[RFC] Change ECC algorithm from userspace

[Javier Martinez Canillas]

On Fri, Oct 28, 2011 at 1:13 PM, Angus CLARK <angus.clark at st.com> wrote:
> Hi Javier, Atlant,
>

Hi Angus,

> NAND Flash, coupled with an appropriate ECC scheme, is every bit (excuse the
> pun!) reliable as NOR Flash (which typically does not employ any ECC).
>
> Many of the current SLC devices that recommend 4-bit ECC, also guarantee block
> zero to be good with 1-bit ECC, allowing the part to be used with existing 1-bit
> Hamming ECC boot-loaders.  However, this is only valid if your boot-code can fit
> into 1 block.
>

Yes, in fact the NAND that we are using guarantee that the first block
can be reliable with 1-bit ECC if it written less than an threshold
value.
That is why I think we can use this NAND for boot.

> We also faced a similar issue in the past, requiring different ECC schemes for
> the boot code and for the remainder of the Flash.  Our approach was based on MTD
> partitions - each partition could have different ECC scheme, with the respective
> 'nand_ecc_ctrl' structures swapped in and out as necessary.
>

What is wasn't clear for me was if I could do it from user-space or if
I had to modify the OMAP NAND driver to switch the layouts and and ECC
scheme for each partition.

> However, as suggested by others, computing the ECC in software, and using raw
> programming is much simpler!
>
> Cheers,

Yes, I think we should also use that approach. It seems simpler and
less intrusive than changing the driver.

Thanks a lot to everyone for the help!

Best regards,

-- 
Javier Martínez Canillas
(+34) 682 39 81 69
Barcelona, Spain