Re: [PATCH] um: Allow builds with Clang

[Masahiro Yamada]

On Fri, Feb 18, 2022 at 2:41 AM Nathan Chancellor <nathan@xxxxxxxxxx> wrote:
>
> On Thu, Feb 17, 2022 at 01:54:58PM +0900, Masahiro Yamada wrote:
> > On Thu, Feb 17, 2022 at 9:28 AM Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> > >
> > > Add x86-64 target for Clang+um and update user-offsets.c to use
> > > Clang-friendly assembler, similar to the fix from commit cf0c3e68aa81
> > > ("kbuild: fix asm-offset generation to work with clang").
> > >
> > > This lets me run KUnit tests with Clang:
> > >
> > > $ ./tools/testing/kunit/kunit.py config --make_options LLVM=1
> > > ...
> > > $ ./tools/testing/kunit/kunit.py run --make_options LLVM=1
> > > ...
> > >
> > > Cc: Jeff Dike <jdike@xxxxxxxxxxx>
> > > Cc: Richard Weinberger <richard@xxxxxx>
> > > Cc: Anton Ivanov <anton.ivanov@xxxxxxxxxxxxxxxxxx>
> > > Cc: Masahiro Yamada <masahiroy@xxxxxxxxxx>
> > > Cc: Nick Desaulniers <ndesaulniers@xxxxxxxxxx>
> > > Cc: Nathan Chancellor <nathan@xxxxxxxxxx>
> > > Cc: David Gow <davidgow@xxxxxxxxxx>
> > > Cc: linux-um@xxxxxxxxxxxxxxxxxxx
> > > Cc: linux-kbuild@xxxxxxxxxxxxxxx
> > > Cc: linux-kselftest@xxxxxxxxxxxxxxx
> > > Cc: kunit-dev@xxxxxxxxxxxxxxxx
> > > Cc: llvm@xxxxxxxxxxxxxxx
> > > Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
> > > ---
> > >  arch/x86/um/user-offsets.c | 4 ++--
> > >  scripts/Makefile.clang     | 1 +
> > >  2 files changed, 3 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/arch/x86/um/user-offsets.c b/arch/x86/um/user-offsets.c
> > > index bae61554abcc..d9071827b515 100644
> > > --- a/arch/x86/um/user-offsets.c
> > > +++ b/arch/x86/um/user-offsets.c
> > > @@ -10,10 +10,10 @@
> > >  #include <asm/types.h>
> > >
> > >  #define DEFINE(sym, val) \
> > > -       asm volatile("\n->" #sym " %0 " #val : : "i" (val))
> > > +       asm volatile("\n.ascii \"->" #sym " %0 " #val "\"": : "i" (val))
> > >
> > >  #define DEFINE_LONGS(sym, val) \
> > > -       asm volatile("\n->" #sym " %0 " #val : : "i" (val/sizeof(unsigned long)))
> > > +       asm volatile("\n.ascii \"->" #sym " %0 " #val "\"": : "i" (val/sizeof(unsigned long)))
> > >
> > >  void foo(void)
> > >  {
> > > diff --git a/scripts/Makefile.clang b/scripts/Makefile.clang
> > > index 51fc23e2e9e5..857b23de51c6 100644
> > > --- a/scripts/Makefile.clang
> > > +++ b/scripts/Makefile.clang
> > > @@ -10,6 +10,7 @@ CLANG_TARGET_FLAGS_powerpc    := powerpc64le-linux-gnu
> > >  CLANG_TARGET_FLAGS_riscv       := riscv64-linux-gnu
> > >  CLANG_TARGET_FLAGS_s390                := s390x-linux-gnu
> > >  CLANG_TARGET_FLAGS_x86         := x86_64-linux-gnu
> > > +CLANG_TARGET_FLAGS_um          := x86_64-linux-gnu
> >
> >
> > Does this work for the i386 host?
> >
> > UML supports i386 and x86_64 as the host architecture as of now,
> > but this always compiles UML for x86_64?
>
> I think the current code will work because arch/x86/Makefile.um includes
> -m32 for CONFIG_X86_32, which will implicitly change x86_64-linux-gnu
> into a 32-bit target triple:

Ah, you are right!


>
> $ echo | clang --target=x86_64-linux-gnu -x c -c -o test.o -
>
> $ file test.o
> test.o: ELF 64-bit LSB relocatable, x86-64, version 1 (SYSV), not stripped
>
> $ echo | clang --target=x86_64-linux-gnu -m32 -x c -c -o test.o -
>
> $ file test.o
> test.o: ELF 32-bit LSB relocatable, Intel 80386, version 1 (SYSV), not stripped
>
> In fact, we rely on this for ARCH=i386 LLVM=1 right now, as it uses
> x86_64-linux-gnu for the target flag.
>
> While UML only supports x86, maybe it is worth using SUBARCH instead of
> hardcoding the triple? No strong opinion around that though.
>
> diff --git a/scripts/Makefile.clang b/scripts/Makefile.clang
> index 51fc23e2e9e5..87285b76adb2 100644
> --- a/scripts/Makefile.clang
> +++ b/scripts/Makefile.clang
> @@ -10,6 +10,7 @@ CLANG_TARGET_FLAGS_powerpc    := powerpc64le-linux-gnu
>  CLANG_TARGET_FLAGS_riscv       := riscv64-linux-gnu
>  CLANG_TARGET_FLAGS_s390                := s390x-linux-gnu
>  CLANG_TARGET_FLAGS_x86         := x86_64-linux-gnu
> +CLANG_TARGET_FLAGS_um          := $(CLANG_TARGET_FLAGS_$(SUBARCH))


LGTM.

I also thought of not passing --target at all for ARCH=um, but
we decided to override --target all the time (for reproducibility?).
Anyway, Nathan's way is clean, and looks OK to me.







>  CLANG_TARGET_FLAGS             := $(CLANG_TARGET_FLAGS_$(SRCARCH))
>
>  ifeq ($(CROSS_COMPILE),)
>
> > >  CLANG_TARGET_FLAGS             := $(CLANG_TARGET_FLAGS_$(SRCARCH))
> > >
> > >  ifeq ($(CROSS_COMPILE),)
> > > --
> > > 2.30.2
> > >
> >
> >
> > --
> > Best Regards
> > Masahiro Yamada



--
Best Regards
Masahiro Yamada