cpu,gbdb: implement better breakpoints

Previously, gbdb had to check several signals, including the current
PC, to determine when do_run() should stop. This code was hotpath and
unecessarily slow.

Instead, leverage a n undefined instruction (0xd3) as a breakpoint instruction.
When the CPU emulation encounters this instruction, it will call a callback
which is implemented by gbdb. This can set a simple flag which is less expensive to
query.

Finally, both the signal handler and the breakpoint callback set specific
"paused" flags and a generic "pause" flag. Now, do_run() can simply check
the generic "paused" flag, then use the specific flags to determine the
stop reason.

This change increased performance by ~10% on Raspberry Pi.

src/apps/gbdb.c

@@ -25,8 +25,9 @@
 
 typedef void (gbdb_cmd)(char *arg_string);
 
-static bool cpu_at_breakpoint(void);
 static void breakpoint_addr_hit(uint16_t id);
+static void breakpoint_cb(struct lr35902_state *lr);
+static void reset_breakpoints(void);
 
 static uint8_t ram[MAX_RAM_LEN];
 static const char prompt[] = "gbdb >";
@@ -40,12 +41,16 @@ static const char usage[] =
 	"peek: view the next instruction to run\n"
 	"exit: quit the program\n";
 
+static struct lr35902_ops cpu_ops;
 static struct lr35902_state cpu;
 static struct gb_video video;
 
 static unsigned char bootrom[0x100] = {0};
 
 static int bootrom_mapped = 1;
+
+static int paused_breakpoint = 0;
+static volatile sig_atomic_t paused_signal = 0;
 static volatile sig_atomic_t paused = 0;
 
 static struct {
@@ -151,7 +156,11 @@ static void mem_dump(char *filename)
 
 static void init(void)
 {
-	lr35902_init(&cpu, mem_read, mem_write);
+	cpu_ops.mem_read = mem_read;
+	cpu_ops.mem_write = mem_write;
+	cpu_ops.undef_d3 = breakpoint_cb;
+
+	lr35902_init(&cpu, &cpu_ops);
 	gb_video_init(&video);
 	memset(&ram, 0, MAX_RAM_LEN);
 	bootrom_mapped = 1;
@@ -184,6 +193,17 @@ static void echo(char *arg_list)
 	}
 }
 
+static void breakpoint_cb(struct lr35902_state *lr)
+{
+  int i;
+  
+  paused_breakpoint = 1;
+  paused = 1;
+
+  breakpoint_addr_hit(cpu.pc);
+}
+
+
 static void step(char *arg_list)
 {
 	uint64_t steps, end_steps;
@@ -199,22 +219,27 @@ static void step(char *arg_list)
 	paused = 0;
 	signal(SIGINT, break_execution_handler);
 	end_steps = cpu.metrics.retired_instrs + steps;
-	do {
-		cycle();
-	} while (end_steps > cpu.metrics.retired_instrs &&
-		 !cpu_at_breakpoint() &&
-		 !paused);
 
+	cycle();
+	reset_breakpoints();
+
+	while(end_steps > cpu.metrics.retired_instrs &&
+	      !paused) {
+	  cycle();
+	}
+	
 	if (end_steps <= cpu.metrics.retired_instrs) {
 		gb_log("CPU stopped after %" PRId64 " instructions\n", steps);
-	} else if (cpu_at_breakpoint()) {
+	} else if (paused_breakpoint) {
 		breakpoint_addr_hit(cpu.pc);
 		gb_log("Breakpoint hit\n");
-	} else if (cpu.halted) {
-		gb_log("CPU halted\n");
-	} else {
+	} else if (paused_signal){
 		gb_log("Interrupted\n");
 	}
+
+	paused = 0;
+	paused_signal = 0;
+	paused_breakpoint = 0;
 }
 
 static void regs(char *arg_list)
@@ -447,18 +472,24 @@ static void quit(char *arg_list)
 static void do_run(void) {
 	paused = 0;
 	signal(SIGINT, break_execution_handler);
-	while(!cpu.halted && !cpu_at_breakpoint() && !paused) {
+
+	cycle();
+	reset_breakpoints();
+	
+	while(!paused) {
 		cycle();
 	}
-
-	if (cpu.halted) {
-		gb_log("CPU halted after %ld cycles\n", cpu.metrics.cycles);
-	} else if (paused) {
+	
+	if (paused_signal) {
 		gb_log("Interrupted.\n");
-	} else {
+	} else if (paused_breakpoint) {
 		breakpoint_addr_hit(cpu.pc);
 		gb_log("Breakpoint hit\n");
 	}
+
+	paused = 0;
+	paused_breakpoint = 0;
+	paused_signal = 0;
 }
 
 static void run(char *arg_list)
@@ -470,6 +501,8 @@ static struct breakpoint {
 	uint16_t addr;
 	int id;
 	bool temp;
+	uint8_t op;
+	uint8_t *ptr;
 } breakpoints[MAX_BREAKPTS];
 
 static int num_breakpoints = 0;
@@ -487,11 +520,29 @@ static struct breakpoint *get_breakpoint(int id)
 	return NULL;
 }
 
+static void reset_breakpoints(void)
+{
+  int i;
+
+  for (i = 0; i < num_breakpoints; i++) {
+    struct breakpoint *bkpt = &breakpoints[i];
+    *bkpt->ptr = 0xd3;
+  }
+}
+
 static int do_set_breakpoint(uint16_t addr, bool temp)
 {
 	static int id = 0;
+	
 	if (num_breakpoints < ARRAY_SIZE(breakpoints)) {
 		breakpoints[num_breakpoints].addr = addr;
+		breakpoints[num_breakpoints].op = mem_read(NULL, addr);
+		if (bootrom_mapped && addr < 0x100) {
+		  breakpoints[num_breakpoints].ptr = &bootrom[addr];
+		} else {
+		  breakpoints[num_breakpoints].ptr = &ram[addr];
+		}
+		*breakpoints[num_breakpoints].ptr = 0xd3;
 		breakpoints[num_breakpoints].id = id++;
 		breakpoints[num_breakpoints].temp = temp;
 		num_breakpoints++;
@@ -543,6 +594,8 @@ static int do_delete_breakpoint(int id)
 		return -1;
 	}
 
+	*bkpt->ptr = bkpt->op;
+	
 	index = bkpt - breakpoints;
 	memmove(&breakpoints[index], &breakpoints[index + 1],
 		num_breakpoints - index - 1);
@@ -659,11 +712,6 @@ static bool breakpoint_is_at_addr(uint16_t addr) {
 	return false;
 }
 
-static bool cpu_at_breakpoint(void)
-{
-	return breakpoint_is_at_addr(cpu.pc);
-}
-
 static void process_cmd(const char *cmd_str, struct tri *commands)
 {
 	char cmd_buffer[INPUT_MAX_LEN]; /* Buffer for command parsing */

src/gbemu/cpu.c

@@ -72,13 +72,13 @@ void lr35902_set_reg_8(struct lr35902_state *cpu, lr35902_regs_8 reg,
 }
 
 void lr35902_init(struct lr35902_state *cpu,
-                  lr35902_mem_read_fn mem_read,
-                  lr35902_mem_write_fn mem_write)
+                  const struct lr35902_ops *ops)
 {
 	int i;
 
-	cpu->mem_read = mem_read;
-	cpu->mem_write = mem_write;
+	cpu->mem_read = ops->mem_read;
+	cpu->mem_write = ops->mem_write;
+	cpu->undef_d3 = ops->undef_d3;
 
 	cpu->int_state = LR35902_INT_OFF;
 	cpu->stall_cycles = 0;
@@ -1353,7 +1353,16 @@ void lr35902_cycle(struct lr35902_state *cpu)
 		POP_16(cpu, cpu->de);
 		break;
 	case 0xd2:
+		break;
 	case 0xd3: /* UNDEF */
+		if (cpu->undef_d3) {
+			/* Undo the incremented PC so the op is called
+			   with the PC at the D3 instruction */
+		  	cpu->metrics.cycles--;
+			cpu->stall_cycles = 0;
+		  	cpu->pc--;
+			cpu->undef_d3(cpu);
+		}
 		break;
 	case 0xd4: /* CALL NC */
 		if (!cpu->nf) {

src/gbemu/cpu.h

@@ -94,8 +94,11 @@ struct lr35902_state {
 
 	int stall_cycles;
 	int halted;
+
 	lr35902_mem_read_fn mem_read;
 	lr35902_mem_write_fn mem_write;
+	void (*undef_d3)(struct lr35902_state *reg);
+
 	lr35902_interrupt_state int_state;
 	struct lr35902_event events[LR35902_MAX_EVENTS];
 
@@ -108,9 +111,15 @@ struct lr35902_state {
 	} metrics;
 };
 
+struct lr35902_ops {
+  lr35902_mem_read_fn mem_read;
+  lr35902_mem_write_fn mem_write;
+
+  void (*undef_d3)(struct lr35902_state *reg);
+};
+
 void lr35902_init(struct lr35902_state *cpu,
-                  lr35902_mem_read_fn mem_read,
-                  lr35902_mem_write_fn mem_write);
+		  const struct lr35902_ops *ops);
 
 uint16_t lr35902_get_reg_16(const struct lr35902_state *cpu,
                             lr35902_regs_16 reg);