Re: [PATCH 1/4] isa: Allow ISA-style drivers on modern systems

[Linus Torvalds]

On Mon, May 23, 2016 at 7:58 AM, William Breathitt Gray
<vilhelm.gray@xxxxxxxxx> wrote:
>
> For now, the ISA_BUS Kconfig option is only be available on X86
> architectures. Support for other architectures may be added as required.

So I'd prefer to see that

> +config ISA_BUS_API
> +       def_bool ISA

part in arch/Kconfig.

Why?

Because other architectures _already_ define that ISA symbol, and we
want the "ISA_BUS_API" to be a complete superset of ISA.

So whenever ISA is enabled, ISA_BUS_API should be enabled.

And the way you did that, that's not true. Now, if you were to enable
ISA on ARM, you'd not get ISA_BUS_API. And that sounds insane to me.
It also sounds *wrong* because it effectively changes the meaning of
this:

  --- a/drivers/base/Makefile
  +++ b/drivers/base/Makefile
  @@ -10,7 +10,7 @@ obj-$(CONFIG_DMA_CMA) += dma-contiguous.o
  -obj-$(CONFIG_ISA)      += isa.o
  +obj-$(CONFIG_ISA_BUS_API)      += isa.o

where now that "isa.c" file gets built only on x86, whereas it *used*
to get built whenever ISA was enabled.

So the reason I suggested a separate ISA_BUS_API config option (that
then a particular architecture can choose to enable, in this case the
x86 choice of selecting ISA_BUS) was _exactly_ this issue. The plain
"ISA" config variable is not limited to x86, and the new subset of it
(the ISA_BUS_API) thus also must not be limited to just x86.

               Linus