Is LTO_REFERENCE_INITCALL() still needed?

Is LTO_REFERENCE_INITCALL() still needed?

[Luis R. Rodriguez]

Andi,

is LTO_REFERENCE_INITCAL() still needed?

I see external stuff like this:

https://github.com/neo-technologies/linux-next/blob/master/scripts/Makefile.lto

But nothing quite obviously setting these upstream. I'm probably just
missing something obvious though. Either way I figured now would be a
good time to ask as I have other code that would use this, the linker
table stuff I've been working. If we can avoid propagating more use of
LTO_REFERENCE_INITCALL() then great. If we still need it its probably
good to document this a bit more as can't see anything super clear.

  Luis

Re: Is LTO_REFERENCE_INITCALL() still needed?

[Andi Kleen]

> But nothing quite obviously setting these upstream. I'm probably just
> missing something obvious though. Either way I figured now would be a
> good time to ask as I have other code that would use this, the linker
> table stuff I've been working. If we can avoid propagating more use of
> LTO_REFERENCE_INITCALL() then great. If we still need it its probably
> good to document this a bit more as can't see anything super clear.

It was a work around for a gcc 4.7 LTO problem. I guess it could be
dropped by now.

Yes I've been slow in submitting the LTO patches, even though they
have a quite active out of tree user base. Will try to do better.

-Andi

Re: Is LTO_REFERENCE_INITCALL() still needed?

[Luis R. Rodriguez]

On Fri, Jun 24, 2016 at 02:07:53PM -0700, Andi Kleen wrote:
> > But nothing quite obviously setting these upstream. I'm probably just
> > missing something obvious though. Either way I figured now would be a
> > good time to ask as I have other code that would use this, the linker
> > table stuff I've been working. If we can avoid propagating more use of
> > LTO_REFERENCE_INITCALL() then great. If we still need it its probably
> > good to document this a bit more as can't see anything super clear.
> 
> It was a work around for a gcc 4.7 LTO problem. I guess it could be
> dropped by now.
> 
> Yes I've been slow in submitting the LTO patches, even though they
> have a quite active out of tree user base. Will try to do better.

Would something like this make sense then? Tested by 0-day.

>From 682579b548ccf5d33349f32fcbe0922f5c29f04d Mon Sep 17 00:00:00 2001
From: "Luis R. Rodriguez" <mcgrof@kernel.org>
Date: Mon, 27 Jun 2016 14:25:14 -0700
Subject: [PATCH] x86: remove LTO flags

The setup for LTO never made it upstream, and although this has
some users, this is now really old stuff for a gcc 4.7 LTO problem.

Signed-off-by: Luis R. Rodriguez <mcgrof@kernel.org>
---
 arch/x86/entry/vdso/Makefile |  6 ++----
 include/linux/init.h         | 20 +-------------------
 kernel/Makefile              |  3 ---
 scripts/Makefile.build       |  2 +-
 4 files changed, 4 insertions(+), 27 deletions(-)

diff --git a/arch/x86/entry/vdso/Makefile b/arch/x86/entry/vdso/Makefile
index 3288dfa63706..d0f1cd210892 100644
--- a/arch/x86/entry/vdso/Makefile
+++ b/arch/x86/entry/vdso/Makefile
@@ -2,7 +2,6 @@
 # Building vDSO images for x86.
 #
 
-KBUILD_CFLAGS += $(DISABLE_LTO)
 KASAN_SANITIZE			:= n
 UBSAN_SANITIZE			:= n
 OBJECT_FILES_NON_STANDARD	:= y
@@ -49,8 +48,7 @@ export CPPFLAGS_vdso.lds += -P -C
 
 VDSO_LDFLAGS_vdso.lds = -m64 -Wl,-soname=linux-vdso.so.1 \
 			-Wl,--no-undefined \
-			-Wl,-z,max-page-size=4096 -Wl,-z,common-page-size=4096 \
-			$(DISABLE_LTO)
+			-Wl,-z,max-page-size=4096 -Wl,-z,common-page-size=4096
 
 $(obj)/vdso64.so.dbg: $(src)/vdso.lds $(vobjs) FORCE
 	$(call if_changed,vdso)
@@ -171,7 +169,7 @@ quiet_cmd_vdso = VDSO    $@
 		 sh $(srctree)/$(src)/checkundef.sh '$(NM)' '$@'
 
 VDSO_LDFLAGS = -fPIC -shared $(call cc-ldoption, -Wl$(comma)--hash-style=both) \
-	$(call cc-ldoption, -Wl$(comma)--build-id) -Wl,-Bsymbolic $(LTO_CFLAGS)
+	$(call cc-ldoption, -Wl$(comma)--build-id) -Wl,-Bsymbolic
 GCOV_PROFILE := n
 
 #
diff --git a/include/linux/init.h b/include/linux/init.h
index 88646cb136e0..0370efacc839 100644
--- a/include/linux/init.h
+++ b/include/linux/init.h
@@ -157,23 +157,6 @@ extern bool initcall_debug;
 
 #ifndef __ASSEMBLY__
 
-#ifdef CONFIG_LTO
-/* Work around a LTO gcc problem: when there is no reference to a variable
- * in a module it will be moved to the end of the program. This causes
- * reordering of initcalls which the kernel does not like.
- * Add a dummy reference function to avoid this. The function is
- * deleted by the linker.
- */
-#define LTO_REFERENCE_INITCALL(x) \
-	; /* yes this is needed */			\
-	static __used __exit void *reference_##x(void)	\
-	{						\
-		return &x;				\
-	}
-#else
-#define LTO_REFERENCE_INITCALL(x)
-#endif
-
 /* initcalls are now grouped by functionality into separate 
  * subsections. Ordering inside the subsections is determined
  * by link order. 
@@ -186,8 +169,7 @@ extern bool initcall_debug;
 
 #define __define_initcall(fn, id) \
 	static initcall_t __initcall_##fn##id __used \
-	__attribute__((__section__(".initcall" #id ".init"))) = fn; \
-	LTO_REFERENCE_INITCALL(__initcall_##fn##id)
+	__attribute__((__section__(".initcall" #id ".init"))) = fn;
 
 /*
  * Early initcalls run before initializing SMP.
diff --git a/kernel/Makefile b/kernel/Makefile
index e2ec54e2b952..f80ce824bb4f 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -29,9 +29,6 @@ KCOV_INSTRUMENT_extable.o := n
 KCOV_INSTRUMENT_kcov.o := n
 KASAN_SANITIZE_kcov.o := n
 
-# cond_syscall is currently not LTO compatible
-CFLAGS_sys_ni.o = $(DISABLE_LTO)
-
 obj-y += sched/
 obj-y += locking/
 obj-y += power/
diff --git a/scripts/Makefile.build b/scripts/Makefile.build
index 11602e5efb3b..a5dd16ffb06e 100644
--- a/scripts/Makefile.build
+++ b/scripts/Makefile.build
@@ -147,7 +147,7 @@ $(multi-objs-y:.o=.s)   : modname = $(modname-multi)
 $(multi-objs-y:.o=.lst) : modname = $(modname-multi)
 
 quiet_cmd_cc_s_c = CC $(quiet_modtag)  $@
-cmd_cc_s_c       = $(CC) $(c_flags) $(DISABLE_LTO) -fverbose-asm -S -o $@ $<
+cmd_cc_s_c       = $(CC) $(c_flags) -fverbose-asm -S -o $@ $<
 
 $(obj)/%.s: $(src)/%.c FORCE
 	$(call if_changed_dep,cc_s_c)
-- 
2.8.4

[PATCH] x86: remove LTO flags

[Luis R. Rodriguez]

The setup for LTO never made it upstream, and although this has
some users, this is now really old stuff for a gcc 4.7 LTO problem.

Signed-off-by: Luis R. Rodriguez <mcgrof@kernel.org>
---
 arch/x86/entry/vdso/Makefile |  6 ++----
 include/linux/init.h         | 20 +-------------------
 kernel/Makefile              |  3 ---
 scripts/Makefile.build       |  2 +-
 4 files changed, 4 insertions(+), 27 deletions(-)

diff --git a/arch/x86/entry/vdso/Makefile b/arch/x86/entry/vdso/Makefile
index 3288dfa63706..d0f1cd210892 100644
--- a/arch/x86/entry/vdso/Makefile
+++ b/arch/x86/entry/vdso/Makefile
@@ -2,7 +2,6 @@
 # Building vDSO images for x86.
 #
 
-KBUILD_CFLAGS += $(DISABLE_LTO)
 KASAN_SANITIZE			:= n
 UBSAN_SANITIZE			:= n
 OBJECT_FILES_NON_STANDARD	:= y
@@ -49,8 +48,7 @@ export CPPFLAGS_vdso.lds += -P -C
 
 VDSO_LDFLAGS_vdso.lds = -m64 -Wl,-soname=linux-vdso.so.1 \
 			-Wl,--no-undefined \
-			-Wl,-z,max-page-size=4096 -Wl,-z,common-page-size=4096 \
-			$(DISABLE_LTO)
+			-Wl,-z,max-page-size=4096 -Wl,-z,common-page-size=4096
 
 $(obj)/vdso64.so.dbg: $(src)/vdso.lds $(vobjs) FORCE
 	$(call if_changed,vdso)
@@ -171,7 +169,7 @@ quiet_cmd_vdso = VDSO    $@
 		 sh $(srctree)/$(src)/checkundef.sh '$(NM)' '$@'
 
 VDSO_LDFLAGS = -fPIC -shared $(call cc-ldoption, -Wl$(comma)--hash-style=both) \
-	$(call cc-ldoption, -Wl$(comma)--build-id) -Wl,-Bsymbolic $(LTO_CFLAGS)
+	$(call cc-ldoption, -Wl$(comma)--build-id) -Wl,-Bsymbolic
 GCOV_PROFILE := n
 
 #
diff --git a/include/linux/init.h b/include/linux/init.h
index 88646cb136e0..0370efacc839 100644
--- a/include/linux/init.h
+++ b/include/linux/init.h
@@ -157,23 +157,6 @@ extern bool initcall_debug;
 
 #ifndef __ASSEMBLY__
 
-#ifdef CONFIG_LTO
-/* Work around a LTO gcc problem: when there is no reference to a variable
- * in a module it will be moved to the end of the program. This causes
- * reordering of initcalls which the kernel does not like.
- * Add a dummy reference function to avoid this. The function is
- * deleted by the linker.
- */
-#define LTO_REFERENCE_INITCALL(x) \
-	; /* yes this is needed */			\
-	static __used __exit void *reference_##x(void)	\
-	{						\
-		return &x;				\
-	}
-#else
-#define LTO_REFERENCE_INITCALL(x)
-#endif
-
 /* initcalls are now grouped by functionality into separate 
  * subsections. Ordering inside the subsections is determined
  * by link order. 
@@ -186,8 +169,7 @@ extern bool initcall_debug;
 
 #define __define_initcall(fn, id) \
 	static initcall_t __initcall_##fn##id __used \
-	__attribute__((__section__(".initcall" #id ".init"))) = fn; \
-	LTO_REFERENCE_INITCALL(__initcall_##fn##id)
+	__attribute__((__section__(".initcall" #id ".init"))) = fn;
 
 /*
  * Early initcalls run before initializing SMP.
diff --git a/kernel/Makefile b/kernel/Makefile
index e2ec54e2b952..f80ce824bb4f 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -29,9 +29,6 @@ KCOV_INSTRUMENT_extable.o := n
 KCOV_INSTRUMENT_kcov.o := n
 KASAN_SANITIZE_kcov.o := n
 
-# cond_syscall is currently not LTO compatible
-CFLAGS_sys_ni.o = $(DISABLE_LTO)
-
 obj-y += sched/
 obj-y += locking/
 obj-y += power/
diff --git a/scripts/Makefile.build b/scripts/Makefile.build
index 11602e5efb3b..a5dd16ffb06e 100644
--- a/scripts/Makefile.build
+++ b/scripts/Makefile.build
@@ -147,7 +147,7 @@ $(multi-objs-y:.o=.s)   : modname = $(modname-multi)
 $(multi-objs-y:.o=.lst) : modname = $(modname-multi)
 
 quiet_cmd_cc_s_c = CC $(quiet_modtag)  $@
-cmd_cc_s_c       = $(CC) $(c_flags) $(DISABLE_LTO) -fverbose-asm -S -o $@ $<
+cmd_cc_s_c       = $(CC) $(c_flags) -fverbose-asm -S -o $@ $<
 
 $(obj)/%.s: $(src)/%.c FORCE
 	$(call if_changed_dep,cc_s_c)
-- 
2.8.4

Re: [PATCH] x86: remove LTO flags

[Andi Kleen]

On Wed, Jul 06, 2016 at 05:22:50PM -0700, Luis R. Rodriguez wrote:
> The setup for LTO never made it upstream, and although this has
> some users, this is now really old stuff for a gcc 4.7 LTO problem.

Sorry, the LTO flags are still needed. Please don't remove

The 4.7 only workaround was only the LTO_REFERENCE_INITCALL macro,
which can be removed (if LTO is not support 4.7 anymore)

-Andi

[PATCH v2] x86: remove LTO_REFERENCE_INITCALL()

[Luis R. Rodriguez]

The setup for LTO never made it upstream, and although this has
some users, this is now really old stuff for a gcc 4.7 LTO problem.
We know that at least LTO_REFERENCE_INITCALL() work around can
be removed if LTO is not supported on v4.7 anymore.

As per Andi the DISABLE_LTO and LTO_CFLAGS are still neeeded though.

Tested-by: 0-day
Signed-off-by: Luis R. Rodriguez <mcgrof@kernel.org>
---
 include/linux/init.h | 20 +-------------------
 1 file changed, 1 insertion(+), 19 deletions(-)

diff --git a/include/linux/init.h b/include/linux/init.h
index 1e5c131d5c9a..aa662ad80d9c 100644
--- a/include/linux/init.h
+++ b/include/linux/init.h
@@ -151,23 +151,6 @@ extern bool initcall_debug;
 
 #ifndef __ASSEMBLY__
 
-#ifdef CONFIG_LTO
-/* Work around a LTO gcc problem: when there is no reference to a variable
- * in a module it will be moved to the end of the program. This causes
- * reordering of initcalls which the kernel does not like.
- * Add a dummy reference function to avoid this. The function is
- * deleted by the linker.
- */
-#define LTO_REFERENCE_INITCALL(x) \
-	; /* yes this is needed */			\
-	static __used __exit void *reference_##x(void)	\
-	{						\
-		return &x;				\
-	}
-#else
-#define LTO_REFERENCE_INITCALL(x)
-#endif
-
 /* initcalls are now grouped by functionality into separate 
  * subsections. Ordering inside the subsections is determined
  * by link order. 
@@ -180,8 +163,7 @@ extern bool initcall_debug;
 
 #define __define_initcall(fn, id) \
 	static initcall_t __initcall_##fn##id __used \
-	__attribute__((__section__(".initcall" #id ".init"))) = fn; \
-	LTO_REFERENCE_INITCALL(__initcall_##fn##id)
+	__attribute__((__section__(".initcall" #id ".init"))) = fn;
 
 /*
  * Early initcalls run before initializing SMP.
-- 
2.8.4

Re: [PATCH v2] x86: remove LTO_REFERENCE_INITCALL()

[Andi Kleen]

On Thu, Jul 07, 2016 at 09:01:49AM -0700, Luis R. Rodriguez wrote:
> The setup for LTO never made it upstream, and although this has
> some users, this is now really old stuff for a gcc 4.7 LTO problem.
> We know that at least LTO_REFERENCE_INITCALL() work around can
> be removed if LTO is not supported on v4.7 anymore.
> 
> As per Andi the DISABLE_LTO and LTO_CFLAGS are still neeeded though.

Looks good. Thanks

Acked-by: Andi Kleen <ak@linux.intel.com>

-Andi

Re: [PATCH v2] x86: remove LTO_REFERENCE_INITCALL()

[Luis R. Rodriguez]

On Thu, Jul 07, 2016 at 09:32:00AM -0700, Andi Kleen wrote:
> On Thu, Jul 07, 2016 at 09:01:49AM -0700, Luis R. Rodriguez wrote:
> > The setup for LTO never made it upstream, and although this has
> > some users, this is now really old stuff for a gcc 4.7 LTO problem.
> > We know that at least LTO_REFERENCE_INITCALL() work around can
> > be removed if LTO is not supported on v4.7 anymore.
> > 
> > As per Andi the DISABLE_LTO and LTO_CFLAGS are still neeeded though.
> 
> Looks good. Thanks
> 
> Acked-by: Andi Kleen <ak@linux.intel.com>

This is not yet merged but since it makes more sense in the
context of linker tables I'm going to fold this patch into
that series and post today.

  Luis