Re: [Xen-devel] [PATCH v2 8/8] efi: drop original xen.efi code and build mechanism

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 04.07.18 at 16:25, <daniel.kiper@xxxxxxxxxx> wrote:
> On Thu, Jun 28, 2018 at 07:51:52AM -0600, Jan Beulich wrote:
>> >>> On 19.06.18 at 16:35, <daniel.kiper@xxxxxxxxxx> wrote:
>> > Then rename xen.mb.efi to xen.efi and drop all related
>> > differentiators in the code.
>>
>> For this you'll first of all need to convince me that the binary you build is
>> a drop-in replacement for xen.efi. As noted in the replies to earlier
>> patches, I'm getting the impression of this not being the case. A further
>> hint towards this is the outright deletion of xen/arch/x86/efi/mkreloc.c:
>> How is the Xen image going to be relocated that way, when loaded from
>> the EFI shell or boot loader?
> 
> It works because all addressing is relative to %rip. To be precise, new
> xen.efi, earlier xen.mb.efi, contains exactly the same code as ELF does.
> So, if ELF works without any relocations why PE should not. Especially on
> x86-64. Additionally, my tests showed that in general UEFI implementations
> just require Base Relocation Table entry in PE Data Directories to relocate
> the image.

The thing I'm missing in the series is the generation of the relocations to
go into the section pointed to by this Data Directory entry.

> Even it can be empty. As it is in current patchset.

No, it can't be empty. Just try removing the relocations from xen.efi and
see what you get.

> Though I am
> afraid about more picky UEFI stuff and considering addition of at least
> one .reloc entry, fake one, as Linux kernel does. And this rises another
> question: should not we add .bss section into PE header? Right now it is
> embedded/hidden in PE .text section.

My xen.efi does have a .bss section.

Jan



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel