Re: [PATCH V4 2/3] mtd: nand: Calculate better default ecc layout

Re: [PATCH V4 2/3] mtd: nand: Calculate better default ecc layout

[David Brownell]

On Friday 26 June 2009, Troy Kisky wrote:
> Looking through the code, I can only see that
> Davinci is effected this way when using 512 bytes
> x 4 steps with hardware ecc.

You mean, with *single-bit* hardware ECC.

Currently each step needs just three bytes,
but it's using the default ECC layout which
allocates six bytes per step.


> It has its ecc 
> bytes moved from oob 40-63 to oob 52-63.

More correctly:  40-42, 46-48, 52-54, 58-60
to 52-63.


> So a Davinci authority will need to ack this,
> or request a change.

Actually I'd just stick with the standard policy
and not make such an incompatible change in the
first place.  You can't guarantee that the change
won't cause regressions...

It would have been nice if the MTD layer were
doing this ECC layout before, but in this specific
case I can't say I think an extra 12 bytes of OOB
data will matter to anyone.

- Dave

Re: [PATCH V4 2/3] mtd: nand: Calculate better default ecc layout

[Troy Kisky]

David Brownell wrote:
> On Friday 26 June 2009, Troy Kisky wrote:
>> Looking through the code, I can only see that
>> Davinci is effected this way when using 512 bytes
>> x 4 steps with hardware ecc.
> 
> You mean, with *single-bit* hardware ECC.

Yes. I do.

> 
> Currently each step needs just three bytes,
> but it's using the default ECC layout which
> allocates six bytes per step.
> 
> 
>> It has its ecc 
>> bytes moved from oob 40-63 to oob 52-63.
> 
> More correctly:  40-42, 46-48, 52-54, 58-60
> to 52-63.

No. More correctly,
40-42  -> 52-54
43-45  -> 55-57
46-48  -> 58-60
49-51  -> 61-63
52-63  -> free space 40-51

> 
> 
>> So a Davinci authority will need to ack this,
>> or request a change.
> 
> Actually I'd just stick with the standard policy
> and not make such an incompatible change in the
> first place.  You can't guarantee that the change
> won't cause regressions...
> 
> It would have been nice if the MTD layer were
> doing this ECC layout before, but in this specific
> case I can't say I think an extra 12 bytes of OOB
> data will matter to anyone.

That was only a side benefit, not the main reason for the patch.

> 
> - Dave


I am perfectly willing to add platform data to the current
evm's to maintain current behavior. As for regressions, could
the next tree help find these?? I have looked fairly closely,
but I definitely could have missed something.


I thank you very much for taking the time to comment upon this.
I'd much rather have a nak than silence.


If anyone ever wants me to post an updated version of the patch
give me a shout. Otherwise, I won't waste anymore of the lists time.


Thanks
Troy