Re: [PATCH v2 1/3] mtd: nand: Cleanup/rework the atmel_nand driver

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

On Tue, 21 Feb 2017 13:02:21 +0200
Andy Shevchenko <andy.shevchenko@gmail.com> wrote:

> On Tue, Feb 21, 2017 at 12:26 PM, Boris Brezillon
> <boris.brezillon@free-electrons.com> wrote:
> > On Tue, 21 Feb 2017 12:03:45 +0200
> > Andy Shevchenko <andy.shevchenko@gmail.com> wrote:  
> 
> >> 1. For example,
> >>
> >> #define ATMEL_NFC_CMD(pos, cmd)                        ((cmd) <<
> >> (((pos) * 8) + 2))  
> >
> > Well, I like to explicitly put parenthesis even when operator
> > precedence guarantees the order of the calculation ('*' is preceding
> > '+').  
> 
> That's my point. I'm not a LISP programmer.
> Personally I think it makes readability worse.

So, it's a matter of taste.
> >>
> >> 4. First of all, why do you need this function in the first place?
> >>
> >> +struct gpio_desc *
> >> +atmel_nand_pdata_get_gpio(struct atmel_nand_controller *nc, int gpioid,
> >> +                         const char *name, bool active_low,
> >> +                         enum gpiod_flags flags)  
> >
> > Because I don't want to duplicate the code done in
> > atmel_nand_pdata_get_gpio() each time I have to convert a GPIO number
> > into a GPIO descriptor, and that is needed to support platforms that
> > haven't moved to DT yet  
> 
> They should use GPIO lookup tables.
> 
> We don't encourage people to use platform data anymore.
> 
> We have unified device properties for something like "timeout-us", we
> have look up tables when you need specifics like pwm, gpio, pinctrl,
> ...
> 
> Abusing platform data with pointers is also not welcome.
> 
> > (in this case, avr32).  
> 
> It's dead de facto.
> 
> When last time did you compile kernel for it? What was the version of kernel?
> Did it get successfully?
> 
> When are we going to remove avr32 support from kernel completely?

I'll let Nicolas answer that one.

> 
> >> 5. BIT() macro:  
> 
> > We could probably use BIT() in a few places.  
> 
> There are more places including data structures assignments.

Yes. These are minor changes. I'll try to fix them.
Note that I sometime prefer to keep (1 << X).

Example:

#define PMECC_CFG_READ_OP			(0 << 12)
#define PMECC_CFG_WRITE_OP			(1 << 12)

> 
> > Again, this has been copied from the old driver. I'll have a closer
> > look.  
> 
> Exactly. You overlooked due to enormous LOC in the one change. See my
> point below.
> 
> >> 7. Question to all that distribution or whatever functions, don't you
> >> have a common helper? Or each vendor requires different logic behind
> >> it?  
> >
> > What are you talking about? nand_chip hooks?  
> 
> That long arithmetic with some data.

Okay, so the code in pmecc.c. See, it's hard to follow a review when
you don't comment inline.

> 
> >> 8. Have you checked what kernel library provides?  
> >
> > I think so, but again, this is really vague, what kind of
> > open-coded functions do you think could be replaced with core libraries
> > helpers?  
> 
> I dunno, I'm asking you. Usually if I see a pattern I got a clue to
> check lib/ and similar places. From time to time I discover something
> new and interesting there.

If you're talking about the code in pmecc.c, yes, I already mentioned
in the header that it should be reworked to use some helpers from
lib/bch.c, but that's not the point of this series, and is left as
future improvements.

> 
> >> And I believe there are still issues like those. After, who is on
> >> topic, might even find some logical and other issues...
> >>
> >> P.S. TBH, so big change is unreviewable in meaningful time. To have a
> >> comprehensive review I, for example, spend ~1h/250LOC, and
> >> ~2.5h/1000LOC, I would estimate ~4h/2000LOC. Imagine one to spend one
> >> day for this. Any volunteer? Not me.  
> >
> > I'm not asking you to review the whole driver, but you started to
> > comment on the code without pointing clearly to the things you wanted
> > me to address.  
> 
> Yes, because my point is *split* this to be reviewable.
> 

And how do you do with new drivers? Do you ask people to split their
submissions in micro changes? I'm regularly reviewing drivers that are
several thousands LOC, and I don't ask people to split things just
because it's too long. When I ask them to split in different commits,
it's because they are doing several unrelated changes at once.

Note that I considered refactoring the existing driver in smaller
steps, but it's almost impossible, because the code is too messy and I
would end up with a huge series of patches that is not easier to review.