Re: [PATCH v4 04/23] mtd: nand: denali: avoid hard-coding ECC step, strength, bytes

[Boris Brezillon <boris.brezillon@free-electrons.com>]

On Wed, 7 Jun 2017 16:21:15 +0900
Masahiro Yamada <yamada.masahiro@socionext.com> wrote:

> Hi Boris,
> 
> 
> 2017-06-07 16:02 GMT+09:00 Boris Brezillon <boris.brezillon@free-electrons.com>:
> > On Wed, 7 Jun 2017 12:09:31 +0900
> > Masahiro Yamada <yamada.masahiro@socionext.com> wrote:
> >  
> >> >> +
> >> >> +static int denali_ecc_setup(struct mtd_info *mtd, struct nand_chip *chip,
> >> >> +                         struct denali_nand_info *denali)
> >> >> +{
> >> >> +     struct nand_ecc_caps caps;
> >> >> +     int ret;
> >> >> +
> >> >> +     caps.stepinfos = denali->stepinfo;
> >> >> +     caps.nstepinfos = 1;
> >> >> +     caps.calc_ecc_bytes = denali_calc_ecc_bytes;
> >> >> +     caps.oob_reserve_bytes = denali->bbtskipbytes;  
> >> >
> >> > If you get rid of this oob_reserve_bytes field, you can define caps as
> >> > a static const and even directly store ecc_caps in denali_nand_info.  
> >>
> >> To make caps static const, denali_calc_ecc_bytes must be exported
> >> to be referenced from denali_dt/denali_pci.
> >> I am reluctant to do it.  
> >
> > You already duplicate other information in denali_dt.c and
> > denali_pci.c,  
> 
> The ECC step-size and strength are tightly associated to each IP variant.
> I see duplication between denali_dt and denali_pci, but it is just because
> Intel and Altera happened to have the same parameters.

It's still duplication.

> 
> On the other hand, denali_calc_ecc_bytes() is common to all variants
> because ECC algorithm is not customizable.

Yes, I agree.

> 
> 
> > so what prevents you from duplicating this one-line
> > function?
> >
> > Also, denali core already exports 2 functions,  
> 
> They are entries for probe/remove.
> 
> > I don't see the problem
> > in exporting the common nand_ecc_caps object. Why are you reluctant to
> > that?  
> 
> denali_calc_ecc_bytes() is independent of DT, PCI, or whatever.
> I see less reason to expose it.

I don't get that one. The fact that it's a generic implementation makes
it a good match for something you want to have in the core and expose
to DT/PCI implems.

> 
> caps is only used on probing, so I used a local variable.
> I do not think it is a big problem.
> 

It is to me, because you'll be the only user of the API at first, and
people tend to copy&paste code from other drivers.
nand_ecc_caps is really something that should be const and attached to
a specific IP revision.