Re: [PATCH v4 1/1] mfd: Add I2C based System Configuaration (SYSCON) access

From: Lee Jones <lee.jones@linaro.org>
To: Michael Walle <michael@walle.cc>, Wolfram Sang <wsa@kernel.org>
Cc: devicetree@vger.kernel.org, arnd@arndb.de,
	gregkh@linuxfoundation.org, linus.walleij@linaro.org,
	linux-kernel@vger.kernel.org, robh+dt@kernel.org,
	bgolaszewski@baylibre.com, broonie@kernel.org,
	andriy.shevchenko@linux.intel.com,
	linux-arm-kernel@lists.infradead.org
Subject: Re: [PATCH v4 1/1] mfd: Add I2C based System Configuaration (SYSCON) access
Date: Thu, 2 Jul 2020 09:18:25 +0100	[thread overview]
Message-ID: <20200702081825.GR1179328@dell> (raw)
In-Reply-To: <c20e8009fadbdc032090a493074396c8@walle.cc>

On Thu, 02 Jul 2020, Michael Walle wrote:

> Am 2020-07-02 08:54, schrieb Lee Jones:
> > On Wed, 01 Jul 2020, Michael Walle wrote:
> > 
> > > Am 2020-07-01 09:04, schrieb Lee Jones:
> > > > On Wed, 01 Jul 2020, Michael Walle wrote:
> > > >
> > > > > Hi Lee,
> > > > >
> > > > > Am 2020-06-30 11:16, schrieb Michael Walle:
> > > > > > I'm just trying to use this for my sl28 driver. Some remarks, see below.
> > > > > >
> > > > > > Am 2020-06-22 09:51, schrieb Lee Jones:
> > > > > > > The existing SYSCON implementation only supports MMIO (memory mapped)
> > > > > > > accesses, facilitated by Regmap.  This extends support for registers
> > > > > > > held behind I2C busses.
> > > > > > >
> > > > > > > Signed-off-by: Lee Jones <lee.jones@linaro.org>
> > > > > > > ---
> > > > > > > Changelog:
> > > > > > >
> > > > > > > v3 => v4
> > > > > > >   - Add ability to provide a non-default Regmap configuration
> > > > > > >
> > > > > > > v2 => v3
> > > > > > >   - Change 'is CONFIG' present check to include loadable modules
> > > > > > >     - s/#ifdef CONFIG_MFD_SYSCON_I2C/#if
> > > > > > > IS_ENABLED(CONFIG_MFD_SYSCON_I2C)/
> > > > > > >
> > > > > > > v1 => v2
> > > > > > >   - Remove legacy references to OF
> > > > > > >   - Allow building as a module (fixes h8300 0-day issue)
> > > > > > >
> > > > > > > drivers/mfd/Kconfig            |   7 +++
> > > > > > >  drivers/mfd/Makefile           |   1 +
> > > > > > >  drivers/mfd/syscon-i2c.c       | 104
> > > > > > > +++++++++++++++++++++++++++++++++
> > > > > > >  include/linux/mfd/syscon-i2c.h |  36 ++++++++++++
> > > > > > >  4 files changed, 148 insertions(+)
> > > > > > >  create mode 100644 drivers/mfd/syscon-i2c.c
> > > > > > >  create mode 100644 include/linux/mfd/syscon-i2c.h
> > 
> > [...]
> > 
> > > > > > This way, (a) a driver doesn't have to use "#include <linux/i2c.h>" just
> > > > > > to call to_i2c_client() (or i2c_verify_client()) and (b) you won't do it
> > > > > > all over again in all sub drivers.
> > > > > >
> > > > > > So you could just do a
> > > > > >   regmap = syscon_i2c_to_regmap(pdev->dev.parent);
> > > > > >
> > > > > > I've also noticed that the mmio syscon uses device_node as parameter.
> > > > > > What
> > > > > > was the reason to divert from that? Just curious.
> > > > >
> > > > > How is this supposed to be used?
> > > > >
> > > > > I had something like the following in mind:
> > > > >
> > > > > &i2c {
> > > > >   cpld@4a {
> > > > >     compatible = "simple-mfd";
> > > > >     reg = <0x4a>;
> > > > >
> > > > >     gpio@4 {
> > > > >       compatible = "vendor,gpio";
> > > > >       reg = <0x4>;
> > > > >     };
> > > > >   };
> > > > > };
> > > >
> > > > Yes, that was the idea.
> > > >
> > > > > But I think the childen are not enumerated if its an I2C device. And
> > > > > the actual i2c driver is also missing.
> > > >
> > > > What do you mean?  Can you elaborate?
> > > 
> > > There is no i2c_driver instance who would create the regmap.
> > 
> > The regmap is created by the first caller of:
> > 
> >  syscon_i2c_to_regmap{_config}()
> 
> But which one is an i2c_driver? All the sub devices are platform drivers
> and there should be no need for them to know that they are behind an
> i2c driver (or spi driver or just mmio). All they have to know is how
> to access the registers.
> 
> > > If I'm
> > > reading the I2C code correctly, it won't probe any i2c device of a
> > > bus if there is no i2c_driver with an associated .probe() or
> > > .probe_new().
> > 
> > Why wouldn't the children be registered using i2c_driver?
> 
> Where is the code which enumerates the children?

Yes, I see the problem now.  So I2C devices depend on a device
registering with the i2c_driver framework, which then probes the
device accordingly.  Thus a physical driver is required to convert I2C
devices to platform devices.  So this stops being an MFD problem and
starts being a 'simple-i2c' issue. :)

(not a genuine suggestion by the way)

Wolfram, do we have this correct?

-- 
Lee Jones [李琼斯]
Senior Technical Lead - Developer Services
Linaro.org │ Open source software for Arm SoCs
Follow Linaro: Facebook | Twitter | Blog

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel