On 12/12/18 12:51, Ard Biesheuvel wrote: > On Wed, 12 Dec 2018 at 12:50, Laszlo Ersek <ler...@redhat.com> wrote: >> >> On 12/12/18 11:33, Ard Biesheuvel wrote: >>> PE/COFF only has a very limited id space for runtime relocations, and >>> so it defines only a single relocation for movw/movt instruction pairs, >>> which can be combined to load a 32-bit symbol reference into a register. >>> For this to work as expected, these instructions must always appear in >>> the same order and adjacently, and this is something few compilers take >>> into account, unless they target PE/COFF explicitly (and this is not the >>> case for our ELF based toolchains) >>> >>> For Clang 3.6 and later, we can pass the -mno-movt option to suppress >>> movw/movt pairs entirely, which works around the issue. Unfortunately, >>> for Clang 3.5, the option is called differently (-mllvm -arm-use-movt=0) >>> and mutually incompatible between 3.5 and 3.6. >>> >>> Since it is desirable for the CLANG35 toolchain to be usable on newer >>> versions of Clang as well (given that it is the only non-LTO alternative >>> to CLANG38), let's work around this issue in a way that permits versions >>> 3.5 and newer of Clang to be used with the CLANG35 profile. >>> >>> So pass the -mkernel flag instead (and drop the -mno-unaligned-access >>> in *_CLANG35_ARM_CC_XIPFLAGS which now becomes redundant, and which >>> Clang complains about). This also inhibits movw/movt generation, along >>> with some other changes (e.g., long calls) which do affect code generation >>> but not in a undesirable manner. >>> >>> Contributed-under: TianoCore Contribution Agreement 1.1 >>> Signed-off-by: Ard Biesheuvel <ard.biesheu...@linaro.org> >>> --- >>> BaseTools/Conf/tools_def.template | 2 +- >>> 1 file changed, 1 insertion(+), 1 deletion(-) >>> >>> diff --git a/BaseTools/Conf/tools_def.template >>> b/BaseTools/Conf/tools_def.template >>> index ac2b95e0f5ba..2ba833e1fb06 100755 >>> --- a/BaseTools/Conf/tools_def.template >>> +++ b/BaseTools/Conf/tools_def.template >>> @@ -5249,7 +5249,7 @@ DEFINE CLANG35_AARCH64_CC_FLAGS = >>> DEF(GCC_AARCH64_CC_FLAGS) DEF(CLANG35_AARCH64 >>> *_CLANG35_ARM_ASM_FLAGS = DEF(GCC_ASM_FLAGS) >>> DEF(CLANG35_ARM_TARGET) $(ARCHASM_FLAGS) $(PLATFORM_FLAGS) >>> -Qunused-arguments >>> *_CLANG35_ARM_DLINK_FLAGS = DEF(CLANG35_ARM_TARGET) >>> DEF(GCC_ARM_DLINK_FLAGS) >>> *_CLANG35_ARM_DLINK2_FLAGS = DEF(GCC_DLINK2_FLAGS_COMMON) >>> -Wl,--defsym=PECOFF_HEADER_SIZE=0x220 >>> -*_CLANG35_ARM_PLATFORM_FLAGS = -march=armv7-a >>> +*_CLANG35_ARM_PLATFORM_FLAGS = -march=armv7-a -mkernel >>> -Qunused-arguments >>> *_CLANG35_ARM_PP_FLAGS = DEF(GCC_PP_FLAGS) >>> DEF(CLANG35_ARM_TARGET) $(ARCHCC_FLAGS) $(PLATFORM_FLAGS) >>> *_CLANG35_ARM_RC_FLAGS = DEF(GCC_ARM_RC_FLAGS) >>> *_CLANG35_ARM_VFRPP_FLAGS = DEF(GCC_VFRPP_FLAGS) >>> DEF(CLANG35_ARM_TARGET) $(ARCHCC_FLAGS) $(PLATFORM_FLAGS) >>> >> >> The commit message speaks about adding -mkernel, and removing >> -mno-unaligned-access (elsewhere). In the patch, I only see -mkernel >> (plus an un-announced -Qunused-arguments option). >> >> Is the commit message slightly out-of-date relative to the code? (Or >> perhaps the other way around?) >> > > Ah yes. The -mno-unaligned-access became redundant, but instead of > removing it, i ended up adding the -Qunused-arguments to stop Clang > from complaining about it. >
OK, thanks! With the commit message updated accordingly: Acked-by: Laszlo Ersek <ler...@redhat.com> _______________________________________________ edk2-devel mailing list edk2-devel@lists.01.org https://lists.01.org/mailman/listinfo/edk2-devel