Re: [PATCH 00/23] mtd: rework ECC layout definition

From: Brian Norris
Date: Tue Jan 26 2016 - 14:27:00 EST


Hi Boris,

On Mon, Dec 07, 2015 at 11:25:55PM +0100, Boris Brezillon wrote:
> Hello,
>
> This patchset aims at getting rid of the nand_ecclayout limitations.
> struct nand_ecclayout is defining fixed eccpos and oobfree arrays which
> can only be increased by modifying the MTD_MAX_ECCPOS_ENTRIES_LARGE and
> MTD_MAX_OOBFREE_ENTRIES_LARGE macros.
> This approach forces us to modify the macro values each time we add a
> new NAND chip with a bigger OOB area, and increasing these arrays also
> penalize all platforms, even those who only support small NAND devices
> (with small OOB area).
>
> The idea to overcome this limitation, is to define the ECC/OOB layout
> by the mean of two functions: ->eccpos() and ->oobfree(), which will
> basically return the same information has those stored in the
> nand_ecclayout struct.
>
> Another advantage of this solution is that ECC layouts are usually
> following a repetitive pattern (i.e. leave X bytes free and put Y bytes
> of ECC per ECC chunk), which allows one to implement the ->eccpos()
> and ->oobfree() functions with a simple logic that can be applied
> to any size of OOB.

Thanks for the work! This definitely needed done. I imagined that it
might be best if we just changed the data structure format and have
drivers allocate it dynamically during probe(), but actually, I kinda
like generating it on the fly. The only concern I'd have is if there is
significant penalty to doing this sort of computation on the fly during
(e.g.) AUTO-layout OOB reads/writes. But I guess if there is such a
penalty, nothing would stop us from caching the results in the MTD/NAND
core code.

> Patches 1 to 10 are just cleanups or trivial fixes that can be taken
> independently.

There were some comments for patch 1, and I want to look more closely at
patch 10. But patches 2 to 9 are pushed to l2-mtd.git. Thanks!

> Patch 19 is just an aggregate of several smaller commits (one per
> driver), and has been submitted this way to limit the size of the
> series. If everybody agrees on this approach, I'll resubmit the series
> will those changes separated in different commits (as done here [1]).
>
> Also note that the last two commits are removing the nand_ecclayout
> definition, thus preventing any new driver to use this structure.
> Of course, this step can be delayed if some of the previous patches
> are not accepted.

I haven't looked in detail at the second half of the series, but I like
the concept. I'll look closer once you fix up things in v2.

Thanks,
Brian

> Best Regards,
>
> Boris
>
> [1]https://github.com/bbrezillon/linux-sunxi/commits/nand/ecclayout2