[PATCH v3 0/3] sframe: Disable by default and support SFRAME_F_FDE_FUNC_START_PCREL

[Adhemerval Zanella Netto]

On 24/07/25 12:58, Andreas K. Huettel wrote:
> Am Mittwoch, 23. Juli 2025, 15:47:22 Mitteleuropäische Sommerzeit schrieb Adhemerval Zanella:
>> The SFrame prior binutils 2.45 is unstable on multiple architectures,
>> so upcoming binutils 2.45 should be considered the minimum required
>> version for SFrame backtracing. This version also includes an erratum,
>> SFRAME_F_FDE_FUNC_START_PCREL, which is not currently supported on
>> glibc. SFrame with this flag is currently ignored by glibc SFrame
>> handling.
>>
>> Due to the experimental nature, the first patch disables SFrame by
>> default, adds extra checks to enable iff the architecture's explicit
>> enablement is on glibc, and fixes some other build issues.
>>
>> The second patch adds support for the SFRAME_F_FDE_FUNC_START_PCREL
>> erratum, which adds SFrame support for binaries built with binutils
>> 2.45.
> 
> OK so I gave patch 1 another read and tested...
> 
> * x86-64, binutils pre-2.45: 
>    - no configure parameter, builds and tests fine with no indications of sframe
>    - --enable-sframe, builds and tests fine with appropriate bits in the logs
> 
> * armv4tl, binutils 2.44: 
>    - no configure parameter, builds and tests (mostly, nspawn) fine
>    - --enable-sframe, errors out with architecture not supported (good)
> 
> * aarch64, binutils 2.44:
>    - no configure parameter, builds and tests fine
>    - --enable-sframe, errors out with too old binutils (good)
> 
> Let's push it.
> 

Alright, I will install them.

> This is hopefully the last real patch, I'll start with the release
> afterwards.
> 
I think only missing patch that we should install is the termios revert
to avoid the gcc and other build breakages [1]

[1] https://sourceware.org/pipermail/libc-alpha/2025-July/168760.html