Compliation issues with x86/libffi

["Cheng Jin" <jincheng@ca.ibm.com>]

Hi,

We are currently exploiting the latest version x86/libffi at
https://github.com/libffi/libffi/tree/master/src/x86 in our project but
there were a few compilation issues we detected on various platforms.


[1] Linux/64bit (compilation was based on the guideline at
https://github.com/libffi/libffi/blob/master/configure.host)

X86_64)
	if test x"$TARGET_X32" = xyes; then
		SOURCES="ffi64.c unix64.S"
	else
		SOURCES="ffi64.c unix64.S ffiw64.c win64.S"  <---------
	fi


The compilation passed on Ubuntu 19 with GCC 9 plus the GLIBC 2.31 but it
failed on Redhat6 with GCC 7.5 plus GLIBC 2.12 as follows:

gcc -O3 -fno-strict-aliasing -fgnu89-inline -g ...-fPIC  -m64
-fstack-protector -I. ...    -Wno-unused-result -c -o closures.o closures.c
gcc -O3 -fno-strict-aliasing -fgnu89-inline -g ...-fPIC -m64
-fstack-protector -I. ...   -c -o debug.o debug.c
gcc -O3 -fno-strict-aliasing -fgnu89-inline -g ...-fPIC  -fstack-protector
-I. ...-c -o ffi64.o x86/ffi64.c
gcc -O3 -fno-strict-aliasing -fgnu89-inline -g ...-fPIC -m64
-fstack-protector -I. ...-c -o ffiw64.o x86/ffiw64.c
gcc -O3 -fno-strict-aliasing -fgnu89-inline -g ...-fPIC   -m64
-fstack-protector -I. ...-c -o java_raw_api.o java_raw_api.c
gcc -O3 -fno-strict-aliasing -fgnu89-inline -g ...-fPIC   -m64
-fstack-protector -I. ...-c -o prep_cif.o prep_cif.c
gcc -O3 -fno-strict-aliasing -fgnu89-inline -g ...-fPIC   -m64
-fstack-protector -I. ...-c -o raw_api.o raw_api.c
gcc -O3 -fno-strict-aliasing -fgnu89-inline -g ... -fPIC  -m64
-fstack-protector -I. ...-c -o types.o types.c
gcc -Xassembler -noexecstack -Xassembler -64 -Xassembler --gdwarf2
-Xassembler -I. -Xassembler -I../include ... -c -o unix64.o x86/unix64.S
gcc -Xassembler -noexecstack -Xassembler -64 -Xassembler --gdwarf2
-Xassembler -I. -Xassembler -I../include ... -c -o win64.o x86/win64.S
ar rcv ../lib/libffi.a closures.o debug.o ffi64.o ffiw64.o java_raw_api.o
prep_cif.o raw_api.o types.o unix64.o win64.o
a - closures.o
a - debug.o
a - ffi64.o
a - ffiw64.o
a - java_raw_api.o
a - prep_cif.o
a - raw_api.o
a - types.o
a - unix64.o
a - win64.o
...
/usr/local/bin/../lib/gcc/x86_64-pc-linux-gnu/7.5.0/../../../../x86_64-pc-linux-gnu/bin/ld: ../lib//libffi.a(unix64.o):
 relocation R_X86_64_PC32 against symbol `abort@@GLIBC_2.2.5' can not be
used when making a shared object; recompile with -fPIC <--------
/usr/local/bin/../lib/gcc/x86_64-pc-linux-gnu/7.5.0/../../../../x86_64-pc-linux-gnu/bin/ld:
 final link failed: Bad value


We'd like to know why failed on the lower version of gcc and whether there
is any chance to fix the issue without upgrading the gcc toolchain to the
latest version (e.g. gcc 9) given there is no specific requirement on the
gcc version in x86/libffi.


------------------------------------

[2] Windows/32bit (compilation with MSVC on x86_64 machines targeting for
32bit based on the guideline at
https://github.com/libffi/libffi/blob/master/configure.host)

X86 | X86_DARWIN | X86_FREEBSD | X86_WIN32)
	if test "$MSVC" = 1; then
		SOURCES="ffi.c sysv_intel.S"  <---------
	else
		SOURCES="ffi.c sysv.S"
	fi


Currently our project includes an older version of x86/libffi (ffi.c &
win32.S) which works fine on Windows/32bit but it failed to compile after
replacing with ffi.c and sysv_intel.S

ffi.lib(prep_cif.c.obj) : warning LNK4049: locally defined symbol
_ffi_type_sint32 imported
ffi.lib(prep_cif.c.obj) : warning LNK4049: locally defined symbol
_ffi_type_float imported
ffi.lib(ffi.c.obj) : error LNK2019: unresolved external symbol
@ffi_call_i386@8 referenced in function _ffi_raw_call
ffi.lib(sysv_intel.asm.obj) : error LNK2019: unresolved external symbol
_@ffi_closure_inner@8 referenced in function _ffi_closure_i386


We tried to add the following code in sysv_intel.S (the same code as in
sysv.S)
/* Handle win32 fastcall name mangling.  */
#ifdef X86_WIN32
# define ffi_call_i386		"@ffi_call_i386@8"
# define ffi_closure_inner	"@ffi_closure_inner@8"
#else
# define ffi_call_i386		C(ffi_call_i386)
# define ffi_closure_inner	C(ffi_closure_inner)
#endif

but still ended up with the misleading failures:
...\libffi\sysv_intel.asm(1124) specified size
 : error A2008:syntax error : @ffi_closure_inner@8
...\libffi\sysv_intel.asm(1139) specified size
 : error A2008:syntax error : @ffi_call_i386@8
...\libffi\sysv_intel.asm(1140) specified size
 : error A2008:syntax error : @ffi_call_i386@8
...\libffi\sysv_intel.asm(1246) specified size
 : error A2008:syntax error : @ffi_call_i386@8
...\libffi\sysv_intel.asm(1366) specified size


So we are not too sure whether there was something we might be unaware of
in terms of compilation options or something else we missed; otherwise, we
might need to switch back to the older version specific to Windows/32bit to
avoid such issue.

Thanks and Best Regards
Cheng Jin