Re: [PATCH 02/24] x86/traps: Move arch/x86/kernel/traps.c to arch/x86/entry/

[Martin Liška]

On 9/7/21 23:12, Nick Desaulniers wrote:
> On Fri, Sep 3, 2021 at 12:36 AM Martin Liška <mliska@xxxxxxx> wrote:
> > On 9/2/21 19:05, Nick Desaulniers wrote:
> > > IIRC GCC only
> > > added the attribute recently in the 10.X release, so it might be too
> > > new to rely on quite yet.
> >
> > The no_stack_protector attribute was actually added in the GCC 11.x release:
> > https://gcc.gnu.org/gcc-11/changes.html
>
> Ah right, that lays more weight though that this feature is still too
> new to rely on quite yet.

Sure.

> Martin, do you know if what happens with
> regards to inlining when the callee and caller mismatch on this
> function attribute in GCC?  This is very much a problem for the
> kernel.

That's a know issue that was already discusses both in a Kernel mailing and
GCC bugzilla:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94722#c9

Martin

> > Note the compiler is definitely used by Fedora, openSUSE Tumbleweed
> > and other cutting edge distributions.
>
> Kernel supports GCC 4.9+ currently. This feature can only be emulated
> with the coarse grain -fno-stack-protector (or gnu_inline with out of
> line assembler definition...:( ).