From 6f617c41755a5d6aff3ef59fc4eae3fd53b261d9 Mon Sep 17 00:00:00 2001
From: Fiora <fioraaeterna@gmail.com>
Date: Fri, 29 Aug 2014 12:19:58 -0700
Subject: [PATCH] JIT64: try enabling dcbz again

This time, check the address carefully beforehand, since apparently some games
do horrible things like running it on non-RAM addresses, or at the very least
virtual addresses.
---
 Source/Core/Core/HW/Memmap.cpp                | 14 ++++++++
 Source/Core/Core/HW/Memmap.h                  |  1 +
 .../Interpreter/Interpreter_LoadStore.cpp     |  2 +-
 .../Core/Core/PowerPC/Jit64/Jit_LoadStore.cpp | 34 +++++++++++++++----
 4 files changed, 43 insertions(+), 8 deletions(-)

diff --git a/Source/Core/Core/HW/Memmap.cpp b/Source/Core/Core/HW/Memmap.cpp
index c3d3099851..f6a4dbe2c2 100644
--- a/Source/Core/Core/HW/Memmap.cpp
+++ b/Source/Core/Core/HW/Memmap.cpp
@@ -235,6 +235,20 @@ void Memset(const u32 _Address, const u8 _iValue, const u32 _iLength)
 	}
 }
 
+void ClearCacheLine(const u32 _Address)
+{
+	u8 *ptr = GetPointer(_Address);
+	if (ptr != nullptr)
+	{
+		memset(ptr, 0, 32);
+	}
+	else
+	{
+		for (u32 i = 0; i < 32; i += 8)
+			Write_U64(0, _Address + i);
+	}
+}
+
 void DMA_LCToMemory(const u32 _MemAddr, const u32 _CacheAddr, const u32 _iNumBlocks)
 {
 	const u8 *src = m_pL1Cache + (_CacheAddr & 0x3FFFF);
diff --git a/Source/Core/Core/HW/Memmap.h b/Source/Core/Core/HW/Memmap.h
index d1ae34711d..f130cf643f 100644
--- a/Source/Core/Core/HW/Memmap.h
+++ b/Source/Core/Core/HW/Memmap.h
@@ -126,6 +126,7 @@ u8* GetPointer(const u32 _Address);
 void DMA_LCToMemory(const u32 _iMemAddr, const u32 _iCacheAddr, const u32 _iNumBlocks);
 void DMA_MemoryToLC(const u32 _iCacheAddr, const u32 _iMemAddr, const u32 _iNumBlocks);
 void Memset(const u32 _Address, const u8 _Data, const u32 _iLength);
+void ClearCacheLine(const u32 _Address); // Zeroes 32 bytes; address should be 32-byte-aligned
 
 // TLB functions
 void SDRUpdated();
diff --git a/Source/Core/Core/PowerPC/Interpreter/Interpreter_LoadStore.cpp b/Source/Core/Core/PowerPC/Interpreter/Interpreter_LoadStore.cpp
index 199b3661f3..fc04cbfb9b 100644
--- a/Source/Core/Core/PowerPC/Interpreter/Interpreter_LoadStore.cpp
+++ b/Source/Core/Core/PowerPC/Interpreter/Interpreter_LoadStore.cpp
@@ -376,7 +376,7 @@ void Interpreter::dcbz(UGeckoInstruction _inst)
 {
 	// HACK but works... we think
 	if (!Core::g_CoreStartupParameter.bDCBZOFF)
-		Memory::Memset(Helper_Get_EA_X(_inst) & (~31), 0, 32);
+		Memory::ClearCacheLine(Helper_Get_EA_X(_inst) & (~31));
 	if (!JitInterface::GetCore())
 		PowerPC::CheckExceptions();
 }
diff --git a/Source/Core/Core/PowerPC/Jit64/Jit_LoadStore.cpp b/Source/Core/Core/PowerPC/Jit64/Jit_LoadStore.cpp
index 9c3e5bd876..a9e7cb94ab 100644
--- a/Source/Core/Core/PowerPC/Jit64/Jit_LoadStore.cpp
+++ b/Source/Core/Core/PowerPC/Jit64/Jit_LoadStore.cpp
@@ -260,17 +260,37 @@ void Jit64::dcbz(UGeckoInstruction inst)
 {
 	INSTRUCTION_START
 	JITDISABLE(bJITLoadStoreOff);
+	if (Core::g_CoreStartupParameter.bDCBZOFF)
+		return;
 
-	// FIXME
-	FALLBACK_IF(true);
+	int a = inst.RA;
+	int b = inst.RB;
 
-	MOV(32, R(EAX), gpr.R(inst.RB));
-	if (inst.RA)
-		ADD(32, R(EAX), gpr.R(inst.RA));
+	u32 mem_mask = Memory::ADDR_MASK_HW_ACCESS;
+	if (Core::g_CoreStartupParameter.bMMU || Core::g_CoreStartupParameter.bTLBHack)
+		mem_mask |= Memory::ADDR_MASK_MEM1;
+
+	MOV(32, R(EAX), gpr.R(b));
+	if (a)
+		ADD(32, R(EAX), gpr.R(a));
 	AND(32, R(EAX), Imm32(~31));
+	TEST(32, R(EAX), Imm32(mem_mask));
+	FixupBranch fast = J_CC(CC_Z, true);
+
+	// Should this code ever run? I can't find any games that use DCBZ on non-physical addresses, but
+	// supposedly there are, at least for some MMU titles. Let's be careful and support it to be sure.
+	MOV(32, M(&PC), Imm32(jit->js.compilerPC));
+	u32 registersInUse = CallerSavedRegistersInUse();
+	ABI_PushRegistersAndAdjustStack(registersInUse, false);
+	ABI_CallFunctionR((void *)&Memory::ClearCacheLine, EAX);
+	ABI_PopRegistersAndAdjustStack(registersInUse, false);
+
+	FixupBranch exit = J();
+	SetJumpTarget(fast);
 	PXOR(XMM0, R(XMM0));
-	MOVAPS(MComplex(EBX, EAX, SCALE_1, 0), XMM0);
-	MOVAPS(MComplex(EBX, EAX, SCALE_1, 16), XMM0);
+	MOVAPS(MComplex(RBX, RAX, SCALE_1, 0), XMM0);
+	MOVAPS(MComplex(RBX, RAX, SCALE_1, 16), XMM0);
+	SetJumpTarget(exit);
 }
 
 void Jit64::stX(UGeckoInstruction inst)