JitArm64: Use farcode in WriteLinkBlock

Now block link nearcode is back to a length of three instructions.

Unfortunately, the code I'm adding to Jit.cpp ends up being a bit messy
because we need to handle the case of already being in farcode...

Source/Core/Core/PowerPC/JitArm64/Jit.cpp

@@ -372,12 +372,29 @@ void JitArm64::WriteExit(u32 destination, bool LK, u32 exit_address_after_return
     STP(IndexType::Pre, ARM64Reg::X0, ARM64Reg::X1, ARM64Reg::SP, -16);
   }
 
+  constexpr size_t primary_farcode_size = 3 * sizeof(u32);
+  const bool switch_to_far_code = !IsInFarCode();
+  const u8* primary_farcode_addr;
+  if (switch_to_far_code)
+  {
+    SwitchToFarCode();
+    primary_farcode_addr = GetCodePtr();
+    SwitchToNearCode();
+  }
+  else
+  {
+    primary_farcode_addr = GetCodePtr() + JitArm64BlockCache::BLOCK_LINK_SIZE +
+                           (LK ? JitArm64BlockCache::BLOCK_LINK_SIZE : 0);
+  }
+  const u8* return_farcode_addr = primary_farcode_addr + primary_farcode_size;
+
   JitBlock* b = js.curBlock;
   JitBlock::LinkData linkData;
   linkData.exitAddress = destination;
   linkData.exitPtrs = GetWritableCodePtr();
   linkData.linkStatus = false;
   linkData.call = LK;
+  linkData.exitFarcode = primary_farcode_addr;
   b->linkData.push_back(linkData);
 
   blocks.WriteLinkBlock(*this, linkData);
@@ -391,10 +408,32 @@ void JitArm64::WriteExit(u32 destination, bool LK, u32 exit_address_after_return
     linkData.exitPtrs = GetWritableCodePtr();
     linkData.linkStatus = false;
     linkData.call = false;
+    linkData.exitFarcode = return_farcode_addr;
     b->linkData.push_back(linkData);
 
     blocks.WriteLinkBlock(*this, linkData);
   }
+
+  if (switch_to_far_code)
+    SwitchToFarCode();
+  DEBUG_ASSERT(GetCodePtr() == primary_farcode_addr || HasWriteFailed());
+  MOVI2R(DISPATCHER_PC, destination);
+  if (LK)
+    BL(GetAsmRoutines()->do_timing);
+  else
+    B(GetAsmRoutines()->do_timing);
+
+  if (LK)
+  {
+    if (GetCodePtr() == return_farcode_addr - sizeof(u32))
+      BRK(101);
+    DEBUG_ASSERT(GetCodePtr() == return_farcode_addr || HasWriteFailed());
+    MOVI2R(DISPATCHER_PC, exit_address_after_return);
+    B(GetAsmRoutines()->do_timing);
+  }
+
+  if (switch_to_far_code)
+    SwitchToNearCode();
 }
 
 void JitArm64::WriteExit(Arm64Gen::ARM64Reg dest, bool LK, u32 exit_address_after_return)
@@ -431,9 +470,27 @@ void JitArm64::WriteExit(Arm64Gen::ARM64Reg dest, bool LK, u32 exit_address_afte
     linkData.exitPtrs = GetWritableCodePtr();
     linkData.linkStatus = false;
     linkData.call = false;
+    const bool switch_to_far_code = !IsInFarCode();
+    if (switch_to_far_code)
+    {
+      SwitchToFarCode();
+      linkData.exitFarcode = GetCodePtr();
+      SwitchToNearCode();
+    }
+    else
+    {
+      linkData.exitFarcode = GetCodePtr() + JitArm64BlockCache::BLOCK_LINK_SIZE;
+    }
     b->linkData.push_back(linkData);
 
     blocks.WriteLinkBlock(*this, linkData);
+
+    if (switch_to_far_code)
+      SwitchToFarCode();
+    MOVI2R(DISPATCHER_PC, exit_address_after_return);
+    B(GetAsmRoutines()->do_timing);
+    if (switch_to_far_code)
+      SwitchToNearCode();
   }
 }
 
@@ -465,10 +522,28 @@ void JitArm64::FakeLKExit(u32 exit_address_after_return)
   linkData.exitPtrs = GetWritableCodePtr();
   linkData.linkStatus = false;
   linkData.call = false;
+  const bool switch_to_far_code = !IsInFarCode();
+  if (switch_to_far_code)
+  {
+    SwitchToFarCode();
+    linkData.exitFarcode = GetCodePtr();
+    SwitchToNearCode();
+  }
+  else
+  {
+    linkData.exitFarcode = GetCodePtr() + JitArm64BlockCache::BLOCK_LINK_SIZE;
+  }
   b->linkData.push_back(linkData);
 
   blocks.WriteLinkBlock(*this, linkData);
 
+  if (switch_to_far_code)
+    SwitchToFarCode();
+  MOVI2R(DISPATCHER_PC, exit_address_after_return);
+  B(GetAsmRoutines()->do_timing);
+  if (switch_to_far_code)
+    SwitchToNearCode();
+
   SetJumpTarget(skip_exit);
 }
 

Source/Core/Core/PowerPC/JitArm64/JitArm64Cache.cpp

@@ -29,8 +29,9 @@ void JitArm64BlockCache::WriteLinkBlock(Arm64Gen::ARM64XEmitter& emit,
     emit.MOVI2R(DISPATCHER_PC, source.exitAddress);
     if (source.call)
     {
-      while (emit.GetCodePtr() < start + BLOCK_LINK_FAST_BL_OFFSET && !emit.HasWriteFailed())
+      if (emit.GetCodePtr() == start + BLOCK_LINK_FAST_BL_OFFSET - sizeof(u32))
         emit.NOP();
+      DEBUG_ASSERT(emit.GetCodePtr() == start + BLOCK_LINK_FAST_BL_OFFSET || emit.HasWriteFailed());
       emit.BL(m_jit.GetAsmRoutines()->dispatcher);
     }
     else
@@ -45,10 +46,8 @@ void JitArm64BlockCache::WriteLinkBlock(Arm64Gen::ARM64XEmitter& emit,
       // The "fast" BL should be the last instruction, so that the return address matches the
       // address that was pushed onto the stack by the function that called WriteLinkBlock
       FixupBranch fast = emit.B(CC_GT);
-      emit.MOVI2R(DISPATCHER_PC, source.exitAddress);
+      emit.B(source.exitFarcode);
-      emit.BL(m_jit.GetAsmRoutines()->do_timing);
+      DEBUG_ASSERT(emit.GetCodePtr() == start + BLOCK_LINK_FAST_BL_OFFSET || emit.HasWriteFailed());
-      while (emit.GetCodePtr() < start + BLOCK_LINK_FAST_BL_OFFSET && !emit.HasWriteFailed())
-        emit.BRK(101);
       emit.SetJumpTarget(fast);
       emit.BL(dest->normalEntry);
     }
@@ -59,16 +58,14 @@ void JitArm64BlockCache::WriteLinkBlock(Arm64Gen::ARM64XEmitter& emit,
       if (block_distance >= -0x40000 && block_distance <= 0x3FFFF)
       {
         emit.B(CC_GT, dest->normalEntry);
-        emit.MOVI2R(DISPATCHER_PC, source.exitAddress);
+        emit.B(source.exitFarcode);
-        emit.B(m_jit.GetAsmRoutines()->do_timing);
       }
       else
       {
         FixupBranch slow = emit.B(CC_LE);
         emit.B(dest->normalEntry);
         emit.SetJumpTarget(slow);
-        emit.MOVI2R(DISPATCHER_PC, source.exitAddress);
+        emit.B(source.exitFarcode);
-        emit.B(m_jit.GetAsmRoutines()->do_timing);
       }
     }
   }

Source/Core/Core/PowerPC/JitArm64/JitArm64Cache.h

@@ -29,7 +29,7 @@ public:
   void WriteLinkBlock(Arm64Gen::ARM64XEmitter& emit, const JitBlock::LinkData& source,
                       const JitBlock* dest = nullptr);
 
-  static constexpr size_t BLOCK_LINK_SIZE = 5 * sizeof(u32);
+  static constexpr size_t BLOCK_LINK_SIZE = 3 * sizeof(u32);
   static constexpr size_t BLOCK_LINK_FAST_BL_OFFSET = BLOCK_LINK_SIZE - sizeof(u32);
 
 private:

Source/Core/Core/PowerPC/JitCommon/JitCache.h

@@ -70,6 +70,9 @@ struct JitBlock : public JitBlockData
   struct LinkData
   {
     u8* exitPtrs;  // to be able to rewrite the exit jump
+#ifdef _M_ARM_64
+    const u8* exitFarcode;
+#endif
     u32 exitAddress;
     bool linkStatus;  // is it already linked?
     bool call;