Re: [patch] video: build fix for drivers/media/video/pvrusb2/

[Mauro Carvalho Chehab <mchehab@infradead.org>]

On Tue, 13 May 2008 00:03:02 -0400
"Michael Krufky" <mkrufky@linuxtv.org> wrote:

> On Mon, May 12, 2008 at 10:54 PM, Mauro Carvalho Chehab
> <mchehab@infradead.org> wrote:
> >
> > On Sun, 11 May 2008 08:34:06 -0400
> >  "Michael Krufky" <mkrufky@linuxtv.org> wrote:
> >
> >  > On Sun, May 11, 2008 at 3:21 AM, Ingo Molnar <mingo@elte.hu> wrote:
> >  > >
> >  > > x86.git testing found the following build failure:
> >  > >
> >  > >  drivers/built-in.o: In function `pvr2_dvb_feed_thread':
> >  > >  pvrusb2-dvb.c:(.text+0x127e78): undefined reference to `dvb_dmx_swfilter'
> >  > >  drivers/built-in.o: In function `pvr2_dvb_adapter_exit':
> >  > >  pvrusb2-dvb.c:(.text+0x128357): undefined reference to `dvb_net_release'
> >  > >  pvrusb2-dvb.c:(.text+0x12836f): undefined reference to `dvb_dmxdev_release'
> >  > >  [...]
> >  > >
> >  > > with this config:
> >  > >
> >  > >  http://redhat.com/~mingo/misc/config-Sun_May_11_07_06_35_CEST_2008.bad
> >  > >
> >  > > the reason for the missing symbols is this combination:
> >  > >
> >  > >  CONFIG_VIDEO_PVRUSB2=y
> >  > >  CONFIG_DVB_CORE=m
> >  > >
> >  > > i.e. pvrusb2 is built-in, dvb-core is modular.
> >  > >
> >  > > This patch solves the problem by adding a dependency on DVB_CORE - this
> >  > > is used by other drivers such as au0828 as well. This way the pvrusb2
> >  > > driver can still be built, but if dvb-core is a module then it will
> >  > > correctly be a module as well and cannot be built-in.
> >  > >
> >  > > Signed-off-by: Ingo Molnar <mingo@elte.hu>
> >  > > ---
> >  > >  drivers/media/video/pvrusb2/Kconfig |    2 +-
> >  > >  1 file changed, 1 insertion(+), 1 deletion(-)
> >  > >
> >  > > Index: linux/drivers/media/video/pvrusb2/Kconfig
> >  > > ===================================================================
> >  > > --- linux.orig/drivers/media/video/pvrusb2/Kconfig
> >  > > +++ linux/drivers/media/video/pvrusb2/Kconfig
> >  > > @@ -1,6 +1,6 @@
> >  > >  config VIDEO_PVRUSB2
> >  > >        tristate "Hauppauge WinTV-PVR USB2 support"
> >  > > -       depends on VIDEO_V4L2 && I2C
> >  > > +       depends on VIDEO_V4L2 && I2C && DVB_CORE
> >  > >        select FW_LOADER
> >  > >        select MEDIA_TUNER
> >  > >        select VIDEO_TVEEPROM
> >  > >
> >  >
> >  > Ingo,
> >  >
> >  > VIDEO_PVRUSB2 should not depend on DVB_CORE unless VIDEO_PVRUSB2_DVB
> >  > is selected, which already depends on DVB_CORE.
> >  >
> >  > For example, if a user has VIDEO_PVRUSB2_DVB not selected, then your
> >  > patch would generate a false dependency on DVB_CORE.
> >
> >  Maybe we can make pvrusb2 dependent of VIDEO_MEDIA. This is a DVB_CORE ||
> >  VIDEO_DEV. So, pvrusb should be 'm' in this case.
> 
> That sounds like it would be OK, although something like this would
> probably be better:
> 
>  config VIDEO_PVRUSB2
>        tristate "Hauppauge WinTV-PVR USB2 support"
> -       depends on VIDEO_V4L2 && I2C
> +       depends on VIDEO_V4L2 && I2C && (DVB_CORE if VIDEO_PVRUSB2_DVB)
>        select FW_LOADER
>        select MEDIA_TUNER
>        select VIDEO_TVEEPROM

This doesn't look to be a good idea, since VIDEO_PVRUSB2_DVB depends on
VIDEO_PVRUSB2. So, you'll create a circular dependency. The syntax I've
proposed seems cleaner. Of course, it needs to be tested. IMO, all hybrid
devices should be dependent of VIDEO_MEDIA. This will help to avoid this kind
of issue.

> I don't know if that syntax works for "depends on" , but it does work
> for select.
> 
> if "depends on FOO if BAR" doesnt work, would adding "select DVB_CORE
> if VIDEO_PVRUSB2_DVB" solve the problem?

Also, this leads into a circular reference.

Cheers,
Mauro