Re: [PATCH] gpio: raspberrypi-ext: fix firmware dependency

[Baruch Siach <baruch@xxxxxxxxxx>]

Hi Arnd,

On Wed, Feb 28, 2018 at 11:08:54PM +0100, Arnd Bergmann wrote:
> On Wed, Feb 28, 2018 at 10:47 PM, Baruch Siach <baruch@xxxxxxxxxx> wrote:
> > Thanks for the fix.
> >
> > On Wed, Feb 28, 2018 at 02:48:08PM +0100, Arnd Bergmann wrote:
> >> When the firmware driver is a loadable module, the gpio driver cannot be
> >> built-in:
> >>
> >> drivers/gpio/gpio-raspberrypi-exp.o: In function `rpi_exp_gpio_set':
> >> gpio-raspberrypi-exp.c:(.text+0xb4): undefined reference to `rpi_firmware_property'
> >> drivers/gpio/gpio-raspberrypi-exp.o: In function `rpi_exp_gpio_get':
> >> gpio-raspberrypi-exp.c:(.text+0x1ec): undefined reference to `rpi_firmware_property'
> >> drivers/gpio/gpio-raspberrypi-exp.o: In function `rpi_exp_gpio_get_direction':
> >> gpio-raspberrypi-exp.c:(.text+0x360): undefined reference to `rpi_firmware_property'
> >> drivers/gpio/gpio-raspberrypi-exp.o: In function `rpi_exp_gpio_get_polarity':
> >> gpio-raspberrypi-exp.c:(.text+0x4d4): undefined reference to `rpi_firmware_property'
> >> drivers/gpio/gpio-raspberrypi-exp.o: In function `rpi_exp_gpio_dir_out':
> >> gpio-raspberrypi-exp.c:(.text+0x670): undefined reference to `rpi_firmware_property'
> >> drivers/gpio/gpio-raspberrypi-exp.o:gpio-raspberrypi-exp.c:(.text+0x7fc): more undefined references to `rpi_firmware_property' follow
> >> drivers/gpio/gpio-raspberrypi-exp.o: In function `rpi_exp_gpio_dir_in':
> >> drivers/gpio/gpio-raspberrypi-exp.o: In function `rpi_exp_gpio_probe':
> >> gpio-raspberrypi-exp.c:(.text+0x93c): undefined reference to `rpi_firmware_get'
> >>
> >> We already have a Kconfig dependency for it, but when compile-testing, it
> >> is disregarded.
> >>
> >> This changes the dependency so that compile-testing is only done when the
> >> firmware driver is completely disabled.
> >
> > What about the CONFIG_ARCH_BCM2835=y case? The combination of
> > CONFIG_GPIO_RASPBERRYPI_EXP=y and CONFIG_RASPBERRYPI_FIRMWARE=m is still
> > valid. Wouldn't that break the build?
> >
> > Isn't there a way in Kconfig to force CONFIG_GPIO_RASPBERRYPI_EXP=m when
> > CONFIG_RASPBERRYPI_FIRMWARE=m?
> 
> The problem I ran into only happens with CONFIG_ARCH_BCM2835=y to
> start with. My fix handles that case correctly, it forces
> CONFIG_GPIO_RASPBERRYPI_EXP to be either 'n' or 'm'
> when CONFIG_RASPBERRYPI_FIRMWARE=m.

Thanks for the explanation.

> > What about 'depends on m || RASPBERRYPI_FIRMWARE=y'?
> 
> That would be (slightly) wrong, it would force CONFIG_GPIO_RASPBERRYPI_EXP
> to be 'm' even if RASPBERRYPI_FIRMWARE=n.
> 
> > Grepping around I also found this:
> >
> > drivers/power/supply/Kconfig:   depends on USB_GADGET || !USB_GADGET # if USB_GADGET=m, this can't be 'y'
> 
> That is what I did here as well, except the !RASPBERRYPI_FIRMWARE
> only applies for COMPILE_TEST.
> 
> > And this:
> >
> > drivers/infiniband/Kconfig:     depends on m || IPV6 != m
> 
> This is a less common way to express it. The idiomatic
> Kconfig expression here would be 'depends on IPV6 || !IPV6'.

I find this way much easier to understand. Without the comment I would never 
have guessed what 'USB_GADGET || !USB_GADGET' actually means. And indeed no 
comment is needed in the infiniband case.

> >> Fixes: a98d90e7d588 ("gpio: raspberrypi-exp: Driver for RPi3 GPIO expander via mailbox service")
> >> Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
> >> ---
> >>  drivers/gpio/Kconfig | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/gpio/Kconfig b/drivers/gpio/Kconfig
> >> index 2ecd2adbaec6..52a8b0a6f4e1 100644
> >> --- a/drivers/gpio/Kconfig
> >> +++ b/drivers/gpio/Kconfig
> >> @@ -126,7 +126,7 @@ config GPIO_RASPBERRYPI_EXP
> >>       tristate "Raspberry Pi 3 GPIO Expander"
> >>       default RASPBERRYPI_FIRMWARE
> >>       depends on OF_GPIO
> >> -     depends on (ARCH_BCM2835 && RASPBERRYPI_FIRMWARE) || COMPILE_TEST
> >> +     depends on (ARCH_BCM2835 && RASPBERRYPI_FIRMWARE) || (COMPILE_TEST && !RASPBERRYPI_FIRMWARE)
> >
> > This is really non-obvious. An inline comment here might help, IMO.
> 
> How about
> 
>        # RASPBERRYPI_FIRMWARE is only available for ARCH_BCM2835, but we want to
>        # allow compile-testing when it is disabled
> 
> ?

The module vs built-in aspect is missing. That is the non-obvious part.

baruch

-- 
     http://baruch.siach.name/blog/                  ~. .~   Tk Open Systems
=}------------------------------------------------ooO--U--Ooo------------{=
   - baruch@xxxxxxxxxx - tel: +972.52.368.4656, http://www.tkos.co.il -
--
To unsubscribe from this list: send the line "unsubscribe linux-gpio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html