Dne 19.2.2015 v 15:53 Arnd Bergmann napsal(a):

On Thursday 19 February 2015 13:11:07 Michal Marek wrote:

quoted

On 2015-02-18 18:12, Arnd Bergmann wrote:

quoted

In the media drivers, the v4l2 core knows about all submodules
and calls into them from a common function. However this cannot
work if the modules that get called are loadable and the
core is built-in. In that case we get

drivers/built-in.o: In function `set_type':
drivers/media/v4l2-core/tuner-core.c:301: undefined reference to `tea5767_attach'
drivers/media/v4l2-core/tuner-core.c:307: undefined reference to `tea5761_attach'
drivers/media/v4l2-core/tuner-core.c:349: undefined reference to `tda9887_attach'
drivers/media/v4l2-core/tuner-core.c:405: undefined reference to `xc4000_attach'
[...]
Ideally Kconfig would be used to avoid the case of a broken dependency,
or the code restructured in a way to turn around the dependency, but either
way would require much larger changes here.

What can be done without extending kbuild is to accept
CONFIG_VIDEO_TUNER=y and CONFIG_MEDIA_TUNER_FOO=m, but build both into
the kernel, e.g.

Right, but

quoted

diff --git a/drivers/media/tuners/Kconfig b/drivers/media/tuners/Kconfig
index 42e5a01..d2c7e89 100644
--- a/drivers/media/tuners/Kconfig
+++ b/drivers/media/tuners/Kconfig

@@ -71,6 +71,11 @@ config MEDIA_TUNER_TEA5767
        help
          Say Y here to include support for the Philips TEA5767 radio tuner.
 
+config MEDIA_TUNER_TEA5767_BUILD
+       tristate
+       default VIDEO_TUNER || MEDIA_TUNER_TEA5767
+       depends on MEDIA_TUNER_TEA5767!=n
+
 config MEDIA_TUNER_MSI001
        tristate "Mirics MSi001"
        depends on MEDIA_SUPPORT && SPI && VIDEO_V4L2

We'd then have to do the same for each tuner driver that we have in the
kernel or that gets added later.

Yes :-(.

quoted

Actually, I have hard time coming up with a kconfig syntactic sugar to
express such dependency. If I understand it correctly, the valid
configurations in this case are

MEDIA_TUNER_TEA5767     n       m       y
VIDEO_TUNER     n       x       x       x
                m       x       x       x
                y       x               x

I.e. only VIDEO_TUNER=y and MEDIA_TUNER_TEA5767=m is incorrect, isn't
it?

Yes, I think that is correct.

OK. In this case, a kconfig extension would be really too specific: "if
set, set to at least X but only if X is set as well."

We have similar problems in other areas
of the kernel. In theory, we could enforce the VIDEO_TUNER driver to
be modular here by adding lots of dependencies to it:

config VIDEO_TUNER
	tristate
	depends on MEDIA_TUNER_TEA5761 || !MEDIA_TUNER_TEA5761
	depends on MEDIA_TUNER_TEA5767 || !MEDIA_TUNER_TEA5767
	depends on MEDIA_TUNER_MSI001  || !MEDIA_TUNER_MSI001

Nah, that's even uglier. I suggest to merge your IS_REACHABLE patch.

Michal