[PATCH v2 2/2] arm64: efi: add vmlinux debug link to the Image binary

[Ard Biesheuvel]

On 25 January 2017 at 11:53, Mark Rutland <mark.rutland at arm.com> wrote:
> On Wed, Jan 25, 2017 at 10:39:19AM +0000, Ard Biesheuvel wrote:
>> When building with debugging symbols, take the absolute path to the
>> vmlinux binary and add it to the special PE/COFF debug table entry.
>>
>> These entries are used internally by EDK2 based* debug builds of UEFI
>> to populate the DebugImageInfo table, which can be used by debuggers
>> as well as by the OS itself to retrieve information about all loaded
>> PE/COFF executables. This is highly useful for source level debugging
>> of the UEFI stub.
>
> Does that mean EFI_IMAGE_DEBUG_DIRECTORY_ENTRY and friends are
> EDK2-specific?
>
> Or just that the way EDK2 happens to use those is EDK2-specific?
>

Those values are defined by the PE/COFF spec, and I assume that a
CodeView type entry in the debug table usually contains a NUL
terminated string as well, given that the EDK2 crowd is very
Wintel-heavy.

The significance of mentioning EDK2 here was that I thought that the
DebugImageInfo table was a PI construct rather than something
described in the UEFI spec. But looking more carefully, it seems that
this table is in fact a UEFI construct, so I should probably drop this
mention from the commit log

>> * for AArch64, this probably means all of them
>>
>> Cc: Matt Fleming <matt at codeblueprint.co.uk>
>> Signed-off-by: Ard Biesheuvel <ard.biesheuvel at linaro.org>
>> ---
>>  arch/arm64/kernel/Makefile |  4 +++
>>  arch/arm64/kernel/head.S   | 34 +++++++++++++++++++-
>>  arch/arm64/kernel/image.h  |  3 ++
>>  3 files changed, 40 insertions(+), 1 deletion(-)
>
>> +efi_debug_table:
>> +     // EFI_IMAGE_DEBUG_DIRECTORY_ENTRY
>> +     .long   0                       // Characteristics
>> +     .long   0                       // TimeDateStamp
>> +     .short  0                       // MajorVersion
>> +     .short  0                       // MinorVersion
>> +     .long   2                       // Type == EFI_IMAGE_DEBUG_TYPE_CODEVIEW
>
>> +ENTRY(efi_debug_entry)
>> +     // EFI_IMAGE_DEBUG_CODEVIEW_NB10_ENTRY
>> +     .long   0x3031424E              // Signature
>
> Certainly not a blocker for this, but this reminds me that it would be
> nice to de-magic the EFI constants used by the stub.
>
> I took a stab a while back [1], for the existing definitions, but that
> fell by the wayside.
>

Good point.