Re: [PATCH 4/4] lib/iomap.c: improve comment about pci anomaly

["Arnd Bergmann" <arnd@kernel.org>]

On Wed, Nov 29, 2023, at 13:40, Philipp Stanner wrote:
> On Tue, 2023-11-21 at 11:03 +0100, Arnd Bergmann wrote:
>> On Mon, Nov 20, 2023, at 22:59, Philipp Stanner wrote:
>> We should be able to define a generic version like
>> 
>> void pci_iounmap(struct pci_dev *dev, void __iomem * addr)
>> {
>> #ifdef CONFIG_HAS_IOPORT
>>        if (iomem_is_ioport(addr)) {
>>               ioport_unmap(addr);
>>               return;
>>        }
>> #endif
>>       iounmap(addr)
>> }
>
> ACK, I think this makes sense – if we agree (as in the other thread)
> that we never need an empty pci_iounmap().
>
>> 
>> and then define iomem_is_ioport() in lib/iomap.c for x86,
>
> Wait a minute.
> lib/ should never contain architecture-specific code, should it? I
> assume your argument is that we write a default version of
> iomem_is_ioport(), that could, in theory, be used by many
> architectures, but ultimately only x86 will use that default.

I would hope that eventually we can build lib/iomap.c
only on x86, as everything else can live without it.

>> while defining it in asm-generic/io.h for the rest,
>
> So we're not talking about the function prototypes here, but about the
> actual implementation that should reside in asm-generic/io.h, aye?
> Because the prototype obviously always has to be identical.

It could live in lib/pci_iomap.c or in
include/asm-generic/pci_iomap.h, it doesn't really matter
since the definition is trivial. asm-generic/io.h is probably
not the right place, unless we want to merge all of
asm-generic/pci_iomap.h into asm-generic/io.h. We could
do that now that all architectures include asm-generic/io.h
and that includes asm-generic/pci_iomap.h unconditionally.

>> with an override in asm/io.h for those architectures
>> that need a custom inb().
>
> So like this in ARCH/include/asm/io.h:
>
> #define iomem_is_ioport iomem_is_ioport
> bool iomem_is_ioport(...);
>
> and in include/asm-generic/io.h:
>
> #ifndef iomem_is_ioport
> bool iomem_is_ioport(...);
>
> correct?

Yes.
 
>> arch/powerpc/platforms/Kconfig: select GENERIC_IOMAP
>> 
>> This is only used for two platforms: cell and powernv,
>> though on Cell it no longer does anything after the
>> commit f4981a00636 ("powerpc: Remove the celleb support");
>> I think the entire io_workarounds code now be folded
>> back into spider_pci.c if we wanted to.
>> 
>> The PowerNV LPC support does seem to still rely on it.
>> This tries to do the exact same thing as lib/logic_pio.c
>> for Huawei arm64 servers. I suspect that neither of them
>> does it entirely correctly since the powerpc side appears
>> to just override any non-LPC PIO support while the arm64
>> side is missing the ioread/iowrite support.
>
> I think by now I get what the issue with GENERIC_IOMAP is. But do you
> want me to do something about GENERIC_IOMAP (besides the things
> directly related to the PCI functionality I'm touching) for you to
> approve of a modified version of this patch series?

It would be nice to clean up some of the architectures
that incorrectly select it at the moment, but that
can be a separate series if you want to get this one
done first, or I can take a look myself.

      Arnd