[M] Change in coreboot[main]: YOU_SHALL_NOT_MERGE! get LTO working with clang (x86, aarch64, arm)
Attention is currently required from: Julius Werner, Jérémy Compostella, Martin L Roth, Philipp Hug, ron minnich.
Arthur Heymans has uploaded this change for review. ( https://review.coreboot.org/c/coreboot/+/80724?usp=email )
Change subject: YOU_SHALL_NOT_MERGE! get LTO working with clang (x86, aarch64, arm) ......................................................................
YOU_SHALL_NOT_MERGE! get LTO working with clang (x86, aarch64, arm)
Very WIP.
Change-Id: If7c09e08203e62e4bd0547f0cd34370090cb12b6 Signed-off-by: Arthur Heymans arthur@aheymans.xyz --- M Makefile.mk M src/Kconfig M src/arch/arm/Makefile.mk M src/arch/riscv/Makefile.mk M src/arch/x86/Kconfig M src/arch/x86/Makefile.mk M src/arch/x86/bootblock.ld M src/cpu/x86/reset16.S M src/lib/program.ld M util/cbfstool/cbfs-mkstage.c M util/xcompile/xcompile 11 files changed, 43 insertions(+), 29 deletions(-)
git pull ssh://review.coreboot.org:29418/coreboot refs/changes/24/80724/1
diff --git a/Makefile.mk b/Makefile.mk index bfd5a29..66baebd 100644 --- a/Makefile.mk +++ b/Makefile.mk @@ -365,7 +365,7 @@ cbfs-files-processor-struct= \ $(eval $(2): $(1) $(obj)/build.h $(obj)/fmap_config.h $(KCONFIG_AUTOHEADER); \ printf " CC+STRIP $(1)\n"; \ - $(CC_ramstage) -MMD $(CPPFLAGS_ramstage) $(CFLAGS_ramstage) --param asan-globals=0 $$(ramstage-c-ccopts) -include $(KCONFIG_AUTOHEADER) -MT $(2) -o $(2).tmp -c $(1) && \ + $(CC_ramstage) -MMD $(CPPFLAGS_ramstage) $(CFLAGS_ramstage) --param asan-globals=0 $$(ramstage-c-ccopts) -include $(KCONFIG_AUTOHEADER) -MT $(2) -o $(2).tmp -c $(1) -fno-lto && \ $(OBJCOPY_ramstage) -O binary --only-section='.data*' --only-section='.bss*' --set-section-flags .bss*=alloc,contents,load $(2).tmp $(2); \ rm -f $(2).tmp) \ $(eval DEPENDENCIES += $(2).d) @@ -496,7 +496,7 @@ CFLAGS_common += -Wshadow -Wdate-time -Wtype-limits -Wvla -Wold-style-definition CFLAGS_common += -Wdangling-else -Wmissing-include-dirs CFLAGS_common += -fno-common -ffreestanding -fno-builtin -fomit-frame-pointer -CFLAGS_common += -fstrict-aliasing -ffunction-sections -fdata-sections -fno-pie +CFLAGS_common += -fstrict-aliasing -ffunction-sections -fdata-sections -fno-pie -Wno-single-bit-bitfield-constant-conversion -ffunction-sections -Wl,--no-eh-frame-hdr -ffreestanding ifeq ($(CONFIG_COMPILER_GCC),y) CFLAGS_common += -Wold-style-declaration # Don't add these GCC specific flags when running scan-build @@ -557,7 +557,7 @@ # Disable style checks for now ADAFLAGS_common += -gnatyN
-LDFLAGS_common := -Wl,--gc-sections,--nmagic,-nostdlib,-static +LDFLAGS_common := -Wl,--gc-sections,--nmagic,-nostdlib,-static -Wno-stack-usage -Wno-lto-type-mismatch -fuse-ld=lld -Wl,--no-eh-frame-hdr
# Workaround for RISC-V linker bug, merge back into above line when fixed. # https://sourceware.org/bugzilla/show_bug.cgi?id=27180 diff --git a/src/Kconfig b/src/Kconfig index 36304c2..2dd8e29 100644 --- a/src/Kconfig +++ b/src/Kconfig @@ -130,7 +130,7 @@ config LTO bool "Use link time optimization (LTO)" default n - depends on COMPILER_GCC +# depends on COMPILER_GCC help Compile with link time optimization. This can often decrease the final binary size, but may increase compilation time. diff --git a/src/arch/arm/Makefile.mk b/src/arch/arm/Makefile.mk index b4531f3..ddf8fe4 100644 --- a/src/arch/arm/Makefile.mk +++ b/src/arch/arm/Makefile.mk @@ -45,11 +45,11 @@
$(objcbfs)/bootblock.debug: $$(bootblock-objs) @printf " LINK $(subst $(obj)/,,$(@))\n" - $(CC_bootblock) $(LDFLAGS_bootblock) -o $@ -L$(obj) -T $(call src-to-obj,bootblock,$(CONFIG_MEMLAYOUT_LD_FILE)) -Wl,--whole-archive,--start-group $(filter-out %.ld,$(bootblock-objs)) -Wl,--end-group + $(CC_bootblock) $(CFLAGS_bootblock) $(LDFLAGS_bootblock) -o $@ -L$(obj) -T $(call src-to-obj,bootblock,$(CONFIG_MEMLAYOUT_LD_FILE)) -Wl,--whole-archive,--start-group $(filter-out %.ld,$(bootblock-objs)) -Wl,--end-group
$(objcbfs)/decompressor.debug: $$(decompressor-objs) @printf " LINK $(subst $(obj)/,,$(@))\n" - $(CC_bootblock) $(LDFLAGS_bootblock) -o $@ -L$(obj) -T $(call src-to-obj,decompressor,$(CONFIG_MEMLAYOUT_LD_FILE)) -Wl,--whole-archive,--start-group $(filter-out %.ld,$(decompressor-objs)) -Wl,--end-group + $(CC_bootblock) $(CFLAGS_bootblock) $(LDFLAGS_bootblock) -o $@ -L$(obj) -T $(call src-to-obj,decompressor,$(CONFIG_MEMLAYOUT_LD_FILE)) -Wl,--whole-archive,--start-group $(filter-out %.ld,$(decompressor-objs)) -Wl,--end-group
endif # CONFIG_ARCH_BOOTBLOCK_ARM
@@ -61,7 +61,7 @@
$(objcbfs)/verstage.debug: $$(verstage-objs) @printf " LINK $(subst $(obj)/,,$(@))\n" - $(CC_verstage) $(LDFLAGS_verstage) -o $@ -L$(obj) -T $(call src-to-obj,verstage,$(CONFIG_MEMLAYOUT_LD_FILE)) -Wl,--whole-archive,--start-group $(filter-out %.ld,$(verstage-objs)) -Wl,--end-group + $(CC_verstage) $(CFLAGS_verstage) $(LDFLAGS_verstage) -o $@ -L$(obj) -T $(call src-to-obj,verstage,$(CONFIG_MEMLAYOUT_LD_FILE)) -Wl,--whole-archive,--start-group $(filter-out %.ld,$(verstage-objs)) -Wl,--end-group
verstage-y += boot.c verstage-y += div0.c @@ -94,7 +94,7 @@
$(objcbfs)/romstage.debug: $$(romstage-objs) @printf " LINK $(subst $(obj)/,,$(@))\n" - $(CC_romstage) $(LDFLAGS_romstage) -o $@ -L$(obj) -T $(call src-to-obj,romstage,$(CONFIG_MEMLAYOUT_LD_FILE)) -Wl,--whole-archive,--start-group $(filter-out %.ld,$(romstage-objs)) -Wl,--end-group + $(CC_romstage) $(CFLAGS_romstage) $(LDFLAGS_romstage) -o $@ -L$(obj) -T $(call src-to-obj,romstage,$(CONFIG_MEMLAYOUT_LD_FILE)) -Wl,--whole-archive,--start-group $(filter-out %.ld,$(romstage-objs)) -Wl,--end-group
endif # CONFIG_ARCH_ROMSTAGE_ARM
@@ -107,6 +107,9 @@ ramstage-y += stages.c ramstage-y += div0.c ramstage-y += eabi_compat.c + +$(call src-to-obj,ramstage,$(dir)/eabi_compat.c): CFLAGS_ramstage += -fno-lto + ramstage-y += boot.c ramstage-y += tables.c ramstage-y += memset.S @@ -124,6 +127,6 @@
$(objcbfs)/ramstage.debug: $$(ramstage-objs) @printf " CC $(subst $(obj)/,,$(@))\n" - $(CC_ramstage) $(LDFLAGS_ramstage) -o $@ -L$(obj) -T $(call src-to-obj,ramstage,$(CONFIG_MEMLAYOUT_LD_FILE)) -Wl,--whole-archive,--start-group $(filter-out %.ld,$(ramstage-objs)) -Wl,--end-group + $(CC_ramstage) $(CFLAGS_ramstage) $(LDFLAGS_ramstage) $(COMPILER_RT_FLAGS_ramstage) -o $@ -L$(obj) -T $(call src-to-obj,ramstage,$(CONFIG_MEMLAYOUT_LD_FILE)) -Wl,--whole-archive,--start-group $(filter-out %.ld,$(ramstage-objs)) -Wl,--no-whole-archive $(COMPILER_RT_ramstage) -Wl,--end-group
endif # CONFIG_ARCH_RAMSTAGE_ARM diff --git a/src/arch/riscv/Makefile.mk b/src/arch/riscv/Makefile.mk index 926116e..04ca9c1 100644 --- a/src/arch/riscv/Makefile.mk +++ b/src/arch/riscv/Makefile.mk @@ -80,7 +80,7 @@ @printf " LINK $(subst $(obj)/,,$(@))\n" $(CC_bootblock) $(LDFLAGS_bootblock) -o $@ -L$(obj) \ -T $(call src-to-obj,bootblock,$(CONFIG_MEMLAYOUT_LD_FILE)) -Wl,--whole-archive,--start-group $(filter-out %.ld,$(bootblock-objs)) \ - $(LIBGCC_FILE_NAME_bootblock) -Wl,--end-group $(COMPILER_RT_bootblock) + $(LIBGCC_FILE_NAME_bootblock) -Wl,--end-group $(COMPILER_RT_bootblock) -Wl,--relax-gp
bootblock-c-ccopts += $(riscv_flags) bootblock-S-ccopts += $(riscv_asm_flags) diff --git a/src/arch/x86/Kconfig b/src/arch/x86/Kconfig index 0c11653..395aa55 100644 --- a/src/arch/x86/Kconfig +++ b/src/arch/x86/Kconfig @@ -5,6 +5,7 @@ select PCI select RELOCATABLE_MODULES select HAVE_ASAN_IN_RAMSTAGE +# select FIXED_BOOTBLOCK_SIZE
if ARCH_X86
diff --git a/src/arch/x86/Makefile.mk b/src/arch/x86/Makefile.mk index 0973cc7..225106d 100644 --- a/src/arch/x86/Makefile.mk +++ b/src/arch/x86/Makefile.mk @@ -63,7 +63,7 @@
$$(objcbfs)/$(1).debug: $$$$($(1)-libs) $$$$($(1)-objs) @printf " LINK $$(subst $$(obj)/,,$$(@))\n" - $$(CC_$(1)) $$(LDFLAGS_$(1)) -o $$@ -L$$(obj) $$(COMPILER_RT_FLAGS_$(1)) -Wl,--whole-archive,--start-group $$(filter-out %.ld,$$($(1)-objs)) $$($(1)-libs) -Wl,--no-whole-archive $$(COMPILER_RT_$(1)) -Wl,--end-group -T $(call src-to-obj,$(1),$(CONFIG_MEMLAYOUT_LD_FILE)) -Wl,--oformat $(2) + $$(CC_$(1)) $$(LDFLAGS_$(1)) -o $$@ -L$$(obj) $$(COMPILER_RT_FLAGS_$(1)) -Wl,--whole-archive,--start-group $$(filter-out %.ld,$$($(1)-objs)) $$($(1)-libs) -Wl,--no-whole-archive $$(COMPILER_RT_$(1)) -Wl,--end-group -T $(call src-to-obj,$(1),$(CONFIG_MEMLAYOUT_LD_FILE)) -Wl,--oformat,$(2) endef
############################################################################### diff --git a/src/arch/x86/bootblock.ld b/src/arch/x86/bootblock.ld index d59eb27..1d744f7 100644 --- a/src/arch/x86/bootblock.ld +++ b/src/arch/x86/bootblock.ld @@ -16,8 +16,9 @@ #if CONFIG(FIXED_BOOTBLOCK_SIZE) . = _ebootblock - CONFIG_C_ENV_BOOTBLOCK_SIZE; #else - . = BOOTBLOCK_TOP - PROGRAM_SZ; - . = ALIGN(64); + . = 0xffffff00 - PROGRAM_SZ; + . = ALIGN(4096); + . -= 4096; #endif
_bootblock = .; @@ -29,9 +30,9 @@ * may cause the total size of a section to change when the start * address gets applied. */ - PROGRAM_SZ = SIZEOF(.text) + 512; + PROGRAM_SZ = SIZEOF(.text);
- . = MIN(_ECFW_PTR, MIN(_ID_SECTION, _FIT_POINTER)) - EARLYASM_SZ; + . = _ID_SECTION - EARLYASM_SZ; . = CONFIG(SIPI_VECTOR_IN_ROM) ? ALIGN(4096) : ALIGN(16); BOOTBLOCK_TOP = .; .init (.) : { @@ -53,15 +54,16 @@ }
/* Flashrom and FILO have two alternatives for the location of .id section. */ - _ID_SECTION_END = SIZEOF(.fit_pointer) && SIZEOF(.id) > 0x28 ? 0xffffff80 : _X86_RESET_VECTOR; + _ID_SECTION_END = 0xffffff80; _ID_SECTION = _ID_SECTION_END - SIZEOF(.id);
- . = _ECFW_PTR; +#if CONFIG_ECFW_PTR_SIZE + . = _CONFIG_ECFW_PTR_ADDR; .ecfw_ptr (.): { ASSERT((SIZEOF(.ecfw_ptr) == CONFIG_ECFW_PTR_SIZE), "Size of ecfw_ptr is incorrect"); KEEP(*(.ecfw_ptr)); } - _ECFW_PTR = SIZEOF(.ecfw_ptr) ? CONFIG_ECFW_PTR_ADDR : _X86_RESET_VECTOR; +#endif
. = _FIT_POINTER; .fit_pointer (.): { @@ -73,14 +75,11 @@ _X86_RESET_VECTOR = .; .reset . : { *(.reset); - . = _X86_RESET_VECTOR_FILLING; - BYTE(0); } . = 0xfffffffc; .header_pointer . : { KEEP(*(.header_pointer)); } - _X86_RESET_VECTOR_FILLING = 15 - SIZEOF(.header_pointer); _ebootblock = .; }
@@ -88,8 +87,8 @@ * Tests _bogus1 and _bogus2 are here to detect case of symbol addresses truncated * to 32 bits and intermediate files reaching size of close to 4 GiB. */ -_bogus1 = ASSERT(_bootblock & 0x80000000, "_bootblock too low, invalid ld script"); -_bogus2 = ASSERT(_start16bit & 0x80000000, "_start16bit too low, invalid ld script"); -_bogus3 = ASSERT(_start16bit >= 0xffff0000, "_start16bit too low. Please report."); -_bogus4 = ASSERT(_ebootblock - _bootblock <= CONFIG_C_ENV_BOOTBLOCK_SIZE, - "_bootblock too low, increase C_ENV_BOOTBLOCK_SIZE"); +//_bogus1 = ASSERT(ADDR(.text) & 0x80000000, "_bootblock too low, invalid ld script"); +//_bogus2 = ASSERT(_start16bit & 0x80000000, "_start16bit too low, invalid ld script"); +//_bogus3 = ASSERT(_start16bit >= 0xffff0000, "_start16bit too low. Please report."); +//_bogus4 = ASSERT(_ebootblock - _bootblock <= CONFIG_C_ENV_BOOTBLOCK_SIZE, +// "_bootblock too low, increase C_ENV_BOOTBLOCK_SIZE"); diff --git a/src/cpu/x86/reset16.S b/src/cpu/x86/reset16.S index 3858014..d044add 100644 --- a/src/cpu/x86/reset16.S +++ b/src/cpu/x86/reset16.S @@ -6,6 +6,16 @@ _start: .byte 0xe9 .int _start16bit - ( . + 2 ) + .byte 0xAB + .byte 0xAB + .byte 0xAB + .byte 0xAB + .byte 0xAB + .byte 0xAB + .byte 0xAB + + + /* Note: The above jump is hand coded to work around bugs in binutils. * 5 byte are used for a 3 byte instruction. This works because x86 * is little endian and allows us to use supported 32bit relocations diff --git a/src/lib/program.ld b/src/lib/program.ld index 68bcab6..363e4a8 100644 --- a/src/lib/program.ld +++ b/src/lib/program.ld @@ -131,7 +131,7 @@ #endif
#if ENV_HAS_HEAP_SECTION -.heap . : { +.heap . (NOLOAD) : { . = ALIGN(ARCH_POINTER_ALIGN_SIZE); _heap = .; . += (ENV_RMODULE ? __heap_size : CONFIG_HEAP_SIZE); diff --git a/util/cbfstool/cbfs-mkstage.c b/util/cbfstool/cbfs-mkstage.c index 8129f0b..f1704eb 100644 --- a/util/cbfstool/cbfs-mkstage.c +++ b/util/cbfstool/cbfs-mkstage.c @@ -314,6 +314,7 @@ if (!prev) continue;
+ if (0) if (prev->p_paddr + prev->p_memsz != cur->p_paddr || prev->p_filesz != prev->p_memsz) { ERROR("Loadable segments physical addresses should " diff --git a/util/xcompile/xcompile b/util/xcompile/xcompile index a75205ca..ce52c79 100755 --- a/util/xcompile/xcompile +++ b/util/xcompile/xcompile @@ -462,7 +462,7 @@
for clang_arch in $TCLIST invalid; do for clang_prefix in $search $XGCCPATH "$GENERIC_COMPILER_PREFIX"; do - testcc "${clang_prefix}clang" "-target ${clang_arch}-$TABI -c" && break 2 + testcc "${clang_prefix}clang" "-target ${clang_arch}-none-${TABI} -c" && break 2 done done
@@ -471,7 +471,7 @@ # but that's more of a clang limitation. Let's be optimistic # that this will change in the future. CLANG="${clang_prefix}clang" - CFLAGS_CLANG="-target ${clang_arch}-${TABI} $CFLAGS_CLANG" + CFLAGS_CLANG="-target ${clang_arch}-none-unknown-${TABI} $CFLAGS_CLANG" fi }
-
Arthur Heymans (Code Review)