bug, I changed the test to "rc < 4", i.e., now I require at least 4 fields, in which case I initialise ENVSECTORS to the default value - 1 sector. If rc < 4 the counter "i" is not incremented and the line is dropped - in the same way as in the original version. Or am I missing something?

Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.

DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-0 Fax: +49-8142-66989-80 Email: [email protected]

#6 lore raw

PATCH 1/1 fw_env: add NAND support

Dear Guennadi,

Quoted 25 lines · In message <[email protected]> you wrote:
Quoted 8 lines

would become:

		if (crc0_ok && !crc1_ok) { 
			dev_current = 0;
		} else if (!crc0_ok && crc1_ok) {
			dev_current = 1;
Quoted 7 lines

If "0" is OK, then use "0"; if "1" is ok then use "1".

Your version: if "1" is OK then use "0", if "2" is OK then use "1" is more difficult to read.

[Or stick with the old identifiers, i. e. use 1 and 2 consistently, but don't mix 0/1 here and 1/2 there.]

Sorry, don't understand. This is the original code:

		if (crc1_ok && !crc2_ok) {
			environment.data = addr1;
			environment.flags = flag1;
			environment.crc = crc1;
			curdev = 0;
			free (addr2);
Quoted 2 lines

Which looks to me like "if (crc1_ok...) {... curdev = 0;...}", which I tried to preserve. So, you would prefer me to _change_ it?

...except that that nor "curdev =0" is present anywhere, but instead the index "1" is used consistently: crc1_ok, addr1, flag1, crc1.

Quoted 5 lines

As far as I understand, with rc == 3 there is no DEVESIZE in the line, which doesn't look like a valid configuration line to me. The orginal code however accepted such lines and only dropped lines with rc < 3. I do not understand how the original code managed to work with rc == 3. As I thought this was a bug, I changed the test to "rc < 4", i.e., now I

Did you verify that is was a bug?

Best regards,

Wolfgang Denk

Signature · 5 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: [email protected]
Ever try. Ever fail. No matter. Try again. Fail again.  Fail  better.
                                                        -- S. Beckett
#7 lore raw

PATCH 1/1 fw_env: add NAND support

Quoted 7 lines · Wolfgang Denk
Quoted 5 lines

As far as I understand, with rc == 3 there is no DEVESIZE in the line, which doesn't look like a valid configuration line to me. The orginal code however accepted such lines and only dropped lines with rc < 3. I do not understand how the original code managed to work with rc == 3. As I thought this was a bug, I changed the test to "rc < 4", i.e., now I

Did you verify that is was a bug?

Yes:

Unlocking flash...
Done
Cannot malloc -32768 bytes: Cannot allocate memory
Error: can't write fw_env to flash

Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.

DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-0 Fax: +49-8142-66989-80 Email: office at denx.de

#8 lore raw

PATCH 1/1 fw_env: add NAND support

In reply to Wolfgang Denk
Quoted 8 lines · Wolfgang Denk

Note that this is a serious impairment compared to the old version.

The whole logic of writing the envrionment (and especially so when redundancy is used) is based on separate logic steps that are per- formed strictly sequentially, and only when the previous one was successfully completed. This essential to maintain consistent data. For the redundant case that means:

Quoted 7 lines

1) write the environment data to flash.
*After* this has completed, make the data valid by
2) writing the CRC checksum.
*After* this has completed, make this copy of the environment valid by
3) writing the valid flag to this copy.
*After* this has completed, make the other copy of the environment obsolete by
4) writing the obsolete flag to the other copy.

I am afraid, I cannot confirm this in the current version. In it first CRC and flag is written with one write, then the data, then optionally any residual data. Which seems like not providing any additional robustness. Would you like me to replicate this behaviour, in which case we could just as well write the whole buffer at once as I did up to now, or change it to follow your scheme above? In the latter case I am not sure if this will work with all kernel versions with all mtd drivers and all NOR flashes.

Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.

DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-0 Fax: +49-8142-66989-80 Email: office at denx.de

#9 lore raw

PATCH 1/1 fw_env: add NAND support

In reply to Guennadi Liakhovetski

Dear Guennadi,

Quoted 9 lines · Guennadi Liakhovetski
tools/env/fw_env.c +9 -0
+static int flash_flag_obsolete (int dev, int fd, off_t offset)
+{
+	struct mtd_info_user mtdinfo;
+	int rc;
+
+	rc = ioctl (fd, MEMGETINFO, &mtdinfo);
+	if (rc < 0) {
+		perror ("Cannot get MTD information");
+		return -1;

tools/env/fw_env.c int fw_setenv (int argc, char *argv[]) · line 729

Quoted 10 lines
tools/env/fw_env.c +10 -0
+	}
+
+	if (mtdinfo.type == MTD_NANDFLASH) {
+		/*
+		 * No luck on NAND - we could save the erase, but have to write
+		 * a whole erase block anyway
+		 */
+		rc = __flash_write_buf (dev, fd, &obsolete_flag,
+					sizeof (obsolete_flag), offset,
+					&mtdinfo);

Hmmm, what exactly are you doing here? IIRC environment in NAND was implemented slightly differently than NOR in that it reused the obsolete flag as a serial number. This number is incremented for every write so the env with the higher number is always the current one (besides the corner case where the number overflows, see env_nand.c). This way the obsoleting can be completly avoided.

I can't see this in your patch?

Best regards
Markus

P.S.: As mentioned before this mechanism of course could save the "obsoleting" operation in NOR too...

Signature · 3 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-0 Fax: +49-8142-66989-80  Email: office at denx.de")
#10 lore raw

PATCH 1/1 fw_env: add NAND support

Dear Markus =?iso-8859-1?Q?Klotzb=FCcher?=,

Quoted 7 lines · In message <20080903115150.GA12440@lisa> you wrote:

Hmmm, what exactly are you doing here? IIRC environment in NAND was implemented slightly differently than NOR in that it reused the obsolete flag as a serial number. This number is incremented for every write so the env with the higher number is always the current one (besides the corner case where the number overflows, see env_nand.c). This way the obsoleting can be completly avoided.

Right, that was how you designed and implemented it.

Quoted 1 line

I can't see this in your patch?

I don't see it either. Thanks for pointing out.

Best regards,

Wolfgang Denk

Signature · 6 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Fools ignore complexity. Pragmatists suffer it. Some  can  avoid  it.
Geniuses remove it.
     - Perlis's Programming Proverb #58, SIGPLAN Notices, Sept.  1982
#11 lore raw

PATCH 1/1 fw_env: add NAND support

In reply to Markus Klotzbücher
Quoted 31 lines · Markus Klotzb?cher

Dear Guennadi,

Quoted 9 lines · Guennadi Liakhovetski
tools/env/fw_env.c +9 -0
+static int flash_flag_obsolete (int dev, int fd, off_t offset)
+{
+	struct mtd_info_user mtdinfo;
+	int rc;
+
+	rc = ioctl (fd, MEMGETINFO, &mtdinfo);
+	if (rc < 0) {
+		perror ("Cannot get MTD information");
+		return -1;

tools/env/fw_env.c int fw_setenv (int argc, char *argv[]) · line 729

Quoted 10 lines
tools/env/fw_env.c +10 -0
+	}
+
+	if (mtdinfo.type == MTD_NANDFLASH) {
+		/*
+		 * No luck on NAND - we could save the erase, but have to write
+		 * a whole erase block anyway
+		 */
+		rc = __flash_write_buf (dev, fd, &obsolete_flag,
+					sizeof (obsolete_flag), offset,
+					&mtdinfo);

Hmmm, what exactly are you doing here? IIRC environment in NAND was implemented slightly differently than NOR in that it reused the obsolete flag as a serial number. This number is incremented for every write so the env with the higher number is always the current one (besides the corner case where the number overflows, see env_nand.c). This way the obsoleting can be completly avoided.

Quoted 1 line

I can't see this in your patch?

No, it is not there, because I had no idea about it:-( Thanks for pointing out - will implement in the next version.

Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.

DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-0 Fax: +49-8142-66989-80 Email: office at denx.de