On Sun, Jul 21, 2019 at 6:10 PM Arnd Bergmann <arnd@xxxxxxxx> wrote:
>
> On Sun, Jul 21, 2019 at 5:45 AM Masahiro Yamada
> <yamada.masahiro@xxxxxxxxxxxxx> wrote:
> > On Sat, Jul 20, 2019 at 10:55 PM Arnd Bergmann <arnd@xxxxxxxx> wrote:
> > > On Sat, Jul 20, 2019 at 5:26 AM Masahiro Yamada <yamada.masahiro@xxxxxxxxxxxxx> wrote:
> > > > So, what shall we do?
> > > >
> > > > Drop the board-file support? Or, keep it
> > > > in case somebody is still using their board-files
> > > > in downstream?
> >>
> > > For this file, all boards got converted to DT, and the old setup
> > > code removed in commit ebc278f15759 ("ARM: mvebu: remove static
> > > LED setup for netxbig boards"), four years ago, so it's a fairly
> > > easy decision to make it DT only.
> >
> > I see another case, which is difficult
> > to make a decision.
> >
> > For example, drivers/spi/spi-tle62x0.c
> >
> > This driver supports only board-file, but the board-file
> > is not found in upstream.
> >
> > Unless I am terribly missing something,
> > there is no one who passes tle62x0_pdata
> > to this driver.
> >
> > $ git grep tle62x0_pdata
> > drivers/spi/spi-tle62x0.c: struct tle62x0_pdata *pdata;
> > include/linux/spi/tle62x0.h:struct tle62x0_pdata {
> >
> > But, removing board-file support
> > makes this driver completely useless...
>
> Adding Ben Dooks to Cc.
>
> I suspect this driver is completely obsolete and should be removed.
>
> For some reason, it's not an SPI controller driver like all the other
> files in that directory, but implements low-level access to the state
> of a particular SPI device.
>
> However, there should not really be a low-level driver for it that
> just exports the pins to user space. It should either be a gpiolib
> driver to let other drivers talk to the pins, or a high-level driver that
> exposes the intended functionality (watchdog, regulator, ...)
> to those respective subsystems.
>
> Arnd
Another example that I have no idea
how it works:
drivers/net/hamradio/yam.c
yam_ioctl() reads data from user-space,
but the data structures for ioctl are
defined in include/linux/yam.h
This header is not exported to user-space
since it is outside of the uapi directory.
I dug the git history, but it has never
exported to user-space in the past.
I do not know how user-space programs can
pass-in data to the kernel.
If we want to fix this, we could move it
to include/uapi/linux/yam.h
But, if nobody has reported any problem about this,
it might be a good proof that nobody is using this driver.
Maybe, can we simply drop odd drivers??
--
Best Regards
Masahiro Yamada
