Re: [PATCH] x86_64/efi: Mapping Boot and Runtime EFI memory regions to different starting virtual address

[joeyli <jlee@suse.com>]

On Thu, Jul 30, 2015 at 02:17:23PM +0100, Matt Fleming wrote:
> On Thu, 30 Jul, at 08:31:16PM, joeyli wrote:
> > 
> > I think hibernate overwrite it.
> 
> We absolutely must get a more detailed answer before going any further.
> 
> Simply put, if we're remapping the EFI regions into the virtual address
> space and calling SetVirtualAddressMap() on hibernate resume there is no
> reason that anyone should be using the old mappings.
> 
> And since you've demonstrated that we *are* using the old mappings,
> we've likely got a bug somewhere that we need to get a handle on before
> we paper over the issue.
> 
> Where exactly is the old mapping address being used? Is it that
> efi.systab->runtime->get_variable is incorrect? If you could paste the
> disassembled output where the page fault occurs, that would be helpful.
> 
> -- 
> Matt Fleming, Intel Open Source Technology Center

OK, understood! Thanks for your suggestion!

But, I have a question about mapping Boot Code/Data to -4G area. I understand
we need Runtime regions, and 1:1 mapping is the workaround of some buggy BIOS.
But why should kernel mapping Boot regions to -4G area?


Thanks a lot!
Joey Lee