[PATCH v2 2/2] arm64/build: Warn on orphan section placement

[Ard Biesheuvel]

On Tue, 23 Jun 2020 at 16:52, Will Deacon <will at kernel.org> wrote:
>
> On Mon, Jun 22, 2020 at 01:58:15PM -0700, Kees Cook wrote:
> > We don't want to depend on the linker's orphan section placement
> > heuristics as these can vary between linkers, and may change between
> > versions. All sections need to be explicitly named in the linker
> > script.
> >
> > Explicitly include debug sections when they're present. Add .eh_frame*
> > to discard as it seems that these are still generated even though
> > -fno-asynchronous-unwind-tables is being specified. Add .plt and
> > .data.rel.ro to discards as they are not actually used. Add .got.plt
> > to the image as it does appear to be mapped near .data. Finally enable
> > orphan section warnings.
>
> Can you elaborate a bit on what .got.plt is being used for, please? I
> wonder if there's an interaction with an erratum workaround in the linker
> or something.
>

.got.plt is not used at all, but it has three magic entries at the
start that the dynamic linker uses for lazy dispatch, so it turns up
as a non-empty section of 0x18 bytes.

We should be able to discard it afaict, but given that it does not
actually take up any space, it doesn't really matter either way.

> > diff --git a/arch/arm64/Makefile b/arch/arm64/Makefile
> > index a0d94d063fa8..3e628983445a 100644
> > --- a/arch/arm64/Makefile
> > +++ b/arch/arm64/Makefile
> > @@ -29,6 +29,10 @@ LDFLAGS_vmlinux    += --fix-cortex-a53-843419
> >    endif
> >  endif
> >
> > +# We never want expected sections to be placed heuristically by the
> > +# linker. All sections should be explicitly named in the linker script.
> > +LDFLAGS_vmlinux += --orphan-handling=warn
> > +
> >  ifeq ($(CONFIG_ARM64_USE_LSE_ATOMICS), y)
> >    ifneq ($(CONFIG_ARM64_LSE_ATOMICS), y)
> >  $(warning LSE atomics not supported by binutils)
> > diff --git a/arch/arm64/kernel/vmlinux.lds.S b/arch/arm64/kernel/vmlinux.lds.S
> > index 5427f502c3a6..c9ecb3b2007d 100644
> > --- a/arch/arm64/kernel/vmlinux.lds.S
> > +++ b/arch/arm64/kernel/vmlinux.lds.S
> > @@ -94,7 +94,8 @@ SECTIONS
> >       /DISCARD/ : {
> >               *(.interp .dynamic)
> >               *(.dynsym .dynstr .hash .gnu.hash)
> > -             *(.eh_frame)
> > +             *(.plt) *(.data.rel.ro)
> > +             *(.eh_frame) *(.init.eh_frame)
>
> Do we need to include .eh_frame_hdr here too?
>

It would be better to build with -fno-unwind-tables, in which case
these sections should not even exist.

> >       }
> >
> >       . = KIMAGE_VADDR + TEXT_OFFSET;
> > @@ -209,6 +210,7 @@ SECTIONS
> >       _data = .;
> >       _sdata = .;
> >       RW_DATA(L1_CACHE_BYTES, PAGE_SIZE, THREAD_ALIGN)
> > +     .got.plt : ALIGN(8) { *(.got.plt) }
> >
> >       /*
> >        * Data written with the MMU off but read with the MMU on requires
> > @@ -244,6 +246,7 @@ SECTIONS
> >       _end = .;
> >
> >       STABS_DEBUG
> > +     DWARF_DEBUG
>
> I know you didn't add it, but do we _really_ care about stabs debug? Who
> generates that for arm64?
>
> Will