diff --git a/Core/HLE/HLEHelperThread.cpp b/Core/HLE/HLEHelperThread.cpp
index 5c8a0278fb..0975d1ea96 100644
--- a/Core/HLE/HLEHelperThread.cpp
+++ b/Core/HLE/HLEHelperThread.cpp
@@ -29,7 +29,7 @@
 HLEHelperThread::HLEHelperThread() : id_(-1), entry_(0) {
 }
 
-HLEHelperThread::HLEHelperThread(const char *threadName, u32 instructions[], u32 instrCount, u32 prio, int stacksize) {
+HLEHelperThread::HLEHelperThread(const char *threadName, const u32 instructions[], u32 instrCount, u32 prio, int stacksize) {
 	u32 instrBytes = instrCount * sizeof(u32);
 	u32 totalBytes = instrBytes + sizeof(u32) * 2;
 	AllocEntry(totalBytes);
diff --git a/Core/HLE/HLEHelperThread.h b/Core/HLE/HLEHelperThread.h
index 39df4c7402..d932c2ea23 100644
--- a/Core/HLE/HLEHelperThread.h
+++ b/Core/HLE/HLEHelperThread.h
@@ -25,7 +25,7 @@ class HLEHelperThread {
 public:
 	// For savestates.
 	HLEHelperThread();
-	HLEHelperThread(const char *threadName, u32 instructions[], u32 instrCount, u32 prio, int stacksize);
+	HLEHelperThread(const char *threadName, const u32 instructions[], u32 instrCount, u32 prio, int stacksize);
 	HLEHelperThread(const char *threadName, const char *module, const char *func, u32 prio, int stacksize);
 	~HLEHelperThread();
 	void DoState(PointerWrap &p);
diff --git a/Core/HLE/sceKernelThread.cpp b/Core/HLE/sceKernelThread.cpp
index 6540147346..89e58a93e1 100644
--- a/Core/HLE/sceKernelThread.cpp
+++ b/Core/HLE/sceKernelThread.cpp
@@ -2025,22 +2025,24 @@ int __KernelStartThread(SceUID threadToStartID, int argSize, u32 argBlockPtr, bo
 	u32 &sp = startThread->context.r[MIPS_REG_SP];
 	// Force args means just use those as a0/a1 without any special treatment.
 	// This is a hack to avoid allocating memory for helper threads which take args.
-	if ((argBlockPtr && argSize > 0) || forceArgs) {
+	if (forceArgs) {
+		startThread->context.r[MIPS_REG_A0] = argSize;
+		startThread->context.r[MIPS_REG_A1] = argBlockPtr;
+	} else if (argBlockPtr && argSize > 0) {
 		// Make room for the arguments, always 0x10 aligned.
-		if (!forceArgs)
-			sp -= (argSize + 0xf) & ~0xf;
+		sp -= (argSize + 0xf) & ~0xf;
 		startThread->context.r[MIPS_REG_A0] = argSize;
 		startThread->context.r[MIPS_REG_A1] = sp;
+
+		// Now copy argument to stack.
+		if (Memory::IsValidAddress(argBlockPtr)) {
+			Memory::Memcpy(sp, argBlockPtr, argSize, "ThreadStartArgs");
+		}
 	} else {
 		startThread->context.r[MIPS_REG_A0] = 0;
 		startThread->context.r[MIPS_REG_A1] = 0;
 	}
 
-	// Now copy argument to stack.
-	if (!forceArgs && Memory::IsValidAddress(argBlockPtr)) {
-		Memory::Memcpy(sp, argBlockPtr, argSize, "ThreadStartArgs");
-	}
-
 	// On the PSP, there's an extra 64 bytes of stack eaten after the args.
 	// This could be stack overflow safety, or just stack eaten by the kernel entry func.
 	sp -= 64;