Message-ID: <ZAi7CPXX0z80mKfQ@linux-8mug>
Date:   Thu, 9 Mar 2023 00:42:48 +0800
From:   Chester Lin <clin@...e.com>
To:     Andy Shevchenko <andy.shevchenko@...il.com>
Cc:     Linus Walleij <linus.walleij@...aro.org>,
        Andreas Färber <afaerber@...e.de>,
        s32@....com, linux-gpio@...r.kernel.org,
        devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
        linux-arm-kernel@...ts.infradead.org,
        Larisa Grigore <larisa.grigore@....com>,
        Ghennadi Procopciuc <Ghennadi.Procopciuc@....nxp.com>,
        Andrei Stefanescu <andrei.stefanescu@....com>,
        Radu Pirea <radu-nicolae.pirea@....com>,
        Matthias Brugger <mbrugger@...e.com>,
        Matthew Nunez <matthew.nunez@....com>,
        Phu Luu An <phu.luuan@....com>,
        Stefan-Gabriel Mirea <stefan-gabriel.mirea@....com>
Subject: Re: [PATCH v5 2/3] pinctrl: add NXP S32 SoC family support

Hi Andy,

On Wed, Mar 08, 2023 at 03:21:00PM +0200, Andy Shevchenko wrote:
> On Wed, Mar 8, 2023 at 7:03 AM Chester Lin <clin@...e.com> wrote:
> > On Tue, Mar 07, 2023 at 01:28:09AM +0200, andy.shevchenko@...il.com wrote:
> > > Mon, Feb 20, 2023 at 10:33:19AM +0800, Chester Lin kirjoitti:
> 
> ...
> 
> > but the driver patch
> > has been merged into the maintainer's for-next so I would not change this part
> > unless the driver patch needs to be reverted and re-submitted in the end.
> 
> As I said you have to keep it in mind for all your future
> contributions to the Linux kernel independently on the destiny of this
> one.
> 
> ...
> 
> > > > +   depends on ARCH_S32 && OF
> > >
> > > Is OF necessary? Can it be
> >
> > I think it's required since the driver file refers to of_* APIs.
> 
> And? Is it functional or compilation dependency? If the latter is the
> case, what API exactly isn't providing a stub?

I was wrong. Looks like the ARM64 arch Kconfig always select OF so it's not
really necessary to have OF here.

> 
> > >       depends OF || COMPILE_TEST
> > >
> > > ?
> 
> So?
> 

Since the OF dependency is not really necessary here, to fulfill the compile test
purpose, the possible dependency might be (ARCH_S32 || COMPILE_TEST), but it
could meet a compiling failure on the reference of pinconf_generic_parse_dt_config()
for those architectures which do not select OF by default since there's no stub
for this function. [pinconf_generic_parse_dt_config() is called in pinctrl-s32cc.c]

> ...
> 
> > > > +   depends on ARCH_S32 && OF
> 
> Ditto.
> 

Based on the previous assumption [OF is not needed and PINCTRL_S32CC doesn't
depend on COMPILE_TEST], selecting PINCTRL_S32G2 wouldn't work if it simply
depends on (ARCH_S32 || COMPILE_TEST), for example:

  WARNING: unmet direct dependencies detected for PINCTRL_S32CC
    Depends on [n]: PINCTRL [=y] && ARCH_S32
      Selected by [y]:
        - PINCTRL_S32G2 [=y] && PINCTRL [=y] && (ARCH_S32 || COMPILE_TEST [=y])

So the better solutions is to still have OF in PINCTRL_S32CC, such as:

config PINCTRL_S32CC
	bool
	depends on ARCH_S32 || (OF && COMPILE_TEST)
	.....

config PINCTRL_S32G2
	depends on ARCH_S32 || COMPILE_TEST
	.....

Regards,
Chester

> > > > +/**
> > > > + * struct s32_pin_group - describes an S32 pin group
> > > > + * @name: the name of this specific pin group
> > > > + * @npins: the number of pins in this group array, i.e. the number of
> > > > + *         elements in pin_ids and pin_sss so we can iterate over that array
> > > > + * @pin_ids: an array of pin IDs in this group
> > > > + * @pin_sss: an array of source signal select configs paired with pin_ids
> > > > + */
> > > > +struct s32_pin_group {
> > > > +   const char *name;
> > > > +   unsigned int npins;
> > > > +   unsigned int *pin_ids;
> > > > +   unsigned int *pin_sss;
> > >
> > > Why didn't you embed struct pingroup?
> >
> > I did think about that but there's an additional 'pin_sss' which could make code
> > a bit messy. For example:
> >
> >         s32_regmap_update(pctldev, grp->grp.pins[i],
> >                           S32_MSCR_SSS_MASK, grp->pin_sss[i]);
> 
> We specifically provide those data types to separate generic things
> with custom ones. I don't think about the code getting longer, the
> access to the proper data seems reasonable to me. Look into other
> drivers that utilise these data types.
> 

Will change it, thanks for your suggestions.

> > > > +};
> > > > +
> > > > +/**
> > > > + * struct s32_pmx_func - describes S32 pinmux functions
> > > > + * @name: the name of this specific function
> > > > + * @groups: corresponding pin groups
> > > > + * @num_groups: the number of groups
> > > > + */
> > > > +struct s32_pmx_func {
> > > > +   const char *name;
> > > > +   const char **groups;
> > > > +   unsigned int num_groups;
> > > > +};
> > >
> > > struct pinfunction.
> >
> > Thanks for your information. I was not aware of this new struct since it just got
> > merged recently.
> 
> That's why the rule is to keep an eye on the subsystem development by
> regular rebasing on top of its tip (pinctrl tree, devel branch in this
> case).
> 
> ...
> 
> > > > +#ifdef CONFIG_PM_SLEEP
> > > > +int __maybe_unused s32_pinctrl_resume(struct device *dev);
> > > > +int __maybe_unused s32_pinctrl_suspend(struct device *dev);
> > > > +#endif
> > >
> > > Please, consider using new PM macros, like pm_ptr().
> >
> > Maybe pm_sleep_ptr() is more accurate?
> 
> You are the author, choose what you think fits the best!
> 
> ...
> 
> 
> > > > +   case PIN_CONFIG_BIAS_PULL_UP:
> > > > +           if (arg)
> > > > +                   *config |= S32_MSCR_PUS;
> > > > +           else
> > > > +                   *config &= ~S32_MSCR_PUS;
> > >
> > > > +           fallthrough;
> > >
> > > It's quite easy to miss this and tell us about how is it supposed to work with PU + PD?
> > >
> > I admit that it's ambiguous and should be improved in order to have better readability.
> >
> > In a S32G2 MSCR register, there are two register bits related to pull up/down functions:
> >
> > PUE (Pull Enable, MSCR[13]): 0'b: Disabled,  1'b: Enabled
> > PUS (Pull Select, MSCR[12]): 0'b: Pull Down, 1'b: Pull Up
> >
> > The dt properties could be like these:
> >
> > 1) 'bias-pull-up' or 'bias-pull-up: true'  --> arg = 1
> >    In this case both PUE and PUS are set.
> >
> > 2) 'bias-pull-up: false'  --> arg = 0
> >    In this case both PUE and PUS are cleared since the pull-up function must be disabled.
> 
> So, split it to a separate function where you do the enabling only once.
> I can point to drivers/pinctrl/intel/pinctrl-intel.c for the idea to take from.
> 

Will do.

> > > > +   case PIN_CONFIG_BIAS_PULL_DOWN:
> > > > +           if (arg)
> > > > +                   *config |= S32_MSCR_PUE;
> > > > +           else
> > > > +                   *config &= ~S32_MSCR_PUE;
> > > > +           *mask |= S32_MSCR_PUE | S32_MSCR_PUS;
> > > > +           break;
> > > > +   case PIN_CONFIG_BIAS_HIGH_IMPEDANCE:
> > > > +           *config &= ~(S32_MSCR_ODE | S32_MSCR_OBE | S32_MSCR_IBE);
> > > > +           *mask |= S32_MSCR_ODE | S32_MSCR_OBE | S32_MSCR_IBE;
> > > > +           fallthrough;
> > >
> > > Ditto.
> > >
> >
> > It's similar to the case 'PIN_CONFIG_BIAS_PULL_UP' although the PUS bit is assumed
> > as 0 via the config variable so only the PUE bit needs to be configured, for example:
> >
> > 1) 'bias-pull-down' or 'bias-pull-down: true'  --> arg = 1
> >    PUE is set and PUS is cleared.
> >
> > 2) 'bias-pull-down: false'  --> arg = 0
> >    In this case both PUE and PUS are cleared since the pull-down function must be disabled.
> >
> > > > +   case PIN_CONFIG_BIAS_DISABLE:
> > > > +           *config &= ~(S32_MSCR_PUS | S32_MSCR_PUE);
> > > > +           *mask |= S32_MSCR_PUS | S32_MSCR_PUE;
> > > > +           break;
> 
> Ditto.
> 
> ...
> 
> I assume that non-commented is equal to silent agreement and will be
> addressed accordingly. If it's not the case, reply again with your
> objections.
> 
> -- 
> With Best Regards,
> Andy Shevchenko

Powered by blists - more mailing lists