[Gumstix-users] NAND bad block regression

[roystonvasey <mik...@ma...>]

Dear List,
There seems to be a regression problem with the MTD driver in kernel 2.6.36.
Essentially any Overos with manufacturer marked bad blocks in the NAND will
fail.
During an erase cycle any bad blocks can be seen:
flash_eraseall -j /dev/mtd4

Erasing 128 Kibyte @ 5400000 -- 33 % complete. Cleanmarker written at
53e0000.
flash_eraseall: /dev/mtd4: MTD Erase failure: Input/output error
Erasing 128 Kibyte @ 6780000 -- 41 % complete. Cleanmarker written at
6760000.
flash_eraseall: /dev/mtd4: MTD Erase failure: Input/output error
Erasing 128 Kibyte @ cac0000 -- 81 % complete. Cleanmarker written at
caa0000.
flash_eraseall: /dev/mtd4: MTD Erase failure: Input/output error
Erasing 128 Kibyte @ f980000 -- 100 % complete.Cleanmarker written at
f960000.

Under 2.6.34 the file system can be mounted and used without any problem.
Under 2.6.36 the console goes mental with "uncorrectable error :" messages
inter dispersed with "mtd->read(0x1fc08 bytes from 0xcac03f8) returned ECC
error" messages until the kernel finally gives up.

I Googled about and found reference to this from August 2010.
"Commit c7b28e25cb9beb943aead770ff14551b55fa8c79 caused a regression
in detection of factory-set bad block markers, especially for certain
small-page NAND. This fix removes some unneeded constraints on using
NAND_SMALL_BADBLOCK_POS, making the detection code more correct.

This regression can be seen, for example, in Hynix HY27US081G1M and
similar."
I tried the suggested patch but that didn't solve the problem.

Has anyone else seen this or have any insight into the problem?

Cheers Mike.

-- 
View this message in context: http://old.nabble.com/NAND-bad-block-regression-tp31891148p31891148.html
Sent from the Gumstix mailing list archive at Nabble.com.