Re: [Xen-devel] Booting signed xen.efi through shim

[Tamas K Lengyel <tamas@xxxxxxxxxxxxx>]

On Wed, Sep 13, 2017 at 11:42 AM, Jan Beulich <JBeulich@xxxxxxxx> wrote:
>>>> On 13.09.17 at 16:40, <tamas@xxxxxxxxxxxxx> wrote:
>> On Wed, Sep 13, 2017 at 3:21 AM, Jan Beulich <JBeulich@xxxxxxxx> wrote:
>>>>>> On 13.09.17 at 07:27, <tamas@xxxxxxxxxxxxx> wrote:
>>>>Sections:
>>>>Idx Name          Size      VMA               LMA               File off  
>>>>Algn
>>>>  0 .text         0017a1ba  ffff82d080200000  ffff82d080200000  00001000  
>>>> 2**12
>>>>                  CONTENTS, ALLOC, LOAD, CODE
>>>>  1 .rodata       000826a0  ffff82d080400000  ffff82d080400000  0017c000  
>>>> 2**5
>>>>                  CONTENTS, ALLOC, LOAD, DATA
>>>>  2 .buildid      00000035  ffff82d0804826a0  ffff82d0804826a0  001fe6a0  
>>>> 2**2
>>>>                  CONTENTS, ALLOC, LOAD, READONLY, DATA
>>>>  3 .init         00077df0  ffff82d080600000  ffff82d080600000  001ff000  
>>>> 2**12
>>>>                  CONTENTS, ALLOC, LOAD, CODE, DATA
>>>>  4 .data.re      0000aa40  ffff82d080800000  ffff82d080800000  00277000  
>>>> 2**7
>>>>                  CONTENTS, ALLOC, LOAD, DATA
>>>>  5 .data         000105a8  ffff82d08080b000  ffff82d08080b000  00282000  
>>>> 2**12
>>>>                  CONTENTS, ALLOC, LOAD, DATA
>>>>  6 .bss          00143280  ffff82d080820000  ffff82d080820000  00000000  
>>>> 2**4
>>>>                  ALLOC, RELOC
>>>
>>> Objdump is apparently ignoring a section attribute bit here - my
>>> own utility properly prints "bss" in addition to "read" (which presumably
>>> matches "ALLOC" above, albeit that's a bogus translation apparently
>>> applying ELF semantics to COFF). You'll want to check that bit 7 in the
>>> section attributes is set. I'm also puzzled by "RELOC", but I do see a
>>> matching bit dumped here; not sure why that's being set.
>>
>> Looking at it with readpe I get:
>>
>> Name:                            .bss
>> Virtual Address:                 0x820000
>> Physical Address:                0x143280
>> Size:                            0 (0 bytes)
>> Pointer To Data:                 0
>> Relocations:                     0
>> Characteristics:                 0xc1000080
>>                                  contains uninitialized data
>>                                  contains extended relocations
>>                                  is readable
>>                                  is writable
>>
>> So bit 7 is set AFAICT.
>
> Good.
>
>>> It is certainly the case that .bss style sections are expected to have a
>>> zero file offset, as there's no data for such sections inside the file (note
>>> the missing "CONTENTS" above. So I would conclude that, unless the
>>> bss flag really got lost, it's a shim loader bug. Since other people can
>>> load xen.efi with the shim, that might be a problem with the particular
>>> version you're using.
>>
>> Perhaps, I'm using the latest master
>> (e22a7b5b772dba6588dd955dc017e572f7e29784) from
>> https://github.com/mjg59/shim, the one being linked to on the wiki. If
>> there is a known good version, I would be happy to give that a shot
>> and see if I can get it working.
>
> I have no idea. What I'd suggest you to try is to zap that stray
> "contains extended relocations" flag. I've written down to go hunt
> for where it comes from, but I don't have the time to do that right
> away.

So I had made some progress using the shim from
https://github.com/rhboot/shim, it actually has been able to jump into
the signed xen.efi. However, Xen bails with the message "Unsupported
relocation type" which is in efi_arch_relocate_image.

Tamas

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel