Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Tri Vo <trong@android.com>
To:     x86@kernel.org, tglx@linutronix.de, mingo@redhat.com, bp@alien8.de,
        hpa@zytor.com
Cc:     grimar@accesssoftek.com, dima@golovin.in, morbo@google.com,
        ndesaulniers@google.com, matz@suse.de, ruiu@google.com,
        linux-kernel@vger.kernel.org, Tri Vo <trong@android.com>
Subject: [PATCH v3] x86_64: Add "-m elf_i386" when linking i386 object files.
Date:   Fri, 11 Jan 2019 12:10:12 -0800
Message-Id: <20190111201012.71210-1-trong@android.com>
In-Reply-To: <CAKwvOdkR=1AzePbuFBcCW32shAK_cvR1iPiTGqHF=7MRH_oAjA@mail.gmail.com>
References: <CAKwvOdkR=1AzePbuFBcCW32shAK_cvR1iPiTGqHF=7MRH_oAjA@mail.gmail.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

From: George Rimar <grimar@accesssoftek.com>

Linux kernel uses OUTPUT_FORMAT in it's linker scripts. Most of the time
-m option is passed to the linker with correct architecture, but
sometimes (at least for x86_64) the -m option contradicts OUTPUT_FORMAT
directive. Specifically, arch/x86/boot and arch/x86/realmode/rm modules
have i386 object files, but are linked with -m elf_x86_64 linker flag
when building for x86_64.

"man ld" doesn't explicitly state any tie-breakers between -m and
OUTPUT_FORMAT. BFD and Gold linkers override -m value with
OUTPUT_FORMAT. But LLVM lld has a different behavior. When supplied with
contradicting -m and OUTPUT_FORMAT values it fails with the following
error message:

  ld.lld: error: arch/x86/realmode/rm/header.o is incompatible with elf_x86_64

Suggested fix: just add correct -m after incorrect one (it overrides
it), so the linker invocation looks like this: ld -m elf_x86_64 -z
max-page-size=0x200000 -m elf_i386 --emit-relocs -T realmode.lds
header.o trampoline_64.o stack.o reboot.o -o realmode.elf

This is not a functional change for GNU ld, because (although not
explicitly documented) it already overrides -m EMULATION with
OUTPUT_FORMAT.

Tested by building x86_64 kernel with GNU gcc/ld toolchain and booting
it in QEMU.

Suggested-by: Dmitry Golovin <dima@golovin.in>
Signed-off-by: George Rimar <grimar@accesssoftek.com>
Signed-off-by: Tri Vo <trong@android.com>
Tested-by: Tri Vo <trong@android.com>
Tested-by: Nick Desaulniers <ndesaulniers@google.com>
---
v2: updated commit message to clarify that ld documentation is ambiguous w.r.t
    -m vs OUTPUT_FORMAT behavior.
v3: fixed/added SOB and "Tested-by" fields.

 arch/x86/boot/Makefile        | 2 +-
 arch/x86/realmode/rm/Makefile | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86/boot/Makefile b/arch/x86/boot/Makefile
index 9b5adae9cc40..e2839b5c246c 100644
--- a/arch/x86/boot/Makefile
+++ b/arch/x86/boot/Makefile
@@ -100,7 +100,7 @@ $(obj)/zoffset.h: $(obj)/compressed/vmlinux FORCE
 AFLAGS_header.o += -I$(objtree)/$(obj)
 $(obj)/header.o: $(obj)/zoffset.h

-LDFLAGS_setup.elf	:= -T
+LDFLAGS_setup.elf	:= -m elf_i386 -T
 $(obj)/setup.elf: $(src)/setup.ld $(SETUP_OBJS) FORCE
 	$(call if_changed,ld)

diff --git a/arch/x86/realmode/rm/Makefile b/arch/x86/realmode/rm/Makefile
index 4463fa72db94..96cb20de08af 100644
--- a/arch/x86/realmode/rm/Makefile
+++ b/arch/x86/realmode/rm/Makefile
@@ -47,7 +47,7 @@ $(obj)/pasyms.h: $(REALMODE_OBJS) FORCE
 targets += realmode.lds
 $(obj)/realmode.lds: $(obj)/pasyms.h

-LDFLAGS_realmode.elf := --emit-relocs -T
+LDFLAGS_realmode.elf := -m elf_i386 --emit-relocs -T
 CPPFLAGS_realmode.lds += -P -C -I$(objtree)/$(obj)

 targets += realmode.elf
--
2.20.1.97.g81188d93c3-goog