On Wed, Apr 04, 2018 at 01:19:29PM +0200, Linus Walleij wrote:
>- There is a series of ISA_BUS_API changes that ended up
>  getting queued in my tree. Most of them have proper ACKs
>  and they have been tested throughly in linux-next.
>  These patches finally split ISA_BUS and ISA_BUS_API
>  apart as I understand it, as there are devices using
>  port-mapped I/O but not strictly speaking using any ISA
>  bus. So the ISA_BUS and the ISA_BUS_API are two
>  things. The most notable effect of ISA_BUS_API is that
>  it compiles drivers/base/isa.c so that we get device driver
>  core infrastructure for these devices. The ISA bus was merged
>  in 2006 (a5117ba7da37) and was used for sound cards and
>  watchdog, but only recently William started
>  to use it for his industrial data acquisition cards and he's found
>  a bit of problems that are now (hopefully) finally fixed up.
>  Any questions: William Breathitt Gray is on the CC for this
>  pull request to answer them.

Some background on the impetus and rationale of the ISA_BUS_API Kconfig

During submission of the Winbond GPIO driver (a0d65009411c), an attempt
to select ISA_BUS_API revealed a recursive dependency bug:

The recursive dependency could be avoided if ISA_BUS_API was selected by
drivers where needed rather than depended on. A select semantic makes
sense as well for these drivers because ISA_BUS_API is simply used to
compile the drivers/base/isa.c file, which is just a thin wrapper to
allow these ISA-style I/O port drivers to fit the device driver model.

ISA_BUS_API is decoupled from the ISA_BUS so that it can be selected
independent of the legacy ISA_BUS drivers, which do not necessary make
use of drivers/base/isa.c functionality. Thus there is a clean
separation between the modern drivers for ISA-style devices which select
ISA_BUS_API, and the legacy drivers for true ISA cards which utilize the
legacy ISA_BUS Kconfig option.

The reason these changes are coming in through the GPIO tree is to avoid
a similar recursive dependency error if the STX104 IIO driver attempts
to select ISA_BUS_API separate from the other patches (due to its
GPIOLIB selection): https://lkml.org/lkml/2017/12/29/384

These updates missed the last merge window, so the Winbond GPIO driver
was merged in at that time with a "depends on ISA_BUS_API" line patch to
pacify the recursive dependency error. Now that these updates were
picked up in the GPIO tree, the revert patch was merged to revert the
"depends on ISA_BUS_API" line to the originally intended
"select ISA_BUS_API" line.

Finally, an explicit EXPERT dependency is added to the PC104 Kconfig
option to keep the PC/104 drivers masked as they have been -- the
PC/104 interface is somewhat niche and unlikely to be a concern for
regular users, so let's keep it tucked away.

William Breathitt Gray

Reply via email to