x86: Fix boot failure for GCC 4.9-built kernels

The current default cross toolchain for building x86 and x86_64 emulator
kernels is GCC 4.8, i.e.

 $AOSP/prebuilts/gcc/linux-x86/x86/x86_64-linux-android-4.8

on Linux hosts. It is also possible to build these kernels with GCC 4.9
($AOSP/prebuilts/gcc/linux-x86/x86/x86_64-linux-android-4.9), but
unfortunately, the generated kernel images are not bootable. This issue
affects both 3.4 and 3.10 kernels, and in the case of 3.10, both classic
(goldfish) and ranchu configurations.

The immediate cause of the boot failure is an invalid opcode exception
(escalating into a triple fault). The offending instruction is MOVDQA,
which belongs to the SSE2 instruction set. A postmortem on the vCPU
shows that the kernel is in an early boot stage, because paging (CR0.PG)
is not yet enabled. Nor is CR4.OSFXSR, which is in fact a prerequisite
for any SSE instructions, hence the exception.

The solution is to instruct GCC not to use SSE when compiling code for
x86 boot. This is done by back-porting the following patches from
upstream:

 5c630089 x86/kbuild: Eliminate duplicate command line options
 8b3b005d x86, build: Pass in additional -mno-mmx, -mno-sse options
 5551a34e x86-64, build: Always pass in -mno-sse

As to why GCC 4.9 behaves differently from GCC 4.8, this line from the
GCC 4.9 changelog (https://gcc.gnu.org/gcc-4.9/changes.html) offers a
possible explanation:

 * -mfpmath=sse is now implied by -ffast-math on all targets where SSE2
   is supported.

See https://android-review.googlesource.com/161639

 for the related
change in the 3.10 kernel.

Change-Id: I4bdf6cda44e3c576dce6510409daa8f3835a0e4f
Signed-off-by: Yu Ning <yu.ning@intel.com>
(cherry picked from commit 6e093637)

arch/x86/Makefile

@@ -128,7 +128,7 @@ KBUILD_CFLAGS += -Wno-sign-compare
 #
 KBUILD_CFLAGS += -fno-asynchronous-unwind-tables
 # prevent gcc from generating any FP code by mistake
-KBUILD_CFLAGS += $(call cc-option,-mno-sse -mno-mmx -mno-sse2 -mno-3dnow,)
+KBUILD_CFLAGS += -mno-sse -mno-mmx -mno-sse2 -mno-3dnow
 KBUILD_CFLAGS += $(call cc-option,-mno-avx,)
 
 ###

arch/x86/boot/Makefile

@@ -52,18 +52,18 @@ $(obj)/cpustr.h: $(obj)/mkcpustr FORCE
 
 # How to compile the 16-bit code.  Note we always compile for -march=i386,
 # that way we can complain to the user if the CPU is insufficient.
-KBUILD_CFLAGS	:= $(LINUXINCLUDE) -g -Os -D_SETUP -D__KERNEL__ \
+KBUILD_CFLAGS	:= $(LINUXINCLUDE) -m32 -g -Os -D_SETUP -D__KERNEL__ \
 		   -DDISABLE_BRANCH_PROFILING \
 		   -Wall -Wstrict-prototypes \
 		   -march=i386 -mregparm=3 \
 		   -include $(srctree)/$(src)/code16gcc.h \
 		   -fno-strict-aliasing -fomit-frame-pointer \
+		   -mno-mmx -mno-sse \
 		   $(call cc-option, -ffreestanding) \
 		   $(call cc-option, -fno-toplevel-reorder,\
 		   $(call cc-option, -fno-unit-at-a-time)) \
 		   $(call cc-option, -fno-stack-protector) \
 		   $(call cc-option, -mpreferred-stack-boundary=2)
-KBUILD_CFLAGS	+= $(call cc-option, -m32)
 KBUILD_AFLAGS	:= $(KBUILD_CFLAGS) -D__ASSEMBLY__
 GCOV_PROFILE := n
 

arch/x86/boot/compressed/Makefile

@@ -12,6 +12,7 @@ KBUILD_CFLAGS += -DDISABLE_BRANCH_PROFILING
 cflags-$(CONFIG_X86_32) := -march=i386
 cflags-$(CONFIG_X86_64) := -mcmodel=small
 KBUILD_CFLAGS += $(cflags-y)
+KBUILD_CFLAGS += -mno-mmx -mno-sse
 KBUILD_CFLAGS += $(call cc-option,-ffreestanding)
 KBUILD_CFLAGS += $(call cc-option,-fno-stack-protector)