On Sat, Jun 24, 2023 at 5:45 AM Sami Tolvanen <samitolvanen@xxxxxxxxxx> wrote:
>
> Hi Pierre-Clément,
>
> On Fri, Jun 23, 2023 at 7:08 AM Pierre-Clément Tosi <ptosi@xxxxxxxxxx> wrote:
> >
> > The (relatively) new KCFI feature in LLVM/Clang encodes type information
> > for C functions by generating symbols named __kcfi_typeid_<fname>, which
> > can then be referenced from assembly. However, some custom build rules
> > (e.g. EFI, nVHE, or early PIE on arm64) use objcopy to add a prefix to
> > all the symbols in their object files, making mksysmap's ignore filter
> > miss those KCFI symbols.
> >
> > Therefore, explicitly list those twice-prefixed KCFI symbols as ignored.
> >
> > Alternatively, this could also be achieved in a less verbose way by
> > ignoring any symbol containing the string "__kcfi_typeid_". However,
> > listing the combined prefixes explicitly saves us from running the small
> > risk of ignoring symbols that should be kept.
> >
> > Signed-off-by: Pierre-Clément Tosi <ptosi@xxxxxxxxxx>
> > ---
> > scripts/mksysmap | 3 +++
> > 1 file changed, 3 insertions(+)
> >
> > diff --git a/scripts/mksysmap b/scripts/mksysmap
> > index 26f39772f7a5..17cf4292e26b 100755
> > --- a/scripts/mksysmap
> > +++ b/scripts/mksysmap
> > @@ -61,7 +61,10 @@ ${NM} -n ${1} | sed >${2} -e "
> > / __microLA25Thunk_/d
> >
> > # CFI type identifiers
> > +/ __efistub___kcfi_typeid_/d
>
> Does the existing __efistub_/d rule not catch this?
Agree with Sami.
This line looks redundant to me.
>
> > / __kcfi_typeid_/d
> > +/ __kvm_nvhe___kcfi_typeid_/d
> > +/ __pi___kcfi_typeid_/d
>
> Either way, otherwise this looks reasonable to me. Thanks for the patch!
>
> Reviewed-by: Sami Tolvanen <samitolvanen@xxxxxxxxxx>
>
> Sami
--
Best Regards
Masahiro Yamada
