qemu-arm: Rework to run bare-metal on boards with Cortex-M CPUs.

Adds support for 3 Cortex-M boards, selectable via "BOARD" in the Makefile:
- microbit, Cortex-M0 via nRF51822
- netduino2, Cortex-M3 via STM32F205
- mps2-an385, Cortex-M3 via FPGA

netduino2 is the default board because it's supported by older qemu
versions (down to at least 2.5.0).

12 files changed, 320 insertions, 18 deletions

diff --git a/ports/qemu-arm/Makefile b/ports/qemu-arm/Makefile
index 95f349beba..b3d9cdf566 100644
--- a/ports/qemu-arm/Makefile
+++ b/ports/qemu-arm/Makefile
@@ -7,6 +7,27 @@ QSTR_DEFS = qstrdefsport.h
 # include py core make definitions
 include $(TOP)/py/py.mk
 
+BOARD ?= netduino2
+
+ifeq ($(BOARD),netduino2)
+CFLAGS += -mthumb -mcpu=cortex-m3 -mfloat-abi=soft
+CFLAGS += -DQEMU_SOC_STM32
+LDSCRIPT = stm32.ld
+endif
+
+ifeq ($(BOARD),microbit)
+CFLAGS += -mthumb -mcpu=cortex-m0 -mfloat-abi=soft
+CFLAGS += -DQEMU_SOC_NRF51
+LDSCRIPT = nrf51.ld
+QEMU_EXTRA = -global nrf51-soc.flash-size=1048576 -global nrf51-soc.sram-size=262144
+endif
+
+ifeq ($(BOARD),mps2-an385)
+CFLAGS += -mthumb -mcpu=cortex-m3 -mfloat-abi=soft
+CFLAGS += -DQEMU_SOC_MPS2
+LDSCRIPT = mps2.ld
+endif
+
 CROSS_COMPILE = arm-none-eabi-
 
 TINYTEST = $(TOP)/lib/tinytest
@@ -16,8 +37,7 @@ INC += -I$(TOP)
 INC += -I$(BUILD)
 INC += -I$(TINYTEST)
 
-CFLAGS_CORTEX_M3 = -mthumb -mcpu=cortex-m3 -mfloat-abi=soft
-CFLAGS = $(INC) -Wall -Wpointer-arith -Werror -std=gnu99 $(CFLAGS_CORTEX_M3) $(COPT) \
+CFLAGS += $(INC) -Wall -Wpointer-arith -Werror -std=gnu99 $(COPT) \
 	 -ffunction-sections -fdata-sections
 
 #Debugging/Optimization
@@ -34,9 +54,12 @@ endif
 ## else instead and according to the following files, this is what we need to pass to `$(CC).
 ## - gcc-arm-none-eabi-4_8-2014q1/share/gcc-arm-none-eabi/samples/src/makefile.conf
 ## - gcc-arm-none-eabi-4_8-2014q1/share/gcc-arm-none-eabi/samples/src/qemu/Makefile
-LDFLAGS= --specs=nano.specs --specs=rdimon.specs -Wl,--gc-sections -Wl,-Map=$(@:.elf=.map)
+LDFLAGS= -T $(LDSCRIPT) --gc-sections -Map=$(@:.elf=.map)
+LIBS = $(shell $(CC) $(CFLAGS) -print-libgcc-file-name)
 
 SRC_COMMON_C = \
+	startup.c \
+	uart.c \
 	moduos.c \
 	modmachine.c \
 
@@ -47,6 +70,7 @@ SRC_TEST_C = \
 	test_main.c \
 
 LIB_SRC_C += $(addprefix lib/,\
+	libc/string0.c \
 	libm/math.c \
 	libm/fmodf.c \
 	libm/nearbyintf.c \
@@ -89,11 +113,11 @@ SRC_QSTR += $(SRC_COMMON_C) $(SRC_RUN_C) $(LIB_SRC_C)
 all: run
 
 run: $(BUILD)/firmware.elf
-	qemu-system-arm -machine integratorcp -cpu cortex-m3 -nographic -monitor null -serial null -semihosting -kernel $(BUILD)/firmware.elf
+	qemu-system-arm -machine $(BOARD) $(QEMU_EXTRA) -nographic -monitor null -semihosting -kernel $<
 
 ## `$(LD)` doesn't seem to like `--specs` for some reason, but we can just use `$(CC)` here.
 $(BUILD)/firmware.elf: $(OBJ_COMMON) $(OBJ_RUN)
-	$(Q)$(CC) $(CFLAGS) $(LDFLAGS) -o $@ $^ $(LIBS)
+	$(Q)$(LD) $(LDFLAGS) -o $@ $^ $(LIBS)
 	$(Q)$(SIZE) $@
 
 include $(TOP)/py/mkrules.mk
diff --git a/ports/qemu-arm/Makefile.test b/ports/qemu-arm/Makefile.test
index a9aace6d07..347c2fefd3 100644
--- a/ports/qemu-arm/Makefile.test
+++ b/ports/qemu-arm/Makefile.test
@@ -15,10 +15,10 @@ $(BUILD)/tinytest.o:
 	$(Q)$(CC) $(CFLAGS) -DNO_FORKING -o $@ -c $(TINYTEST)/tinytest.c
 
 $(BUILD)/firmware-test.elf: $(OBJ_COMMON) $(OBJ_TEST)
-	$(Q)$(CC) $(CFLAGS) $(LDFLAGS) -o $@ $^ $(LIBS)
+	$(Q)$(LD) $(LDFLAGS) -o $@ $^ $(LIBS)
 	$(Q)$(SIZE) $@
 
 test: $(BUILD)/firmware-test.elf
-	qemu-system-arm -machine integratorcp -cpu cortex-m3 -nographic -monitor null -serial null -semihosting -kernel $(BUILD)/firmware-test.elf > $(BUILD)/console.out
+	qemu-system-arm -machine $(BOARD) $(QEMU_EXTRA) -nographic -monitor null -semihosting -kernel $< > $(BUILD)/console.out
 	$(Q)tail -n2 $(BUILD)/console.out
 	$(Q)tail -n1 $(BUILD)/console.out | grep -q "status: 0"
diff --git a/ports/qemu-arm/main.c b/ports/qemu-arm/main.c
index d23ef576f9..4cdd148287 100644
--- a/ports/qemu-arm/main.c
+++ b/ports/qemu-arm/main.c
@@ -30,7 +30,7 @@ void do_str(const char *src, mp_parse_input_kind_t input_kind) {
 int main(int argc, char **argv) {
     mp_stack_ctrl_init();
     mp_stack_set_limit(10240);
-    void *heap = malloc(16 * 1024);
+    uint32_t heap[16*1024 / 4];
     gc_init(heap, (char*)heap + 16 * 1024);
     mp_init();
     do_str("print('hello world!')", MP_PARSE_SINGLE_INPUT);
diff --git a/ports/qemu-arm/mpconfigport.h b/ports/qemu-arm/mpconfigport.h
index 5d86191988..3d4abd52ff 100644
--- a/ports/qemu-arm/mpconfigport.h
+++ b/ports/qemu-arm/mpconfigport.h
@@ -38,7 +38,7 @@
 #define MICROPY_PY_UHASHLIB         (1)
 #define MICROPY_PY_MACHINE          (1)
 #define MICROPY_PY_MICROPYTHON_MEM_INFO (1)
-#define MICROPY_USE_INTERNAL_PRINTF (0)
+#define MICROPY_USE_INTERNAL_PRINTF (1)
 #define MICROPY_VFS                 (1)
 
 // type definitions for the specific machine
@@ -54,9 +54,6 @@ typedef int32_t mp_int_t; // must be pointer size
 typedef uint32_t mp_uint_t; // must be pointer size
 typedef long mp_off_t;
 
-#include <unistd.h>
-#define MP_PLAT_PRINT_STRN(str, len) write(1, str, len)
-
 // extra built in names to add to the global namespace
 #define MICROPY_PORT_BUILTINS \
     { MP_ROM_QSTR(MP_QSTR_open), MP_ROM_PTR(&mp_builtin_open_obj) },
diff --git a/ports/qemu-arm/mphalport.h b/ports/qemu-arm/mphalport.h
index d996402ae4..29f080805f 100644
--- a/ports/qemu-arm/mphalport.h
+++ b/ports/qemu-arm/mphalport.h
@@ -1,2 +1,5 @@
+#include <stddef.h>
+#include "uart.h"
+
 #define mp_hal_stdin_rx_chr() (0)
-#define mp_hal_stdout_tx_strn_cooked(s, l) write(1, (s), (l))
+#define mp_hal_stdout_tx_strn_cooked(s, l) uart_tx_strn((s), (l))
diff --git a/ports/qemu-arm/mps2.ld b/ports/qemu-arm/mps2.ld
new file mode 100644
index 0000000000..5c1aa21ca2
--- /dev/null
+++ b/ports/qemu-arm/mps2.ld
@@ -0,0 +1,38 @@
+MEMORY
+{
+    RAM : ORIGIN = 0x00000000, LENGTH = 4M
+}
+
+_estack = ORIGIN(RAM) + LENGTH(RAM);
+
+SECTIONS
+{
+    .text : {
+        . = ALIGN(4);
+        KEEP(*(.isr_vector))
+        *(.text*)
+        *(.rodata*)
+        . = ALIGN(4);
+        _etext = .;
+        _sidata = _etext;
+    } > RAM
+
+    .data : AT ( _sidata )
+    {
+        . = ALIGN(4);
+        _sdata = .;
+        *(.data*)
+        . = ALIGN(4);
+        _edata = .;
+    } >RAM
+
+    .bss :
+    {
+        . = ALIGN(4);
+        _sbss = .;
+        *(.bss*)
+        *(COMMON)
+        . = ALIGN(4);
+        _ebss = .;
+    } >RAM
+}
diff --git a/ports/qemu-arm/nrf51.ld b/ports/qemu-arm/nrf51.ld
new file mode 100644
index 0000000000..70269773cb
--- /dev/null
+++ b/ports/qemu-arm/nrf51.ld
@@ -0,0 +1,39 @@
+MEMORY
+{
+    ROM : ORIGIN = 0x00000000, LENGTH = 1M
+    RAM : ORIGIN = 0x20000000, LENGTH = 256K
+}
+
+_estack = ORIGIN(RAM) + LENGTH(RAM);
+
+SECTIONS
+{
+    .text : {
+        . = ALIGN(4);
+        KEEP(*(.isr_vector))
+        *(.text*)
+        *(.rodata*)
+        . = ALIGN(4);
+        _etext = .;
+        _sidata = _etext;
+    } > ROM
+
+    .data : AT ( _sidata )
+    {
+        . = ALIGN(4);
+        _sdata = .;
+        *(.data*)
+        . = ALIGN(4);
+        _edata = .;
+    } >RAM
+
+    .bss :
+    {
+        . = ALIGN(4);
+        _sbss = .;
+        *(.bss*)
+        *(COMMON)
+        . = ALIGN(4);
+        _ebss = .;
+    } >RAM
+}
diff --git a/ports/qemu-arm/startup.c b/ports/qemu-arm/startup.c
new file mode 100644
index 0000000000..5198d089f9
--- /dev/null
+++ b/ports/qemu-arm/startup.c
@@ -0,0 +1,84 @@
+#include <stdint.h>
+#include <stdlib.h>
+
+#include "uart.h"
+
+extern uint32_t _estack, _sidata, _sdata, _edata, _sbss, _ebss;
+
+__attribute__((naked)) void Reset_Handler(void) {
+    // set stack pointer
+    __asm volatile ("ldr r0, =_estack");
+    __asm volatile ("mov sp, r0");
+    // copy .data section from flash to RAM
+    for (uint32_t *src = &_sidata, *dest = &_sdata; dest < &_edata;) {
+        *dest++ = *src++;
+    }
+    // zero out .bss section
+    for (uint32_t *dest = &_sbss; dest < &_ebss;) {
+        *dest++ = 0;
+    }
+    // jump to board initialisation
+    void _start(void);
+    _start();
+}
+
+void Default_Handler(void) {
+    for (;;) {
+    }
+}
+
+const uint32_t isr_vector[] __attribute__((section(".isr_vector"))) = {
+    (uint32_t)&_estack,
+    (uint32_t)&Reset_Handler,
+    (uint32_t)&Default_Handler, // NMI_Handler
+    (uint32_t)&Default_Handler, // HardFault_Handler
+    (uint32_t)&Default_Handler, // MemManage_Handler
+    (uint32_t)&Default_Handler, // BusFault_Handler
+    (uint32_t)&Default_Handler, // UsageFault_Handler
+    0,
+    0,
+    0,
+    0,
+    (uint32_t)&Default_Handler, // SVC_Handler
+    (uint32_t)&Default_Handler, // DebugMon_Handler
+    0,
+    (uint32_t)&Default_Handler, // PendSV_Handler
+    (uint32_t)&Default_Handler, // SysTick_Handler
+};
+
+void _start(void) {
+    // Enable the UART
+    uart_init();
+
+    // Now that we have a basic system up and running we can call main
+    extern int main();
+    main(0, 0);
+
+    // Finished
+    exit(0);
+}
+
+__attribute__((naked)) void exit(int status) {
+    // Force qemu to exit using ARM Semihosting
+    __asm volatile (
+        "mov r1, r0\n"
+        "cmp r1, #0\n"
+        "bne .notclean\n"
+        "ldr r1, =0x20026\n" // ADP_Stopped_ApplicationExit, a clean exit
+        ".notclean:\n"
+        "movs r0, #0x18\n" // SYS_EXIT
+        "bkpt 0xab\n"
+    );
+    for (;;) {
+    }
+}
+
+// The following are needed for tinytest
+
+#include <stdio.h>
+
+int setvbuf(FILE *stream, char *buf, int mode, size_t size) {
+    return 0;
+}
+
+struct _reent *_impure_ptr;
diff --git a/ports/qemu-arm/stm32.ld b/ports/qemu-arm/stm32.ld
new file mode 100644
index 0000000000..4e541526b7
--- /dev/null
+++ b/ports/qemu-arm/stm32.ld
@@ -0,0 +1,39 @@
+MEMORY
+{
+    ROM : ORIGIN = 0x00000000, LENGTH = 1M
+    RAM : ORIGIN = 0x20000000, LENGTH = 128K
+}
+
+_estack = ORIGIN(RAM) + LENGTH(RAM);
+
+SECTIONS
+{
+    .text : {
+        . = ALIGN(4);
+        KEEP(*(.isr_vector))
+        *(.text*)
+        *(.rodata*)
+        . = ALIGN(4);
+        _etext = .;
+        _sidata = _etext;
+    } > ROM
+
+    .data : AT ( _sidata )
+    {
+        . = ALIGN(4);
+        _sdata = .;
+        *(.data*)
+        . = ALIGN(4);
+        _edata = .;
+    } >RAM
+
+    .bss :
+    {
+        . = ALIGN(4);
+        _sbss = .;
+        *(.bss*)
+        *(COMMON)
+        . = ALIGN(4);
+        _ebss = .;
+    } >RAM
+}
diff --git a/ports/qemu-arm/test_main.c b/ports/qemu-arm/test_main.c
index adbdf04e18..2cb9e73859 100644
--- a/ports/qemu-arm/test_main.c
+++ b/ports/qemu-arm/test_main.c
@@ -15,15 +15,14 @@
 #include "tinytest.h"
 #include "tinytest_macros.h"
 
-#define HEAP_SIZE (128 * 1024)
-STATIC void *heap;
+#define HEAP_SIZE (100 * 1024)
 
 #include "genhdr/tests.h"
 
 int main() {
     mp_stack_ctrl_init();
     mp_stack_set_limit(10240);
-    heap = malloc(HEAP_SIZE);
+    static uint32_t heap[HEAP_SIZE / sizeof(uint32_t)];
     upytest_set_heap(heap, (char*)heap + HEAP_SIZE);
     int r = tinytest_main(0, NULL, groups);
     printf("status: %d\n", r);
@@ -34,8 +33,7 @@ void gc_collect(void) {
     gc_collect_start();
 
     // get the registers and the sp
-    jmp_buf env;
-    setjmp(env);
+    // TODO get registers
     volatile mp_uint_t dummy;
     void *sp = (void*)&dummy;
 
diff --git a/ports/qemu-arm/uart.c b/ports/qemu-arm/uart.c
new file mode 100644
index 0000000000..a0ce737c06
--- /dev/null
+++ b/ports/qemu-arm/uart.c
@@ -0,0 +1,78 @@
+#include <stdint.h>
+#include <stddef.h>
+
+#include "uart.h"
+
+#if defined(QEMU_SOC_STM32)
+
+typedef struct _UART_t {
+    volatile uint32_t SR;
+    volatile uint32_t DR;
+} UART_t;
+
+#define UART0 ((UART_t*)(0x40011000))
+
+void uart_init(void) {
+}
+
+void uart_tx_strn(const char *buf, size_t len) {
+    for (size_t i = 0; i < len; ++i) {
+        UART0->DR = buf[i];
+    }
+}
+
+#elif defined(QEMU_SOC_NRF51)
+
+typedef struct _UART_t {
+    volatile uint32_t r0[2];
+    volatile uint32_t STARTTX;  // 0x008
+    volatile uint32_t r1[(0x500 - 0x008) / 4 - 1];
+    volatile uint32_t ENABLE;   // 0x500
+    volatile uint32_t r2[(0x51c - 0x500) / 4 - 1];
+    volatile uint32_t TXD;      // 0x51c
+} UART_t;
+
+#define UART0 ((UART_t*)(0x40002000))
+
+void uart_init(void) {
+    UART0->ENABLE = 4;
+    UART0->STARTTX = 1;
+}
+
+void uart_tx_strn(const char *buf, size_t len) {
+    for (size_t i = 0; i < len; ++i) {
+        UART0->TXD = buf[i];
+    }
+}
+
+#elif defined(QEMU_SOC_MPS2)
+
+#define UART_STATE_TXFULL (1 << 0)
+
+#define UART_CTRL_TX_EN (1 << 0)
+#define UART_CTRL_RX_EN (1 << 1)
+
+typedef struct _UART_t {
+    volatile uint32_t DATA;
+    volatile uint32_t STATE;
+    volatile uint32_t CTRL;
+    volatile uint32_t INTSTATUS;
+    volatile uint32_t BAUDDIV;
+} UART_t;
+
+#define UART0 ((UART_t*)(0x40004000))
+
+void uart_init(void) {
+    UART0->BAUDDIV = 16;
+    UART0->CTRL = UART_CTRL_TX_EN;
+}
+
+void uart_tx_strn(const char *buf, size_t len) {
+    for (size_t i = 0; i < len; ++i) {
+        while (UART0->STATE & UART_STATE_TXFULL) {
+        }
+        UART0->DATA = buf[i];
+    }
+}
+
+#endif
diff --git a/ports/qemu-arm/uart.h b/ports/qemu-arm/uart.h
new file mode 100644
index 0000000000..a89e3f26e2
--- /dev/null
+++ b/ports/qemu-arm/uart.h
@@ -0,0 +1,2 @@
+void uart_init(void);
+void uart_tx_strn(const char *buf, size_t len);