NAND bitflip problems

Hello everyone, I need some urgent help over an issue that I have just discovered a few days ago:

I have bit flip problems when handling big files (180MB) in my UBIFS filesystem, and I really need to fix this problem as soon as possible. The system seems to handle small files without problems.

My system is a TI DM816X processor using a Micron 29F4G08ABADA NAND flash chip. The problem manifests even on the TI DM8168 that we have running latest U-Boot and Kernel found on the EZSDK v5.05.01.04 . The EVM is booting entirely from NAND, where we store U-Boot, Kernel, and UBIFS rootfs.

Using the information found in http://processors.wiki.ti.com/index.php/UBIFS_Support , this is the UBI partition configuration we are using:

mkfs.ubifs -r /path/to/rootfs -m 2048 -e 126976 -c 3977 -o rootfs.ubifs

ubinize -o rootfs.ubinized.img -m 2048 -p 128KiB -s 2048 -O 2048 ubinize.cfg

ubinize.cfg content:

[ubifs]

mode=ubi

image=rootfs.ubifs

vol_id=0

vol_size=481MiB

vol_type=dynamic

vol_name=rootfs

vol_flags=autoresize

=============================

These are the bootargs passed to the kernel:

console=ttyO2,115200n8 rootwait rw mem=256M quiet notifyk.vpssm3_sva=0xBF900000 vram=50M ti816xfb.vram=0:16M,1:16M,2:6M mtdparts=omap2-nand.0:1024k(uboot),3072k(kernel),-(rootfs) ubi.mtd=2,2048 root=ubi0:rootfs rw rootfstype=ubifs"

The problem happens when copying big files. Sometimes you can see the efects of the bit-flip immediately, and sometimes you need to reboot the system in order for the problem to become visible. Many reboots result in different md5sums of the same file.

We have verified that the internal ECC mechanism of the NAND chip is disabled, and reading the kernel code, it seems that the DM8168 should be using its BCH8 ECC scheme.

Any thoughts on how we should proceed in order to debug/fix this? We will be willing to share more details if they are helpful to find a solution. Thank you very much in advance.

Xabier.