On Tuesday 26 January 2016 14:36:44 Mauro Carvalho Chehab wrote:

Em Tue, 26 Jan 2016 16:53:38 +0100
Arnd Bergmann [off-list ref] escreveu:

quoted

On Tuesday 26 January 2016 12:33:08 Mauro Carvalho Chehab wrote:

quoted

Em Tue, 26 Jan 2016 15:10:00 +0100
Advanced users may, instead, manually select the media tuner that his
hardware needs. In such case, it doesn't matter if MEDIA_TUNER
is enabled or not.

As this is due to a Kconfig limitation, I've no idea how to fix or get
hid of it, but making em28xx dependent of MEDIA_TUNER is wrong.  

I don't understand what limitation you see here. 

Before MEDIA_TUNER, what we had was something like:

	config MEDIA_driver_foo
	select VIDEO_tuner_bar if MEDIA_SUBDRV_AUTOSELECT
	select MEDIA_frontend_foobar if MEDIA_SUBDRV_AUTOSELECT
	...

However, as different I2C drivers had different dependencies, this
used to cause lots of troubles. So, one of the Kbuild maintainers
came out with the idea of converting from select into depends on.
The MEDIA_TUNER is just an ancillary invisible option to make it
work at the tuner's side, as usually what we want is to have all
tuners selected, as we don't have a one to one mapping about what
driver supports what tuner (nor we wanted to do it, as this would
mean lots of work for not much gain).

Ok

quoted

The definition
of the VIDEO_TUNER symbol is an empty 'tristate' symbol with a
dependency on MEDIA_TUNER to ensure we get a warning if MEDIA_TUNER
is not enabled, and to ensure it is set to 'm' if MEDIA_TUNER=m and
a "bool" driver selects VIDEO_TUNER.

No, VIDEO_TUNER is there because we wanted to be able to use select
to enable V4L2 tuner core support and let people to manually select
the needed I2C devices with MEDIA_SUBDRV_AUTOSELECT unselected.

I meant what the dependency is there for, not the symbol itself.
It's clear what the symbol does.

quoted

diff --git a/drivers/media/v4l2-core/Kconfig b/drivers/media/v4l2-core/Kconfig
index 9beece00869b..1050bdf1848f 100644
--- a/drivers/media/v4l2-core/Kconfig
+++ b/drivers/media/v4l2-core/Kconfig

@@ -37,7 +37,11 @@ config VIDEO_PCI_SKELETON
 # Used by drivers that need tuner.ko
 config VIDEO_TUNER
 	tristate
-	depends on MEDIA_TUNER
+
+config VIDEO_TUNER_MODULE
+	tristate # must not be built-in if MEDIA_TUNER=m because of I2C
+	default y if VIDEO_TUNER=y || MEDIA_TUNER=y
+	default m if VIDEO_TUNER=m

Doesn't need to worry about that, because all drivers that select VIDEO_TUNER
depend on I2C:

Ok, then the dependency does not do anything other than generate a
warning.

So this means it's now enabled if MEDIA_TUNER is enabled, whereas
before it was only enabled when explicitly selected. That sounds
like a useful change, but it also seems unrelated to the warning
fix, and should probably be a separate change, while for now
we can simply remove the 'depends on' line without any replacement.

Ok, if we'll have platform drivers for analog TV using the I2C bus
at directly in SoC, then your solution is better, but the tuner core
driver may not be the best way of doing it. So, for now, I would use
the simpler version.

Ok. Do you want me to submit a new version or do you prefer to write
one yourself? With or without the 'default'?

	Arnd