BaseTools/tools_def: AARCH64: disable LTO type mismatch warnings

On AARCH64, any code that may execute with the MMU off needs to be built
with -mstrict-align, given that unaligned accesses are not allowed unless
the MMU is enabled. This does not only affect SEC and PEI modules, but
also static libraries of the BASE type, which may be linked into such
modules, as well as into modules of other types. As it turns out, the
presence of -mstrict-align is reflected in the internal representations
of the types defined in those libraries.

When -fstrict-aliasing is passed to GCC, it assumes that pointers to
objects of different types cannot refer to the same memory location, and
attempts to exploit this fact when optimizing the code. Since such
assumptions are only valid under very strict conditions which are not
guaranteed to be met in EDK2, we disable this optimization by passing
-fno-strict-aliasing by default. [*]

When LTO is in effect, this applies equally to the code generation that
may occur at link time, which is why the linker warns about unexpected
differences in type definitions between the intermediate representations
that are present in the object files being linked. This may result in
warnings such as the one below, even if -fno-strict-aliasing is used:

  MdePkg/Include/Library/BaseLib.h:1712:1:
  warning: type of 'StrToGuid' does not match original declaration
  [-Wlto-type-mismatch]
   StrToGuid (
   ^
  MdePkg/Library/BaseLib/SafeString.c:1506:1:
  note: 'StrToGuid' was previously declared here
   StrToGuid (
   ^
  MdePkg/Library/BaseLib/SafeString.c:1506:1:
  note: code may be misoptimized unless -fno-strict-aliasing is used

This warning is inadvertently triggered when linking BASE libraries built
with -mstrict-align into modules of types other than SEC or PEI, since the
types are subtly different, even though the use of code that maintains
strict alignment in a module that does not care about this is unlikely to
cause problems. And even if it did, it would still only affect code built
with -fstrict-aliasing enabled, which we disable unconditionally. So let's
just silence the warning by passing -Wno-lto-type-mismatch.

[*] Leif adds: "-fstrict-aliasing is GCC default, because it is a
    restriction in the C language. Because it's a bit non-obvious, things
    can go hilariously wrong in very non-obvious ways, and the potential
    optimization gains are unlikely to be generally relevant,
    -fno-strict-aliasing is a sensible thing to always have set (like we
    do)."

Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Reviewed-by: Leif Lindholm <leif.lindholm@linaro.org>

diff --git a/BaseTools/Conf/tools_def.template b/BaseTools/Conf/tools_def.template
index 39486b8dfbd0e20fe768105fbc3b897d636d2f6d..17681e8ce3f6c1ab33f7ee5687246b72a75c17b2 100755 (executable)
--- a/BaseTools/Conf/tools_def.template
+++ b/BaseTools/Conf/tools_def.template
@@ -5407,7 +5407,7 @@ RELEASE_GCC5_ARM_DLINK_FLAGS     = DEF(GCC5_ARM_DLINK_FLAGS) -flto -Os -L$(WORKS
   DEBUG_GCC5_AARCH64_DLINK_XIPFLAGS = -z common-page-size=0x20\r
 \r
 RELEASE_GCC5_AARCH64_CC_FLAGS    = DEF(GCC5_AARCH64_CC_FLAGS) -flto -Wno-unused-but-set-variable -mcmodel=tiny -fomit-frame-pointer\r
-RELEASE_GCC5_AARCH64_DLINK_FLAGS = DEF(GCC5_AARCH64_DLINK_FLAGS) -flto -Os -L$(WORKSPACE)/ArmPkg/Library/GccLto -llto-aarch64 -Wl,-plugin-opt=-pass-through=-llto-aarch64\r
+RELEASE_GCC5_AARCH64_DLINK_FLAGS = DEF(GCC5_AARCH64_DLINK_FLAGS) -flto -Os -L$(WORKSPACE)/ArmPkg/Library/GccLto -llto-aarch64 -Wl,-plugin-opt=-pass-through=-llto-aarch64 -Wno-lto-type-mismatch\r
 \r
   NOOPT_GCC5_AARCH64_CC_FLAGS    = DEF(GCC5_AARCH64_CC_FLAGS) -O0 -mcmodel=small\r
   NOOPT_GCC5_AARCH64_DLINK_FLAGS = DEF(GCC5_AARCH64_DLINK_FLAGS) -z common-page-size=0x1000 -O0\r