Re: [PATCH] arm64: acpi: Export symbol for acpi_os_ioremap

[Sudeep Holla]

On Mon, May 29, 2023 at 03:31:12PM +0200, Ard Biesheuvel wrote:
> On Fri, 26 May 2023 at 15:12, lihuisong (C) <lihuisong@xxxxxxxxxx> wrote:
> >
> >
> > 在 2023/5/26 20:39, Ard Biesheuvel 写道:
> > > (cc Lorenzo)
> > >
> > > On Fri, 26 May 2023 at 14:20, Huisong Li <lihuisong@xxxxxxxxxx> wrote:
> > >> The driver who calls the acpi_os_ioremap() cannot be compiled if the 'M'
> > >> is selected for the driver. The compiling log is as follows:
> > >> -->
> > >> MODPOST Module.symvers
> > >> ERROR: modpost: "acpi_os_ioremap" [drivers/soc/hisilicon/xxx.ko] undefined!
> > >> scripts/Makefile.modpost:136: recipe for target 'Module.symvers' failed
> > >> make[1]: *** [Module.symvers] Error 1
> > >>
> > >> So this patch exports symbol for acpi_os_ioremap.
> > >>
> > > That driver does not exist in mainline.
> >
> > We have an uploading driver [1] that may use it.
> >
> > [1]
> > https://patchwork.kernel.org/project/linux-soc/patch/20230522072211.8894-2-lihuisong@xxxxxxxxxx/
> >
> > >
> > > Why does it need to use acpi_os_ioremap() instead of the ordinary
> > > memremap/ioremap routines?
> > This driver needs to ioremap the shared memory space of a PCC subspace.
> > And @Sudeep suggested that we use this interface.
> > It is suitable here.
> 
> I disagree. acpi_io_ioremap() is internal arch plumbing for the ACPI
> subsystem. I don't see why we should use it here.
>

Yes. One reason I suggested this was in past firmware authors had mixed
the memory allocated for PCC and using acpi_io_ioremap() made sense. But
I hear you and it make sense to avoid it especially if the driver must
know what type of memory it is and must be dealing with.

> On arm64, acpi_os_ioremap() cross references the EFI memory map to
> figure out whether a physical region is memory or device, but a driver
> should already know that.

Agreed.

-- 
Regards,
Sudeep