On Mon, Apr 8, 2024, at 11:54, Ilpo Järvinen wrote:
> On Fri, 5 Apr 2024, Niklas Schnelle wrote:
>> config SERIAL_8250_CS
>> tristate "8250/16550 PCMCIA device support"
>> - depends on PCMCIA && SERIAL_8250
>> + depends on PCMCIA && SERIAL_8250 && HAS_IOPORT
>> help
>> Say Y here to enable support for 16-bit PCMCIA serial devices,
>> including serial port cards, modems, and the modem functions of
>
> What about drivers that use SERIAL8250_PORT()?
It probably makes sense to hide these, since they won't ever
work. I probably missed them in my initial series because they
don't cause a compile-time error, but I agree that there is no
use in showing the options here.
> Also port provided in 8250_PNP might expect it I think.
I don't think these need any change: 8250_pnp.c supports
both IORESOURCE_IO and IORESOURCE_MEM based ports. It will
still create a 8250 port for the I/O based ones but they
will now correctly fail to probe in the main driver rather
than crashing the kernel. PNP devices that only use
memory BARs will keep working as before, on both machines
with and without CONFIG_HAS_IOPORT.
I think that most 8250_pnp variants are probably used only
with ISAPNP or PNPBIOS, neither of which exists without
HAS_IOPORT, but you could certainly have PNPACPI on arm
or riscv machines that don't have port I/O but come with
a memory-mapped 8250 port described by firmware.
Arnd
