Re: [PATCH] media: i2c: fix max9271 build dependencies

[Jacopo Mondi <jacopo@jmondi.org>]

On Mon, Feb 08, 2021 at 01:32:38PM +0200, Laurent Pinchart wrote:
> Hi Mauro,
>
> On Mon, Feb 08, 2021 at 11:07:23AM +0100, Mauro Carvalho Chehab wrote:
> > Em Mon, 8 Feb 2021 11:24:25 +0200 Sakari Ailus escreveu:
> > > On Mon, Feb 08, 2021 at 10:08:22AM +0100, Mauro Carvalho Chehab wrote:
> > > > Em Mon, 8 Feb 2021 10:41:47 +0200 Sakari Ailus escreveu:
> > > > > On Mon, Feb 08, 2021 at 09:36:16AM +0100, Jacopo Mondi wrote:
> > > > > > On Mon, Feb 08, 2021 at 09:27:01AM +0200, Sakari Ailus wrote:
> > > > > > > Hi Mauro,
> > > > > > >
> > > > > > > Thanks for the patch.
> > > > > >
> > > > > > Sorry, that's cleary an oversight from my side.
> > > > > > Thanks for tackling it.
> > > > > >
> > > > > > > On Mon, Feb 08, 2021 at 07:53:15AM +0100, Mauro Carvalho Chehab wrote:
> > > > > > > > As described on its c file, the Maxim MAX9271 GMSL serializer isn't a
> > > > > > > > self-contained driver, as MAX9271 is usually embedded in camera modules
> > > > > > > > with at least one image sensor and optional additional components,
> > > > > > > > such as uController units or ISPs/DSPs.
> > > > > > > >
> > > > > > > > After chanseset a59f853b3b4b ("media: i2c: Add driver for RDACM21 camera module"),
> > > > > > > > there are now two drivers currently needing it: rdacm20 and rdacm21.
> > > > > > > >
> > > > > > > > Building with allmodconfig is now causing those warnings:
> > > > > > > >
> > > > > > > > 	WARNING: modpost: drivers/media/i2c/rdacm21-camera_module: 'max9271_set_serial_link' exported twice. Previous export was in drivers/media/i2c/rdacm20-camera_module.ko
> > > > > > > > 	WARNING: modpost: drivers/media/i2c/rdacm21-camera_module: 'max9271_configure_i2c' exported twice. Previous export was in drivers/media/i2c/rdacm20-camera_module.ko
> > > > > > > > 	WARNING: modpost: drivers/media/i2c/rdacm21-camera_module: 'max9271_set_high_threshold' exported twice. Previous export was in drivers/media/i2c/rdacm20-camera_module.ko
> > > > > > > > 	WARNING: modpost: drivers/media/i2c/rdacm21-camera_module: 'max9271_configure_gmsl_link' exported twice. Previous export was in drivers/media/i2c/rdacm20-camera_module.ko
> > > > > > > > 	WARNING: modpost: drivers/media/i2c/rdacm21-camera_module: 'max9271_set_gpios' exported twice. Previous export was in drivers/media/i2c/rdacm20-camera_module.ko
> > > > > > > > 	WARNING: modpost: drivers/media/i2c/rdacm21-camera_module: 'max9271_clear_gpios' exported twice. Previous export was in drivers/media/i2c/rdacm20-camera_module.ko
> > > > > > > > 	WARNING: modpost: drivers/media/i2c/rdacm21-camera_module: 'max9271_enable_gpios' exported twice. Previous export was in drivers/media/i2c/rdacm20-camera_module.ko
> > > > > > > > 	WARNING: modpost: drivers/media/i2c/rdacm21-camera_module: 'max9271_disable_gpios' exported twice. Previous export was in drivers/media/i2c/rdacm20-camera_module.ko
> > > > > > > > 	WARNING: modpost: drivers/media/i2c/rdacm21-camera_module: 'max9271_verify_id' exported twice. Previous export was in drivers/media/i2c/rdacm20-camera_module.ko
> > > > > > > > 	WARNING: modpost: drivers/media/i2c/rdacm21-camera_module: 'max9271_set_address' exported twice. Previous export was in drivers/media/i2c/rdacm20-camera_module.ko
> > > > > > > > 	WARNING: modpost: drivers/media/i2c/rdacm21-camera_module: 'max9271_set_deserializer_address' exported twice. Previous export was in drivers/media/i2c/rdacm20-camera_module.ko
> > > > > > > > 	WARNING: modpost: drivers/media/i2c/rdacm21-camera_module: 'max9271_set_translation' exported twice. Previous export was in drivers/media/i2c/rdacm20-camera_module.ko
> > > > > > > >
> > > > > > > > Address the issue by adding a Kconfig item for it, that it is
> > > > > > > > seleced if either one of the modules that need max9271 is used.
> > > > > > > >
> > > > > > > > Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
> > > > > > > > Fixes: a59f853b3b4b ("media: i2c: Add driver for RDACM21 camera module")
> > > > > > > > Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
> > > > > > > > ---
> > > > > > > >  drivers/media/i2c/Kconfig  | 10 ++++++++++
> > > > > > > >  drivers/media/i2c/Makefile |  8 ++++----
> > > > > > > >  2 files changed, 14 insertions(+), 4 deletions(-)
> > > > > > > >
> > > > > > > > diff --git a/drivers/media/i2c/Kconfig b/drivers/media/i2c/Kconfig
> > > > > > > > index 2d3dc0d82f9e..a6802195d583 100644
> > > > > > > > --- a/drivers/media/i2c/Kconfig
> > > > > > > > +++ b/drivers/media/i2c/Kconfig
> > > > > > > > @@ -712,6 +712,16 @@ config VIDEO_ST_MIPID02
> > > > > > > >  	  module will be called st-mipid02.
> > > > > > > >  endmenu
> > > > > > > >
> > > > > > > > +#
> > > > > > > > +# Camera ancillary chips
> > > > > > > > +#
> > > > > > > > +
> > > > > > > > +# MAX9271 is usually embedded in camera modules
> > > > > > > > +config VIDEO_MAX9271_SERIALIZER
> > > > > > > > +	tristate
> > > > > > > > +	default y
> > > > > > > > +	depends on VIDEO_RDACM20 || VIDEO_RDACM21
> > > > > > > > +
> > > > > > >
> > > > > > > I'd instead make the RDACM drivers depend on this one instead. The RDACM20
> > > > > > > driver directly depends on the symbols in the MAX9271 driver.
> > > >
> > > > While such solution is technically OK, it will make harder for the ones
> > > > wanting to use the RDACM drivers, because those two drivers will be
> > > > ridden at the config menu, until MAX9271 got selected.
> > > >
> > > > With the above solution, this driver will be auto-selected if either
> > > > RDACM20 or RDACM21 is needed.
> > > >
> > > > Btw, this is exactly the same strategy we use for tuner I2C modules:
> > > > the user doesn't need to know that his board uses tuner "foo" or "bar".
> > > > All it needs is to know that it will require, for instance, the em28xx
> > > > driver.
> > > >
> > > > > > >
> > > > > >
> > > > > > OTOH I it makes sense to have MAX9271 depend on the camera modules, as
> > > > > > selecting the serializer alone is not that useful.
> > > > > >
> > > > > > Could the two camera modules symbols instead select the MAX9271 one ?
> > > > >
> > > > > MAX9271 could be used elsewhere than in RDACM* devices.
> > > >
> > > > Sure, but then it is just a matter of adding the other driver there:
> > > >
> > > > 	depends on VIDEO_RDACM20 || VIDEO_RDACM21 || VIDEO_FOO
> > > >
> > > > To be frank, I doubt that we'll end having dozens of boards with this.
> > > > If we end having lots of drivers, we can work on a different
> > > > solution.
> > >
> > > Also note that there will be combinations of something compiled as a module
> > > and another driver linked directly to the kernel.
> > >
> > > So each of the RDACM drivers would likely also need:
> > >
> > > 	depends on VIDEO_MAX9271_SERIALIZER || !VIDEO_MAX9271_SERIALIZER
> >
> > No. I fail to see where this would work, as MAX9271 is a mandatory
> > dependency of those drivers. See:
> >
> > 	static int rdacm20_s_stream(struct v4l2_subdev *sd, int enable)
> > 	{
> > 	        struct rdacm20_device *dev = sd_to_rdacm20(sd);
> >
> > 	        return max9271_set_serial_link(dev->serializer, enable);
> > 	}
> >
> >
> > Without compiling max9271, the device won't stream ;-)
> >
> > -
> >
> > Yet, assuming that this could be an optional feature, if you use this:
> >
> > 	config VIDEO_MAX9271_SERIALIZER
> > 		tristate
> > 		default y
> > 		depends on VIDEO_RDACM20 || VIDEO_RDACM21
> >
> > You should *not* never do:
> >
> >  	depends on VIDEO_MAX9271_SERIALIZER || !VIDEO_MAX9271_SERIALIZER
> >
> > for VIDEO_RDACM20 or VIDEO_RDACM21, as it would cause a dependency
> > loop.
> >
> > -
> >
> > If you do, instead:
> >
> >     if VIDEO_V4L2 && I2C
> > 	config VIDEO_MAX9271_SERIALIZER
> > 		tristate
> >
> > 	config VIDEO_RDACM20
> > 		select VIDEO_MAX9271_SERIALIZER
> > 		...
> >
> > 	config VIDEO_RDACM21
> > 		select VIDEO_MAX9271_SERIALIZER
> > 		...
> >     endif
> >
> > Then you also won't need:
> > 	depends on VIDEO_MAX9271_SERIALIZER || !VIDEO_MAX9271_SERIALIZER
> >
> > As select should do the right thing in this case, ensuring that MAX9271
> > will be builtin either if RDACM20 or RDACM21 is builtin.
>
> I also vote for usage of "select".
>

I would prefer that too, I was concerned about possible un-met
dependencies, as Sakari pointed out, but the current situation is no
better, as the only Kconfig symbols where those can be listed are the
camera modules one.

> > -
> >
> > This should also work:
> >
> > 	config VIDEO_MAX9271_SERIALIZER
> > 		tristate
> > 		...
> >
> > 	if VIDEO_MAX9271_SERIALIZER
> >
> > 	config VIDEO_RDACM20
> > 		tristate
> > 		...
> >
> > 	config VIDEO_RDACM21
> > 		tristate
> > 		...
> >
> > 	endif
> >
> > without needing a "depend on FOO || !FOO" logic, but, as I said before, this
> > case will force the user that want to build the RDACM20/RDACM21 drivers to
> > be aware that a MAX9271 driver is required.
> >
> > -
> >
> > You would only need a "depend on FOO || !FOO" if the driver FOO is
> > optional and if you try do something like:
> >
> > 	config FOO
> > 		tristate
> > 		...
> >
> > 	config VIDEO_RDACM20
> > 		tristate
> > 		depends on FOO
> > 		...
> >
> > 	config VIDEO_RDACM21
> > 		tristate
> > 		depends on FOO
> > 		...
> >
> > as, on such case, it would allow the vars to be:
> >
> > 	FOO="m"
> > 	VIDEO_RDACM20="y"
> > 	VIDEO_RDACM21="m"
> >
> > which would cause build errors.
>
> --
> Regards,
>
> Laurent Pinchart