Re: [PATCH 4/4] sh: machvec: remove custom ioport_{un,}map()

["Arnd Bergmann" <arnd@arndb.de>]

On Wed, Sep 13, 2023, at 16:13, Geert Uytterhoeven wrote:
> On Wed, Sep 13, 2023 at 4:08 PM Arnd Bergmann <arnd@arndb.de> wrote:
>> On Wed, Sep 13, 2023, at 14:32, Geert Uytterhoeven wrote:
>> > On Wed, Aug 2, 2023 at 8:49 PM Arnd Bergmann <arnd@kernel.org> wrote:
>>
>> Do you have a link to that series? I don't understand why you'd
>> want to enable GENERIC_IOMAP on sh, given that its PIO accesses
>> are always memory mapped in the end.
>
> "[RESEND RFC PATCH 00/12] DeviceTree support for SH7751 based boards."
> https://lore.kernel.org/linux-sh/cover.1693444193.git.ysato@users.sourceforge.jp/

Ok, thanks.

> In the meantime, there is a v2, which I wasn't aware of when I wrote
> my previous email, so perhaps my comment is no longer valid.
> "[RFC PATCH v2 00/30] Device Tree support for SH7751 based board"
> https://lore.kernel.org/linux-sh/cover.1694596125.git.ysato@users.sourceforge.jp

Right, it looks like the GENERIC_IOMAP part if gone from that
series, and I also see that the PCI host bridge does not actually
map the port I/O window. That's usually fine because very few
drivers actually need it, and it also means that there should be
no need for GENERIC_IOMAP or the simpler alternative.

The first version probably only did it accidentally, which is a
common mistake, and I think the ones for hexagon, m68k, and
mips can probably be removed as well with some simplifiations.

x86 and ia64 want GENERIC_IOMAP because they require using
custom instructions for accessing IORESOURCE_IO registers,
but it's not really generic.

    Arnd