[PATCH v4 4/5] arm: Add [U]EFI runtime services support

[Leif Lindholm]

On Mon, Jan 13, 2014 at 07:43:09PM +0100, Arnd Bergmann wrote:
> > This patch implements basic support for UEFI runtime services in the
> > ARM architecture - a requirement for using efibootmgr to read and update
> > the system boot configuration.
> > 
> > It uses the generic configuration table scanning to populate ACPI and
> > SMBIOS pointers.
> 
> As far as I'm concerned there are no plans to have ACPI support on ARM32,
> so I wonder what the code to populate the ACPI tables is about. Can
> you clarify this?

Are you suggesting that I should #ifndef ARM in common code, or that I
should neglect to document what the common code will do with data it is
given by UEFI?

> > diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
> > index 78a79a6a..1ab24cc 100644
> > --- a/arch/arm/Kconfig
> > +++ b/arch/arm/Kconfig
> > @@ -1853,6 +1853,20 @@ config EARLY_IOREMAP
> >  	  the same virtual memory range as kmap so all early mappings must
> >  	  be unapped before paging_init() is called.
> >  
> > +config EFI
> > +	bool "UEFI runtime service support"
> > +	depends on OF && !CPU_BIG_ENDIAN
> 
> What is the dependency on !CPU_BIG_ENDIAN?

Mainly on code not being implemented to byte-reverse UCS strings.

> We try hard to have
> all kernel code support both big-endian and little-endian, and
> I'm guessing there is a significant overlap between the people
> that want UEFI support and those that want big-endian kernels.

Not really. There might be some. Also, it is not necessarily the case
that those people want to run the big-endian kernel at EL2.

If a need is seen, this support can be added at a later date.

> > +struct efi_memory_map memmap;
> 
> "memmap" is not a good name for a global identifier, particularly because
> it's easily confused with the well-known "mem_map" array. This
> wants namespace prefix like you other variable, or a "static" tag,
> preferably both.

It is defined by include/linux/efi.h.

> > +/*
> > + * This function switches the UEFI runtime services to virtual mode.
> > + * This operation must be performed only once in the system's lifetime,
> > + * including any kecec calls.
> 
>                    kexec

Ok.

> > diff --git a/include/linux/efi.h b/include/linux/efi.h
> > index fa7d950..afaeb85 100644
> > --- a/include/linux/efi.h
> > +++ b/include/linux/efi.h
> > @@ -664,7 +664,7 @@ extern int __init efi_setup_pcdp_console(char *);
> >  #define EFI_64BIT		5	/* Is the firmware 64-bit? */
> >  
> >  #ifdef CONFIG_EFI
> > -# ifdef CONFIG_X86
> > +# if defined(CONFIG_X86) || defined(CONFIG_ARM)
> >  extern int efi_enabled(int facility);
> >  # else
> >  static inline int efi_enabled(int facility)
> 
> Maybe better #ifndef CONFIG_IA64? It seems that is the odd one out and
> all possible future architectures would be like x86 and ARM.

This was pointed out by Matt Fleming earlier, so it will change.
Mark Salter suggested introducing something like ARCH_USES_EFI_FACILITY
would be a bit cleaner.

/
    Leif