Re-worked the complete loader code: Re-implemented the LiBounceOneChu…

…nk and LiWaitOneChunk functions and replaced the whole earlier loader functions by one LiWaitOneChunk function. (LiBounceOneChunk is just for...good to know ;-))

Comment:
This corrects the loading behavior. I don't think this will increase/decrease the compatibility (might be) as the loader behavior was almost correct before too but this reduces our loader code size by a lot, makes the loader code a lot more understandable and a lot easier to port to other system menu versions. Also since we don't need to fear GCC overwriting some registers anymore, the code optimizations can be used again which i re-enabled now too.

installer/launcher.c

@@ -630,9 +630,9 @@ static void InstallLoader(private_data_t *private_data)
     /* Get our functions */
     struct magic_t
     {
-        const void* func; // our replacement function which is called
-        const void* call; // address where to place the jump to our function
-        uint        orig_instr;
+        const void * repl_func;           // our replacement function which is called
+        const void * repl_addr;           // address where to place the jump to the our function
+        const unsigned int call_type;     // call type, e.g. 0x48000000 for branch and 0x48000001 for bl
     } *magic = (struct magic_t *)loader_magic;
     int magic_len = loader_magic_len / sizeof(struct magic_t);
 
@@ -641,16 +641,16 @@ static void InstallLoader(private_data_t *private_data)
     int i;
     for (i = 0; i < magic_len; i ++)
     {
-        int func_addr  = (int)magic[i].func;
-        int call_addr  = (int)magic[i].call;
-//        int orig_instr = (int)magic[i].orig_instr;
+        unsigned int repl_func  = (unsigned int)magic[i].repl_func;
+        unsigned int repl_addr  = (unsigned int)magic[i].repl_addr;
+        unsigned int call_type  = magic[i].call_type;
 
         // Install function hook only if needed
-        int jump_addr = func_addr - call_addr; // Compute jump length to jump from current instruction address to our function address
-        *((volatile uint32_t *)(0xC1000000 + call_addr)) = 0x48000001 | jump_addr; // Replace the instruction in the loader by the jump to our function
+        int jump_addr = (repl_func - repl_addr) & 0x03fffffc;                       // Compute jump length to jump from current instruction address to our function address
+        *((volatile uint32_t *)(0xC1000000 + repl_addr)) = call_type | jump_addr;   // Replace the instruction in the loader by the jump to our function
         // flush caches and invalidate instruction cache
-        DCFlushRange((void*)(0xC1000000 + call_addr), 4);
-        ICInvalidateRange((void*)(0xC1000000 + call_addr), 4);
+        DCFlushRange((void*)(0xC1000000 + repl_addr), 4);
+        ICInvalidateRange((void*)(0xC1000000 + repl_addr), 4);
     }
 
 }

loader/Makefile

@@ -31,7 +31,7 @@ SFLAGS := -mgekko -mregnames
 # so for now i reverted it as there is still enough space in payload
 # adding it back in will give another 600 - 1000 bytes of space
 # CFLAGS := -Os -nostdinc -nostdlib -Wall -x c -std=gnu99
-CFLAGS := -O0 -nostdinc -nostdlib -Wall -x c -std=gnu99 \
+CFLAGS := -Os -nostdinc -nostdlib -Wall -x c -std=gnu99 \
           -ffreestanding \
           -mrvl -mcpu=750 -meabi -mhard-float -fshort-wchar \
           -msdata=none -memb -ffunction-sections -fdata-sections  \
@@ -56,8 +56,8 @@ build/loader%.magic.bin: build/loader%.elf $(BUILD_REQ)
 	$(OBJCOPY) -j .magic -O binary $< $@
 
 build/loader%.elf: loader%.ld $(OBJ) $(BUILD_REQ)
-	$(LD) -T $< $(OBJ)
-#	$(LD) -T $< $(OBJ) -s -L"$(DEVKITPPC)/lib/gcc/powerpc-eabi/4.8.2" -lgcc
+#	$(LD) -T $< $(OBJ)
+	$(LD) -T $< $(OBJ) -s -L"$(DEVKITPPC)/lib/gcc/powerpc-eabi/4.8.2" -lgcc
 
 build/%.o: %.c $(BUILD_REQ)
 	$(CC) -c $(CFLAGS) -o $@ $<