ARM: LPC32XX SLC ECC Handling

[Brian Norris]

Hi,

On Wed, Aug 29, 2012 at 4:52 AM, Roland Stigge <stigge at antcom.de> wrote:
> On 08/29/2012 10:33 AM, Artem Bityutskiy wrote:
>>>>> the LPC32XX SLC controller has a ECC feature that covers only
>>>>> the main data area, but not the OOB data.  Is it the
>>>>> responsibility of the user to protect its OOB data via an
>>>>> ECC?  The LPC32XX MLC controller on the other hand covers the
>>>>> complete OOB data with its ECC.  How can a user determine if
>>>>> it has to protect its OOB data?

I haven't dealt with this question, since I basically don't support
user data in OOB on my system. I'm basically using only UBI. But see
my other comments below.

>>>> I guess it is good idea to CC the author of the drivers.
>>>
>>> Thanks for CC'ing, would have missed it.
>>>
>>> Yes, it's exactly as Sebastian described - only the MLC
>>> controller covers OOB in ECC.
>>>
>>> At this point, I need to give back the question to the MTD
>>> maintainers: Is there an API how we can help users at this point?
>>> I would be happy to implement it.
>>
>> Well, I guess from the user's POW the driver either protects OOB
>> data with ECC or not. There is no interface for detecting this in
>> MTD, but I do not see why someone could not create it.
>>
>> Am I right that we are talking about the situation when OOB and
>> data areas are covered by different ECCs, or this is about user
>> writing data +oob at one go, and everything is covered by the same
>> ECC?
>
> In the above case of LPC32xx/MLC NAND (not sure if this is common with
> other controllers),

By my experience in interacting with the community, few controllers
have support for protecting OOB with ECC, and even fewer drivers
actual implement it. I think my (out-of-tree) driver was the only one
that could ever reported bitflips for OOB transactions. And for
instance, I added the read_oob/read_oob_raw and
write_oob/write_oob_raw clarification in nand_ecc_ctrl, but there are
no in-tree users that differentiate the two...

> the controller covers 512 (sub)pages plus some OOB
> bytes with a single ECC operation, storing it behind those covered OOB.

My controller is basically the same.

Brian