linux-kernel - Re: [PATCH v3 3/4] MIPS: VDSO: Use $(LD) instead of $(CC) to link VDSO

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20200428021737.GA7351@ubuntu-s3-xlarge-x86>
Date:   Mon, 27 Apr 2020 19:17:37 -0700
From:   Nathan Chancellor <natechancellor@...il.com>
To:     Thomas Bogendoerfer <tsbogend@...ha.franken.de>
Cc:     Masahiro Yamada <masahiroy@...nel.org>, linux-mips@...r.kernel.org,
        linux-kernel@...r.kernel.org, clang-built-linux@...glegroups.com,
        linux-kbuild@...r.kernel.org,
        Nick Desaulniers <ndesaulniers@...gle.com>,
        Sami Tolvanen <samitolvanen@...gle.com>,
        Dmitry Golovin <dima@...ovin.in>,
        Sedat Dilek <sedat.dilek@...il.com>
Subject: Re: [PATCH v3 3/4] MIPS: VDSO: Use $(LD) instead of $(CC) to link
 VDSO

On Sun, Apr 26, 2020 at 06:27:37PM +0200, Thomas Bogendoerfer wrote:
> On Thu, Apr 23, 2020 at 10:18:06AM -0700, Nathan Chancellor wrote:
> > Currently, the VDSO is being linked through $(CC). This does not match
> > how the rest of the kernel links objects, which is through the $(LD)
> > variable.
> 
> this causes build errors for me when (cross) compiling a big endian target:
> 
> target is little endian
> mips64-linux-gnu-ld: arch/mips/vdso/elf.o: endianness incompatible with that of the selected emulation
> mips64-linux-gnu-ld: failed to merge target specific data of file arch/mips/vdso/elf.o
> 
> Thomas.
> 

Thanks for reporting this, I figured it out. This is the solution that I
came up with, I'll send out a v4 tomorrow once I do some more testing.

Cheers,
Nathan

>From 256e3b6c8fff7a66aa29961ebefc0fe653ec34b6 Mon Sep 17 00:00:00 2001
From: Nathan Chancellor <natechancellor@...il.com>
Date: Mon, 27 Apr 2020 17:02:55 -0700
Subject: [PATCH] MIPS: Unconditionally specify '-EL' or '-EB'

This was all done to work around a GCC bug that has been fixed after
4.2. The kernel requires GCC 4.6 or newer so remove all of these hacks
and just use the traditional flags.

 $ mips64-linux-gcc --version | head -n1
 mips64-linux-gcc (GCC) 4.6.3

 $ mips64-linux-gcc -EB -dM -E -C -x c /dev/null | grep MIPSE
 #define MIPSEB 1
 #define __MIPSEB__ 1
 #define _MIPSEB 1
 #define __MIPSEB 1

 $ mips64-linux-gcc -EL -dM -E -C -x c /dev/null | grep MIPSE
 #define __MIPSEL__ 1
 #define MIPSEL 1
 #define _MIPSEL 1
 #define __MIPSEL 1

This is necessary when converting the MIPS VDSO to use $(LD) instead of
$(CC) to link because the OUTPUT_FORMAT is defaulted to little endian
and only flips to big endian when -EB is set on the command line, which
is inherited from KBUILD_CFLAGS. Without this, we will see the following
error when compiling for big endian (64r2_defconfig):

$ make -j$(nproc) ARCH=mips CROSS_COMPILE=mips64-linux- \
  64r2el_defconfig arch/mips/vdso/
...
mips64-linux-ld: arch/mips/vdso/elf.o: compiled for a big endian system
and target is little endian
mips64-linux-ld: arch/mips/vdso/elf.o: endianness incompatible with that
of the selected emulation
mips64-linux-ld: failed to merge target specific data of file
arch/mips/vdso/elf.o
...

Reported-by: Thomas Bogendoerfer <tsbogend@...ha.franken.de>
Signed-off-by: Nathan Chancellor <natechancellor@...il.com>
---
 arch/mips/Makefile | 25 -------------------------
 1 file changed, 25 deletions(-)

diff --git a/arch/mips/Makefile b/arch/mips/Makefile
index e1c44aed81565..301efb90b51ed 100644
--- a/arch/mips/Makefile
+++ b/arch/mips/Makefile
@@ -116,33 +116,8 @@ endif
 
 cflags-y += -ffreestanding
 
-#
-# We explicitly add the endianness specifier if needed, this allows
-# to compile kernels with a toolchain for the other endianness. We
-# carefully avoid to add it redundantly because gcc 3.3/3.4 complains
-# when fed the toolchain default!
-#
-# Certain gcc versions up to gcc 4.1.1 (probably 4.2-subversion as of
-# 2006-10-10 don't properly change the predefined symbols if -EB / -EL
-# are used, so we kludge that here.  A bug has been filed at
-# http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29413.
-#
-# clang doesn't suffer from these issues and our checks against -dumpmachine
-# don't work so well when cross compiling, since without providing --target
-# clang's output will be based upon the build machine. So for clang we simply
-# unconditionally specify -EB or -EL as appropriate.
-#
-ifdef CONFIG_CC_IS_CLANG
 cflags-$(CONFIG_CPU_BIG_ENDIAN)		+= -EB
 cflags-$(CONFIG_CPU_LITTLE_ENDIAN)	+= -EL
-else
-undef-all += -UMIPSEB -U_MIPSEB -U__MIPSEB -U__MIPSEB__
-undef-all += -UMIPSEL -U_MIPSEL -U__MIPSEL -U__MIPSEL__
-predef-be += -DMIPSEB -D_MIPSEB -D__MIPSEB -D__MIPSEB__
-predef-le += -DMIPSEL -D_MIPSEL -D__MIPSEL -D__MIPSEL__
-cflags-$(CONFIG_CPU_BIG_ENDIAN)		+= $(shell $(CC) -dumpmachine |grep -q 'mips.*el-.*' && echo -EB $(undef-all) $(predef-be))
-cflags-$(CONFIG_CPU_LITTLE_ENDIAN)	+= $(shell $(CC) -dumpmachine |grep -q 'mips.*el-.*' || echo -EL $(undef-all) $(predef-le))
-endif
 
 cflags-$(CONFIG_SB1XXX_CORELIS)	+= $(call cc-option,-mno-sched-prolog) \
 				   -fno-omit-frame-pointer
-- 
2.26.2