Adding DMA transfer "Sleep" mode

This accounts for DMA stealing cycles from the CPU whenever the CPU
triggers a DMA (does not affect H/V blank or sound DMAs).
Works by moving the CPU to a PAUSED state where the cycles are accounted
for, reusing a similar mechanism for HALT/STOP.

Fixes a couple of games, notably GTA that has a DMA/IRQ race condition
(likely a bug really) if cycles are grossly miscalculated.

arm/arm64_stub.S

@@ -57,7 +57,7 @@ _##symbol:
 #define REG_Z_FLAG        (21 * 4)
 #define REG_C_FLAG        (22 * 4)
 #define REG_V_FLAG        (23 * 4)
-#define REG_UNUSED_1      (24 * 4)
+#define REG_SLEEP_CYCLES  (24 * 4)
 #define OAM_UPDATED       (25 * 4)
 #define REG_SAVE          (26 * 4)
 

arm/arm_stub.S

@@ -39,7 +39,7 @@ _##symbol:
 #define REG_Z_FLAG        (21 * 4)
 #define REG_C_FLAG        (22 * 4)
 #define REG_V_FLAG        (23 * 4)
-#define REG_UNUSED_1      (24 * 4)
+#define REG_SLEEP_CYCLES  (24 * 4)
 #define OAM_UPDATED       (25 * 4)
 
 #define CPU_ALERT_HALT    (1 << 0)

cpu.c

@@ -1489,7 +1489,12 @@ u32 check_and_raise_interrupts()
     reg[REG_PC] = 0x00000018;
 
     set_cpu_mode(MODE_IRQ);
+
+    // Wake up CPU if it is stopped/sleeping.
+    if (reg[CPU_HALT_STATE] == CPU_STOP ||
+        reg[CPU_HALT_STATE] == CPU_HALT)
       reg[CPU_HALT_STATE] = CPU_ACTIVE;
+
     return 1;
   }
   return 0;
@@ -3690,6 +3695,7 @@ void init_cpu(void)
     spsr[i] = 0x00000010;
 
   reg[CPU_HALT_STATE] = CPU_ACTIVE;
+  reg[REG_SLEEP_CYCLES] = 0;
 
   if (selected_boot_mode == boot_game) {
     reg[REG_SP] = 0x03007F00;

cpu.h

@@ -45,6 +45,7 @@ typedef u32 cpu_mode_type;
 #define CPU_ACTIVE          0
 #define CPU_HALT            1
 #define CPU_STOP            2
+#define CPU_DMA             3  /* CPU is idling due to DMA transfer */
 
 typedef u8 cpu_alert_type;
 
@@ -91,7 +92,7 @@ typedef enum
   REG_Z_FLAG        = 21,
   REG_C_FLAG        = 22,
   REG_V_FLAG        = 23,
-  REG_UNUSED_1      = 24,
+  REG_SLEEP_CYCLES  = 24,
   OAM_UPDATED       = 25,
   REG_SAVE          = 26,
   REG_SAVE2         = 27,

gba_memory.c

@@ -761,8 +761,17 @@ static cpu_alert_type trigger_dma(u32 dma_number, u32 value)
       }
 
       write_dmareg(REG_DMA0CNT_H, dma_number, value);
-      if(start_type == DMA_START_IMMEDIATELY)
-        return dma_transfer(dma_number, NULL);
+      if(start_type == DMA_START_IMMEDIATELY) {
+        // Excutes the DMA now! Copies the data and returns side effects.
+        int dma_cycles = 0;
+        cpu_alert_type ret = dma_transfer(dma_number, &dma_cycles);
+        if (!dma_cycles)
+          return ret;
+        // Sleep CPU for N cycles and return HALT as side effect (so it does).
+        reg[CPU_HALT_STATE] = CPU_DMA;
+        reg[REG_SLEEP_CYCLES] = 0x80000000 | (u32)dma_cycles;
+        return CPU_ALERT_HALT | ret;
+      }
     }
   }
   else

main.c

@@ -247,6 +247,22 @@ u32 function_cc update_gba(int remaining_cycles)
     // a video event or a timer event, whatever happens first.
     execute_cycles = MAX(video_count, 0);
 
+    // If we are paused due to a DMA, cap the number of cyles to that amount.
+    if (reg[CPU_HALT_STATE] == CPU_DMA) {
+      u32 dma_cyc = reg[REG_SLEEP_CYCLES];
+      // The first iteration is marked by bit 31 set, just do nothing now.
+      if (dma_cyc & 0x80000000)
+        dma_cyc &= 0x7FFFFFFF;  // Start counting DMA cycles from now on.
+      else
+        dma_cyc -= MIN(dma_cyc, completed_cycles);  // Account DMA cycles.
+
+      reg[REG_SLEEP_CYCLES] = dma_cyc;
+      if (!dma_cyc)
+        reg[CPU_HALT_STATE] = CPU_ACTIVE;   // DMA finished, resume execution.
+      else
+        execute_cycles = MIN(execute_cycles, dma_cyc);  // Continue sleeping.
+    }
+
     for (i = 0; i < 4; i++)
     {
        if (timer[i].status == TIMER_PRESCALE &&
@@ -294,7 +310,8 @@ bool main_check_savestate(const u8 *src)
 
   if (!bson_contains_key(p1, "cpu-ticks", BSON_TYPE_INT32) ||
       !bson_contains_key(p1, "exec-cycles", BSON_TYPE_INT32) ||
-      !bson_contains_key(p1, "video-count", BSON_TYPE_INT32))
+      !bson_contains_key(p1, "video-count", BSON_TYPE_INT32) ||
+      !bson_contains_key(p1, "sleep-cycles", BSON_TYPE_INT32))
     return false;
 
   for (i = 0; i < 4; i++)
@@ -327,7 +344,8 @@ bool main_read_savestate(const u8 *src)
 
   if (!(bson_read_int32(p1, "cpu-ticks", &cpu_ticks) &&
          bson_read_int32(p1, "exec-cycles", &execute_cycles) &&
-         bson_read_int32(p1, "video-count", (u32*)&video_count)))
+         bson_read_int32(p1, "video-count", (u32*)&video_count) &&
+         bson_read_int32(p1, "sleep-cycles", &reg[REG_SLEEP_CYCLES])))
     return false;
 
   for (i = 0; i < 4; i++)
@@ -357,6 +375,7 @@ unsigned main_write_savestate(u8* dst)
   bson_write_int32(dst, "cpu-ticks", cpu_ticks);
   bson_write_int32(dst, "exec-cycles", execute_cycles);
   bson_write_int32(dst, "video-count", video_count);
+  bson_write_int32(dst, "sleep-cycles", reg[REG_SLEEP_CYCLES]);
   bson_finish_document(dst, wbptr);
 
   bson_start_document(dst, "timers", wbptr);

mips/mips_stub.S

@@ -104,7 +104,7 @@ symbol:
 .equ REG_Z_FLAG,          (21 * 4)
 .equ REG_C_FLAG,          (22 * 4)
 .equ REG_V_FLAG,          (23 * 4)
-.equ REG_UNUSED_1,        (24 * 4)
+.equ REG_SLEEP_CYCLES,    (24 * 4)
 .equ OAM_UPDATED,         (25 * 4)
 .equ REG_SAVE,            (26 * 4)
 .equ REG_SAVE2,           (27 * 4)

savestate.h

@@ -92,7 +92,7 @@ bool bson_read_bytes(const u8 *srcp, const char *key, void* buffer, unsigned cnt
 /* this is an upper limit, leave room for future (?) stuff */
 #define GBA_STATE_MEM_SIZE                    (416*1024)
 #define GBA_STATE_MAGIC                       0x06BAC0DE
-#define GBA_STATE_VERSION                     0x00010001
+#define GBA_STATE_VERSION                     0x00010002
 
 bool gba_load_state(const void *src);
 void gba_save_state(void *dst);

x86/x86_stub.S

@@ -95,7 +95,7 @@ _##symbol:
 .equ REG_Z_FLAG,        (21 * 4)
 .equ REG_C_FLAG,        (22 * 4)
 .equ REG_V_FLAG,        (23 * 4)
-.equ REG_UNUSED_1,      (24 * 4)
+.equ REG_SLEEP_CYCLES,  (24 * 4)
 .equ OAM_UPDATED,       (25 * 4)
 .equ REG_SAVE,          (26 * 4)