Message ID | 87tuslxhry.fsf@mpe.ellerman.id.au (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | powerpc VDSO files being unnecessarily rebuilt | expand |
On Thu, Dec 17, 2020 at 11:56 AM Michael Ellerman <mpe@ellerman.id.au> wrote: > > Hi all, > > Since the merge of the C VDSO I see we are repeatedly rebuilding some > files in the VDSO, eg: > > $ make V=2 > make[1]: Entering directory '/home/michael/linux/build~' > GEN Makefile > CALL /home/michael/linux/scripts/checksyscalls.sh - due to target missing > CALL /home/michael/linux/scripts/atomic/check-atomics.sh - due to target missing > CHK include/generated/compile.h > CC arch/powerpc/kernel/vdso64/vgettimeofday.o - due to vgettimeofday.o not in $(targets) > > This then causes multiple other files to be rebuilt. > > So the obvious fix is to add it to targets: > > diff --git a/arch/powerpc/kernel/vdso64/Makefile b/arch/powerpc/kernel/vdso64/Makefile > index d365810a689a..5386532866ce 100644 > --- a/arch/powerpc/kernel/vdso64/Makefile > +++ b/arch/powerpc/kernel/vdso64/Makefile > @@ -5,6 +5,7 @@ ARCH_REL_TYPE_ABS := R_PPC_JUMP_SLOT|R_PPC_GLOB_DAT|R_PPC_ADDR32|R_PPC_ADDR24|R_ > include $(srctree)/lib/vdso/Makefile > > obj-vdso64 = sigtramp.o gettimeofday.o datapage.o cacheflush.o note.o getcpu.o > +targets := $(obj-vdso64) vdso64.so.dbg > > ifneq ($(c-gettimeofday-y),) > CFLAGS_vgettimeofday.o += -include $(c-gettimeofday-y) > @@ -13,11 +14,11 @@ ifneq ($(c-gettimeofday-y),) > CFLAGS_vgettimeofday.o += -DDISABLE_BRANCH_PROFILING > CFLAGS_vgettimeofday.o += -ffreestanding -fasynchronous-unwind-tables > CFLAGS_REMOVE_vgettimeofday.o = $(CC_FLAGS_FTRACE) > + targets += vgettimeofday.o > endif > > # Build rules > > -targets := $(obj-vdso64) vdso64.so.dbg > obj-vdso64 := $(addprefix $(obj)/, $(obj-vdso64)) > > GCOV_PROFILE := n > > > But then I see it still rebuilt: > > CC arch/powerpc/kernel/vdso64/vgettimeofday.o - due to command line change > > > I'm not changing the command line, and AFAICS the .cmd file is not > changing either: > > $ make V=2 > ... > CC arch/powerpc/kernel/vdso64/vgettimeofday.o - due to command line change > > $ sha256sum build\~/arch/powerpc/kernel/vdso64/vgettimeofday.o > 7f635546bc2768c7b929d3de1724d83285f3cd54394fcd7104f8b1301d689d65 build~/arch/powerpc/kernel/vdso64/vgettimeofday.o > > $ make V=2 > ... > CC arch/powerpc/kernel/vdso64/vgettimeofday.o - due to command line change > > $ sha256sum build\~/arch/powerpc/kernel/vdso64/vgettimeofday.o > 7f635546bc2768c7b929d3de1724d83285f3cd54394fcd7104f8b1301d689d65 build~/arch/powerpc/kernel/vdso64/vgettimeofday.o > > > So any hints on what I'm missing here? > > cheers This is because PPC builds the vdso twice with different command arguments. First time: vdso_prepare: prepare0 $(if $(CONFIG_VDSO32),$(Q)$(MAKE) \ $(build)=arch/powerpc/kernel/vdso32 include/generated/vdso32-offsets.h) $(if $(CONFIG_PPC64),$(Q)$(MAKE) \ $(build)=arch/powerpc/kernel/vdso64 include/generated/vdso64-offsets.h) Second time: from arch/powerpc/kernel/Makefile For the first build, -Werror is missing because Kbuild directly descends into arch/powerpc/kernel/vdso[32,64]/. For the second build, arch/powerpc/Kbuild appends the following: subdir-ccflags-$(CONFIG_PPC_WERROR) := -Werror
On Thu, Dec 17, 2020 at 6:23 PM Masahiro Yamada <masahiroy@kernel.org> wrote: > > On Thu, Dec 17, 2020 at 11:56 AM Michael Ellerman <mpe@ellerman.id.au> wrote: > > > > Hi all, > > > > Since the merge of the C VDSO I see we are repeatedly rebuilding some > > files in the VDSO, eg: > > > > $ make V=2 > > make[1]: Entering directory '/home/michael/linux/build~' > > GEN Makefile > > CALL /home/michael/linux/scripts/checksyscalls.sh - due to target missing > > CALL /home/michael/linux/scripts/atomic/check-atomics.sh - due to target missing > > CHK include/generated/compile.h > > CC arch/powerpc/kernel/vdso64/vgettimeofday.o - due to vgettimeofday.o not in $(targets) > > > > This then causes multiple other files to be rebuilt. > > > > So the obvious fix is to add it to targets: > > > > diff --git a/arch/powerpc/kernel/vdso64/Makefile b/arch/powerpc/kernel/vdso64/Makefile > > index d365810a689a..5386532866ce 100644 > > --- a/arch/powerpc/kernel/vdso64/Makefile > > +++ b/arch/powerpc/kernel/vdso64/Makefile > > @@ -5,6 +5,7 @@ ARCH_REL_TYPE_ABS := R_PPC_JUMP_SLOT|R_PPC_GLOB_DAT|R_PPC_ADDR32|R_PPC_ADDR24|R_ > > include $(srctree)/lib/vdso/Makefile > > > > obj-vdso64 = sigtramp.o gettimeofday.o datapage.o cacheflush.o note.o getcpu.o > > +targets := $(obj-vdso64) vdso64.so.dbg > > > > ifneq ($(c-gettimeofday-y),) > > CFLAGS_vgettimeofday.o += -include $(c-gettimeofday-y) > > @@ -13,11 +14,11 @@ ifneq ($(c-gettimeofday-y),) > > CFLAGS_vgettimeofday.o += -DDISABLE_BRANCH_PROFILING > > CFLAGS_vgettimeofday.o += -ffreestanding -fasynchronous-unwind-tables > > CFLAGS_REMOVE_vgettimeofday.o = $(CC_FLAGS_FTRACE) > > + targets += vgettimeofday.o > > endif > > > > # Build rules > > > > -targets := $(obj-vdso64) vdso64.so.dbg > > obj-vdso64 := $(addprefix $(obj)/, $(obj-vdso64)) > > > > GCOV_PROFILE := n > > > > > > But then I see it still rebuilt: > > > > CC arch/powerpc/kernel/vdso64/vgettimeofday.o - due to command line change > > > > > > I'm not changing the command line, and AFAICS the .cmd file is not > > changing either: > > > > $ make V=2 > > ... > > CC arch/powerpc/kernel/vdso64/vgettimeofday.o - due to command line change > > > > $ sha256sum build\~/arch/powerpc/kernel/vdso64/vgettimeofday.o > > 7f635546bc2768c7b929d3de1724d83285f3cd54394fcd7104f8b1301d689d65 build~/arch/powerpc/kernel/vdso64/vgettimeofday.o > > > > $ make V=2 > > ... > > CC arch/powerpc/kernel/vdso64/vgettimeofday.o - due to command line change > > > > $ sha256sum build\~/arch/powerpc/kernel/vdso64/vgettimeofday.o > > 7f635546bc2768c7b929d3de1724d83285f3cd54394fcd7104f8b1301d689d65 build~/arch/powerpc/kernel/vdso64/vgettimeofday.o > > > > > > So any hints on what I'm missing here? > > > > cheers > > > This is because PPC builds the vdso twice > with different command arguments. > > > First time: > > vdso_prepare: prepare0 > $(if $(CONFIG_VDSO32),$(Q)$(MAKE) \ > $(build)=arch/powerpc/kernel/vdso32 > include/generated/vdso32-offsets.h) > $(if $(CONFIG_PPC64),$(Q)$(MAKE) \ > $(build)=arch/powerpc/kernel/vdso64 > include/generated/vdso64-offsets.h) > > > Second time: > from arch/powerpc/kernel/Makefile > > > > > > For the first build, -Werror is missing because > Kbuild directly descends into arch/powerpc/kernel/vdso[32,64]/. > > > For the second build, > > arch/powerpc/Kbuild appends the following: > > subdir-ccflags-$(CONFIG_PPC_WERROR) := -Werror > This is very dangerous because the vdso offsets are wrong. I think ppc followed arm64 implementation. Let me figure out how to fix arm64 vdso.
diff --git a/arch/powerpc/kernel/vdso64/Makefile b/arch/powerpc/kernel/vdso64/Makefile index d365810a689a..5386532866ce 100644 --- a/arch/powerpc/kernel/vdso64/Makefile +++ b/arch/powerpc/kernel/vdso64/Makefile @@ -5,6 +5,7 @@ ARCH_REL_TYPE_ABS := R_PPC_JUMP_SLOT|R_PPC_GLOB_DAT|R_PPC_ADDR32|R_PPC_ADDR24|R_ include $(srctree)/lib/vdso/Makefile obj-vdso64 = sigtramp.o gettimeofday.o datapage.o cacheflush.o note.o getcpu.o +targets := $(obj-vdso64) vdso64.so.dbg ifneq ($(c-gettimeofday-y),) CFLAGS_vgettimeofday.o += -include $(c-gettimeofday-y) @@ -13,11 +14,11 @@ ifneq ($(c-gettimeofday-y),) CFLAGS_vgettimeofday.o += -DDISABLE_BRANCH_PROFILING CFLAGS_vgettimeofday.o += -ffreestanding -fasynchronous-unwind-tables CFLAGS_REMOVE_vgettimeofday.o = $(CC_FLAGS_FTRACE) + targets += vgettimeofday.o endif # Build rules -targets := $(obj-vdso64) vdso64.so.dbg obj-vdso64 := $(addprefix $(obj)/, $(obj-vdso64)) GCOV_PROFILE := n