Re: [PATCH v5 6/8] arm64/efi: move SetVirtualAddressMap() to UEFI stub

[Matt Fleming <matt@codeblueprint.co.uk>]

On Thu, 08 Jan, at 06:48:32PM, Ard Biesheuvel wrote:
> @@ -46,4 +54,26 @@ extern void efi_idmap_init(void);
>  
>  #define EFI_ALLOC_ALIGN		SZ_64K
>  
> +/*
> + * On ARM systems, virtually remapped UEFI runtime services are set up in three
> + * distinct stages:
> + * - The stub retrieves the final version of the memory map from UEFI, populates
> + *   the virt_addr fields and calls the SetVirtualAddressMap() [SVAM] runtime
> + *   service to communicate the new mapping to the firmware (Note that the new
> + *   mapping is not live at this time)
> + * - During early boot, the page tables are allocated and populated based on the
> + *   virt_addr fields in the memory map, but only if all descriptors with the
> + *   EFI_MEMORY_RUNTIME attribute have a non-zero value for virt_addr. If this
> + *   succeeds, the EFI_VIRTMAP flag is set to indicate that the virtual mappings
> + *   have been installed successfully.
> + * - During an early initcall(), the UEFI Runtime Services are enabled and the
> + *   EFI_RUNTIME_SERVICES bit set if some conditions are met, i.e., we need a
> + *   non-early mapping of the UEFI system table, and we need to have the virtmap
> + *   installed.
> + */
> +#define EFI_VIRTMAP		EFI_ARCH_1

Lucid. Thanks, Ard!

-- 
Matt Fleming, Intel Open Source Technology Center