Re: [RFC 01/13] binutils: Support for the NT_X86_CPUID core dump note

[John Baldwin <jhb@FreeBSD.org>]

On 10/16/23 2:23 AM, Lancelot SIX wrote:
> Hi John
> 
> I am aware this is a RFC series where you are mostly looking for
> feedback on the overall direction, but I have an implemenation question
> below (feel free to discard as part of the RFC process).t
> 
> On Mon, Oct 09, 2023 at 11:36:03AM -0700, John Baldwin wrote:
>> This core dump note contains an array of CPUID leaf values.  Each
>> entry in the array contains six 32-bit integers describing the inputs
>> to the CPUID instruction (%eax and %ecx) and the outputs of the
>> instruction (%eax, %ebx, %ecx, and %edx) similar to the C structure:
>>
>> struct cpuid_leaf
>> {
>>      uint32_t leaf;
>>      uint32_t subleaf;
>>      uint32_t eax;
>>      uint32_t ebx;
>>      uint32_t ecx;
>>      uint32_t edx;
>> };
>>
>> Current implementations of this note contain leaves associated with
>> the XSAVE state area (major leaf 0xd), but future implementations may
>> add other leaf values in the future.
>> ---
>>   bfd/elf-bfd.h        |  2 ++
>>   bfd/elf.c            | 35 +++++++++++++++++++++++++++++++++++
>>   binutils/readelf.c   |  2 ++
>>   include/elf/common.h |  2 ++
>>   4 files changed, 41 insertions(+)
>>
>> @@ -12640,6 +12660,19 @@ elfcore_write_xstatereg (bfd *abfd, char *buf, int *bufsiz,
>>   			     note_name, NT_X86_XSTATE, xfpregs, size);
>>   }
>>   
>> +char *
>> +elfcore_write_x86_cpuid (bfd *abfd, char *buf, int *bufsiz,
>> +			 const void *cpuid, int size)
>> +{
>> +  char *note_name;
>> +  if (get_elf_backend_data (abfd)->elf_osabi == ELFOSABI_FREEBSD)
>> +    note_name = "FreeBSD";
> 
> The code above (in elfcore_grok_note) only accept "LINUX", and the
> comment in "include/elf/common.h" says "note name must be "LINUX".
> 
> Is this something you intend to adjust later when adding full FreeBSD
> support?

I really just kept that to match what is there for NT_X86_XSTATE, but
this patch already includes FreeBSD support (FreeBSD notes use
elfcore_grok_freebsd_note rather than elfcore_grok_note).  I almost
left that comment about the note name out, and probably I should just
do that.  (Arguably I should also remove that comment from NT_X86_XSTATE
and possibly some others that both FreeBSD and Linux generate.)

On a related note, abusing elf_osabi to determine the note name is a
bit of a hack.  It would be nice if elfcore_write_register_note
accepted the note name as an argument, but I'm not sure how much
that would break (e.g. if that's a public binutils API used by
other things than gdb).

-- 
John Baldwin