Rebase branch 'profiling' from master

9 files changed, 43 insertions, 21 deletions

diff --git a/Makefile b/Makefile
index 12e7df8..e349ab1 100644
--- a/Makefile
+++ b/Makefile
@@ -117,6 +117,9 @@ CFLAGS += -I$(MBED_DIR)/targets/cmsis/TARGET_STM/TARGET_STM32F4
 CFLAGS += -I$(MBED_DIR)/targets/cmsis/TARGET_STM/TARGET_STM32F4/$(BOARD)
 CFLAGS += -I$(MBED_DIR)/targets/hal/TARGET_STM/TARGET_STM32F4
 CFLAGS += -I$(MBED_DIR)/targets/hal/TARGET_STM/TARGET_STM32F4/$(BOARD)
+ifdef DO_TASK_METRICS
+CFLAGS += -DDO_TASK_METRICS
+endif
 
 %.o : %.c
 	$(CC) $(CFLAGS) -c -o $@ $<
@@ -124,7 +127,11 @@ CFLAGS += -I$(MBED_DIR)/targets/hal/TARGET_STM/TARGET_STM32F4/$(BOARD)
 %.o : %.S
 	$(CC) $(CFLAGS) -c -o $@ $<
 
+ifdef DO_PROFILING
+all: hsm
+else
 all: board-test cli-test libhal-test hsm bootloader
+endif
 
 $(MBED_DIR)/libstmf4.a: .FORCE
 	$(MAKE) -C $(MBED_DIR)
@@ -152,9 +159,6 @@ libhal-test: $(BOARD_OBJS) $(LIBS) $(LIBHAL_BLD)/libhal.a .FORCE
 
 ifdef DO_PROFILING
 CFLAGS += -pg -DDO_PROFILING
-endif
-ifdef DO_TASK_METRICS
-CFLAGS += -DDO_TASK_METRICS
 hsm: $(BOARD_OBJS) $(LIBS) $(LIBHAL_BLD)/libhal.a $(LIBCLI_BLD)/libcli.a $(LIBPROF_BLD)/libprof.a .FORCE
 	$(MAKE) -C projects/hsm
 else
diff --git a/libraries/libprof/README.txt b/libraries/libprof/README.txt
index 1fe378c..f0b8ee8 100644
--- a/libraries/libprof/README.txt
+++ b/libraries/libprof/README.txt
@@ -52,7 +52,7 @@ In the OpenOCD console, enable semihosting:
 
 In another window, start the debugger:
 
-    $ sw/stm32/bin/debug projects/hsm/hsm
+    $ ../../bin/debug hsm
 
 In the CLI, type `profile start`, then start the unit test or whatever
 will be exercising the hsm. Afterwards, in the CLI, type `profile stop`.
diff --git a/libraries/libprof/profil.c b/libraries/libprof/profil.c
index 0654879..b0d8d55 100644
--- a/libraries/libprof/profil.c
+++ b/libraries/libprof/profil.c
@@ -25,27 +25,30 @@ static struct profinfo prof = {
   PROFILE_NOT_INIT, 0, 0, 0, 0
 };
 
-extern void set_SysTick_hook(void (*hook)(void));
-
 /* sample the current program counter */
-static void SysTick_hook(void) {
-  size_t pc = (size_t)((uint32_t *)__get_MSP())[8];
-  if (pc >= prof.lowpc && pc < prof.highpc) {
+void profil_callback(void) {
+  if (prof.state == PROFILE_ON) {
+    /* The interrupt mechanism pushes xPSR, PC, LR, R12, and R3-R0 onto the
+     * stack, so PC is the 6th word from the top at that point. However, the
+     * normal function entry code pushes registers as well, so the stack
+     * offset right now depends on the call tree that got us here.
+     */
+    size_t pc = (size_t)((uint32_t *)__get_MSP())[6 + 6];
+    if (pc >= prof.lowpc && pc < prof.highpc) {
       size_t idx = PROFIDX (pc, prof.lowpc, prof.scale);
       prof.counter[idx]++;
+    }
   }
 }
 
 /* Stop profiling to the profiling buffer pointed to by p. */
 static int profile_off (struct profinfo *p) {
-  set_SysTick_hook(NULL);
   p->state = PROFILE_OFF;
   return 0;
 }
 
 /* Create a timer thread and pass it a pointer P to the profiling buffer. */
 static int profile_on (struct profinfo *p) {
-  set_SysTick_hook(SysTick_hook);
   p->state = PROFILE_ON;
   return 0; /* ok */
 }
diff --git a/libraries/libtfm/Makefile b/libraries/libtfm/Makefile
index ceb8541..5be45f4 100644
--- a/libraries/libtfm/Makefile
+++ b/libraries/libtfm/Makefile
@@ -30,7 +30,12 @@ LIB  := tomsfastmath/libtfm.a
 # for now, at least until we confirm that it hasn't already been fixed
 # in a newer version of libtfm.
 
+ifdef DO_PROFILING
+# arm-none-eabi-gcc: error: -pg and -fomit-frame-pointer are incompatible
+STM32_LIBTFM_CFLAGS_OPTIMIZATION := -O3 -funroll-loops
+else
 STM32_LIBTFM_CFLAGS_OPTIMIZATION := -O3 -funroll-loops -fomit-frame-pointer
+endif
 
 CFLAGS := $(subst ${STM32_CFLAGS_OPTIMIZATION},${STM32_LIBTFM_CFLAGS_OPTIMIZATION},${CFLAGS})
 CFLAGS += -DTFM_ARM -Dasm=__asm__ -Wa,-mimplicit-it=thumb
diff --git a/projects/hsm/Makefile b/projects/hsm/Makefile
index 429069d..9a75b92 100644
--- a/projects/hsm/Makefile
+++ b/projects/hsm/Makefile
@@ -12,8 +12,6 @@ OBJS =	mgmt-cli.o \
 	log.o \
 	$(TOPLEVEL)/task.o
 
-CFLAGS += -DTASK_METRICS
-
 CFLAGS += -DNUM_RPC_TASK=4
 
 CFLAGS += -I$(LIBHAL_SRC)
@@ -32,6 +30,10 @@ LIBS += $(LIBPROF_BLD)/libprof.a
 LDFLAGS += --specs=rdimon.specs -lc -lrdimon
 endif
 
+ifdef DO_TASK_METRICS
+CFLAGS += -DDO_TASK_METRICS
+endif
+
 all: $(PROJ:=.elf)
 
 %.elf: %.o $(BOARD_OBJS) $(OBJS) $(LIBS)
diff --git a/projects/hsm/hsm.c b/projects/hsm/hsm.c
index b6b8820..5ef2ccc 100644
--- a/projects/hsm/hsm.c
+++ b/projects/hsm/hsm.c
@@ -268,6 +268,11 @@ size_t uart_rx_max = 0;
 
 void HAL_SYSTICK_Callback(void)
 {
+#ifdef DO_PROFILING
+    extern void profil_callback(void);
+    profil_callback();
+#endif
+
     size_t count = RINGBUF_COUNT(uart_ringbuf);
     if (uart_rx_max < count) uart_rx_max = count;
 
diff --git a/projects/hsm/mgmt-task.c b/projects/hsm/mgmt-task.c
index 9f6a908..4668585 100644
--- a/projects/hsm/mgmt-task.c
+++ b/projects/hsm/mgmt-task.c
@@ -73,7 +73,7 @@ static int cmd_task_show(struct cli_def *cli, const char *command, char *argv[],
     return CLI_OK;
 }
 
-#ifdef TASK_METRICS
+#ifdef DO_TASK_METRICS
 static int cmd_task_show_metrics(struct cli_def *cli, const char *command, char *argv[], int argc)
 {
     struct task_metrics tm;
@@ -99,9 +99,12 @@ void configure_cli_task(struct cli_def *cli)
     struct cli_command *c = cli_register_command(cli, NULL, "task", NULL, 0, 0, NULL);
 
     /* task show */
-    struct cli_command *c_show = cli_register_command(cli, c, "show", cmd_task_show, 0, 0, "Show the active tasks");
+#ifdef DO_TASK_METRICS
+    struct cli_command *c_show =
+#endif
+    cli_register_command(cli, c, "show", cmd_task_show, 0, 0, "Show the active tasks");
 
-#ifdef TASK_METRICS
+#ifdef DO_TASK_METRICS
     /* task show metrics */
     cli_register_command(cli, c_show, "metrics", cmd_task_show_metrics, 0, 0, "Show task metrics");
 
diff --git a/task.c b/task.c
index d8af217..36e8580 100644
--- a/task.c
+++ b/task.c
@@ -81,7 +81,7 @@ static tcb_t *cur_task = NULL;
 
 #define STACK_GUARD_WORD 0x55AA5A5A
 
-#ifdef TASK_METRICS
+#ifdef DO_TASK_METRICS
 static uint32_t tick_start = 0;
 static uint32_t tick_idle  = 0;
 static uint32_t tick_max   = 0;
@@ -192,7 +192,7 @@ void task_yield(void)
     if (tail == NULL)
 	return;
 
-#ifdef TASK_METRICS
+#ifdef DO_TASK_METRICS
     uint32_t tick0 = HAL_GetTick();
 #endif
 
@@ -213,7 +213,7 @@ void task_yield(void)
      * } while (next == NULL);
      */
 
-#ifdef TASK_METRICS
+#ifdef DO_TASK_METRICS
     uint32_t tick = HAL_GetTick();
     tick_idle += (tick - tick0);
     if (tick_start == 0)
@@ -395,7 +395,7 @@ void task_mutex_unlock(task_mutex_t *mutex)
 	mutex->locked = 0;
 }
 
-#ifdef TASK_METRICS
+#ifdef DO_TASK_METRICS
 void task_get_metrics(struct task_metrics *tm)
 {
     if (tm != NULL) {
diff --git a/task.h b/task.h
index 73ff33f..47d62d6 100644
--- a/task.h
+++ b/task.h
@@ -74,7 +74,7 @@ extern void task_delay(uint32_t delay);
 extern void task_mutex_lock(task_mutex_t *mutex);
 extern void task_mutex_unlock(task_mutex_t *mutex);
 
-#ifdef TASK_METRICS
+#ifdef DO_TASK_METRICS
 #include <sys/time.h>
 
 struct task_metrics {