is nand ecc correction logic sane enough?

for some time now, I have gone with the belief that the nand ecc imlementation such as this which I helped write good enough for nand devices, but Vimal Singh pointed me to these two articles:
http://lists.infradead.org/pipermail/linux-mtd/2008-August/022611.html
http://lists.infradead.org/pipermail/linux-mtd/2008-August/022613.html

Essentially:
If we do just old_ecc^new_ecc to detect/correct ecc errors - the logic handles the case of bit flip error at 63rd bit(as e.g.) faultily. I was completely unaware of this weakness of the algo.. thanks Vimal for pointing me here..

I have plans of setting up a bit of code to verify this (essentially hack up the current drivers by overriding the read path ;) ).. will post if I can proove(I suspect this would be the case) this is the case with gpmc..

Comments

Popular posts from this blog

simpler alternative to minicom

Quick and Dirty userspace testing of regulator with virtual consumer

Setting up Email forwarding System for GIT