ECC in Nand_write_page

ECC in Nand_write_page

[Steve Tsai]

The board I used will display the message, "nand_write_ecc: Failed ECC
write", so I try to trace the function nand_write_ecc, I found it can
not write the ECC code into OOB sometime because the ECC in OOB are not
set as 0xff. There are two conditions to cause the problem, one
condition is the erase does not success, another condition is that the
page  was written before. Nand_write_page could write partial page to
the flash. Does it cause the ECC in OOB was written and ECC can not be
written next time. For example, nand_write_page write 100 byte the first
time and if it want to write 412 bytes to the same page next time, it
will fail. Does anyone have this problem?


Steve Tsai

Re: ECC in Nand_write_page

[Ken Offer]

On Thursday, July 11, 2002, at 05:19  AM, Steve Tsai wrote:

> The board I used will display the message, "nand_write_ecc: Failed ECC
> write", so I try to trace the function nand_write_ecc, I found it can
> not write the ECC code into OOB sometime because the ECC in OOB are not
> set as 0xff. There are two conditions to cause the problem, one
> condition is the erase does not success, another condition is that the
> page  was written before. Nand_write_page could write partial page to
> the flash. Does it cause the ECC in OOB was written and ECC can not be
> written next time. For example, nand_write_page write 100 byte 
> the first
> time and if it want to write 412 bytes to the same page next time, it
> will fail. Does anyone have this problem?

I have read that some NAND memory devices can read partial pages 
but must write full pages.  Could this be your problem?  In such 
cases you might have to read the page, make a write to part of 
the page and then write out the modified full page.

+-----------------------------------------------------------+
|  Ken Offer        koffer@arlut.utexas.edu         ARL:UT  |
|  Office: 512-835-3859                  Fax: 512-835-3259  |
+-----------------------------------------------------------+

RE: ECC in Nand_write_page

[Steve Tsai]

I don't think nand_write_page can write partial page, if someone enable
ECC. For example, when I write the partial page to the page 100, it will
write ECC code into the page also, but when I write the other data to
the empty area of the same page, I can write data into  the empty area
of the page, but I can not write the ECC code, because the area storing
ECC code was written at the previous time.  


Steve Tsai


-----Original Message-----
From: linux-mtd-admin@lists.infradead.org
[mailto:linux-mtd-admin@lists.infradead.org] On Behalf Of Ken Offer
Sent: Thursday, July 11, 2002 9:25 PM
To: Steve Tsai
Cc: LinuxMTD
Subject: Re: ECC in Nand_write_page


On Thursday, July 11, 2002, at 05:19  AM, Steve Tsai wrote:

> The board I used will display the message, "nand_write_ecc: Failed ECC

> write", so I try to trace the function nand_write_ecc, I found it can 
> not write the ECC code into OOB sometime because the ECC in OOB are 
> not set as 0xff. There are two conditions to cause the problem, one 
> condition is the erase does not success, another condition is that the

> page  was written before. Nand_write_page could write partial page to 
> the flash. Does it cause the ECC in OOB was written and ECC can not be

> written next time. For example, nand_write_page write 100 byte the 
> first time and if it want to write 412 bytes to the same page next 
> time, it will fail. Does anyone have this problem?

I have read that some NAND memory devices can read partial pages 
but must write full pages.  Could this be your problem?  In such 
cases you might have to read the page, make a write to part of 
the page and then write out the modified full page.

+-----------------------------------------------------------+
|  Ken Offer        koffer@arlut.utexas.edu         ARL:UT  |
|  Office: 512-835-3859                  Fax: 512-835-3259  |
+-----------------------------------------------------------+


______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/

Re: ECC in Nand_write_page

[Jörn Engel]

On Fri, 12 July 2002 10:39:36 +0800, Steve Tsai wrote:
> I don't think nand_write_page can write partial page, if someone enable
> ECC. For example, when I write the partial page to the page 100, it will
> write ECC code into the page also, but when I write the other data to
> the empty area of the same page, I can write data into  the empty area
> of the page, but I can not write the ECC code, because the area storing
> ECC code was written at the previous time.  

You are right. I've had a similar problem with ECC flash lately - NOR
flash, but the principle is the same. The ECC checksum is written to
the flash. By writing the same page twice, the checksum would have to
toggle some bits from 0 to 1, which is impossible for flash.
Thus you have a constraint to write any given page but once between
two erase cycles.

Joern

Re: ECC in Nand_write_page

[Thomas Gleixner]

On Thursday, 11. July 2002 12:19, Steve Tsai wrote:
> The board I used will display the message, "nand_write_ecc: Failed ECC
> write", so I try to trace the function nand_write_ecc, I found it can
> not write the ECC code into OOB sometime because the ECC in OOB are not
> set as 0xff. There are two conditions to cause the problem, one
> condition is the erase does not success, another condition is that the
> page  was written before. Nand_write_page could write partial page to
> the flash. Does it cause the ECC in OOB was written and ECC can not be
> written next time. For example, nand_write_page write 100 byte the first
> time and if it want to write 412 bytes to the same page next time, it
> will fail. Does anyone have this problem?

Do you use latest CVS code ?
Which filesystem do you use ?
There's only JFFS2, which can handle the nand problem correct.

Read http://www.linux-mtd.infradead.org/tech/nand.html !

1. If an erase fails, the JFFS2 does not use this block.

2. JFFS2 writes only full pages

3. ECC is definitely not written, if you write 100 bytes

	/* Calculate and write the ECC if we have enough data */
	if ((col < mtd->eccsize) && (last >= mtd->eccsize)) {
		nand_calculate_ecc (&this->data_buf[0], &(ecc_code[0]));
		for (i = 0; i < 3; i++)
			this->data_buf[(mtd->oobblock + oob_config.ecc_pos[i])] = ecc_code[i];
		if (oob_config.eccvalid_pos != -1)
			this->data_buf[mtd->oobblock + oob_config.eccvalid_pos] = 0xf0;
	}

	/* Calculate and write the second ECC if we have enough data */
	if ((mtd->oobblock == 512) && (last == mtd->oobblock)) {
		nand_calculate_ecc (&this->data_buf[256], &(ecc_code[3]));
		for (i = 3; i < 6; i++)
			this->data_buf[(mtd->oobblock + oob_config.ecc_pos[i])] = ecc_code[i];
		if (oob_config.eccvalid_pos != -1)
			this->data_buf[mtd->oobblock + oob_config.eccvalid_pos] &= 0x0f;
	}


-- 
Thomas
___________________________________________
linutronix - competence in embedded & realtime linux
http://www.linutronix.de
mail: tglx@linutronix.de

RE: ECC in Nand_write_page

[Steve Tsai]

Hi,
I am using 20020619 and JFFS2 file system. I did not set the following
option. Will it cause the error? Thanks.

CONFIG_MTD_NAND_ECC_JFFS2=y
CONFIG_JFFS2_FS_NAND=y

Steve Tsai

-----Original Message-----
From: linux-mtd-admin@lists.infradead.org
[mailto:linux-mtd-admin@lists.infradead.org] On Behalf Of Thomas
Gleixner
Sent: Friday, July 12, 2002 2:12 PM
To: Steve Tsai; Linux MTD mailing list
Subject: Re: ECC in Nand_write_page


On Thursday, 11. July 2002 12:19, Steve Tsai wrote:
> The board I used will display the message, "nand_write_ecc: Failed ECC

> write", so I try to trace the function nand_write_ecc, I found it can 
> not write the ECC code into OOB sometime because the ECC in OOB are 
> not set as 0xff. There are two conditions to cause the problem, one 
> condition is the erase does not success, another condition is that the

> page  was written before. Nand_write_page could write partial page to 
> the flash. Does it cause the ECC in OOB was written and ECC can not be

> written next time. For example, nand_write_page write 100 byte the 
> first time and if it want to write 412 bytes to the same page next 
> time, it will fail. Does anyone have this problem?

Do you use latest CVS code ?
Which filesystem do you use ?
There's only JFFS2, which can handle the nand problem correct.

Read http://www.linux-mtd.infradead.org/tech/nand.html !

1. If an erase fails, the JFFS2 does not use this block.

2. JFFS2 writes only full pages

3. ECC is definitely not written, if you write 100 bytes

	/* Calculate and write the ECC if we have enough data */
	if ((col < mtd->eccsize) && (last >= mtd->eccsize)) {
		nand_calculate_ecc (&this->data_buf[0], &(ecc_code[0]));
		for (i = 0; i < 3; i++)
			this->data_buf[(mtd->oobblock +
oob_config.ecc_pos[i])] = ecc_code[i];
		if (oob_config.eccvalid_pos != -1)
			this->data_buf[mtd->oobblock +
oob_config.eccvalid_pos] = 0xf0;
	}

	/* Calculate and write the second ECC if we have enough data */
	if ((mtd->oobblock == 512) && (last == mtd->oobblock)) {
		nand_calculate_ecc (&this->data_buf[256],
&(ecc_code[3]));
		for (i = 3; i < 6; i++)
			this->data_buf[(mtd->oobblock +
oob_config.ecc_pos[i])] = ecc_code[i];
		if (oob_config.eccvalid_pos != -1)
			this->data_buf[mtd->oobblock +
oob_config.eccvalid_pos] &= 0x0f;
	}


-- 
Thomas
___________________________________________
linutronix - competence in embedded & realtime linux
http://www.linutronix.de
mail: tglx@linutronix.de

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/