x86/build: Mostly disable '-maccumulate-outgoing-args'

The GCC '-maccumulate-outgoing-args' flag is enabled for most configs,
mostly because of issues which are no longer relevant.  For most
configs, and with most recent versions of GCC, it's no longer needed.

Clarify which cases need it, and only enable it for those cases.  Also
produce a compile-time error for the ftrace graph + mcount + '-Os' case,
which will otherwise cause runtime failures.

The main benefit of '-maccumulate-outgoing-args' is that it prevents an
ugly prologue for functions which have aligned stacks.  But removing the
option also has some benefits: more readable argument saves, smaller
text size, and (presumably) slightly improved performance.

Here are the object size savings for 32-bit and 64-bit defconfig
kernels:

      text	   data	    bss	     dec	    hex	filename
  10006710	3543328	1773568	15323606	 e9d1d6	vmlinux.x86-32.before
   9706358	3547424	1773568	15027350	 e54c96	vmlinux.x86-32.after

      text	   data	    bss	     dec	    hex	filename
  10652105	4537576	 843776	16033457	 f4a6b1	vmlinux.x86-64.before
  10639629	4537576	 843776	16020981	 f475f5	vmlinux.x86-64.after

That comes out to a 3% text size improvement on x86-32 and a 0.1% text
size improvement on x86-64.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Lutomirski <luto@kernel.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20170316193133.zrj6gug53766m6nn@treble
Signed-off-by: Ingo Molnar <mingo@kernel.org>

1 files changed, 31 insertions, 4 deletions

diff --git a/arch/x86/Makefile b/arch/x86/Makefile
index 2d449337a3605..a94a4d10f2dfa 100644
--- a/arch/x86/Makefile
+++ b/arch/x86/Makefile
@@ -120,10 +120,6 @@ else
         # -funit-at-a-time shrinks the kernel .text considerably
         # unfortunately it makes reading oopses harder.
         KBUILD_CFLAGS += $(call cc-option,-funit-at-a-time)
-
-        # this works around some issues with generating unwind tables in older gccs
-        # newer gccs do it by default
-        KBUILD_CFLAGS += $(call cc-option,-maccumulate-outgoing-args)
 endif
 
 ifdef CONFIG_X86_X32
@@ -147,6 +143,37 @@ ifeq ($(CONFIG_KMEMCHECK),y)
 	KBUILD_CFLAGS += $(call cc-option,-fno-builtin-memcpy)
 endif
 
+#
+# If the function graph tracer is used with mcount instead of fentry,
+# '-maccumulate-outgoing-args' is needed to prevent a GCC bug
+# (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=42109)
+#
+ifdef CONFIG_FUNCTION_GRAPH_TRACER
+  ifndef CONFIG_HAVE_FENTRY
+	ACCUMULATE_OUTGOING_ARGS := 1
+  else
+    ifeq ($(call cc-option-yn, -mfentry), n)
+	ACCUMULATE_OUTGOING_ARGS := 1
+    endif
+  endif
+endif
+
+#
+# Jump labels need '-maccumulate-outgoing-args' for gcc < 4.5.2 to prevent a
+# GCC bug (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=46226).  There's no way
+# to test for this bug at compile-time because the test case needs to execute,
+# which is a no-go for cross compilers.  So check the GCC version instead.
+#
+ifdef CONFIG_JUMP_LABEL
+  ifneq ($(ACCUMULATE_OUTGOING_ARGS), 1)
+	ACCUMULATE_OUTGOING_ARGS = $(call cc-if-fullversion, -lt, 040502, 1)
+  endif
+endif
+
+ifeq ($(ACCUMULATE_OUTGOING_ARGS), 1)
+	KBUILD_CFLAGS += -maccumulate-outgoing-args
+endif
+
 # Stackpointer is addressed different for 32 bit and 64 bit x86
 sp-$(CONFIG_X86_32) := esp
 sp-$(CONFIG_X86_64) := rsp