MTD/YAFFS2 page problem, data missing

MTD/YAFFS2 page problem, data missing

[Kjetil Aamodt]

>From time to time we have seen data corruption on our nand flash. Closer investigation with a 300MB large file showed us that 32 of 98 write-boot-read cycles got corrupted. The corruption were of two very similar forms:

a) Somewhere in the image one complete page, page M, were replaced by the contents of M+1. Page M+1, and all other data in the file were correct (data from the same page is returned twice).

b) Somewhere in the image there were two consecutive page errors. In the corrupted file page N were missing and replaced with N+1. Page N+1 contained only zeros. The rest of the data in the file were correct.

Our Linux kernel version are 2.6.22.1. The YAFFS version is per CVS of November 13. 2007.

Have anyone else seen such behavior with MTD and/or YAFFS2?
Do you have any suggestion in how to debug this?

Best regards,
Kjetil Aamodt
TANDBERG

Re: MTD/YAFFS2 page problem, data missing

[falls huang]

Hello Kjetil!

    I don't the exactly status about your board and testing method .
Maybe I can give u some suggests:

1.  Do testing with such cycle ->
erase the yaffs2 partition that you write to , write , boot , read

2.
Did u open the debug options in /proc/yaffs ?
FYI: http://www.yaffs.net/yaffs-debugging , you can open the option:
YAFFS_TRACE_WRITE

3.
Read your nand flash driver carefully, yaffs2 based on mtd , mtd based
on nandflash driver .

Gook Luck

2007/12/19, Kjetil Aamodt <kjetil.aamodt@tandberg.com>:
> >From time to time we have seen data corruption on our nand flash. Closer investigation with a 300MB large file showed us that 32 of 98 write-boot-read cycles got corrupted. The corruption were of two very similar forms:
>
> a) Somewhere in the image one complete page, page M, were replaced by the contents of M+1. Page M+1, and all other data in the file were correct (data from the same page is returned twice).
>
> b) Somewhere in the image there were two consecutive page errors. In the corrupted file page N were missing and replaced with N+1. Page N+1 contained only zeros. The rest of the data in the file were correct.
>
> Our Linux kernel version are 2.6.22.1. The YAFFS version is per CVS of November 13. 2007.
>
> Have anyone else seen such behavior with MTD and/or YAFFS2?
> Do you have any suggestion in how to debug this?
>
> Best regards,
> Kjetil Aamodt
> TANDBERG
>
>
> ______________________________________________________
> Linux MTD discussion mailing list
> http://lists.infradead.org/mailman/listinfo/linux-mtd/
>


-- 
Regards
Falls Huang