On Tue, 2021-03-23 at 21:27 +0200, Sakari Ailus wrote:
> Hi Ezequiel,
>
> On Tue, Mar 23, 2021 at 02:46:04PM -0300, Ezequiel Garcia wrote:
> > On Fri, 2021-03-12 at 14:56 +0200, Sakari Ailus wrote:
> > > The V4L2 async framework is generally used with the V4L2 fwnode, which
> > > also depends on the former. There is only one exception, the CAFE_CCIC
> > > driver, which uses V4L2 async but does not need V4L2 fwnode.
> > >
> > > At the same time there is a vast number of systems that need videodev
> > > module, but have no use for v4l2-async that's now part of videodev.
> > >
> > > In order to improve, build v4l2-async and v4l2-fwnode as a single module
> > > called v4l2-async (the v4l2-async.c file is renamed as v4l2-async-core.c).
> > > Also the menu item V4L2_FWNODE is renamed as V4L2_ASYNC.
> > >
> >
> > Seems this is not what this patch is doing: the symbol is not renamed,
> > and now we have two modules v4l2-fwnode.ko and v4l2-async.ko, unless
> > I'm confused.
>
> Oh, well, I changed what the patch does but forgot to update this. :-I
>
> Basically this now splits v4l2-async out of videodev.
>
> >
> > I personally was more fond of having just one v4l2-async.ko module,
> > but I'm not sure if you found any obstacles.
> >
> > > This also moves the initialisation of the debufs entries for async subdevs
> > > to loading of the v4l2-async module. The directory is named as
> > > "v4l2-async".
> > >
> > > Signed-off-by: Sakari Ailus <sakari.ailus@xxxxxxxxxxxxxxx>
> > > ---
> > > drivers/media/i2c/Kconfig | 8 ++++++++
> > > drivers/media/v4l2-core/Kconfig | 4 ++++
> > > drivers/media/v4l2-core/Makefile | 11 +++++++++--
> > > drivers/media/v4l2-core/v4l2-async.c | 23 +++++++++++++++++++++--
> > > drivers/media/v4l2-core/v4l2-dev.c | 5 -----
> > > 5 files changed, 42 insertions(+), 9 deletions(-)
> > >
> > [..]
> > > diff --git a/drivers/media/v4l2-core/Makefile b/drivers/media/v4l2-core/Makefile
> > > index e4cd589b99a5..eb5ebb5461fd 100644
> > > --- a/drivers/media/v4l2-core/Makefile
> > > +++ b/drivers/media/v4l2-core/Makefile
> > > @@ -7,15 +7,22 @@ tuner-objs := tuner-core.o
> > >
> > > videodev-objs := v4l2-dev.o v4l2-ioctl.o v4l2-device.o v4l2-fh.o \
> > > v4l2-event.o v4l2-ctrls.o v4l2-subdev.o \
> > > - v4l2-async.o v4l2-common.o
> > > + v4l2-common.o
> > > videodev-$(CONFIG_COMPAT) += v4l2-compat-ioctl32.o
> > > videodev-$(CONFIG_TRACEPOINTS) += v4l2-trace.o
> > > videodev-$(CONFIG_MEDIA_CONTROLLER) += v4l2-mc.o
> > > videodev-$(CONFIG_SPI) += v4l2-spi.o
> > > videodev-$(CONFIG_VIDEO_V4L2_I2C) += v4l2-i2c.o
> > >
> > > -obj-$(CONFIG_V4L2_FWNODE) += v4l2-fwnode.o
> > > obj-$(CONFIG_VIDEO_V4L2) += videodev.o
> > > +ifneq ($(findstring y,$(CONFIG_V4L2_ASYNC)$(CONFIG_V4L2_FWNODE)),)
> >
> > Hm, to me this ifdefery is an indication that we are trying
> > to do something fishy.
>
> What's happening here is that v4l2-fwnode depends on v4l2-async, but for
> drivers to avoid selecting both, v4l2-async is built if only
> CONFIG_V4L2_FWNODE is set.
>
> So this perhaps isn't too pretty but it saves hassle in Kconfig files.
>
How about:
v4l2-async-objs := v4l2-async-core.o
v4l2-async-$(CONFIG_V4L2_FWNODE) += v4l2-fwnode.o
obj-$(CONFIG_V4L2_ASYNC) += v4l2-async.o
config V4L2_FWNODE
bool
select V4L2_ASYNC
config V4L2_ASYNC
tristate
This avoids the ifdefs and expresses the same behavior
you stated above:
* v4l2-async is now out of videodev
* v4l2-fwnode depends on v4l2-async (through select in this case,
given they are helper modules).
* v4l2-async is built if CONFIG_V4L2_FWNODE is set.
And v4l2-async is a module or built-in, depending on the
media subsystem (MEDIA_SUPPORT) being module or built-in,
which I think makes sense.
Thanks,
Ezequiel
