Re: [PATCH v3 08/19] x86/head64: Replace pointer fixups with PIE codegen

[Ard Biesheuvel <ardb@kernel.org>]

On Mon, 12 Feb 2024 at 11:29, Borislav Petkov <bp@alien8.de> wrote:
>
> On Mon, Jan 29, 2024 at 07:05:11PM +0100, Ard Biesheuvel wrote:
> > From: Ard Biesheuvel <ardb@kernel.org>
> >
> > Some of the C code in head64.c may be called from a different virtual
> > address than it was linked at. Currently, we deal with this by using
>
> Yeah, make passive pls: "Currently, this is done by using... "
>
> > ordinary, position dependent codegen, and fixing up all symbol
> > references on the fly. This is fragile and tricky to maintain. It is
> > also unnecessary: we can use position independent codegen (with hidden
>                    ^^^
> Ditto: "use ..."
>
> In the comments below too, pls, where it says "we".
>

Ack.

> > visibility) to ensure that all compiler generated symbol references are
> > RIP-relative, removing the need for fixups entirely.
> >
> > It does mean we need explicit references to kernel virtual addresses to
> > be generated by hand, so generate those using a movabs instruction in
> > inline asm in the handful places where we actually need this.
> >
> > Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
> > ---
> >  arch/x86/Makefile                 |  8 ++
> >  arch/x86/boot/compressed/Makefile |  2 +-
> >  arch/x86/include/asm/desc.h       |  3 +-
> >  arch/x86/include/asm/setup.h      |  4 +-
> >  arch/x86/kernel/Makefile          |  5 ++
> >  arch/x86/kernel/head64.c          | 88 +++++++-------------
> >  arch/x86/kernel/head_64.S         |  5 +-
> >  7 files changed, 51 insertions(+), 64 deletions(-)
> >
> > diff --git a/arch/x86/Makefile b/arch/x86/Makefile
> > index 1a068de12a56..2b5954e75318 100644
> > --- a/arch/x86/Makefile
> > +++ b/arch/x86/Makefile
> > @@ -168,6 +168,14 @@ else
> >          KBUILD_CFLAGS += -mcmodel=kernel
> >          KBUILD_RUSTFLAGS += -Cno-redzone=y
> >          KBUILD_RUSTFLAGS += -Ccode-model=kernel
> > +
> > +     PIE_CFLAGS-$(CONFIG_STACKPROTECTOR)     += -fno-stack-protector
>
> Main Makefile has
>
> KBUILD_CFLAGS += -fno-PIE
>
> and this ends up being:
>
> gcc -Wp,-MMD,arch/x86/kernel/.head64.s.d -nostdinc ... -fno-PIE ... -fpie ... -fverbose-asm -S -o arch/x86/kernel/head64.s arch/x86/kernel/head64.c
>
> Can you pls remove -fno-PIE from those TUs which use PIE_CFLAGS so that
> there's no confusion when staring at V=1 output?
>

Yeah. That would means adding PIE_CFLAGS_REMOVE alongside PIE_CFLAGS
and applying both in every place it is used, but we are only dealing
with a handful of object files here.


> > +     PIE_CFLAGS-$(CONFIG_LTO)                += -fno-lto
> > +
> > +     PIE_CFLAGS := -fpie -mcmodel=small $(PIE_CFLAGS-y) \
> > +                   -include $(srctree)/include/linux/hidden.h
> > +
> > +     export PIE_CFLAGS
> >  endif
> >
> >  #
>
> Other than that, that code becomes much more readable, cool!
>

Thanks. But now that we have RIP_REL_REF(), I might split the cleanup
from the actual switch to -fpie, which I am still a bit on the fence
about, given different compiler versions, LTO, etc.

RIP_REL_REF(foo) just turns into 'foo' when compiling with -fpie and
we could drop those piecemeal once we are confident that -fpie does
not cause any regressions.

Note that I have some reservations now about .pi.text as well: it is a
bit intrusive, and on x86, we might just as well move everything that
executes from the 1:1 mapping into .head.text, and teach objtool that
those sections should not contain any ELF relocations involving
absolute addresses. But this is another thing that I want to spend a
bit more time on before I respin it, so I will just do the cleanup in
the next revision, and add the rigid correctness checks the next
cycle.