[PATCH] x32: Allow R_X86_64_PC64 for SFrame V3

[Indu Bhagat]

On 1/19/26 12:16 PM, Indu Bhagat wrote:
> On 1/18/26 6:19 PM, H.J. Lu wrote:
>> SFrame V3 generates R_X86_64_PC64 relocation, instead of R_X86_64_PC32,
>> in .sframe section for x86-64.  Although x32 doesn't support SFrame,
>> .sframe section can be found in x32 object converted from x86-64 object
>> with objcopy, which only changes the ELF file class from ELFCLASS64 to
>> ELFCLASS32 with all section contents unchanged.
>>
>> Update elf_x86_64_scan_relocs to allow R_X86_64_PC64 for x32 so that
>> x32 object file with .sframe section can be used as x32 linker input.
>>
>> bfd/
>>
>> PR ld/33807
>> * elf64-x86-64.c (elf_x86_64_scan_relocs): Allow R_X86_64_PC64
>> for x32.
>>
>> ld/
>>
>> PR ld/33807
>> * testsuite/ld-x86-64/sframe.rd: New file.
>> * testsuite/ld-x86-64/sframe.s: Likewise.
>> * testsuite/ld-x86-64/x86-64.exp: Run PR ld/33807 tests.
>>
>>
> 
> Thanks HJ for the patch.
> 
> (Adding Mathieu, Steve and Josh.)
> 
> For context, we see the issue with kernel builds with SFrame V3 enabled 
> (https://sourceware.org/bugzilla/show_bug.cgi?id=33807).  SFrame V3 
> sections use 64-bit start pc offset by  default, and hence see the 
> R_X86_64_PC64 relas.
> 
> The proposed patch fixes ld to not complain about R_X86_64_PC64 in x32 
> objects.  IOW, with this patch, users will have a way to create .sframe 
> sections in x32 objects.  FWIW, GAS does not generate .sframe for x32. 
> But users could have used objcopy and then ld even in previous GNU 
> Binutils (which supported SFrame V2) to create SFrame V2 sections (which 
> had R_X86_64_PC32) for x32 binaries..
> 
> Mathieu, I am reminded of a discussion with you sometime ago regarding a 
> concern that if there is a way to create .sframe sections for x32 
> binaries, it may expose some vulnerable code paths in the Linux kernel 
> stacktracer/SFrame related bits.  Can you remind what the issue was/is ?
> 
> In general, I am wondering what is the right thing to do here:
>    - restrict the kernel's linker scripts to discard .sframe for x32 
> objects or,
>    - should 'objcopy -O elf32-x86-64' pick one of the behaviors:
>      + silently skip .sframe/.rela.sframe, or
>      + alert the user with non-fatal error and not 
> copy .sframe/.rela.sframe
> 

In the SFrame meeting earlier this week, we talked a bit about this PR.

As for the vulnerable code paths related x32, perhaps I had been 
misremembering something.  Either way, it is acceptable to expect the 
associated kernel script vdsox32.lds be adapted to discard SFrame 
section for x32 binary as necessary.

Summary: we can move forward with HJ's patch.