.gitignore | 3 ++ MAINTAINERS | 5 +++ Makefile | 40 ++++++++++++++++++++--- arch/Kconfig | 19 +++++++++++ scripts/Makefile.build | 52 +++++++++++++++++++++++++++--- scripts/Makefile.lib | 7 +++- scripts/Makefile.vmlinux_o | 16 +++++++--- scripts/Makefile.vmlinux_thinlink | 53 +++++++++++++++++++++++++++++++ scripts/head-object-list.txt | 1 + 9 files changed, 181 insertions(+), 15 deletions(-) create mode 100644 scripts/Makefile.vmlinux_thinlink
From: Rong Xu <xur@google.com>
Add distributed ThinLTO build support for the Linux kernel.
This new mode offers several advantages: (1) Increased
flexibility in handling user-specified build options.
(2) Improved user-friendliness for developers. (3) Greater
convenience for integrating with objtool and livepatch.
Note that "distributed" in this context refers to a term
that differentiates in-process ThinLTO builds by invoking
backend compilation through the linker, not necessarily
building in distributed environments.
Distributed ThinLTO is enabled via the
`CONFIG_LTO_CLANG_THIN_DIST` Kconfig option. For example:
> make LLVM=1 defconfig
> scripts/config -e LTO_CLANG_THIN_DIST
> make LLVM=1 oldconfig
> make LLVM=1 vmlinux -j <..>
The implementation changes the top-level Makefile with a
macro for generating `vmlinux.o` for distributed ThinLTO
builds. It uses the existing Kbuild infrastructure to
perform two recursive passes through the subdirectories.
The first pass generates LLVM IR object files, similar to
in-process ThinLTO. Following the thin-link stage, a second
pass compiles these IR files into the final native object
files. The build rules and actions for this two-pass process
are primarily implemented in `scripts/Makefile.build`.
Currently, this patch focuses on building the main kernel
image (`vmlinux`) only. Support for building kernel modules
using this method is planned for a subsequent patch.
Tested on the following arch: x86, arm64, loongarch, and
riscv.
Some implementation details can be found here:
https://discourse.llvm.org/t/rfc-distributed-thinlto-build-for-kernel/85934
Signed-off-by: Rong Xu <xur@google.com>
---
Changelog since v1:
- Updated the description in arch/Kconfig based on feedback
from Nathan Chancellor
- Revised file suffixes: .final_o -> .o.thinlto.native, and
.final_a -> .a.thinlto.native
- Updated list of ignored files in .gitignore
Changelog since v2:
- Changed file suffixes: .o.thinlto.native -> .o_thinlto_native,
and .a.thinlto.native -> .a_thinlto_native so that basename
works as intended.
- Tested the patch with AutoFDO and Propeller.
---
.gitignore | 3 ++
MAINTAINERS | 5 +++
Makefile | 40 ++++++++++++++++++++---
arch/Kconfig | 19 +++++++++++
scripts/Makefile.build | 52 +++++++++++++++++++++++++++---
scripts/Makefile.lib | 7 +++-
scripts/Makefile.vmlinux_o | 16 +++++++---
scripts/Makefile.vmlinux_thinlink | 53 +++++++++++++++++++++++++++++++
scripts/head-object-list.txt | 1 +
9 files changed, 181 insertions(+), 15 deletions(-)
create mode 100644 scripts/Makefile.vmlinux_thinlink
diff --git a/.gitignore b/.gitignore
index f2f63e47fb886..b83a68185ef46 100644
--- a/.gitignore
+++ b/.gitignore
@@ -12,6 +12,7 @@
#
.*
*.a
+*.a_thinlto_native
*.asn1.[ch]
*.bin
*.bz2
@@ -39,6 +40,7 @@
*.mod.c
*.o
*.o.*
+*.o_thinlto_native
*.patch
*.rmeta
*.rpm
@@ -64,6 +66,7 @@ modules.order
/vmlinux
/vmlinux.32
/vmlinux.map
+/vmlinux.thinlink
/vmlinux.symvers
/vmlinux.unstripped
/vmlinux-gdb.py
diff --git a/MAINTAINERS b/MAINTAINERS
index d48dd6726fe6b..f54090f364c93 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -5790,6 +5790,11 @@ F: scripts/Makefile.clang
F: scripts/clang-tools/
K: \b(?i:clang|llvm)\b
+CLANG/LLVM THINLTO DISTRIBUTED BUILD
+M: Rong Xu <xur@google.com>
+S: Supported
+F: scripts/Makefile.vmlinux_thinlink
+
CLK API
M: Russell King <linux@armlinux.org.uk>
L: linux-clk@vger.kernel.org
diff --git a/Makefile b/Makefile
index a9edd03036537..8fbff2ab87ebd 100644
--- a/Makefile
+++ b/Makefile
@@ -298,7 +298,8 @@ no-dot-config-targets := $(clean-targets) \
outputmakefile rustavailable rustfmt rustfmtcheck
no-sync-config-targets := $(no-dot-config-targets) %install modules_sign kernelrelease \
image_name
-single-targets := %.a %.i %.ko %.lds %.ll %.lst %.mod %.o %.rsi %.s %/
+single-targets := %.a %.i %.ko %.lds %.ll %.lst %.mod %.o %.rsi %.s %.o_thinlto_native \
+ %.a_thinlto_native %.o.thinlto.bc %/
config-build :=
mixed-build :=
@@ -991,10 +992,10 @@ export CC_FLAGS_SCS
endif
ifdef CONFIG_LTO_CLANG
-ifdef CONFIG_LTO_CLANG_THIN
-CC_FLAGS_LTO := -flto=thin -fsplit-lto-unit
-else
+ifdef CONFIG_LTO_CLANG_FULL
CC_FLAGS_LTO := -flto
+else # for CONFIG_LTO_CLANG_THIN or CONFIG_LTO_CLANG_THIN_DIST
+CC_FLAGS_LTO := -flto=thin -fsplit-lto-unit
endif
CC_FLAGS_LTO += -fvisibility=hidden
@@ -1213,8 +1214,34 @@ vmlinux.a: $(KBUILD_VMLINUX_OBJS) scripts/head-object-list.txt FORCE
$(call if_changed,ar_vmlinux.a)
PHONY += vmlinux_o
+ifdef CONFIG_LTO_CLANG_THIN_DIST
+vmlinux.thinlink: vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
+ $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.vmlinux_thinlink
+targets += vmlinux.thinlink
+
+vmlinux.a_thinlto_native := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_OBJS))
+quiet_cmd_ar_vmlinux.a_thinlto_native = AR $@
+ cmd_ar_vmlinux.a_thinlto_native = \
+ rm -f $@; \
+ $(AR) cDPrST $@ $(vmlinux.a_thinlto_native); \
+ $(AR) mPiT $$($(AR) t $@ | sed -n 1p) $@ $$($(AR) t $@ | grep -F -f $(srctree)/scripts/head-object-list.txt)
+
+define rule_gen_vmlinux.a_thinlto_native
+ +$(Q)$(MAKE) $(build)=. need-builtin=1 thinlto_final_pass=1 need-modorder=1 built-in.a_thinlto_native
+ $(call cmd_and_savecmd,ar_vmlinux.a_thinlto_native)
+endef
+
+vmlinux.a_thinlto_native: vmlinux.thinlink scripts/head-object-list.txt FORCE
+ $(call if_changed_rule,gen_vmlinux.a_thinlto_native)
+
+targets += vmlinux.a_thinlto_native
+
+vmlinux_o: vmlinux.a_thinlto_native
+ $(Q)$(MAKE) thinlto_final_pass=1 -f $(srctree)/scripts/Makefile.vmlinux_o
+else
vmlinux_o: vmlinux.a $(KBUILD_VMLINUX_LIBS)
$(Q)$(MAKE) -f $(srctree)/scripts/Makefile.vmlinux_o
+endif
vmlinux.o modules.builtin.modinfo modules.builtin: vmlinux_o
@:
@@ -1572,7 +1599,8 @@ CLEAN_FILES += vmlinux.symvers modules-only.symvers \
modules.builtin.ranges vmlinux.o.map vmlinux.unstripped \
compile_commands.json rust/test \
rust-project.json .vmlinux.objs .vmlinux.export.c \
- .builtin-dtbs-list .builtin-dtb.S
+ .builtin-dtbs-list .builtin-dtb.S \
+ .vmlinux_thinlto_bc_files vmlinux.thinlink
# Directories & files removed with 'make mrproper'
MRPROPER_FILES += include/config include/generated \
@@ -2023,6 +2051,8 @@ clean: $(clean-dirs)
-o -name '*.symtypes' -o -name 'modules.order' \
-o -name '*.c.[012]*.*' \
-o -name '*.ll' \
+ -o -name '*.a_thinlto_native' -o -name '*.o_thinlto_native' \
+ -o -name '*.o.thinlto.bc' \
-o -name '*.gcno' \
\) -type f -print \
-o -name '.tmp_*' -print \
diff --git a/arch/Kconfig b/arch/Kconfig
index b0adb665041f1..30dccda07c671 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -810,6 +810,25 @@ config LTO_CLANG_THIN
https://clang.llvm.org/docs/ThinLTO.html
If unsure, say Y.
+
+config LTO_CLANG_THIN_DIST
+ bool "Clang ThinLTO in distributed mode (EXPERIMENTAL)"
+ depends on HAS_LTO_CLANG && ARCH_SUPPORTS_LTO_CLANG_THIN
+ select LTO_CLANG
+ help
+ This option enables Clang's ThinLTO in distributed build mode.
+ In this mode, the linker performs the thin-link, generating
+ ThinLTO index files. Subsequently, the build system explicitly
+ invokes ThinLTO backend compilation using these index files
+ and pre-linked IR objects. The resulting native object files
+ are with the .o_thinlto_native suffix.
+
+ This build mode offers improved visibility into the ThinLTO
+ process through explicit subcommand exposure. It also makes
+ final native object files directly available, benefiting
+ tools like objtool and kpatch. Additionally, it provides
+ crucial granular control over back-end options, enabling
+ module-specific compiler options, and simplifies debugging.
endchoice
config ARCH_SUPPORTS_AUTOFDO_CLANG
diff --git a/scripts/Makefile.build b/scripts/Makefile.build
index 13dcd86e74ca8..338e1aec0eaa3 100644
--- a/scripts/Makefile.build
+++ b/scripts/Makefile.build
@@ -50,18 +50,23 @@ endif
# ===========================================================================
+builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a)
+ifeq ($(thinlto_final_pass),1)
+builtin_suffix :=.a_thinlto_native
+endif
+
# subdir-builtin and subdir-modorder may contain duplications. Use $(sort ...)
-subdir-builtin := $(sort $(filter %/built-in.a, $(real-obj-y)))
+subdir-builtin := $(sort $(filter %/built-in$(builtin_suffix), $(real-obj-y)))
subdir-modorder := $(sort $(filter %/modules.order, $(obj-m)))
targets-for-builtin := $(extra-y)
ifneq ($(strip $(lib-y) $(lib-m) $(lib-)),)
-targets-for-builtin += $(obj)/lib.a
+targets-for-builtin += $(obj)/lib$(builtin_suffix)
endif
ifdef need-builtin
-targets-for-builtin += $(obj)/built-in.a
+targets-for-builtin += $(obj)/built-in$(builtin_suffix)
endif
targets-for-modules := $(foreach x, o mod, \
@@ -337,6 +342,10 @@ $(obj)/%.o: $(obj)/%.S FORCE
targets += $(filter-out $(subdir-builtin), $(real-obj-y))
targets += $(filter-out $(subdir-modorder), $(real-obj-m))
targets += $(lib-y) $(always-y)
+ifeq ($(builtin_suffix),.a_thinlto_native)
+native_targets = $(patsubst,%.o,%.o_thinlto_native,$(targets))
+targets += $(native_targets)
+endif
# Linker scripts preprocessor (.lds.S -> .lds)
# ---------------------------------------------------------------------------
@@ -347,6 +356,24 @@ quiet_cmd_cpp_lds_S = LDS $@
$(obj)/%.lds: $(src)/%.lds.S FORCE
$(call if_changed_dep,cpp_lds_S)
+ifdef CONFIG_LTO_CLANG_THIN_DIST
+# Generate .o_thinlto_native (obj) from .o (bitcode) file
+# ---------------------------------------------------------------------------
+quiet_cmd_cc_o_bc = CC $(quiet_modtag) $@
+
+cmd_cc_o_bc = $(if $(filter bitcode, $(shell file -b $<)),$(CC) \
+ $(filter-out -Wp% $(LINUXINCLUDE) %.h.gch %.h -D% \
+ -flto=thin, $(c_flags)) \
+ -Wno-unused-command-line-argument \
+ -x ir -fthinlto-index=$<.thinlto.bc -c -o $@ \
+ $(if $(findstring ../,$<), \
+ $$(realpath --relative-to=$(srcroot) $<), $<), \
+ cp $< $@)
+
+$(obj)/%.o_thinlto_native: $(obj)/%.o FORCE
+ $(call if_changed,cc_o_bc)
+endif
+
# ASN.1 grammar
# ---------------------------------------------------------------------------
quiet_cmd_asn1_compiler = ASN.1 $(basename $@).[ch]
@@ -360,7 +387,7 @@ $(obj)/%.asn1.c $(obj)/%.asn1.h: $(src)/%.asn1 $(objtree)/scripts/asn1_compiler
# ---------------------------------------------------------------------------
# To build objects in subdirs, we need to descend into the directories
-$(subdir-builtin): $(obj)/%/built-in.a: $(obj)/% ;
+$(subdir-builtin): $(obj)/%/built-in$(builtin_suffix): $(obj)/% ;
$(subdir-modorder): $(obj)/%/modules.order: $(obj)/% ;
#
@@ -377,6 +404,12 @@ quiet_cmd_ar_builtin = AR $@
$(obj)/built-in.a: $(real-obj-y) FORCE
$(call if_changed,ar_builtin)
+ifdef CONFIG_LTO_CLANG_THIN_DIST
+# Rule to compile a set of .o_thinlto_native files into one .a_thinlto_native file.
+$(obj)/built-in.a_thinlto_native: $(patsubst %.o,%.o_thinlto_native,$(real-obj-y)) FORCE
+ $(call if_changed,ar_builtin)
+endif
+
# This is a list of build artifacts from the current Makefile and its
# sub-directories. The timestamp should be updated when any of the member files.
@@ -394,6 +427,14 @@ $(obj)/modules.order: $(obj-m) FORCE
$(obj)/lib.a: $(lib-y) FORCE
$(call if_changed,ar)
+ifdef CONFIG_LTO_CLANG_THIN_DIST
+quiet_cmd_ar_native = AR $@
+ cmd_ar_native = rm -f $@; $(AR) cDPrsT $@ $(patsubst %.o,%.o_thinlto_native,$(real-prereqs))
+
+$(obj)/lib.a_thinlto_native: $(patsubst %.o,%.o_thinlto_native,$(lib-y)) FORCE
+ $(call if_changed,ar_native)
+endif
+
quiet_cmd_ld_multi_m = LD [M] $@
cmd_ld_multi_m = $(LD) $(ld_flags) -r -o $@ @$< $(cmd_objtool)
@@ -459,7 +500,8 @@ $(single-subdir-goals): $(single-subdirs)
PHONY += $(subdir-ym)
$(subdir-ym):
$(Q)$(MAKE) $(build)=$@ \
- need-builtin=$(if $(filter $@/built-in.a, $(subdir-builtin)),1) \
+ need-builtin=$(if $(filter $@/built-in$(builtin_suffix), $(subdir-builtin)),1) \
+ thinlto_final_pass=$(if $(filter .a_thinlto_native, $(builtin_suffix)),1) \
need-modorder=$(if $(filter $@/modules.order, $(subdir-modorder)),1) \
$(filter $@/%, $(single-subdir-goals))
diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
index 2fe73cda0bddb..9cfd23590334d 100644
--- a/scripts/Makefile.lib
+++ b/scripts/Makefile.lib
@@ -34,8 +34,13 @@ else
obj-m := $(filter-out %/, $(obj-m))
endif
+builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a)
+ifeq ($(thinlto_final_pass),1)
+ builtin_suffix :=.a_thinlto_native
+endif
+
ifdef need-builtin
-obj-y := $(patsubst %/, %/built-in.a, $(obj-y))
+obj-y := $(patsubst %/, %/built-in$(builtin_suffix), $(obj-y))
else
obj-y := $(filter-out %/, $(obj-y))
endif
diff --git a/scripts/Makefile.vmlinux_o b/scripts/Makefile.vmlinux_o
index b024ffb3e2018..f9abc45a68b36 100644
--- a/scripts/Makefile.vmlinux_o
+++ b/scripts/Makefile.vmlinux_o
@@ -9,6 +9,14 @@ include $(srctree)/scripts/Kbuild.include
# for objtool
include $(srctree)/scripts/Makefile.lib
+ifeq ($(thinlto_final_pass),1)
+vmlinux_a := vmlinux.a_thinlto_native
+vmlinux_libs := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_LIBS))
+else
+vmlinux_a := vmlinux.a
+vmlinux_libs := $(KBUILD_VMLINUX_LIBS)
+endif
+
# Generate a linker script to ensure correct ordering of initcalls for Clang LTO
# ---------------------------------------------------------------------------
@@ -18,7 +26,7 @@ quiet_cmd_gen_initcalls_lds = GEN $@
$(PERL) $(real-prereqs) > $@
.tmp_initcalls.lds: $(srctree)/scripts/generate_initcall_order.pl \
- vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
+ $(vmlinux_a) $(vmlinux_libs) FORCE
$(call if_changed,gen_initcalls_lds)
targets := .tmp_initcalls.lds
@@ -59,8 +67,8 @@ quiet_cmd_ld_vmlinux.o = LD $@
$(LD) ${KBUILD_LDFLAGS} -r -o $@ \
$(vmlinux-o-ld-args-y) \
$(addprefix -T , $(initcalls-lds)) \
- --whole-archive vmlinux.a --no-whole-archive \
- --start-group $(KBUILD_VMLINUX_LIBS) --end-group \
+ --whole-archive $(vmlinux_a) --no-whole-archive \
+ --start-group $(vmlinux_libs) --end-group \
$(cmd_objtool)
define rule_ld_vmlinux.o
@@ -68,7 +76,7 @@ define rule_ld_vmlinux.o
$(call cmd,gen_objtooldep)
endef
-vmlinux.o: $(initcalls-lds) vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
+vmlinux.o: $(initcalls-lds) $(vmlinux_a) $(vmlinux_libs) FORCE
$(call if_changed_rule,ld_vmlinux.o)
targets += vmlinux.o
diff --git a/scripts/Makefile.vmlinux_thinlink b/scripts/Makefile.vmlinux_thinlink
new file mode 100644
index 0000000000000..13e4026c7d45b
--- /dev/null
+++ b/scripts/Makefile.vmlinux_thinlink
@@ -0,0 +1,53 @@
+# SPDX-License-Identifier: GPL-2.0-only
+
+PHONY := __default
+__default: vmlinux.thinlink
+
+include include/config/auto.conf
+include $(srctree)/scripts/Kbuild.include
+
+
+# Generate a linker script to ensure correct ordering of initcalls for Clang LTO
+# ---------------------------------------------------------------------------
+
+quiet_cmd_gen_initcalls_lds = GEN $@
+ cmd_gen_initcalls_lds = \
+ $(PYTHON3) $(srctree)/scripts/jobserver-exec \
+ $(PERL) $(real-prereqs) > $@
+
+.tmp_initcalls_thinlink.lds: $(srctree)/scripts/generate_initcall_order.pl \
+ vmlinux.a FORCE
+ $(call if_changed,gen_initcalls_lds)
+
+targets := .tmp_initcalls_thinlink.lds
+
+initcalls-lds := .tmp_initcalls_thinlink.lds
+
+quiet_cmd_ld_vmlinux.thinlink = LD $@
+ cmd_ld_vmlinux.thinlink = \
+ $(AR) t vmlinux.a > .vmlinux_thinlto_bc_files; \
+ $(LD) ${KBUILD_LDFLAGS} -r $(addprefix -T , $(initcalls-lds)) \
+ --thinlto-index-only @.vmlinux_thinlto_bc_files; \
+ touch vmlinux.thinlink
+
+vmlinux.thinlink: vmlinux.a $(initcalls-lds) FORCE
+ $(call if_changed,ld_vmlinux.thinlink)
+
+targets += vmlinux.thinlink
+
+# Add FORCE to the prerequisites of a target to force it to be always rebuilt.
+# ---------------------------------------------------------------------------
+
+PHONY += FORCE
+FORCE:
+
+# Read all saved command lines and dependencies for the $(targets) we
+# may be building above, using $(if_changed{,_dep}). As an
+# optimization, we don't need to read them if the target does not
+# exist, we will rebuild anyway in that case.
+
+existing-targets := $(wildcard $(sort $(targets)))
+
+-include $(foreach f,$(existing-targets),$(dir $(f)).$(notdir $(f)).cmd)
+
+.PHONY: $(PHONY)
diff --git a/scripts/head-object-list.txt b/scripts/head-object-list.txt
index 7274dfc65af60..90710b87a3877 100644
--- a/scripts/head-object-list.txt
+++ b/scripts/head-object-list.txt
@@ -18,6 +18,7 @@ arch/arm/kernel/head.o
arch/csky/kernel/head.o
arch/hexagon/kernel/head.o
arch/loongarch/kernel/head.o
+arch/loongarch/kernel/head.o_thinlto_native
arch/m68k/68000/head.o
arch/m68k/coldfire/head.o
arch/m68k/kernel/head.o
--
2.49.0.1143.g0be31eac6b-googOn Thu, May 22, 2025 at 6:35 AM <xur@google.com> wrote: > > From: Rong Xu <xur@google.com> > > Add distributed ThinLTO build support for the Linux kernel. > This new mode offers several advantages: (1) Increased > flexibility in handling user-specified build options. > (2) Improved user-friendliness for developers. (3) Greater > convenience for integrating with objtool and livepatch. > > Note that "distributed" in this context refers to a term > that differentiates in-process ThinLTO builds by invoking > backend compilation through the linker, not necessarily > building in distributed environments. > > Distributed ThinLTO is enabled via the > `CONFIG_LTO_CLANG_THIN_DIST` Kconfig option. For example: > > make LLVM=1 defconfig > > scripts/config -e LTO_CLANG_THIN_DIST > > make LLVM=1 oldconfig > > make LLVM=1 vmlinux -j <..> > > The implementation changes the top-level Makefile with a > macro for generating `vmlinux.o` for distributed ThinLTO > builds. It uses the existing Kbuild infrastructure to > perform two recursive passes through the subdirectories. > The first pass generates LLVM IR object files, similar to > in-process ThinLTO. Following the thin-link stage, a second > pass compiles these IR files into the final native object > files. The build rules and actions for this two-pass process > are primarily implemented in `scripts/Makefile.build`. > > Currently, this patch focuses on building the main kernel > image (`vmlinux`) only. Support for building kernel modules > using this method is planned for a subsequent patch. > > Tested on the following arch: x86, arm64, loongarch, and > riscv. > > Some implementation details can be found here: > https://discourse.llvm.org/t/rfc-distributed-thinlto-build-for-kernel/85934 > > Signed-off-by: Rong Xu <xur@google.com> > --- > Changelog since v1: > - Updated the description in arch/Kconfig based on feedback > from Nathan Chancellor > - Revised file suffixes: .final_o -> .o.thinlto.native, and > .final_a -> .a.thinlto.native > - Updated list of ignored files in .gitignore > > Changelog since v2: > - Changed file suffixes: .o.thinlto.native -> .o_thinlto_native, > and .a.thinlto.native -> .a_thinlto_native so that basename > works as intended. > - Tested the patch with AutoFDO and Propeller. > --- > .gitignore | 3 ++ > MAINTAINERS | 5 +++ > Makefile | 40 ++++++++++++++++++++--- > arch/Kconfig | 19 +++++++++++ > scripts/Makefile.build | 52 +++++++++++++++++++++++++++--- > scripts/Makefile.lib | 7 +++- > scripts/Makefile.vmlinux_o | 16 +++++++--- > scripts/Makefile.vmlinux_thinlink | 53 +++++++++++++++++++++++++++++++ > scripts/head-object-list.txt | 1 + > 9 files changed, 181 insertions(+), 15 deletions(-) > create mode 100644 scripts/Makefile.vmlinux_thinlink I re-implemented the Makefiles to avoid the second recursion and hacky ifdefs. Attached. The topic branch is available in git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild.git thinlto-dist-refactor I only compile and boot tested on x86 QEMU. -- Best Regards Masahiro Yamada
On Mon, May 26, 2025 at 6:12 AM Masahiro Yamada <masahiroy@kernel.org> wrote: > > On Thu, May 22, 2025 at 6:35 AM <xur@google.com> wrote: > > > > From: Rong Xu <xur@google.com> > > > > Add distributed ThinLTO build support for the Linux kernel. > > This new mode offers several advantages: (1) Increased > > flexibility in handling user-specified build options. > > (2) Improved user-friendliness for developers. (3) Greater > > convenience for integrating with objtool and livepatch. > > > > Note that "distributed" in this context refers to a term > > that differentiates in-process ThinLTO builds by invoking > > backend compilation through the linker, not necessarily > > building in distributed environments. > > > > Distributed ThinLTO is enabled via the > > `CONFIG_LTO_CLANG_THIN_DIST` Kconfig option. For example: > > > make LLVM=1 defconfig > > > scripts/config -e LTO_CLANG_THIN_DIST > > > make LLVM=1 oldconfig > > > make LLVM=1 vmlinux -j <..> > > > > The implementation changes the top-level Makefile with a > > macro for generating `vmlinux.o` for distributed ThinLTO > > builds. It uses the existing Kbuild infrastructure to > > perform two recursive passes through the subdirectories. > > The first pass generates LLVM IR object files, similar to > > in-process ThinLTO. Following the thin-link stage, a second > > pass compiles these IR files into the final native object > > files. The build rules and actions for this two-pass process > > are primarily implemented in `scripts/Makefile.build`. > > > > Currently, this patch focuses on building the main kernel > > image (`vmlinux`) only. Support for building kernel modules > > using this method is planned for a subsequent patch. > > > > Tested on the following arch: x86, arm64, loongarch, and > > riscv. > > > > Some implementation details can be found here: > > https://discourse.llvm.org/t/rfc-distributed-thinlto-build-for-kernel/85934 > > > > Signed-off-by: Rong Xu <xur@google.com> > > --- > > Changelog since v1: > > - Updated the description in arch/Kconfig based on feedback > > from Nathan Chancellor > > - Revised file suffixes: .final_o -> .o.thinlto.native, and > > .final_a -> .a.thinlto.native > > - Updated list of ignored files in .gitignore > > > > Changelog since v2: > > - Changed file suffixes: .o.thinlto.native -> .o_thinlto_native, > > and .a.thinlto.native -> .a_thinlto_native so that basename > > works as intended. > > - Tested the patch with AutoFDO and Propeller. > > --- > > .gitignore | 3 ++ > > MAINTAINERS | 5 +++ > > Makefile | 40 ++++++++++++++++++++--- > > arch/Kconfig | 19 +++++++++++ > > scripts/Makefile.build | 52 +++++++++++++++++++++++++++--- > > scripts/Makefile.lib | 7 +++- > > scripts/Makefile.vmlinux_o | 16 +++++++--- > > scripts/Makefile.vmlinux_thinlink | 53 +++++++++++++++++++++++++++++++ > > scripts/head-object-list.txt | 1 + > > 9 files changed, 181 insertions(+), 15 deletions(-) > > create mode 100644 scripts/Makefile.vmlinux_thinlink > > I re-implemented the Makefiles to avoid > the second recursion and hacky ifdefs. > Attached. > > The topic branch is available in > > git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild.git > thinlto-dist-refactor > > I only compile and boot tested on x86 QEMU. This implementation does look cleaner. But with one issue: it has a unified cflags for all the BE compilations. This means per-file flags, such as CFLAGS_fork.o = -fabc, are lost during the BE compilation for fork.thinlto_native.o. This exact issue was what I aimed to prevent with the two-recursion approach. If we must avoid two recursions, perhaps we could leverage the saved per-file pre-link commands (i.e., .*.o.cmd)? Regards, -Rong > > > -- > Best Regards > Masahiro Yamada
On Wed, May 28, 2025 at 6:05 AM Rong Xu <xur@google.com> wrote: > > On Mon, May 26, 2025 at 6:12 AM Masahiro Yamada <masahiroy@kernel.org> wrote: > > > > On Thu, May 22, 2025 at 6:35 AM <xur@google.com> wrote: > > > > > > From: Rong Xu <xur@google.com> > > > > > > Add distributed ThinLTO build support for the Linux kernel. > > > This new mode offers several advantages: (1) Increased > > > flexibility in handling user-specified build options. > > > (2) Improved user-friendliness for developers. (3) Greater > > > convenience for integrating with objtool and livepatch. > > > > > > Note that "distributed" in this context refers to a term > > > that differentiates in-process ThinLTO builds by invoking > > > backend compilation through the linker, not necessarily > > > building in distributed environments. > > > > > > Distributed ThinLTO is enabled via the > > > `CONFIG_LTO_CLANG_THIN_DIST` Kconfig option. For example: > > > > make LLVM=1 defconfig > > > > scripts/config -e LTO_CLANG_THIN_DIST > > > > make LLVM=1 oldconfig > > > > make LLVM=1 vmlinux -j <..> > > > > > > The implementation changes the top-level Makefile with a > > > macro for generating `vmlinux.o` for distributed ThinLTO > > > builds. It uses the existing Kbuild infrastructure to > > > perform two recursive passes through the subdirectories. > > > The first pass generates LLVM IR object files, similar to > > > in-process ThinLTO. Following the thin-link stage, a second > > > pass compiles these IR files into the final native object > > > files. The build rules and actions for this two-pass process > > > are primarily implemented in `scripts/Makefile.build`. > > > > > > Currently, this patch focuses on building the main kernel > > > image (`vmlinux`) only. Support for building kernel modules > > > using this method is planned for a subsequent patch. > > > > > > Tested on the following arch: x86, arm64, loongarch, and > > > riscv. > > > > > > Some implementation details can be found here: > > > https://discourse.llvm.org/t/rfc-distributed-thinlto-build-for-kernel/85934 > > > > > > Signed-off-by: Rong Xu <xur@google.com> > > > --- > > > Changelog since v1: > > > - Updated the description in arch/Kconfig based on feedback > > > from Nathan Chancellor > > > - Revised file suffixes: .final_o -> .o.thinlto.native, and > > > .final_a -> .a.thinlto.native > > > - Updated list of ignored files in .gitignore > > > > > > Changelog since v2: > > > - Changed file suffixes: .o.thinlto.native -> .o_thinlto_native, > > > and .a.thinlto.native -> .a_thinlto_native so that basename > > > works as intended. > > > - Tested the patch with AutoFDO and Propeller. > > > --- > > > .gitignore | 3 ++ > > > MAINTAINERS | 5 +++ > > > Makefile | 40 ++++++++++++++++++++--- > > > arch/Kconfig | 19 +++++++++++ > > > scripts/Makefile.build | 52 +++++++++++++++++++++++++++--- > > > scripts/Makefile.lib | 7 +++- > > > scripts/Makefile.vmlinux_o | 16 +++++++--- > > > scripts/Makefile.vmlinux_thinlink | 53 +++++++++++++++++++++++++++++++ > > > scripts/head-object-list.txt | 1 + > > > 9 files changed, 181 insertions(+), 15 deletions(-) > > > create mode 100644 scripts/Makefile.vmlinux_thinlink > > > > I re-implemented the Makefiles to avoid > > the second recursion and hacky ifdefs. > > Attached. > > > > The topic branch is available in > > > > git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild.git > > thinlto-dist-refactor > > > > I only compile and boot tested on x86 QEMU. > This implementation does look cleaner. But with one issue: it has a > unified cflags for > all the BE compilations. This means per-file flags, such as > CFLAGS_fork.o = -fabc, > are lost during the BE compilation for fork.thinlto_native.o. > > This exact issue was what I aimed to prevent with the two-recursion approach. > If we must avoid two recursions, perhaps we could leverage the saved per-file > pre-link commands (i.e., .*.o.cmd)? How important is this? I do not know which compiler flags are consumed in the distributed thin lto stage. Are only compiler flags starting "-f" relevant? In your implementation, you filter-out many compiler flags. -- Best Regards Masahiro Yamada
On Fri, May 30, 2025 at 11:48 PM Masahiro Yamada <masahiroy@kernel.org> wrote: > > On Wed, May 28, 2025 at 6:05 AM Rong Xu <xur@google.com> wrote: > > > > On Mon, May 26, 2025 at 6:12 AM Masahiro Yamada <masahiroy@kernel.org> wrote: > > > > > > On Thu, May 22, 2025 at 6:35 AM <xur@google.com> wrote: > > > > > > > > From: Rong Xu <xur@google.com> > > > > > > > > Add distributed ThinLTO build support for the Linux kernel. > > > > This new mode offers several advantages: (1) Increased > > > > flexibility in handling user-specified build options. > > > > (2) Improved user-friendliness for developers. (3) Greater > > > > convenience for integrating with objtool and livepatch. > > > > > > > > Note that "distributed" in this context refers to a term > > > > that differentiates in-process ThinLTO builds by invoking > > > > backend compilation through the linker, not necessarily > > > > building in distributed environments. > > > > > > > > Distributed ThinLTO is enabled via the > > > > `CONFIG_LTO_CLANG_THIN_DIST` Kconfig option. For example: > > > > > make LLVM=1 defconfig > > > > > scripts/config -e LTO_CLANG_THIN_DIST > > > > > make LLVM=1 oldconfig > > > > > make LLVM=1 vmlinux -j <..> > > > > > > > > The implementation changes the top-level Makefile with a > > > > macro for generating `vmlinux.o` for distributed ThinLTO > > > > builds. It uses the existing Kbuild infrastructure to > > > > perform two recursive passes through the subdirectories. > > > > The first pass generates LLVM IR object files, similar to > > > > in-process ThinLTO. Following the thin-link stage, a second > > > > pass compiles these IR files into the final native object > > > > files. The build rules and actions for this two-pass process > > > > are primarily implemented in `scripts/Makefile.build`. > > > > > > > > Currently, this patch focuses on building the main kernel > > > > image (`vmlinux`) only. Support for building kernel modules > > > > using this method is planned for a subsequent patch. > > > > > > > > Tested on the following arch: x86, arm64, loongarch, and > > > > riscv. > > > > > > > > Some implementation details can be found here: > > > > https://discourse.llvm.org/t/rfc-distributed-thinlto-build-for-kernel/85934 > > > > > > > > Signed-off-by: Rong Xu <xur@google.com> > > > > --- > > > > Changelog since v1: > > > > - Updated the description in arch/Kconfig based on feedback > > > > from Nathan Chancellor > > > > - Revised file suffixes: .final_o -> .o.thinlto.native, and > > > > .final_a -> .a.thinlto.native > > > > - Updated list of ignored files in .gitignore > > > > > > > > Changelog since v2: > > > > - Changed file suffixes: .o.thinlto.native -> .o_thinlto_native, > > > > and .a.thinlto.native -> .a_thinlto_native so that basename > > > > works as intended. > > > > - Tested the patch with AutoFDO and Propeller. > > > > --- > > > > .gitignore | 3 ++ > > > > MAINTAINERS | 5 +++ > > > > Makefile | 40 ++++++++++++++++++++--- > > > > arch/Kconfig | 19 +++++++++++ > > > > scripts/Makefile.build | 52 +++++++++++++++++++++++++++--- > > > > scripts/Makefile.lib | 7 +++- > > > > scripts/Makefile.vmlinux_o | 16 +++++++--- > > > > scripts/Makefile.vmlinux_thinlink | 53 +++++++++++++++++++++++++++++++ > > > > scripts/head-object-list.txt | 1 + > > > > 9 files changed, 181 insertions(+), 15 deletions(-) > > > > create mode 100644 scripts/Makefile.vmlinux_thinlink > > > > > > I re-implemented the Makefiles to avoid > > > the second recursion and hacky ifdefs. > > > Attached. > > > > > > The topic branch is available in > > > > > > git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild.git > > > thinlto-dist-refactor > > > > > > I only compile and boot tested on x86 QEMU. > > This implementation does look cleaner. But with one issue: it has a > > unified cflags for > > all the BE compilations. This means per-file flags, such as > > CFLAGS_fork.o = -fabc, > > are lost during the BE compilation for fork.thinlto_native.o. > > > > This exact issue was what I aimed to prevent with the two-recursion approach. > > If we must avoid two recursions, perhaps we could leverage the saved per-file > > pre-link commands (i.e., .*.o.cmd)? > > How important is this? It is one of the main reasons for using distributed mode. For example, for some source, the user chooses to use -O1 (or disable some loop transformations), there is no way to do this with one-unified options. > > I do not know which compiler flags are consumed in the > distributed thin lto stage. > > Are only compiler flags starting "-f" relevant? All the flags that have impact on BE compilation need to pass to BE. Yes, usually most -f flags are relevant, all the internal options, like -mllvm -foo need to be passed to BE. -D, -i, -include are OK to filter. > > In your implementation, you filter-out many compiler flags. These are flags for the compiler front-end (clang). They don't have impact on BE compilation. > > > -- > Best Regards > Masahiro Yamada
On Thu, May 22, 2025 at 6:35 AM <xur@google.com> wrote:
> +ifdef CONFIG_LTO_CLANG_THIN_DIST
> +# Generate .o_thinlto_native (obj) from .o (bitcode) file
> +# ---------------------------------------------------------------------------
> +quiet_cmd_cc_o_bc = CC $(quiet_modtag) $@
> +
> +cmd_cc_o_bc = $(if $(filter bitcode, $(shell file -b $<)),$(CC) \
> + $(filter-out -Wp% $(LINUXINCLUDE) %.h.gch %.h -D% \
> + -flto=thin, $(c_flags)) \
I think this filter-out is fragile.
I think it is better to use $(_c_flags), like this:
$(filter-out -flto=thin, $(_c_flags))
> diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
> index 2fe73cda0bddb..9cfd23590334d 100644
> --- a/scripts/Makefile.lib
> +++ b/scripts/Makefile.lib
> @@ -34,8 +34,13 @@ else
> obj-m := $(filter-out %/, $(obj-m))
> endif
>
> +builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a)
> +ifeq ($(thinlto_final_pass),1)
> + builtin_suffix :=.a_thinlto_native
> +endif
> +
> ifdef need-builtin
> -obj-y := $(patsubst %/, %/built-in.a, $(obj-y))
> +obj-y := $(patsubst %/, %/built-in$(builtin_suffix), $(obj-y))
> else
> obj-y := $(filter-out %/, $(obj-y))
> endif
> diff --git a/scripts/Makefile.vmlinux_o b/scripts/Makefile.vmlinux_o
> index b024ffb3e2018..f9abc45a68b36 100644
> --- a/scripts/Makefile.vmlinux_o
> +++ b/scripts/Makefile.vmlinux_o
> @@ -9,6 +9,14 @@ include $(srctree)/scripts/Kbuild.include
> # for objtool
> include $(srctree)/scripts/Makefile.lib
>
> +ifeq ($(thinlto_final_pass),1)
> +vmlinux_a := vmlinux.a_thinlto_native
> +vmlinux_libs := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_LIBS))
> +else
> +vmlinux_a := vmlinux.a
> +vmlinux_libs := $(KBUILD_VMLINUX_LIBS)
> +endif
> +
> # Generate a linker script to ensure correct ordering of initcalls for Clang LTO
> # ---------------------------------------------------------------------------
>
> @@ -18,7 +26,7 @@ quiet_cmd_gen_initcalls_lds = GEN $@
> $(PERL) $(real-prereqs) > $@
>
> .tmp_initcalls.lds: $(srctree)/scripts/generate_initcall_order.pl \
> - vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> + $(vmlinux_a) $(vmlinux_libs) FORCE
> $(call if_changed,gen_initcalls_lds)
>
> targets := .tmp_initcalls.lds
> @@ -59,8 +67,8 @@ quiet_cmd_ld_vmlinux.o = LD $@
> $(LD) ${KBUILD_LDFLAGS} -r -o $@ \
> $(vmlinux-o-ld-args-y) \
> $(addprefix -T , $(initcalls-lds)) \
> - --whole-archive vmlinux.a --no-whole-archive \
> - --start-group $(KBUILD_VMLINUX_LIBS) --end-group \
> + --whole-archive $(vmlinux_a) --no-whole-archive \
> + --start-group $(vmlinux_libs) --end-group \
> $(cmd_objtool)
>
> define rule_ld_vmlinux.o
> @@ -68,7 +76,7 @@ define rule_ld_vmlinux.o
> $(call cmd,gen_objtooldep)
> endef
>
> -vmlinux.o: $(initcalls-lds) vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> +vmlinux.o: $(initcalls-lds) $(vmlinux_a) $(vmlinux_libs) FORCE
> $(call if_changed_rule,ld_vmlinux.o)
>
> targets += vmlinux.o
> diff --git a/scripts/Makefile.vmlinux_thinlink b/scripts/Makefile.vmlinux_thinlink
> new file mode 100644
> index 0000000000000..13e4026c7d45b
> --- /dev/null
> +++ b/scripts/Makefile.vmlinux_thinlink
> @@ -0,0 +1,53 @@
> +# SPDX-License-Identifier: GPL-2.0-only
> +
> +PHONY := __default
> +__default: vmlinux.thinlink
> +
> +include include/config/auto.conf
> +include $(srctree)/scripts/Kbuild.include
> +
> +
> +# Generate a linker script to ensure correct ordering of initcalls for Clang LTO
> +# ---------------------------------------------------------------------------
> +
> +quiet_cmd_gen_initcalls_lds = GEN $@
> + cmd_gen_initcalls_lds = \
> + $(PYTHON3) $(srctree)/scripts/jobserver-exec \
> + $(PERL) $(real-prereqs) > $@
This is a copy-paste from scripts/Makefile.vmlinux_o.
Is this linker script critical when generating the index file?
--
Best Regards
Masahiro Yamada
On Mon, May 26, 2025 at 2:05 AM Masahiro Yamada <masahiroy@kernel.org> wrote:
>
> On Thu, May 22, 2025 at 6:35 AM <xur@google.com> wrote:
>
> > +ifdef CONFIG_LTO_CLANG_THIN_DIST
> > +# Generate .o_thinlto_native (obj) from .o (bitcode) file
> > +# ---------------------------------------------------------------------------
> > +quiet_cmd_cc_o_bc = CC $(quiet_modtag) $@
> > +
> > +cmd_cc_o_bc = $(if $(filter bitcode, $(shell file -b $<)),$(CC) \
> > + $(filter-out -Wp% $(LINUXINCLUDE) %.h.gch %.h -D% \
> > + -flto=thin, $(c_flags)) \
>
> I think this filter-out is fragile.
>
> I think it is better to use $(_c_flags), like this:
>
> $(filter-out -flto=thin, $(_c_flags))
Ack.
>
>
>
>
>
>
>
>
>
>
>
>
> > diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
> > index 2fe73cda0bddb..9cfd23590334d 100644
> > --- a/scripts/Makefile.lib
> > +++ b/scripts/Makefile.lib
> > @@ -34,8 +34,13 @@ else
> > obj-m := $(filter-out %/, $(obj-m))
> > endif
> >
> > +builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a)
> > +ifeq ($(thinlto_final_pass),1)
> > + builtin_suffix :=.a_thinlto_native
> > +endif
> > +
> > ifdef need-builtin
> > -obj-y := $(patsubst %/, %/built-in.a, $(obj-y))
> > +obj-y := $(patsubst %/, %/built-in$(builtin_suffix), $(obj-y))
> > else
> > obj-y := $(filter-out %/, $(obj-y))
> > endif
> > diff --git a/scripts/Makefile.vmlinux_o b/scripts/Makefile.vmlinux_o
> > index b024ffb3e2018..f9abc45a68b36 100644
> > --- a/scripts/Makefile.vmlinux_o
> > +++ b/scripts/Makefile.vmlinux_o
> > @@ -9,6 +9,14 @@ include $(srctree)/scripts/Kbuild.include
> > # for objtool
> > include $(srctree)/scripts/Makefile.lib
> >
> > +ifeq ($(thinlto_final_pass),1)
> > +vmlinux_a := vmlinux.a_thinlto_native
> > +vmlinux_libs := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_LIBS))
> > +else
> > +vmlinux_a := vmlinux.a
> > +vmlinux_libs := $(KBUILD_VMLINUX_LIBS)
> > +endif
> > +
> > # Generate a linker script to ensure correct ordering of initcalls for Clang LTO
> > # ---------------------------------------------------------------------------
> >
> > @@ -18,7 +26,7 @@ quiet_cmd_gen_initcalls_lds = GEN $@
> > $(PERL) $(real-prereqs) > $@
> >
> > .tmp_initcalls.lds: $(srctree)/scripts/generate_initcall_order.pl \
> > - vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> > + $(vmlinux_a) $(vmlinux_libs) FORCE
> > $(call if_changed,gen_initcalls_lds)
> >
> > targets := .tmp_initcalls.lds
> > @@ -59,8 +67,8 @@ quiet_cmd_ld_vmlinux.o = LD $@
> > $(LD) ${KBUILD_LDFLAGS} -r -o $@ \
> > $(vmlinux-o-ld-args-y) \
> > $(addprefix -T , $(initcalls-lds)) \
> > - --whole-archive vmlinux.a --no-whole-archive \
> > - --start-group $(KBUILD_VMLINUX_LIBS) --end-group \
> > + --whole-archive $(vmlinux_a) --no-whole-archive \
> > + --start-group $(vmlinux_libs) --end-group \
> > $(cmd_objtool)
> >
> > define rule_ld_vmlinux.o
> > @@ -68,7 +76,7 @@ define rule_ld_vmlinux.o
> > $(call cmd,gen_objtooldep)
> > endef
> >
> > -vmlinux.o: $(initcalls-lds) vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> > +vmlinux.o: $(initcalls-lds) $(vmlinux_a) $(vmlinux_libs) FORCE
> > $(call if_changed_rule,ld_vmlinux.o)
> >
> > targets += vmlinux.o
> > diff --git a/scripts/Makefile.vmlinux_thinlink b/scripts/Makefile.vmlinux_thinlink
> > new file mode 100644
> > index 0000000000000..13e4026c7d45b
> > --- /dev/null
> > +++ b/scripts/Makefile.vmlinux_thinlink
> > @@ -0,0 +1,53 @@
> > +# SPDX-License-Identifier: GPL-2.0-only
> > +
> > +PHONY := __default
> > +__default: vmlinux.thinlink
> > +
> > +include include/config/auto.conf
> > +include $(srctree)/scripts/Kbuild.include
> > +
> > +
> > +# Generate a linker script to ensure correct ordering of initcalls for Clang LTO
> > +# ---------------------------------------------------------------------------
> > +
> > +quiet_cmd_gen_initcalls_lds = GEN $@
> > + cmd_gen_initcalls_lds = \
> > + $(PYTHON3) $(srctree)/scripts/jobserver-exec \
> > + $(PERL) $(real-prereqs) > $@
>
>
> This is a copy-paste from scripts/Makefile.vmlinux_o.
>
> Is this linker script critical when generating the index file?
Our usual guideline for the ThinLTO indexing command is to use the
same arguments
as the final link. For this particular case, it probably does not
matter (i.e we can skip the linker script option).
>
>
>
> --
> Best Regards
> Masahiro Yamada
Hi,
kernel test robot noticed the following build warnings:
[auto build test WARNING on masahiroy-kbuild/for-next]
[also build test WARNING on masahiroy-kbuild/fixes linus/master v6.15-rc7 next-20250523]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]
url: https://github.com/intel-lab-lkp/linux/commits/xur-google-com/kbuild-distributed-build-support-for-Clang-ThinLTO/20250522-053726
base: https://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild.git for-next
patch link: https://lore.kernel.org/r/20250521213534.3159514-1-xur%40google.com
patch subject: [PATCH v3] kbuild: distributed build support for Clang ThinLTO
config: arm64-randconfig-r131-20250525 (https://download.01.org/0day-ci/archive/20250525/202505251836.qXs81x9F-lkp@intel.com/config)
compiler: clang version 21.0.0git (https://github.com/llvm/llvm-project f819f46284f2a79790038e1f6649172789734ae8)
reproduce: (https://download.01.org/0day-ci/archive/20250525/202505251836.qXs81x9F-lkp@intel.com/reproduce)
If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202505251836.qXs81x9F-lkp@intel.com/
sparse warnings: (new ones prefixed by >>)
>> mm/sparse-vmemmap.c:442:21: sparse: sparse: cast to non-scalar
>> mm/sparse-vmemmap.c:442:21: sparse: sparse: cast from non-scalar
mm/sparse-vmemmap.c: note: in included file (through include/linux/pgtable.h, include/linux/mm.h):
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast to non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast from non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast to non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast from non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast to non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast from non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast to non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast from non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast to non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast from non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast to non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast from non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast to non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast from non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast to non-scalar
arch/arm64/include/asm/pgtable.h:339:16: sparse: sparse: cast from non-scalar
vim +442 mm/sparse-vmemmap.c
2045a3b8911b6e Feiyang Chen 2022-10-27 415
2045a3b8911b6e Feiyang Chen 2022-10-27 416 int __meminit vmemmap_populate_hugepages(unsigned long start, unsigned long end,
2045a3b8911b6e Feiyang Chen 2022-10-27 417 int node, struct vmem_altmap *altmap)
2045a3b8911b6e Feiyang Chen 2022-10-27 418 {
2045a3b8911b6e Feiyang Chen 2022-10-27 419 unsigned long addr;
2045a3b8911b6e Feiyang Chen 2022-10-27 420 unsigned long next;
2045a3b8911b6e Feiyang Chen 2022-10-27 421 pgd_t *pgd;
2045a3b8911b6e Feiyang Chen 2022-10-27 422 p4d_t *p4d;
2045a3b8911b6e Feiyang Chen 2022-10-27 423 pud_t *pud;
2045a3b8911b6e Feiyang Chen 2022-10-27 424 pmd_t *pmd;
2045a3b8911b6e Feiyang Chen 2022-10-27 425
2045a3b8911b6e Feiyang Chen 2022-10-27 426 for (addr = start; addr < end; addr = next) {
2045a3b8911b6e Feiyang Chen 2022-10-27 427 next = pmd_addr_end(addr, end);
2045a3b8911b6e Feiyang Chen 2022-10-27 428
2045a3b8911b6e Feiyang Chen 2022-10-27 429 pgd = vmemmap_pgd_populate(addr, node);
2045a3b8911b6e Feiyang Chen 2022-10-27 430 if (!pgd)
2045a3b8911b6e Feiyang Chen 2022-10-27 431 return -ENOMEM;
2045a3b8911b6e Feiyang Chen 2022-10-27 432
2045a3b8911b6e Feiyang Chen 2022-10-27 433 p4d = vmemmap_p4d_populate(pgd, addr, node);
2045a3b8911b6e Feiyang Chen 2022-10-27 434 if (!p4d)
2045a3b8911b6e Feiyang Chen 2022-10-27 435 return -ENOMEM;
2045a3b8911b6e Feiyang Chen 2022-10-27 436
2045a3b8911b6e Feiyang Chen 2022-10-27 437 pud = vmemmap_pud_populate(p4d, addr, node);
2045a3b8911b6e Feiyang Chen 2022-10-27 438 if (!pud)
2045a3b8911b6e Feiyang Chen 2022-10-27 439 return -ENOMEM;
2045a3b8911b6e Feiyang Chen 2022-10-27 440
2045a3b8911b6e Feiyang Chen 2022-10-27 441 pmd = pmd_offset(pud, addr);
2045a3b8911b6e Feiyang Chen 2022-10-27 @442 if (pmd_none(READ_ONCE(*pmd))) {
2045a3b8911b6e Feiyang Chen 2022-10-27 443 void *p;
2045a3b8911b6e Feiyang Chen 2022-10-27 444
2045a3b8911b6e Feiyang Chen 2022-10-27 445 p = vmemmap_alloc_block_buf(PMD_SIZE, node, altmap);
2045a3b8911b6e Feiyang Chen 2022-10-27 446 if (p) {
2045a3b8911b6e Feiyang Chen 2022-10-27 447 vmemmap_set_pmd(pmd, p, node, addr, next);
2045a3b8911b6e Feiyang Chen 2022-10-27 448 continue;
2045a3b8911b6e Feiyang Chen 2022-10-27 449 } else if (altmap) {
2045a3b8911b6e Feiyang Chen 2022-10-27 450 /*
2045a3b8911b6e Feiyang Chen 2022-10-27 451 * No fallback: In any case we care about, the
2045a3b8911b6e Feiyang Chen 2022-10-27 452 * altmap should be reasonably sized and aligned
2045a3b8911b6e Feiyang Chen 2022-10-27 453 * such that vmemmap_alloc_block_buf() will always
2045a3b8911b6e Feiyang Chen 2022-10-27 454 * succeed. For consistency with the PTE case,
2045a3b8911b6e Feiyang Chen 2022-10-27 455 * return an error here as failure could indicate
2045a3b8911b6e Feiyang Chen 2022-10-27 456 * a configuration issue with the size of the altmap.
2045a3b8911b6e Feiyang Chen 2022-10-27 457 */
2045a3b8911b6e Feiyang Chen 2022-10-27 458 return -ENOMEM;
2045a3b8911b6e Feiyang Chen 2022-10-27 459 }
2045a3b8911b6e Feiyang Chen 2022-10-27 460 } else if (vmemmap_check_pmd(pmd, node, addr, next))
2045a3b8911b6e Feiyang Chen 2022-10-27 461 continue;
2045a3b8911b6e Feiyang Chen 2022-10-27 462 if (vmemmap_populate_basepages(addr, next, node, altmap))
2045a3b8911b6e Feiyang Chen 2022-10-27 463 return -ENOMEM;
2045a3b8911b6e Feiyang Chen 2022-10-27 464 }
2045a3b8911b6e Feiyang Chen 2022-10-27 465 return 0;
2045a3b8911b6e Feiyang Chen 2022-10-27 466 }
2045a3b8911b6e Feiyang Chen 2022-10-27 467
--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
On Thu, May 22, 2025 at 6:35 AM <xur@google.com> wrote:
>
> From: Rong Xu <xur@google.com>
>
> Add distributed ThinLTO build support for the Linux kernel.
> This new mode offers several advantages: (1) Increased
> flexibility in handling user-specified build options.
> (2) Improved user-friendliness for developers. (3) Greater
> convenience for integrating with objtool and livepatch.
I did not set up for a distributed build environment, but
does CC='distcc clang' work to process thin LTO
on back-end build machines?
You mentioned the benefits (3), but you did not touch
anything about objtool.
How will this patch change the objtool integration?
>
> Note that "distributed" in this context refers to a term
> that differentiates in-process ThinLTO builds by invoking
> backend compilation through the linker, not necessarily
> building in distributed environments.
>
> Distributed ThinLTO is enabled via the
> `CONFIG_LTO_CLANG_THIN_DIST` Kconfig option. For example:
> > make LLVM=1 defconfig
> > scripts/config -e LTO_CLANG_THIN_DIST
> > make LLVM=1 oldconfig
> > make LLVM=1 vmlinux -j <..>
>
> The implementation changes the top-level Makefile with a
> macro for generating `vmlinux.o` for distributed ThinLTO
> builds. It uses the existing Kbuild infrastructure to
> perform two recursive passes through the subdirectories.
> The first pass generates LLVM IR object files, similar to
> in-process ThinLTO. Following the thin-link stage, a second
> pass compiles these IR files into the final native object
> files. The build rules and actions for this two-pass process
> are primarily implemented in `scripts/Makefile.build`.
>
> Currently, this patch focuses on building the main kernel
> image (`vmlinux`) only. Support for building kernel modules
> using this method is planned for a subsequent patch.
>
> Tested on the following arch: x86, arm64, loongarch, and
> riscv.
>
> Some implementation details can be found here:
> https://discourse.llvm.org/t/rfc-distributed-thinlto-build-for-kernel/85934
>
> Signed-off-by: Rong Xu <xur@google.com>
> ---
> Changelog since v1:
> - Updated the description in arch/Kconfig based on feedback
> from Nathan Chancellor
> - Revised file suffixes: .final_o -> .o.thinlto.native, and
> .final_a -> .a.thinlto.native
> - Updated list of ignored files in .gitignore
>
> Changelog since v2:
> - Changed file suffixes: .o.thinlto.native -> .o_thinlto_native,
> and .a.thinlto.native -> .a_thinlto_native so that basename
> works as intended.
> - Tested the patch with AutoFDO and Propeller.
> ---
> diff --git a/.gitignore b/.gitignore
> index f2f63e47fb886..b83a68185ef46 100644
> --- a/.gitignore
> +++ b/.gitignore
> @@ -12,6 +12,7 @@
> #
> .*
> *.a
> +*.a_thinlto_native
This is unneeded.
> *.asn1.[ch]
> *.bin
> *.bz2
> @@ -39,6 +40,7 @@
> *.mod.c
> *.o
> *.o.*
> +*.o_thinlto_native
I would rename this to *.thinlto-native.o
so we do not need to touch .gitignore at all.
> *.patch
> *.rmeta
> *.rpm
> @@ -64,6 +66,7 @@ modules.order
> /vmlinux
> /vmlinux.32
> /vmlinux.map
> +/vmlinux.thinlink
> /vmlinux.symvers
> /vmlinux.unstripped
> /vmlinux-gdb.py
> diff --git a/MAINTAINERS b/MAINTAINERS
> index d48dd6726fe6b..f54090f364c93 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -5790,6 +5790,11 @@ F: scripts/Makefile.clang
> F: scripts/clang-tools/
> K: \b(?i:clang|llvm)\b
>
> +CLANG/LLVM THINLTO DISTRIBUTED BUILD
> +M: Rong Xu <xur@google.com>
> +S: Supported
> +F: scripts/Makefile.vmlinux_thinlink
> +
> CLK API
> M: Russell King <linux@armlinux.org.uk>
> L: linux-clk@vger.kernel.org
> diff --git a/Makefile b/Makefile
> index a9edd03036537..8fbff2ab87ebd 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -298,7 +298,8 @@ no-dot-config-targets := $(clean-targets) \
> outputmakefile rustavailable rustfmt rustfmtcheck
> no-sync-config-targets := $(no-dot-config-targets) %install modules_sign kernelrelease \
> image_name
> -single-targets := %.a %.i %.ko %.lds %.ll %.lst %.mod %.o %.rsi %.s %/
> +single-targets := %.a %.i %.ko %.lds %.ll %.lst %.mod %.o %.rsi %.s %.o_thinlto_native \
> + %.a_thinlto_native %.o.thinlto.bc %/
You are adding all new suffixes, but in my understanding there is no way
to generate %.o.thinlto.bc as a single target.
>
> config-build :=
> mixed-build :=
> @@ -991,10 +992,10 @@ export CC_FLAGS_SCS
> endif
>
> ifdef CONFIG_LTO_CLANG
> -ifdef CONFIG_LTO_CLANG_THIN
> -CC_FLAGS_LTO := -flto=thin -fsplit-lto-unit
> -else
> +ifdef CONFIG_LTO_CLANG_FULL
> CC_FLAGS_LTO := -flto
> +else # for CONFIG_LTO_CLANG_THIN or CONFIG_LTO_CLANG_THIN_DIST
> +CC_FLAGS_LTO := -flto=thin -fsplit-lto-unit
> endif
> CC_FLAGS_LTO += -fvisibility=hidden
>
> @@ -1213,8 +1214,34 @@ vmlinux.a: $(KBUILD_VMLINUX_OBJS) scripts/head-object-list.txt FORCE
> $(call if_changed,ar_vmlinux.a)
>
> PHONY += vmlinux_o
> +ifdef CONFIG_LTO_CLANG_THIN_DIST
> +vmlinux.thinlink: vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> + $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.vmlinux_thinlink
> +targets += vmlinux.thinlink
> +
> +vmlinux.a_thinlto_native := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_OBJS))
> +quiet_cmd_ar_vmlinux.a_thinlto_native = AR $@
> + cmd_ar_vmlinux.a_thinlto_native = \
> + rm -f $@; \
> + $(AR) cDPrST $@ $(vmlinux.a_thinlto_native); \
> + $(AR) mPiT $$($(AR) t $@ | sed -n 1p) $@ $$($(AR) t $@ | grep -F -f $(srctree)/scripts/head-object-list.txt)
> +
> +define rule_gen_vmlinux.a_thinlto_native
> + +$(Q)$(MAKE) $(build)=. need-builtin=1 thinlto_final_pass=1 need-modorder=1 built-in.a_thinlto_native
> + $(call cmd_and_savecmd,ar_vmlinux.a_thinlto_native)
> +endef
> +
> +vmlinux.a_thinlto_native: vmlinux.thinlink scripts/head-object-list.txt FORCE
> + $(call if_changed_rule,gen_vmlinux.a_thinlto_native)
> +
> +targets += vmlinux.a_thinlto_native
> +
> +vmlinux_o: vmlinux.a_thinlto_native
> + $(Q)$(MAKE) thinlto_final_pass=1 -f $(srctree)/scripts/Makefile.vmlinux_o
> +else
> vmlinux_o: vmlinux.a $(KBUILD_VMLINUX_LIBS)
> $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.vmlinux_o
> +endif
>
> vmlinux.o modules.builtin.modinfo modules.builtin: vmlinux_o
> @:
> @@ -1572,7 +1599,8 @@ CLEAN_FILES += vmlinux.symvers modules-only.symvers \
> modules.builtin.ranges vmlinux.o.map vmlinux.unstripped \
> compile_commands.json rust/test \
> rust-project.json .vmlinux.objs .vmlinux.export.c \
> - .builtin-dtbs-list .builtin-dtb.S
> + .builtin-dtbs-list .builtin-dtb.S \
> + .vmlinux_thinlto_bc_files vmlinux.thinlink
>
> # Directories & files removed with 'make mrproper'
> MRPROPER_FILES += include/config include/generated \
> @@ -2023,6 +2051,8 @@ clean: $(clean-dirs)
> -o -name '*.symtypes' -o -name 'modules.order' \
> -o -name '*.c.[012]*.*' \
> -o -name '*.ll' \
> + -o -name '*.a_thinlto_native' -o -name '*.o_thinlto_native' \
> + -o -name '*.o.thinlto.bc' \
> -o -name '*.gcno' \
> \) -type f -print \
> -o -name '.tmp_*' -print \
> diff --git a/arch/Kconfig b/arch/Kconfig
> index b0adb665041f1..30dccda07c671 100644
> --- a/arch/Kconfig
> +++ b/arch/Kconfig
> @@ -810,6 +810,25 @@ config LTO_CLANG_THIN
> https://clang.llvm.org/docs/ThinLTO.html
>
> If unsure, say Y.
> +
> +config LTO_CLANG_THIN_DIST
> + bool "Clang ThinLTO in distributed mode (EXPERIMENTAL)"
> + depends on HAS_LTO_CLANG && ARCH_SUPPORTS_LTO_CLANG_THIN
> + select LTO_CLANG
> + help
> + This option enables Clang's ThinLTO in distributed build mode.
> + In this mode, the linker performs the thin-link, generating
> + ThinLTO index files. Subsequently, the build system explicitly
> + invokes ThinLTO backend compilation using these index files
> + and pre-linked IR objects. The resulting native object files
> + are with the .o_thinlto_native suffix.
> +
> + This build mode offers improved visibility into the ThinLTO
> + process through explicit subcommand exposure. It also makes
> + final native object files directly available, benefiting
> + tools like objtool and kpatch. Additionally, it provides
> + crucial granular control over back-end options, enabling
> + module-specific compiler options, and simplifies debugging.
> endchoice
>
> config ARCH_SUPPORTS_AUTOFDO_CLANG
> diff --git a/scripts/Makefile.build b/scripts/Makefile.build
> index 13dcd86e74ca8..338e1aec0eaa3 100644
> --- a/scripts/Makefile.build
> +++ b/scripts/Makefile.build
> @@ -50,18 +50,23 @@ endif
>
> # ===========================================================================
>
> +builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a)
> +ifeq ($(thinlto_final_pass),1)
> +builtin_suffix :=.a_thinlto_native
> +endif
> +
> # subdir-builtin and subdir-modorder may contain duplications. Use $(sort ...)
> -subdir-builtin := $(sort $(filter %/built-in.a, $(real-obj-y)))
> +subdir-builtin := $(sort $(filter %/built-in$(builtin_suffix), $(real-obj-y)))
> subdir-modorder := $(sort $(filter %/modules.order, $(obj-m)))
>
> targets-for-builtin := $(extra-y)
>
> ifneq ($(strip $(lib-y) $(lib-m) $(lib-)),)
> -targets-for-builtin += $(obj)/lib.a
> +targets-for-builtin += $(obj)/lib$(builtin_suffix)
> endif
>
> ifdef need-builtin
> -targets-for-builtin += $(obj)/built-in.a
> +targets-for-builtin += $(obj)/built-in$(builtin_suffix)
> endif
>
> targets-for-modules := $(foreach x, o mod, \
> @@ -337,6 +342,10 @@ $(obj)/%.o: $(obj)/%.S FORCE
> targets += $(filter-out $(subdir-builtin), $(real-obj-y))
> targets += $(filter-out $(subdir-modorder), $(real-obj-m))
> targets += $(lib-y) $(always-y)
> +ifeq ($(builtin_suffix),.a_thinlto_native)
> +native_targets = $(patsubst,%.o,%.o_thinlto_native,$(targets))
> +targets += $(native_targets)
> +endif
>
> # Linker scripts preprocessor (.lds.S -> .lds)
> # ---------------------------------------------------------------------------
> @@ -347,6 +356,24 @@ quiet_cmd_cpp_lds_S = LDS $@
> $(obj)/%.lds: $(src)/%.lds.S FORCE
> $(call if_changed_dep,cpp_lds_S)
>
> +ifdef CONFIG_LTO_CLANG_THIN_DIST
> +# Generate .o_thinlto_native (obj) from .o (bitcode) file
> +# ---------------------------------------------------------------------------
> +quiet_cmd_cc_o_bc = CC $(quiet_modtag) $@
> +
> +cmd_cc_o_bc = $(if $(filter bitcode, $(shell file -b $<)),$(CC) \
> + $(filter-out -Wp% $(LINUXINCLUDE) %.h.gch %.h -D% \
> + -flto=thin, $(c_flags)) \
> + -Wno-unused-command-line-argument \
> + -x ir -fthinlto-index=$<.thinlto.bc -c -o $@ \
> + $(if $(findstring ../,$<), \
> + $$(realpath --relative-to=$(srcroot) $<), $<), \
> + cp $< $@)
> +
> +$(obj)/%.o_thinlto_native: $(obj)/%.o FORCE
> + $(call if_changed,cc_o_bc)
> +endif
> +
> # ASN.1 grammar
> # ---------------------------------------------------------------------------
> quiet_cmd_asn1_compiler = ASN.1 $(basename $@).[ch]
> @@ -360,7 +387,7 @@ $(obj)/%.asn1.c $(obj)/%.asn1.h: $(src)/%.asn1 $(objtree)/scripts/asn1_compiler
> # ---------------------------------------------------------------------------
>
> # To build objects in subdirs, we need to descend into the directories
> -$(subdir-builtin): $(obj)/%/built-in.a: $(obj)/% ;
> +$(subdir-builtin): $(obj)/%/built-in$(builtin_suffix): $(obj)/% ;
> $(subdir-modorder): $(obj)/%/modules.order: $(obj)/% ;
>
> #
> @@ -377,6 +404,12 @@ quiet_cmd_ar_builtin = AR $@
> $(obj)/built-in.a: $(real-obj-y) FORCE
> $(call if_changed,ar_builtin)
>
> +ifdef CONFIG_LTO_CLANG_THIN_DIST
> +# Rule to compile a set of .o_thinlto_native files into one .a_thinlto_native file.
> +$(obj)/built-in.a_thinlto_native: $(patsubst %.o,%.o_thinlto_native,$(real-obj-y)) FORCE
> + $(call if_changed,ar_builtin)
> +endif
> +
> # This is a list of build artifacts from the current Makefile and its
> # sub-directories. The timestamp should be updated when any of the member files.
>
> @@ -394,6 +427,14 @@ $(obj)/modules.order: $(obj-m) FORCE
> $(obj)/lib.a: $(lib-y) FORCE
> $(call if_changed,ar)
>
> +ifdef CONFIG_LTO_CLANG_THIN_DIST
> +quiet_cmd_ar_native = AR $@
> + cmd_ar_native = rm -f $@; $(AR) cDPrsT $@ $(patsubst %.o,%.o_thinlto_native,$(real-prereqs))
> +
> +$(obj)/lib.a_thinlto_native: $(patsubst %.o,%.o_thinlto_native,$(lib-y)) FORCE
> + $(call if_changed,ar_native)
> +endif
> +
> quiet_cmd_ld_multi_m = LD [M] $@
> cmd_ld_multi_m = $(LD) $(ld_flags) -r -o $@ @$< $(cmd_objtool)
>
> @@ -459,7 +500,8 @@ $(single-subdir-goals): $(single-subdirs)
> PHONY += $(subdir-ym)
> $(subdir-ym):
> $(Q)$(MAKE) $(build)=$@ \
> - need-builtin=$(if $(filter $@/built-in.a, $(subdir-builtin)),1) \
> + need-builtin=$(if $(filter $@/built-in$(builtin_suffix), $(subdir-builtin)),1) \
> + thinlto_final_pass=$(if $(filter .a_thinlto_native, $(builtin_suffix)),1) \
> need-modorder=$(if $(filter $@/modules.order, $(subdir-modorder)),1) \
> $(filter $@/%, $(single-subdir-goals))
>
> diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
> index 2fe73cda0bddb..9cfd23590334d 100644
> --- a/scripts/Makefile.lib
> +++ b/scripts/Makefile.lib
> @@ -34,8 +34,13 @@ else
> obj-m := $(filter-out %/, $(obj-m))
> endif
>
> +builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a)
> +ifeq ($(thinlto_final_pass),1)
> + builtin_suffix :=.a_thinlto_native
> +endif
> +
> ifdef need-builtin
> -obj-y := $(patsubst %/, %/built-in.a, $(obj-y))
> +obj-y := $(patsubst %/, %/built-in$(builtin_suffix), $(obj-y))
> else
> obj-y := $(filter-out %/, $(obj-y))
> endif
> diff --git a/scripts/Makefile.vmlinux_o b/scripts/Makefile.vmlinux_o
> index b024ffb3e2018..f9abc45a68b36 100644
> --- a/scripts/Makefile.vmlinux_o
> +++ b/scripts/Makefile.vmlinux_o
> @@ -9,6 +9,14 @@ include $(srctree)/scripts/Kbuild.include
> # for objtool
> include $(srctree)/scripts/Makefile.lib
>
> +ifeq ($(thinlto_final_pass),1)
> +vmlinux_a := vmlinux.a_thinlto_native
> +vmlinux_libs := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_LIBS))
> +else
> +vmlinux_a := vmlinux.a
> +vmlinux_libs := $(KBUILD_VMLINUX_LIBS)
> +endif
> +
> # Generate a linker script to ensure correct ordering of initcalls for Clang LTO
> # ---------------------------------------------------------------------------
>
> @@ -18,7 +26,7 @@ quiet_cmd_gen_initcalls_lds = GEN $@
> $(PERL) $(real-prereqs) > $@
>
> .tmp_initcalls.lds: $(srctree)/scripts/generate_initcall_order.pl \
> - vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> + $(vmlinux_a) $(vmlinux_libs) FORCE
> $(call if_changed,gen_initcalls_lds)
>
> targets := .tmp_initcalls.lds
> @@ -59,8 +67,8 @@ quiet_cmd_ld_vmlinux.o = LD $@
> $(LD) ${KBUILD_LDFLAGS} -r -o $@ \
> $(vmlinux-o-ld-args-y) \
> $(addprefix -T , $(initcalls-lds)) \
> - --whole-archive vmlinux.a --no-whole-archive \
> - --start-group $(KBUILD_VMLINUX_LIBS) --end-group \
> + --whole-archive $(vmlinux_a) --no-whole-archive \
> + --start-group $(vmlinux_libs) --end-group \
> $(cmd_objtool)
>
> define rule_ld_vmlinux.o
> @@ -68,7 +76,7 @@ define rule_ld_vmlinux.o
> $(call cmd,gen_objtooldep)
> endef
>
> -vmlinux.o: $(initcalls-lds) vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> +vmlinux.o: $(initcalls-lds) $(vmlinux_a) $(vmlinux_libs) FORCE
> $(call if_changed_rule,ld_vmlinux.o)
>
> targets += vmlinux.o
> diff --git a/scripts/Makefile.vmlinux_thinlink b/scripts/Makefile.vmlinux_thinlink
> new file mode 100644
> index 0000000000000..13e4026c7d45b
> --- /dev/null
> +++ b/scripts/Makefile.vmlinux_thinlink
> @@ -0,0 +1,53 @@
> +# SPDX-License-Identifier: GPL-2.0-only
> +
> +PHONY := __default
> +__default: vmlinux.thinlink
> +
> +include include/config/auto.conf
> +include $(srctree)/scripts/Kbuild.include
> +
> +
> +# Generate a linker script to ensure correct ordering of initcalls for Clang LTO
> +# ---------------------------------------------------------------------------
> +
> +quiet_cmd_gen_initcalls_lds = GEN $@
> + cmd_gen_initcalls_lds = \
> + $(PYTHON3) $(srctree)/scripts/jobserver-exec \
> + $(PERL) $(real-prereqs) > $@
> +
> +.tmp_initcalls_thinlink.lds: $(srctree)/scripts/generate_initcall_order.pl \
> + vmlinux.a FORCE
> + $(call if_changed,gen_initcalls_lds)
> +
> +targets := .tmp_initcalls_thinlink.lds
> +
> +initcalls-lds := .tmp_initcalls_thinlink.lds
> +
> +quiet_cmd_ld_vmlinux.thinlink = LD $@
> + cmd_ld_vmlinux.thinlink = \
> + $(AR) t vmlinux.a > .vmlinux_thinlto_bc_files; \
Question: Is this a workaround for a linker bug?
I wonder why we cannot directly pass vmlinux.a just like
$(LD) ... --thinlto-index-only vmlinux.a
> + $(LD) ${KBUILD_LDFLAGS} -r $(addprefix -T , $(initcalls-lds)) \
> + --thinlto-index-only @.vmlinux_thinlto_bc_files; \
> + touch vmlinux.thinlink
Instead of 'touch', I think it is better to use
--thinlto-index-only=$@
> +
> +vmlinux.thinlink: vmlinux.a $(initcalls-lds) FORCE
> + $(call if_changed,ld_vmlinux.thinlink)
> +
> +targets += vmlinux.thinlink
> +
> +# Add FORCE to the prerequisites of a target to force it to be always rebuilt.
> +# ---------------------------------------------------------------------------
> +
> +PHONY += FORCE
> +FORCE:
> +
> +# Read all saved command lines and dependencies for the $(targets) we
> +# may be building above, using $(if_changed{,_dep}). As an
> +# optimization, we don't need to read them if the target does not
> +# exist, we will rebuild anyway in that case.
> +
> +existing-targets := $(wildcard $(sort $(targets)))
> +
> +-include $(foreach f,$(existing-targets),$(dir $(f)).$(notdir $(f)).cmd)
> +
> +.PHONY: $(PHONY)
> diff --git a/scripts/head-object-list.txt b/scripts/head-object-list.txt
> index 7274dfc65af60..90710b87a3877 100644
> --- a/scripts/head-object-list.txt
> +++ b/scripts/head-object-list.txt
> @@ -18,6 +18,7 @@ arch/arm/kernel/head.o
> arch/csky/kernel/head.o
> arch/hexagon/kernel/head.o
> arch/loongarch/kernel/head.o
> +arch/loongarch/kernel/head.o_thinlto_native
> arch/m68k/68000/head.o
> arch/m68k/coldfire/head.o
> arch/m68k/kernel/head.o
> --
> 2.49.0.1143.g0be31eac6b-goog
>
>
--
Best Regards
Masahiro Yamada
On Sun, May 25, 2025 at 1:59 AM Masahiro Yamada <masahiroy@kernel.org> wrote:
>
> On Thu, May 22, 2025 at 6:35 AM <xur@google.com> wrote:
> >
> > From: Rong Xu <xur@google.com>
> >
> > Add distributed ThinLTO build support for the Linux kernel.
> > This new mode offers several advantages: (1) Increased
> > flexibility in handling user-specified build options.
> > (2) Improved user-friendliness for developers. (3) Greater
> > convenience for integrating with objtool and livepatch.
>
> I did not set up for a distributed build environment, but
> does CC='distcc clang' work to process thin LTO
> on back-end build machines?
I'm sure I get the question. I don't think "distcc clang" works with
in-process ThinLTO build. But I think it works with distributed
ThinLTO (I did not
try myself).
>
>
> You mentioned the benefits (3), but you did not touch
> anything about objtool.
I fixed some bugs in objtool for ThinLTO in some early 5.xx kernel.
>
> How will this patch change the objtool integration?
I haven't looked into the details. But will the availability of native objects
help here?
>
>
>
> >
> > Note that "distributed" in this context refers to a term
> > that differentiates in-process ThinLTO builds by invoking
> > backend compilation through the linker, not necessarily
> > building in distributed environments.
> >
> > Distributed ThinLTO is enabled via the
> > `CONFIG_LTO_CLANG_THIN_DIST` Kconfig option. For example:
> > > make LLVM=1 defconfig
> > > scripts/config -e LTO_CLANG_THIN_DIST
> > > make LLVM=1 oldconfig
> > > make LLVM=1 vmlinux -j <..>
> >
> > The implementation changes the top-level Makefile with a
> > macro for generating `vmlinux.o` for distributed ThinLTO
> > builds. It uses the existing Kbuild infrastructure to
> > perform two recursive passes through the subdirectories.
> > The first pass generates LLVM IR object files, similar to
> > in-process ThinLTO. Following the thin-link stage, a second
> > pass compiles these IR files into the final native object
> > files. The build rules and actions for this two-pass process
> > are primarily implemented in `scripts/Makefile.build`.
> >
> > Currently, this patch focuses on building the main kernel
> > image (`vmlinux`) only. Support for building kernel modules
> > using this method is planned for a subsequent patch.
> >
> > Tested on the following arch: x86, arm64, loongarch, and
> > riscv.
> >
> > Some implementation details can be found here:
> > https://discourse.llvm.org/t/rfc-distributed-thinlto-build-for-kernel/85934
> >
> > Signed-off-by: Rong Xu <xur@google.com>
> > ---
> > Changelog since v1:
> > - Updated the description in arch/Kconfig based on feedback
> > from Nathan Chancellor
> > - Revised file suffixes: .final_o -> .o.thinlto.native, and
> > .final_a -> .a.thinlto.native
> > - Updated list of ignored files in .gitignore
> >
> > Changelog since v2:
> > - Changed file suffixes: .o.thinlto.native -> .o_thinlto_native,
> > and .a.thinlto.native -> .a_thinlto_native so that basename
> > works as intended.
> > - Tested the patch with AutoFDO and Propeller.
> > ---
>
> > diff --git a/.gitignore b/.gitignore
> > index f2f63e47fb886..b83a68185ef46 100644
> > --- a/.gitignore
> > +++ b/.gitignore
> > @@ -12,6 +12,7 @@
> > #
> > .*
> > *.a
> > +*.a_thinlto_native
>
> This is unneeded.
ack.
>
> > *.asn1.[ch]
> > *.bin
> > *.bz2
> > @@ -39,6 +40,7 @@
> > *.mod.c
> > *.o
> > *.o.*
> > +*.o_thinlto_native
>
> I would rename this to *.thinlto-native.o
> so we do not need to touch .gitignore at all.
ack.
>
>
>
>
>
> > *.patch
> > *.rmeta
> > *.rpm
> > @@ -64,6 +66,7 @@ modules.order
> > /vmlinux
> > /vmlinux.32
> > /vmlinux.map
> > +/vmlinux.thinlink
> > /vmlinux.symvers
> > /vmlinux.unstripped
> > /vmlinux-gdb.py
> > diff --git a/MAINTAINERS b/MAINTAINERS
> > index d48dd6726fe6b..f54090f364c93 100644
> > --- a/MAINTAINERS
> > +++ b/MAINTAINERS
> > @@ -5790,6 +5790,11 @@ F: scripts/Makefile.clang
> > F: scripts/clang-tools/
> > K: \b(?i:clang|llvm)\b
> >
> > +CLANG/LLVM THINLTO DISTRIBUTED BUILD
> > +M: Rong Xu <xur@google.com>
> > +S: Supported
> > +F: scripts/Makefile.vmlinux_thinlink
> > +
> > CLK API
> > M: Russell King <linux@armlinux.org.uk>
> > L: linux-clk@vger.kernel.org
> > diff --git a/Makefile b/Makefile
> > index a9edd03036537..8fbff2ab87ebd 100644
> > --- a/Makefile
> > +++ b/Makefile
> > @@ -298,7 +298,8 @@ no-dot-config-targets := $(clean-targets) \
> > outputmakefile rustavailable rustfmt rustfmtcheck
> > no-sync-config-targets := $(no-dot-config-targets) %install modules_sign kernelrelease \
> > image_name
> > -single-targets := %.a %.i %.ko %.lds %.ll %.lst %.mod %.o %.rsi %.s %/
> > +single-targets := %.a %.i %.ko %.lds %.ll %.lst %.mod %.o %.rsi %.s %.o_thinlto_native \
> > + %.a_thinlto_native %.o.thinlto.bc %/
>
>
> You are adding all new suffixes, but in my understanding there is no way
> to generate %.o.thinlto.bc as a single target.
You are correct. they are generated in batch.
>
>
>
>
>
>
>
>
> >
> > config-build :=
> > mixed-build :=
> > @@ -991,10 +992,10 @@ export CC_FLAGS_SCS
> > endif
> >
> > ifdef CONFIG_LTO_CLANG
> > -ifdef CONFIG_LTO_CLANG_THIN
> > -CC_FLAGS_LTO := -flto=thin -fsplit-lto-unit
> > -else
> > +ifdef CONFIG_LTO_CLANG_FULL
> > CC_FLAGS_LTO := -flto
> > +else # for CONFIG_LTO_CLANG_THIN or CONFIG_LTO_CLANG_THIN_DIST
> > +CC_FLAGS_LTO := -flto=thin -fsplit-lto-unit
> > endif
> > CC_FLAGS_LTO += -fvisibility=hidden
> >
> > @@ -1213,8 +1214,34 @@ vmlinux.a: $(KBUILD_VMLINUX_OBJS) scripts/head-object-list.txt FORCE
> > $(call if_changed,ar_vmlinux.a)
> >
> > PHONY += vmlinux_o
> > +ifdef CONFIG_LTO_CLANG_THIN_DIST
> > +vmlinux.thinlink: vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> > + $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.vmlinux_thinlink
> > +targets += vmlinux.thinlink
> > +
> > +vmlinux.a_thinlto_native := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_OBJS))
> > +quiet_cmd_ar_vmlinux.a_thinlto_native = AR $@
> > + cmd_ar_vmlinux.a_thinlto_native = \
> > + rm -f $@; \
> > + $(AR) cDPrST $@ $(vmlinux.a_thinlto_native); \
> > + $(AR) mPiT $$($(AR) t $@ | sed -n 1p) $@ $$($(AR) t $@ | grep -F -f $(srctree)/scripts/head-object-list.txt)
> > +
> > +define rule_gen_vmlinux.a_thinlto_native
> > + +$(Q)$(MAKE) $(build)=. need-builtin=1 thinlto_final_pass=1 need-modorder=1 built-in.a_thinlto_native
> > + $(call cmd_and_savecmd,ar_vmlinux.a_thinlto_native)
> > +endef
> > +
> > +vmlinux.a_thinlto_native: vmlinux.thinlink scripts/head-object-list.txt FORCE
> > + $(call if_changed_rule,gen_vmlinux.a_thinlto_native)
> > +
> > +targets += vmlinux.a_thinlto_native
> > +
> > +vmlinux_o: vmlinux.a_thinlto_native
> > + $(Q)$(MAKE) thinlto_final_pass=1 -f $(srctree)/scripts/Makefile.vmlinux_o
> > +else
> > vmlinux_o: vmlinux.a $(KBUILD_VMLINUX_LIBS)
> > $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.vmlinux_o
> > +endif
> >
> > vmlinux.o modules.builtin.modinfo modules.builtin: vmlinux_o
> > @:
> > @@ -1572,7 +1599,8 @@ CLEAN_FILES += vmlinux.symvers modules-only.symvers \
> > modules.builtin.ranges vmlinux.o.map vmlinux.unstripped \
> > compile_commands.json rust/test \
> > rust-project.json .vmlinux.objs .vmlinux.export.c \
> > - .builtin-dtbs-list .builtin-dtb.S
> > + .builtin-dtbs-list .builtin-dtb.S \
> > + .vmlinux_thinlto_bc_files vmlinux.thinlink
> >
> > # Directories & files removed with 'make mrproper'
> > MRPROPER_FILES += include/config include/generated \
> > @@ -2023,6 +2051,8 @@ clean: $(clean-dirs)
> > -o -name '*.symtypes' -o -name 'modules.order' \
> > -o -name '*.c.[012]*.*' \
> > -o -name '*.ll' \
> > + -o -name '*.a_thinlto_native' -o -name '*.o_thinlto_native' \
> > + -o -name '*.o.thinlto.bc' \
> > -o -name '*.gcno' \
> > \) -type f -print \
> > -o -name '.tmp_*' -print \
> > diff --git a/arch/Kconfig b/arch/Kconfig
> > index b0adb665041f1..30dccda07c671 100644
> > --- a/arch/Kconfig
> > +++ b/arch/Kconfig
> > @@ -810,6 +810,25 @@ config LTO_CLANG_THIN
> > https://clang.llvm.org/docs/ThinLTO.html
> >
> > If unsure, say Y.
> > +
> > +config LTO_CLANG_THIN_DIST
> > + bool "Clang ThinLTO in distributed mode (EXPERIMENTAL)"
> > + depends on HAS_LTO_CLANG && ARCH_SUPPORTS_LTO_CLANG_THIN
> > + select LTO_CLANG
> > + help
> > + This option enables Clang's ThinLTO in distributed build mode.
> > + In this mode, the linker performs the thin-link, generating
> > + ThinLTO index files. Subsequently, the build system explicitly
> > + invokes ThinLTO backend compilation using these index files
> > + and pre-linked IR objects. The resulting native object files
> > + are with the .o_thinlto_native suffix.
> > +
> > + This build mode offers improved visibility into the ThinLTO
> > + process through explicit subcommand exposure. It also makes
> > + final native object files directly available, benefiting
> > + tools like objtool and kpatch. Additionally, it provides
> > + crucial granular control over back-end options, enabling
> > + module-specific compiler options, and simplifies debugging.
> > endchoice
> >
> > config ARCH_SUPPORTS_AUTOFDO_CLANG
> > diff --git a/scripts/Makefile.build b/scripts/Makefile.build
> > index 13dcd86e74ca8..338e1aec0eaa3 100644
> > --- a/scripts/Makefile.build
> > +++ b/scripts/Makefile.build
> > @@ -50,18 +50,23 @@ endif
> >
> > # ===========================================================================
> >
> > +builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a)
> > +ifeq ($(thinlto_final_pass),1)
> > +builtin_suffix :=.a_thinlto_native
> > +endif
> > +
> > # subdir-builtin and subdir-modorder may contain duplications. Use $(sort ...)
> > -subdir-builtin := $(sort $(filter %/built-in.a, $(real-obj-y)))
> > +subdir-builtin := $(sort $(filter %/built-in$(builtin_suffix), $(real-obj-y)))
> > subdir-modorder := $(sort $(filter %/modules.order, $(obj-m)))
> >
> > targets-for-builtin := $(extra-y)
> >
> > ifneq ($(strip $(lib-y) $(lib-m) $(lib-)),)
> > -targets-for-builtin += $(obj)/lib.a
> > +targets-for-builtin += $(obj)/lib$(builtin_suffix)
> > endif
> >
> > ifdef need-builtin
> > -targets-for-builtin += $(obj)/built-in.a
> > +targets-for-builtin += $(obj)/built-in$(builtin_suffix)
> > endif
> >
> > targets-for-modules := $(foreach x, o mod, \
> > @@ -337,6 +342,10 @@ $(obj)/%.o: $(obj)/%.S FORCE
> > targets += $(filter-out $(subdir-builtin), $(real-obj-y))
> > targets += $(filter-out $(subdir-modorder), $(real-obj-m))
> > targets += $(lib-y) $(always-y)
> > +ifeq ($(builtin_suffix),.a_thinlto_native)
> > +native_targets = $(patsubst,%.o,%.o_thinlto_native,$(targets))
> > +targets += $(native_targets)
> > +endif
> >
> > # Linker scripts preprocessor (.lds.S -> .lds)
> > # ---------------------------------------------------------------------------
> > @@ -347,6 +356,24 @@ quiet_cmd_cpp_lds_S = LDS $@
> > $(obj)/%.lds: $(src)/%.lds.S FORCE
> > $(call if_changed_dep,cpp_lds_S)
> >
> > +ifdef CONFIG_LTO_CLANG_THIN_DIST
> > +# Generate .o_thinlto_native (obj) from .o (bitcode) file
> > +# ---------------------------------------------------------------------------
> > +quiet_cmd_cc_o_bc = CC $(quiet_modtag) $@
> > +
> > +cmd_cc_o_bc = $(if $(filter bitcode, $(shell file -b $<)),$(CC) \
> > + $(filter-out -Wp% $(LINUXINCLUDE) %.h.gch %.h -D% \
> > + -flto=thin, $(c_flags)) \
> > + -Wno-unused-command-line-argument \
> > + -x ir -fthinlto-index=$<.thinlto.bc -c -o $@ \
> > + $(if $(findstring ../,$<), \
> > + $$(realpath --relative-to=$(srcroot) $<), $<), \
> > + cp $< $@)
> > +
> > +$(obj)/%.o_thinlto_native: $(obj)/%.o FORCE
> > + $(call if_changed,cc_o_bc)
> > +endif
> > +
> > # ASN.1 grammar
> > # ---------------------------------------------------------------------------
> > quiet_cmd_asn1_compiler = ASN.1 $(basename $@).[ch]
> > @@ -360,7 +387,7 @@ $(obj)/%.asn1.c $(obj)/%.asn1.h: $(src)/%.asn1 $(objtree)/scripts/asn1_compiler
> > # ---------------------------------------------------------------------------
> >
> > # To build objects in subdirs, we need to descend into the directories
> > -$(subdir-builtin): $(obj)/%/built-in.a: $(obj)/% ;
> > +$(subdir-builtin): $(obj)/%/built-in$(builtin_suffix): $(obj)/% ;
> > $(subdir-modorder): $(obj)/%/modules.order: $(obj)/% ;
> >
> > #
> > @@ -377,6 +404,12 @@ quiet_cmd_ar_builtin = AR $@
> > $(obj)/built-in.a: $(real-obj-y) FORCE
> > $(call if_changed,ar_builtin)
> >
> > +ifdef CONFIG_LTO_CLANG_THIN_DIST
> > +# Rule to compile a set of .o_thinlto_native files into one .a_thinlto_native file.
> > +$(obj)/built-in.a_thinlto_native: $(patsubst %.o,%.o_thinlto_native,$(real-obj-y)) FORCE
> > + $(call if_changed,ar_builtin)
> > +endif
> > +
> > # This is a list of build artifacts from the current Makefile and its
> > # sub-directories. The timestamp should be updated when any of the member files.
> >
> > @@ -394,6 +427,14 @@ $(obj)/modules.order: $(obj-m) FORCE
> > $(obj)/lib.a: $(lib-y) FORCE
> > $(call if_changed,ar)
> >
> > +ifdef CONFIG_LTO_CLANG_THIN_DIST
> > +quiet_cmd_ar_native = AR $@
> > + cmd_ar_native = rm -f $@; $(AR) cDPrsT $@ $(patsubst %.o,%.o_thinlto_native,$(real-prereqs))
> > +
> > +$(obj)/lib.a_thinlto_native: $(patsubst %.o,%.o_thinlto_native,$(lib-y)) FORCE
> > + $(call if_changed,ar_native)
> > +endif
> > +
> > quiet_cmd_ld_multi_m = LD [M] $@
> > cmd_ld_multi_m = $(LD) $(ld_flags) -r -o $@ @$< $(cmd_objtool)
> >
> > @@ -459,7 +500,8 @@ $(single-subdir-goals): $(single-subdirs)
> > PHONY += $(subdir-ym)
> > $(subdir-ym):
> > $(Q)$(MAKE) $(build)=$@ \
> > - need-builtin=$(if $(filter $@/built-in.a, $(subdir-builtin)),1) \
> > + need-builtin=$(if $(filter $@/built-in$(builtin_suffix), $(subdir-builtin)),1) \
> > + thinlto_final_pass=$(if $(filter .a_thinlto_native, $(builtin_suffix)),1) \
> > need-modorder=$(if $(filter $@/modules.order, $(subdir-modorder)),1) \
> > $(filter $@/%, $(single-subdir-goals))
> >
> > diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
> > index 2fe73cda0bddb..9cfd23590334d 100644
> > --- a/scripts/Makefile.lib
> > +++ b/scripts/Makefile.lib
> > @@ -34,8 +34,13 @@ else
> > obj-m := $(filter-out %/, $(obj-m))
> > endif
> >
> > +builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a)
> > +ifeq ($(thinlto_final_pass),1)
> > + builtin_suffix :=.a_thinlto_native
> > +endif
> > +
> > ifdef need-builtin
> > -obj-y := $(patsubst %/, %/built-in.a, $(obj-y))
> > +obj-y := $(patsubst %/, %/built-in$(builtin_suffix), $(obj-y))
> > else
> > obj-y := $(filter-out %/, $(obj-y))
> > endif
> > diff --git a/scripts/Makefile.vmlinux_o b/scripts/Makefile.vmlinux_o
> > index b024ffb3e2018..f9abc45a68b36 100644
> > --- a/scripts/Makefile.vmlinux_o
> > +++ b/scripts/Makefile.vmlinux_o
> > @@ -9,6 +9,14 @@ include $(srctree)/scripts/Kbuild.include
> > # for objtool
> > include $(srctree)/scripts/Makefile.lib
> >
> > +ifeq ($(thinlto_final_pass),1)
> > +vmlinux_a := vmlinux.a_thinlto_native
> > +vmlinux_libs := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_LIBS))
> > +else
> > +vmlinux_a := vmlinux.a
> > +vmlinux_libs := $(KBUILD_VMLINUX_LIBS)
> > +endif
> > +
> > # Generate a linker script to ensure correct ordering of initcalls for Clang LTO
> > # ---------------------------------------------------------------------------
> >
> > @@ -18,7 +26,7 @@ quiet_cmd_gen_initcalls_lds = GEN $@
> > $(PERL) $(real-prereqs) > $@
> >
> > .tmp_initcalls.lds: $(srctree)/scripts/generate_initcall_order.pl \
> > - vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> > + $(vmlinux_a) $(vmlinux_libs) FORCE
> > $(call if_changed,gen_initcalls_lds)
> >
> > targets := .tmp_initcalls.lds
> > @@ -59,8 +67,8 @@ quiet_cmd_ld_vmlinux.o = LD $@
> > $(LD) ${KBUILD_LDFLAGS} -r -o $@ \
> > $(vmlinux-o-ld-args-y) \
> > $(addprefix -T , $(initcalls-lds)) \
> > - --whole-archive vmlinux.a --no-whole-archive \
> > - --start-group $(KBUILD_VMLINUX_LIBS) --end-group \
> > + --whole-archive $(vmlinux_a) --no-whole-archive \
> > + --start-group $(vmlinux_libs) --end-group \
> > $(cmd_objtool)
> >
> > define rule_ld_vmlinux.o
> > @@ -68,7 +76,7 @@ define rule_ld_vmlinux.o
> > $(call cmd,gen_objtooldep)
> > endef
> >
> > -vmlinux.o: $(initcalls-lds) vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> > +vmlinux.o: $(initcalls-lds) $(vmlinux_a) $(vmlinux_libs) FORCE
> > $(call if_changed_rule,ld_vmlinux.o)
> >
> > targets += vmlinux.o
> > diff --git a/scripts/Makefile.vmlinux_thinlink b/scripts/Makefile.vmlinux_thinlink
> > new file mode 100644
> > index 0000000000000..13e4026c7d45b
> > --- /dev/null
> > +++ b/scripts/Makefile.vmlinux_thinlink
> > @@ -0,0 +1,53 @@
> > +# SPDX-License-Identifier: GPL-2.0-only
> > +
> > +PHONY := __default
> > +__default: vmlinux.thinlink
> > +
> > +include include/config/auto.conf
> > +include $(srctree)/scripts/Kbuild.include
> > +
> > +
> > +# Generate a linker script to ensure correct ordering of initcalls for Clang LTO
> > +# ---------------------------------------------------------------------------
> > +
> > +quiet_cmd_gen_initcalls_lds = GEN $@
> > + cmd_gen_initcalls_lds = \
> > + $(PYTHON3) $(srctree)/scripts/jobserver-exec \
> > + $(PERL) $(real-prereqs) > $@
> > +
> > +.tmp_initcalls_thinlink.lds: $(srctree)/scripts/generate_initcall_order.pl \
> > + vmlinux.a FORCE
> > + $(call if_changed,gen_initcalls_lds)
> > +
> > +targets := .tmp_initcalls_thinlink.lds
> > +
> > +initcalls-lds := .tmp_initcalls_thinlink.lds
> > +
> > +quiet_cmd_ld_vmlinux.thinlink = LD $@
> > + cmd_ld_vmlinux.thinlink = \
> > + $(AR) t vmlinux.a > .vmlinux_thinlto_bc_files; \
>
>
> Question: Is this a workaround for a linker bug?
You can call it a bug as clang does not handle this.
>
> I wonder why we cannot directly pass vmlinux.a just like
>
> $(LD) ... --thinlto-index-only vmlinux.a
I also think tt should be this way. But currently it does not work.
>
>
>
>
> > + $(LD) ${KBUILD_LDFLAGS} -r $(addprefix -T , $(initcalls-lds)) \
> > + --thinlto-index-only @.vmlinux_thinlto_bc_files; \
> > + touch vmlinux.thinlink
>
>
> Instead of 'touch', I think it is better to use
>
> --thinlto-index-only=$@
ack.
>
>
>
>
>
>
>
>
>
>
>
>
>
>
> > +
> > +vmlinux.thinlink: vmlinux.a $(initcalls-lds) FORCE
> > + $(call if_changed,ld_vmlinux.thinlink)
> > +
> > +targets += vmlinux.thinlink
> > +
> > +# Add FORCE to the prerequisites of a target to force it to be always rebuilt.
> > +# ---------------------------------------------------------------------------
> > +
> > +PHONY += FORCE
> > +FORCE:
> > +
> > +# Read all saved command lines and dependencies for the $(targets) we
> > +# may be building above, using $(if_changed{,_dep}). As an
> > +# optimization, we don't need to read them if the target does not
> > +# exist, we will rebuild anyway in that case.
> > +
> > +existing-targets := $(wildcard $(sort $(targets)))
> > +
> > +-include $(foreach f,$(existing-targets),$(dir $(f)).$(notdir $(f)).cmd)
> > +
> > +.PHONY: $(PHONY)
> > diff --git a/scripts/head-object-list.txt b/scripts/head-object-list.txt
> > index 7274dfc65af60..90710b87a3877 100644
> > --- a/scripts/head-object-list.txt
> > +++ b/scripts/head-object-list.txt
> > @@ -18,6 +18,7 @@ arch/arm/kernel/head.o
> > arch/csky/kernel/head.o
> > arch/hexagon/kernel/head.o
> > arch/loongarch/kernel/head.o
> > +arch/loongarch/kernel/head.o_thinlto_native
> > arch/m68k/68000/head.o
> > arch/m68k/coldfire/head.o
> > arch/m68k/kernel/head.o
> > --
> > 2.49.0.1143.g0be31eac6b-goog
> >
> >
>
>
> --
> Best Regards
> Masahiro Yamada
On Thu, May 22, 2025 at 6:35 AM <xur@google.com> wrote:
>
> From: Rong Xu <xur@google.com>
>
> Add distributed ThinLTO build support for the Linux kernel.
> This new mode offers several advantages: (1) Increased
> flexibility in handling user-specified build options.
> (2) Improved user-friendliness for developers. (3) Greater
> convenience for integrating with objtool and livepatch.
I did not understand any of these benefits
despite a lot of added complexity.
> Note that "distributed" in this context refers to a term
> that differentiates in-process ThinLTO builds by invoking
> backend compilation through the linker, not necessarily
> building in distributed environments.
>
> Distributed ThinLTO is enabled via the
> `CONFIG_LTO_CLANG_THIN_DIST` Kconfig option. For example:
> > make LLVM=1 defconfig
> > scripts/config -e LTO_CLANG_THIN_DIST
> > make LLVM=1 oldconfig
> > make LLVM=1 vmlinux -j <..>
>
> The implementation changes the top-level Makefile with a
> macro for generating `vmlinux.o` for distributed ThinLTO
> builds. It uses the existing Kbuild infrastructure to
> perform two recursive passes through the subdirectories.
> The first pass generates LLVM IR object files, similar to
> in-process ThinLTO. Following the thin-link stage, a second
> pass compiles these IR files into the final native object
> files. The build rules and actions for this two-pass process
> are primarily implemented in `scripts/Makefile.build`.
>
> Currently, this patch focuses on building the main kernel
> image (`vmlinux`) only. Support for building kernel modules
> using this method is planned for a subsequent patch.
The announcement of the future plan is even scary,
especially after looking at your implementation.
>
> Tested on the following arch: x86, arm64, loongarch, and
> riscv.
>
> Some implementation details can be found here:
> https://discourse.llvm.org/t/rfc-distributed-thinlto-build-for-kernel/85934
>
> Signed-off-by: Rong Xu <xur@google.com>
> ---
> Changelog since v1:
> - Updated the description in arch/Kconfig based on feedback
> from Nathan Chancellor
> - Revised file suffixes: .final_o -> .o.thinlto.native, and
> .final_a -> .a.thinlto.native
> - Updated list of ignored files in .gitignore
>
> Changelog since v2:
> - Changed file suffixes: .o.thinlto.native -> .o_thinlto_native,
> and .a.thinlto.native -> .a_thinlto_native so that basename
> works as intended.
> - Tested the patch with AutoFDO and Propeller.
> ---
> .gitignore | 3 ++
> MAINTAINERS | 5 +++
> Makefile | 40 ++++++++++++++++++++---
> arch/Kconfig | 19 +++++++++++
> scripts/Makefile.build | 52 +++++++++++++++++++++++++++---
> scripts/Makefile.lib | 7 +++-
> scripts/Makefile.vmlinux_o | 16 +++++++---
> scripts/Makefile.vmlinux_thinlink | 53 +++++++++++++++++++++++++++++++
> scripts/head-object-list.txt | 1 +
> 9 files changed, 181 insertions(+), 15 deletions(-)
> create mode 100644 scripts/Makefile.vmlinux_thinlink
>
> diff --git a/.gitignore b/.gitignore
> index f2f63e47fb886..b83a68185ef46 100644
> --- a/.gitignore
> +++ b/.gitignore
> @@ -12,6 +12,7 @@
> #
> .*
> *.a
> +*.a_thinlto_native
> *.asn1.[ch]
> *.bin
> *.bz2
> @@ -39,6 +40,7 @@
> *.mod.c
> *.o
> *.o.*
> +*.o_thinlto_native
> *.patch
> *.rmeta
> *.rpm
> @@ -64,6 +66,7 @@ modules.order
> /vmlinux
> /vmlinux.32
> /vmlinux.map
> +/vmlinux.thinlink
> /vmlinux.symvers
> /vmlinux.unstripped
> /vmlinux-gdb.py
> diff --git a/MAINTAINERS b/MAINTAINERS
> index d48dd6726fe6b..f54090f364c93 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -5790,6 +5790,11 @@ F: scripts/Makefile.clang
> F: scripts/clang-tools/
> K: \b(?i:clang|llvm)\b
>
> +CLANG/LLVM THINLTO DISTRIBUTED BUILD
> +M: Rong Xu <xur@google.com>
> +S: Supported
> +F: scripts/Makefile.vmlinux_thinlink
> +
> CLK API
> M: Russell King <linux@armlinux.org.uk>
> L: linux-clk@vger.kernel.org
> diff --git a/Makefile b/Makefile
> index a9edd03036537..8fbff2ab87ebd 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -298,7 +298,8 @@ no-dot-config-targets := $(clean-targets) \
> outputmakefile rustavailable rustfmt rustfmtcheck
> no-sync-config-targets := $(no-dot-config-targets) %install modules_sign kernelrelease \
> image_name
> -single-targets := %.a %.i %.ko %.lds %.ll %.lst %.mod %.o %.rsi %.s %/
> +single-targets := %.a %.i %.ko %.lds %.ll %.lst %.mod %.o %.rsi %.s %.o_thinlto_native \
> + %.a_thinlto_native %.o.thinlto.bc %/
>
> config-build :=
> mixed-build :=
> @@ -991,10 +992,10 @@ export CC_FLAGS_SCS
> endif
>
> ifdef CONFIG_LTO_CLANG
> -ifdef CONFIG_LTO_CLANG_THIN
> -CC_FLAGS_LTO := -flto=thin -fsplit-lto-unit
> -else
> +ifdef CONFIG_LTO_CLANG_FULL
> CC_FLAGS_LTO := -flto
> +else # for CONFIG_LTO_CLANG_THIN or CONFIG_LTO_CLANG_THIN_DIST
> +CC_FLAGS_LTO := -flto=thin -fsplit-lto-unit
> endif
> CC_FLAGS_LTO += -fvisibility=hidden
>
> @@ -1213,8 +1214,34 @@ vmlinux.a: $(KBUILD_VMLINUX_OBJS) scripts/head-object-list.txt FORCE
> $(call if_changed,ar_vmlinux.a)
>
> PHONY += vmlinux_o
> +ifdef CONFIG_LTO_CLANG_THIN_DIST
> +vmlinux.thinlink: vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> + $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.vmlinux_thinlink
> +targets += vmlinux.thinlink
> +
> +vmlinux.a_thinlto_native := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_OBJS))
> +quiet_cmd_ar_vmlinux.a_thinlto_native = AR $@
> + cmd_ar_vmlinux.a_thinlto_native = \
> + rm -f $@; \
> + $(AR) cDPrST $@ $(vmlinux.a_thinlto_native); \
> + $(AR) mPiT $$($(AR) t $@ | sed -n 1p) $@ $$($(AR) t $@ | grep -F -f $(srctree)/scripts/head-object-list.txt)
> +
> +define rule_gen_vmlinux.a_thinlto_native
> + +$(Q)$(MAKE) $(build)=. need-builtin=1 thinlto_final_pass=1 need-modorder=1 built-in.a_thinlto_native
I really do not understand why this needs the second
recursion, as you already know the list of objects
by 'ar t vmlinux.a'.
> + $(call cmd_and_savecmd,ar_vmlinux.a_thinlto_native)
> +endef
> +
> +vmlinux.a_thinlto_native: vmlinux.thinlink scripts/head-object-list.txt FORCE
> + $(call if_changed_rule,gen_vmlinux.a_thinlto_native)
> +
> +targets += vmlinux.a_thinlto_native
> +
> +vmlinux_o: vmlinux.a_thinlto_native
> + $(Q)$(MAKE) thinlto_final_pass=1 -f $(srctree)/scripts/Makefile.vmlinux_o
> +else
> vmlinux_o: vmlinux.a $(KBUILD_VMLINUX_LIBS)
> $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.vmlinux_o
> +endif
>
> vmlinux.o modules.builtin.modinfo modules.builtin: vmlinux_o
> @:
> @@ -1572,7 +1599,8 @@ CLEAN_FILES += vmlinux.symvers modules-only.symvers \
> modules.builtin.ranges vmlinux.o.map vmlinux.unstripped \
> compile_commands.json rust/test \
> rust-project.json .vmlinux.objs .vmlinux.export.c \
> - .builtin-dtbs-list .builtin-dtb.S
> + .builtin-dtbs-list .builtin-dtb.S \
> + .vmlinux_thinlto_bc_files vmlinux.thinlink
>
> # Directories & files removed with 'make mrproper'
> MRPROPER_FILES += include/config include/generated \
> @@ -2023,6 +2051,8 @@ clean: $(clean-dirs)
> -o -name '*.symtypes' -o -name 'modules.order' \
> -o -name '*.c.[012]*.*' \
> -o -name '*.ll' \
> + -o -name '*.a_thinlto_native' -o -name '*.o_thinlto_native' \
> + -o -name '*.o.thinlto.bc' \
> -o -name '*.gcno' \
> \) -type f -print \
> -o -name '.tmp_*' -print \
> diff --git a/arch/Kconfig b/arch/Kconfig
> index b0adb665041f1..30dccda07c671 100644
> --- a/arch/Kconfig
> +++ b/arch/Kconfig
> @@ -810,6 +810,25 @@ config LTO_CLANG_THIN
> https://clang.llvm.org/docs/ThinLTO.html
>
> If unsure, say Y.
> +
> +config LTO_CLANG_THIN_DIST
> + bool "Clang ThinLTO in distributed mode (EXPERIMENTAL)"
> + depends on HAS_LTO_CLANG && ARCH_SUPPORTS_LTO_CLANG_THIN
> + select LTO_CLANG
> + help
> + This option enables Clang's ThinLTO in distributed build mode.
> + In this mode, the linker performs the thin-link, generating
> + ThinLTO index files. Subsequently, the build system explicitly
> + invokes ThinLTO backend compilation using these index files
> + and pre-linked IR objects. The resulting native object files
> + are with the .o_thinlto_native suffix.
> +
> + This build mode offers improved visibility into the ThinLTO
> + process through explicit subcommand exposure. It also makes
> + final native object files directly available, benefiting
> + tools like objtool and kpatch. Additionally, it provides
> + crucial granular control over back-end options, enabling
> + module-specific compiler options, and simplifies debugging.
> endchoice
>
> config ARCH_SUPPORTS_AUTOFDO_CLANG
> diff --git a/scripts/Makefile.build b/scripts/Makefile.build
> index 13dcd86e74ca8..338e1aec0eaa3 100644
> --- a/scripts/Makefile.build
> +++ b/scripts/Makefile.build
> @@ -50,18 +50,23 @@ endif
>
> # ===========================================================================
>
> +builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a)
> +ifeq ($(thinlto_final_pass),1)
> +builtin_suffix :=.a_thinlto_native
> +endif
> +
> # subdir-builtin and subdir-modorder may contain duplications. Use $(sort ...)
> -subdir-builtin := $(sort $(filter %/built-in.a, $(real-obj-y)))
> +subdir-builtin := $(sort $(filter %/built-in$(builtin_suffix), $(real-obj-y)))
> subdir-modorder := $(sort $(filter %/modules.order, $(obj-m)))
>
> targets-for-builtin := $(extra-y)
>
> ifneq ($(strip $(lib-y) $(lib-m) $(lib-)),)
> -targets-for-builtin += $(obj)/lib.a
> +targets-for-builtin += $(obj)/lib$(builtin_suffix)
> endif
>
> ifdef need-builtin
> -targets-for-builtin += $(obj)/built-in.a
> +targets-for-builtin += $(obj)/built-in$(builtin_suffix)
> endif
>
> targets-for-modules := $(foreach x, o mod, \
> @@ -337,6 +342,10 @@ $(obj)/%.o: $(obj)/%.S FORCE
> targets += $(filter-out $(subdir-builtin), $(real-obj-y))
> targets += $(filter-out $(subdir-modorder), $(real-obj-m))
> targets += $(lib-y) $(always-y)
> +ifeq ($(builtin_suffix),.a_thinlto_native)
> +native_targets = $(patsubst,%.o,%.o_thinlto_native,$(targets))
> +targets += $(native_targets)
> +endif
>
> # Linker scripts preprocessor (.lds.S -> .lds)
> # ---------------------------------------------------------------------------
> @@ -347,6 +356,24 @@ quiet_cmd_cpp_lds_S = LDS $@
> $(obj)/%.lds: $(src)/%.lds.S FORCE
> $(call if_changed_dep,cpp_lds_S)
>
> +ifdef CONFIG_LTO_CLANG_THIN_DIST
> +# Generate .o_thinlto_native (obj) from .o (bitcode) file
> +# ---------------------------------------------------------------------------
> +quiet_cmd_cc_o_bc = CC $(quiet_modtag) $@
> +
> +cmd_cc_o_bc = $(if $(filter bitcode, $(shell file -b $<)),$(CC) \
> + $(filter-out -Wp% $(LINUXINCLUDE) %.h.gch %.h -D% \
> + -flto=thin, $(c_flags)) \
> + -Wno-unused-command-line-argument \
> + -x ir -fthinlto-index=$<.thinlto.bc -c -o $@ \
> + $(if $(findstring ../,$<), \
> + $$(realpath --relative-to=$(srcroot) $<), $<), \
> + cp $< $@)
> +
> +$(obj)/%.o_thinlto_native: $(obj)/%.o FORCE
> + $(call if_changed,cc_o_bc)
> +endif
> +
> # ASN.1 grammar
> # ---------------------------------------------------------------------------
> quiet_cmd_asn1_compiler = ASN.1 $(basename $@).[ch]
> @@ -360,7 +387,7 @@ $(obj)/%.asn1.c $(obj)/%.asn1.h: $(src)/%.asn1 $(objtree)/scripts/asn1_compiler
> # ---------------------------------------------------------------------------
>
> # To build objects in subdirs, we need to descend into the directories
> -$(subdir-builtin): $(obj)/%/built-in.a: $(obj)/% ;
> +$(subdir-builtin): $(obj)/%/built-in$(builtin_suffix): $(obj)/% ;
> $(subdir-modorder): $(obj)/%/modules.order: $(obj)/% ;
>
> #
> @@ -377,6 +404,12 @@ quiet_cmd_ar_builtin = AR $@
> $(obj)/built-in.a: $(real-obj-y) FORCE
> $(call if_changed,ar_builtin)
>
> +ifdef CONFIG_LTO_CLANG_THIN_DIST
> +# Rule to compile a set of .o_thinlto_native files into one .a_thinlto_native file.
> +$(obj)/built-in.a_thinlto_native: $(patsubst %.o,%.o_thinlto_native,$(real-obj-y)) FORCE
> + $(call if_changed,ar_builtin)
> +endif
> +
> # This is a list of build artifacts from the current Makefile and its
> # sub-directories. The timestamp should be updated when any of the member files.
>
> @@ -394,6 +427,14 @@ $(obj)/modules.order: $(obj-m) FORCE
> $(obj)/lib.a: $(lib-y) FORCE
> $(call if_changed,ar)
>
> +ifdef CONFIG_LTO_CLANG_THIN_DIST
> +quiet_cmd_ar_native = AR $@
> + cmd_ar_native = rm -f $@; $(AR) cDPrsT $@ $(patsubst %.o,%.o_thinlto_native,$(real-prereqs))
> +
> +$(obj)/lib.a_thinlto_native: $(patsubst %.o,%.o_thinlto_native,$(lib-y)) FORCE
> + $(call if_changed,ar_native)
> +endif
> +
> quiet_cmd_ld_multi_m = LD [M] $@
> cmd_ld_multi_m = $(LD) $(ld_flags) -r -o $@ @$< $(cmd_objtool)
>
> @@ -459,7 +500,8 @@ $(single-subdir-goals): $(single-subdirs)
> PHONY += $(subdir-ym)
> $(subdir-ym):
> $(Q)$(MAKE) $(build)=$@ \
> - need-builtin=$(if $(filter $@/built-in.a, $(subdir-builtin)),1) \
> + need-builtin=$(if $(filter $@/built-in$(builtin_suffix), $(subdir-builtin)),1) \
> + thinlto_final_pass=$(if $(filter .a_thinlto_native, $(builtin_suffix)),1) \
> need-modorder=$(if $(filter $@/modules.order, $(subdir-modorder)),1) \
> $(filter $@/%, $(single-subdir-goals))
>
> diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
> index 2fe73cda0bddb..9cfd23590334d 100644
> --- a/scripts/Makefile.lib
> +++ b/scripts/Makefile.lib
> @@ -34,8 +34,13 @@ else
> obj-m := $(filter-out %/, $(obj-m))
> endif
>
> +builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a)
> +ifeq ($(thinlto_final_pass),1)
> + builtin_suffix :=.a_thinlto_native
> +endif
> +
> ifdef need-builtin
> -obj-y := $(patsubst %/, %/built-in.a, $(obj-y))
> +obj-y := $(patsubst %/, %/built-in$(builtin_suffix), $(obj-y))
> else
> obj-y := $(filter-out %/, $(obj-y))
> endif
> diff --git a/scripts/Makefile.vmlinux_o b/scripts/Makefile.vmlinux_o
> index b024ffb3e2018..f9abc45a68b36 100644
> --- a/scripts/Makefile.vmlinux_o
> +++ b/scripts/Makefile.vmlinux_o
> @@ -9,6 +9,14 @@ include $(srctree)/scripts/Kbuild.include
> # for objtool
> include $(srctree)/scripts/Makefile.lib
>
> +ifeq ($(thinlto_final_pass),1)
> +vmlinux_a := vmlinux.a_thinlto_native
> +vmlinux_libs := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_LIBS))
> +else
> +vmlinux_a := vmlinux.a
> +vmlinux_libs := $(KBUILD_VMLINUX_LIBS)
> +endif
> +
> # Generate a linker script to ensure correct ordering of initcalls for Clang LTO
> # ---------------------------------------------------------------------------
>
> @@ -18,7 +26,7 @@ quiet_cmd_gen_initcalls_lds = GEN $@
> $(PERL) $(real-prereqs) > $@
>
> .tmp_initcalls.lds: $(srctree)/scripts/generate_initcall_order.pl \
> - vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> + $(vmlinux_a) $(vmlinux_libs) FORCE
> $(call if_changed,gen_initcalls_lds)
>
> targets := .tmp_initcalls.lds
> @@ -59,8 +67,8 @@ quiet_cmd_ld_vmlinux.o = LD $@
> $(LD) ${KBUILD_LDFLAGS} -r -o $@ \
> $(vmlinux-o-ld-args-y) \
> $(addprefix -T , $(initcalls-lds)) \
> - --whole-archive vmlinux.a --no-whole-archive \
> - --start-group $(KBUILD_VMLINUX_LIBS) --end-group \
> + --whole-archive $(vmlinux_a) --no-whole-archive \
> + --start-group $(vmlinux_libs) --end-group \
> $(cmd_objtool)
>
> define rule_ld_vmlinux.o
> @@ -68,7 +76,7 @@ define rule_ld_vmlinux.o
> $(call cmd,gen_objtooldep)
> endef
>
> -vmlinux.o: $(initcalls-lds) vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> +vmlinux.o: $(initcalls-lds) $(vmlinux_a) $(vmlinux_libs) FORCE
> $(call if_changed_rule,ld_vmlinux.o)
>
> targets += vmlinux.o
> diff --git a/scripts/Makefile.vmlinux_thinlink b/scripts/Makefile.vmlinux_thinlink
> new file mode 100644
> index 0000000000000..13e4026c7d45b
> --- /dev/null
> +++ b/scripts/Makefile.vmlinux_thinlink
> @@ -0,0 +1,53 @@
> +# SPDX-License-Identifier: GPL-2.0-only
> +
> +PHONY := __default
> +__default: vmlinux.thinlink
> +
> +include include/config/auto.conf
> +include $(srctree)/scripts/Kbuild.include
> +
> +
> +# Generate a linker script to ensure correct ordering of initcalls for Clang LTO
> +# ---------------------------------------------------------------------------
> +
> +quiet_cmd_gen_initcalls_lds = GEN $@
> + cmd_gen_initcalls_lds = \
> + $(PYTHON3) $(srctree)/scripts/jobserver-exec \
> + $(PERL) $(real-prereqs) > $@
> +
> +.tmp_initcalls_thinlink.lds: $(srctree)/scripts/generate_initcall_order.pl \
> + vmlinux.a FORCE
> + $(call if_changed,gen_initcalls_lds)
> +
> +targets := .tmp_initcalls_thinlink.lds
> +
> +initcalls-lds := .tmp_initcalls_thinlink.lds
> +
> +quiet_cmd_ld_vmlinux.thinlink = LD $@
> + cmd_ld_vmlinux.thinlink = \
> + $(AR) t vmlinux.a > .vmlinux_thinlto_bc_files; \
> + $(LD) ${KBUILD_LDFLAGS} -r $(addprefix -T , $(initcalls-lds)) \
> + --thinlto-index-only @.vmlinux_thinlto_bc_files; \
> + touch vmlinux.thinlink
> +
> +vmlinux.thinlink: vmlinux.a $(initcalls-lds) FORCE
> + $(call if_changed,ld_vmlinux.thinlink)
> +
> +targets += vmlinux.thinlink
> +
> +# Add FORCE to the prerequisites of a target to force it to be always rebuilt.
> +# ---------------------------------------------------------------------------
> +
> +PHONY += FORCE
> +FORCE:
> +
> +# Read all saved command lines and dependencies for the $(targets) we
> +# may be building above, using $(if_changed{,_dep}). As an
> +# optimization, we don't need to read them if the target does not
> +# exist, we will rebuild anyway in that case.
> +
> +existing-targets := $(wildcard $(sort $(targets)))
> +
> +-include $(foreach f,$(existing-targets),$(dir $(f)).$(notdir $(f)).cmd)
> +
> +.PHONY: $(PHONY)
> diff --git a/scripts/head-object-list.txt b/scripts/head-object-list.txt
> index 7274dfc65af60..90710b87a3877 100644
> --- a/scripts/head-object-list.txt
> +++ b/scripts/head-object-list.txt
Adding a new entry to this file is NACK.
> @@ -18,6 +18,7 @@ arch/arm/kernel/head.o
> arch/csky/kernel/head.o
> arch/hexagon/kernel/head.o
> arch/loongarch/kernel/head.o
> +arch/loongarch/kernel/head.o_thinlto_native
> arch/m68k/68000/head.o
> arch/m68k/coldfire/head.o
> arch/m68k/kernel/head.o
> --
> 2.49.0.1143.g0be31eac6b-goog
>
>
--
Best Regards
Masahiro Yamada
On Sat, May 24, 2025 at 11:15 AM Masahiro Yamada <masahiroy@kernel.org> wrote:
>
> On Thu, May 22, 2025 at 6:35 AM <xur@google.com> wrote:
> >
> > From: Rong Xu <xur@google.com>
> >
> > Add distributed ThinLTO build support for the Linux kernel.
> > This new mode offers several advantages: (1) Increased
> > flexibility in handling user-specified build options.
> > (2) Improved user-friendliness for developers. (3) Greater
> > convenience for integrating with objtool and livepatch.
>
> I did not understand any of these benefits
> despite a lot of added complexity.
I'm sorry you felt this way. For someone with my background in
compilers, especially with extensive work in ThinLTO, distributed
ThinLTO is unquestionably the superior choice.
>
>
>
>
>
>
> > Note that "distributed" in this context refers to a term
> > that differentiates in-process ThinLTO builds by invoking
> > backend compilation through the linker, not necessarily
> > building in distributed environments.
> >
> > Distributed ThinLTO is enabled via the
> > `CONFIG_LTO_CLANG_THIN_DIST` Kconfig option. For example:
> > > make LLVM=1 defconfig
> > > scripts/config -e LTO_CLANG_THIN_DIST
> > > make LLVM=1 oldconfig
> > > make LLVM=1 vmlinux -j <..>
> >
> > The implementation changes the top-level Makefile with a
> > macro for generating `vmlinux.o` for distributed ThinLTO
> > builds. It uses the existing Kbuild infrastructure to
> > perform two recursive passes through the subdirectories.
> > The first pass generates LLVM IR object files, similar to
> > in-process ThinLTO. Following the thin-link stage, a second
> > pass compiles these IR files into the final native object
> > files. The build rules and actions for this two-pass process
> > are primarily implemented in `scripts/Makefile.build`.
> >
> > Currently, this patch focuses on building the main kernel
> > image (`vmlinux`) only. Support for building kernel modules
> > using this method is planned for a subsequent patch.
>
> The announcement of the future plan is even scary,
> especially after looking at your implementation.
I'm sure this patch can be improved (or rewritten).
>
>
> >
> > Tested on the following arch: x86, arm64, loongarch, and
> > riscv.
> >
> > Some implementation details can be found here:
> > https://discourse.llvm.org/t/rfc-distributed-thinlto-build-for-kernel/85934
> >
> > Signed-off-by: Rong Xu <xur@google.com>
> > ---
> > Changelog since v1:
> > - Updated the description in arch/Kconfig based on feedback
> > from Nathan Chancellor
> > - Revised file suffixes: .final_o -> .o.thinlto.native, and
> > .final_a -> .a.thinlto.native
> > - Updated list of ignored files in .gitignore
> >
> > Changelog since v2:
> > - Changed file suffixes: .o.thinlto.native -> .o_thinlto_native,
> > and .a.thinlto.native -> .a_thinlto_native so that basename
> > works as intended.
> > - Tested the patch with AutoFDO and Propeller.
> > ---
> > .gitignore | 3 ++
> > MAINTAINERS | 5 +++
> > Makefile | 40 ++++++++++++++++++++---
> > arch/Kconfig | 19 +++++++++++
> > scripts/Makefile.build | 52 +++++++++++++++++++++++++++---
> > scripts/Makefile.lib | 7 +++-
> > scripts/Makefile.vmlinux_o | 16 +++++++---
> > scripts/Makefile.vmlinux_thinlink | 53 +++++++++++++++++++++++++++++++
> > scripts/head-object-list.txt | 1 +
> > 9 files changed, 181 insertions(+), 15 deletions(-)
> > create mode 100644 scripts/Makefile.vmlinux_thinlink
> >
> > diff --git a/.gitignore b/.gitignore
> > index f2f63e47fb886..b83a68185ef46 100644
> > --- a/.gitignore
> > +++ b/.gitignore
> > @@ -12,6 +12,7 @@
> > #
> > .*
> > *.a
> > +*.a_thinlto_native
> > *.asn1.[ch]
> > *.bin
> > *.bz2
> > @@ -39,6 +40,7 @@
> > *.mod.c
> > *.o
> > *.o.*
> > +*.o_thinlto_native
> > *.patch
> > *.rmeta
> > *.rpm
> > @@ -64,6 +66,7 @@ modules.order
> > /vmlinux
> > /vmlinux.32
> > /vmlinux.map
> > +/vmlinux.thinlink
> > /vmlinux.symvers
> > /vmlinux.unstripped
> > /vmlinux-gdb.py
> > diff --git a/MAINTAINERS b/MAINTAINERS
> > index d48dd6726fe6b..f54090f364c93 100644
> > --- a/MAINTAINERS
> > +++ b/MAINTAINERS
> > @@ -5790,6 +5790,11 @@ F: scripts/Makefile.clang
> > F: scripts/clang-tools/
> > K: \b(?i:clang|llvm)\b
> >
> > +CLANG/LLVM THINLTO DISTRIBUTED BUILD
> > +M: Rong Xu <xur@google.com>
> > +S: Supported
> > +F: scripts/Makefile.vmlinux_thinlink
> > +
> > CLK API
> > M: Russell King <linux@armlinux.org.uk>
> > L: linux-clk@vger.kernel.org
> > diff --git a/Makefile b/Makefile
> > index a9edd03036537..8fbff2ab87ebd 100644
> > --- a/Makefile
> > +++ b/Makefile
> > @@ -298,7 +298,8 @@ no-dot-config-targets := $(clean-targets) \
> > outputmakefile rustavailable rustfmt rustfmtcheck
> > no-sync-config-targets := $(no-dot-config-targets) %install modules_sign kernelrelease \
> > image_name
> > -single-targets := %.a %.i %.ko %.lds %.ll %.lst %.mod %.o %.rsi %.s %/
> > +single-targets := %.a %.i %.ko %.lds %.ll %.lst %.mod %.o %.rsi %.s %.o_thinlto_native \
> > + %.a_thinlto_native %.o.thinlto.bc %/
> >
> > config-build :=
> > mixed-build :=
> > @@ -991,10 +992,10 @@ export CC_FLAGS_SCS
> > endif
> >
> > ifdef CONFIG_LTO_CLANG
> > -ifdef CONFIG_LTO_CLANG_THIN
> > -CC_FLAGS_LTO := -flto=thin -fsplit-lto-unit
> > -else
> > +ifdef CONFIG_LTO_CLANG_FULL
> > CC_FLAGS_LTO := -flto
> > +else # for CONFIG_LTO_CLANG_THIN or CONFIG_LTO_CLANG_THIN_DIST
> > +CC_FLAGS_LTO := -flto=thin -fsplit-lto-unit
> > endif
> > CC_FLAGS_LTO += -fvisibility=hidden
> >
> > @@ -1213,8 +1214,34 @@ vmlinux.a: $(KBUILD_VMLINUX_OBJS) scripts/head-object-list.txt FORCE
> > $(call if_changed,ar_vmlinux.a)
> >
> > PHONY += vmlinux_o
> > +ifdef CONFIG_LTO_CLANG_THIN_DIST
> > +vmlinux.thinlink: vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> > + $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.vmlinux_thinlink
> > +targets += vmlinux.thinlink
> > +
> > +vmlinux.a_thinlto_native := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_OBJS))
> > +quiet_cmd_ar_vmlinux.a_thinlto_native = AR $@
> > + cmd_ar_vmlinux.a_thinlto_native = \
> > + rm -f $@; \
> > + $(AR) cDPrST $@ $(vmlinux.a_thinlto_native); \
> > + $(AR) mPiT $$($(AR) t $@ | sed -n 1p) $@ $$($(AR) t $@ | grep -F -f $(srctree)/scripts/head-object-list.txt)
> > +
> > +define rule_gen_vmlinux.a_thinlto_native
> > + +$(Q)$(MAKE) $(build)=. need-builtin=1 thinlto_final_pass=1 need-modorder=1 built-in.a_thinlto_native
>
>
> I really do not understand why this needs the second
> recursion, as you already know the list of objects
> by 'ar t vmlinux.a'.
The second recursion is to get the per-file cflags reliably.
>
>
>
>
>
>
>
>
> > + $(call cmd_and_savecmd,ar_vmlinux.a_thinlto_native)
> > +endef
> > +
> > +vmlinux.a_thinlto_native: vmlinux.thinlink scripts/head-object-list.txt FORCE
> > + $(call if_changed_rule,gen_vmlinux.a_thinlto_native)
> > +
> > +targets += vmlinux.a_thinlto_native
> > +
> > +vmlinux_o: vmlinux.a_thinlto_native
> > + $(Q)$(MAKE) thinlto_final_pass=1 -f $(srctree)/scripts/Makefile.vmlinux_o
> > +else
> > vmlinux_o: vmlinux.a $(KBUILD_VMLINUX_LIBS)
> > $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.vmlinux_o
> > +endif
> >
> > vmlinux.o modules.builtin.modinfo modules.builtin: vmlinux_o
> > @:
> > @@ -1572,7 +1599,8 @@ CLEAN_FILES += vmlinux.symvers modules-only.symvers \
> > modules.builtin.ranges vmlinux.o.map vmlinux.unstripped \
> > compile_commands.json rust/test \
> > rust-project.json .vmlinux.objs .vmlinux.export.c \
> > - .builtin-dtbs-list .builtin-dtb.S
> > + .builtin-dtbs-list .builtin-dtb.S \
> > + .vmlinux_thinlto_bc_files vmlinux.thinlink
> >
> > # Directories & files removed with 'make mrproper'
> > MRPROPER_FILES += include/config include/generated \
> > @@ -2023,6 +2051,8 @@ clean: $(clean-dirs)
> > -o -name '*.symtypes' -o -name 'modules.order' \
> > -o -name '*.c.[012]*.*' \
> > -o -name '*.ll' \
> > + -o -name '*.a_thinlto_native' -o -name '*.o_thinlto_native' \
> > + -o -name '*.o.thinlto.bc' \
> > -o -name '*.gcno' \
> > \) -type f -print \
> > -o -name '.tmp_*' -print \
> > diff --git a/arch/Kconfig b/arch/Kconfig
> > index b0adb665041f1..30dccda07c671 100644
> > --- a/arch/Kconfig
> > +++ b/arch/Kconfig
> > @@ -810,6 +810,25 @@ config LTO_CLANG_THIN
> > https://clang.llvm.org/docs/ThinLTO.html
> >
> > If unsure, say Y.
> > +
> > +config LTO_CLANG_THIN_DIST
> > + bool "Clang ThinLTO in distributed mode (EXPERIMENTAL)"
> > + depends on HAS_LTO_CLANG && ARCH_SUPPORTS_LTO_CLANG_THIN
> > + select LTO_CLANG
> > + help
> > + This option enables Clang's ThinLTO in distributed build mode.
> > + In this mode, the linker performs the thin-link, generating
> > + ThinLTO index files. Subsequently, the build system explicitly
> > + invokes ThinLTO backend compilation using these index files
> > + and pre-linked IR objects. The resulting native object files
> > + are with the .o_thinlto_native suffix.
> > +
> > + This build mode offers improved visibility into the ThinLTO
> > + process through explicit subcommand exposure. It also makes
> > + final native object files directly available, benefiting
> > + tools like objtool and kpatch. Additionally, it provides
> > + crucial granular control over back-end options, enabling
> > + module-specific compiler options, and simplifies debugging.
> > endchoice
> >
> > config ARCH_SUPPORTS_AUTOFDO_CLANG
> > diff --git a/scripts/Makefile.build b/scripts/Makefile.build
> > index 13dcd86e74ca8..338e1aec0eaa3 100644
> > --- a/scripts/Makefile.build
> > +++ b/scripts/Makefile.build
> > @@ -50,18 +50,23 @@ endif
> >
> > # ===========================================================================
> >
> > +builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a)
> > +ifeq ($(thinlto_final_pass),1)
> > +builtin_suffix :=.a_thinlto_native
> > +endif
> > +
> > # subdir-builtin and subdir-modorder may contain duplications. Use $(sort ...)
> > -subdir-builtin := $(sort $(filter %/built-in.a, $(real-obj-y)))
> > +subdir-builtin := $(sort $(filter %/built-in$(builtin_suffix), $(real-obj-y)))
> > subdir-modorder := $(sort $(filter %/modules.order, $(obj-m)))
> >
> > targets-for-builtin := $(extra-y)
> >
> > ifneq ($(strip $(lib-y) $(lib-m) $(lib-)),)
> > -targets-for-builtin += $(obj)/lib.a
> > +targets-for-builtin += $(obj)/lib$(builtin_suffix)
> > endif
> >
> > ifdef need-builtin
> > -targets-for-builtin += $(obj)/built-in.a
> > +targets-for-builtin += $(obj)/built-in$(builtin_suffix)
> > endif
> >
> > targets-for-modules := $(foreach x, o mod, \
> > @@ -337,6 +342,10 @@ $(obj)/%.o: $(obj)/%.S FORCE
> > targets += $(filter-out $(subdir-builtin), $(real-obj-y))
> > targets += $(filter-out $(subdir-modorder), $(real-obj-m))
> > targets += $(lib-y) $(always-y)
> > +ifeq ($(builtin_suffix),.a_thinlto_native)
> > +native_targets = $(patsubst,%.o,%.o_thinlto_native,$(targets))
> > +targets += $(native_targets)
> > +endif
> >
> > # Linker scripts preprocessor (.lds.S -> .lds)
> > # ---------------------------------------------------------------------------
> > @@ -347,6 +356,24 @@ quiet_cmd_cpp_lds_S = LDS $@
> > $(obj)/%.lds: $(src)/%.lds.S FORCE
> > $(call if_changed_dep,cpp_lds_S)
> >
> > +ifdef CONFIG_LTO_CLANG_THIN_DIST
> > +# Generate .o_thinlto_native (obj) from .o (bitcode) file
> > +# ---------------------------------------------------------------------------
> > +quiet_cmd_cc_o_bc = CC $(quiet_modtag) $@
> > +
> > +cmd_cc_o_bc = $(if $(filter bitcode, $(shell file -b $<)),$(CC) \
> > + $(filter-out -Wp% $(LINUXINCLUDE) %.h.gch %.h -D% \
> > + -flto=thin, $(c_flags)) \
> > + -Wno-unused-command-line-argument \
> > + -x ir -fthinlto-index=$<.thinlto.bc -c -o $@ \
> > + $(if $(findstring ../,$<), \
> > + $$(realpath --relative-to=$(srcroot) $<), $<), \
> > + cp $< $@)
> > +
> > +$(obj)/%.o_thinlto_native: $(obj)/%.o FORCE
> > + $(call if_changed,cc_o_bc)
> > +endif
> > +
> > # ASN.1 grammar
> > # ---------------------------------------------------------------------------
> > quiet_cmd_asn1_compiler = ASN.1 $(basename $@).[ch]
> > @@ -360,7 +387,7 @@ $(obj)/%.asn1.c $(obj)/%.asn1.h: $(src)/%.asn1 $(objtree)/scripts/asn1_compiler
> > # ---------------------------------------------------------------------------
> >
> > # To build objects in subdirs, we need to descend into the directories
> > -$(subdir-builtin): $(obj)/%/built-in.a: $(obj)/% ;
> > +$(subdir-builtin): $(obj)/%/built-in$(builtin_suffix): $(obj)/% ;
> > $(subdir-modorder): $(obj)/%/modules.order: $(obj)/% ;
> >
> > #
> > @@ -377,6 +404,12 @@ quiet_cmd_ar_builtin = AR $@
> > $(obj)/built-in.a: $(real-obj-y) FORCE
> > $(call if_changed,ar_builtin)
> >
> > +ifdef CONFIG_LTO_CLANG_THIN_DIST
> > +# Rule to compile a set of .o_thinlto_native files into one .a_thinlto_native file.
> > +$(obj)/built-in.a_thinlto_native: $(patsubst %.o,%.o_thinlto_native,$(real-obj-y)) FORCE
> > + $(call if_changed,ar_builtin)
> > +endif
> > +
> > # This is a list of build artifacts from the current Makefile and its
> > # sub-directories. The timestamp should be updated when any of the member files.
> >
> > @@ -394,6 +427,14 @@ $(obj)/modules.order: $(obj-m) FORCE
> > $(obj)/lib.a: $(lib-y) FORCE
> > $(call if_changed,ar)
> >
> > +ifdef CONFIG_LTO_CLANG_THIN_DIST
> > +quiet_cmd_ar_native = AR $@
> > + cmd_ar_native = rm -f $@; $(AR) cDPrsT $@ $(patsubst %.o,%.o_thinlto_native,$(real-prereqs))
> > +
> > +$(obj)/lib.a_thinlto_native: $(patsubst %.o,%.o_thinlto_native,$(lib-y)) FORCE
> > + $(call if_changed,ar_native)
> > +endif
> > +
> > quiet_cmd_ld_multi_m = LD [M] $@
> > cmd_ld_multi_m = $(LD) $(ld_flags) -r -o $@ @$< $(cmd_objtool)
> >
> > @@ -459,7 +500,8 @@ $(single-subdir-goals): $(single-subdirs)
> > PHONY += $(subdir-ym)
> > $(subdir-ym):
> > $(Q)$(MAKE) $(build)=$@ \
> > - need-builtin=$(if $(filter $@/built-in.a, $(subdir-builtin)),1) \
> > + need-builtin=$(if $(filter $@/built-in$(builtin_suffix), $(subdir-builtin)),1) \
> > + thinlto_final_pass=$(if $(filter .a_thinlto_native, $(builtin_suffix)),1) \
> > need-modorder=$(if $(filter $@/modules.order, $(subdir-modorder)),1) \
> > $(filter $@/%, $(single-subdir-goals))
> >
> > diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
> > index 2fe73cda0bddb..9cfd23590334d 100644
> > --- a/scripts/Makefile.lib
> > +++ b/scripts/Makefile.lib
> > @@ -34,8 +34,13 @@ else
> > obj-m := $(filter-out %/, $(obj-m))
> > endif
> >
> > +builtin_suffix := $(if $(filter %.a_thinlto_native, $(MAKECMDGOALS)),.a_thinlto_native,.a)
> > +ifeq ($(thinlto_final_pass),1)
> > + builtin_suffix :=.a_thinlto_native
> > +endif
> > +
> > ifdef need-builtin
> > -obj-y := $(patsubst %/, %/built-in.a, $(obj-y))
> > +obj-y := $(patsubst %/, %/built-in$(builtin_suffix), $(obj-y))
> > else
> > obj-y := $(filter-out %/, $(obj-y))
> > endif
> > diff --git a/scripts/Makefile.vmlinux_o b/scripts/Makefile.vmlinux_o
> > index b024ffb3e2018..f9abc45a68b36 100644
> > --- a/scripts/Makefile.vmlinux_o
> > +++ b/scripts/Makefile.vmlinux_o
> > @@ -9,6 +9,14 @@ include $(srctree)/scripts/Kbuild.include
> > # for objtool
> > include $(srctree)/scripts/Makefile.lib
> >
> > +ifeq ($(thinlto_final_pass),1)
> > +vmlinux_a := vmlinux.a_thinlto_native
> > +vmlinux_libs := $(patsubst %.a,%.a_thinlto_native,$(KBUILD_VMLINUX_LIBS))
> > +else
> > +vmlinux_a := vmlinux.a
> > +vmlinux_libs := $(KBUILD_VMLINUX_LIBS)
> > +endif
> > +
> > # Generate a linker script to ensure correct ordering of initcalls for Clang LTO
> > # ---------------------------------------------------------------------------
> >
> > @@ -18,7 +26,7 @@ quiet_cmd_gen_initcalls_lds = GEN $@
> > $(PERL) $(real-prereqs) > $@
> >
> > .tmp_initcalls.lds: $(srctree)/scripts/generate_initcall_order.pl \
> > - vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> > + $(vmlinux_a) $(vmlinux_libs) FORCE
> > $(call if_changed,gen_initcalls_lds)
> >
> > targets := .tmp_initcalls.lds
> > @@ -59,8 +67,8 @@ quiet_cmd_ld_vmlinux.o = LD $@
> > $(LD) ${KBUILD_LDFLAGS} -r -o $@ \
> > $(vmlinux-o-ld-args-y) \
> > $(addprefix -T , $(initcalls-lds)) \
> > - --whole-archive vmlinux.a --no-whole-archive \
> > - --start-group $(KBUILD_VMLINUX_LIBS) --end-group \
> > + --whole-archive $(vmlinux_a) --no-whole-archive \
> > + --start-group $(vmlinux_libs) --end-group \
> > $(cmd_objtool)
> >
> > define rule_ld_vmlinux.o
> > @@ -68,7 +76,7 @@ define rule_ld_vmlinux.o
> > $(call cmd,gen_objtooldep)
> > endef
> >
> > -vmlinux.o: $(initcalls-lds) vmlinux.a $(KBUILD_VMLINUX_LIBS) FORCE
> > +vmlinux.o: $(initcalls-lds) $(vmlinux_a) $(vmlinux_libs) FORCE
> > $(call if_changed_rule,ld_vmlinux.o)
> >
> > targets += vmlinux.o
> > diff --git a/scripts/Makefile.vmlinux_thinlink b/scripts/Makefile.vmlinux_thinlink
> > new file mode 100644
> > index 0000000000000..13e4026c7d45b
> > --- /dev/null
> > +++ b/scripts/Makefile.vmlinux_thinlink
> > @@ -0,0 +1,53 @@
> > +# SPDX-License-Identifier: GPL-2.0-only
> > +
> > +PHONY := __default
> > +__default: vmlinux.thinlink
> > +
> > +include include/config/auto.conf
> > +include $(srctree)/scripts/Kbuild.include
> > +
> > +
> > +# Generate a linker script to ensure correct ordering of initcalls for Clang LTO
> > +# ---------------------------------------------------------------------------
> > +
> > +quiet_cmd_gen_initcalls_lds = GEN $@
> > + cmd_gen_initcalls_lds = \
> > + $(PYTHON3) $(srctree)/scripts/jobserver-exec \
> > + $(PERL) $(real-prereqs) > $@
> > +
> > +.tmp_initcalls_thinlink.lds: $(srctree)/scripts/generate_initcall_order.pl \
> > + vmlinux.a FORCE
> > + $(call if_changed,gen_initcalls_lds)
> > +
> > +targets := .tmp_initcalls_thinlink.lds
> > +
> > +initcalls-lds := .tmp_initcalls_thinlink.lds
> > +
> > +quiet_cmd_ld_vmlinux.thinlink = LD $@
> > + cmd_ld_vmlinux.thinlink = \
> > + $(AR) t vmlinux.a > .vmlinux_thinlto_bc_files; \
> > + $(LD) ${KBUILD_LDFLAGS} -r $(addprefix -T , $(initcalls-lds)) \
> > + --thinlto-index-only @.vmlinux_thinlto_bc_files; \
> > + touch vmlinux.thinlink
> > +
> > +vmlinux.thinlink: vmlinux.a $(initcalls-lds) FORCE
> > + $(call if_changed,ld_vmlinux.thinlink)
> > +
> > +targets += vmlinux.thinlink
> > +
> > +# Add FORCE to the prerequisites of a target to force it to be always rebuilt.
> > +# ---------------------------------------------------------------------------
> > +
> > +PHONY += FORCE
> > +FORCE:
> > +
> > +# Read all saved command lines and dependencies for the $(targets) we
> > +# may be building above, using $(if_changed{,_dep}). As an
> > +# optimization, we don't need to read them if the target does not
> > +# exist, we will rebuild anyway in that case.
> > +
> > +existing-targets := $(wildcard $(sort $(targets)))
> > +
> > +-include $(foreach f,$(existing-targets),$(dir $(f)).$(notdir $(f)).cmd)
> > +
> > +.PHONY: $(PHONY)
> > diff --git a/scripts/head-object-list.txt b/scripts/head-object-list.txt
> > index 7274dfc65af60..90710b87a3877 100644
> > --- a/scripts/head-object-list.txt
> > +++ b/scripts/head-object-list.txt
>
> Adding a new entry to this file is NACK.
Then this platform cannot use LTO?
>
>
>
> > @@ -18,6 +18,7 @@ arch/arm/kernel/head.o
> > arch/csky/kernel/head.o
> > arch/hexagon/kernel/head.o
> > arch/loongarch/kernel/head.o
> > +arch/loongarch/kernel/head.o_thinlto_native
> > arch/m68k/68000/head.o
> > arch/m68k/coldfire/head.o
> > arch/m68k/kernel/head.o
> > --
> > 2.49.0.1143.g0be31eac6b-goog
> >
> >
>
>
> --
> Best Regards
> Masahiro Yamada
© 2016 - 2025 Red Hat, Inc.