Re: [GIT PULL] mtd: Changes for v5.13-rc4

From: Miquel Raynal
Date: Tue Sep 28 2021 - 18:24:48 EST


Hi Linus,

miquel.raynal@xxxxxxxxxxx wrote on Wed, 26 May 2021 18:46:12 +0200:

> Hi Linus,
>
> Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote on Wed, 26 May
> 2021 06:20:35 -1000:
>
> > On Wed, May 26, 2021 at 5:59 AM Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote:
> > >
> > > Raw NAND:
> > > * txx9ndfmc, tmio, sharpsl, ndfc, lpc32xx_slc, fsmc, cs553x:
> > > - Fix external use of SW Hamming ECC helper
> >
> > Why are these guys all pointlessly duplicating the ecc wrapper
> > functions for their ecc 'correct' functions?
> >
> > The whole "the Hamming software ECC engine has been updated to become
> > a proper and independent ECC engine" excuse makes no sense. If
> > multiple chips just want a basic sw hamming helper, then they should
> > have one. Not have to be forced to each write their own pointless
> > wrapper like this.
> >
> > These chip drivers just want 'ecc_sw_hamming_correct()' with the
> > proper arguments, and it seems entirely wrong to duplicate the helper
> > five times or whatever. There should just be a generic helper - the
> > way there used to be.
> >
> > In fact, I would generally strongly recommend that if there used to be
> > a generic helper that different chip drivers used (ie the old
> > rawnand_sw_hamming_correct()), then such a helper should be left alone
> > and not change the semantics of it.
>
> I am not happy neither with the fix (which I wrote myself) as my first
> goal was to uniformize the way the Hamming helpers are being called (as
> part of a much bigger work). I assumed that all drivers either used the
> Hamming software engine or simply didn't, without thinking about the
> "intermediate" situations where a particular driver would just want to
> call a particular Hamming helper to workaround its "missing" hardware
> capabilities.
>
> Unfortunately when I spotted that many drivers were broken by my rework
> I decided to provide per-driver fixes, while, as you suggest, I should
> probably have declared a generic 'hamming correct' core helper and use
> that directly instead of duplicating the logic in each broken driver.
>
> > The new "proper independent ECC engine" that had new semantics should
> > have been the one that got a new name, rather than breaking an old and
> > existing helper function and then making the chip drivers pointlessly
> > write their own new helper functions.
> >
> > I've pulled this, but under protest. The patch honestly just looks
> > like mindless duplication.

Just to let you know that I proposed there [1] a series to clean this
up.

[1] https://lore.kernel.org/linux-mtd/20210928221507.199198-1-miquel.raynal@xxxxxxxxxxx/T/#t

Thanks,
Miquèl