Received-SPF: pass (google.com: domain of linux-kernel+bounces-139542-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Precedence: bulk
MIME-Version: 1.0
References: <20240411-as-instr-opt-wrussq-v1-1-99b1a1a99f93@gmail.com>
In-Reply-To: <20240411-as-instr-opt-wrussq-v1-1-99b1a1a99f93@gmail.com>
From: Dmitry Safonov <0x7f454c46@gmail.com>
Date: Thu, 11 Apr 2024 00:41:27 +0100
Message-ID: <CAJwJo6aBbw96uo_eiDuKP5nX4EsoSYkEmaKjdMY01_Pc2z-1DA@mail.gmail.com>
Subject: Re: [PATCH] kbuild/x86: Use $(KBUILD_AFLAGS) in Kbuild's version of $(as-instr)
To: 0x7f454c46@gmail.com
Cc: Borislav Petkov <bp@alien8.de>, Dave Hansen <dave.hansen@linux.intel.com>, 
	Ingo Molnar <mingo@redhat.com>, "H. Peter Anvin" <hpa@zytor.com>, Masahiro Yamada <masahiroy@kernel.org>, 
	Thomas Gleixner <tglx@linutronix.de>, x86@kernel.org, linux-kernel@vger.kernel.org, 
	linux-kbuild@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"

Please, ignore this version.

the previous one was to add a helper function and that seemed to work,
but on rework to actually use KBUILD_AFLAGS instead of adding a new
function I certainly haven't tested it enough.
Going to prepare version 2, sorry for the noise.

On Thu, 11 Apr 2024 at 00:24, Dmitry Safonov via B4 Relay
<devnull+0x7f454c46.gmail.com@kernel.org> wrote:
>
> From: Dmitry Safonov <0x7f454c46@gmail.com>
>
> At Arista some products use compatible 32-bit userspace running on x86.
> As a part of disto build for ia32 it also compiles the 64-bit kernel.
> While the toolchain for the kernel build is yet the same, with 64-bit gcc:
> > / @Bru-na-Boinne% file /usr/bin/gcc-11
> > /usr/bin/gcc-11: ELF 64-bit LSB pie executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, BuildID[sha1]=6571ad50d8f12eece053f1bac7a95a2c767f32c9, for GNU/Linux 3.2.0, stripped
>
> It seems that gcc is being smart and detects that it's running in
> a 32-bit container (personality flag? 32-bit mmap base? something else
> inherited post-exec?  haven't yet figured it out) and by default tries
> to build 32-bit binaries.
>
> That results in a failing toolchain check:
> > / @Bru-na-Boinne% printf "%b\n" "wrussq %rax, (%rbx)" | /usr/bin/gcc-11 -Wa,--fatal-warnings -c -x assembler-with-cpp -o /dev/null -
> > <stdin>: Assembler messages:
> > <stdin>:1: Error: `wrussq' is only supported in 64-bit mode
>
> Which passes when -m64 is directly specify for the build check:
> > / @Bru-na-Boinne% printf "%b\n" "wrussq %rax, (%rbx)" | /usr/bin/gcc-11 -m64 -Wa,--fatal-warnings -c -x assembler-with-cpp -o /dev/null -
> > / @Bru-na-Boinne% echo $?
> > 0
>
> As a result, kbuild produces different value for CONFIG_AS_WRUSS
> for native 64-bit containers and ia32 containers with 64-bit gcc,
> which produces different kernels with enabled/disabled
> CONFIG_X86_USER_SHADOW_STACK.
>
> arch/x86/Makefile already properly defines KBUILD_AFLAGS += -m64,
> which is luckly already available at the point of toolchain check
> in arch/x86/Kconfig.assembler
>
> By hacking around Kbuild variable the following way:
> > --- a/scripts/Kconfig.include
> > +++ b/scripts/Kconfig.include
> > @@ -13,7 +13,8 @@ left_paren  := (
> >
> >  # $(if-success,<command>,<then>,<else>)
> >  # Return <then> if <command> exits with 0, <else> otherwise.
> > -if-success = $(shell,{ $(1); } >/dev/null 2>&1 && echo "$(2)" || echo "$(3)")
> > +if-success = $(shell,echo '$(1)' 1>&2;{ $(1); } >/dev/null 2>&1 && echo "$(2)" || echo "$(3)")
>
> I got the following output for the toolchain check, before:
> > linux @Bru-na-Boinne% make ARCH=x86_64 oldconfig V=1 2>&1 | grep wrus
> > printf "%b\n" "wrussq %rax,(%rbx)" | gcc  -c -x assembler-with-cpp -o /dev/null -
>
> and after:
> > linux @Bru-na-Boinne% make ARCH=x86_64 oldconfig V=1 2>&1 | grep wrus
> > printf "%b\n" "wrussq %rax,(%rbx)" | gcc -D__ASSEMBLY__ -fno-PIE -m64 -c -x assembler-with-cpp -o /dev/null -
>
> Which seems appropriate to me.
> This also reflects the existing definition in scripts/Makefile.compiler
> for $(as-instr) that already has $(KBUILD_AFLAGS).
>
> Signed-off-by: Dmitry Safonov <0x7f454c46@gmail.com>
> ---
>  scripts/Kconfig.include | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/scripts/Kconfig.include b/scripts/Kconfig.include
> index 3ee8ecfb8c04..d8574c1faf2d 100644
> --- a/scripts/Kconfig.include
> +++ b/scripts/Kconfig.include
> @@ -33,7 +33,7 @@ ld-option = $(success,$(LD) -v $(1))
>
>  # $(as-instr,<instr>)
>  # Return y if the assembler supports <instr>, n otherwise
> -as-instr = $(success,printf "%b\n" "$(1)" | $(CC) $(CLANG_FLAGS) -Wa$(comma)--fatal-warnings -c -x assembler-with-cpp -o /dev/null -)
> +as-instr = $(success,printf "%b\n" "$(1)" | $(CC) $(CLANG_FLAGS) $(KBUILD_AFLAGS) -Wa$(comma)--fatal-warnings -c -x assembler-with-cpp -o /dev/null -)
>
>  # check if $(CC) and $(LD) exist
>  $(error-if,$(failure,command -v $(CC)),C compiler '$(CC)' not found)
>
> ---
> base-commit: 2c71fdf02a95b3dd425b42f28fd47fb2b1d22702
> change-id: 20240410-as-instr-opt-wrussq-48ec37e36204
>
> Best regards,
> --
> Dmitry Safonov <0x7f454c46@gmail.com>
>
>