Re: [PATCH] powerpc: remove checks for binutils older than 2.25

[Nicholas Piggin]

On Mon Jun 26, 2023 at 1:11 PM AEST, Masahiro Yamada wrote:
> On Mon, Jun 26, 2023 at 10:58 AM Nicholas Piggin <npiggin@xxxxxxxxx> wrote:
> >
> > On Sun Jun 25, 2023 at 6:02 PM AEST, Masahiro Yamada wrote:
> > > On Thu, Jan 19, 2023 at 9:37 PM Masahiro Yamada <masahiroy@xxxxxxxxxx> wrote:
> > > >
> > > > On Thu, Jan 19, 2023 at 9:12 PM Joel Stanley <joel@xxxxxxxxx> wrote:
> > > > >
> > > > > On Thu, 19 Jan 2023 at 08:24, Masahiro Yamada <masahiroy@xxxxxxxxxx> wrote:
> > > > > >
> > > > > > Commit e4412739472b ("Documentation: raise minimum supported version of
> > > > > > binutils to 2.25") allows us to remove the checks for old binutils.
> > > > > >
> > > > > > There is no more user for ld-ifversion. Remove it as well.
> > > > >
> > > > > ppc kernels fail to link with 2.27 under some configurations:
> > > > >
> > > > >  https://github.com/linuxppc/issues/issues/388
> > > > >
> > > > > We may want to use ld-ifversion to exclude that version.
> > > >
> > > >
> > >
> > >
> > > Ping?
> > >
> > >
> > >
> > >
> > > >
> > > > For LLD, CONFIG option is directly checked.
> >
> > Yeah, doesn't seem too difficult to add new linker version tests if needed.
> >
> > > >
> > > >
> > > > masahiro@zoe:~/ref/linux(master)$ git grep  CONFIG_LLD_VERSION
> > > > Makefile:ifeq ($(call test-lt, $(CONFIG_LLD_VERSION), 130000),y)
> > > > arch/riscv/Makefile:ifeq ($(call test-lt, $(CONFIG_LLD_VERSION), 150000),y)
> > > > arch/x86/Makefile:ifeq ($(call test-lt, $(CONFIG_LLD_VERSION), 130000),y)
> > > > scripts/Kbuild.include:# Usage: $(call test-lt, $(CONFIG_LLD_VERSION), 150000)
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > > >
> > > > > > Signed-off-by: Masahiro Yamada <masahiroy@xxxxxxxxxx>
> > > > > > ---
> > > > > >
> > > > > >  arch/powerpc/Makefile     | 22 +---------------------
> > > > > >  arch/powerpc/lib/Makefile |  2 +-
> > > > > >  scripts/Makefile.compiler |  4 ----
> > > > > >  3 files changed, 2 insertions(+), 26 deletions(-)
> > > > > >
> > > > > > diff --git a/arch/powerpc/Makefile b/arch/powerpc/Makefile
> > > > > > index dc4cbf0a5ca9..3d265b16c0ae 100644
> > > > > > --- a/arch/powerpc/Makefile
> > > > > > +++ b/arch/powerpc/Makefile
> > > > > > @@ -42,18 +42,13 @@ machine-$(CONFIG_PPC64) += 64
> > > > > >  machine-$(CONFIG_CPU_LITTLE_ENDIAN) += le
> > > > > >  UTS_MACHINE := $(subst $(space),,$(machine-y))
> > > > > >
> > > > > > -# XXX This needs to be before we override LD below
> > > > > > -ifdef CONFIG_PPC32
> > > > > > -KBUILD_LDFLAGS_MODULE += arch/powerpc/lib/crtsavres.o
> > > > > > -else
> > > > > > -ifeq ($(call ld-ifversion, -ge, 22500, y),y)
> > > > > > +ifeq ($(CONFIG_PPC64)$(CONFIG_LD_IS_BFD),yy)
> > > > > >  # Have the linker provide sfpr if possible.
> > > > > >  # There is a corresponding test in arch/powerpc/lib/Makefile
> > > > > >  KBUILD_LDFLAGS_MODULE += --save-restore-funcs
> > > > > >  else
> > > > > >  KBUILD_LDFLAGS_MODULE += arch/powerpc/lib/crtsavres.o
> > > > > >  endif
> > > > > > -endif
> > > > > >
> > > > > >  ifdef CONFIG_CPU_LITTLE_ENDIAN
> > > > > >  KBUILD_CFLAGS  += -mlittle-endian
> > > > > > @@ -389,8 +384,6 @@ vdso_prepare: prepare0
> > > > > >                 $(build)=arch/powerpc/kernel/vdso include/generated/vdso64-offsets.h)
> > > > > >  endif
> > > > > >
> > > > > > -archprepare: checkbin
> > > > > > -
> > > > > >  archheaders:
> > > > > >         $(Q)$(MAKE) $(build)=arch/powerpc/kernel/syscalls all
> > > > > >
> > > > > > @@ -405,16 +398,3 @@ else
> > > > > >         $(eval KBUILD_CFLAGS += -mstack-protector-guard-offset=$(shell awk '{if ($$2 == "TASK_CANARY") print $$3;}' include/generated/asm-offsets.h))
> > > > > >  endif
> > > > > >  endif
> > > > > > -
> > > > > > -PHONY += checkbin
> > > > > > -# Check toolchain versions:
> > > > > > -# - gcc-4.6 is the minimum kernel-wide version so nothing required.
> > > > > > -checkbin:
> > > > > > -       @if test "x${CONFIG_LD_IS_LLD}" != "xy" -a \
> > > > > > -               "x$(call ld-ifversion, -le, 22400, y)" = "xy" ; then \
> > > > > > -               echo -n '*** binutils 2.24 miscompiles weak symbols ' ; \
> > > > > > -               echo 'in some circumstances.' ; \
> > > > > > -               echo    '*** binutils 2.23 do not define the TOC symbol ' ; \
> > > > > > -               echo -n '*** Please use a different binutils version.' ; \
> > > > > > -               false ; \
> > > > > > -       fi
> > > > > > diff --git a/arch/powerpc/lib/Makefile b/arch/powerpc/lib/Makefile
> > > > > > index 4de71cbf6e8e..c53618c34b70 100644
> > > > > > --- a/arch/powerpc/lib/Makefile
> > > > > > +++ b/arch/powerpc/lib/Makefile
> > > > > > @@ -42,7 +42,7 @@ obj-$(CONFIG_FUNCTION_ERROR_INJECTION)        += error-inject.o
> > > > > >  # 64-bit linker creates .sfpr on demand for final link (vmlinux),
> > > > > >  # so it is only needed for modules, and only for older linkers which
> > > > > >  # do not support --save-restore-funcs
> > > > > > -ifeq ($(call ld-ifversion, -lt, 22500, y),y)
> > > > > > +ifndef CONFIG_LD_IS_BFD
> > > > > >  extra-$(CONFIG_PPC64)  += crtsavres.o
> > > > > >  endif
> >
> > This test got inverted for LLD now AFAIKS?
>
>
> In my understanding,
> For a BFD linker, use --save-restore-funcs.
> Otherwise (i.e. CONFIG_LD_IS_BFD is unset), link crtsavres.o to modules.
>
> I hope I did not change the logic.

I think I misread it, ignore me.
>
>
>
> >
> > Does LLVM support --save-restore-funcs and supply .sfpr already I
> > wonder? We could remove this stuff entirely.
>
> I don't know.
>
> If LLVM 11.0.0  (the minimum supported LLVM version) supports
> --save-restore-funcs, you can remove this check entirely.
> Even so, it should be done in a follow-up patch.

True. I don't have an lld < 14 to test, but 14 does accept the option
and emits the functions if needed. But my clang 14 does not even
generate calls to them so I assume earlier ones do not either. So we
might be able to get rid of this stuff. I'll open an issue for it.

> This patch is just dropping dead code for BFD linker < 2.25,
> which is now unsupported. Nothing else has changed (I believe).

Seems okay then.

Reviewed-by: Nicholas Piggin <npiggin@xxxxxxxxx>

Thanks,
Nick