Re: [PATCH v7 1/5] dt-bindings: Add the r9a06g032-sysctrl.h file
From: Geert Uytterhoeven
Date: Fri Jun 01 2018 - 04:18:54 EST
Hi Michel,
On Thu, May 31, 2018 at 12:01 PM, M P <buserror@xxxxxxxxx> wrote:
> On Thu, 31 May 2018 at 10:32, Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
>> On Thu, May 31, 2018 at 11:11 AM, M P <buserror@xxxxxxxxx> wrote:
>> > On Fri, 25 May 2018 at 11:31, Geert Uytterhoeven <geert@xxxxxxxxxxxxxx>
>> > wrote:
>> >> On Thu, May 24, 2018 at 11:28 AM, Michel Pollet
>> >> <michel.pollet@xxxxxxxxxxxxxx> wrote:
>> >> > This adds the constants necessary to use the renesas,r9a06g032-sysctrl
>> > node.
>>
>> >> > @@ -0,0 +1,187 @@
>> >> > +/* SPDX-License-Identifier: GPL-2.0 */
>> >> > +/*
>> >> > + * R9A06G032 sysctrl IDs
>> >> > + *
>> >> > + * Copyright (C) 2018 Renesas Electronics Europe Limited
>> >> > + *
>> >> > + * Michel Pollet <michel.pollet@xxxxxxxxxxxxxx>, <buserror@xxxxxxxxx>
>> >> > + * Derived from zx-reboot.c
>> >> > + */
>> >> > +
>> >> > +#ifndef __DT_BINDINGS_R9A06G032_SYSCTRL_H__
>> >> > +#define __DT_BINDINGS_R9A06G032_SYSCTRL_H__
>> >> > +
>> >> > +#define R9A06G032_CLKOUT 0
>> >> > +#define R9A06G032_CLK_PLL_USB 1
>> >> > +#define R9A06G032_CLK_48 1 /* AKA CLK_PLL_USB */
>> >> > +#define R9A06G032_CLKOUT_D10 2
>> >> > +#define R9A06G032_CLKOUT_D16 3
>> >> > +#define R9A06G032_MSEBIS_CLK 3 /* AKA CLKOUT_D16 */
>> >> > +#define R9A06G032_MSEBIM_CLK 3 /* AKA CLKOUT_D16 */
>> >> > +#define R9A06G032_CLKOUT_D160 4
>> >> > +#define R9A06G032_CLKOUT_D1OR2 5
>> >> > +#define R9A06G032_CLK_DDRPHY_PLLCLK 5 /* AKA CLKOUT_D1OR2 */
>> >
>> >> [...]
>> >
>> >> I have 3 comments:
>> >
>> >> 1. I had expected this list to match (both name- and order-wise)
>> > Appendix
>> >> C ("Clock Tree Structure") in the RZ/N1[DSL] Userâs Manual: System
>> >> Introduction, Multiplexing, Electrical and Mechanical Information.
>> >> That would make it easier to review.
>> >
>> > Well, that document was made a *long* time after the internal documentation
>> > we used to generate the clock lists. There are a few things we had to do:
>> >
>> > * Renumber peripherals. We decided a long time ago that u-boot and linux
>> > would stick to zero based peripherals, and not one based numbers. It's next
>> > to impossible to keep mixing number based across software base, so we use
>> > UART0 while the hardware manual mentions UART1 -- It *is* documented
>> > extensively with out SDK, and makes life using linux a lot easier. It's
>> > across all our SDK, u-boot, webapps readmes, howtos etc.
>> >
>> > * Rename some peripherals. Plenty of peripherals names made little sense
>> > and had zero consistency, in fact, many names were different depending on
>> > the place they were used! -- "flashnand"+"nand_flash"+"FNAND" etc,
>> > "QUADSPI"+"QSPI" etc etc so we also re-normalized the names to match linux
>> > conventions.
>> >
>> > * Other internal reasons I can't document here
>> >
>> > Also, the value here are made up anyway -- so I've decided to sort them to
>> > make sure any clock that has a parent is numbered *after* the parent...
>>
>> Well, all of that combines makes it very hard for us to review the list.
>
> I understand -- the previous format was a lot more readable, here I had to
> work to make the table as small as I could, and quite a few bits of readability
> were lost in the process. It's already a huge file.
>
> I'm not sure if that would help, but here is the link to the old table
> format I used
> https://github.com/renesas-rz/rzn1_linux/blob/rzn1-stable/drivers/clk/rzn1/rzn1-clkctrl-tables.h
>
> I had to throw that away to make up the new composite table that now contains
> what was in the device tree file before. However the names are the same.
>
> Perhaps I could change how I encode the register/bit pairs to use 8+8 bits to
> make the hex constants more readable?
Alternatively, you can use a macro to pack them, cfr. MOD_CLK_PACK()
in renesas-cpg-mssr.
>> >> 2. These definitions seems to be a mix of:
>> >> 1. Internal core clocks,
>> >> 2. Other core clocks,
>> >> 3. Module clocks.
>> >
>> >> The internal clocks do not need to be referenced from DT, so I would
>> >> not make them part of the DT bindings.
>> >
>> > Why? I'm told that "Bindings aren't for linux" -- why can't I imagine
>> > 'something' needing them? Why would I decide NOT to include them,
>> > as they are there? I 'factored' a few of them to the same number when
>>
>> Just general safety w.r.t. unchangeable DT definitions: anything that is
>> not listed here cannot be declared wrong later.
>
> Well, I need these #define *somewhere* as I use them in my driver. So what
> do you want me to do?
> + Remove the header, move the constants into the driver?
> + Use hard coded numbers in the DT and remove the header entirely?
> + Duplicate the header, keep the full one with the driver, and only list
> the CA7+UART0 one in the dt-binding and duplicate the ones I need as
> I go along?
None of these. You can just move the #defines that will never be needed
from DT into the driver source file.
Cfr. "Internal Core Clocks" in drivers/clk/renesas/r8a7795-cpg-mssr.c.
>> > applicable.
>>
>> You're 100% sure they're the same?
>
> Yes, the script logic hasn't changed in 2 years, and we've been using that
> hierarchy extensively since.
> Basically the logic is that if a clock doesn't have a gate and isn't a divisor
> of some sort, it's factored with it's parent clock... it used to be
> done with a DT
> alias of the same name.
OK.
>> > This is all done automatically BTW -- a script takes the original clocking
>> > spreadsheet, and converts it into a table, correcting 'human input' as it
>> > goes along.
>>
>> So the internal spreadsheet doesn't match the public documentation neither?
>
> Nope, as usual, people who wrote the documentation went their own way at
> some point, and didn't backport any changes they made.
Well, I guess we'll have to live with that...
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds