Hi Christophe,
On Thu, Aug 14, 2025 at 03:43:09PM +0200, Christophe Leroy wrote:
> Le 12/08/2025 à 07:44, Thomas Weißschuh a écrit :
> > The compiler can emit absolute relocations in vDSO code,
> > which are invalid in vDSO code.
> > Detect them at compile-time.
>
> I'm a bit puzzled with this series.
>
> If I understand correctly, the check will be done only when you have RUST
> available ?
Yes, this new check will only be performed if a rust toolchain is available.
CONFIG_RUST however is *not* required.
> I wouldn't expect having RUST to build a C kernel.
The build will work fine without Rust present and will fall back to the
simplistic readelf test. A single report of breakage will allow us to fix the code,
not everybody needs to run the full thing.
> By the way, aren't relocations already detected by command cmd_vdso_check in
> lib/vdso/Makefile.include , using readelf ? Why is a new tool needed
The current cmd_vdso_check only validates the final vDSO image.
However that is not sufficient, as some problematic relocations will not show
up in the final image anymore but only the intermediary object files.
And there the logic is more complex than can be reasonably expressed in inline
shell scripts, see the previous revisions of this series for the attempts.
The valid relocations depend on each architecture and the specific ELF section
they appear in.
For the real example that triggered all of this, see commit
0c314cda9325 ("arm64: vdso: Work around invalid absolute relocations from GCC")
> and why does it have to be written in RUST langage ?
There is no hard requirement for Rust. I chose it for convenience of
implementation, especially around descriptive error handling and generic
functions. tglx was fine with it.
Thomas