Merge pull request RIOT-OS#5742 from kaspar030/fix_lto

Joakim Nohlgård · web-flow · commit 455fb6cd4d1b · 2016-08-29T19:00:09.000+02:00
LTO fixes
diff --git a/Makefile.base b/Makefile.base
@@ -23,7 +23,7 @@ ${DIRS:%=CLEAN--%}:
 	"$(MAKE)" -C ${@:CLEAN--%=%} clean
 
 ifeq ($(strip $(SRC)),)
-    SRC := $(wildcard *.c)
+    SRC := $(filter-out $(SRC_NOLTO), $(wildcard *.c))
 endif
 ifeq ($(strip $(SRCXX)),)
     SRCXX := $(wildcard *.cpp)
@@ -35,10 +35,12 @@ ifeq ($(strip $(ASSMSRC)),)
     ASSMSRC := $(wildcard *.S)
 endif
 
-OBJC    := $(SRC:%.c=$(BINDIR)$(MODULE)/%.o)
-OBJCXX  := $(SRCXX:%.cpp=$(BINDIR)$(MODULE)/%.o)
-ASMOBJ  := $(ASMSRC:%.s=$(BINDIR)$(MODULE)/%.o)
-ASSMOBJ := $(ASSMSRC:%.S=$(BINDIR)$(MODULE)/%.o)
+OBJC_LTO    := $(SRC:%.c=$(BINDIR)$(MODULE)/%.o)
+OBJC_NOLTO  := $(SRC_NOLTO:%.c=$(BINDIR)$(MODULE)/%.o)
+OBJC        := $(OBJC_NOLTO) $(OBJC_LTO)
+OBJCXX      := $(SRCXX:%.cpp=$(BINDIR)$(MODULE)/%.o)
+ASMOBJ      := $(ASMSRC:%.s=$(BINDIR)$(MODULE)/%.o)
+ASSMOBJ     := $(ASSMSRC:%.S=$(BINDIR)$(MODULE)/%.o)
 
 OBJ := $(OBJC) $(OBJCXX) $(ASMOBJ) $(ASSMOBJ)
 DEP := $(OBJC:.o=.d) $(OBJCXX:.o=.d) $(ASSMOBJ:.o=.d)
@@ -56,6 +58,8 @@ CXXFLAGS = $(filter-out $(CXXUWFLAGS), $(CFLAGS)) $(CXXEXFLAGS)
 
 # compile and generate dependency info
 
+$(OBJC_LTO): CFLAGS+=$(LTOFLAGS)
+
 $(OBJC): $(BINDIR)$(MODULE)/%.o: %.c $(RIOTBUILD_CONFIG_HEADER_C)
 	$(AD)$(CCACHE) $(CC) \
 		-DRIOT_FILE_RELATIVE=\"$(patsubst $(RIOTBASE)/%,%,$(abspath $<))\" \
diff --git a/Makefile.cflags b/Makefile.cflags
@@ -42,10 +42,9 @@ endif
 # Unwanted flags for c++
 CXXUWFLAGS += -std=%
 
-ifeq ($(LTO),yes)
-  $(info Building with Link-Time-Optimizations is currently an experimental feature. Expect broken binaries.)
-  LTOFLAGS = -flto -ffat-lto-objects
-  CFLAGS += ${LTOFLAGS}
+ifeq ($(LTO),1)
+  $(warning Building with Link-Time-Optimizations is currently an experimental feature. Expect broken binaries.)
+  LTOFLAGS = -flto
   LINKFLAGS += ${LTOFLAGS}
 endif
 
diff --git a/Makefile.vars b/Makefile.vars
@@ -35,6 +35,7 @@ export AS                    # The assembler.
 export ASFLAGS               # Flags for the assembler.
 export LINK                  # The command used to link the files. Must take the same parameters as GCC, i.e. "ld" won't work.
 export LINKFLAGS             # Flags to supply in the linking step.
+export LTOFLAGS              # extra CFLAGS for compiling with link time optimization
 export OBJCOPY               # The command used to create the HEXFILE.
 export OFLAGS                # The parameter for OBJCOPY, e.g. to strip the debug information.
 export OBJDUMP               # The command used to create the assembly listing.
diff --git a/boards/arduino-mega2560/Makefile.include b/boards/arduino-mega2560/Makefile.include
@@ -44,3 +44,10 @@ export ASFLAGS += $(CFLAGS_CPU) $(CFLAGS_DBG)
 export LINKFLAGS += $(CFLAGS_CPU) $(CFLAGS_DBG) $(CFLAGS_OPT) -Wl,--gc-sections -static -lgcc -e reset_handler
 export OFLAGS += -j .text -j .data -O ihex
 export FFLAGS += -p m2560 -c $(PROGRAMMER) $(PROGRAMMER_FLAGS) -F -D -U flash:w:bin/$(BOARD)/$(PROJECT)$(APPLICATION).hex
+
+ifeq ($(LTO),1)
+  # avr-gcc <4.8.3 has a bug when using LTO which causes a warning to be printed always:
+  # '_vector_25' appears to be a misspelled signal handler [enabled by default]
+  # See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59396
+  export LINKFLAGS += -Wno-error
+endif
diff --git a/boards/native/Makefile.include b/boards/native/Makefile.include
@@ -14,7 +14,11 @@ USEMODULE += native-drivers
 export PREFIX =
 export CC ?= $(PREFIX)gcc
 export CXX ?= $(PREFIX)g++
-export AR ?= $(PREFIX)ar
+ifeq ($(LTO),1)
+export AR = $(PREFIX)gcc-ar
+else
+export AR = $(PREFIX)ar
+endif
 export AS ?= $(PREFIX)as
 export LINK ?= $(PREFIX)gcc
 export SIZE ?= $(PREFIX)size
@@ -41,7 +45,6 @@ export GPROF ?= gprof
 
 # basic cflags:
 export CFLAGS += -Wall -Wextra -pedantic
-export CFLAGS += -ffunction-sections -fdata-sections
 ifeq ($(shell uname -m),x86_64)
 export CFLAGS += -m32
 endif
@@ -73,6 +76,10 @@ else
 export LINKFLAGS += -ldl
 endif
 
+# clean up unused functions
+export CFLAGS += -ffunction-sections -fdata-sections
+export LINKFLAGS += -Wl,--gc-sections
+
 # set the tap interface for term/valgrind
 ifneq (,$(filter netdev2_tap,$(USEMODULE)))
 	export PORT ?= tap0
diff --git a/boards/x86-multiboot-common/Makefile.include b/boards/x86-multiboot-common/Makefile.include
@@ -32,7 +32,11 @@ export CPU = x86
 
 # toolchain config
 export CC ?= $(PREFIX)gcc
-export AR ?= $(PREFIX)ar
+ifeq ($(LTO),1)
+export AR = $(PREFIX)gcc-ar
+else
+export AR = $(PREFIX)ar
+endif
 export AS ?= $(PREFIX)as
 export RANLIB ?= $(PREFIX)ranlib
 export LINK ?= $(RIOTBASE)/boards/x86-multiboot-common/dist/link $(PREFIX)gcc
@@ -46,6 +50,11 @@ LINKFLAGS += -m32 -nostdlib -nostdinc -nostartfiles -nodefaultlibs \
              --prefix=$(NEWLIB_BASE) \
              -Wl,-rpath,$(NEWLIB_BASE)/lib \
              -T$(RIOTBASE)/boards/x86-multiboot-common/linker.ld
+
+# clean up unused functions
+export CFLAGS += -ffunction-sections -fdata-sections
+export LINKFLAGS += -Wl,--gc-sections
+
 UNDEF += $(BINDIR)x86-multiboot-common/startup.o
 
 BASELIBS += $(NEWLIB_BASE)/lib/libc.a \
diff --git a/core/sched.c b/core/sched.c
@@ -58,7 +58,7 @@ static void (*sched_cb) (uint32_t timestamp, uint32_t value) = NULL;
 schedstat sched_pidlist[KERNEL_PID_LAST + 1];
 #endif
 
-int sched_run(void)
+int __attribute__((used)) sched_run(void)
 {
     sched_context_switch_request = 0;
 
diff --git a/cpu/Makefile.include.gnu b/cpu/Makefile.include.gnu
@@ -1,7 +1,11 @@
 export GDBPREFIX ?= $(PREFIX)
 export CC = $(PREFIX)gcc
 export CXX = $(PREFIX)g++
+ifeq ($(LTO),1)
+export AR = $(PREFIX)gcc-ar
+else
 export AR = $(PREFIX)ar
+endif
 export AS = $(PREFIX)as
 export LINK = $(PREFIX)gcc
 export SIZE = $(PREFIX)size
diff --git a/cpu/atmega2560/startup.c b/cpu/atmega2560/startup.c
@@ -43,27 +43,28 @@ extern void __libc_init_array(void);
  * which should never be reached but just in case jumps to exit.
  * This way there should be no way to call main directly.
  */
-void init7_ovr(void) __attribute__((naked)) __attribute__((section(".init7")));
-void init8_ovr(void) __attribute__((naked)) __attribute__((section(".init8")));
+void init7_ovr(void) __attribute__((section(".init7")));
+void init8_ovr(void) __attribute__((section(".init8")));
 
 
-void init7_ovr(void)
+__attribute__((used,naked)) void init7_ovr(void)
 {
     __asm__("call reset_handler");
 }
 
-void init8_ovr(void)
+__attribute__((used,naked)) void init8_ovr(void)
 {
     __asm__("jmp exit");
 }
+
 /**
  * @brief This function is the entry point after a system reset
  *
  * After a system reset, the following steps are necessary and carried out:
  * 1. initialize the board (sync clock, setup std-IO)
  * 2. initialize and start RIOTs kernel
  */
-void reset_handler(void)
+__attribute__((used)) void reset_handler(void)
 {
     /* initialize the board and startup the kernel */
     board_init();
diff --git a/cpu/cc2538/Makefile b/cpu/cc2538/Makefile
@@ -9,4 +9,7 @@ ifneq (,$(filter cc2538_rf,$(USEMODULE)))
     DIRS += radio
 endif
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/cc26x0/Makefile b/cpu/cc26x0/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # Add a list of subdirectories, that should also be built:
 DIRS = periph $(RIOTCPU)/cortexm_common
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/cortexm_common/Makefile b/cpu/cortexm_common/Makefile
@@ -1 +1,5 @@
+# thread_arch.c's inline assembler breaks when compiling with link time
+# optimization. see #5774.
+SRC_NOLTO += thread_arch.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/cortexm_common/thread_arch.c b/cpu/cortexm_common/thread_arch.c
@@ -296,17 +296,12 @@ void thread_arch_yield(void)
     SCB->ICSR |= SCB_ICSR_PENDSVSET_Msk;
 }
 
-__attribute__((naked)) void arch_context_switch(void)
-{
+void __attribute__((naked)) __attribute__((used)) isr_pendsv(void) {
     __asm__ volatile (
     /* PendSV handler entry point */
-    ".global isr_pendsv               \n"
-    ".thumb_func                      \n"
-    "isr_pendsv:                      \n"
     /* save context by pushing unsaved registers to the stack */
     /* {r0-r3,r12,LR,PC,xPSR} are saved automatically on exception entry */
     ".thumb_func                      \n"
-    "context_save:"
     "mrs    r0, psp                   \n" /* get stack pointer from user mode */
 #if defined(CPU_ARCH_CORTEX_M0) || defined(CPU_ARCH_CORTEX_M0PLUS)
     "mov    r12, sp                   \n" /* remember the exception SP */
@@ -332,11 +327,15 @@ __attribute__((naked)) void arch_context_switch(void)
     "ldr    r1, =sched_active_thread  \n" /* load address of current tcb */
     "ldr    r1, [r1]                  \n" /* dereference pdc */
     "str    r0, [r1]                  \n" /* write r0 to pdc->sp */
+    "bl     isr_svc                   \n" /* continue with svc */
+    );
+}
+
+void __attribute__((naked)) __attribute__((used)) isr_svc(void) {
+    __asm__ volatile (
     /* SVC handler entry point */
-    /* PendSV will continue from above and through this part as well */
-    ".global isr_svc                  \n"
+    /* PendSV will continue here as well (via jump) */
     ".thumb_func                      \n"
-    "isr_svc:                         \n"
     /* perform scheduling */
     "bl     sched_run                 \n"
     /* restore context and return from exception */
diff --git a/cpu/ezr32wg/Makefile b/cpu/ezr32wg/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/k60/Makefile b/cpu/k60/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common $(KINETIS_COMMON)
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/k60/Makefile.include b/cpu/k60/Makefile.include
@@ -19,7 +19,7 @@ export CFLAGS += -DCPU_ARCH_$(ARCH)
 export COMMON_STARTUP = $(KINETIS_COMMON)
 
 # add the CPU specific system calls implementations for the linker
-export UNDEF += $(BINDIR)cpu/vector.o
+export UNDEF += $(BINDIR)cpu/vectors.o
 export UNDEF += $(BINDIR)cpu/ssp.o
 
 include $(RIOTCPU)/Makefile.include.cortexm_common
diff --git a/cpu/k60/vectors.c b/cpu/k60/vectors.c
diff --git a/cpu/k64f/Makefile b/cpu/k64f/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common $(KINETIS_COMMON)
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/kw2x/Makefile b/cpu/kw2x/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common $(KINETIS_COMMON)
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/kw2x/Makefile.include b/cpu/kw2x/Makefile.include
@@ -19,6 +19,6 @@ export CFLAGS += -DCPU_ARCH_$(ARCH)
 export COMMON_STARTUP = $(KINETIS_COMMON)
 
 # add the CPU specific system calls implementations for the linker
-export UNDEF += $(BINDIR)cpu/vector.o
+export UNDEF += $(BINDIR)cpu/vectors.o
 
 include $(RIOTCPU)/Makefile.include.cortexm_common
diff --git a/cpu/kw2x/vectors.c b/cpu/kw2x/vectors.c
diff --git a/cpu/lm4f120/Makefile b/cpu/lm4f120/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/lpc11u34/Makefile b/cpu/lpc11u34/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/lpc1768/Makefile b/cpu/lpc1768/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/lpc1768/vectors.c b/cpu/lpc1768/vectors.c
@@ -70,7 +70,7 @@ WEAK_DEFAULT void isr_qei(void);
 WEAK_DEFAULT void isr_pll1(void);
 
 /* interrupt vector table */
-__attribute__ ((section(".vectors")))
+__attribute__ ((used,section(".vectors")))
 const void *interrupt_vector[] = {
     /* Exception stack pointer */
     (void*) (&_estack),             /* pointer to the top of the stack */
diff --git a/cpu/nrf51/Makefile b/cpu/nrf51/Makefile
@@ -9,4 +9,7 @@ ifneq (,$(filter radio_nrfmin,$(USEMODULE)))
     DIRS += radio/nrfmin
 endif
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/nrf52/Makefile b/cpu/nrf52/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common $(RIOTCPU)/nrf5x_common
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/sam3/Makefile b/cpu/sam3/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/samd21/Makefile b/cpu/samd21/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common $(RIOTCPU)/sam21_common
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/saml21/Makefile b/cpu/saml21/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common $(RIOTCPU)/sam21_common
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/stm32f0/Makefile b/cpu/stm32f0/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common $(RIOTCPU)/stm32_common
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/stm32f1/Makefile b/cpu/stm32f1/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS += periph $(RIOTCPU)/cortexm_common $(RIOTCPU)/stm32_common
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/stm32f3/Makefile b/cpu/stm32f3/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common $(RIOTCPU)/stm32_common
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/stm32f4/Makefile b/cpu/stm32f4/Makefile
@@ -4,4 +4,7 @@ MODULE = cpu
 # add a list of subdirectories, that should also be build
 DIRS = periph $(RIOTCPU)/cortexm_common $(RIOTCPU)/stm32_common
 
+# (file triggers compiler bug. see #5775)
+SRC_NOLTO += vectors.c
+
 include $(RIOTBASE)/Makefile.base
diff --git a/cpu/stm32l1/Makefile b/cpu/stm32l1/Makefile
diff --git a/cpu/x86/x86_interrupts.c b/cpu/x86/x86_interrupts.c
diff --git a/sys/newlib/syscalls.c b/sys/newlib/syscalls.c

Original file line number	Diff line number	Diff line change
`@@ -58,7 +58,7 @@ static void (*sched_cb) (uint32_t timestamp, uint32_t value) = NULL;`
`58`	`58`	`schedstat sched_pidlist[KERNEL_PID_LAST + 1];`
`59`	`59`	`#endif`
`60`	`60`
`61`		`-int sched_run(void)`
	`61`	`+int __attribute__((used)) sched_run(void)`
`62`	`62`	`{`
`63`	`63`	`sched_context_switch_request = 0;`
`64`	`64`
Original file line number	Diff line number	Diff line change
`@@ -43,27 +43,28 @@ extern void __libc_init_array(void);`
`43`	`43`	`* which should never be reached but just in case jumps to exit.`
`44`	`44`	`* This way there should be no way to call main directly.`
`45`	`45`	`*/`
`46`		`-void init7_ovr(void) __attribute__((naked)) __attribute__((section(".init7")));`
`47`		`-void init8_ovr(void) __attribute__((naked)) __attribute__((section(".init8")));`
	`46`	`+void init7_ovr(void) __attribute__((section(".init7")));`
	`47`	`+void init8_ovr(void) __attribute__((section(".init8")));`
`48`	`48`
`49`	`49`
`50`		`-void init7_ovr(void)`
	`50`	`+__attribute__((used,naked)) void init7_ovr(void)`
`51`	`51`	`{`
`52`	`52`	`__asm__("call reset_handler");`
`53`	`53`	`}`
`54`	`54`
`55`		`-void init8_ovr(void)`
	`55`	`+__attribute__((used,naked)) void init8_ovr(void)`
`56`	`56`	`{`
`57`	`57`	`__asm__("jmp exit");`
`58`	`58`	`}`
	`59`	`+`
`59`	`60`	`/**`
`60`	`61`	`* @brief This function is the entry point after a system reset`
`61`	`62`	`*`
`62`	`63`	`* After a system reset, the following steps are necessary and carried out:`
`63`	`64`	`* 1. initialize the board (sync clock, setup std-IO)`
`64`	`65`	`* 2. initialize and start RIOTs kernel`
`65`	`66`	`*/`
`66`		`-void reset_handler(void)`
	`67`	`+__attribute__((used)) void reset_handler(void)`
`67`	`68`	`{`
`68`	`69`	`/* initialize the board and startup the kernel */`
`69`	`70`	`board_init();`