Docs: Update for release FEX-2511

f80 stack xchg optimization for fast path

CodeEmitter/CodeEmitter/ALUOps.inl

@@ -36,24 +36,33 @@ public:
     DataProcessing_PCRel_Imm(Op, rd, Imm);
   }
 
-  void adr(ARMEmitter::Register rd, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded adr(ARMEmitter::Register rd, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
     LOGMAN_THROW_A_FMT(IsADRRange(Imm), "Unscaled offset too large");
 
-    constexpr uint32_t Op = 0b0001'0000 << 24;
-    DataProcessing_PCRel_Imm(Op, rd, Imm);
+    if (IsADRRange(Imm)) {
+      constexpr uint32_t Op = 0b0001'0000 << 24;
+      DataProcessing_PCRel_Imm(Op, rd, Imm);
+      return BranchEncodeSucceeded::Success;
+    }
+
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
   }
-  void adr(ARMEmitter::Register rd, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded adr(ARMEmitter::Register rd, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::ADR});
     constexpr uint32_t Op = 0b0001'0000 << 24;
     DataProcessing_PCRel_Imm(Op, rd, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
   }
 
-  void adr(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded adr(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      adr(rd, &Label->Backward);
+      return adr(rd, &Label->Backward);
     } else {
-      adr(rd, &Label->Forward);
+      return adr(rd, &Label->Forward);
     }
   }
 
@@ -62,32 +71,42 @@ public:
     DataProcessing_PCRel_Imm(Op, rd, Imm);
   }
 
-  void adrp(ARMEmitter::Register rd, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded adrp(ARMEmitter::Register rd, const BackwardLabel* Label) {
     int64_t Imm = reinterpret_cast<int64_t>(Label->Location) - (GetCursorAddress<int64_t>() & ~0xFFFLL);
     LOGMAN_THROW_A_FMT(IsADRPRange(Imm) && IsADRPAligned(Imm), "Unscaled offset too large");
 
-    constexpr uint32_t Op = 0b1001'0000 << 24;
-    DataProcessing_PCRel_Imm(Op, rd, Imm);
+    if (IsADRPRange(Imm) && IsADRPAligned(Imm)) {
+      constexpr uint32_t Op = 0b1001'0000 << 24;
+      DataProcessing_PCRel_Imm(Op, rd, Imm);
+      return BranchEncodeSucceeded::Success;
+    }
+
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
   }
-  void adrp(ARMEmitter::Register rd, ForwardLabel* Label) {
+
+  [[nodiscard]] BranchEncodeSucceeded adrp(ARMEmitter::Register rd, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::ADRP});
     constexpr uint32_t Op = 0b1001'0000 << 24;
     DataProcessing_PCRel_Imm(Op, rd, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
   }
 
-  void adrp(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded adrp(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      adrp(rd, &Label->Backward);
+      return adrp(rd, &Label->Backward);
     } else {
-      adrp(rd, &Label->Forward);
+      return adrp(rd, &Label->Forward);
     }
   }
 
-  void LongAddressGen(ARMEmitter::Register rd, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded LongAddressGen(ARMEmitter::Register rd, const BackwardLabel* Label) {
     int64_t Imm = reinterpret_cast<int64_t>(Label->Location) - (GetCursorAddress<int64_t>());
     if (IsADRRange(Imm)) {
       // If the range is in ADR range then we can just use ADR.
-      adr(rd, Label);
+      return adr(rd, Label);
     } else if (IsADRPRange(Imm)) {
       int64_t ADRPImm = (reinterpret_cast<int64_t>(Label->Location) & ~0xFFFLL) - (GetCursorAddress<int64_t>() & ~0xFFFLL);
 
@@ -102,23 +121,28 @@ public:
         // Now even an add
         add(ARMEmitter::Size::i64Bit, rd, rd, AlignedOffset);
       }
-    } else {
-      LOGMAN_MSG_A_FMT("Unscaled offset too large");
-      FEX_UNREACHABLE;
+
+      return BranchEncodeSucceeded::Success;
     }
+
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
   }
-  void LongAddressGen(ARMEmitter::Register rd, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded LongAddressGen(ARMEmitter::Register rd, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::LONG_ADDRESS_GEN});
     // Emit a register index and a nop. These will be backpatched.
     dc32(rd.Idx());
     nop();
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
   }
 
-  void LongAddressGen(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded LongAddressGen(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      LongAddressGen(rd, &Label->Backward);
+      return LongAddressGen(rd, &Label->Backward);
     } else {
-      LongAddressGen(rd, &Label->Forward);
+      return LongAddressGen(rd, &Label->Forward);
     }
   }
 
@@ -862,12 +886,6 @@ public:
   }
 
 private:
-  static constexpr Condition InvertCondition(Condition cond) {
-    // These behave as always, so it makes no sense to allow inverting these.
-    LOGMAN_THROW_A_FMT(cond != Condition::CC_AL && cond != Condition::CC_NV, "Cannot invert CC_AL or CC_NV");
-    return static_cast<Condition>(FEXCore::ToUnderlying(cond) ^ 1);
-  }
-
   void and_(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, uint32_t n, uint32_t immr, uint32_t imms) {
     constexpr uint32_t Op = 0b001'0010'00 << 22;
     DataProcessing_Logical_Imm(Op, s, rd, rn, n, immr, imms);

CodeEmitter/CodeEmitter/BranchOps.inl

@@ -20,23 +20,31 @@ public:
     constexpr uint32_t Op = 0b0101'010 << 25;
     Branch_Conditional(Op, 0, 0, Cond, Imm);
   }
-  void b(ARMEmitter::Condition Cond, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded b(ARMEmitter::Condition Cond, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
-    constexpr uint32_t Op = 0b0101'010 << 25;
-    Branch_Conditional(Op, 0, 0, Cond, Imm >> 2);
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 0, Cond, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }
+
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
   }
-  void b(ARMEmitter::Condition Cond, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded b(ARMEmitter::Condition Cond, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::BC});
     constexpr uint32_t Op = 0b0101'010 << 25;
     Branch_Conditional(Op, 0, 0, Cond, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
   }
 
-  void b(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded b(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      b(Cond, &Label->Backward);
+      return b(Cond, &Label->Backward);
     } else {
-      b(Cond, &Label->Forward);
+      return b(Cond, &Label->Forward);
     }
   }
 
@@ -45,24 +53,32 @@ public:
     constexpr uint32_t Op = 0b0101'010 << 25;
     Branch_Conditional(Op, 0, 1, Cond, Imm);
   }
-  void bc(ARMEmitter::Condition Cond, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded bc(ARMEmitter::Condition Cond, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
-    constexpr uint32_t Op = 0b0101'010 << 25;
-    Branch_Conditional(Op, 0, 1, Cond, Imm >> 2);
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 1, Cond, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }
+
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
   }
 
-  void bc(ARMEmitter::Condition Cond, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded bc(ARMEmitter::Condition Cond, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::BC});
     constexpr uint32_t Op = 0b0101'010 << 25;
     Branch_Conditional(Op, 0, 1, Cond, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
   }
 
-  void bc(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded bc(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      bc(Cond, &Label->Backward);
+      return bc(Cond, &Label->Backward);
     } else {
-      bc(Cond, &Label->Forward);
+      return bc(Cond, &Label->Forward);
     }
   }
 
@@ -98,25 +114,32 @@ public:
 
     UnconditionalBranch(Op, Imm);
   }
-  void b(const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded b(const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
-    constexpr uint32_t Op = 0b0001'01 << 26;
+    if (Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0001'01 << 26;
+      UnconditionalBranch(Op, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }
 
-    UnconditionalBranch(Op, Imm >> 2);
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
   }
-  void b(ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded b(ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::B});
     constexpr uint32_t Op = 0b0001'01 << 26;
 
     UnconditionalBranch(Op, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
   }
 
-  void b(BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded b(BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      b(&Label->Backward);
+      return b(&Label->Backward);
     } else {
-      b(&Label->Forward);
+      return b(&Label->Forward);
     }
   }
 
@@ -126,25 +149,33 @@ public:
     UnconditionalBranch(Op, Imm);
   }
 
-  void bl(const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded bl(const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
-    constexpr uint32_t Op = 0b1001'01 << 26;
+    if (Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b1001'01 << 26;
+      UnconditionalBranch(Op, Imm >> 2);
 
-    UnconditionalBranch(Op, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }
+
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
   }
-  void bl(ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded bl(ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::B});
     constexpr uint32_t Op = 0b1001'01 << 26;
 
     UnconditionalBranch(Op, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
   }
 
-  void bl(BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded bl(BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      bl(&Label->Backward);
+      return bl(&Label->Backward);
     } else {
-      bl(&Label->Forward);
+      return bl(&Label->Forward);
     }
   }
 
@@ -155,28 +186,35 @@ public:
     CompareAndBranch(Op, s, rt, Imm);
   }
 
-  void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded cbz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
 
-    constexpr uint32_t Op = 0b0011'0100 << 24;
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0011'0100 << 24;
+      CompareAndBranch(Op, s, rt, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }
 
-    CompareAndBranch(Op, s, rt, Imm >> 2);
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
   }
 
-  void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded cbz(ARMEmitter::Size s, ARMEmitter::Register rt, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::BC});
 
     constexpr uint32_t Op = 0b0011'0100 << 24;
 
     CompareAndBranch(Op, s, rt, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
   }
 
-  void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded cbz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      cbz(s, rt, &Label->Backward);
+      return cbz(s, rt, &Label->Backward);
     } else {
-      cbz(s, rt, &Label->Forward);
+      return cbz(s, rt, &Label->Forward);
     }
   }
 
@@ -186,28 +224,35 @@ public:
     CompareAndBranch(Op, s, rt, Imm);
   }
 
-  void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
 
-    constexpr uint32_t Op = 0b0011'0101 << 24;
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0011'0101 << 24;
+      CompareAndBranch(Op, s, rt, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }
 
-    CompareAndBranch(Op, s, rt, Imm >> 2);
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
   }
 
-  void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::BC});
 
     constexpr uint32_t Op = 0b0011'0101 << 24;
 
     CompareAndBranch(Op, s, rt, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
   }
 
-  void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      cbnz(s, rt, &Label->Backward);
+      return cbnz(s, rt, &Label->Backward);
     } else {
-      cbnz(s, rt, &Label->Forward);
+      return cbnz(s, rt, &Label->Forward);
     }
   }
 
@@ -217,28 +262,35 @@ public:
 
     TestAndBranch(Op, rt, Bit, Imm);
   }
-  void tbz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded tbz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");
 
-    constexpr uint32_t Op = 0b0011'0110 << 24;
+    if (Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0011'0110 << 24;
+      TestAndBranch(Op, rt, Bit, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }
 
-    TestAndBranch(Op, rt, Bit, Imm >> 2);
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
   }
 
-  void tbz(ARMEmitter::Register rt, uint32_t Bit, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded tbz(ARMEmitter::Register rt, uint32_t Bit, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::TEST_BRANCH});
 
     constexpr uint32_t Op = 0b0011'0110 << 24;
 
     TestAndBranch(Op, rt, Bit, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
   }
 
-  void tbz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded tbz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      tbz(rt, Bit, &Label->Backward);
+      return tbz(rt, Bit, &Label->Backward);
     } else {
-      tbz(rt, Bit, &Label->Forward);
+      return tbz(rt, Bit, &Label->Forward);
     }
   }
 
@@ -247,27 +299,35 @@ public:
 
     TestAndBranch(Op, rt, Bit, Imm);
   }
-  void tbnz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded tbnz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* Label) {
     int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
     LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");
 
-    constexpr uint32_t Op = 0b0011'0111 << 24;
+    if (Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0011'0111 << 24;
+      TestAndBranch(Op, rt, Bit, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }
 
-    TestAndBranch(Op, rt, Bit, Imm >> 2);
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
   }
 
-  void tbnz(ARMEmitter::Register rt, uint32_t Bit, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded tbnz(ARMEmitter::Register rt, uint32_t Bit, ForwardLabel* Label) {
     AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::TEST_BRANCH});
     constexpr uint32_t Op = 0b0011'0111 << 24;
 
     TestAndBranch(Op, rt, Bit, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
   }
 
-  void tbnz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded tbnz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
     if (Label->Backward.Location) {
-      tbnz(rt, Bit, &Label->Backward);
+      return tbnz(rt, Bit, &Label->Backward);
     } else {
-      tbnz(rt, Bit, &Label->Forward);
+      return tbnz(rt, Bit, &Label->Forward);
     }
   }
 

CodeEmitter/CodeEmitter/Emitter.h

@@ -586,6 +586,15 @@ concept IsXOrWRegister = std::is_same_v<T, XRegister> || std::is_same_v<T, WRegi
 template<typename T>
 concept IsQOrDRegister = std::is_same_v<T, QRegister> || std::is_same_v<T, DRegister>;
 
+template<typename T>
+concept IsLabel = std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
+                  std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>;
+
+enum class BranchEncodeSucceeded {
+  Success,
+  Failure,
+};
+
 // Whether or not a given set of vector registers are sequential
 // in increasing order as far as the register file is concerned (modulo its size)
 //
@@ -638,19 +647,25 @@ public:
 
   // Bind a backward label to an address.
   // Address that is bound is the current emitter location.
-  void Bind(BackwardLabel* Label) {
+  [[nodiscard]] bool Bind(BackwardLabel* Label) {
     LOGMAN_THROW_A_FMT(Label->Location == nullptr, "Trying to bind a label twice");
     Label->Location = GetCursorAddress<uint8_t*>();
+
+    // Always binds because it is only storing a location.
+    return true;
   }
 
-  void Bind(const ForwardLabel::Reference* Label) {
+  [[nodiscard]] bool Bind(const ForwardLabel::Reference* Label) {
     uint8_t* CurrentAddress = GetCursorAddress<uint8_t*>();
     // Patch up the instructions
     switch (Label->Type) {
     case ForwardLabel::InstType::ADR: {
       uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
       int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      LOGMAN_THROW_A_FMT(IsADRRange(Imm), "Unscaled offset too large");
+      if (!IsADRRange(Imm)) {
+        // Can't bind.
+        return false;
+      }
       uint32_t InstMask = 0b11 << 29 | 0b1111'1111'1111'1111'111 << 5;
       uint32_t Offset = static_cast<uint32_t>(Imm) & 0x3F'FFFF;
       uint32_t Inst = *Instruction & ~InstMask;
@@ -662,7 +677,12 @@ public:
     case ForwardLabel::InstType::ADRP: {
       uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
       int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      LOGMAN_THROW_A_FMT(IsADRPRange(Imm) && IsADRPAligned(Imm), "Unscaled offset too large");
+
+      if (!(IsADRPRange(Imm) && IsADRPAligned(Imm))) {
+        // Can't bind.
+        return false;
+      }
+
       Imm >>= 12;
       uint32_t InstMask = 0b11 << 29 | 0b1111'1111'1111'1111'111 << 5;
       uint32_t Offset = static_cast<uint32_t>(Imm) & 0x3F'FFFF;
@@ -672,11 +692,13 @@ public:
       *Instruction = Inst;
       break;
     }
-
     case ForwardLabel::InstType::B: {
       uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
       int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+      if (!(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0))) {
+        // Can't bind.
+        return false;
+      }
       Imm >>= 2;
       uint32_t InstMask = 0x3FF'FFFF;
       uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
@@ -686,11 +708,13 @@ public:
 
       break;
     }
-
     case ForwardLabel::InstType::TEST_BRANCH: {
       uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
       int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+      if (!(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0))) {
+        // Can't bind.
+        return false;
+      }
       Imm >>= 2;
       uint32_t InstMask = 0x3FFF;
       uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
@@ -704,7 +728,10 @@ public:
     case ForwardLabel::InstType::RELATIVE_LOAD: {
       uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
       int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+      if (!(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0))) {
+        // Can't bind.
+        return false;
+      }
       Imm >>= 2;
       uint32_t InstMask = 0x7'FFFF;
       uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
@@ -753,27 +780,41 @@ public:
     }
     default: LOGMAN_MSG_A_FMT("Unexpected inst type in label fixup");
     }
+
+    return true;
   }
 
   // Bind a forward label to a location.
   // This walks all the instructions in the label's vector.
   // Then backpatching all instructions that have used the label.
-  void Bind(ForwardLabel* Label) {
+  [[nodiscard]] bool Bind(ForwardLabel* Label) {
+    bool Bound = true;
     if (Label->FirstInst.Location) {
-      Bind(&Label->FirstInst);
+      Bound &= Bind(&Label->FirstInst);
     }
     for (auto& Inst : Label->Insts) {
-      Bind(&Inst);
+      Bound &= Bind(&Inst);
     }
+
+    return Bound;
   }
 
   // Bind a bidirectional location to a location.
   // Binds both forwards and backwards depending on how the label was used.
-  void Bind(BiDirectionalLabel* Label) {
+  [[nodiscard]] bool Bind(BiDirectionalLabel* Label) {
+    bool Bound = true;
     if (!Label->Backward.Location) {
-      Bind(&Label->Backward);
+      Bound &= Bind(&Label->Backward);
     }
-    Bind(&Label->Forward);
+    Bound &= Bind(&Label->Forward);
+
+    return Bound;
+  }
+
+  static constexpr Condition InvertCondition(Condition cond) {
+    // These behave as always, so it makes no sense to allow inverting these.
+    LOGMAN_THROW_A_FMT(cond != Condition::CC_AL && cond != Condition::CC_NV, "Cannot invert CC_AL or CC_NV");
+    return static_cast<Condition>(FEXCore::ToUnderlying(cond) ^ 1);
   }
 
 #include <CodeEmitter/VixlUtils.inl>

FEXCore/Source/CMakeLists.txt

@@ -66,6 +66,7 @@ set (SRCS
   Interface/IR/Passes/RedundantFlagCalculationElimination.cpp
   Interface/IR/Passes/RegisterAllocationPass.cpp
   Interface/IR/Passes/x87StackOptimizationPass.cpp
+  Utils/LongJump.cpp
   Utils/Telemetry.cpp
   Utils/Threads.cpp
   Utils/Profiler.cpp

FEXCore/Source/Common/JitSymbols.h

@@ -1,5 +1,6 @@
 // SPDX-License-Identifier: MIT
 #pragma once
+#include <FEXCore/Utils/TypeDefines.h>
 
 #include <FEXCore/fextl/memory.h>
 
@@ -12,7 +13,7 @@ namespace FEXCore {
 // Buffered JIT symbol tracking.
 struct JITSymbolBuffer {
   // Maximum buffer size to ensure we are a page in size.
-  constexpr static size_t BUFFER_SIZE = 4096 - (8 * 2);
+  constexpr static size_t BUFFER_SIZE = FEXCore::Utils::FEX_PAGE_SIZE - (8 * 2);
   // Maximum distance until the end of the buffer to do a write.
   constexpr static size_t NEEDS_WRITE_DISTANCE = BUFFER_SIZE - 64;
   // Maximum time threshhold to wait before a buffer write occurs.
@@ -27,7 +28,7 @@ struct JITSymbolBuffer {
   size_t Offset {};
   char Buffer[BUFFER_SIZE] {};
 };
-static_assert(sizeof(JITSymbolBuffer) == 4096, "Ensure this is one page in size");
+static_assert(sizeof(JITSymbolBuffer) == FEXCore::Utils::FEX_PAGE_SIZE, "Ensure this is one page in size");
 
 class JITSymbols final {
 public:

FEXCore/Source/Common/SoftFloat.h

@@ -504,47 +504,6 @@ struct FEX_PACKED X80SoftFloat {
     return std::bit_cast<float>(Result);
   }
 
-  bool IsSignalingNaN() const {
-    return (Exponent == 0x7FFF) && (Significand & 0x8000000000000000ULL) && !(Significand & 0x4000000000000000ULL) && // Bit 62 clear (signaling)
-           (Significand & 0x3FFFFFFFFFFFFFFFULL);
-  }
-
-  bool IsQuietNaN() const {
-    return (Exponent == 0x7FFF) && (Significand & 0x8000000000000000ULL) && (Significand & 0x4000000000000000ULL); // Bit 62 set (quiet)
-  }
-
-  // Helper to detect if this is any NaN
-  bool IsNaN() const {
-    return IsSignalingNaN() || IsQuietNaN();
-  }
-
-  // X87 value to F64 while preserving signaling nan property
-  double ToF64_PreserveNan(softfloat_state* state) const {
-    if (IsSignalingNaN()) {
-      // we keep it as a signaling nan in ieee754 in 64bits
-      uint64_t sign_bit = Sign ? 0x8000000000000000ULL : 0;
-      uint64_t exp_bits = 0x7FF0000000000000ULL;
-      uint64_t x87_frac = Significand & 0x3FFFFFFFFFFFFFFFULL;
-      uint64_t ieee_frac = (x87_frac >> 11) & 0x0007FFFFFFFFFFFFULL;
-
-      if (ieee_frac == 0) {
-        ieee_frac = 1;
-      }
-      ieee_frac &= ~0x0008000000000000ULL;
-
-      uint64_t result_bits = sign_bit | exp_bits | ieee_frac;
-      return std::bit_cast<double>(result_bits);
-    } else if (IsQuietNaN()) {
-      const float64_t Result = extF80_to_f64(state, *this);
-      uint64_t result_bits = std::bit_cast<uint64_t>(Result);
-      result_bits |= 0x0008000000000000ULL;
-      return std::bit_cast<double>(result_bits);
-    } else {
-      const float64_t Result = extF80_to_f64(state, *this);
-      return std::bit_cast<double>(Result);
-    }
-  }
-
   double ToF64(softfloat_state* state) const {
     const float64_t Result = extF80_to_f64(state, *this);
     return std::bit_cast<double>(Result);
@@ -625,39 +584,6 @@ struct FEX_PACKED X80SoftFloat {
     *this = f64_to_extF80(state, std::bit_cast<float64_t>(rhs));
   }
 
-  // Create X80SoftFloat from double while preserving NaN signaling properties
-  static X80SoftFloat FromF64_PreserveNaN(softfloat_state* state, double value) {
-    uint64_t bits = std::bit_cast<uint64_t>(value);
-
-    // Check if it's a nan
-    if ((bits & 0x7FF0000000000000ULL) == 0x7FF0000000000000ULL && (bits & 0x000FFFFFFFFFFFFFULL) != 0) {
-
-      X80SoftFloat result;
-      result.Sign = (bits >> 63) & 1;
-      result.Exponent = 0x7FFF;
-
-      bool is_signaling = !(bits & 0x0008000000000000ULL);
-      uint64_t ieee_payload = bits & 0x0007FFFFFFFFFFFFULL;
-
-      // set bit 63 required for x87
-      result.Significand = 0x8000000000000000ULL;
-
-      if (is_signaling) { // clear bit 62 for signaling nan
-        result.Significand &= ~0x4000000000000000ULL;
-      } else { // clear bit 62 for quiet nan
-        result.Significand |= 0x4000000000000000ULL;
-      }
-
-      // ieee754 51-bit payload -> x87 62-bit payload
-      result.Significand |= (ieee_payload << 11) & 0x3FFFFFFFFFFFFFFFULL;
-
-      return result;
-    }
-
-    // For non-NaN values, use standard conversion
-    return X80SoftFloat(state, value);
-  }
-
   X80SoftFloat(softfloat_state* state, BIGFLOAT rhs) {
 #if BIGFLOATSIZE == 16
     *this = f128_to_extF80(state, std::bit_cast<float128_t>(rhs));

FEXCore/Source/Common/VectorRegType.h

@@ -4,6 +4,8 @@
 #ifdef _M_X86_64
 #include <xmmintrin.h>
 #include <immintrin.h>
+#else
+#include <cstdint>
 #endif
 
 namespace FEXCore {

FEXCore/Source/Interface/Config/Config.json.in

@@ -161,6 +161,44 @@
         "Desc": [
           "Allows the user to pass additional arguments to the application"
         ]
+      },
+      "DisableL2Cache": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Disables FEXCore's JIT L2 cache lookup. Saving memory.",
+          "Can potentially introduce more stutters."
+        ]
+      },
+      "DynamicL1Cache": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Switches FEXCore's JIT L1 cache to be dynamically sized. Saving memory.",
+          "Can potentially introduce more stutters."
+        ]
+      },
+      "DynamicL1CacheIncreaseCountHeuristic": {
+        "Type": "uint64",
+        "Default": "250",
+        "Desc": [
+          "Threshold of lookups per second that the L1 dynamic cache should increase its size.",
+          "Lower numbers means more aggressive scaling upward to the maximum size.",
+          "Higher numbers means more conservative scaling, using less memory.",
+          "Can potentially introduce stutters, more likely the higher the number.",
+          "Don't have this number smaller than the decrease count!"
+        ]
+      },
+      "DynamicL1CacheDecreaseCountHeuristic": {
+        "Type": "uint64",
+        "Default": "50",
+        "Desc": [
+          "Threshold of lookups per second that the L1 dynamic cache should decrease its size.",
+          "The higher the number, the more aggressively it reduces the L1 cache size.",
+          "Lower numbers means more conservative memory savings.",
+          "Can potentially introduce more stutters, more likely the higher the number.",
+          "Don't have this number larger than the increase count!"
+        ]
       }
     },
     "Debug": {
@@ -330,6 +368,13 @@
           "Enables FEX's low-overhead sampling profile statistics.",
           "Requires a supported version of Mangohud to see the results"
         ]
+      },
+      "EnableGpuvisProfiling": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Enables profiling when FEX was built with the gpuvis profiler backend."
+        ]
       }
     },
     "Hacks": {
@@ -389,7 +434,7 @@
         "Default": "true",
         "Desc": [
           "Use volatile metadata in PE files to inform TSO instructions when available.",
-	  "When metadata is unavailable falls back to the currently enabled TSO options."
+          "When metadata is unavailable falls back to the currently enabled TSO options."
         ]
       },
       "X87ReducedPrecision": {
@@ -399,31 +444,6 @@
           "Emulates X87 floating point using 64-bit precision. This reduces emulation accuracy and may result in rendering bugs."
         ]
       },
-      "X87StrictReducedPrecision": {
-        "Type": "bool",
-        "Default": "false",
-        "Desc": [
-          "Enables stricter X87 floating point behavior when X87ReducedPrecision is enabled.",
-          "Adds additional checks and implementations like NaN propagation for better compatibility."
-        ]
-      },
-      "ABILocalFlags": {
-        "Type": "bool",
-        "Default": "false",
-        "Desc": [
-          "When enabled enables an optimization around flags.",
-          "Assumes flags are not used across cals.",
-          "Hand-written assembly can violate this assumption."
-        ]
-      },
-      "ParanoidTSO": {
-        "Type": "bool",
-        "Default": "false",
-        "Desc": [
-          "Makes TSO operations even more strict.",
-          "Forces vector loadstores to also become atomic."
-        ]
-      },
       "StallProcess": {
         "Type": "bool",
         "Default": "false",

FEXCore/Source/Interface/Context/Context.h

@@ -29,6 +29,7 @@
 namespace FEXCore {
 class SignalDelegator;
 class ThunkHandler;
+struct LookupCacheWriteLockToken;
 
 namespace Core {
   struct DebugData;
@@ -197,7 +198,6 @@ public:
     FEX_CONFIG_OPT(TSOEnabled, TSOENABLED);
     FEX_CONFIG_OPT(VectorTSOEnabled, VECTORTSOENABLED);
     FEX_CONFIG_OPT(MemcpySetTSOEnabled, MEMCPYSETTSOENABLED);
-    FEX_CONFIG_OPT(ABILocalFlags, ABILOCALFLAGS);
     FEX_CONFIG_OPT(SMCChecks, SMCCHECKS);
     FEX_CONFIG_OPT(MaxInstPerBlock, MAXINST);
     FEX_CONFIG_OPT(RootFSPath, ROOTFS);
@@ -205,9 +205,7 @@ public:
     FEX_CONFIG_OPT(LibraryJITNaming, LIBRARYJITNAMING);
     FEX_CONFIG_OPT(BlockJITNaming, BLOCKJITNAMING);
     FEX_CONFIG_OPT(GDBSymbols, GDBSYMBOLS);
-    FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
     FEX_CONFIG_OPT(x87ReducedPrecision, X87REDUCEDPRECISION);
-    FEX_CONFIG_OPT(x87StrictReducedPrecision, X87STRICTREDUCEDPRECISION);
     FEX_CONFIG_OPT(DisableTelemetry, DISABLETELEMETRY);
     FEX_CONFIG_OPT(DisableVixlIndirectCalls, DISABLE_VIXL_INDIRECT_RUNTIME_CALLS);
     FEX_CONFIG_OPT(SmallTSCScale, SMALLTSCSCALE);
@@ -233,7 +231,7 @@ public:
 
   ContextImpl(const FEXCore::HostFeatures& Features);
 
-  static bool ThreadRemoveCodeEntry(FEXCore::Core::InternalThreadState* Thread, uint64_t GuestRIP);
+  static bool ThreadRemoveCodeEntry(FEXCore::Core::InternalThreadState* Thread, uint64_t GuestRIP, const FEXCore::LookupCacheWriteLockToken& lk);
 
   static void ThreadRemoveCodeEntryFromJit(FEXCore::Core::CpuStateFrame* Frame, uint64_t GuestRIP);
 
@@ -269,9 +267,9 @@ public:
 
   FEXCore::JITSymbols Symbols;
 
-  FEXCore::Utils::PooledAllocatorVirtual OpDispatcherAllocator;
-  FEXCore::Utils::PooledAllocatorVirtual FrontendAllocator;
-  FEXCore::Utils::PooledAllocatorVirtual CPUBackendAllocator;
+  FEXCore::Utils::PooledAllocatorVirtual OpDispatcherAllocator {"FEXMem_OpDispatcher"};
+  FEXCore::Utils::PooledAllocatorVirtual FrontendAllocator {"FEXMem_Frontend"};
+  FEXCore::Utils::PooledAllocatorVirtual CPUBackendAllocator {"FEXMem_CPUBackend"};
 
   // If Atomic-based TSO emulation is enabled or not.
   bool IsAtomicTSOEnabled() const {
@@ -312,10 +310,6 @@ protected:
       AtomicTSOEmulationEnabled = false;
       VectorAtomicTSOEmulationEnabled = false;
       MemcpyAtomicTSOEmulationEnabled = false;
-    } else if (Config.ParanoidTSO) {
-      AtomicTSOEmulationEnabled = true;
-      VectorAtomicTSOEmulationEnabled = true;
-      MemcpyAtomicTSOEmulationEnabled = true;
     } else {
       AtomicTSOEmulationEnabled = Config.TSOEnabled;
       VectorAtomicTSOEmulationEnabled = Config.TSOEnabled && Config.VectorTSOEnabled;
@@ -323,13 +317,6 @@ protected:
     }
   }
 
-  void UpdateX87PrecisionConfig() {
-    // If strict reduced precision is enabled, automatically enable reduced precision
-    if (Config.x87StrictReducedPrecision() && !Config.x87ReducedPrecision()) {
-      FEXCore::Config::Set(FEXCore::Config::CONFIG_X87REDUCEDPRECISION, "1");
-    }
-  }
-
 private:
   /**
    * @brief Initializes the JIT compilers for the thread

FEXCore/Source/Interface/Core/CPUBackend.cpp

@@ -360,9 +360,7 @@ namespace CPU {
       LogMan::Msg::EFmt("Failed to mprotect last page of code buffer.");
     }
 
-#ifndef _WIN32
-    prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, Ptr, Size, "FEXMemJIT");
-#endif
+    FEXCore::Allocator::VirtualName("FEXMemJIT", reinterpret_cast<void*>(Ptr), Size);
 
     LookupCache = fextl::make_unique<GuestToHostMap>();
   }

FEXCore/Source/Interface/Core/CPUID.cpp

@@ -88,6 +88,7 @@ namespace ProductNames {
   static const char ARM_Blizzard_M2Pro[] = "Apple Blizzard (M2 Pro)";
   static const char ARM_Avalanche_M2Max[] = "Apple Avalanche (M2 Max)";
   static const char ARM_Blizzard_M2Max[] = "Apple Blizzard (M2 Max)";
+  static const char ARM_AppleSilicon[] = "Apple Silicon";
 
   static const char ARM_ORYON_1[] = "Oryon-1";
   static const char ARM_Ampere_1[] = "AmpereOne";
@@ -188,6 +189,7 @@ void CPUIDEmu::SetupHostHybridFlag() {
     {0x61, 0x029, 1, ProductNames::ARM_Firestorm_M1Max}, // Apple Firestorm (M1 Max)
     {0x61, 0x025, 1, ProductNames::ARM_Firestorm_M1Pro}, // Apple Firestorm (M1 Pro)
     {0x61, 0x023, 1, ProductNames::ARM_Firestorm_M1},    // Apple Firestorm (M1)
+    {0x61, 0, 1, ProductNames::ARM_AppleSilicon},        // QEmu Apple Silicon
 
     {0x41, 0xd8c, 1, ProductNames::ARM_C1Ultra},   // C1-Ultra
     {0x41, 0xd90, 1, ProductNames::ARM_C1Premium}, // C1-Premium

FEXCore/Source/Interface/Core/Core.cpp

@@ -57,13 +57,20 @@ $end_info$
 #include <algorithm>
 #include <array>
 #include <atomic>
+#include <chrono>
+#include <condition_variable>
 #include <fcntl.h>
 #include <functional>
 #include <mutex>
+#include <queue>
 #include <shared_mutex>
+#include <signal.h>
 #include <stdio.h>
+#include <string_view>
 #include <sys/stat.h>
+#include <type_traits>
 #include <unistd.h>
+#include <unordered_map>
 #include <utility>
 #include <xxhash.h>
 
@@ -93,8 +100,6 @@ ContextImpl::ContextImpl(const FEXCore::HostFeatures& Features)
 
   // Track atomic TSO emulation configuration.
   UpdateAtomicTSOEmulationConfig();
-  // Ensure X87 precision constraints are respected.
-  UpdateX87PrecisionConfig();
 }
 
 struct GetFrameBlockInfoResult {
@@ -371,7 +376,9 @@ void ContextImpl::InitializeCompiler(FEXCore::Core::InternalThreadState* Thread)
   Thread->FrontendDecoder = fextl::make_unique<FEXCore::Frontend::Decoder>(Thread);
   Thread->PassManager = fextl::make_unique<FEXCore::IR::PassManager>();
 
-  Thread->CurrentFrame->Pointers.Common.L1Pointer = Thread->LookupCache->GetL1Pointer();
+  Thread->CurrentFrame->State.L1Pointer = Thread->LookupCache->GetL1Pointer();
+  Thread->CurrentFrame->State.L1Mask = Thread->LookupCache->GetScaledL1PointerMask();
+
   Thread->CurrentFrame->Pointers.Common.L2Pointer = Thread->LookupCache->GetPagePointer();
 
   Dispatcher->InitThreadPointers(Thread);
@@ -393,6 +400,7 @@ ContextImpl::CreateThread(uint64_t InitialRIP, uint64_t StackPointer, const FEXC
   FEXCore::Core::InternalThreadState* Thread = new FEXCore::Core::InternalThreadState {
     .CTX = this,
   };
+  FEXCore::Allocator::VirtualName("FEXMem_ThreadState", Thread, sizeof(*Thread));
 
   Thread->CurrentFrame->State.gregs[X86State::REG_RSP] = StackPointer;
   Thread->CurrentFrame->State.rip = InitialRIP;
@@ -465,7 +473,8 @@ void ContextImpl::ClearCodeCache(FEXCore::Core::InternalThreadState* Thread, boo
     Thread->CPUBackend->ClearCache();
   } else {
     // Clear L1+L2 cache of this thread, and clear L3 cache across any threads using it
-    Thread->LookupCache->ClearCache();
+    auto lk = Thread->LookupCache->AcquireWriteLock();
+    Thread->LookupCache->ClearCache(lk);
   }
   Allocator::VirtualDontNeed(Thread->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
 }
@@ -637,10 +646,10 @@ ContextImpl::GenerateIR(FEXCore::Core::InternalThreadState* Thread, uint64_t Gue
               LogMan::Msg::EFmt("Invalid or Unknown instruction: {} 0x{:x}", TableInfo->Name ?: "UND", Block.Entry - GuestRIP);
             }
 
-            if (Block.BlockStatus == Frontend::Decoder::DecodedBlockStatus::NOEXEC_INST) {
-              Thread->OpDispatcher->NoExecOp(DecodedInfo);
-            } else {
+            if (Block.BlockStatus == Frontend::Decoder::DecodedBlockStatus::INVALID_INST) {
               Thread->OpDispatcher->InvalidOp(DecodedInfo);
+            } else {
+              Thread->OpDispatcher->NoExecOp(DecodedInfo);
             }
           }
 
@@ -723,7 +732,7 @@ ContextImpl::CompileCodeResult ContextImpl::CompileCode(FEXCore::Core::InternalT
   // but this would increase lock contention. Redundant frontend runs aren't
   // as expensive and are easily reverted.
   if (MaxInst != 1) {
-    if (auto Block = Thread->LookupCache->FindBlock(GuestRIP)) {
+    if (auto Block = Thread->LookupCache->FindBlock(Thread, GuestRIP)) {
       Thread->OpDispatcher->DelayedDisownBuffer();
       return {.CompiledCode = {.BlockBegin = reinterpret_cast<uint8_t*>(Block), .EntryPoints = {{GuestRIP, reinterpret_cast<uint8_t*>(Block)}}},
               .DebugData = nullptr,
@@ -764,10 +773,13 @@ uintptr_t ContextImpl::CompileBlock(FEXCore::Core::CpuStateFrame* Frame, uint64_
 
   // Is the code in the cache?
   // The backends only check L1 and L2, not L3
-  if (auto HostCode = Thread->LookupCache->FindBlock(GuestRIP)) {
+  if (auto HostCode = Thread->LookupCache->FindBlock(Thread, GuestRIP)) {
     return HostCode;
   }
 
+  // Accumulate a JIT count now, as even if another thread raced us, it should count as a compile.
+  FEXCORE_PROFILE_INSTANT_INCREMENT(Thread, AccumulatedJITCount, 1);
+
   auto [CompiledCode, DebugData, StartAddr, Length, NeedsAddGuestCodeRanges] = CompileCode(Thread, GuestRIP, MaxInst);
   auto CodePtr = CompiledCode.EntryPoints[GuestRIP];
   if (CodePtr == nullptr) {
@@ -826,7 +838,7 @@ uintptr_t ContextImpl::CompileBlock(FEXCore::Core::CpuStateFrame* Frame, uint64_
     // contain code, inform the frontend so it can setup SMC detection.
     auto BlockInfo = Thread->FrontendDecoder->GetDecodedBlockInfo();
     for (auto CodePage : BlockInfo->CodePages) {
-      if (Thread->LookupCache->AddBlockExecutableRange(BlockInfo->EntryPoints, CodePage, FEXCore::Utils::FEX_PAGE_SIZE)) {
+      if (Thread->LookupCache->AddBlockExecutableRange(Thread, BlockInfo->EntryPoints, CodePage, FEXCore::Utils::FEX_PAGE_SIZE)) {
         SyscallHandler->MarkGuestExecutableRange(Thread, CodePage, FEXCore::Utils::FEX_PAGE_SIZE);
       }
     }
@@ -834,7 +846,7 @@ uintptr_t ContextImpl::CompileBlock(FEXCore::Core::CpuStateFrame* Frame, uint64_
 
   // Insert to lookup cache
   for (auto [GuestAddr, HostAddr] : CompiledCode.EntryPoints) {
-    Thread->LookupCache->AddBlockMapping(GuestAddr, HostAddr);
+    Thread->LookupCache->AddBlockMapping(Thread, GuestAddr, HostAddr);
   }
 
   return (uintptr_t)CodePtr;
@@ -867,7 +879,7 @@ static void InvalidateGuestThreadCodeRange(FEXCore::Core::InternalThreadState* T
   // Accessing FrontendDecoder is safe as the thread's code invalidation mutex must be locked here.
   Thread->FrontendDecoder->ResetExecutableRangeCache();
 
-  auto lk = Thread->LookupCache->AcquireLock();
+  auto lk = Thread->LookupCache->AcquireWriteLock();
   auto& CodePages = Thread->LookupCache->Shared->CodePages;
 
   auto lower = CodePages.lower_bound(Start >> 12);
@@ -880,7 +892,7 @@ static void InvalidateGuestThreadCodeRange(FEXCore::Core::InternalThreadState* T
   bool InvalidatedAnyEntries = false;
   for (const auto& PageEntries : Accumulator) {
     for (const auto& Entry : PageEntries) {
-      if (ContextImpl::ThreadRemoveCodeEntry(Thread, Entry)) {
+      if (ContextImpl::ThreadRemoveCodeEntry(Thread, Entry, lk)) {
         InvalidatedAnyEntries = true;
       }
     }
@@ -897,11 +909,12 @@ void ContextImpl::InvalidateGuestCodeRange(FEXCore::Core::InternalThreadState* T
   InvalidateGuestThreadCodeRange(Thread, Accumulator, Start, Length);
 }
 
-bool ContextImpl::ThreadRemoveCodeEntry(FEXCore::Core::InternalThreadState* Thread, uint64_t GuestRIP) {
+bool ContextImpl::ThreadRemoveCodeEntry(FEXCore::Core::InternalThreadState* Thread, uint64_t GuestRIP,
+                                        const FEXCore::LookupCacheWriteLockToken& lk) {
   LogMan::Throw::AFmt(static_cast<ContextImpl*>(Thread->CTX)->CodeInvalidationMutex.try_lock() == false, "CodeInvalidationMutex needs to "
                                                                                                          "be unique_locked here");
 
-  return Thread->LookupCache->Erase(Thread->CurrentFrame, GuestRIP);
+  return Thread->LookupCache->Erase(Thread->CurrentFrame, GuestRIP, lk);
 }
 
 void ContextImpl::ThreadRemoveCodeEntryFromJit(FEXCore::Core::CpuStateFrame* Frame, uint64_t GuestRIP) {
@@ -971,7 +984,7 @@ void ContextImpl::AddThunkTrampolineIRHandler(uintptr_t Entrypoint, uintptr_t Gu
 void ContextImpl::AddForceTSOInformation(const IntervalList<uint64_t>& ValidRanges, fextl::set<uint64_t>&& Instructions) {
   LogMan::Throw::AFmt(CodeInvalidationMutex.try_lock() == false, "CodeInvalidationMutex needs to be unique_locked here");
   ForceTSOValidRanges.Insert(ValidRanges);
-  ForceTSOInstructions.merge(Instructions);
+  ForceTSOInstructions.merge(std::move(Instructions));
 }
 
 void ContextImpl::RemoveForceTSOInformation(uint64_t Address, uint64_t Size) {

FEXCore/Source/Interface/Core/Dispatcher/Dispatcher.cpp

@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: MIT
 
-#include "Common/SoftFloat.h"
+#include "Common/VectorRegType.h"
 #include "Interface/Context/Context.h"
 #include "Interface/Core/CPUBackend.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
@@ -26,9 +26,7 @@
 #endif
 
 #include <array>
-#include <atomic>
 #include <bit>
-#include <condition_variable>
 #include <csignal>
 #include <cstring>
 
@@ -38,12 +36,14 @@ static void SleepThread(FEXCore::Context::ContextImpl* CTX, FEXCore::Core::CpuSt
   CTX->SyscallHandler->SleepThread(CTX, Frame);
 }
 
-constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096 * 4;
+constexpr size_t MAX_DISPATCHER_CODE_SIZE = FEXCore::Utils::FEX_PAGE_SIZE * 4;
 
 Dispatcher::Dispatcher(FEXCore::Context::ContextImpl* ctx)
   : Arm64Emitter(ctx, FEXCore::Allocator::VirtualAlloc(MAX_DISPATCHER_CODE_SIZE, true), MAX_DISPATCHER_CODE_SIZE)
   , CTX {ctx} {
   EmitDispatcher();
+
+  FEXCore::Allocator::VirtualName("FEXMem_Misc", reinterpret_cast<void*>(GetBufferBase()), MAX_DISPATCHER_CODE_SIZE);
 }
 
 Dispatcher::~Dispatcher() {
@@ -93,12 +93,12 @@ void Dispatcher::EmitDispatcher() {
 
   FillStaticRegs();
   ldr(RipReg, STATE_PTR(CpuStateFrame, State.rip));
-  cbnz(ARMEmitter::Size::i32Bit, ENTRY_FILL_SRA_SINGLE_INST_REG, &CompileSingleStep);
+  (void)cbnz(ARMEmitter::Size::i32Bit, ENTRY_FILL_SRA_SINGLE_INST_REG, &CompileSingleStep);
 
   ARMEmitter::BiDirectionalLabel LoopTop {};
 
 #ifdef _M_ARM_64EC
-  b(&LoopTop);
+  (void)b(&LoopTop);
 
   AbsoluteLoopTopAddressEnterECFillSRA = GetCursorAddress<uint64_t>();
   ldr(STATE, EC_ENTRY_CPUAREA_REG, CPU_AREA_EMULATOR_DATA_OFFSET);
@@ -106,10 +106,10 @@ void Dispatcher::EmitDispatcher() {
 
   ldr(RipReg, STATE_PTR(CpuStateFrame, State.rip));
   // Force a single instruction block if ENTRY_FILL_SRA_SINGLE_INST_REG is nonzero entering the JIT, used for inline SMC handling.
-  cbnz(ARMEmitter::Size::i32Bit, ENTRY_FILL_SRA_SINGLE_INST_REG, &CompileSingleStep);
+  (void)cbnz(ARMEmitter::Size::i32Bit, ENTRY_FILL_SRA_SINGLE_INST_REG, &CompileSingleStep);
 
   // Enter JIT
-  b(&LoopTop);
+  (void)b(&LoopTop);
 
   AbsoluteLoopTopAddressEnterEC = GetCursorAddress<uint64_t>();
   // Load ThreadState and write the target PC there
@@ -130,7 +130,7 @@ void Dispatcher::EmitDispatcher() {
   ldp<ARMEmitter::IndexType::OFFSET>(TMP1, TMP2, REG_CALLRET_SP);
   // EC_CALL_CHECKER_PC_REG is REG_PF which isn't touched by any of the above
   sub(ARMEmitter::Size::i64Bit, TMP1, EC_CALL_CHECKER_PC_REG, TMP1);
-  cbnz(ARMEmitter::Size::i64Bit, TMP1, &LoopTop);
+  (void)cbnz(ARMEmitter::Size::i64Bit, TMP1, &LoopTop);
 
   // If the entry at the TOS is for the target address, pop it and return to the JIT code
   add(ARMEmitter::Size::i64Bit, REG_CALLRET_SP, REG_CALLRET_SP, 0x10);
@@ -142,7 +142,7 @@ void Dispatcher::EmitDispatcher() {
   // We want to ensure that we are 16 byte aligned at the top of this loop
   Align16B();
 
-  Bind(&LoopTop);
+  (void)Bind(&LoopTop);
   AbsoluteLoopTopAddress = GetCursorAddress<uint64_t>();
 
   // Load in our RIP
@@ -169,66 +169,73 @@ void Dispatcher::EmitDispatcher() {
   ldr(TMP2, STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionEC));
   br(TMP2);
 
-  Bind(&l_NotECCode);
+  (void)Bind(&l_NotECCode);
 #endif
 
   ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
-  cbnz(ARMEmitter::Size::i32Bit, TMP1, &CompileSingleStep);
-
-  // This is the block cache lookup routine
-  // It matches what is going on it LookupCache.h::FindBlock
-  ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.L2Pointer));
-
-  // Mask the address by the virtual address size so we can check for aliases
-  uint64_t VirtualMemorySize = CTX->Config.VirtualMemSize;
-  if (std::popcount(VirtualMemorySize) == 1) {
-    and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), VirtualMemorySize - 1);
-  } else {
-    LoadConstant(ARMEmitter::Size::i64Bit, TMP4, VirtualMemorySize);
-    and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), TMP4);
-  }
+  (void)cbnz(ARMEmitter::Size::i32Bit, TMP1, &CompileSingleStep);
 
   ARMEmitter::ForwardLabel NoBlock;
 
-  {
-    // Offset the address and add to our page pointer
-    lsr(ARMEmitter::Size::i64Bit, TMP2, TMP4, 12);
+  if (DisableL2Cache()) {
+    (void)b(&NoBlock);
+  } else {
+    // This is the block cache lookup routine
+    // It matches what is going on it LookupCache.h::FindBlock
+    ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.L2Pointer));
 
-    // Load the pointer from the offset
-    ldr(TMP1, TMP1, TMP2, ARMEmitter::ExtendedType::LSL_64, 3);
+    // Mask the address by the virtual address size so we can check for aliases
+    uint64_t VirtualMemorySize = CTX->Config.VirtualMemSize;
+    if (std::popcount(VirtualMemorySize) == 1) {
+      and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), VirtualMemorySize - 1);
+    } else {
+      LoadConstant(ARMEmitter::Size::i64Bit, TMP4, VirtualMemorySize);
+      and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), TMP4);
+    }
 
-    // If page pointer is zero then we have no block
-    cbz(ARMEmitter::Size::i64Bit, TMP1, &NoBlock);
-
-    // Steal the page offset
-    and_(ARMEmitter::Size::i64Bit, TMP2, TMP4, 0x0FFF);
-
-    // Shift the offset by the size of the block cache entry
-    add(TMP1, TMP1, TMP2, ARMEmitter::ShiftType::LSL, (int)log2(sizeof(FEXCore::LookupCache::LookupCacheEntry)));
-
-    // The the full LookupCacheEntry with a single LDP.
-    // Check the guest address first to ensure it maps to the address we are currently at.
-    // This fixes aliasing problems
-    ldp<ARMEmitter::IndexType::OFFSET>(TMP4, TMP2, TMP1, 0);
-
-    // If the guest address doesn't match, Compile the block.
-    sub(TMP2, TMP2, RipReg);
-    cbnz(ARMEmitter::Size::i64Bit, TMP2, &NoBlock);
-
-    // Check the host address to see if it matches, else compile the block.
-    cbz(ARMEmitter::Size::i64Bit, TMP4, &NoBlock);
-
-    // If we've made it here then we have a real compiled block
     {
-      // update L1 cache
-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
+      // Offset the address and add to our page pointer
+      lsr(ARMEmitter::Size::i64Bit, TMP2, TMP4, 12);
 
-      and_(ARMEmitter::Size::i64Bit, TMP2, RipReg.R(), LookupCache::L1_ENTRIES_MASK);
-      add(TMP1, TMP1, TMP2, ARMEmitter::ShiftType::LSL, 4);
-      stp<ARMEmitter::IndexType::OFFSET>(TMP4, RipReg, TMP1);
+      // Load the pointer from the offset
+      ldr(TMP1, TMP1, TMP2, ARMEmitter::ExtendedType::LSL_64, 3);
 
-      // Jump to the block
-      br(TMP4);
+      // If page pointer is zero then we have no block
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &NoBlock);
+
+      // Steal the page offset
+      and_(ARMEmitter::Size::i64Bit, TMP2, TMP4, 0x0FFF);
+
+      // Shift the offset by the size of the block cache entry
+      add(TMP1, TMP1, TMP2, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(sizeof(LookupCache::LookupCacheEntry)));
+
+      // The the full LookupCacheEntry with a single LDP.
+      // Check the guest address first to ensure it maps to the address we are currently at.
+      // This fixes aliasing problems
+      ldp<ARMEmitter::IndexType::OFFSET>(TMP4, TMP2, TMP1, 0);
+
+      // If the guest address doesn't match, Compile the block.
+      sub(TMP2, TMP2, RipReg);
+      (void)cbnz(ARMEmitter::Size::i64Bit, TMP2, &NoBlock);
+
+      // Check the host address to see if it matches, else compile the block.
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP4, &NoBlock);
+
+      // If we've made it here then we have a real compiled block
+      {
+        // update L1 cache
+        ldp<ARMEmitter::IndexType::OFFSET>(TMP1, TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.L1Pointer));
+
+        // Calculate (tmp1 + ((ripreg & L1_ENTRIES_MASK) << 4)) for the address
+        // L1Mask is pre-shifted.
+        and_(ARMEmitter::Size::i64Bit, TMP2, TMP2, RipReg.R(), ARMEmitter::ShiftType::LSL, FEXCore::ilog2(sizeof(LookupCache::LookupCacheEntry)));
+        add(TMP1, TMP1, TMP2);
+
+        stp<ARMEmitter::IndexType::OFFSET>(TMP4, RipReg, TMP1);
+
+        // Jump to the block
+        br(TMP4);
+      }
     }
   }
 
@@ -304,7 +311,7 @@ void Dispatcher::EmitDispatcher() {
 
   // Need to create the block
   {
-    Bind(&NoBlock);
+    (void)Bind(&NoBlock);
 
     EmitSignalGuardedRegion([&]() {
       SpillStaticRegs(TMP1);
@@ -338,7 +345,7 @@ void Dispatcher::EmitDispatcher() {
   }
 
   {
-    Bind(&CompileSingleStep);
+    (void)Bind(&CompileSingleStep);
 
     EmitSignalGuardedRegion([&]() {
       SpillStaticRegs(TMP1);
@@ -500,7 +507,7 @@ void Dispatcher::EmitDispatcher() {
     stp<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::zr, ARMEmitter::XReg::zr, REG_CALLRET_SP, -0x10);
 
     // Now go back to the regular dispatcher loop
-    b(&LoopTop);
+    (void)b(&LoopTop);
   }
 
   auto EmitLongALUOpHandler = [&](auto R, auto Offset) {
@@ -569,14 +576,15 @@ void Dispatcher::EmitDispatcher() {
     }
   }
 
-  Bind(&l_CTX);
+  (void)Bind(&l_CTX);
   dc64(reinterpret_cast<uintptr_t>(CTX));
-  Bind(&l_Sleep);
+  (void)Bind(&l_Sleep);
   dc64(reinterpret_cast<uint64_t>(SleepThread));
-  Bind(&l_CompileBlock);
+  (void)Bind(&l_CompileBlock);
   FEXCore::Utils::MemberFunctionToPointerCast PMFCompileBlock(&FEXCore::Context::ContextImpl::CompileBlock);
   dc64(PMFCompileBlock.GetConvertedPointer());
-  Bind(&l_CompileSingleStep);
+  (void)Bind(&l_CompileSingleStep);
+
   FEXCore::Utils::MemberFunctionToPointerCast PMFCompileSingleStep(&FEXCore::Context::ContextImpl::CompileSingleStep);
   dc64(PMFCompileSingleStep.GetConvertedPointer());
 

FEXCore/Source/Interface/Core/Dispatcher/Dispatcher.h

@@ -4,6 +4,7 @@
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 
+#include <FEXCore/Config/Config.h>
 #include <FEXCore/fextl/memory.h>
 
 #include <array>
@@ -92,6 +93,8 @@ private:
 
   void EmitDispatcher();
   uint64_t GenerateABICall(FallbackABI ABI);
+
+  FEX_CONFIG_OPT(DisableL2Cache, DISABLEL2CACHE);
 };
 
 } // namespace FEXCore::CPU

FEXCore/Source/Interface/Core/Frontend.cpp

@@ -1047,8 +1047,11 @@ Decoder::DecodedBlockStatus Decoder::DecodeInstruction(uint64_t PC) {
     // Put an invalid instruction in the stream so the core can raise SIGILL if hit
     // Error while decoding instruction. We don't know the table or instruction size
     DecodeInst->TableInfo = nullptr;
+    auto Result = ErrorDuringDecoding  ? DecodedBlockStatus::INVALID_INST :
+                  DecodeInst->InstSize ? DecodedBlockStatus::PARTIAL_DECODE_INST :
+                                         DecodedBlockStatus::NOEXEC_INST;
     DecodeInst->InstSize = 0;
-    return ErrorDuringDecoding ? DecodedBlockStatus::INVALID_INST : DecodedBlockStatus::NOEXEC_INST;
+    return Result;
   } else if (!DecodeInst->TableInfo || (DecodeInst->TableInfo->Type == TYPE_INST && !DecodeInst->TableInfo->OpcodeDispatcher.OpDispatch)) {
     // If there wasn't an error during decoding but we have no dispatcher for the instruction then claim invalid instruction.
     return DecodedBlockStatus::INVALID_INST;
@@ -1450,7 +1453,10 @@ void Decoder::DecodeInstructionsAtEntry(FEXCore::Core::InternalThreadState* Thre
           EraseBlock = true;
         } else {
           LogMan::Msg::EFmt("{} instruction in entry block: {:X}",
-                            BlockIt->BlockStatus == DecodedBlockStatus::INVALID_INST ? "Invalid" : "NoExec", OpAddress);
+                            BlockIt->BlockStatus == DecodedBlockStatus::INVALID_INST ? "Invalid" :
+                            BlockIt->BlockStatus == DecodedBlockStatus::NOEXEC_INST  ? "NoExec" :
+                                                                                       "PartialDecode",
+                            OpAddress);
         }
         break;
       }

FEXCore/Source/Interface/Core/Frontend.h

@@ -27,6 +27,7 @@ public:
     SUCCESS,
     INVALID_INST,
     NOEXEC_INST,
+    PARTIAL_DECODE_INST,
   };
 
   // New Frontend decoding

FEXCore/Source/Interface/Core/Interpreter/Fallbacks/F80Fallbacks.h

@@ -2,13 +2,11 @@
 #pragma once
 #include "Common/SoftFloat.h"
 
-#include "Interface/Context/Context.h"
 #include "Interface/Core/Interpreter/Fallbacks/FallbackOpHandler.h"
 #include "Interface/IR/IR.h"
 
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/SHMStats.h>
-#include <FEXCore/Config/Config.h>
 
 namespace FEXCore::CPU {
 FEXCORE_PRESERVE_ALL_ATTR static softfloat_state SoftFloatStateFromFCW(uint16_t FCW, bool Force80BitPrecision = false) {
@@ -79,12 +77,6 @@ struct OpHandlers<IR::OP_F80CVTTO> {
   FEXCORE_PRESERVE_ALL_ATTR static VectorRegType handle8(uint16_t FCW, double src, FEXCore::Core::CpuStateFrame* Frame) {
     FEXCORE_PROFILE_INSTANT_INCREMENT(Frame->Thread, AccumulatedFloatFallbackCount, 1);
     ScopedSoftFloatState State {FCW, Frame};
-    auto Context = static_cast<Context::ContextImpl*>(Frame->Thread->CTX);
-    auto ReducedPrecisionMode = Context->Config.x87ReducedPrecision;
-    auto StrictReducedPrecisionMode = Context->Config.x87StrictReducedPrecision;
-    if (!ReducedPrecisionMode || StrictReducedPrecisionMode) {
-      return X80SoftFloat::FromF64_PreserveNaN(&State.State, src);
-    }
     return X80SoftFloat(&State.State, src);
   }
 };
@@ -123,12 +115,6 @@ struct OpHandlers<IR::OP_F80CVT> {
   FEXCORE_PRESERVE_ALL_ATTR static double handle8(uint16_t FCW, VectorRegType src, FEXCore::Core::CpuStateFrame* Frame) {
     FEXCORE_PROFILE_INSTANT_INCREMENT(Frame->Thread, AccumulatedFloatFallbackCount, 1);
     ScopedSoftFloatState State {FCW, Frame};
-    auto Context = static_cast<Context::ContextImpl*>(Frame->Thread->CTX);
-    auto ReducedPrecisionMode = Context->Config.x87ReducedPrecision;
-    auto StrictReducedPrecisionMode = Context->Config.x87StrictReducedPrecision;
-    if (!ReducedPrecisionMode || StrictReducedPrecisionMode) {
-      return X80SoftFloat(src).ToF64_PreserveNan(&State.State);
-    }
     return X80SoftFloat(src).ToF64(&State.State);
   }
 };

FEXCore/Source/Interface/Core/JIT/ALUOps.cpp

@@ -588,7 +588,7 @@ DEF_OP(ShiftFlags) {
   and_(ARMEmitter::Size::i32Bit, TMP1, Src2, OpSize == IR::OpSize::i64Bit ? 0x3f : 0x1f);
 
   ARMEmitter::ForwardLabel Done;
-  cbz(EmitSize, TMP1, &Done);
+  (void)cbz(EmitSize, TMP1, &Done);
   {
     // PF/SF/ZF/OF
     if (OpSize >= IR::OpSize::i32Bit) {
@@ -652,7 +652,7 @@ DEF_OP(ShiftFlags) {
       msr(ARMEmitter::SystemRegister::NZCV, TMP2);
     }
   }
-  Bind(&Done);
+  (void)Bind(&Done);
 
   // TODO: Make RA less dumb so this can't happen (e.g. with late-kill).
   if (PFOutput != PFTemp) {
@@ -669,7 +669,7 @@ DEF_OP(RotateFlags) {
 
   // If shift=0, flags are unaffected. Wrap the whole implementation in a cbz.
   ARMEmitter::ForwardLabel Done;
-  cbz(EmitSize, Shift, &Done);
+  (void)cbz(EmitSize, Shift, &Done);
   {
     // Extract the last bit shifted in to CF
     const auto BitSize = IR::OpSizeToSize(Op->Size) * 8;
@@ -701,7 +701,7 @@ DEF_OP(RotateFlags) {
       msr(ARMEmitter::SystemRegister::NZCV, TMP3);
     }
   }
-  Bind(&Done);
+  (void)Bind(&Done);
 }
 
 DEF_OP(Extr) {
@@ -767,14 +767,14 @@ DEF_OP(PDep) {
     // Now, they're copied, so we can start setting Dest (even if it overlaps with
     // one of them).  Handle early exit case
     mov(EmitSize, Dest, 0);
-    cbz(EmitSize, OrigMask, &Done);
+    (void)cbz(EmitSize, OrigMask, &Done);
 
     // Setup for first iteration
     neg(EmitSize, T0, Mask);
     and_(EmitSize, T0, T0, Mask);
 
     // Main loop
-    Bind(&NextBit);
+    (void)Bind(&NextBit);
     sbfx(EmitSize, T1, Input, 0, 1);
     eor(EmitSize, Mask, Mask, T0);
     and_(EmitSize, T0, T1, T0);
@@ -782,10 +782,10 @@ DEF_OP(PDep) {
     orr(EmitSize, Dest, Dest, T0);
     lsr(EmitSize, Input, Input, 1);
     and_(EmitSize, T0, Mask, T1);
-    cbnz(EmitSize, T0, &NextBit);
+    (void)cbnz(EmitSize, T0, &NextBit);
 
     // All done with nothing to do.
-    Bind(&Done);
+    (void)Bind(&Done);
   }
 }
 
@@ -821,27 +821,27 @@ DEF_OP(PExt) {
     ARMEmitter::BackwardLabel NextBit;
     ARMEmitter::ForwardLabel Done;
 
-    cbz(EmitSize, Mask, &EarlyExit);
+    (void)cbz(EmitSize, Mask, &EarlyExit);
     mov(EmitSize, MaskReg, Mask);
     mov(EmitSize, ValueReg, Input);
     mov(EmitSize, Dest, ARMEmitter::Reg::zr);
 
     // Main loop
-    Bind(&NextBit);
-    cbz(EmitSize, MaskReg, &Done);
+    (void)Bind(&NextBit);
+    (void)cbz(EmitSize, MaskReg, &Done);
     clz(EmitSize, BitReg, MaskReg);
     lslv(EmitSize, ValueReg, ValueReg, BitReg);
     lslv(EmitSize, MaskReg, MaskReg, BitReg);
     extr(EmitSize, Dest, Dest, ValueReg, OpSizeBitsM1);
     bfc(EmitSize, MaskReg, OpSizeBitsM1, 1);
-    b(&NextBit);
+    (void)b(&NextBit);
 
     // Early exit
-    Bind(&EarlyExit);
+    (void)Bind(&EarlyExit);
     mov(EmitSize, Dest, ARMEmitter::Reg::zr);
 
     // All done with nothing to do.
-    Bind(&Done);
+    (void)Bind(&Done);
   }
 }
 
@@ -909,7 +909,7 @@ DEF_OP(Div) {
     eor(EmitSize, TMP1, TMP1, Upper);
 
     // If the sign bit matches then the result is zero
-    cbz(EmitSize, TMP1, &Only64Bit);
+    (void)cbz(EmitSize, TMP1, &Only64Bit);
 
     // Long divide
     {
@@ -928,17 +928,17 @@ DEF_OP(Div) {
       mov(EmitSize, Remainder, TMP2);
 
       // Skip 64-bit path
-      b(&LongDIVRet);
+      (void)b(&LongDIVRet);
     }
 
-    Bind(&Only64Bit);
+    (void)Bind(&Only64Bit);
     // 64-Bit only
     {
       sdiv(EmitSize, Quotient, Lower, Divisor);
       msub(EmitSize, Remainder, Quotient, Divisor, Lower);
     }
 
-    Bind(&LongDIVRet);
+    (void)Bind(&LongDIVRet);
     break;
   }
   default: LOGMAN_MSG_A_FMT("Unknown DIV Size: {}", OpSize); break;
@@ -992,7 +992,7 @@ DEF_OP(UDiv) {
 
     // Check the upper bits for zero
     // If the upper bits are zero then we can do a 64-bit divide
-    cbz(EmitSize, Upper, &Only64Bit);
+    (void)cbz(EmitSize, Upper, &Only64Bit);
 
     // Long divide
     {
@@ -1011,17 +1011,17 @@ DEF_OP(UDiv) {
       mov(EmitSize, Remainder, TMP2);
 
       // Skip 64-bit path
-      b(&LongDIVRet);
+      (void)b(&LongDIVRet);
     }
 
-    Bind(&Only64Bit);
+    (void)Bind(&Only64Bit);
     // 64-Bit only
     {
       udiv(EmitSize, Quotient, Lower, Divisor);
       msub(EmitSize, Remainder, Quotient, Divisor, Lower);
     }
 
-    Bind(&LongDIVRet);
+    (void)Bind(&LongDIVRet);
     break;
   }
   default: LOGMAN_MSG_A_FMT("Unknown LUDIV Size: {}", OpSize); break;

FEXCore/Source/Interface/Core/JIT/Arm64Relocations.cpp

@@ -63,7 +63,7 @@ void Arm64JITCore::PlaceNamedSymbolLiteral(NamedSymbolLiteralPair& Lit) {
   auto CurrentCursor = GetCursorAddress<uint8_t*>();
   Lit.MoveABI.NamedSymbolLiteral.Offset = CurrentCursor - CodeData.BlockBegin;
 
-  Bind(&Lit.Loc);
+  BindOrRestart(&Lit.Loc);
   dc64(Lit.Lit);
   Relocations.emplace_back(Lit.MoveABI);
 }

FEXCore/Source/Interface/Core/JIT/AtomicOps.cpp

@@ -62,27 +62,27 @@ DEF_OP(CASPair) {
     ARMEmitter::BackwardLabel LoopTop;
     ARMEmitter::ForwardLabel LoopNotExpected;
     ARMEmitter::ForwardLabel LoopExpected;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
 
     // This instruction sequence must be synced with HandleCASPAL_Armv8.
     ldaxp(EmitSize, TMP2, TMP3, MemSrc);
     cmp(EmitSize, TMP2, Expected0);
     ccmp(EmitSize, TMP3, Expected1, ARMEmitter::StatusFlags::None, ARMEmitter::Condition::CC_EQ);
-    b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
+    (void)b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
     stlxp(EmitSize, TMP2, Desired0, Desired1, MemSrc);
-    cbnz(EmitSize, TMP2, &LoopTop);
+    (void)cbnz(EmitSize, TMP2, &LoopTop);
     mov(EmitSize, Dst0, Expected0);
     mov(EmitSize, Dst1, Expected1);
 
-    b(&LoopExpected);
+    (void)b(&LoopExpected);
 
-    Bind(&LoopNotExpected);
+    (void)Bind(&LoopNotExpected);
     mov(EmitSize, Dst0, TMP2.R());
     mov(EmitSize, Dst1, TMP3.R());
     // exclusive monitor needs to be cleared here
     // Might have hit the case where ldaxr was hit but stlxr wasn't
     clrex();
-    Bind(&LoopExpected);
+    (void)Bind(&LoopExpected);
 
     // Restore
     msr(ARMEmitter::SystemRegister::NZCV, TMP1);
@@ -114,7 +114,7 @@ DEF_OP(CAS) {
     ARMEmitter::BackwardLabel LoopTop;
     ARMEmitter::ForwardLabel LoopNotExpected;
     ARMEmitter::ForwardLabel LoopExpected;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     if (IROp->Size == IR::OpSize::i8Bit) {
       cmp(EmitSize, TMP2, Expected, ARMEmitter::ExtendedType::UXTB, 0);
@@ -123,18 +123,18 @@ DEF_OP(CAS) {
     } else {
       cmp(EmitSize, TMP2, Expected);
     }
-    b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
+    (void)b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
     stlxr(SubEmitSize, TMP3, Desired, MemSrc);
-    cbnz(EmitSize, TMP3, &LoopTop);
+    (void)cbnz(EmitSize, TMP3, &LoopTop);
     mov(EmitSize, Dst, Expected);
-    b(&LoopExpected);
+    (void)b(&LoopExpected);
 
-    Bind(&LoopNotExpected);
+    (void)Bind(&LoopNotExpected);
     mov(EmitSize, Dst, TMP2.R());
     // exclusive monitor needs to be cleared here
     // Might have hit the case where ldaxr was hit but stlxr wasn't
     clrex();
-    Bind(&LoopExpected);
+    (void)Bind(&LoopExpected);
   }
 }
 
@@ -150,11 +150,11 @@ DEF_OP(AtomicXor) {
     steorl(SubEmitSize, Src, MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     eor(EmitSize, TMP2, TMP2, Src);
     stlxr(SubEmitSize, TMP2, TMP2, MemSrc);
-    cbnz(EmitSize, TMP2, &LoopTop);
+    (void)cbnz(EmitSize, TMP2, &LoopTop);
   }
 }
 
@@ -179,10 +179,10 @@ DEF_OP(AtomicSwap) {
     ldswpal(SubEmitSize, Src, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     stlxr(SubEmitSize, TMP4, Src, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
     ubfm(EmitSize, GetReg(Node), TMP2, 0, IR::OpSizeAsBits(OpSize) - 1);
   }
 }
@@ -199,11 +199,11 @@ DEF_OP(AtomicFetchAdd) {
     ldaddal(SubEmitSize, Src, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     add(EmitSize, TMP3, TMP2, Src);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -221,11 +221,11 @@ DEF_OP(AtomicFetchSub) {
     ldaddal(SubEmitSize, TMP2, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     sub(EmitSize, TMP3, TMP2, Src);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -243,11 +243,11 @@ DEF_OP(AtomicFetchAnd) {
     ldclral(SubEmitSize, TMP2, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     and_(EmitSize, TMP3, TMP2, Src);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -264,11 +264,11 @@ DEF_OP(AtomicFetchCLR) {
     ldclral(SubEmitSize, Src, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     bic(EmitSize, TMP3, TMP2, Src);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -285,11 +285,11 @@ DEF_OP(AtomicFetchOr) {
     ldsetal(SubEmitSize, Src, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     orr(EmitSize, TMP3, TMP2, Src);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -306,11 +306,11 @@ DEF_OP(AtomicFetchXor) {
     ldeoral(SubEmitSize, Src, GetReg(Node), MemSrc);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     eor(EmitSize, TMP3, TMP2, Src);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -326,20 +326,20 @@ DEF_OP(AtomicFetchNeg) {
     // Use a CAS loop to avoid needing to emulate unaligned LLSC atomics
     ldr(SubEmitSize, TMP2, MemSrc);
     ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
     mov(EmitSize, TMP4, TMP2);
     neg(EmitSize, TMP3, TMP2);
     casal(SubEmitSize, TMP2, TMP3, MemSrc);
     sub(EmitSize, TMP3, TMP2, TMP4);
-    cbnz(EmitSize, TMP3, &LoopTop);
+    (void)cbnz(EmitSize, TMP3, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
     ldaxr(SubEmitSize, TMP2, MemSrc);
     neg(EmitSize, TMP3, TMP2);
     stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
     mov(EmitSize, GetReg(Node), TMP2.R());
   }
 }
@@ -359,11 +359,11 @@ DEF_OP(TelemetrySetValue) {
     stsetl(ARMEmitter::SubRegSize::i64Bit, TMP1, TMP2);
   } else {
     ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
     ldaxr(ARMEmitter::SubRegSize::i64Bit, TMP3, TMP2);
     orr(ARMEmitter::Size::i32Bit, TMP3, TMP3, Src);
     stlxr(ARMEmitter::SubRegSize::i64Bit, TMP3, TMP3, TMP2);
-    cbnz(ARMEmitter::Size::i32Bit, TMP3, &LoopTop);
+    (void)cbnz(ARMEmitter::Size::i32Bit, TMP3, &LoopTop);
   }
 #endif
 }

FEXCore/Source/Interface/Core/JIT/BranchOps.cpp

@@ -141,7 +141,7 @@ DEF_OP(ExitFunction) {
         if (!Op->CallReturnBlock.IsInvalid()) {
           auto CallReturnAddressReg = GetReg(Op->CallReturnAddress).X();
           PendingCallReturnTargetLabel = &CallReturnTargets.try_emplace(Op->CallReturnBlock.ID()).first->second;
-          adr(TMP1, &l_CallReturn);
+          (void)adr(TMP1, &l_CallReturn);
           stp<ARMEmitter::IndexType::PRE>(CallReturnAddressReg, TMP1, REG_CALLRET_SP, -0x10);
         } else {
           stp<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::zr, ARMEmitter::XReg::zr, REG_CALLRET_SP, -0x10);
@@ -149,16 +149,16 @@ DEF_OP(ExitFunction) {
       } else if (Op->Hint == IR::BranchHint::CheckTF) {
         ARMEmitter::ForwardLabel TFUnset;
         ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
-        cbz(ARMEmitter::Size::i32Bit, TMP1, &TFUnset);
+        (void)cbz(ARMEmitter::Size::i32Bit, TMP1, &TFUnset);
         LoadConstant(ARMEmitter::Size::i64Bit, TMP1, NewRIP);
         str(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip));
         ldr(TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop));
         blr(TMP2);
-        Bind(&TFUnset);
+        (void)Bind(&TFUnset);
       }
 
       EmitLinkedBranch(NewRIP, Op->Hint == IR::BranchHint::Call);
-      Bind(&l_CallReturn);
+      (void)Bind(&l_CallReturn);
 #ifdef _M_ARM_64EC
     }
 #endif
@@ -170,40 +170,38 @@ DEF_OP(ExitFunction) {
       // First try to pop from the call-ret stack, otherwise follow the normal path (but ending in a ret)
       ldp<ARMEmitter::IndexType::POST>(TMP1, TMP2, REG_CALLRET_SP, 0x10);
       sub(TMP1, TMP1, RipReg.X());
-      cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
     }
 
     // L1 Cache
-    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.L1Pointer));
+    ldp<ARMEmitter::IndexType::OFFSET>(TMP1, TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.L1Pointer));
 
     // Calculate (tmp1 + ((ripreg & L1_ENTRIES_MASK) << 4)) for the address
-    // arithmetic. ubfiz+add is marginally faster on Firestorm than
-    // and+add(shift). Same performance on Cortex.
-    static_assert(LookupCache::L1_ENTRIES_MASK == ((1u << 20) - 1));
-    ubfiz(ARMEmitter::Size::i64Bit, TMP4, RipReg, 4, 20);
-    add(TMP1, TMP1, TMP4);
+    // L1Mask is pre-shifted.
+    and_(ARMEmitter::Size::i64Bit, TMP2, TMP2, RipReg, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(sizeof(LookupCache::LookupCacheEntry)));
+    add(TMP1, TMP1, TMP2);
 
     ldp<ARMEmitter::IndexType::OFFSET>(TMP2, TMP1, TMP1, 0);
 
     // Note: sub+cbnz used over cmp+br to preserve flags.
     sub(TMP1, TMP1, RipReg.X());
-    cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
+    (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
     ldr(TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop));
     str(RipReg.X(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip));
 
-    Bind(&SkipFullLookup);
+    (void)Bind(&SkipFullLookup);
     if (Op->Hint == IR::BranchHint::Call) {
       ARMEmitter::ForwardLabel l_CallReturn;
       if (!Op->CallReturnBlock.IsInvalid()) {
         auto CallReturnAddressReg = GetReg(Op->CallReturnAddress).X();
         PendingCallReturnTargetLabel = &CallReturnTargets.try_emplace(Op->CallReturnBlock.ID()).first->second;
-        adr(TMP1, &l_CallReturn);
+        (void)adr(TMP1, &l_CallReturn);
         stp<ARMEmitter::IndexType::PRE>(CallReturnAddressReg, TMP1, REG_CALLRET_SP, -0x10);
       } else {
         stp<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::zr, ARMEmitter::XReg::zr, REG_CALLRET_SP, -0x10);
       }
       blr(TMP2);
-      Bind(&l_CallReturn);
+      (void)Bind(&l_CallReturn);
     } else if (Op->Hint == IR::BranchHint::Return) {
       ret(TMP2);
     } else {
@@ -224,7 +222,7 @@ DEF_OP(CondJump) {
   auto TrueTargetLabel = JumpTarget(Op->TrueBlock);
 
   if (Op->FromNZCV) {
-    b(MapCC(Op->Cond), TrueTargetLabel);
+    b_OrRestart(MapCC(Op->Cond), TrueTargetLabel);
   } else {
     uint64_t Const;
     const bool isConst = IsInlineConstant(Op->Cmp2, &Const);
@@ -237,16 +235,16 @@ DEF_OP(CondJump) {
 
     if (Op->Cond == IR::CondClass::EQ) {
       LOGMAN_THROW_A_FMT(Const == 0, "CondJump: Expected 0 source");
-      cbz(Size, Reg, TrueTargetLabel);
+      cbz_OrRestart(Size, Reg, TrueTargetLabel);
     } else if (Op->Cond == IR::CondClass::NEQ) {
       LOGMAN_THROW_A_FMT(Const == 0, "CondJump: Expected 0 source");
-      cbnz(Size, Reg, TrueTargetLabel);
+      cbnz_OrRestart(Size, Reg, TrueTargetLabel);
     } else if (Op->Cond == IR::CondClass::TSTZ) {
       LOGMAN_THROW_A_FMT(Const < 64, "CondJump: Expected valid bit source");
-      tbz(Reg, Const, TrueTargetLabel);
+      tbz_OrRestart(Reg, Const, TrueTargetLabel);
     } else if (Op->Cond == IR::CondClass::TSTNZ) {
       LOGMAN_THROW_A_FMT(Const < 64, "CondJump: Expected valid bit source");
-      tbnz(Reg, Const, TrueTargetLabel);
+      tbnz_OrRestart(Reg, Const, TrueTargetLabel);
     } else {
       LOGMAN_THROW_A_FMT(false, "CondJump expected simple condition");
     }
@@ -262,16 +260,10 @@ DEF_OP(Syscall) {
   // X1: ThreadState
   // X2: Pointer to SyscallArguments
 
-  FEXCore::IR::SyscallFlags Flags = Op->Flags;
   PushDynamicRegs(TMP1);
 
   uint32_t GPRSpillMask = ~0U;
   uint32_t FPRSpillMask = ~0U;
-  if ((Flags & FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) == FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) {
-    // Need to spill all caller saved registers still
-    GPRSpillMask = CALLER_GPR_MASK;
-    FPRSpillMask = CALLER_FPR_MASK;
-  }
 
   SpillStaticRegs(TMP1, true, GPRSpillMask, FPRSpillMask);
 
@@ -305,117 +297,22 @@ DEF_OP(Syscall) {
 
   add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, SPOffset);
 
-  if ((Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
-    // Result is now in x0
-    // Fix the stack and any values that were stepped on
-    FillStaticRegs(true, GPRSpillMask, FPRSpillMask, ARMEmitter::Reg::r1, ARMEmitter::Reg::r2);
+  // Result is now in x0
+  // Fix the stack and any values that were stepped on
+  FillStaticRegs(true, GPRSpillMask, FPRSpillMask, ARMEmitter::Reg::r1, ARMEmitter::Reg::r2);
 
-    // Now the registers we've spilled are back in their original host registers
-    // We can safely claim we are no longer in a syscall
-    str(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
+  // Now the registers we've spilled are back in their original host registers
+  // We can safely claim we are no longer in a syscall
+  str(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
 
-    PopDynamicRegs();
+  PopDynamicRegs();
 
-    if ((Flags & FEXCore::IR::SyscallFlags::NORETURNEDRESULT) != FEXCore::IR::SyscallFlags::NORETURNEDRESULT) {
-      // Move result to its destination register.
-      // Only if `NORETURNEDRESULT` wasn't set, otherwise we might overwrite the CPUState refilled with `FillStaticRegs`
-      mov(ARMEmitter::Size::i64Bit, GetReg(Node), ARMEmitter::Reg::r0);
-    }
-  }
-}
+  const auto OSABI = CTX->SyscallHandler->GetOSABI();
 
-DEF_OP(InlineSyscall) {
-  auto Op = IROp->C<IR::IROp_InlineSyscall>();
-  // Arguments are passed as follows:
-  // X8: SyscallNumber - RA INTERSECT
-  // X0: Arg0 & Return
-  // X1: Arg1
-  // X2: Arg2
-  // X3: Arg3
-  // X4: Arg4 - RA INTERSECT
-  // X5: Arg5 - RA INTERSECT
-  // X6: Arg6 - Doesn't exist in x86-64 land. RA INTERSECT
-
-  // One argument is removed from the SyscallArguments::MAX_ARGS since the first argument was syscall number
-  const static std::array<ARMEmitter::XRegister, FEXCore::HLE::SyscallArguments::MAX_ARGS - 1> RegArgs = {
-    {ARMEmitter::XReg::x0, ARMEmitter::XReg::x1, ARMEmitter::XReg::x2, ARMEmitter::XReg::x3, ARMEmitter::XReg::x4, ARMEmitter::XReg::x5}};
-
-  bool Intersects {};
-  // We always need to spill x8 since we can't know if it is live at this SSA location
-  uint32_t SpillMask = 1U << 8;
-  for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS - 1; ++i) {
-    if (Op->Header.Args[i].IsInvalid()) {
-      break;
-    }
-
-    auto Reg = GetReg(Op->Header.Args[i]);
-    if (Reg == ARMEmitter::Reg::r8 || Reg == ARMEmitter::Reg::r4 || Reg == ARMEmitter::Reg::r5) {
-
-      SpillMask |= (1U << Reg.Idx());
-      Intersects = true;
-    }
-  }
-
-  // Ordering is incredibly important here
-  // We must spill any overlapping registers first THEN claim we are in a syscall without invalidating state at all
-  // Only spill the registers that intersect with our usage
-  SpillStaticRegs(TMP1, false, SpillMask);
-
-  // Now that we are spilled, store in the state that we are in a syscall
-  // Still without overwriting registers that matter
-  // 16bit LoadConstant to be a single instruction
-  // We must always spill at least one register (x8) so this value always has a bit set
-  // This gives the signal handler a value to check to see if we are in a syscall at all
-  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, SpillMask & 0xFFFF);
-  str(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
-
-  // Now that we have claimed to be a syscall we can set up the arguments
-  const auto EmitSize = CTX->Config.Is64BitMode() ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
-  const auto EmitSubSize = CTX->Config.Is64BitMode() ? ARMEmitter::SubRegSize::i64Bit : ARMEmitter::SubRegSize::i32Bit;
-  if (Intersects) {
-    for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS - 1; ++i) {
-      if (Op->Header.Args[i].IsInvalid()) {
-        break;
-      }
-
-      auto Reg = GetReg(Op->Header.Args[i]);
-      if (SpillMask & (1U << Reg.Idx())) {
-        // In the case of intersection with x4, x5, or x8 then these are currently SRA
-        // for registers RAX, RDX, and RSP. Which have just been spilled
-        // Just load back from the context.
-        auto Correlation = GetX86RegRelationToARMReg(Reg);
-        LOGMAN_THROW_A_FMT(Correlation != X86State::REG_INVALID, "Invalid register mapping");
-        ldr(EmitSubSize, RegArgs[i].R(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[Correlation]));
-      } else {
-        mov(EmitSize, RegArgs[i].R(), Reg);
-      }
-    }
-  } else {
-    for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS - 1; ++i) {
-      if (Op->Header.Args[i].IsInvalid()) {
-        break;
-      }
-
-      mov(EmitSize, RegArgs[i].R(), GetReg(Op->Header.Args[i]));
-    }
-  }
-
-  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r8, Op->HostSyscallNumber);
-  svc(0);
-  // On updated signal mask we can receive a signal RIGHT HERE
-
-  if ((Op->Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
-    // Now that we are done in the syscall we need to carefully peel back the state
-    // First unspill the registers from before
-    FillStaticRegs(false, SpillMask, ~0U, ARMEmitter::Reg::r8, ARMEmitter::Reg::r1);
-
-    // Now the registers we've spilled are back in their original host registers
-    // We can safely claim we are no longer in a syscall
-    str(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
-
-    // Result is now in x0
-    // Move result to its destination register
-    mov(EmitSize, GetReg(Node), ARMEmitter::Reg::r0);
+  if (OSABI != FEXCore::HLE::SyscallOSABI::OS_GENERIC) {
+    // Move result to its destination register.
+    // Only if `NORETURNEDRESULT` wasn't set, otherwise we might overwrite the CPUState refilled with `FillStaticRegs`
+    mov(ARMEmitter::Size::i64Bit, GetReg(Node), ARMEmitter::Reg::r0);
   }
 }
 
@@ -458,7 +355,7 @@ DEF_OP(ValidateCode) {
     while (len >= Size) {
       LoadData();
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, TMP2);
-      cbnz(ARMEmitter::Size::i64Bit, TMP1, &Fail);
+      cbnz_OrRestart(ARMEmitter::Size::i64Bit, TMP1, &Fail);
       len -= Size;
       Offset += Size;
     }
@@ -486,10 +383,10 @@ DEF_OP(ValidateCode) {
 
   ARMEmitter::ForwardLabel End;
   LoadConstant(ARMEmitter::Size::i32Bit, Dst, 0);
-  b(&End);
-  Bind(&Fail);
+  b_OrRestart(&End);
+  BindOrRestart(&Fail);
   LoadConstant(ARMEmitter::Size::i32Bit, Dst, 1);
-  Bind(&End);
+  BindOrRestart(&End);
 }
 
 DEF_OP(ThreadRemoveCodeEntry) {

FEXCore/Source/Interface/Core/JIT/JIT.cpp

@@ -11,8 +11,6 @@ desc: Main glue logic of the arm64 splatter backend
 $end_info$
 */
 
-#include "Common/SoftFloat.h"
-
 #include "Interface/Context/Context.h"
 #include "Interface/Core/LookupCache.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
@@ -30,6 +28,7 @@ $end_info$
 #include <FEXCore/Utils/CompilerDefs.h>
 #include <FEXCore/Utils/EnumUtils.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/LongJump.h>
 #include <FEXCore/Utils/Profiler.h>
 #include <FEXCore/Utils/Telemetry.h>
 #include <FEXCore/Utils/TypeDefines.h>
@@ -37,7 +36,6 @@ $end_info$
 
 #include <cstdio>
 #include <cstring>
-#include <limits>
 #include <unistd.h>
 
 namespace {
@@ -540,7 +538,7 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
       // Guard the LookupCache lock with the code invalidation mutex, to avoid issues with forking
       auto lk_inval =
         GuardSignalDeferringSection<std::shared_lock>(static_cast<Context::ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
-      HostCode = Thread->LookupCache->FindBlock(GuestRip);
+      HostCode = Thread->LookupCache->FindBlock(Thread, GuestRip);
     }
     if (!HostCode) {
       // Hold a reference to the code buffer, to avoid linking unmapped code if compilation triggers a recreation.
@@ -565,7 +563,7 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
   auto lk_inval = GuardSignalDeferringSection<std::shared_lock>(static_cast<Context::ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
 
   // Lock here is necessary to prevent simultaneous linking and delinking
-  auto lk = Thread->LookupCache->AcquireLock();
+  auto lk = Thread->LookupCache->AcquireWriteLock();
 
   // For non-calls, this would extend into the block's code, however that's fine as an out-of-range adr would never
   // be generated avoiding any false positives.
@@ -578,14 +576,17 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
 
     if (KnownCallMarkerInst == ExpectedKnownCallMarkerInst) {
       BranchEmit.bl(BranchOffset);
-      Thread->LookupCache->AddBlockLink(GuestRip, Record, [](FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record) {
-        DirectBlockDelinker(Frame, Record, true);
-      });
+      Thread->LookupCache->AddBlockLink(
+        GuestRip, Record,
+        [](FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record) { DirectBlockDelinker(Frame, Record, true); }, lk);
     } else {
       BranchEmit.b(BranchOffset);
-      Thread->LookupCache->AddBlockLink(GuestRip, Record, [](FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record) {
-        DirectBlockDelinker(Frame, Record, false);
-      });
+      Thread->LookupCache->AddBlockLink(
+        GuestRip, Record,
+        [](FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record) {
+          DirectBlockDelinker(Frame, Record, false);
+        },
+        lk);
     }
 
     std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(CallerAddress)).store(BranchInst, std::memory_order::relaxed);
@@ -604,7 +605,7 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
     std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(JumpThunkStartAddress)).store(LdrInst, std::memory_order::relaxed);
     ARMEmitter::Emitter::ClearICache(reinterpret_cast<void*>(JumpThunkStartAddress), 4);
 
-    Thread->LookupCache->AddBlockLink(GuestRip, Record, IndirectBlockDelinker);
+    Thread->LookupCache->AddBlockLink(GuestRip, Record, IndirectBlockDelinker, lk);
   }
 
   return HostCode;
@@ -669,15 +670,6 @@ Arm64JITCore::Arm64JITCore(FEXCore::Context::ContextImpl* ctx, FEXCore::Core::In
 
   CurrentCodeBuffer = CodeBuffers.GetLatest();
   ThreadState->LookupCache->Shared = CurrentCodeBuffer->LookupCache.get();
-
-  // Setup dynamic dispatch.
-  if (ParanoidTSO()) {
-    RT_LoadMemTSO = &Arm64JITCore::Op_ParanoidLoadMemTSO;
-    RT_StoreMemTSO = &Arm64JITCore::Op_ParanoidStoreMemTSO;
-  } else {
-    RT_LoadMemTSO = &Arm64JITCore::Op_LoadMemTSO;
-    RT_StoreMemTSO = &Arm64JITCore::Op_StoreMemTSO;
-  }
 }
 
 void Arm64JITCore::EmitDetectionString() {
@@ -689,13 +681,13 @@ void Arm64JITCore::EmitDetectionString() {
 void Arm64JITCore::ClearCache() {
   // NOTE: Holding on to the reference here is required to ensure validity of the WriteLock mutex
   auto PrevCodeBuffer = CurrentCodeBuffer;
-  std::lock_guard lk(PrevCodeBuffer->LookupCache->WriteLock);
+  auto lk = PrevCodeBuffer->LookupCache->AcquireWriteLock();
 
   auto CodeBuffer = GetEmptyCodeBuffer();
   SetBuffer(CodeBuffer->Ptr, CodeBuffer->Size);
   EmitDetectionString();
 
-  ThreadState->LookupCache->ChangeGuestToHostMapping(*PrevCodeBuffer, *CurrentCodeBuffer->LookupCache);
+  ThreadState->LookupCache->ChangeGuestToHostMapping(*PrevCodeBuffer, *CurrentCodeBuffer->LookupCache, lk);
 }
 
 Arm64JITCore::~Arm64JITCore() {}
@@ -748,11 +740,11 @@ void Arm64JITCore::EmitTFCheck() {
   // Note that this needs to be before the below suspend checks, as X86 checks this flag immediately after executing an instruction.
   ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
 
-  cbz(ARMEmitter::Size::i32Bit, TMP1, &l_TFUnset);
+  (void)cbz(ARMEmitter::Size::i32Bit, TMP1, &l_TFUnset);
 
   // X86 semantically checks TF after executing each instruction, so e.g. setting a context with TF set will execute a single instruction
   // and then raise an exception. However on the FEX side this is simpler to implement by checking at the start of each instruction, handle this by having bit 1 being unset in the flag state indicate that TF is blocked for a single instruction.
-  tbz(TMP1, 1, &l_TFBlocked);
+  (void)tbz(TMP1, 1, &l_TFBlocked);
 
   // Block TF for a single instruction when the frontend jumps to a new context by unsetting bit 1.
   ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
@@ -775,11 +767,11 @@ void Arm64JITCore::EmitTFCheck() {
   ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP));
   br(TMP1);
 
-  Bind(&l_TFBlocked);
+  (void)Bind(&l_TFBlocked);
   // If TF was blocked for this instruction, unblock it for the next.
   LoadConstant(ARMEmitter::Size::i32Bit, TMP1, 0b11);
   strb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
-  Bind(&l_TFUnset);
+  (void)Bind(&l_TFUnset);
 }
 
 void Arm64JITCore::EmitSuspendInterruptCheck() {
@@ -795,16 +787,16 @@ void Arm64JITCore::EmitSuspendInterruptCheck() {
 
   ldr(TMP2.W(), STATE_PTR(CpuStateFrame, SuspendDoorbell));
   ARMEmitter::ForwardLabel l_NoSuspend;
-  cbz(ARMEmitter::Size::i32Bit, TMP2, &l_NoSuspend);
+  (void)cbz(ARMEmitter::Size::i32Bit, TMP2, &l_NoSuspend);
   brk(SuspendMagic);
-  Bind(&l_NoSuspend);
+  (void)Bind(&l_NoSuspend);
 #endif
 }
 
 void Arm64JITCore::EmitEntryPoint(ARMEmitter::BackwardLabel& HeaderLabel, bool CheckTF) {
   // Get the address of the JITCodeHeader and store in to the core state.
   // Two instruction cost, each 1 cycle.
-  adr(TMP1, &HeaderLabel);
+  adr_OrRestart(TMP1, &HeaderLabel);
   str(TMP1, STATE, offsetof(FEXCore::Core::CPUState, InlineJITBlockHeader));
 
   if (CheckTF) {
@@ -821,21 +813,34 @@ void Arm64JITCore::EmitEntryPoint(ARMEmitter::BackwardLabel& HeaderLabel, bool C
       sub(ARMEmitter::Size::i64Bit, ARMEmitter::XReg::rsp, ARMEmitter::XReg::rsp, TMP1, ARMEmitter::ExtendedType::LSL_64, 0);
     }
   }
+
+  EmitSuspendInterruptCheck();
 }
 
 CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size, bool SingleInst, const FEXCore::IR::IRListView* IR,
                                                    FEXCore::Core::DebugData* DebugData, bool CheckTF) {
   FEXCORE_PROFILE_SCOPED("Arm64::CompileCode");
-
-  JumpTargets.clear();
-  CallReturnTargets.clear();
-  PendingJumpThunks.clear();
-  uint32_t SSACount = IR->GetSSACount();
-  JumpTargets.resize(IR->GetHeader()->BlockCount, {});
-
   this->Entry = Entry;
   this->DebugData = DebugData;
   this->IR = IR;
+  RequiresFarARM64Jumps = false;
+
+  // Prepare restart via long jump in case branch encoding fails.
+  // This uses UncheckedLongJump since we don't implement std::longjmp in WoA setups
+  switch (static_cast<RestartOptions::Control>(FEXCore::UncheckedLongJump::SetJump(RestartControl.RestartJump))) {
+  case RestartOptions::Control::Incoming:
+    // Nothing
+    break;
+  case RestartOptions::Control::EnableFarARM64Jumps: RequiresFarARM64Jumps = true; break;
+  default: ERROR_AND_DIE_FMT("Unhandled Arm64 restart condition!");
+  }
+
+  uint32_t SSACount = IR->GetSSACount();
+  JumpTargets.clear();
+  CallReturnTargets.clear();
+  PendingJumpThunks.clear();
+  JumpTargets.resize(IR->GetHeader()->BlockCount, {});
+
   CodeData.EntryPoints.clear();
 
   // Fairly excessive buffer range to make sure we don't overflow
@@ -850,7 +855,7 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
 
   // Put the code header at the start of the data block.
   ARMEmitter::BackwardLabel JITCodeHeaderLabel {};
-  Bind(&JITCodeHeaderLabel);
+  (void)Bind(&JITCodeHeaderLabel);
   JITCodeHeader* CodeHeader = GetCursorAddress<JITCodeHeader*>();
   CursorIncrement(sizeof(JITCodeHeader));
 
@@ -898,7 +903,7 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
         if (PendingTargetLabel->Backward.Location) {
           EmitSuspendInterruptCheck();
         }
-        b(PendingTargetLabel);
+        b_OrRestart(PendingTargetLabel);
         PendingTargetLabel = nullptr;
       }
 
@@ -908,14 +913,14 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
         const auto IsReturnTarget = CallReturnTargets.try_emplace(Node).first;
         if (PendingTargetLabel) {
           // If there is a fallthrough branch to this block, skip over the entrypoint code.
-          b(Target);
+          b_OrRestart(Target);
         } else if (PendingCallReturnTargetLabel && PendingCallReturnTargetLabel != &IsReturnTarget->second) {
           // If we just emitted a call, but the block we're now emitting is not the return block so don't fallthrough.
-          b(PendingCallReturnTargetLabel);
+          b_OrRestart(PendingCallReturnTargetLabel);
         }
         PendingCallReturnTargetLabel = nullptr;
 
-        Bind(&IsReturnTarget->second);
+        BindOrRestart(&IsReturnTarget->second);
         CodeData.EntryPoints.emplace(BlockStartRIP, GetCursorAddress<uint8_t*>());
         DebugData->GuestOpcodes.push_back({BlockIROp->GuestEntryOffset, GetCursorAddress<uint8_t*>() - CodeData.BlockBegin});
 
@@ -924,18 +929,16 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
 
       if (PendingCallReturnTargetLabel) {
         // If there is still a pending call return target, then the block we're emitting is not the return block so don't fallthrough.
-        b(PendingCallReturnTargetLabel);
+        b_OrRestart(PendingCallReturnTargetLabel);
         PendingCallReturnTargetLabel = nullptr;
       }
       PendingTargetLabel = nullptr;
 
-      Bind(Target);
+      BindOrRestart(Target);
     }
 
     for (auto [CodeNode, IROp] : IR->GetCode(BlockNode)) {
       switch (IROp->Op) {
-#define REGISTER_OP_RT(op, x) \
-  case FEXCore::IR::IROps::OP_##op: std::invoke(RT_##x, this, IROp, CodeNode); break
 #define REGISTER_OP(op, x) \
   case FEXCore::IR::IROps::OP_##op: Op_##x(IROp, CodeNode); break
 
@@ -956,7 +959,7 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
     if (PendingTargetLabel->Backward.Location) {
       EmitSuspendInterruptCheck();
     }
-    b(PendingTargetLabel);
+    b_OrRestart(PendingTargetLabel);
   }
   PendingTargetLabel = nullptr;
 
@@ -967,21 +970,21 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
 
     ARMEmitter::ForwardLabel l_DoLink;
     uint64_t ThunkAddress = GetCursorAddress<uint64_t>();
-    Bind(&PendingJumpThunk.Label);
-    b(&l_DoLink);
+    BindOrRestart(&PendingJumpThunk.Label);
+    b_OrRestart(&l_DoLink);
     br(TMP1);
-    Bind(&l_DoLink);
+    BindOrRestart(&l_DoLink);
     ldr(TMP1, &l_ExitLink);
     blr(TMP1);
 
     // This is a ExitFunctionLinkData struct
-    Bind(&l_ExitLink);
+    BindOrRestart(&l_ExitLink);
     dc64(0);                                             // HostCode
     dc64(PendingJumpThunk.GuestRIP);                     // GuestRIP
     dc64(PendingJumpThunk.CallerAddress - ThunkAddress); // CallerOffset
   }
 
-  Bind(&l_ExitLink);
+  BindOrRestart(&l_ExitLink);
   dc64(ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker);
 
   // CodeSize not including the header or tail data.
@@ -1065,7 +1068,8 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
                                                                                                    "doesn't match up!\n");
       if (auto Prev = CheckCodeBufferUpdate()) {
         Allocator::VirtualDontNeed(ThreadState->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
-        ThreadState->LookupCache->ChangeGuestToHostMapping(*Prev, *CurrentCodeBuffer->LookupCache);
+        auto lk = ThreadState->LookupCache->AcquireWriteLock();
+        ThreadState->LookupCache->ChangeGuestToHostMapping(*Prev, *CurrentCodeBuffer->LookupCache, lk);
       }
 
       // NOTE: 16-byte alignment of the new cursor offset must be preserved for block linking records

FEXCore/Source/Interface/Core/JIT/JITClass.h

@@ -23,6 +23,7 @@ $end_info$
 #include <FEXCore/fextl/memory.h>
 #include <FEXCore/fextl/string.h>
 #include <FEXCore/fextl/vector.h>
+#include <FEXCore/Utils/LongJump.h>
 
 #include <CodeEmitter/Emitter.h>
 
@@ -60,14 +61,25 @@ public:
   }
 
 private:
-  FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
-
   const bool HostSupportsSVE128 {};
   const bool HostSupportsSVE256 {};
   const bool HostSupportsAVX256 {};
   const bool HostSupportsRPRES {};
   const bool HostSupportsAFP {};
 
+  struct RestartOptions {
+    FEXCore::UncheckedLongJump::JumpBuf RestartJump;
+    enum class Control : uint64_t {
+      Incoming = 0,
+      EnableFarARM64Jumps = 1,
+    };
+  };
+
+  // FEXCore makes assumptions in the JIT about certain conditions being true.
+  // In the rare case when those assumptions are broken, FEX needs to safely restart the JIT.
+  RestartOptions RestartControl {};
+  bool RequiresFarARM64Jumps {};
+
   ARMEmitter::BiDirectionalLabel* PendingTargetLabel {};
   ARMEmitter::BiDirectionalLabel* PendingCallReturnTargetLabel {};
   FEXCore::Context::ContextImpl* CTX {};
@@ -331,14 +343,179 @@ private:
   void EmitLinkedBranch(uint64_t GuestRIP, bool Call) {
     PendingJumpThunks.push_back({GetCursorAddress<uint64_t>(), GuestRIP, {}});
     auto& Thunk = PendingJumpThunks.back();
-    Bind(&Thunk.Label);
+    BindOrRestart(&Thunk.Label);
     if (Call) {
-      bl(&Thunk.Label);
+      bl_OrRestart(&Thunk.Label);
     } else {
-      b(&Thunk.Label);
+      b_OrRestart(&Thunk.Label);
     }
   }
 
+  // Restart helpers
+  template<ARMEmitter::IsLabel T>
+  void bl_OrRestart(T* Label) {
+    if (bl(Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    // We can support this but currently unnecessary.
+    ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+    FEXCore::UncheckedLongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void b_OrRestart(T* Label) {
+    if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    // We can support this but currently unnecessary.
+    ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+    FEXCore::UncheckedLongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void b_OrRestart(ARMEmitter::Condition Cond, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      ARMEmitter::ForwardLabel Skip {};
+      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
+      (void)b(InvertCondition(Cond), &Skip);
+      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+      }
+
+      (void)Bind(&Skip);
+      return;
+    }
+
+    if (b(Cond, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void cbz_OrRestart(ARMEmitter::Size s, ARMEmitter::Register rt, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      ARMEmitter::ForwardLabel Skip {};
+      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
+      (void)cbnz(s, rt, &Skip);
+      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+      }
+
+      (void)Bind(&Skip);
+      return;
+    }
+
+    if (cbz(s, rt, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void cbnz_OrRestart(ARMEmitter::Size s, ARMEmitter::Register rt, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      ARMEmitter::ForwardLabel Skip {};
+      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
+      (void)cbz(s, rt, &Skip);
+      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+      }
+
+      (void)Bind(&Skip);
+      return;
+    }
+
+    if (cbnz(s, rt, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void tbz_OrRestart(ARMEmitter::Register rt, uint32_t Bit, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      ARMEmitter::ForwardLabel Skip {};
+      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
+      (void)tbnz(rt, Bit, &Skip);
+      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+      }
+
+      (void)Bind(&Skip);
+      return;
+    }
+
+    if (tbz(rt, Bit, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void tbnz_OrRestart(ARMEmitter::Register rt, uint32_t Bit, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      ARMEmitter::ForwardLabel Skip {};
+      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
+      (void)tbz(rt, Bit, &Skip);
+      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+      }
+
+      (void)Bind(&Skip);
+      return;
+    }
+
+    if (tbnz(rt, Bit, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void adr_OrRestart(ARMEmitter::Register rd, T* Label) {
+    if (adr(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    // We can support this but currently unnecessary.
+    ERROR_AND_DIE_FMT("Long ADR currently unsupported!");
+    FEXCore::UncheckedLongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void adrp_OrRestart(ARMEmitter::Register rd, T* Label) {
+    if (adrp(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    // We can support this but currently unnecessary.
+    ERROR_AND_DIE_FMT("Long ADRP currently unsupported!");
+    FEXCore::UncheckedLongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void BindOrRestart(T* Label) {
+    if (Bind(Label)) {
+      return;
+    }
+
+    if (RequiresFarARM64Jumps) {
+      // This should have been caught before this point.
+      ERROR_AND_DIE_FMT("Unhandled long bind");
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
   // This is purely a debugging aid for developers to see if they are in JIT code space when inspecting raw memory
   void EmitDetectionString();
   IR::RegisterAllocationPass* RAPass {};
@@ -430,17 +607,8 @@ private:
 
   void EmitEntryPoint(ARMEmitter::BackwardLabel& HeaderLabel, bool CheckTF);
 
-  // Runtime selection;
-  // Load and store TSO memory style
-  OpType RT_LoadMemTSO;
-  OpType RT_StoreMemTSO;
-
 #define DEF_OP(x) void Op_##x(IR::IROp_Header const* IROp, IR::Ref Node)
 
-  // Dynamic Dispatcher supporting operations
-  DEF_OP(ParanoidLoadMemTSO);
-  DEF_OP(ParanoidStoreMemTSO);
-
   ///< Unhandled handler
   DEF_OP(Unhandled);
 

FEXCore/Source/Interface/Core/JIT/MemoryOps.cpp

@@ -912,7 +912,7 @@ DEF_OP(VLoadVectorMasked) {
 
       // If the sign bit is zero then skip the load
       ARMEmitter::ForwardLabel Skip {};
-      tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
+      (void)tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
       // Do the gather load for this element into the destination
       switch (IROp->ElementSize) {
       case IR::OpSize::i8Bit: ld1<ARMEmitter::SubRegSize::i8Bit>(TempDst.Q(), i, TempMemReg); break;
@@ -923,7 +923,7 @@ DEF_OP(VLoadVectorMasked) {
       default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, IROp->ElementSize); return;
       }
 
-      Bind(&Skip);
+      (void)Bind(&Skip);
 
       if ((i + 1) != NumElements) {
         // Handle register rename to save a move.
@@ -1013,7 +1013,7 @@ DEF_OP(VStoreVectorMasked) {
 
       // If the sign bit is zero then skip the load
       ARMEmitter::ForwardLabel Skip {};
-      tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
+      (void)tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
       // Do the gather load for this element into the destination
       switch (IROp->ElementSize) {
       case IR::OpSize::i8Bit: st1<ARMEmitter::SubRegSize::i8Bit>(RegData.Q(), i, TempMemReg); break;
@@ -1024,7 +1024,7 @@ DEF_OP(VStoreVectorMasked) {
       default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, IROp->ElementSize); return;
       }
 
-      Bind(&Skip);
+      (void)Bind(&Skip);
 
       if ((i + 1) != NumElements) {
         // Handle register rename to save a move.
@@ -1102,7 +1102,7 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
     PerformMove(ElementSize, WorkingReg, MaskReg, i);
 
     // Skip if the mask's sign bit isn't set
-    tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
+    (void)tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
 
     // Extract Index Element
     if ((IndexElement * IR::OpSizeToSize(VectorIndexSize)) >= 16) {
@@ -1140,7 +1140,7 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
     default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, ElementSize); FEX_UNREACHABLE;
     }
 
-    Bind(&Skip);
+    (void)Bind(&Skip);
   }
 
   if (NeedsDestTmp) {
@@ -1874,7 +1874,7 @@ DEF_OP(MemSet) {
 
   if (!DirectionIsInline) {
     // Backward or forwards implementation depends on flag
-    tbnz(DirectionReg, 1, &BackwardImpl);
+    (void)tbnz(DirectionReg, 1, &BackwardImpl);
   }
 
   auto MemStore = [this](auto Value, uint32_t OpSize, int32_t Size) {
@@ -1922,7 +1922,7 @@ DEF_OP(MemSet) {
     ARMEmitter::ForwardLabel DoneInternal {};
 
     // Early exit if zero count.
-    cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+    (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
 
     if (!IsAtomic) {
       ARMEmitter::ForwardLabel AgainInternal256Exit {};
@@ -1939,50 +1939,50 @@ DEF_OP(MemSet) {
       // Do this in two parts, to fallback to the byte by byte loop if size < 32, and to the
       // single copy loop if size < 64.
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbnz(TMP1, 63, &AgainInternal128Exit);
+      (void)tbnz(TMP1, 63, &AgainInternal128Exit);
 
       // Fill VTMP2 with the set pattern
       dup(SubRegSize, VTMP2.Q(), Value);
 
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbnz(TMP1, 63, &AgainInternal256Exit);
+      (void)tbnz(TMP1, 63, &AgainInternal256Exit);
 
-      Bind(&AgainInternal256);
+      (void)Bind(&AgainInternal256);
       stp<ARMEmitter::IndexType::POST>(VTMP2.Q(), VTMP2.Q(), TMP2, 32 * Direction);
       stp<ARMEmitter::IndexType::POST>(VTMP2.Q(), VTMP2.Q(), TMP2, 32 * Direction);
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
-      tbz(TMP1, 63, &AgainInternal256);
+      (void)tbz(TMP1, 63, &AgainInternal256);
 
-      Bind(&AgainInternal256Exit);
+      (void)Bind(&AgainInternal256Exit);
       add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
-      cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
 
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbnz(TMP1, 63, &AgainInternal128Exit);
-      Bind(&AgainInternal128);
+      (void)tbnz(TMP1, 63, &AgainInternal128Exit);
+      (void)Bind(&AgainInternal128);
       stp<ARMEmitter::IndexType::POST>(VTMP2.Q(), VTMP2.Q(), TMP2, 32 * Direction);
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbz(TMP1, 63, &AgainInternal128);
+      (void)tbz(TMP1, 63, &AgainInternal128);
 
-      Bind(&AgainInternal128Exit);
+      (void)Bind(&AgainInternal128Exit);
       add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
 
       if (Direction == -1) {
         add(ARMEmitter::Size::i64Bit, TMP2, TMP2, 32 - Size);
       }
     }
 
-    Bind(&AgainInternal);
+    (void)Bind(&AgainInternal);
     if (IsAtomic) {
       MemStoreTSO(Value, OpSize, SizeDirection);
     } else {
       MemStore(Value, OpSize, SizeDirection);
     }
     sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 1);
-    cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
+    (void)cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
 
-    Bind(&DoneInternal);
+    (void)Bind(&DoneInternal);
 
     if (SizeDirection >= 0) {
       switch (OpSize) {
@@ -2012,12 +2012,12 @@ DEF_OP(MemSet) {
       EmitMemset(Direction);
 
       if (Direction == 1) {
-        b(&Done);
-        Bind(&BackwardImpl);
+        (void)b(&Done);
+        (void)Bind(&BackwardImpl);
       }
     }
 
-    Bind(&Done);
+    (void)Bind(&Done);
     // Destination already set to the final pointer.
   }
 }
@@ -2067,7 +2067,7 @@ DEF_OP(MemCpy) {
 
   if (!DirectionIsInline) {
     // Backward or forwards implementation depends on flag
-    tbnz(DirectionReg, 1, &BackwardImpl);
+    (void)tbnz(DirectionReg, 1, &BackwardImpl);
   }
 
   auto MemCpy = [this](uint32_t OpSize, int32_t Size) {
@@ -2164,7 +2164,7 @@ DEF_OP(MemCpy) {
     ARMEmitter::ForwardLabel DoneInternal {};
 
     // Early exit if zero count.
-    cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+    (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
 
     if (!IsAtomic) {
       ARMEmitter::ForwardLabel AbsPos {};
@@ -2174,11 +2174,11 @@ DEF_OP(MemCpy) {
       ARMEmitter::BackwardLabel AgainInternal256 {};
 
       sub(ARMEmitter::Size::i64Bit, TMP4, TMP2, TMP3);
-      tbz(TMP4, 63, &AbsPos);
+      (void)tbz(TMP4, 63, &AbsPos);
       neg(ARMEmitter::Size::i64Bit, TMP4, TMP4);
-      Bind(&AbsPos);
+      (void)Bind(&AbsPos);
       sub(ARMEmitter::Size::i64Bit, TMP4, TMP4, 32);
-      tbnz(TMP4, 63, &AgainInternal);
+      (void)tbnz(TMP4, 63, &AgainInternal);
 
       if (Direction == -1) {
         sub(ARMEmitter::Size::i64Bit, TMP2, TMP2, 32 - Size);
@@ -2190,30 +2190,30 @@ DEF_OP(MemCpy) {
       // Do this in two parts, to fallback to the byte by byte loop if size < 32, and to the
       // single copy loop if size < 64.
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbnz(TMP1, 63, &AgainInternal128Exit);
+      (void)tbnz(TMP1, 63, &AgainInternal128Exit);
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbnz(TMP1, 63, &AgainInternal256Exit);
+      (void)tbnz(TMP1, 63, &AgainInternal256Exit);
 
-      Bind(&AgainInternal256);
+      (void)Bind(&AgainInternal256);
       MemCpy(32, 32 * Direction);
       MemCpy(32, 32 * Direction);
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
-      tbz(TMP1, 63, &AgainInternal256);
+      (void)tbz(TMP1, 63, &AgainInternal256);
 
-      Bind(&AgainInternal256Exit);
+      (void)Bind(&AgainInternal256Exit);
       add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
-      cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
 
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbnz(TMP1, 63, &AgainInternal128Exit);
-      Bind(&AgainInternal128);
+      (void)tbnz(TMP1, 63, &AgainInternal128Exit);
+      (void)Bind(&AgainInternal128);
       MemCpy(32, 32 * Direction);
       sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbz(TMP1, 63, &AgainInternal128);
+      (void)tbz(TMP1, 63, &AgainInternal128);
 
-      Bind(&AgainInternal128Exit);
+      (void)Bind(&AgainInternal128Exit);
       add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
 
       if (Direction == -1) {
         add(ARMEmitter::Size::i64Bit, TMP2, TMP2, 32 - Size);
@@ -2221,16 +2221,16 @@ DEF_OP(MemCpy) {
       }
     }
 
-    Bind(&AgainInternal);
+    (void)Bind(&AgainInternal);
     if (IsAtomic) {
       MemCpyTSO(OpSize, SizeDirection);
     } else {
       MemCpy(OpSize, SizeDirection);
     }
     sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 1);
-    cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
+    (void)cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
 
-    Bind(&DoneInternal);
+    (void)Bind(&DoneInternal);
 
     // Needs to use temporaries just in case of overwrite
     mov(TMP1, MemRegDest.X());
@@ -2288,186 +2288,15 @@ DEF_OP(MemCpy) {
     for (int32_t Direction : {1, -1}) {
       EmitMemcpy(Direction);
       if (Direction == 1) {
-        b(&Done);
-        Bind(&BackwardImpl);
+        (void)b(&Done);
+        (void)Bind(&BackwardImpl);
       }
     }
-    Bind(&Done);
+    (void)Bind(&Done);
     // Destination already set to the final pointer.
   }
 }
 
-DEF_OP(ParanoidLoadMemTSO) {
-  const auto Op = IROp->C<IR::IROp_LoadMemTSO>();
-  const auto OpSize = IROp->Size;
-
-  auto MemReg = GetReg(Op->Addr);
-
-  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == IR::RegClass::GPR) {
-    const auto Dst = GetReg(Node);
-    uint64_t Offset = 0;
-    if (!Op->Offset.IsInvalid()) {
-      if (!IsInlineConstant(Op->Offset, &Offset)) {
-        MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
-      }
-    }
-
-    if (OpSize == IR::OpSize::i8Bit) {
-      // 8bit load is always aligned to natural alignment
-      const auto Dst = GetReg(Node);
-      ldapurb(Dst, MemReg, Offset);
-    } else {
-      switch (OpSize) {
-      case IR::OpSize::i16Bit: ldapurh(Dst, MemReg, Offset); break;
-      case IR::OpSize::i32Bit: ldapur(Dst.W(), MemReg, Offset); break;
-      case IR::OpSize::i64Bit: ldapur(Dst.X(), MemReg, Offset); break;
-      default: LOGMAN_MSG_A_FMT("Unhandled ParanoidLoadMemTSO size: {}", OpSize); break;
-      }
-    }
-  } else if (CTX->HostFeatures.SupportsRCPC && Op->Class == IR::RegClass::GPR) {
-    const auto Dst = GetReg(Node);
-    MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
-    if (OpSize == IR::OpSize::i8Bit) {
-      // 8bit load is always aligned to natural alignment
-      ldaprb(Dst.W(), MemReg);
-    } else {
-      switch (OpSize) {
-      case IR::OpSize::i16Bit: ldaprh(Dst.W(), MemReg); break;
-      case IR::OpSize::i32Bit: ldapr(Dst.W(), MemReg); break;
-      case IR::OpSize::i64Bit: ldapr(Dst.X(), MemReg); break;
-      default: LOGMAN_MSG_A_FMT("Unhandled ParanoidLoadMemTSO size: {}", OpSize); break;
-      }
-    }
-  } else if (Op->Class == IR::RegClass::GPR) {
-    const auto Dst = GetReg(Node);
-    MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
-    switch (OpSize) {
-    case IR::OpSize::i8Bit: ldarb(Dst, MemReg); break;
-    case IR::OpSize::i16Bit: ldarh(Dst, MemReg); break;
-    case IR::OpSize::i32Bit: ldar(Dst.W(), MemReg); break;
-    case IR::OpSize::i64Bit: ldar(Dst.X(), MemReg); break;
-    default: LOGMAN_MSG_A_FMT("Unhandled ParanoidLoadMemTSO size: {}", OpSize); break;
-    }
-  } else {
-    const auto Dst = GetVReg(Node);
-    MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
-    switch (OpSize) {
-    case IR::OpSize::i8Bit:
-      ldarb(TMP1, MemReg);
-      fmov(ARMEmitter::Size::i32Bit, Dst.S(), TMP1.W());
-      break;
-    case IR::OpSize::i16Bit:
-      ldarh(TMP1, MemReg);
-      fmov(ARMEmitter::Size::i32Bit, Dst.S(), TMP1.W());
-      break;
-    case IR::OpSize::i32Bit:
-      ldar(TMP1.W(), MemReg);
-      fmov(ARMEmitter::Size::i32Bit, Dst.S(), TMP1.W());
-      break;
-    case IR::OpSize::i64Bit:
-      ldar(TMP1, MemReg);
-      fmov(ARMEmitter::Size::i64Bit, Dst.D(), TMP1);
-      break;
-    case IR::OpSize::i128Bit:
-      ldaxp(ARMEmitter::Size::i64Bit, TMP1, TMP2, MemReg);
-      clrex();
-      ins(ARMEmitter::SubRegSize::i64Bit, Dst, 0, TMP1);
-      ins(ARMEmitter::SubRegSize::i64Bit, Dst, 1, TMP2);
-      break;
-    case IR::OpSize::i256Bit:
-      LOGMAN_THROW_A_FMT(HostSupportsSVE256, "Need SVE256 support in order to use {} with 256-bit operation", __func__);
-      dmb(ARMEmitter::BarrierScope::ISH);
-      ld1b<ARMEmitter::SubRegSize::i8Bit>(Dst.Z(), PRED_TMP_32B.Zeroing(), MemReg);
-      dmb(ARMEmitter::BarrierScope::ISH);
-      break;
-    default: LOGMAN_MSG_A_FMT("Unhandled ParanoidLoadMemTSO size: {}", OpSize); break;
-    }
-  }
-}
-
-DEF_OP(ParanoidStoreMemTSO) {
-  const auto Op = IROp->C<IR::IROp_StoreMemTSO>();
-  const auto OpSize = IROp->Size;
-
-  auto MemReg = GetReg(Op->Addr);
-
-  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == IR::RegClass::GPR) {
-    const auto Src = GetZeroableReg(Op->Value);
-    uint64_t Offset = 0;
-    if (!Op->Offset.IsInvalid()) {
-      if (!IsInlineConstant(Op->Offset, &Offset)) {
-        MemReg = ApplyMemOperand(OpSize, MemReg, TMP1, Op->Offset, Op->OffsetType, Op->OffsetScale);
-      }
-    }
-
-    if (OpSize == IR::OpSize::i8Bit) {
-      // 8bit load is always aligned to natural alignment
-      stlurb(Src, MemReg, Offset);
-    } else {
-      switch (OpSize) {
-      case IR::OpSize::i16Bit: stlurh(Src, MemReg, Offset); break;
-      case IR::OpSize::i32Bit: stlur(Src.W(), MemReg, Offset); break;
-      case IR::OpSize::i64Bit: stlur(Src.X(), MemReg, Offset); break;
-      default: LOGMAN_MSG_A_FMT("Unhandled ParanoidStoreMemTSO size: {}", OpSize); break;
-      }
-    }
-  } else if (Op->Class == IR::RegClass::GPR) {
-    const auto Src = GetZeroableReg(Op->Value);
-    MemReg = ApplyMemOperand(OpSize, MemReg, TMP1, Op->Offset, Op->OffsetType, Op->OffsetScale);
-    switch (OpSize) {
-    case IR::OpSize::i8Bit: stlrb(Src, MemReg); break;
-    case IR::OpSize::i16Bit: stlrh(Src, MemReg); break;
-    case IR::OpSize::i32Bit: stlr(Src.W(), MemReg); break;
-    case IR::OpSize::i64Bit: stlr(Src.X(), MemReg); break;
-    default: LOGMAN_MSG_A_FMT("Unhandled ParanoidStoreMemTSO size: {}", OpSize); break;
-    }
-  } else {
-    const auto Src = GetVReg(Op->Value);
-
-    MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
-
-    switch (OpSize) {
-    case IR::OpSize::i8Bit:
-      umov<ARMEmitter::SubRegSize::i8Bit>(TMP1, Src, 0);
-      stlrb(TMP1, MemReg);
-      break;
-    case IR::OpSize::i16Bit:
-      umov<ARMEmitter::SubRegSize::i16Bit>(TMP1, Src, 0);
-      stlrh(TMP1, MemReg);
-      break;
-    case IR::OpSize::i32Bit:
-      umov<ARMEmitter::SubRegSize::i32Bit>(TMP1, Src, 0);
-      stlr(TMP1.W(), MemReg);
-      break;
-    case IR::OpSize::i64Bit:
-      umov<ARMEmitter::SubRegSize::i64Bit>(TMP1, Src, 0);
-      stlr(TMP1, MemReg);
-      break;
-    case IR::OpSize::i128Bit: {
-      // Move vector to GPRs
-      umov<ARMEmitter::SubRegSize::i64Bit>(TMP1, Src, 0);
-      umov<ARMEmitter::SubRegSize::i64Bit>(TMP2, Src, 1);
-      ARMEmitter::BackwardLabel B;
-      Bind(&B);
-
-      // ldaxp must not have both the destination registers be the same
-      ldaxp(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::zr, TMP3, MemReg); // <- Can hit SIGBUS. Overwritten with DMB
-      stlxp(ARMEmitter::Size::i64Bit, TMP3, TMP1, TMP2, MemReg);          // <- Can also hit SIGBUS
-      cbnz(ARMEmitter::Size::i64Bit, TMP3, &B);                           // < Overwritten with DMB
-      break;
-    }
-    case IR::OpSize::i256Bit: {
-      LOGMAN_THROW_A_FMT(HostSupportsSVE256, "Need SVE256 support in order to use {} with 256-bit operation", __func__);
-      dmb(ARMEmitter::BarrierScope::ISH);
-      st1b<ARMEmitter::SubRegSize::i8Bit>(Src.Z(), PRED_TMP_32B, MemReg, 0);
-      dmb(ARMEmitter::BarrierScope::ISH);
-      break;
-    }
-    default: LOGMAN_MSG_A_FMT("Unhandled ParanoidStoreMemTSO size: {}", OpSize); break;
-    }
-  }
-}
-
 DEF_OP(CacheLineClear) {
   if (!CTX->HostFeatures.SupportsCacheMaintenanceOps) {
     dmb(ARMEmitter::BarrierScope::SY);

FEXCore/Source/Interface/Core/LookupCache.cpp

@@ -15,7 +15,7 @@ $end_info$
 
 namespace FEXCore {
 GuestToHostMap::GuestToHostMap()
-  : BlockLinks_mbr {fextl::pmr::get_default_resource()} {
+  : BlockLinks_mbr {"FEXMem_BlockLinks"} {
   BlockLinks_pma = fextl::make_unique<std::pmr::polymorphic_allocator<std::byte>>(&BlockLinks_mbr);
   // Setup our PMR map.
   BlockLinks = BlockLinks_pma->new_object<BlockLinksMapType>();
@@ -24,7 +24,7 @@ GuestToHostMap::GuestToHostMap()
 LookupCache::LookupCache(FEXCore::Context::ContextImpl* CTX)
   : ctx {CTX} {
 
-  TotalCacheSize = ctx->Config.VirtualMemSize / 4096 * 8 + CODE_SIZE + L1_SIZE;
+  TotalCacheSize = ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8 + CODE_SIZE + MAX_L1_SIZE;
 
   // Block cache ends up looking like this
   // PageMemoryMap[VirtualMemoryRegion >> 12]
@@ -39,6 +39,10 @@ LookupCache::LookupCache(FEXCore::Context::ContextImpl* CTX)
   // We need one pointer per page of virtual memory
   // At 64GB of virtual memory this will allocate 128MB of virtual memory space
   PagePointer = reinterpret_cast<uintptr_t>(FEXCore::Allocator::VirtualAlloc(TotalCacheSize, false, false));
+  LOGMAN_THROW_A_FMT(PagePointer != -1ULL, "Failed to allocate PagePointer");
+
+  FEXCore::Allocator::VirtualName("FEXMem_Lookup", reinterpret_cast<void*>(PagePointer),
+                                  ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8 + CODE_SIZE);
   CTX->SyscallHandler->MarkOvercommitRange(PagePointer, TotalCacheSize);
 
   // Allocate our memory backing our pages
@@ -46,14 +50,21 @@ LookupCache::LookupCache(FEXCore::Context::ContextImpl* CTX)
   // XXX: We can drop down to 16KB if we store 4byte offsets from the code base
   // We currently limit to 128MB of real memory for caching for the total cache size.
   // Can end up being inefficient if we compile a small number of blocks per page
-  PageMemory = PagePointer + ctx->Config.VirtualMemSize / 4096 * 8;
-  LOGMAN_THROW_A_FMT(PageMemory != -1ULL, "Failed to allocate page memory");
+  PageMemory = PagePointer + ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8;
 
   // L1 Cache
   L1Pointer = PageMemory + CODE_SIZE;
-  LOGMAN_THROW_A_FMT(L1Pointer != -1ULL, "Failed to allocate L1Pointer");
+  FEXCore::Allocator::VirtualName("FEXMem_Lookup_L1", reinterpret_cast<void*>(L1Pointer), MAX_L1_SIZE);
 
   VirtualMemSize = ctx->Config.VirtualMemSize;
+
+  if (DynamicL1Cache()) {
+    // Start at minimum size when dynamic.
+    L1PointerMask = MIN_L1_ENTRIES - 1;
+  } else {
+    // Start at maximum instead.
+    L1PointerMask = MAX_L1_ENTRIES - 1;
+  }
 }
 
 LookupCache::~LookupCache() {
@@ -64,31 +75,27 @@ LookupCache::~LookupCache() {
   // These will get freed when their memory allocators are deallocated.
 }
 
-void LookupCache::ClearL2Cache() {
-  auto lk = Shared->AcquireLock();
+void LookupCache::ClearL2Cache(const FEXCore::LookupCacheReadLockToken& lk) {
   // Clear out the page memory
   // PagePointer and PageMemory are sequential with each other. Clear both at once.
-  FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer), ctx->Config.VirtualMemSize / 4096 * 8 + CODE_SIZE, false);
+  FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer),
+                                      ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8 + CODE_SIZE, false);
   AllocateOffset = 0;
 }
 
-void LookupCache::ClearThreadLocalCaches() {
-  auto lk = Shared->AcquireLock();
-
+void LookupCache::ClearThreadLocalCaches(const LookupCacheWriteLockToken&) {
   // Clear L1 and L2 by clearing the full cache.
   FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer), TotalCacheSize, false);
 }
 
-void LookupCache::ClearCache() {
-  auto lk = Shared->AcquireLock();
-
+void LookupCache::ClearCache(const LookupCacheWriteLockToken& lk) {
   // Clear L1 and L2 by clearing the full cache.
   FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer), TotalCacheSize, false);
 
   Shared->ClearCache(lk);
 }
 
-void GuestToHostMap::ClearCache(const LockToken&) {
+void GuestToHostMap::ClearCache(const LookupCacheWriteLockToken&) {
   // Allocate a new pointer from the BlockLinks pma again.
   BlockLinks = BlockLinks_pma->new_object<BlockLinksMapType>();
   // All code is gone, clear the block list

FEXCore/Source/Interface/Core/LookupCache.h

@@ -2,6 +2,9 @@
 #pragma once
 #include "Interface/Context/Context.h"
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/SHMStats.h>
+#include "Utils/WritePriorityMutex.h"
+
 #include <FEXCore/fextl/map.h>
 #include <FEXCore/fextl/memory_resource.h>
 #include <FEXCore/fextl/robin_map.h>
@@ -9,23 +12,40 @@
 #include <FEXCore/fextl/memory_resource.h>
 
 #include <cstdint>
-#include <functional>
 #include <stddef.h>
 #include <utility>
 #include <mutex>
 
 namespace FEXCore {
+struct LookupCacheWriteLockToken {
+private:
+  // Only constructible by GuestToHostMap
+  friend struct GuestToHostMap;
+  LookupCacheWriteLockToken(FEXCore::Utils::WritePriorityMutex::Mutex& Mutex)
+    : Lock {Mutex} {}
+  std::lock_guard<FEXCore::Utils::WritePriorityMutex::Mutex> Lock;
+};
+
+struct LookupCacheReadLockToken {
+private:
+  // Only constructible by GuestToHostMap
+  friend struct GuestToHostMap;
+  LookupCacheReadLockToken(FEXCore::Utils::WritePriorityMutex::Mutex& Mutex)
+    : Lock {Mutex} {}
+  std::shared_lock<FEXCore::Utils::WritePriorityMutex::Mutex> Lock;
+};
 
 struct GuestToHostMap {
-  std::recursive_mutex WriteLock;
-
-  struct LockToken {
-    std::lock_guard<std::recursive_mutex> Lock;
-  };
+  FEXCore::Utils::WritePriorityMutex::Mutex Lock {};
 
   [[nodiscard]]
-  LockToken AcquireLock() {
-    return LockToken {std::lock_guard {WriteLock}};
+  LookupCacheWriteLockToken AcquireWriteLock() {
+    return LookupCacheWriteLockToken {Lock};
+  }
+
+  [[nodiscard]]
+  LookupCacheReadLockToken AcquireReadLock() {
+    return LookupCacheReadLockToken {Lock};
   }
 
   struct BlockLinkTag {
@@ -49,7 +69,7 @@ struct GuestToHostMap {
   // walking each block member and destructing objects.
   //
   // This makes `BlockLinks` look like a raw pointer that could memory leak, but since it is backed by the MBR, it won't.
-  std::pmr::monotonic_buffer_resource BlockLinks_mbr;
+  fextl::pmr::named_monotonic_page_buffer_resource BlockLinks_mbr;
   using BlockLinksMapType = std::pmr::map<BlockLinkTag, FEXCore::Context::BlockDelinkerFunc>;
   fextl::unique_ptr<std::pmr::polymorphic_allocator<std::byte>> BlockLinks_pma;
   BlockLinksMapType* BlockLinks;
@@ -61,7 +81,7 @@ struct GuestToHostMap {
   GuestToHostMap();
 
   // Adds to Guest -> Host code mapping
-  void AddBlockMapping(uint64_t Address, void* HostCode, const LockToken&) {
+  void AddBlockMapping(uint64_t Address, void* HostCode, const LookupCacheWriteLockToken&) {
     // This may replace an existing mapping
     // NOTE: Generally no previous entry should exist, however there is one exception:
     //       If the backend updates the active thread's CodeBuffer, the new associated LookupCache
@@ -70,7 +90,7 @@ struct GuestToHostMap {
     BlockList[Address] = (uintptr_t)HostCode;
   }
 
-  std::optional<uintptr_t> FindBlock(uint64_t Address, const LockToken&) {
+  std::optional<uintptr_t> FindBlock(uint64_t Address, const LookupCacheReadLockToken&) {
     auto HostCode = BlockList.find(Address);
     if (HostCode == BlockList.end()) {
       return std::nullopt;
@@ -78,7 +98,7 @@ struct GuestToHostMap {
     return HostCode->second;
   }
 
-  bool Erase(FEXCore::Core::CpuStateFrame* Frame, uint64_t Address, const LockToken&) {
+  bool Erase(FEXCore::Core::CpuStateFrame* Frame, uint64_t Address, const LookupCacheWriteLockToken&) {
     // Sever any links to this block
     auto lower = BlockLinks->lower_bound({Address, nullptr});
     auto upper = BlockLinks->upper_bound({Address, reinterpret_cast<FEXCore::Context::ExitFunctionLinkData*>(UINTPTR_MAX)});
@@ -91,11 +111,11 @@ struct GuestToHostMap {
   }
 
   void AddBlockLink(uint64_t GuestDestination, FEXCore::Context::ExitFunctionLinkData* HostLink,
-                    const FEXCore::Context::BlockDelinkerFunc& delinker, const LockToken&) {
+                    const FEXCore::Context::BlockDelinkerFunc& delinker, const LookupCacheWriteLockToken&) {
     BlockLinks->insert({{GuestDestination, HostLink}, delinker});
   }
 
-  bool AddBlockExecutableRange(const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length, const LockToken&) {
+  bool AddBlockExecutableRange(const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length, const LookupCacheWriteLockToken&) {
     bool rv = false;
 
     for (auto CurrentPage = Start >> 12, EndPage = (Start + Length - 1) >> 12; CurrentPage <= EndPage; CurrentPage++) {
@@ -107,7 +127,7 @@ struct GuestToHostMap {
     return rv;
   }
 
-  void ClearCache(const LockToken&);
+  void ClearCache(const LookupCacheWriteLockToken&);
 };
 
 class LookupCache {
@@ -122,69 +142,134 @@ public:
 
   // Swaps out the underlying GuestToHostMap and clears all associated caches.
   // This interface requires the previous CodeBuffer to be provided despite not using it. This ensures the shared write lock is still valid.
-  void ChangeGuestToHostMapping([[maybe_unused]] CPU::CodeBuffer& Prev, GuestToHostMap& NewMap) {
-    ClearThreadLocalCaches();
+  void ChangeGuestToHostMapping([[maybe_unused]] CPU::CodeBuffer& Prev, GuestToHostMap& NewMap, const LookupCacheWriteLockToken& lk) {
+    ClearThreadLocalCaches(lk);
     Shared = &NewMap;
   }
 
-  uintptr_t FindBlock(uint64_t Address) {
+  uintptr_t FindBlock(FEXCore::Core::InternalThreadState* Thread, uint64_t Address) {
     // Try L1, no lock needed
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
+    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1PointerMask];
     if (L1Entry.GuestCode == Address) {
       return L1Entry.HostCode;
     }
 
     // L2 and L3 need to be locked
-    auto lk = Shared->AcquireLock();
+    uintptr_t HostPtr {};
+    {
+      std::optional<FEXCore::SHMStats::AccumulationBlock<uint64_t>> LockTime(
+        Thread->ThreadStats ? &Thread->ThreadStats->AccumulatedCacheReadLockTime : nullptr);
+      auto lk = Shared->AcquireReadLock();
+      LockTime.reset();
 
-    // Try L2
-    const auto PageIndex = (Address & (VirtualMemSize - 1)) >> 12;
-    const auto PageOffset = Address & (0x0FFF);
+      if (!DisableL2Cache()) {
+        // Try L2
+        const auto PageIndex = (Address & (VirtualMemSize - 1)) >> 12;
+        const auto PageOffset = Address & (0x0FFF);
 
-    const auto Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
-    auto LocalPagePointer = Pointers[PageIndex];
+        const auto Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
+        auto LocalPagePointer = Pointers[PageIndex];
 
-    // Do we a page pointer for this address?
-    if (LocalPagePointer) {
-      // Find there pointer for the address in the blocks
-      auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+        // Do we a page pointer for this address?
+        if (LocalPagePointer) {
+          // Find there pointer for the address in the blocks
+          auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
 
-      if (BlockPointers[PageOffset].GuestCode == Address) {
-        L1Entry.GuestCode = Address;
-        L1Entry.HostCode = BlockPointers[PageOffset].HostCode;
-        return L1Entry.HostCode;
+          if (BlockPointers[PageOffset].GuestCode == Address) {
+            L1Entry.GuestCode = Address;
+            L1Entry.HostCode = BlockPointers[PageOffset].HostCode;
+            HostPtr = L1Entry.HostCode;
+          }
+        }
+      }
+
+      if (!HostPtr) {
+        // Try L3
+        auto HostCode = Shared->FindBlock(Address, lk);
+        if (HostCode) {
+          CacheBlockMapping(Address, HostCode.value(), lk);
+          HostPtr = HostCode.value();
+        }
       }
     }
 
-    // Try L3
-    auto HostCode = Shared->FindBlock(Address, lk);
-    if (HostCode) {
-      CacheBlockMapping(Address, HostCode.value());
-      return HostCode.value();
+    if (HostPtr && DynamicL1Cache()) {
+      UpdateDynamicL1Stats(Thread);
     }
 
-    // Failed to find
-    return 0;
+    FEXCORE_PROFILE_INSTANT_INCREMENT(Thread, AccumulatedCacheMissCount, 1);
+
+    return HostPtr;
+  }
+
+  void UpdateDynamicL1Stats(FEXCore::Core::InternalThreadState* Thread) {
+    // If host pointer was found in L2 or L3, then add it to the counter.
+    // Keeping track not L1 misses, but specifically L2/L3 hits.
+    ++L2L3CacheHits;
+
+    const auto CurrentTime = std::chrono::system_clock::now();
+    const auto Period = CurrentTime - LastPeriod;
+    if (Period >= SamplePeriod) {
+      // If larger than the sample period then check if we need to increase L1 cache size.
+      const double AveragePerSecond = static_cast<double>(L2L3CacheHits) /
+                                      static_cast<double>(std::chrono::duration_cast<std::chrono::milliseconds>(Period).count()) * 1000.0;
+
+      if (AveragePerSecond >= DynamicL1CacheIncreaseCountHeuristic()) {
+        if (CurrentL1Entries < MAX_L1_ENTRIES) {
+          CurrentL1Entries <<= 1;
+          L1PointerMask = CurrentL1Entries - 1;
+
+          // Update the thread's L1 pointer mask to increase how much cache it uses.
+          // Since we're in C-code, this is safe to update here.
+          Thread->CurrentFrame->State.L1Mask = GetScaledL1PointerMask();
+        }
+      } else if (AveragePerSecond < DynamicL1CacheDecreaseCountHeuristic()) {
+        if (CurrentL1Entries > MIN_L1_ENTRIES) {
+          CurrentL1Entries >>= 1;
+          L1PointerMask = CurrentL1Entries - 1;
+
+          // Madvise the entries that we are dropping. Gives the memory back to the OS.
+          LookupCacheEntry* FirstZeroL1Entry = &reinterpret_cast<LookupCacheEntry*>(L1Pointer)[CurrentL1Entries];
+          size_t ZeroMemorySize = (MAX_L1_ENTRIES - CurrentL1Entries) * sizeof(LookupCacheEntry);
+          FEXCore::Allocator::VirtualDontNeed(FirstZeroL1Entry, ZeroMemorySize, false);
+
+          // Update the thread's L1 pointer mask to increase how much cache it uses.
+          // Since we're in C-code, this is safe to update here.
+          Thread->CurrentFrame->State.L1Mask = GetScaledL1PointerMask();
+        }
+      }
+
+      // Update Last period to start again.
+      LastPeriod = CurrentTime;
+      L2L3CacheHits = 0;
+    }
   }
 
   GuestToHostMap* Shared = nullptr;
 
   // Appends a list of Block {Address} to CodePages [Start, Start + Length)
   // Returns true if new pages are marked as containing code
-  bool AddBlockExecutableRange(const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length) {
-    auto lk = Shared->AcquireLock();
+  bool AddBlockExecutableRange(FEXCore::Core::InternalThreadState* Thread, const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length) {
+    std::optional<FEXCore::SHMStats::AccumulationBlock<uint64_t>> LockTime(
+      Thread->ThreadStats ? &Thread->ThreadStats->AccumulatedCacheWriteLockTime : nullptr);
+    auto lk = Shared->AcquireWriteLock();
+    LockTime.reset();
+
     return Shared->AddBlockExecutableRange(Addresses, Start, Length, lk);
   }
 
   // Adds to Guest -> Host code mapping
-  void AddBlockMapping(uint64_t Address, void* HostCode) {
-    auto lk = Shared->AcquireLock();
+  void AddBlockMapping(FEXCore::Core::InternalThreadState* Thread, uint64_t Address, void* HostCode) {
+    std::optional<FEXCore::SHMStats::AccumulationBlock<uint64_t>> LockTime(
+      Thread->ThreadStats ? &Thread->ThreadStats->AccumulatedCacheWriteLockTime : nullptr);
+    auto lk = Shared->AcquireWriteLock();
+    LockTime.reset();
 
     Shared->AddBlockMapping(Address, HostCode, lk);
 
     // There is no need to update L1 or L2, they will get updated on first lookup
     // However, adding to L1 here increases performance
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
+    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1PointerMask];
     L1Entry.GuestCode = Address;
     L1Entry.HostCode = (uintptr_t)HostCode;
   }
@@ -192,13 +277,11 @@ public:
   // NOTE: It's the caller's responsibility to call Erase() for all other
   //       GuestToHostMaps that share the same LookupCache. Otherwise, the
   //       L1/L2 caches will contain stale references to deallocated memory.
-  bool Erase(FEXCore::Core::CpuStateFrame* Frame, uint64_t Address) {
-    auto lk = Shared->AcquireLock();
-
+  bool Erase(FEXCore::Core::CpuStateFrame* Frame, uint64_t Address, const LookupCacheWriteLockToken& lk) {
     bool ErasedAny = Shared->Erase(Frame, Address, lk);
 
     // Do L1
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
+    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1PointerMask];
     if (L1Entry.GuestCode == Address) {
       L1Entry.GuestCode = 0;
       ErasedAny = true;
@@ -207,37 +290,42 @@ public:
       // and it hasn't been thoroughly tested
     }
 
-    // Do full map
-    Address = Address & (VirtualMemSize - 1);
-    uint64_t PageOffset = Address & (0x0FFF);
-    Address >>= 12;
+    if (!DisableL2Cache()) {
+      // Do full map
+      Address = Address & (VirtualMemSize - 1);
+      uint64_t PageOffset = Address & (0x0FFF);
+      Address >>= 12;
 
-    uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
-    uint64_t LocalPagePointer = Pointers[Address];
-    if (!LocalPagePointer) {
-      // Page for this code didn't even exist, nothing to do
-      return ErasedAny;
+      uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
+      uint64_t LocalPagePointer = Pointers[Address];
+      if (!LocalPagePointer) {
+        // Page for this code didn't even exist, nothing to do
+        return ErasedAny;
+      }
+
+      // Page exists, just set the offset to zero
+      auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+      BlockPointers[PageOffset].GuestCode = 0;
+      BlockPointers[PageOffset].HostCode = 0;
     }
-
-    // Page exists, just set the offset to zero
-    auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
-    BlockPointers[PageOffset].GuestCode = 0;
-    BlockPointers[PageOffset].HostCode = 0;
     return true;
   }
 
-  void AddBlockLink(uint64_t GuestDestination, FEXCore::Context::ExitFunctionLinkData* HostLink, const FEXCore::Context::BlockDelinkerFunc& delinker) {
-    auto lk = Shared->AcquireLock();
+  void AddBlockLink(uint64_t GuestDestination, FEXCore::Context::ExitFunctionLinkData* HostLink,
+                    const FEXCore::Context::BlockDelinkerFunc& delinker, const LookupCacheWriteLockToken& lk) {
     Shared->AddBlockLink(GuestDestination, HostLink, delinker, lk);
   }
 
-  void ClearCache();
-  void ClearL2Cache();
-  void ClearThreadLocalCaches();
+  void ClearCache(const LookupCacheWriteLockToken&);
+  void ClearL2Cache(const LookupCacheReadLockToken&);
+  void ClearThreadLocalCaches(const LookupCacheWriteLockToken&);
 
   uintptr_t GetL1Pointer() const {
     return L1Pointer;
   }
+  uintptr_t GetScaledL1PointerMask() const {
+    return L1PointerMask << FEXCore::ilog2(sizeof(LookupCache::LookupCacheEntry));
+  }
   uintptr_t GetPagePointer() const {
     return PagePointer;
   }
@@ -245,9 +333,6 @@ public:
     return VirtualMemSize;
   }
 
-  constexpr static size_t L1_ENTRIES = 1 * 1024 * 1024; // Must be a power of 2
-  constexpr static size_t L1_ENTRIES_MASK = L1_ENTRIES - 1;
-
   // This needs to be taken before reads or writes to L2, L3, CodePages,
   // and before writes to L1. Concurrent access from a thread that this LookupCache doesn't belong to
   // may only happen during cross thread invalidation (::Erase).
@@ -255,45 +340,48 @@ public:
   // Some care is taken so that L1 lookups can be done without locks, and even tearing is unlikely to lead to a crash.
   // This approach has not been fully vetted yet.
   // Also note that L1 lookups might be inlined in the JIT Dispatcher and/or block ends.
-  auto AcquireLock() {
-    return Shared->AcquireLock();
+  auto AcquireWriteLock() {
+    return Shared->AcquireWriteLock();
   }
 
 private:
-  void CacheBlockMapping(uint64_t Address, uintptr_t HostCode) {
+  void CacheBlockMapping(uint64_t Address, uintptr_t HostCode, const LookupCacheReadLockToken& lk) {
     // Do L1
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
+    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1PointerMask];
     L1Entry.GuestCode = Address;
     L1Entry.HostCode = HostCode;
 
-    // Do ful map
-    auto FullAddress = Address;
-    Address = Address & (VirtualMemSize - 1);
+    if (!DisableL2Cache()) {
+      // Do ful map
+      auto FullAddress = Address;
+      Address = Address & (VirtualMemSize - 1);
 
-    uint64_t PageOffset = Address & (0x0FFF);
-    Address >>= 12;
-    uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
-    uint64_t LocalPagePointer = Pointers[Address];
-    if (!LocalPagePointer) {
-      // We don't have a page pointer for this address
-      // Allocate one now if we can
-      uintptr_t NewPageBacking = AllocateBackingForPage();
-      if (!NewPageBacking) {
-        // Couldn't allocate, clear L2 and retry
-        ClearL2Cache();
-        CacheBlockMapping(Address, HostCode);
-        return;
+      uint64_t PageOffset = Address & (0x0FFF);
+      Address >>= 12;
+
+      uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
+      uint64_t LocalPagePointer = Pointers[Address];
+      if (!LocalPagePointer) {
+        // We don't have a page pointer for this address
+        // Allocate one now if we can
+        uintptr_t NewPageBacking = AllocateBackingForPage();
+        if (!NewPageBacking) {
+          // Couldn't allocate, clear L2 and retry
+          ClearL2Cache(lk);
+          CacheBlockMapping(Address, HostCode, lk);
+          return;
+        }
+        Pointers[Address] = NewPageBacking;
+        LocalPagePointer = NewPageBacking;
       }
-      Pointers[Address] = NewPageBacking;
-      LocalPagePointer = NewPageBacking;
+
+      // Add the new pointer to the page block
+      auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+
+      // This silently replaces existing mappings
+      BlockPointers[PageOffset].GuestCode = FullAddress;
+      BlockPointers[PageOffset].HostCode = HostCode;
     }
-
-    // Add the new pointer to the page block
-    auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
-
-    // This silently replaces existing mappings
-    BlockPointers[PageOffset].GuestCode = FullAddress;
-    BlockPointers[PageOffset].HostCode = HostCode;
   }
 
   uintptr_t AllocateBackingForPage() {
@@ -313,16 +401,32 @@ private:
   uintptr_t PagePointer;
   uintptr_t PageMemory;
   uintptr_t L1Pointer;
+  uintptr_t L1PointerMask;
 
   size_t TotalCacheSize;
 
+  // Start with 8k entries in L1 to give 128KB of L1 cache to each thread.
+  // Max out at 1 million entries to give each thread 16MB of L1 cache maximum.
+  constexpr static size_t MIN_L1_ENTRIES = 8 * 1024;        // Must be a power of 2
+  constexpr static size_t MAX_L1_ENTRIES = 1 * 1024 * 1024; // Must be a power of 2
+
   constexpr static size_t CODE_SIZE = 128 * 1024 * 1024;
-  constexpr static size_t SIZE_PER_PAGE = 4096 * sizeof(LookupCacheEntry);
-  constexpr static size_t L1_SIZE = L1_ENTRIES * sizeof(LookupCacheEntry);
+  constexpr static size_t SIZE_PER_PAGE = FEXCore::Utils::FEX_PAGE_SIZE * sizeof(LookupCacheEntry);
+  constexpr static size_t MAX_L1_SIZE = MAX_L1_ENTRIES * sizeof(LookupCacheEntry);
 
   size_t AllocateOffset {};
 
   FEXCore::Context::ContextImpl* ctx;
   uint64_t VirtualMemSize {};
+
+  size_t CurrentL1Entries = MIN_L1_ENTRIES;
+  uint64_t L2L3CacheHits {};
+  std::chrono::time_point<std::chrono::system_clock> LastPeriod {};
+  constexpr static std::chrono::seconds SamplePeriod {1};
+  FEX_CONFIG_OPT(DynamicL1CacheIncreaseCountHeuristic, DYNAMICL1CACHEINCREASECOUNTHEURISTIC);
+  FEX_CONFIG_OPT(DynamicL1CacheDecreaseCountHeuristic, DYNAMICL1CACHEDECREASECOUNTHEURISTIC);
+
+  FEX_CONFIG_OPT(DynamicL1Cache, DYNAMICL1CACHE);
+  FEX_CONFIG_OPT(DisableL2Cache, DISABLEL2CACHE);
 };
 } // namespace FEXCore

FEXCore/Source/Interface/Core/OpcodeDispatcher.cpp

@@ -28,7 +28,6 @@ $end_info$
 #include <algorithm>
 #include <array>
 #include <cstdint>
-#include <tuple>
 
 namespace FEXCore::IR {
 
@@ -51,8 +50,6 @@ void OpDispatchBuilder::SyscallOp(OpcodeArgs, bool IsSyscallInst) {
     FEXCore::X86State::REG_RSI, FEXCore::X86State::REG_RDI, FEXCore::X86State::REG_RBP,
   };
 
-  SyscallFlags DefaultSyscallFlags = FEXCore::IR::SyscallFlags::DEFAULT;
-
   const auto OSABI = CTX->SyscallHandler->GetOSABI();
   if (OSABI == FEXCore::HLE::SyscallOSABI::OS_LINUX64) {
     NumArguments = GPRIndexes_64.size();
@@ -64,7 +61,6 @@ void OpDispatchBuilder::SyscallOp(OpcodeArgs, bool IsSyscallInst) {
     // All registers will be spilled before the syscall and filled afterwards so no JIT-side argument handling is necessary.
     NumArguments = 0;
     GPRIndexes = nullptr;
-    DefaultSyscallFlags = FEXCore::IR::SyscallFlags::NORETURNEDRESULT;
   } else {
     ERROR_AND_DIE_FMT("Unhandled OSABI syscall");
   }
@@ -98,9 +94,10 @@ void OpDispatchBuilder::SyscallOp(OpcodeArgs, bool IsSyscallInst) {
   }
 
   FlushRegisterCache();
-  auto SyscallOp = _Syscall(Arguments[0], Arguments[1], Arguments[2], Arguments[3], Arguments[4], Arguments[5], Arguments[6], DefaultSyscallFlags);
+  auto SyscallOp = _Syscall(Arguments[0], Arguments[1], Arguments[2], Arguments[3], Arguments[4], Arguments[5], Arguments[6]);
 
-  if ((DefaultSyscallFlags & FEXCore::IR::SyscallFlags::NORETURNEDRESULT) != FEXCore::IR::SyscallFlags::NORETURNEDRESULT) {
+  // Generic ABI doesn't store result in RAX.
+  if (OSABI != FEXCore::HLE::SyscallOSABI::OS_GENERIC) {
     StoreGPRRegister(X86State::REG_RAX, SyscallOp);
   }
 
@@ -153,12 +150,6 @@ void OpDispatchBuilder::NOPOp(OpcodeArgs) {}
 void OpDispatchBuilder::RETOp(OpcodeArgs) {
   const auto GPRSize = GetGPROpSize();
 
-  // ABI Optimization: Flags don't survive calls or rets
-  if (CTX->Config.ABILocalFlags) {
-    _InvalidateFlags(~0UL); // all flags
-    InvalidatePF_AF();
-  }
-
   Ref SP = _RMWHandle(LoadGPRRegister(X86State::REG_RSP));
   Ref NewRIP = Pop(GPRSize, SP);
 
@@ -522,12 +513,6 @@ void OpDispatchBuilder::CALLOp(OpcodeArgs) {
 
   BlockSetRIP = true;
 
-  // ABI Optimization: Flags don't survive calls or rets
-  if (CTX->Config.ABILocalFlags) {
-    _InvalidateFlags(~0UL); // all flags
-    InvalidatePF_AF();
-  }
-
   // Call instruction only uses up to 32-bit signed displacement
   int64_t TargetOffset = Op->Src[0].Literal();
 
@@ -3230,7 +3215,11 @@ void OpDispatchBuilder::STOSOp(OpcodeArgs) {
     Ref Dest = MakeSegmentAddress(X86State::REG_RDI, 0, X86Tables::DecodeFlags::FLAG_ES_PREFIX, true);
 
     // Store to memory where RDI points
-    _StoreMemGPRAutoTSO(Size, Dest, Src, Size);
+    if (CTX->IsMemcpyAtomicTSOEnabled()) {
+      _StoreMemGPRAutoTSO(Size, Dest, Src, Size);
+    } else {
+      _StoreMem(RegClass::GPR, Size, Src, Dest, Invalid(), OpSize::i8Bit, MemOffsetType::SXTX, 1);
+    }
 
     // Offset the pointer
     Ref TailDest = LoadGPRRegister(X86State::REG_RDI);
@@ -3298,10 +3287,15 @@ void OpDispatchBuilder::MOVSOp(OpcodeArgs) {
     Ref RSI = MakeSegmentAddress(X86State::REG_RSI, Op->Flags, X86Tables::DecodeFlags::FLAG_DS_PREFIX);
     Ref RDI = MakeSegmentAddress(X86State::REG_RDI, 0, X86Tables::DecodeFlags::FLAG_ES_PREFIX, true);
 
-    auto Src = _LoadMemGPRAutoTSO(Size, RSI, Size);
+    if (CTX->IsMemcpyAtomicTSOEnabled()) {
+      auto Src = _LoadMemGPRAutoTSO(Size, RSI, Size);
 
-    // Store to memory where RDI points
-    _StoreMemGPRAutoTSO(Size, RDI, Src, Size);
+      // Store to memory where RDI points
+      _StoreMemGPRAutoTSO(Size, RDI, Src, Size);
+    } else {
+      auto Src = _LoadMem(RegClass::GPR, Size, RSI, Invalid(), OpSize::i8Bit, MemOffsetType::SXTX, 1);
+      _StoreMem(RegClass::GPR, Size, Src, RDI, Invalid(), OpSize::i8Bit, MemOffsetType::SXTX, 1);
+    }
 
     auto PtrDir = LoadDir(IR::OpSizeToSize(Size));
     RSI = Add(OpSize::i64Bit, RSI, PtrDir);

FEXCore/Source/Interface/Core/OpcodeDispatcher.h

@@ -1330,7 +1330,6 @@ protected:
 
 private:
   FEX_CONFIG_OPT(ReducedPrecisionMode, X87REDUCEDPRECISION);
-  FEX_CONFIG_OPT(StrictReducedPrecisionMode, X87STRICTREDUCEDPRECISION);
 
   struct JumpTargetInfo {
     Ref BlockEntry;
@@ -2528,7 +2527,7 @@ private:
     const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
 
     // Use ldp if possible, otherwise fallback on two loads.
-    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit & Size <= OpSize::i128Bit) {
+    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit && Size <= OpSize::i128Bit) {
       const auto B = SelectPairAddressMode(A, Size);
       return LoadMemPair(Class, Size, B.Base, B.Offset);
     }
@@ -2567,7 +2566,7 @@ private:
     const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
 
     // Use stp if possible, otherwise fallback on two stores.
-    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit & Size <= OpSize::i128Bit) {
+    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit && Size <= OpSize::i128Bit) {
       const auto B = SelectPairAddressMode(A, Size);
       _StoreMemPair(Class, Size, Value1, Value2, B.Base, B.Offset);
     } else {

FEXCore/Source/Interface/Core/OpcodeDispatcher/AVX_128.cpp

@@ -1956,7 +1956,7 @@ void OpDispatchBuilder::AVX128_VFMAImpl(OpcodeArgs, IROps IROp, uint8_t Src1Idx,
 }
 
 void OpDispatchBuilder::AVX128_VFMAScalarImpl(OpcodeArgs, IROps IROp, uint8_t Src1Idx, uint8_t Src2Idx, uint8_t AddendIdx) {
-  const auto SrcSize = OpSizeFromSrc(Op);
+  const OpSize ElementSize = Op->Flags & X86Tables::DecodeFlags::FLAG_OPTION_AVX_W ? OpSize::i64Bit : OpSize::i32Bit;
 
   auto Dest = AVX128_LoadSource_WithOpSize(Op, Op->Dest, Op->Flags, false).Low;
   auto Src1 = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false).Low;
@@ -1964,13 +1964,13 @@ void OpDispatchBuilder::AVX128_VFMAScalarImpl(OpcodeArgs, IROps IROp, uint8_t Sr
   if (Op->Src[1].IsGPR()) {
     Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false).Low;
   } else {
-    Src2 = LoadSourceFPR_WithOpSize(Op, Op->Src[1], SrcSize, Op->Flags);
+    Src2 = LoadSourceFPR_WithOpSize(Op, Op->Src[1], ElementSize, Op->Flags);
   }
 
   Ref Sources[3] = {Dest, Src1, Src2};
 
   DeriveOp(Result_Low, IROp,
-           _VFMLAScalarInsert(OpSize::i128Bit, SrcSize, Dest, Sources[Src1Idx - 1], Sources[Src2Idx - 1], Sources[AddendIdx - 1]));
+           _VFMLAScalarInsert(OpSize::i128Bit, ElementSize, Dest, Sources[Src1Idx - 1], Sources[Src2Idx - 1], Sources[AddendIdx - 1]));
   AVX128_StoreResult_WithOpSize(Op, Op->Dest, AVX128_Zext(Result_Low));
 }
 

FEXCore/Source/Interface/Core/OpcodeDispatcher/SecondaryTables.h

@@ -145,7 +145,7 @@ constexpr DispatchTableEntry OpDispatch_TwoByteOpTable[] = {
 
 #ifndef _WIN32
   // FEX reserved instructions
-  {0x37, 1, &OpDispatchBuilder::CallbackReturnOp},
+  {0x3E, 1, &OpDispatchBuilder::CallbackReturnOp},
   {0x3F, 1, &OpDispatchBuilder::ThunkOp},
 #endif
 };

FEXCore/Source/Interface/Core/OpcodeDispatcher/X87.cpp

@@ -17,7 +17,6 @@ $end_info$
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/FPState.h>
 
-#include <cmath>
 #include <stddef.h>
 #include <stdint.h>
 
@@ -69,7 +68,7 @@ void OpDispatchBuilder::FLD(OpcodeArgs, IR::OpSize Width) {
   if (Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
     ConvertedData = _F80CVTTo(Data, ReadWidth);
   }
-  _PushStack(ConvertedData, Data, ReadWidth, true);
+  _PushStack(ConvertedData, Data, ReadWidth);
 }
 
 // Float LoaD operation with memory operand
@@ -81,7 +80,7 @@ void OpDispatchBuilder::FBLD(OpcodeArgs) {
   // Read from memory
   Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::f80Bit, Op->Flags);
   Ref ConvertedData = _F80BCDLoad(Data);
-  _PushStack(ConvertedData, Data, OpSize::i128Bit, true);
+  _PushStack(ConvertedData, Data, OpSize::i128Bit);
 }
 
 void OpDispatchBuilder::FBSTP(OpcodeArgs) {
@@ -93,7 +92,7 @@ void OpDispatchBuilder::FBSTP(OpcodeArgs) {
 void OpDispatchBuilder::FLD_Const(OpcodeArgs, NamedVectorConstant K) {
   // Update TOP
   Ref Data = LoadAndCacheNamedVectorConstant(OpSize::i128Bit, K);
-  _PushStack(Data, Data, OpSize::i128Bit, true);
+  _PushStack(Data, Data, OpSize::i128Bit);
 }
 
 void OpDispatchBuilder::FILD(OpcodeArgs) {
@@ -124,7 +123,7 @@ void OpDispatchBuilder::FILD(OpcodeArgs) {
   auto upper = _Or(OpSize::i64Bit, sign, zeroed_exponent);
 
   Ref ConvertedData = _VLoadTwoGPRs(shifted, upper);
-  _PushStack(ConvertedData, Data, ReadWidth, false);
+  _PushStack(ConvertedData, Invalid(), ReadWidth);
 }
 
 void OpDispatchBuilder::FST(OpcodeArgs, IR::OpSize Width) {
@@ -132,7 +131,7 @@ void OpDispatchBuilder::FST(OpcodeArgs, IR::OpSize Width) {
   AddressMode A = DecodeAddress(Op, Op->Dest, MemoryAccessType::DEFAULT, false);
 
   A = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, false, false, Width);
-  _StoreStackMem(SourceSize, Width, A.Base, A.Index, OpSize::iInvalid, A.IndexType, A.IndexScale, /*Float=*/true);
+  _StoreStackMem(SourceSize, Width, A.Base, A.Index, OpSize::iInvalid, A.IndexType, A.IndexScale);
 
   if (Op->TableInfo->Flags & X86Tables::InstFlags::FLAGS_POP) {
     _PopStackDestroy();
@@ -878,8 +877,8 @@ void OpDispatchBuilder::X87FXTRACT(OpcodeArgs) {
   _PopStackDestroy();
   auto Exp = _F80XTRACT_EXP(Top);
   auto Sig = _F80XTRACT_SIG(Top);
-  _PushStack(Exp, Exp, OpSize::f80Bit, true);
-  _PushStack(Sig, Sig, OpSize::f80Bit, true);
+  _PushStack(Exp, Invalid(), OpSize::f80Bit);
+  _PushStack(Sig, Invalid(), OpSize::f80Bit);
 }
 
 } // namespace FEXCore::IR

FEXCore/Source/Interface/Core/OpcodeDispatcher/X87F64.cpp

@@ -68,7 +68,7 @@ void OpDispatchBuilder::FLDF64(OpcodeArgs, IR::OpSize Width) {
   } else if (Width == OpSize::f80Bit) {
     ConvertedData = _F80CVT(OpSize::i64Bit, Data);
   }
-  _PushStack(ConvertedData, Data, ReadWidth, true);
+  _PushStack(ConvertedData, Data, ReadWidth);
 }
 
 void OpDispatchBuilder::FBLDF64(OpcodeArgs) {
@@ -76,7 +76,7 @@ void OpDispatchBuilder::FBLDF64(OpcodeArgs) {
   Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::f80Bit, Op->Flags);
   Ref ConvertedData = _F80BCDLoad(Data);
   ConvertedData = _F80CVT(OpSize::i64Bit, ConvertedData);
-  _PushStack(ConvertedData, Data, OpSize::i64Bit, true);
+  _PushStack(ConvertedData, Data, OpSize::i64Bit);
 }
 
 void OpDispatchBuilder::FBSTPF64(OpcodeArgs) {
@@ -88,7 +88,7 @@ void OpDispatchBuilder::FBSTPF64(OpcodeArgs) {
 
 void OpDispatchBuilder::FLDF64_Const(OpcodeArgs, uint64_t Num) {
   auto Data = _VCastFromGPR(OpSize::i64Bit, OpSize::i64Bit, Constant(Num));
-  _PushStack(Data, Data, OpSize::i64Bit, true);
+  _PushStack(Data, Data, OpSize::i64Bit);
 }
 
 void OpDispatchBuilder::FILDF64(OpcodeArgs) {
@@ -100,7 +100,7 @@ void OpDispatchBuilder::FILDF64(OpcodeArgs) {
     Data = _Sbfe(OpSize::i64Bit, IR::OpSizeAsBits(ReadWidth), 0, Data);
   }
   auto ConvertedData = _Float_FromGPR_S(OpSize::i64Bit, ReadWidth == OpSize::i32Bit ? OpSize::i32Bit : OpSize::i64Bit, Data);
-  _PushStack(ConvertedData, Data, ReadWidth, false);
+  _PushStack(ConvertedData, Invalid(), ReadWidth);
 }
 
 void OpDispatchBuilder::FISTF64(OpcodeArgs, bool Truncate) {
@@ -397,7 +397,7 @@ void OpDispatchBuilder::X87FXTRACTF64(OpcodeArgs) {
   Ref Exp = _NZCVSelectV(OpSize::i64Bit, CondClass::EQ, ExpZV, ExpNZV);
 
   _PopStackDestroy();
-  _PushStack(Exp, Exp, OpSize::i64Bit, true);
-  _PushStack(Sig, Sig, OpSize::i64Bit, true);
+  _PushStack(Exp, Invalid(), OpSize::i64Bit);
+  _PushStack(Sig, Invalid(), OpSize::i64Bit);
 }
 } // namespace FEXCore::IR

FEXCore/Source/Interface/Core/X86HelperGen.cpp

@@ -28,7 +28,7 @@ X86GeneratedCode::X86GeneratedCode() {
   CodePtr = AllocateGuestCodeSpace(CODE_SIZE);
 
   constexpr std::array<uint8_t, 2> SignalReturnCode = {
-    0x0F, 0x37, // CALLBACKRET FEX Instruction
+    0x0F, 0x3E, // CALLBACKRET FEX Instruction
   };
 
   CallbackReturn = reinterpret_cast<uint64_t>(CodePtr);
@@ -51,7 +51,9 @@ void* X86GeneratedCode::AllocateGuestCodeSpace(size_t Size) {
 
   if (Is64BitMode()) {
     // 64bit mode can have its sigret handler anywhere
-    return FEXCore::Allocator::VirtualAlloc(Size);
+    auto Result = FEXCore::Allocator::VirtualAlloc(Size);
+    FEXCore::Allocator::VirtualName("FEXMem_Misc", reinterpret_cast<void*>(Result), Size);
+    return Result;
   }
 
   // First 64bit page

FEXCore/Source/Interface/Core/X86Tables/SecondaryTables.cpp

@@ -100,10 +100,11 @@ constexpr std::array<X86InstInfo, MAX_SECOND_TABLE_SIZE> SecondBaseOps = []() co
     {0x34, 1, X86InstInfo{"SYSENTER",   TYPE_INST, FLAGS_NO_OVERLAY,                                                                             0}},
     {0x35, 1, X86InstInfo{"SYSEXIT",    TYPE_INST, FLAGS_NO_OVERLAY,                                                                             0}},
     {0x36, 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NO_OVERLAY,                                                                          0}},
+    {0x37, 1, X86InstInfo{"GETSEC",     TYPE_INVALID, FLAGS_NO_OVERLAY,                                                                          0}},
     {0x38, 1, X86InstInfo{"",           TYPE_0F38_TABLE, FLAGS_NO_OVERLAY,                                                                       0}},
     {0x39, 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NO_OVERLAY,                                                                          0}},
     {0x3A, 1, X86InstInfo{"",           TYPE_0F3A_TABLE, FLAGS_NO_OVERLAY,                                                                       0}},
-    {0x3B, 4, X86InstInfo{"",           TYPE_INVALID, FLAGS_NO_OVERLAY,                                                                          0}},
+    {0x3B, 3, X86InstInfo{"",           TYPE_INVALID, FLAGS_NO_OVERLAY,                                                                          0}},
 
     {0x40, 1, X86InstInfo{"CMOVO",      TYPE_INST, FLAGS_MODRM | FLAGS_NO_OVERLAY,                                                               0}},
     {0x41, 1, X86InstInfo{"CMOVNO",     TYPE_INST, FLAGS_MODRM | FLAGS_NO_OVERLAY,                                                               0}},
@@ -299,7 +300,7 @@ constexpr std::array<X86InstInfo, MAX_SECOND_TABLE_SIZE> SecondBaseOps = []() co
     // FEX reserved instructions
     // Unused x86 encoding instruction.
 
-    {0x37, 1, X86InstInfo{"CALLBACKRET",  TYPE_INST, FLAGS_BLOCK_END | FLAGS_NO_OVERLAY | FLAGS_SETS_RIP,                                                                          0}},
+    {0x3E, 1, X86InstInfo{"CALLBACKRET",  TYPE_INST, FLAGS_BLOCK_END | FLAGS_NO_OVERLAY | FLAGS_SETS_RIP,                                                                          0}},
 
     // This was originally used by VIA to jump to its alternative instruction set. Used for OP_THUNK
     {0x3F, 1, X86InstInfo{"ALTINST",      TYPE_INST, FLAGS_BLOCK_END | FLAGS_NO_OVERLAY | FLAGS_SETS_RIP,                                                            0}},

FEXCore/Source/Interface/IR/IR.json

@@ -138,7 +138,6 @@
     "FenceType": "FenceType",
     "RegisterClass": "RegClass",
     "CondClass": "CondClass",
-    "SyscallFlags": "FEXCore::IR::SyscallFlags",
     "SHA256Sum": "SHA256Sum",
     "MemOffsetType": "MemOffsetType",
     "BreakDefinition": "BreakDefinition",
@@ -314,25 +313,13 @@
       "CallbackReturn": {
         "HasSideEffects": true
       },
-      "GPR = Syscall GPR:$SyscallID, GPR:$Arg0, GPR:$Arg1, GPR:$Arg2, GPR:$Arg3, GPR:$Arg4, GPR:$Arg5, SyscallFlags:$Flags": {
+      "GPR = Syscall GPR:$SyscallID, GPR:$Arg0, GPR:$Arg1, GPR:$Arg2, GPR:$Arg3, GPR:$Arg4, GPR:$Arg5": {
         "HasSideEffects": true,
         "Desc": ["Dispatches a guest syscall through to the SyscallHandler class"
                 ],
         "DestSize": "OpSize::i64Bit"
       },
 
-      "GPR = InlineSyscall GPR:$Arg0, GPR:$Arg1, GPR:$Arg2, GPR:$Arg3, GPR:$Arg4, GPR:$Arg5, i32:$HostSyscallNumber, SyscallFlags:$Flags": {
-        "HasSideEffects": true,
-        "Desc": ["Dispatches a guest syscall directly to the host syscall interface,",
-                 "bypassing the SyscallHandler class used by Syscall.",
-                 "This has significantly less overhead than Syscall, which needs to save JIT state first.",
-                 "Can only be used for syscalls that match across architecture,",
-                 "such as gettid (matches on x86/x86-64/Arm64)."
-                ],
-
-        "DestSize": "OpSize::i64Bit"
-      },
-
       "Thunk GPR:$ArgPtr, SHA256Sum:$ThunkNameHash": {
         "HasSideEffects": true
       },
@@ -578,8 +565,7 @@
         "Desc": ["Does a x86 TSO compatible load from memory. Offset must be Invalid()."
                 ],
         "Inline": ["", "Memtso"],
-        "DestSize": "Size",
-        "DynamicDispatch": true
+        "DestSize": "Size"
       },
 
       "StoreMemTSO RegisterClass:$Class, OpSize:#Size, SSA:$Value, GPR:$Addr, GPR:$Offset, OpSize:$Align, MemOffsetType:$OffsetType, u8:$OffsetScale": {
@@ -587,8 +573,7 @@
                 ],
         "Inline": ["Zero", "", "Memtso"],
         "HasSideEffects": true,
-        "DestSize": "Size",
-        "DynamicDispatch": true
+        "DestSize": "Size"
       },
 
       "FPR = VLoadVectorMasked OpSize:#RegisterSize, OpSize:#ElementSize, FPR:$Mask, GPR:$Addr, GPR:$Offset, MemOffsetType:$OffsetType, u8:$OffsetScale": {
@@ -2833,17 +2818,13 @@
         "X87": true,
         "HasSideEffects": true
       },
-      "PushStack FPR:$X80Src, SSA:$OriginalValue, OpSize:$LoadSize, i1:$Float": {
+      "PushStack FPR:$X80Src, FPR:$OriginalValue, OpSize:$LoadSize": {
         "Desc": [
           "Pushes the provided X80Src source on to the x87 stack.",
-          "Tracks OriginalValue as the original value of X80Src.",
+          "Tracks OriginalValue as the original value of X80Src. OriginalValue can be Invalid() in which case no tracking is done.",
           "Opsize is 128bit for F80 values, 64-bit for low precision.",
           "LoadSize the original load size, i.e. of size of OriginalValue.",
-          "Float: 80-bit, 64-bit, 32-bit",
-          "Int: 64-bit, 32-bit, 16-bit"
-        ],
-        "EmitValidation": [
-          "WalkFindRegClass($OriginalValue) == RegClass::FPR || WalkFindRegClass($OriginalValue) == RegClass::GPR"
+          "Float: 80-bit, 64-bit, 32-bit"
         ],
         "HasSideEffects": true,
         "X87": true
@@ -2855,13 +2836,12 @@
         "HasSideEffects": true,
         "X87": true
       },
-      "StoreStackMem OpSize:$SourceSize, OpSize:$StoreSize, GPR:$Addr, GPR:$Offset, OpSize:$Align, MemOffsetType:$OffsetType, u8:$OffsetScale, i1:$Float": {
+      "StoreStackMem OpSize:$SourceSize, OpSize:$StoreSize, GPR:$Addr, GPR:$Offset, OpSize:$Align, MemOffsetType:$OffsetType, u8:$OffsetScale": {
         "Desc": [
           "Takes the top value off the x87 stack and stores it to memory.",
           "SourceSize is 128bit for F80 values, 64-bit for low precision.",
           "StoreSize is the store size for conversion:",
-          "Float: 80-bit, 64-bit, or 32-bit",
-          "Int: 64-bit, 32-bit, 16-bit"
+          "Float: 80-bit, 64-bit, or 32-bit"
         ],
         "HasSideEffects": true,
         "X87": true

FEXCore/Source/Interface/IR/IRDumper.cpp

@@ -149,20 +149,6 @@ static void PrintArg(fextl::stringstream* out, const IRListView*, RoundMode Arg)
   }();
 }
 
-static void PrintArg(fextl::stringstream* out, const IRListView*, SyscallFlags Arg) {
-  *out << [Arg] {
-    switch (Arg) {
-    case SyscallFlags::DEFAULT: return "Default";
-    case SyscallFlags::OPTIMIZETHROUGH: return "Optimize Through";
-    case SyscallFlags::NOSYNCSTATEONENTRY: return "No Sync State on Entry";
-    case SyscallFlags::NORETURN: return "No Return";
-    case SyscallFlags::NOSIDEEFFECTS: return "No Side Effects";
-    case SyscallFlags::NORETURNEDRESULT: return "No Returned Result";
-    }
-    return "<Unknown Syscall Flags>";
-  }();
-}
-
 static void PrintArg(fextl::stringstream* out, const IRListView*, NamedVectorConstant Arg) {
   *out << [Arg] {
     // clang-format off

FEXCore/Source/Interface/IR/PassManager.h

@@ -39,7 +39,7 @@ public:
   }
 
 protected:
-  PassManager* Manager;
+  PassManager* Manager {};
 };
 
 class PassManager final {

FEXCore/Source/Interface/IR/Passes/RegisterAllocationPass.cpp

@@ -63,9 +63,9 @@ public:
 private:
   RegisterClassData Classes[IR::NumClasses];
 
-  IREmitter* IREmit;
-  IRListView* IR;
-  const FEXCore::CPUIDEmu* CPUID;
+  IREmitter* IREmit {};
+  IRListView* IR {};
+  const FEXCore::CPUIDEmu* CPUID {};
 
   // Map of nodes to their preferred register, to coalesce load/store reg.
   fextl::vector<PhysicalRegister> PreferredReg;
@@ -83,7 +83,7 @@ private:
   fextl::vector<bool> Seen;
 
   // SourcesNextUses is read backwards, this tracks the index
-  int64_t SourceIndex;
+  int64_t SourceIndex {};
 
   bool Rematerializable(IROp_Header* IROp) {
     return IROp->Op == OP_CONSTANT;
@@ -110,7 +110,7 @@ private:
   // block, so we don't need to size the block up-front.
   fextl::vector<uint32_t> NextUses;
 
-  bool AnySpilled;
+  bool AnySpilled {};
 
   bool IsValidArg(OrderedNodeWrapper Arg) {
     if (Arg.IsInvalid()) {

FEXCore/Source/Interface/IR/Passes/RegisterAllocationPass.h

@@ -19,7 +19,7 @@ public:
   virtual void AddRegisters(RegClass Class, uint32_t RegisterCount) = 0;
 
   // Number of GPRs usable for pairs at start of GPR set. Must be even.
-  uint32_t PairRegs;
+  uint32_t PairRegs {};
 };
 
 } // namespace FEXCore::IR

FEXCore/Source/Interface/IR/Passes/x87StackOptimizationPass.cpp

@@ -6,7 +6,6 @@
 #include "Interface/IR/PassManager.h"
 #include "FEXCore/IR/IR.h"
 #include "FEXCore/Utils/Profiler.h"
-#include "FEXCore/Utils/MathUtils.h"
 #include "FEXCore/Core/HostFeatures.h"
 #include "Interface/Core/Addressing.h"
 
@@ -158,9 +157,7 @@ public:
     : Features(Features)
     , GPROpSize(GPROpSize) {
     FEX_CONFIG_OPT(ReducedPrecision, X87REDUCEDPRECISION);
-    FEX_CONFIG_OPT(StrictReducedPrecision, X87STRICTREDUCEDPRECISION);
     ReducedPrecisionMode = ReducedPrecision;
-    StrictReducedPrecisionMode = StrictReducedPrecision;
   }
   void Run(IREmitter* Emit) override;
 
@@ -168,12 +165,10 @@ private:
   const FEXCore::HostFeatures& Features;
   const OpSize GPROpSize;
   bool ReducedPrecisionMode;
-  bool StrictReducedPrecisionMode;
   FEX_CONFIG_OPT(DisableVixlIndirectCalls, DISABLE_VIXL_INDIRECT_RUNTIME_CALLS);
 
   // Helpers
   Ref RotateRight8(uint32_t V, Ref Amount);
-  Ref SilenceNaN(Ref Value);
 
   void F80SplitStore_Helper(const IROp_StoreStackMem* Op, Ref StackNode) {
     Ref AddrNode = IR->GetNode(Op->Addr);
@@ -208,9 +203,6 @@ private:
     case OpSize::i32Bit:
     case OpSize::i64Bit: {
       StackNode = IREmit->_F80CVT(Op->StoreSize, StackNode);
-      if (!ReducedPrecisionMode || StrictReducedPrecisionMode) {
-        StackNode = SilenceNaN(StackNode);
-      }
       IREmit->_StoreMemFPR(Op->StoreSize, StackNode, AddrNode, Offset, Align, OffsetType, OffsetScale);
       break;
     }
@@ -242,10 +234,6 @@ private:
     MemOffsetType OffsetType = Op->OffsetType;
     uint8_t OffsetScale = Op->OffsetScale;
 
-    if ((!ReducedPrecisionMode || StrictReducedPrecisionMode) && Op->StoreSize != OpSize::f80Bit) {
-      StackNode = SilenceNaN(StackNode);
-    }
-
     switch (Op->StoreSize) {
     case OpSize::i32Bit: {
       StackNode = IREmit->_Float_FToF(OpSize::i32Bit, OpSize::i64Bit, StackNode);
@@ -304,10 +292,9 @@ private:
     StackMemberInfo() {}
     StackMemberInfo(Ref Data)
       : StackDataNode(Data) {}
-    StackMemberInfo(Ref Data, Ref Source, OpSize Size, bool Float)
+    StackMemberInfo(Ref Data, Ref Source, OpSize Size)
       : StackDataNode(Data)
-      , Source({Size, Source})
-      , InterpretAsFloat(Float) {}
+      , Source({Size, Source}) {}
     Ref StackDataNode {}; // Reference to the data in the Stack.
                           // This is the source data node in the stack format, possibly converted to 64/80 bits.
     struct StackMemberData final {
@@ -317,7 +304,6 @@ private:
     // Tuple is only valid if we have information about the Source of the Stack Data Node.
     // In it's valid then OpSize is the original source size and Ref is the original source node.
     std::optional<StackMemberData> Source {};
-    bool InterpretAsFloat {false}; // True if this is a floating point value, false if integer
   };
 
   // StackData, TopCache need to be always properly set to ensure
@@ -499,15 +485,6 @@ inline Ref X87StackOptimization::RotateRight8(uint32_t V, Ref Amount) {
   return IREmit->_Lshr(OpSize::i32Bit, GetConstant(V | (V << 8)), Amount);
 }
 
-inline Ref X87StackOptimization::SilenceNaN(Ref Value) {
-  Ref GPRValue = IREmit->_VExtractToGPR(OpSize::i64Bit, OpSize::i64Bit, Value, 0);
-
-  IREmit->_FCmp(OpSize::i64Bit, Value, Value); // Comparison with itself should set VS if nan
-  Ref QuietNaNGPR = IREmit->_Or(OpSize::i64Bit, GPRValue, IREmit->_Constant(0x0008000000000000ULL));
-  Ref SilencedValue = IREmit->_VCastFromGPR(OpSize::i64Bit, OpSize::i64Bit, QuietNaNGPR);
-  return IREmit->_NZCVSelectV(OpSize::i64Bit, CondClass::VS, SilencedValue, Value);
-}
-
 inline std::optional<X87StackOptimization::StackMemberInfo> X87StackOptimization::MigrateToSlowPath_IfInvalid(uint8_t Offset) {
   const auto& [Valid, StackMember] = StackData.top(Offset);
   MigrateToSlowPathIf(Valid != StackSlot::VALID);
@@ -947,8 +924,13 @@ void X87StackOptimization::Run(IREmitter* Emit) {
           StoreStackValueAtOffset_Slow(SourceNode);
         } else {
           auto* SourceNode = CurrentIR.GetNode(Op->X80Src);
-          auto* OriginalNode = CurrentIR.GetNode(Op->OriginalValue);
-          StackData.push(StackMemberInfo {SourceNode, OriginalNode, Op->LoadSize, Op->Float});
+          if (Op->OriginalValue.IsInvalid()) {
+            // No original value to track - just push the converted data
+            StackData.push(StackMemberInfo {SourceNode});
+          } else {
+            auto* OriginalNode = CurrentIR.GetNode(Op->OriginalValue);
+            StackData.push(StackMemberInfo {SourceNode, OriginalNode, Op->LoadSize});
+          }
         }
         break;
       }
@@ -1013,9 +995,8 @@ void X87StackOptimization::Run(IREmitter* Emit) {
         // str w2, [x1]
         // or similar. As long as the source size and dest size are one and the same.
         // This will avoid any conversions between source and stack element size and conversion back.
-        if (!SlowPath && Value->Source && Value->Source->Size == Op->StoreSize && Value->InterpretAsFloat) {
-          const auto ClassType = Value->InterpretAsFloat ? RegClass::FPR : RegClass::GPR;
-          IREmit->_StoreMem(ClassType, Op->StoreSize, Value->Source->Node, AddrNode, Offset, Align, OffsetType, OffsetScale);
+        if (!SlowPath && Value->Source && Value->Source->Size == Op->StoreSize) {
+          IREmit->_StoreMemFPR(Op->StoreSize, Value->Source->Node, AddrNode, Offset, Align, OffsetType, OffsetScale);
           break;
         }
 
@@ -1055,11 +1036,26 @@ void X87StackOptimization::Run(IREmitter* Emit) {
       case OP_F80STACKXCHANGE: {
         const auto* Op = IROp->C<IROp_F80StackXchange>();
         auto Offset = Op->SrcStack;
-        Ref ValueTop = LoadStackValue();
-        Ref ValueOffset = LoadStackValue(Offset);
 
-        StoreStackValue(ValueOffset);
-        StoreStackValue(ValueTop, Offset);
+        if (Offset == 0) {
+          // No-op
+          break;
+        }
+
+        const auto [ValidTop, StackMemberTop] = StackData.top(0);
+        const auto [ValidOffset, StackMemberOffset] = StackData.top(Offset);
+
+        if (ValidTop != StackSlot::VALID || ValidOffset != StackSlot::VALID) {
+          // Slow path: do actual memory operations
+          Ref ValueTop = LoadStackValue();
+          Ref ValueOffset = LoadStackValue(Offset);
+          StoreStackValue(ValueOffset);
+          StoreStackValue(ValueTop, Offset);
+        } else {
+          // Fast path: swap complete StackMemberInfo preserving Source metadata
+          StackData.setTop(StackMemberOffset, 0);
+          StackData.setTop(StackMemberTop, Offset);
+        }
         break;
       }
 

FEXCore/Source/Utils/Allocator.cpp

@@ -53,10 +53,15 @@ void* FEX_mmap(void* addr, size_t length, int prot, int flags, int fd, off_t off
   }
 
   if (flags & MAP_ANONYMOUS) {
-    prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, Result, length, "FEXMem");
+    VirtualName("FEXMem", Result, length);
   }
   return Result;
 }
+
+void VirtualName(const char* Name, void* Ptr, size_t Size) {
+  prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, Ptr, Size, Name);
+}
+
 int FEX_munmap(void* addr, size_t length) {
   int Result = Alloc64->Munmap(addr, length);
 
@@ -88,7 +93,7 @@ void* DisableSBRKAllocations() {
   // calls won't allocate any memory through that.
   void* AlignedBRK = reinterpret_cast<void*>(FEXCore::AlignUp(reinterpret_cast<uintptr_t>(StartingSBRK), FEXCore::Utils::FEX_PAGE_SIZE));
   void* AfterBRK =
-    mmap(AlignedBRK, FEXCore::Utils::FEX_PAGE_SIZE, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED_NOREPLACE | MAP_NORESERVE, -1, 0);
+    ::mmap(AlignedBRK, FEXCore::Utils::FEX_PAGE_SIZE, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED_NOREPLACE | MAP_NORESERVE, -1, 0);
   if (AfterBRK == INVALID_PTR) {
     // Couldn't allocate the page after the aligned brk? This should never happen.
     // FEXCore::LogMan isn't configured yet so we just need to print the message.
@@ -135,10 +140,7 @@ void ClearHooks() {
   FEXCore::Allocator::mmap = ::mmap;
   FEXCore::Allocator::munmap = ::munmap;
 
-  // XXX: This is currently a leak.
-  // We can't work around this yet until static initializers that allocate memory are completely removed from our codebase
-  // Luckily we only remove this on process shutdown, so the kernel will do the cleanup for us
-  Alloc64.release();
+  Alloc::OSAllocator::ReleaseAllocatorWorkaround(std::move(Alloc64));
 }
 #pragma GCC diagnostic pop
 
@@ -289,7 +291,7 @@ fextl::vector<MemoryRegion> StealMemoryRegion(uintptr_t Begin, uintptr_t End) {
       --StackRegionIt;
 
       auto Alloc =
-        mmap(StackRegionIt->Ptr, StackRegionIt->Size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_NORESERVE | MAP_PRIVATE | MAP_FIXED, -1, 0);
+        ::mmap(StackRegionIt->Ptr, StackRegionIt->Size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_NORESERVE | MAP_PRIVATE | MAP_FIXED, -1, 0);
 
       LogMan::Throw::AFmt(Alloc != MAP_FAILED, "mmap({},{:x}) failed", fmt::ptr(StackRegionIt->Ptr), StackRegionIt->Size);
       LogMan::Throw::AFmt(Alloc == StackRegionIt->Ptr, "mmap returned {} instead of {}", Alloc, fmt::ptr(StackRegionIt->Ptr));
@@ -300,7 +302,7 @@ fextl::vector<MemoryRegion> StealMemoryRegion(uintptr_t Begin, uintptr_t End) {
 
   // Block remaining memory gaps
   for (auto RegionIt = Regions.begin(); RegionIt != Regions.end(); ++RegionIt) {
-    auto Alloc = mmap(RegionIt->Ptr, RegionIt->Size, PROT_NONE, MAP_ANONYMOUS | MAP_NORESERVE | MAP_PRIVATE | MAP_FIXED_NOREPLACE, -1, 0);
+    auto Alloc = ::mmap(RegionIt->Ptr, RegionIt->Size, PROT_NONE, MAP_ANONYMOUS | MAP_NORESERVE | MAP_PRIVATE | MAP_FIXED_NOREPLACE, -1, 0);
 
     LogMan::Throw::AFmt(Alloc != MAP_FAILED, "mmap({},{:x}) failed", fmt::ptr(RegionIt->Ptr), RegionIt->Size);
     LogMan::Throw::AFmt(Alloc == RegionIt->Ptr, "mmap returned {} instead of {}", Alloc, fmt::ptr(RegionIt->Ptr));

FEXCore/Source/Utils/Allocator/64BitAllocator.cpp

@@ -98,7 +98,7 @@ private:
     // Align UsedPages so it pads to the next page.
     // Necessary to take advantage of madvise zero page pooling.
     using FlexBitElementType = uint64_t;
-    alignas(4096) FEXCore::FlexBitSet<FlexBitElementType> UsedPages;
+    alignas(FEXCore::Utils::FEX_PAGE_SIZE) FEXCore::FlexBitSet<FlexBitElementType> UsedPages;
 
     // This returns the size of the LiveVMARegion in addition to the flex set that tracks the used data
     // The LiveVMARegion lives at the start of the VMA region which means on initialization we need to set that
@@ -140,7 +140,7 @@ private:
     }
   };
 
-  static_assert(sizeof(LiveVMARegion) == 4096, "Needs to be the size of a page");
+  static_assert(sizeof(LiveVMARegion) == FEXCore::Utils::FEX_PAGE_SIZE, "Needs to be the size of a page");
 
   static_assert(std::is_trivially_copyable<LiveVMARegion>::value, "Needs to be trivially copyable");
   static_assert(offsetof(LiveVMARegion, UsedPages) == sizeof(LiveVMARegion), "FlexBitSet needs to be at the end");
@@ -168,6 +168,7 @@ private:
     LOGMAN_THROW_A_FMT(Res != -1, "Couldn't mprotect region: {} '{}' Likely occurs when running out of memory or Maximum VMAs", errno,
                        strerror(errno));
 
+    FEXCore::Allocator::VirtualName("FEXMem_Misc", reinterpret_cast<void*>(ReservedRegion->Base), SizePlusManagedData);
     LiveVMARegion* LiveRange = new (reinterpret_cast<void*>(ReservedRegion->Base)) LiveVMARegion();
 
     // Copy over the reserved data
@@ -206,7 +207,7 @@ OSAllocator_64Bit::LiveVMARegion* OSAllocator_64Bit::FindLiveRegionForAddress(ui
     uintptr_t RegionBegin = (*it)->SlabInfo->Base;
     uintptr_t RegionEnd = RegionBegin + (*it)->SlabInfo->RegionSize;
 
-    if (Addr >= RegionBegin && Addr < RegionEnd) {
+    if (Addr >= RegionBegin && AddrEnd < RegionEnd) {
       LiveRegion = *it;
       // Leave our loop
       break;
@@ -404,14 +405,18 @@ again:
     // Mark the pages as used
     uintptr_t RegionBegin = LiveRegion->SlabInfo->Base;
     uintptr_t MappedBegin = (AllocatedOffset - RegionBegin) >> FEXCore::Utils::FEX_PAGE_SHIFT;
+    size_t PagesSet {};
 
     for (size_t i = 0; i < NumberOfPages; ++i) {
-      LiveRegion->UsedPages.Set(MappedBegin + i);
+      PagesSet += LiveRegion->UsedPages.TestAndSet(MappedBegin + i) == false;
     }
 
     // Change our last allocation region
     LiveRegion->LastPageAllocation = MappedBegin + NumberOfPages;
-    LiveRegion->FreeSpace -= length;
+    LiveRegion->FreeSpace -= PagesSet * FEXCore::Utils::FEX_PAGE_SIZE;
+    LOGMAN_THROW_A_FMT(LiveRegion->FreeSpace <= LiveRegion->SlabInfo->RegionSize,
+                       "Corrupt LiveRegion free space! 0x{:x} > 0x{:x}. After allocating 0x{:x} (0x{:x} overlapped)", LiveRegion->FreeSpace,
+                       LiveRegion->SlabInfo->RegionSize, length, PagesSet);
   }
 
   if (!AllocatedOffset) {
@@ -473,7 +478,7 @@ int OSAllocator_64Bit::Munmap(void* addr, size_t length) {
         ::mmap(addr, length, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
       }
 
-      (*it)->FreeSpace += FreedPages * 4096;
+      (*it)->FreeSpace += FreedPages * FEXCore::Utils::FEX_PAGE_SIZE;
 
       // Set the last allocated page to the minimum of last page allocation or this slab
       // This will let us more quickly fill holes
@@ -505,6 +510,8 @@ void OSAllocator_64Bit::AllocateMemoryRegions(fextl::vector<FEXCore::Allocator::
     // This enables the kernel to use transparent large pages in the allocator which can reduce memory pressure
     ::madvise(it.Ptr, ObjectAllocSize, MADV_HUGEPAGE);
 
+    FEXCore::Allocator::VirtualName("FEXMem_Misc", reinterpret_cast<void*>(it.Ptr), ObjectAllocSize);
+
     ObjectAlloc = new (it.Ptr) Alloc::ForwardOnlyIntrusiveArenaAllocator(it.Ptr, ObjectAllocSize);
     ReservedRegions = ObjectAlloc->new_construct(ReservedRegions, ObjectAlloc);
     LiveRegions = ObjectAlloc->new_construct(LiveRegions, ObjectAlloc);
@@ -602,6 +609,8 @@ fextl::unique_ptr<T> make_alloc_unique(FEXCore::Allocator::MemoryRegion& Base, A
     ERROR_AND_DIE_FMT("Couldn't allocate memory region");
   }
 
+  FEXCore::Allocator::VirtualName("FEXMem_Misc", reinterpret_cast<void*>(ptr), MinPage);
+
   // Remove the page from the base region.
   // Could be zero after this.
   Base.Size -= MinPage;

FEXCore/Source/Utils/Allocator/FlexBitSet.h

@@ -27,6 +27,11 @@ struct FlexBitSet final {
     Memory[Element / MinimumSizeBits] &= ~(1ULL << (Element % MinimumSizeBits));
     return Value;
   }
+  bool TestAndSet(size_t Element) {
+    bool Value = Get(Element);
+    Memory[Element / MinimumSizeBits] |= (1ULL << (Element % MinimumSizeBits));
+    return Value;
+  }
   void Set(size_t Element) {
     Memory[Element / MinimumSizeBits] |= (1ULL << (Element % MinimumSizeBits));
   }
@@ -70,12 +75,17 @@ struct FlexBitSet final {
   template<bool WantUnset>
   BitsetScanResults BackwardScanForRange(size_t BeginningElement, size_t ElementCount, size_t MinimumElement) {
     bool FoundHole {};
-    for (size_t CurrentPage = BeginningElement; CurrentPage >= (MinimumElement + ElementCount);) {
+
+    // Final element to iterate to.
+    const size_t FinalElement = MinimumElement + ElementCount - 1;
+
+    for (size_t CurrentPage = BeginningElement; CurrentPage >= FinalElement;) {
       size_t Remaining = ElementCount;
-      LOGMAN_THROW_A_FMT(Remaining <= CurrentPage, "Scanning less than available range");
+      LOGMAN_THROW_A_FMT(CurrentPage <= BeginningElement && CurrentPage >= FinalElement, "BackwardScanForRange: Scanning less than "
+                                                                                         "available range");
 
       while (Remaining) {
-        if (this->Get(CurrentPage - Remaining) == WantUnset) {
+        if (this->Get(CurrentPage - Remaining + 1) == WantUnset) {
           // Has an intersecting range
           break;
         }
@@ -92,7 +102,7 @@ struct FlexBitSet final {
         CurrentPage -= Remaining;
       } else {
         // We have a slab range
-        return BitsetScanResults {CurrentPage - ElementCount, FoundHole};
+        return BitsetScanResults {CurrentPage - ElementCount + 1, FoundHole};
       }
     }
 
@@ -108,11 +118,15 @@ struct FlexBitSet final {
   BitsetScanResults ForwardScanForRange(size_t BeginningElement, size_t ElementCount, size_t ElementsInSet) {
     bool FoundHole {};
 
-    for (size_t CurrentElement = BeginningElement; CurrentElement < (ElementsInSet - ElementCount);) {
+    // Final element to iterate to.
+    const size_t FinalElement = ElementsInSet - ElementCount + 1;
+
+    for (size_t CurrentElement = BeginningElement; CurrentElement <= FinalElement;) {
       // If we have enough free space, check if we have enough free pages that are contiguous
       size_t Remaining = ElementCount;
 
-      LOGMAN_THROW_A_FMT((CurrentElement + Remaining - 1) < ElementsInSet, "Scanning less than available range");
+      LOGMAN_THROW_A_FMT(CurrentElement >= BeginningElement && CurrentElement <= FinalElement, "ForwardScanForRange: Scanning less than "
+                                                                                               "available range");
 
       while (Remaining) {
         if (this->Get(CurrentElement + Remaining - 1) == WantUnset) {

FEXCore/Source/Utils/Allocator/HostAllocator.h

@@ -53,4 +53,12 @@ public:
 namespace Alloc::OSAllocator {
 fextl::unique_ptr<Alloc::HostAllocator> Create64BitAllocator();
 fextl::unique_ptr<Alloc::HostAllocator> Create64BitAllocatorWithRegions(fextl::vector<FEXCore::Allocator::MemoryRegion>& Regions);
+static inline void ReleaseAllocatorWorkaround(fextl::unique_ptr<Alloc::HostAllocator> Allocator) {
+  // XXX: This is currently a leak.
+  // We can't work around this yet until static initializers that allocate memory are completely removed from our codebase
+  // The allocator is also intrusively allocated, so the unique_ptr tries to double free the HostAllocator object.
+  // Luckily we only remove this on process shutdown, so the kernel will do the cleanup for us
+  Allocator.release();
+}
+
 } // namespace Alloc::OSAllocator

FEXCore/Source/Utils/ArchHelpers/Arm64.cpp

@@ -144,33 +144,33 @@ static __uint128_t LoadAcquire128(uint64_t Addr) {
 }
 
 static uint64_t LoadAcquire64(uint64_t Addr) {
-  std::atomic<uint64_t>* Atom = reinterpret_cast<std::atomic<uint64_t>*>(Addr);
-  return Atom->load(std::memory_order_acquire);
+  auto Atom = std::atomic_ref<uint64_t>(*reinterpret_cast<uint64_t*>(Addr));
+  return Atom.load(std::memory_order_acquire);
 }
 
 static bool StoreCAS64(uint64_t& Expected, uint64_t Val, uint64_t Addr) {
-  std::atomic<uint64_t>* Atom = reinterpret_cast<std::atomic<uint64_t>*>(Addr);
-  return Atom->compare_exchange_strong(Expected, Val);
+  auto Atom = std::atomic_ref<uint64_t>(*reinterpret_cast<uint64_t*>(Addr));
+  return Atom.compare_exchange_strong(Expected, Val);
 }
 
 static uint32_t LoadAcquire32(uint64_t Addr) {
-  std::atomic<uint32_t>* Atom = reinterpret_cast<std::atomic<uint32_t>*>(Addr);
-  return Atom->load(std::memory_order_acquire);
+  auto Atom = std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(Addr));
+  return Atom.load(std::memory_order_acquire);
 }
 
 static bool StoreCAS32(uint32_t& Expected, uint32_t Val, uint64_t Addr) {
-  std::atomic<uint32_t>* Atom = reinterpret_cast<std::atomic<uint32_t>*>(Addr);
-  return Atom->compare_exchange_strong(Expected, Val);
+  auto Atom = std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(Addr));
+  return Atom.compare_exchange_strong(Expected, Val);
 }
 
 static uint8_t LoadAcquire8(uint64_t Addr) {
-  std::atomic<uint8_t>* Atom = reinterpret_cast<std::atomic<uint8_t>*>(Addr);
-  return Atom->load(std::memory_order_acquire);
+  auto Atom = std::atomic_ref<uint8_t>(*reinterpret_cast<uint8_t*>(Addr));
+  return Atom.load(std::memory_order_acquire);
 }
 
 static bool StoreCAS8(uint8_t& Expected, uint8_t Val, uint64_t Addr) {
-  std::atomic<uint8_t>* Atom = reinterpret_cast<std::atomic<uint8_t>*>(Addr);
-  return Atom->compare_exchange_strong(Expected, Val);
+  auto Atom = std::atomic_ref<uint8_t>(*reinterpret_cast<uint8_t*>(Addr));
+  return Atom.compare_exchange_strong(Expected, Val);
 }
 
 static uint16_t DoLoad16(uint64_t Addr) {
@@ -211,8 +211,8 @@ static uint16_t DoLoad16(uint64_t Addr) {
         uint64_t Alignment = Addr & AlignmentMask;
         Addr &= ~AlignmentMask;
 
-        std::atomic<uint64_t>* Atomic = reinterpret_cast<std::atomic<uint64_t>*>(Addr);
-        uint64_t TmpResult = Atomic->load();
+        auto Atomic = std::atomic_ref<uint64_t>(*reinterpret_cast<uint64_t*>(Addr));
+        uint64_t TmpResult = Atomic.load();
 
         // Zexts the result
         uint16_t Result = TmpResult >> (Alignment * 8);
@@ -224,8 +224,8 @@ static uint16_t DoLoad16(uint64_t Addr) {
         uint64_t Alignment = Addr & AlignmentMask;
         Addr &= ~AlignmentMask;
 
-        std::atomic<uint32_t>* Atomic = reinterpret_cast<std::atomic<uint32_t>*>(Addr);
-        uint32_t TmpResult = Atomic->load();
+        auto Atomic = std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(Addr));
+        uint32_t TmpResult = Atomic.load();
 
         // Zexts the result
         uint16_t Result = TmpResult >> (Alignment * 8);
@@ -272,8 +272,8 @@ static uint32_t DoLoad32(uint64_t Addr) {
       uint64_t Alignment = Addr & AlignmentMask;
       Addr &= ~AlignmentMask;
 
-      std::atomic<uint64_t>* Atomic = reinterpret_cast<std::atomic<uint64_t>*>(Addr);
-      uint64_t TmpResult = Atomic->load();
+      auto Atomic = std::atomic_ref<uint64_t>(*reinterpret_cast<uint64_t*>(Addr));
+      uint64_t TmpResult = Atomic.load();
 
       return TmpResult >> (Alignment * 8);
     }
@@ -465,7 +465,7 @@ static bool RunCASPAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg1, uint3
       // Fits within a 16byte region
       uint64_t Alignment = Addr & 0b1111;
       Addr &= ~0b1111ULL;
-      std::atomic<__uint128_t>* Atomic128 = reinterpret_cast<std::atomic<__uint128_t>*>(Addr);
+      auto Atomic128 = std::atomic_ref<__uint128_t>(*reinterpret_cast<__uint128_t*>(Addr));
 
       __uint128_t Mask = ~0ULL;
       Mask <<= Alignment * 8;
@@ -480,7 +480,7 @@ static bool RunCASPAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg1, uint3
       Expected <<= Alignment * 8;
 
       while (1) {
-        TmpExpected = Atomic128->load();
+        TmpExpected = Atomic128.load();
 
         // Set up expected
         TmpExpected &= NegMask;
@@ -491,7 +491,7 @@ static bool RunCASPAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg1, uint3
         TmpDesired &= NegMask;
         TmpDesired |= Desired;
 
-        bool CASResult = Atomic128->compare_exchange_strong(TmpExpected, TmpDesired);
+        bool CASResult = Atomic128.compare_exchange_strong(TmpExpected, TmpDesired);
         if (CASResult) {
           // Successful, so we are done
           return true;
@@ -617,36 +617,6 @@ static uint64_t HandleCASPAL_ARMv8(uint32_t Instr, uintptr_t ProgramCounter, uin
   }
 }
 
-static bool HandleAtomicVectorStore(uint32_t Instr, uintptr_t ProgramCounter) {
-  uint32_t* PC = (uint32_t*)ProgramCounter;
-
-  uint32_t Size = (Instr >> 30) & 1;
-  uint32_t DataReg = Instr & 0x1F;
-
-  if (Size == 1) {
-    // 64-bit pair happens on paranoid vector stores
-    // [0] ldaxp(xzr, TMP3, MemSrc); // <- Can hit SIGBUS. Overwritten with DMB
-    // [1] stlxp(TMP3, TMP1, TMP2, MemSrc); // <- Can also hit SIGBUS
-    // [2] cbnz(TMP3, &B); // < Overwritten with DMB
-    if (DataReg == 31) {
-      uint32_t NextInstr = PC[1];
-      uint32_t AddrReg = (NextInstr >> 5) & 0x1F;
-      DataReg = NextInstr & 0x1F;
-      uint32_t DataReg2 = (NextInstr >> 10) & 0x1F;
-      uint32_t STP = (0b10 << 30) | (0b101001000000000 << 15) | (DataReg2 << 10) | (AddrReg << 5) | DataReg;
-
-      PC[0] = DMB;
-      PC[1] = STP;
-      PC[2] = DMB;
-      // Back up one instruction and have another go
-      ClearICache(&PC[0], 12);
-      return true;
-    }
-  }
-
-  return false;
-}
-
 template<typename T>
 using CASExpectedFn = T (*)(T Src, T Expected);
 template<typename T>
@@ -740,7 +710,7 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
       // Fits within a 16byte region
       uint64_t Alignment = Addr & 0b1111;
       Addr &= ~0b1111ULL;
-      std::atomic<__uint128_t>* Atomic128 = reinterpret_cast<std::atomic<__uint128_t>*>(Addr);
+      auto Atomic128 = std::atomic_ref<__uint128_t>(*reinterpret_cast<__uint128_t*>(Addr));
 
       __uint128_t Mask = 0xFFFF;
       Mask <<= Alignment * 8;
@@ -749,7 +719,7 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
       __uint128_t TmpDesired {};
 
       while (1) {
-        TmpExpected = Atomic128->load();
+        TmpExpected = Atomic128.load();
 
         __uint128_t Desired = DesiredFunction(TmpExpected >> (Alignment * 8), DesiredSrc);
         Desired <<= Alignment * 8;
@@ -766,7 +736,7 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
         TmpDesired &= NegMask;
         TmpDesired |= Desired;
 
-        bool CASResult = Atomic128->compare_exchange_strong(TmpExpected, TmpDesired);
+        bool CASResult = Atomic128.compare_exchange_strong(TmpExpected, TmpDesired);
         if (CASResult) {
           // Successful, so we are done
           return Expected >> (Alignment * 8);
@@ -810,9 +780,9 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
         uint64_t TmpExpected {};
         uint64_t TmpDesired {};
 
-        std::atomic<uint64_t>* Atomic = reinterpret_cast<std::atomic<uint64_t>*>(Addr);
+        auto Atomic = std::atomic_ref<uint64_t>(*reinterpret_cast<uint64_t*>(Addr));
         while (1) {
-          TmpExpected = Atomic->load();
+          TmpExpected = Atomic.load();
 
           uint64_t Desired = DesiredFunction(TmpExpected >> (Alignment * 8), DesiredSrc);
           Desired <<= Alignment * 8;
@@ -829,7 +799,7 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
           TmpDesired &= NegMask;
           TmpDesired |= Desired;
 
-          bool CASResult = Atomic->compare_exchange_strong(TmpExpected, TmpDesired);
+          bool CASResult = Atomic.compare_exchange_strong(TmpExpected, TmpDesired);
           if (CASResult) {
             // Successful, so we are done
             return Expected >> (Alignment * 8);
@@ -873,9 +843,9 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
         uint32_t TmpExpected {};
         uint32_t TmpDesired {};
 
-        std::atomic<uint32_t>* Atomic = reinterpret_cast<std::atomic<uint32_t>*>(Addr);
+        auto Atomic = std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(Addr));
         while (1) {
-          TmpExpected = Atomic->load();
+          TmpExpected = Atomic.load();
 
 
           uint32_t Desired = DesiredFunction(TmpExpected >> (Alignment * 8), DesiredSrc);
@@ -893,7 +863,7 @@ static uint16_t DoCAS16(uint16_t DesiredSrc, uint16_t ExpectedSrc, uint64_t Addr
           TmpDesired &= NegMask;
           TmpDesired |= Desired;
 
-          bool CASResult = Atomic->compare_exchange_strong(TmpExpected, TmpDesired);
+          bool CASResult = Atomic.compare_exchange_strong(TmpExpected, TmpDesired);
           if (CASResult) {
             // Successful, so we are done
             return Expected >> (Alignment * 8);
@@ -1040,7 +1010,7 @@ static uint32_t DoCAS32(uint32_t DesiredSrc, uint32_t ExpectedSrc, uint64_t Addr
       // Fits within a 16byte region
       uint64_t Alignment = Addr & 0b1111;
       Addr &= ~0b1111ULL;
-      std::atomic<__uint128_t>* Atomic128 = reinterpret_cast<std::atomic<__uint128_t>*>(Addr);
+      auto Atomic128 = std::atomic_ref<__uint128_t>(*reinterpret_cast<__uint128_t*>(Addr));
 
       __uint128_t Mask = ~0U;
       Mask <<= Alignment * 8;
@@ -1049,7 +1019,7 @@ static uint32_t DoCAS32(uint32_t DesiredSrc, uint32_t ExpectedSrc, uint64_t Addr
       __uint128_t TmpDesired {};
 
       while (1) {
-        __uint128_t TmpActual = Atomic128->load();
+        __uint128_t TmpActual = Atomic128.load();
 
         __uint128_t Desired = DesiredFunction(TmpActual >> (Alignment * 8), DesiredSrc);
         __uint128_t Expected = ExpectedFunction(TmpActual >> (Alignment * 8), ExpectedSrc);
@@ -1064,7 +1034,7 @@ static uint32_t DoCAS32(uint32_t DesiredSrc, uint32_t ExpectedSrc, uint64_t Addr
         TmpDesired &= NegMask;
         TmpDesired |= Desired << (Alignment * 8);
 
-        bool CASResult = Atomic128->compare_exchange_strong(TmpExpected, TmpDesired);
+        bool CASResult = Atomic128.compare_exchange_strong(TmpExpected, TmpDesired);
         if (CASResult) {
           // Stored successfully
           return Expected;
@@ -1108,9 +1078,9 @@ static uint32_t DoCAS32(uint32_t DesiredSrc, uint32_t ExpectedSrc, uint64_t Addr
       uint64_t TmpExpected {};
       uint64_t TmpDesired {};
 
-      std::atomic<uint64_t>* Atomic = reinterpret_cast<std::atomic<uint64_t>*>(Addr);
+      auto Atomic = std::atomic_ref<uint64_t>(*reinterpret_cast<uint64_t*>(Addr));
       while (1) {
-        uint64_t TmpActual = Atomic->load();
+        uint64_t TmpActual = Atomic.load();
 
         uint64_t Desired = DesiredFunction(TmpActual >> (Alignment * 8), DesiredSrc);
         uint64_t Expected = ExpectedFunction(TmpActual >> (Alignment * 8), ExpectedSrc);
@@ -1125,7 +1095,7 @@ static uint32_t DoCAS32(uint32_t DesiredSrc, uint32_t ExpectedSrc, uint64_t Addr
         TmpDesired &= NegMask;
         TmpDesired |= Desired << (Alignment * 8);
 
-        bool CASResult = Atomic->compare_exchange_strong(TmpExpected, TmpDesired);
+        bool CASResult = Atomic.compare_exchange_strong(TmpExpected, TmpDesired);
         if (CASResult) {
           // Stored successfully
           return Expected;
@@ -1270,7 +1240,7 @@ static uint64_t DoCAS64(uint64_t DesiredSrc, uint64_t ExpectedSrc, uint64_t Addr
     // Fits within a 16byte region
     uint64_t Alignment = Addr & AlignmentMask;
     Addr &= ~AlignmentMask;
-    std::atomic<__uint128_t>* Atomic128 = reinterpret_cast<std::atomic<__uint128_t>*>(Addr);
+    auto Atomic128 = std::atomic_ref<__uint128_t>(*reinterpret_cast<__uint128_t*>(Addr));
 
     __uint128_t Mask = ~0ULL;
     Mask <<= Alignment * 8;
@@ -1279,7 +1249,7 @@ static uint64_t DoCAS64(uint64_t DesiredSrc, uint64_t ExpectedSrc, uint64_t Addr
     __uint128_t TmpDesired {};
 
     while (1) {
-      __uint128_t TmpActual = Atomic128->load();
+      __uint128_t TmpActual = Atomic128.load();
 
       __uint128_t Desired = DesiredFunction(TmpActual >> (Alignment * 8), DesiredSrc);
       __uint128_t Expected = ExpectedFunction(TmpActual >> (Alignment * 8), ExpectedSrc);
@@ -1294,7 +1264,7 @@ static uint64_t DoCAS64(uint64_t DesiredSrc, uint64_t ExpectedSrc, uint64_t Addr
       TmpDesired &= NegMask;
       TmpDesired |= Desired << (Alignment * 8);
 
-      bool CASResult = Atomic128->compare_exchange_strong(TmpExpected, TmpDesired);
+      bool CASResult = Atomic128.compare_exchange_strong(TmpExpected, TmpDesired);
       if (CASResult) {
         // Stored successfully
         return Expected;
@@ -1326,9 +1296,7 @@ static uint64_t DoCAS64(uint64_t DesiredSrc, uint64_t ExpectedSrc, uint64_t Addr
   }
 }
 
-static bool RunCASAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg, uint32_t ExpectedReg, uint32_t AddressReg, uint32_t* StrictSplitLockMutex) {
-  uint64_t Addr = GPRs[AddressReg];
-
+static std::optional<uint64_t> DoCAS(uint32_t Size, uint64_t Desired, uint64_t Expected, uint64_t Addr, uint32_t* StrictSplitLockMutex) {
   // Cross-cacheline CAS doesn't work on ARM
   // It isn't even guaranteed to work on x86
   // Intel will do a "split lock" which locks the full bus
@@ -1341,7 +1309,7 @@ static bool RunCASAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg, uint32_
   // Only need to handle 16, 32, 64
   if (Size == 2) {
     auto Res = DoCAS16<false>(
-      GPRs[DesiredReg], GPRs[ExpectedReg], Addr,
+      Desired, Expected, Addr,
       [](uint16_t, uint16_t Expected) -> uint16_t {
         // Expected is just Expected
         return Expected;
@@ -1351,16 +1319,10 @@ static bool RunCASAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg, uint32_
         return Desired;
       },
       StrictSplitLockMutex);
-
-    // Regardless of pass or fail
-    // We set the result register if it isn't a zero register
-    if (ExpectedReg != 31) {
-      GPRs[ExpectedReg] = Res;
-    }
-    return true;
+    return Res;
   } else if (Size == 4) {
     auto Res = DoCAS32<false>(
-      GPRs[DesiredReg], GPRs[ExpectedReg], Addr,
+      Desired, Expected, Addr,
       [](uint32_t, uint32_t Expected) -> uint32_t {
         // Expected is just Expected
         return Expected;
@@ -1370,16 +1332,10 @@ static bool RunCASAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg, uint32_
         return Desired;
       },
       StrictSplitLockMutex);
-
-    // Regardless of pass or fail
-    // We set the result register if it isn't a zero register
-    if (ExpectedReg != 31) {
-      GPRs[ExpectedReg] = Res;
-    }
-    return true;
+    return Res;
   } else if (Size == 8) {
     auto Res = DoCAS64<false>(
-      GPRs[DesiredReg], GPRs[ExpectedReg], Addr,
+      Desired, Expected, Addr,
       [](uint64_t, uint64_t Expected) -> uint64_t {
         // Expected is just Expected
         return Expected;
@@ -1389,16 +1345,24 @@ static bool RunCASAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg, uint32_
         return Desired;
       },
       StrictSplitLockMutex);
-
-    // Regardless of pass or fail
-    // We set the result register if it isn't a zero register
-    if (ExpectedReg != 31) {
-      GPRs[ExpectedReg] = Res;
-    }
-    return true;
+    return Res;
   }
 
-  return false;
+  return std::nullopt;
+}
+
+static bool RunCASAL(uint64_t* GPRs, uint32_t Size, uint32_t DesiredReg, uint32_t ExpectedReg, uint32_t AddressReg, uint32_t* StrictSplitLockMutex) {
+  std::optional<uint64_t> Res = DoCAS(Size, GPRs[DesiredReg], GPRs[ExpectedReg], GPRs[AddressReg], StrictSplitLockMutex);
+  if (!Res.has_value()) {
+    return false;
+  }
+
+  // Regardless of pass or fail
+  // We set the result register if it isn't a zero register
+  if (ExpectedReg != 31) {
+    GPRs[ExpectedReg] = *Res;
+  }
+  return true;
 }
 
 static bool HandleCASAL(uint64_t* GPRs, uint32_t Instr, uint32_t* StrictSplitLockMutex) {
@@ -1560,38 +1524,43 @@ static bool HandleAtomicMemOp(uint32_t Instr, uint64_t* GPRs, uint32_t* StrictSp
   return false;
 }
 
-static bool HandleAtomicLoad(uint32_t Instr, uint64_t* GPRs, int64_t Offset) {
+static bool HandleAtomicLoad(uint32_t Instr, uint64_t* GPRs, int64_t Offset, Core::UnalignedExclusiveStore* Store = nullptr) {
   uint32_t Size = 1 << (Instr >> 30);
 
   uint32_t ResultReg = Instr & 0b11111;
   uint32_t AddressReg = (Instr >> 5) & 0b11111;
 
   uint64_t Addr = GPRs[AddressReg] + Offset;
+  uint64_t Res;
 
   if (Size == 2) {
-    auto Res = DoLoad16(Addr);
+    Res = DoLoad16(Addr);
     // We set the result register if it isn't a zero register
     if (ResultReg != 31) {
       GPRs[ResultReg] = Res;
     }
-    return true;
   } else if (Size == 4) {
-    auto Res = DoLoad32(Addr);
+    Res = DoLoad32(Addr);
     // We set the result register if it isn't a zero register
     if (ResultReg != 31) {
       GPRs[ResultReg] = Res;
     }
-    return true;
   } else if (Size == 8) {
-    auto Res = DoLoad64(Addr);
+    Res = DoLoad64(Addr);
     // We set the result register if it isn't a zero register
     if (ResultReg != 31) {
       GPRs[ResultReg] = Res;
     }
-    return true;
+  } else {
+    return false;
   }
 
-  return false;
+  if (Store) {
+    Store->Addr = Addr;
+    Store->Store = Res;
+    Store->Size = Size;
+  }
+  return true;
 }
 
 static bool HandleAtomicStore(uint32_t Instr, uint64_t* GPRs, int64_t Offset, uint32_t* StrictSplitLockMutex) {
@@ -1952,8 +1921,8 @@ static uint64_t HandleAtomicLoadstoreExclusive(uintptr_t ProgramCounter, uint64_
 }
 
 [[nodiscard]]
-std::optional<int32_t>
-HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandlerType HandleType, uintptr_t ProgramCounter, uint64_t* GPRs) {
+std::optional<int32_t> HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandlerType HandleType,
+                                             uintptr_t ProgramCounter, uint64_t* GPRs, bool IsJIT) {
 #ifdef _M_ARM_64
   constexpr bool is_arm64 = true;
 #else
@@ -1977,8 +1946,7 @@ HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandl
   auto CTX = static_cast<Context::ContextImpl*>(Thread->CTX);
   uint32_t* StrictSplitLockMutex {CTX->Config.StrictInProcessSplitLocks ? &CTX->StrictSplitLockMutex : nullptr};
 
-  // ParanoidTSO path doesn't modify any code.
-  if (HandleType == UnalignedHandlerType::Paranoid) [[unlikely]] {
+  if (!IsJIT) [[unlikely]] {
     if ((Instr & LDAXR_MASK) == LDAR_INST ||  // LDAR*
         (Instr & LDAXR_MASK) == LDAPR_INST) { // LDAPR*
       if (ArchHelpers::Arm64::HandleAtomicLoad(Instr, GPRs, 0)) {
@@ -2016,7 +1984,29 @@ HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandl
         LogMan::Msg::EFmt("Unhandled JIT SIGBUS LDLUR*: PC: 0x{:x} Instruction: 0x{:08x}\n", ProgramCounter, PC[0]);
         return std::nullopt;
       }
+    } else if ((Instr & ArchHelpers::Arm64::LDAXR_MASK) == ArchHelpers::Arm64::LDAXR_INST) { // LDAXR*
+      if (ArchHelpers::Arm64::HandleAtomicLoad(Instr, GPRs, 0, &Thread->ExclusiveStore)) {
+        return 4;
+      }
+    } else if ((Instr & ArchHelpers::Arm64::STLXR_MASK) == ArchHelpers::Arm64::STLXR_INST) { // STLXR*
+      uint32_t StatusReg = Instr << 11 >> 27;
+      // // Emulate exclusive store by validating the address and value against the last unaligned LDAXR*.
+      if (GPRs[AddrReg] != Thread->ExclusiveStore.Addr || Size > Thread->ExclusiveStore.Size) {
+        if (StatusReg != 31) {
+          GPRs[StatusReg] = 1;
+        }
+        return 4;
+      }
+      if (std::optional<uint64_t> Prev =
+            DoCAS(Size, DataReg == 31 ? 0 : GPRs[DataReg], Thread->ExclusiveStore.Store, GPRs[AddrReg], StrictSplitLockMutex)) {
+        if (StatusReg != 31) {
+          GPRs[StatusReg] = !!memcmp(&Thread->ExclusiveStore.Store, &*Prev, Size);
+        }
+        Thread->ExclusiveStore.Size = 0;
+        return 4;
+      }
     }
+    return 0;
   }
 
   const auto Frame = Thread->CurrentFrame;
@@ -2068,6 +2058,9 @@ HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandl
     if (BytesToSkip) {
       // Skip this instruction now
       return BytesToSkip;
+    } else {
+      LogMan::Msg::EFmt("Unhandled JIT SIGBUS CASPAL: PC: 0x{:x} Instruction: 0x{:08x}\n", ProgramCounter, PC[0]);
+      return std::nullopt;
     }
   }
 
@@ -2131,19 +2124,6 @@ HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandl
     ClearICache(&PC[-1], 8);
     // Back up one instruction and have another go
     return -4;
-  } else if ((Instr & ArchHelpers::Arm64::LDAXP_MASK) == ArchHelpers::Arm64::LDAXP_INST) { // LDAXP
-    /// This is handling the case of paranoid ARMv8.0-a atomic stores.
-    /// This backpatches the ldaxp+stlxp+cbnz if the previous `HandleCASPAL_ARMv8` didn't handle the case.
-    if (ArchHelpers::Arm64::HandleAtomicVectorStore(Instr, ProgramCounter)) {
-      return 0;
-    } else {
-      LogMan::Msg::EFmt("Unhandled JIT SIGBUS LDAXP: PC: 0x{:x} Instruction: 0x{:08x}\n", ProgramCounter, PC[0]);
-      return std::nullopt;
-    }
-  } else if ((Instr & ArchHelpers::Arm64::STLXP_MASK) == ArchHelpers::Arm64::STLXP_INST) { // STLXP
-    // Should not trigger - middle of an LDAXP/STAXP pair.
-    LogMan::Msg::EFmt("Unhandled JIT SIGBUS STLXP: PC: 0x{:x} Instruction: 0x{:08x}\n", ProgramCounter, PC[0]);
-    return std::nullopt;
   }
 
   // Check if another thread backpatched this instruction before this thread got here

FEXCore/Source/Utils/LongJump.cpp

@@ -0,0 +1,119 @@
+// SPDX-License-Identifier: MIT
+#include <FEXCore/Utils/LongJump.h>
+
+namespace FEXCore::UncheckedLongJump {
+#if defined(_M_ARM_64)
+[[nodiscard]]
+FEX_DEFAULT_VISIBILITY FEX_NAKED uint64_t SetJump(JumpBuf& Buffer) {
+  __asm volatile(R"(
+      // x0 contains the jumpbuffer
+      stp x19, x20, [x0, #( 0 * 8)];
+      stp x21, x22, [x0, #( 2 * 8)];
+      stp x23, x24, [x0, #( 4 * 8)];
+      stp x25, x26, [x0, #( 6 * 8)];
+      stp x27, x28, [x0, #( 8 * 8)];
+      stp x29, x30, [x0, #(10 * 8)];
+
+      // FPRs
+      stp d8,   d9, [x0, #(12 * 8)];
+      stp d10, d11, [x0, #(14 * 8)];
+      stp d12, d13, [x0, #(16 * 8)];
+      stp d14, d15, [x0, #(18 * 8)];
+
+      // Move SP in to a temporary to store.
+      mov x1, sp;
+      str x1,  [x0, #(20 * 8)];
+
+      // Return zero to signify this is the SetJump.
+      mov x0, #0;
+      ret;
+    )" ::
+                   : "memory");
+}
+
+[[noreturn]]
+FEX_DEFAULT_VISIBILITY FEX_NAKED void LongJump(JumpBuf& Buffer, uint64_t Value) {
+  __asm volatile(R"(
+      // x0 contains the jumpbuffer
+      ldp x19, x20, [x0, #( 0 * 8)];
+      ldp x21, x22, [x0, #( 2 * 8)];
+      ldp x23, x24, [x0, #( 4 * 8)];
+      ldp x25, x26, [x0, #( 6 * 8)];
+      ldp x27, x28, [x0, #( 8 * 8)];
+      ldp x29, x30, [x0, #(10 * 8)];
+
+      // FPRs
+      ldp d8,   d9, [x0, #(12 * 8)];
+      ldp d10, d11, [x0, #(14 * 8)];
+      ldp d12, d13, [x0, #(16 * 8)];
+      ldp d14, d15, [x0, #(18 * 8)];
+
+      // Load SP in to temporary then move
+      ldr x0,  [x0, #(20 * 8)];
+      mov sp, x0;
+
+      // Move value in to result register
+      mov x0, x1;
+      ret;
+    )" ::
+                   : "memory");
+}
+#else
+[[nodiscard]]
+FEX_DEFAULT_VISIBILITY FEX_NAKED uint64_t SetJump(JumpBuf& Buffer) {
+  __asm volatile(R"(
+    .intel_syntax noprefix;
+    // rdi contains the jumpbuffer
+    mov [rdi + (0 * 8)], rbx;
+    mov [rdi + (1 * 8)], rsp;
+    mov [rdi + (2 * 8)], rbp;
+    mov [rdi + (3 * 8)], r12;
+    mov [rdi + (4 * 8)], r13;
+    mov [rdi + (5 * 8)], r14;
+    mov [rdi + (6 * 8)], r15;
+
+    // Return address is on the stack, load it and store
+    mov rsi, [rsp];
+    mov [rdi + (7 * 8)], rsi;
+
+    // Return zero to signify this is the SetJump.
+    mov rax, 0;
+    ret;
+
+    .att_syntax prefix;
+    )" ::
+                   : "memory");
+}
+
+[[noreturn]]
+FEX_DEFAULT_VISIBILITY FEX_NAKED void LongJump(JumpBuf& Buffer, uint64_t Value) {
+  __asm volatile(R"(
+    .intel_syntax noprefix;
+    // rdi contains the jumpbuffer
+    mov rbx, [rdi + (0 * 8)];
+    mov rsp, [rdi + (1 * 8)];
+    mov rbp, [rdi + (2 * 8)];
+    mov r12, [rdi + (3 * 8)];
+    mov r13, [rdi + (4 * 8)];
+    mov r14, [rdi + (5 * 8)];
+    mov r15, [rdi + (6 * 8)];
+
+    // Move value in to result register
+    mov rax, rsi;
+
+    // Pop the dead return address off the stack
+    pop rsi;
+
+    // Load the original return address from the jumpbuffer
+    mov rsi, [rdi + (7 * 8)];
+
+    // Return using a jump
+    jmp rsi;
+
+    .att_syntax prefix;
+    )" ::
+                   : "memory");
+}
+
+#endif
+} // namespace FEXCore::UncheckedLongJump

FEXCore/Source/Utils/Profiler.cpp

@@ -9,6 +9,7 @@
 
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/Profiler.h>
+#include <FEXCore/Config/Config.h>
 #include <FEXCore/fextl/fmt.h>
 #include <FEXCore/fextl/string.h>
 
@@ -70,6 +71,10 @@ static std::array<const char*, 2> TraceFSDirectories {
 };
 
 void Init() {
+  FEX_CONFIG_OPT(EnableGpuvisProfiling, ENABLEGPUVISPROFILING);
+  if (!EnableGpuvisProfiling()) {
+    return;
+  }
   for (auto Path : TraceFSDirectories) {
 #ifdef _WIN32
     constexpr auto flags = O_WRONLY;

FEXCore/Source/Utils/SpinWaitLock.h

@@ -1,4 +1,6 @@
 // SPDX-License-Identifier: MIT
+#pragma once
+
 #include <atomic>
 #include <chrono>
 #include <mutex>
@@ -125,8 +127,8 @@ static inline uint64_t WFELoadAtomic(uint64_t* Futex) {
 
 template<typename T, typename TT = T>
 static inline void Wait(T* Futex, TT ExpectedValue) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
-  T Result = AtomicFutex->load();
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
+  T Result = AtomicFutex.load();
 
   // Early exit if possible.
   if (Result == ExpectedValue) {
@@ -149,9 +151,9 @@ template void Wait<uint64_t>(uint64_t*, uint64_t);
 
 template<typename T, typename TT>
 static inline bool Wait(T* Futex, TT ExpectedValue, const std::chrono::nanoseconds& Timeout) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
 
-  T Result = AtomicFutex->load();
+  T Result = AtomicFutex.load();
 
   // Early exit if possible.
   if (Result == ExpectedValue) {
@@ -184,11 +186,31 @@ template bool Wait<uint16_t>(uint16_t*, uint16_t, const std::chrono::nanoseconds
 template bool Wait<uint32_t>(uint32_t*, uint32_t, const std::chrono::nanoseconds&);
 template bool Wait<uint64_t>(uint64_t*, uint64_t, const std::chrono::nanoseconds&);
 
+template<typename T>
+static inline T OneShotWFEBitComparison(T* Futex, T Mask, T Comp) {
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
+  T Result = AtomicFutex.load();
+
+  // Early exit if possible.
+  if ((Result & Mask) == Comp) {
+    return Result;
+  }
+
+  Result = LoadExclusive(Futex);
+  if ((Result & Mask) == Comp) {
+    return Result;
+  }
+
+  // Waits for write and returns result.
+  Result = WFELoadAtomic(Futex);
+  return Result;
+}
+
 #else
 template<typename T, typename TT>
 static inline void Wait(T* Futex, TT ExpectedValue) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
-  T Result = AtomicFutex->load();
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
+  T Result = AtomicFutex.load();
 
   // Early exit if possible.
   if (Result == ExpectedValue) {
@@ -196,15 +218,15 @@ static inline void Wait(T* Futex, TT ExpectedValue) {
   }
 
   do {
-    Result = AtomicFutex->load();
+    Result = AtomicFutex.load();
   } while (Result != ExpectedValue);
 }
 
 template<typename T, typename TT>
 static inline bool Wait(T* Futex, TT ExpectedValue, const std::chrono::nanoseconds& Timeout) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
 
-  T Result = AtomicFutex->load();
+  T Result = AtomicFutex.load();
 
   // Early exit if possible.
   if (Result == ExpectedValue) {
@@ -214,7 +236,7 @@ static inline bool Wait(T* Futex, TT ExpectedValue, const std::chrono::nanosecon
   const auto Begin = std::chrono::high_resolution_clock::now();
 
   do {
-    Result = AtomicFutex->load();
+    Result = AtomicFutex.load();
 
     const auto CurrentCycleCounter = std::chrono::high_resolution_clock::now();
     if ((CurrentCycleCounter - Begin) >= Timeout) {
@@ -230,12 +252,12 @@ static inline bool Wait(T* Futex, TT ExpectedValue, const std::chrono::nanosecon
 
 template<typename T>
 static inline void lock(T* Futex) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
   T Expected {};
   T Desired {1};
 
   // Try to CAS immediately.
-  if (AtomicFutex->compare_exchange_strong(Expected, Desired)) {
+  if (AtomicFutex.compare_exchange_strong(Expected, Desired)) {
     return;
   }
 
@@ -243,17 +265,17 @@ static inline void lock(T* Futex) {
     // Wait until the futex is unlocked.
     Wait(Futex, 0);
     Expected = 0;
-  } while (!AtomicFutex->compare_exchange_strong(Expected, Desired));
+  } while (!AtomicFutex.compare_exchange_strong(Expected, Desired));
 }
 
 template<typename T>
 static inline bool try_lock(T* Futex) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
   T Expected {};
   T Desired {1};
 
   // Try to CAS immediately.
-  if (AtomicFutex->compare_exchange_strong(Expected, Desired)) {
+  if (AtomicFutex.compare_exchange_strong(Expected, Desired)) {
     return true;
   }
 
@@ -262,8 +284,8 @@ static inline bool try_lock(T* Futex) {
 
 template<typename T>
 static inline void unlock(T* Futex) {
-  std::atomic<T>* AtomicFutex = reinterpret_cast<std::atomic<T>*>(Futex);
-  AtomicFutex->store(0);
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
+  AtomicFutex.store(0);
 }
 
 #undef SPINLOOP_8BIT

FEXCore/Source/Utils/WritePriorityMutex.h

@@ -0,0 +1,384 @@
+// SPDX-License-Identifier: MIT
+#pragma once
+#include <atomic>
+#include <cstdint>
+
+#if !defined(_WIN32)
+#include <linux/futex.h> /* Definition of FUTEX_* constants */
+#include <sys/syscall.h> /* Definition of SYS_* constants */
+#include <unistd.h>
+#else
+#include <synchapi.h>
+#endif
+
+#include <FEXCore/Utils/LogManager.h>
+
+#include "Utils/SpinWaitLock.h"
+
+namespace FEXCore::Utils::WritePriorityMutex {
+
+#if !defined(_WIN32)
+// A custom mutex that prioritizes exclusive locks.
+// In highly contested scenarios, this can help minimize overall contention time.
+//
+// Features:
+//  - Up to 32767 pending exclusive locks ("writers")
+//  - Up to 32767 pending shared_locks ("readers")
+//  - Low-overhead waiting via WFE with a fallback to futex on timeout
+//  - Direct writer->reader hand-off and vice-versa to further reduce overhead
+//
+// Trade-offs:
+//  - No guaranteed order of wake-ups besides prioritizing writers
+//  - No support for recursive locking
+//  - We can't use FUTEX_LOCK_PI to enable priority inheritance
+class Mutex final {
+public:
+  Mutex() = default;
+
+  // Move-only type
+  Mutex(const Mutex&) = delete;
+  Mutex& operator=(const Mutex&) = delete;
+  Mutex(Mutex&& rhs) = delete;
+  Mutex& operator=(Mutex&&) = delete;
+
+  void lock() {
+    // Try a non-blocking lock first.
+    if (try_lock()) {
+      return;
+    }
+
+    // Try a quick WFE write-lock.
+    if (Attempt_WFE_WriteLock()) {
+      return;
+    }
+
+    // Still couldn't get it. Start waiting.
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+
+    uint32_t Expected {};
+    uint32_t Desired {};
+#if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
+    Expected = AtomicFutex.load(std::memory_order_relaxed);
+    do {
+      // Increment the number of write waiters.
+      Desired = Expected + WRITE_WAITER_INCREMENT;
+
+      LOGMAN_THROW_A_FMT((Desired & WRITE_WAITER_COUNT_MASK) != 0, "Overflow in write-waiters!");
+    } while (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire) == false);
+#else
+    Expected = AtomicFutex.fetch_add(WRITE_WAITER_INCREMENT);
+    Desired = Expected + WRITE_WAITER_INCREMENT;
+#endif
+
+    // Thread added to waiter list.
+    Expected = Desired;
+
+    while (true) {
+      bool Sleep = false;
+
+      do {
+        if ((Expected & WRITE_OWNED_BIT) == 0 && (Expected & READ_OWNER_COUNT_MASK) == 0) {
+          // If not write-owned, and no read-owners, try to acquire.
+          LOGMAN_THROW_A_FMT((Expected & WRITE_WAITER_COUNT_MASK) != 0, "Underflow in write-waiters!");
+
+          // Add write-owned bit.
+          Desired = Expected | WRITE_OWNED_BIT;
+
+          // Remove ourselves from the wait list.
+          Desired -= WRITE_WAITER_INCREMENT;
+
+          Sleep = false;
+        } else {
+          // Already write-owned or read-locked. Go to sleep.
+          Desired = Expected;
+          Sleep = true;
+          break;
+        }
+
+      } while (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire) == false);
+
+      if (!Sleep) {
+        // Acquired early.
+        LOGMAN_THROW_A_FMT((Desired & WRITE_OWNED_BIT) == WRITE_OWNED_BIT, "Somehow acquired a write-lock without it being set!");
+        return;
+      }
+      FutexWaitForWriteAvailable(Desired);
+
+      Expected = AtomicFutex.load(std::memory_order_relaxed);
+    }
+  }
+
+  void lock_shared() {
+    // Try an uncontended lock first.
+    if (try_lock_shared()) {
+      return;
+    }
+
+    // Try a quick WFE read-lock.
+    if (Attempt_WFE_ReadLock()) {
+      return;
+    }
+
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+
+    uint32_t Expected = AtomicFutex.load(std::memory_order_relaxed);
+    uint32_t Desired {};
+
+    while (true) {
+
+      bool Sleep = false;
+      do {
+        if ((Expected & WRITE_OWNED_BIT) == 0 && (Expected & WRITE_WAITER_COUNT_MASK) == 0) {
+          // If no write-owner and no write-waiting, try and acquire.
+
+          Desired = Expected + READ_OWNER_INCREMENT;
+          LOGMAN_THROW_A_FMT((Desired & READ_OWNER_COUNT_MASK) != 0, "Overflow in read-owners!");
+          Sleep = false;
+        } else {
+          // Waiting for lock to become available. Add to waiters.
+          Desired = Expected | READ_WAITER_BIT;
+          Sleep = true;
+        }
+      } while (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire) == false);
+
+      if (!Sleep) {
+        // Acquired early.
+        LOGMAN_THROW_A_FMT((Desired & WRITE_OWNED_BIT) != WRITE_OWNED_BIT, "Somehow read-locked and got a write lock!");
+        return;
+      }
+
+      FutexWaitForReadAvailable(Desired);
+
+      Expected = AtomicFutex.load(std::memory_order_relaxed);
+    }
+  }
+
+  void unlock() {
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+
+    uint32_t Expected = AtomicFutex.load(std::memory_order_relaxed);
+    uint32_t Desired {};
+    do {
+      LOGMAN_THROW_A_FMT((Expected & WRITE_OWNED_BIT) == WRITE_OWNED_BIT, "Trying to write-unlock something not write-locked!");
+      // Remove the exclusive lock bit.
+      Desired = Expected & ~WRITE_OWNED_BIT;
+
+      // If no more writers, then make sure to clear the read-waiters bit as well.
+      if ((Desired & WRITE_WAITER_COUNT_MASK) == 0) {
+        Desired &= ~READ_WAITER_BIT;
+      }
+    } while (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire) == false);
+
+    // If success, then `Expected` has old value. Containing `READ_WAITER_BIT` which was just masked off, and also `WRITE_WAITER_COUNT_MASK`.
+    if ((Expected & WRITE_WAITER_COUNT_MASK)) {
+      // Handle write-write handoff.
+      FutexWakeWriter();
+    } else if ((Expected & READ_WAITER_BIT)) {
+      // Handle write-reader handoff.
+      FutexWakeReaders();
+    }
+  }
+
+  void unlock_shared() {
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+
+    uint32_t Desired {};
+#if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
+    uint32_t Expected = AtomicFutex.load(std::memory_order_relaxed);
+    do {
+      LOGMAN_THROW_A_FMT((Expected & WRITE_OWNED_BIT) != WRITE_OWNED_BIT, "Trying to read-unlock something write-locked!");
+      LOGMAN_THROW_A_FMT((Expected & READ_OWNER_COUNT_MASK) != 0, "Trying to read-unlock something not read-locked!");
+
+      // Decrement the shared counter.
+      Desired = Expected - READ_OWNER_INCREMENT;
+    } while (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire) == false);
+#else
+    Desired = AtomicFutex.fetch_sub(READ_OWNER_INCREMENT) - READ_OWNER_INCREMENT;
+#endif
+
+    // Handle read->write handoff if there are any waiting writers, and no readers left.
+    if ((Desired & WRITE_WAITER_COUNT_MASK) && (Desired & READ_OWNER_COUNT_MASK) == 0) {
+      FutexWakeWriter();
+    }
+  }
+
+  bool try_lock() {
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+
+    uint32_t Expected = 0;
+
+    // Try and grab the owned bit.
+    uint32_t Desired = WRITE_OWNED_BIT;
+
+    // try to CAS immediately.
+    return AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire);
+  }
+
+  // Can race with other threads trying to lock shared!
+  bool try_lock_shared() {
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+    uint32_t Expected = AtomicFutex.load(std::memory_order_relaxed);
+
+    // Exclusively owned or has a list of waiting owners. Can't pass.
+    if ((Expected & WRITE_OWNED_BIT) || (Expected & WRITE_WAITER_COUNT_MASK)) {
+      return false;
+    }
+
+    // Try to add reader.
+    uint32_t Desired = Expected + READ_OWNER_INCREMENT;
+    LOGMAN_THROW_A_FMT((Desired & READ_OWNER_COUNT_MASK) != 0, "Overflow in read-owners!");
+
+    // Uncontended mutex check
+    return AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire);
+  }
+
+private:
+
+  void FutexWaitForWriteAvailable(uint32_t Expected) {
+    ::syscall(SYS_futex, &Futex, FUTEX_PRIVATE_FLAG | FUTEX_WAIT_BITSET, Expected, nullptr, nullptr, FUTEX_BITSET_WAIT_WRITERS);
+  }
+
+  // Read-lock waiting for writers to drain out.
+  void FutexWaitForReadAvailable(uint32_t Expected) {
+    ::syscall(SYS_futex, &Futex, FUTEX_PRIVATE_FLAG | FUTEX_WAIT_BITSET, Expected, nullptr, nullptr, FUTEX_BITSET_WAIT_READERS);
+  }
+
+  // Read-Lock or Write-lock unlocked, wake one writer.
+  // - Read->Write handoff.
+  // - Write->Write handoff.
+  void FutexWakeWriter() {
+    ::syscall(SYS_futex, &Futex, FUTEX_PRIVATE_FLAG | FUTEX_WAKE_BITSET, 1, nullptr, nullptr, FUTEX_BITSET_WAIT_WRITERS);
+  }
+
+  // Write-lock unlocked, wake read-locks waiting.
+  void FutexWakeReaders() {
+    // Wake all readers.
+    ::syscall(SYS_futex, &Futex, FUTEX_PRIVATE_FLAG | FUTEX_WAKE_BITSET, INT_MAX, nullptr, nullptr, FUTEX_BITSET_WAIT_READERS);
+  }
+
+  // Reuse the SpinWaitLock WFE implementations for read/write lock acquiring with WFE.
+  // Can't reuse the spin-lock directly as some bit-representations are different.
+  // WFE-write-lock is less likely to occur the more read-lock threads are participating. Can still occur so good to try.
+  // WFE-read-lock is actually quite likely to succeed.
+  // Return: true if the lock was acquired.
+  bool Attempt_WFE_WriteLock() {
+#ifdef _M_ARM_64
+    const auto Begin = FEXCore::Utils::SpinWaitLock::GetCycleCounter();
+    auto Now = Begin;
+    const auto Duration = FEXCore::Utils::SpinWaitLock::CycleCounterFrequency / CYCLECOUNT_DIVISOR;
+
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+    uint32_t Expected = AtomicFutex.load(std::memory_order_relaxed);
+
+    while ((Now - Begin) < Duration) {
+      if (Expected == 0) {
+        // Try and grab the owned bit.
+        uint32_t Desired = WRITE_OWNED_BIT;
+
+        if (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire)) {
+          return true;
+        }
+      }
+
+      // One-shot attempt to wait for mask to be zero.
+      Expected = FEXCore::Utils::SpinWaitLock::OneShotWFEBitComparison(&Futex, ~0U, 0U);
+      Now = FEXCore::Utils::SpinWaitLock::GetCycleCounter();
+    }
+#endif
+
+    return false;
+  }
+
+  // Return: true if the lock was acquired.
+  bool Attempt_WFE_ReadLock() {
+#ifdef _M_ARM_64
+    // Spin on a WFE for a short-amount of time, waiting for write-owned and writer-count to be zero.
+    //  - Attempt to acquire read-lock at that point.
+    //  - Don't add read-waiters bit on failure, return false.
+    const auto Begin = FEXCore::Utils::SpinWaitLock::GetCycleCounter();
+    auto Now = Begin;
+    const auto Duration = FEXCore::Utils::SpinWaitLock::CycleCounterFrequency / CYCLECOUNT_DIVISOR;
+
+    auto AtomicFutex = std::atomic_ref<uint32_t>(Futex);
+    uint32_t Expected = AtomicFutex.load(std::memory_order_relaxed);
+    uint32_t Desired {};
+
+    while ((Now - Begin) < Duration) {
+      if ((Expected & WRITE_OWNED_BIT) == 0 && (Expected & WRITE_WAITER_COUNT_MASK) == 0) {
+        // If no write-owner and no write-waiting, try and acquire.
+
+        Desired = Expected + READ_OWNER_INCREMENT;
+        LOGMAN_THROW_A_FMT((Desired & READ_OWNER_COUNT_MASK) != 0, "Overflow in read-owners!");
+        if (AtomicFutex.compare_exchange_strong(Expected, Desired, std::memory_order_acq_rel, std::memory_order_acquire)) {
+          return true;
+        }
+      }
+
+      // One-shot attempt to wait for mask to be zero.
+      Expected = FEXCore::Utils::SpinWaitLock::OneShotWFEBitComparison(&Futex, WRITE_OWNED_BIT | WRITE_WAITER_COUNT_MASK, 0U);
+      Now = FEXCore::Utils::SpinWaitLock::GetCycleCounter();
+    }
+#endif
+
+    return false;
+  }
+
+  constexpr static uint32_t WRITE_OWNED_BIT = 1U << 31;
+  constexpr static uint32_t READ_WAITER_BIT = 1U << 15;
+  constexpr static uint32_t WRITE_WAITER_OFFSET = 16;
+  constexpr static uint32_t WRITE_WAITER_INCREMENT = 1U << WRITE_WAITER_OFFSET;
+  constexpr static uint32_t READ_OWNER_INCREMENT = 1;
+
+  // Count masks
+  constexpr static uint32_t WRITE_WAITER_COUNT_MASK = 0x7FFFU << WRITE_WAITER_OFFSET;
+  constexpr static uint32_t READ_OWNER_COUNT_MASK = 0x7FFFU;
+
+  // Independent futex bit-set masks.
+  // Wait for readers to drain.
+  constexpr static uint32_t FUTEX_BITSET_WAIT_READERS = 1U << 0;
+  // Wait for writers to drain.
+  constexpr static uint32_t FUTEX_BITSET_WAIT_WRITERS = 1U << 1;
+
+  // Only spin on WFE for 0.01ms (10k ns).
+  constexpr static uint64_t CYCLECOUNT_DIVISOR = 1'000'000'000ULL / 10'000U;
+
+  // Layout:
+  //    Bits[31]: Write-lock bit.
+  // Bits[30:16]: Write-waiter count.
+  //    Bits[15]: Read-waiter bit.
+  //  Bits[14:0]: Read-owner count.
+  uint32_t Futex {};
+};
+#else
+// SRWLocks are already write-priority locks in WINE and Windows. Use them to avoid lock stampeding.
+class Mutex final {
+public:
+  void lock() {
+    AcquireSRWLockExclusive(&Futex);
+  }
+
+  void lock_shared() {
+    AcquireSRWLockShared(&Futex);
+  }
+
+  void unlock() {
+    ReleaseSRWLockExclusive(&Futex);
+  }
+
+  void unlock_shared() {
+    ReleaseSRWLockShared(&Futex);
+  }
+
+  bool try_lock() {
+    return TryAcquireSRWLockExclusive(&Futex);
+  }
+
+  bool try_lock_shared() {
+    return TryAcquireSRWLockShared(&Futex);
+  }
+
+private:
+  SRWLOCK Futex = SRWLOCK_INIT;
+};
+#endif
+} // namespace FEXCore::Utils::WritePriorityMutex

FEXCore/include/FEXCore/Core/CoreState.h

@@ -104,6 +104,10 @@ struct CPUState {
   uint64_t avx_high[16][2];
 
   uint64_t gregs[16] {};
+  uint64_t L1Pointer {};
+  uint64_t L1Mask {};
+  uint64_t callret_sp {};
+  uint64_t _pad1 {};
   XMMRegs xmm {};
 
   // Raw segment register indexes
@@ -116,8 +120,6 @@ struct CPUState {
   uint64_t gs_cached {};
   uint64_t fs_cached {};
   uint8_t flags[48] {};
-  uint64_t callret_sp {};
-  uint64_t _pad1 {};
   uint64_t mm[8][2] {};
 
   // 32bit x86 state
@@ -247,6 +249,8 @@ static_assert(offsetof(CPUState, xmm) % 32 == 0, "xmm needs to be 256-bit aligne
 static_assert(offsetof(CPUState, mm) % 16 == 0, "mm needs to be 128-bit aligned!");
 static_assert(offsetof(CPUState, gregs[15]) <= 504, "gregs maximum offset must be <= 504 for ldp/stp to work");
 static_assert(offsetof(CPUState, DeferredSignalRefCount) % 8 == 0, "Needs to be 8-byte aligned");
+static_assert(offsetof(CPUState, L1Pointer) <= 504, "This needs to be <= 504 for ldp");
+static_assert(offsetof(CPUState, L1Mask) == (offsetof(CPUState, L1Pointer) + 8), "These two variables are paired");
 
 struct InternalThreadState;
 
@@ -349,13 +353,11 @@ struct JITPointers {
     uint64_t ExitFunctionLinker {};
     uint64_t ThreadStopHandlerSpillSRA {};
     uint64_t ThreadPauseHandlerSpillSRA {};
-    uint64_t UnimplementedInstructionHandler {};
     uint64_t GuestSignal_SIGILL {};
     uint64_t GuestSignal_SIGTRAP {};
     uint64_t GuestSignal_SIGSEGV {};
     uint64_t SignalReturnHandler {};
     uint64_t SignalReturnHandlerRT {};
-    uint64_t L1Pointer {};
     uint64_t L2Pointer {};
     /**  @} */
 
@@ -368,8 +370,6 @@ struct JITPointers {
       // Process specific
       uint64_t LUDIV {};
       uint64_t LDIV {};
-      uint64_t LUREM {};
-      uint64_t LREM {};
 
       // Thread Specific
 
@@ -378,8 +378,6 @@ struct JITPointers {
        * @{ */
       uint64_t LUDIVHandler {};
       uint64_t LDIVHandler {};
-      uint64_t LUREMHandler {};
-      uint64_t LREMHandler {};
       /**  @} */
     } AArch64;
 

FEXCore/include/FEXCore/Debug/InternalThreadState.h

@@ -80,7 +80,14 @@ private:
 static_assert(!std::is_move_constructible_v<NonMovableUniquePtr<int>>);
 static_assert(!std::is_move_assignable_v<NonMovableUniquePtr<int>>);
 
-struct InternalThreadState : public FEXCore::Allocator::FEXAllocOperators {
+// Store used for unaligned LDAXR*/STLXR* emulation.
+struct UnalignedExclusiveStore {
+  uint64_t Addr;
+  uint64_t Store;
+  uint8_t Size;
+};
+
+struct alignas(FEXCore::Utils::FEX_PAGE_SIZE) InternalThreadState : public FEXCore::Allocator::FEXAllocOperators {
   FEXCore::Core::CpuStateFrame* const CurrentFrame = &BaseFrameState;
 
   FEXCore::Context::Context* const CTX;
@@ -101,6 +108,8 @@ struct InternalThreadState : public FEXCore::Allocator::FEXAllocOperators {
   // This pointer is owned by the frontend.
   FEXCore::SHMStats::ThreadStats* ThreadStats {};
 
+  UnalignedExclusiveStore ExclusiveStore;
+
   ///< Data pointer for exclusive use by the frontend
   void* FrontendPtr;
 
@@ -116,8 +125,9 @@ struct InternalThreadState : public FEXCore::Allocator::FEXAllocOperators {
   alignas(FEXCore::Utils::FEX_PAGE_SIZE) uint8_t InterruptFaultPage[FEXCore::Utils::FEX_PAGE_SIZE];
 };
 static_assert(std::is_standard_layout_v<FEXCore::Core::InternalThreadState>);
-static_assert(
-  (offsetof(FEXCore::Core::InternalThreadState, InterruptFaultPage) - offsetof(FEXCore::Core::InternalThreadState, BaseFrameState)) < 4096,
-  "Fault page is outside of immediate range from CPU state");
+static_assert((offsetof(FEXCore::Core::InternalThreadState, InterruptFaultPage) - offsetof(FEXCore::Core::InternalThreadState, BaseFrameState)) <
+                FEXCore::Utils::FEX_PAGE_SIZE,
+              "Fault page is outside of immediate range from CPU state");
+static_assert(sizeof(FEXCore::Core::InternalThreadState) == (FEXCore::Utils::FEX_PAGE_SIZE * 2));
 
 } // namespace FEXCore::Core

FEXCore/include/FEXCore/HLE/SyscallHandler.h

@@ -54,10 +54,6 @@ public:
   virtual ~SyscallHandler() = default;
 
   virtual uint64_t HandleSyscall(FEXCore::Core::CpuStateFrame* Frame, FEXCore::HLE::SyscallArguments* Args) = 0;
-  virtual SyscallABI GetSyscallABI(uint64_t Syscall) = 0;
-  virtual FEXCore::IR::SyscallFlags GetSyscallFlags(uint64_t Syscall) const {
-    return FEXCore::IR::SyscallFlags::DEFAULT;
-  }
 
   SyscallOSABI GetOSABI() const {
     return OSABI;

FEXCore/include/FEXCore/IR/IR.h

@@ -8,26 +8,6 @@
 
 namespace FEXCore::IR {
 
-enum class SyscallFlags : uint8_t {
-  DEFAULT = 0,
-  // Syscalldoesn't care about CPUState being serialized up to the syscall instruction.
-  // Means dead code elimination can optimize through a syscall operation.
-  OPTIMIZETHROUGH = 1 << 0,
-  // Syscall only reads the passed in arguments. Doesn't read CPUState.
-  NOSYNCSTATEONENTRY = 1 << 1,
-  // Syscall doesn't return. Code generation after syscall return can be removed.
-  NORETURN = 1 << 2,
-  // Syscall doesn't have any side-effects, so if the result isn't used then it can be removed.
-  NOSIDEEFFECTS = 1 << 3,
-  // Syscall doesn't return a result.
-  // Means the resulting register shouldn't be written (Usually RAX).
-  // Usually used with !NOSYNCSTATEONENTRY, so the syscall can modify CPU state entirely.
-  // Then on return FEXCore picks up the new state.
-  NORETURNEDRESULT = 1 << 4,
-};
-
-FEX_DEF_NUM_OPS(SyscallFlags)
-
 // This enum of named vector constants are linked to an array in CPUBackend.cpp.
 // This is used with the IROp `LoadNamedVectorConstant` to load a vector constant
 // that would otherwise be costly to materialize.

FEXCore/include/FEXCore/Utils/AllocatorHooks.h

@@ -82,12 +82,15 @@ inline bool VirtualProtect(void* Ptr, size_t Size, ProtectOptions options) {
   return ::VirtualProtect(Ptr, Size, prot, nullptr) == 0;
 }
 
+inline void VirtualName(const char*, void*, size_t) {}
+
 #else
 using MMAP_Hook = void* (*)(void*, size_t, int, int, int, off_t);
 using MUNMAP_Hook = int (*)(void*, size_t);
 
 FEX_DEFAULT_VISIBILITY extern MMAP_Hook mmap;
 FEX_DEFAULT_VISIBILITY extern MUNMAP_Hook munmap;
+FEX_DEFAULT_VISIBILITY extern void VirtualName(const char* Name, void* Ptr, size_t Size);
 
 // All commit parameters are ignored here, they are unnecessary as Linux supports overcommit
 

FEXCore/include/FEXCore/Utils/ArchHelpers/Arm64.h

@@ -12,8 +12,6 @@ struct InternalThreadState;
 
 namespace FEXCore::ArchHelpers::Arm64 {
 enum class UnalignedHandlerType {
-  ///< Don't backpatch code, instead handle inside SIGBUS handler.
-  Paranoid,
   ///< Backpatch unaligned access to half-barrier based atomic.
   HalfBarrier,
   ///< Backpatch unaligned access to non-atomic.
@@ -26,7 +24,7 @@ enum class UnalignedHandlerType {
  * This is an OS agnostic handler where the frontend must provide FEXCore with the information necessary to know if this is safe.
  * This does not check if the PC is within a JIT code buffer, the frontend must provide that safety with `CPUBackend::IsAddressInCodeBuffer`.
  *
- * @param ParanoidTSO If the unaligned fault needs to handled directly or can be backpatched.
+ * @param HandleType Type of TSO handling to use.
  * @param ProgramCounter The location in memory for the instruction that did the access
  * @param GPRs The array of GPRs from the signal context. This will be modified and the host context needs to be updated on signal return.
  *
@@ -34,6 +32,6 @@ enum class UnalignedHandlerType {
  * by. FEXCore will return a positive or negative offset depending on internal handling.
  */
 [[nodiscard]]
-FEX_DEFAULT_VISIBILITY std::optional<int32_t>
-HandleUnalignedAccess(FEXCore::Core::InternalThreadState* Thread, UnalignedHandlerType HandleType, uintptr_t ProgramCounter, uint64_t* GPRs);
+FEX_DEFAULT_VISIBILITY std::optional<int32_t> HandleUnalignedAccess(
+  FEXCore::Core::InternalThreadState* Thread, UnalignedHandlerType HandleType, uintptr_t ProgramCounter, uint64_t* GPRs, bool IsJIT = true);
 } // namespace FEXCore::ArchHelpers::Arm64

FEXCore/include/FEXCore/Utils/LongJump.h

@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MIT
+#pragma once
+#include <FEXCore/Utils/CompilerDefs.h>
+
+#include <cstdint>
+
+// Reimplementation of longjmp without glibc fortification checks.
+// This is useful when false positives need to be avoided or when using
+// a libc implementation that does not implement std::longjmp.
+namespace FEXCore::UncheckedLongJump {
+// JumpBuf definition needs to be public because the frontend needs to understand it.
+#if defined(_M_ARM_64)
+struct JumpBuf {
+  // All the registers that are required by AAPCS64 to save.
+  // GPRs
+  // X19, X20, X21, X22,
+  // X23, X24, X25, X26,
+  // X27, X28, X29, X30,
+  //
+  // Lower 64-bits:
+  //  V8,  V9, V10, V11,
+  // V12, V13, V14, V15,
+  //
+  // SP,
+  uint64_t Registers[21];
+};
+#else
+struct JumpBuf {
+  // Registers to preserve
+  // RBX, RSP, RBP, R12, R13, R14, R15,
+  // <return address>
+  uint64_t Registers[8];
+};
+#endif
+
+[[nodiscard]] FEX_DEFAULT_VISIBILITY uint64_t SetJump(JumpBuf& Buffer);
+[[noreturn]] FEX_DEFAULT_VISIBILITY void LongJump(JumpBuf& Buffer, uint64_t Value);
+} // namespace FEXCore::UncheckedLongJump

FEXCore/include/FEXCore/Utils/SHMStats.h

@@ -39,10 +39,12 @@ enum class AppType : uint8_t {
   WIN_WOW64,
 };
 
+// Only append new members to the end of {ThreadStatsHeader, ThreadStats} to allow old tools time to support new information.
+// FEX isn't guaranteeing /not/ breaking compatibility with versions, but trying to not cause too much churn.
 struct ThreadStatsHeader {
   uint8_t Version;
   AppType app_type;
-  uint8_t _pad[2];
+  uint16_t ThreadStatsSize;
   char fex_version[48];
   std::atomic<uint32_t> Head;
   std::atomic<uint32_t> Size;
@@ -61,8 +63,17 @@ struct ThreadStats {
   uint64_t AccumulatedSIGBUSCount;
   uint64_t AccumulatedSMCCount;
   uint64_t AccumulatedFloatFallbackCount;
+
+  uint64_t AccumulatedCacheMissCount;
+  uint64_t AccumulatedCacheReadLockTime;
+  uint64_t AccumulatedCacheWriteLockTime;
+
+  uint64_t AccumulatedJITCount;
 };
 
+// Ensure 16-byte alignment to take advantage of ARM single-copy atomicity.
+static_assert(sizeof(ThreadStats) % 16 == 0, "Needs to be 16-byte aligned!");
+
 template<typename T, size_t FlatOffset = 0>
 class AccumulationBlock final {
 public:

FEXCore/include/FEXCore/Utils/ThreadPoolAllocator.h

@@ -380,6 +380,8 @@ private:
 class PooledAllocatorVirtual final : public IntrusivePooledAllocator {
 public:
   PooledAllocatorVirtual() = default;
+  PooledAllocatorVirtual(const char* Name)
+    : Name {Name} {}
 
   virtual ~PooledAllocatorVirtual() {
     FreeAllBuffers();
@@ -387,12 +389,18 @@ public:
 
 private:
   void* Alloc(size_t Size) override {
-    return FEXCore::Allocator::VirtualAlloc(Size);
+    auto Result = FEXCore::Allocator::VirtualAlloc(Size);
+    if (Name) {
+      FEXCore::Allocator::VirtualName(Name, Result, Size);
+    }
+    return Result;
   }
 
   void Free(void* Ptr, size_t Size) override {
     FEXCore::Allocator::VirtualFree(Ptr, Size);
   }
+
+  const char* Name {};
 };
 
 /**

FEXCore/include/FEXCore/fextl/memory_resource.h

@@ -2,7 +2,9 @@
 #pragma once
 #include <FEXCore/Utils/MathUtils.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/TypeDefines.h>
 #include <FEXCore/fextl/allocator.h>
+#include <FEXCore/fextl/list.h>
 
 #include <memory_resource>
 #include <fmt/format.h>
@@ -26,6 +28,90 @@ namespace pmr {
 
   FEX_DEFAULT_VISIBILITY std::pmr::memory_resource* get_default_resource();
 
+  /**
+   * @brief A `std::pmr::monotonic_buffer_resource` compatible class.
+   *
+   * Allocates internal buffers on page boundaries and names them for buffer tracking.
+   */
+  class named_monotonic_page_buffer_resource final : public std::pmr::memory_resource {
+  public:
+    explicit named_monotonic_page_buffer_resource(const char* Name)
+      : Name {Name} {}
+
+    void release() noexcept {
+      for (auto& Iter : Buffers) {
+        FEXCore::Allocator::VirtualFree(Iter.Buffer, Iter.BufferSize);
+      }
+      Buffers.clear();
+
+      CurrentBufferRemaining = 0;
+      CurrentAllocationSize = FEXCore::Utils::FEX_PAGE_SIZE;
+    }
+
+  protected:
+    void* do_allocate(std::size_t bytes, std::size_t alignment) override {
+      LOGMAN_THROW_A_FMT(bytes != 0, "Nope");
+      LOGMAN_THROW_A_FMT(alignment <= FEXCore::Utils::FEX_PAGE_SIZE, "Nope");
+
+      // Wow, an actual use case of std::align in the wild.
+      void* NewPointer = std::align(alignment, bytes, CurrentBuffer, CurrentBufferRemaining);
+      if (!NewPointer) [[unlikely]] {
+        AllocateNewBuffer(bytes, alignment);
+        NewPointer = CurrentBuffer;
+      }
+
+      CurrentBuffer = static_cast<char*>(CurrentBuffer) + bytes;
+      CurrentBufferRemaining -= bytes;
+
+      return NewPointer;
+    }
+
+    void do_deallocate(void*, std::size_t, std::size_t) override {
+      // Explicit no-op.
+    }
+
+    bool do_is_equal(const std::pmr::memory_resource& other) const noexcept override {
+      return this == &other;
+    }
+
+  private:
+    const char* Name;
+
+    // Allocate a new buffer that can at least fit the passed in bytes with alignment.
+    void AllocateNewBuffer(std::size_t bytes, std::size_t) {
+      bytes = FEXCore::AlignUp(bytes, CurrentAllocationSize);
+      void* Ptr = FEXCore::Allocator::VirtualAlloc(bytes);
+      if (Name) {
+        FEXCore::Allocator::VirtualName(Name, Ptr, bytes);
+      }
+
+      Buffers.emplace_back(BufferData {
+        .Buffer = Ptr,
+        .BufferSize = bytes,
+      });
+
+      CurrentBuffer = Ptr;
+      CurrentBufferRemaining = bytes;
+
+      // Multiply the allocation size by 1.5 for the next allocation
+      // Avoid double math because of ugly conversions.
+      CurrentAllocationSize = FEXCore::AlignUp(CurrentAllocationSize + (CurrentAllocationSize >> 1), FEXCore::Utils::FEX_PAGE_SIZE);
+    }
+
+    // Current buffer management.
+    void* CurrentBuffer {};
+    size_t CurrentBufferRemaining {};
+
+    struct BufferData final {
+      void* Buffer;
+      size_t BufferSize;
+    };
+
+    fextl::list<BufferData> Buffers {};
+
+    size_t CurrentAllocationSize = FEXCore::Utils::FEX_PAGE_SIZE;
+  };
+
   /**
    * @brief This is similar to the std::pmr::monotonic_buffer_resource.
    *

FEXCore/unittests/APITests/Allocator.cpp

@@ -0,0 +1,66 @@
+// SPDX-License-Identifier: MIT
+#include <catch2/catch_test_macros.hpp>
+#include <catch2/generators/catch_generators_range.hpp>
+
+#include "Utils/Allocator/HostAllocator.h"
+#include <FEXCore/Utils/Allocator.h>
+#include <sys/mman.h>
+
+template<typename T>
+bool HasSyscallError(T Result) {
+  constexpr uint64_t MAX_ERRNO = 0xFFFF'FFFF'FFFF'0001ULL;
+  return reinterpret_cast<uint64_t>(Result) >= MAX_ERRNO;
+}
+
+TEST_CASE("Allocator - Fixed replacement") {
+  const auto RegionSize = 128 * 1024 * 1024;
+  fextl::vector<FEXCore::Allocator::MemoryRegion> MemoryRegions {};
+  for (size_t i = 0; i < 2; ++i) {
+    auto Ptr = mmap(nullptr, RegionSize, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+    MemoryRegions.emplace_back(FEXCore::Allocator::MemoryRegion {
+      .Ptr = Ptr,
+      .Size = RegionSize,
+    });
+  }
+
+  auto Allocator = Alloc::OSAllocator::Create64BitAllocatorWithRegions(MemoryRegions);
+  auto Base = Allocator->Mmap(nullptr, 4096, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+  REQUIRE(!HasSyscallError(Base));
+
+  // Allocate perfectly overlapping pages. Allocate as many pages as the region.
+  // FEX had a bug where the allocator could run out of memory with MAP_FIXED.
+  for (size_t i = 0; i < (RegionSize / 4096); ++i) {
+    auto NewBase = Allocator->Mmap(Base, 4096, PROT_NONE, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+    REQUIRE(Base == NewBase);
+  }
+
+  Alloc::OSAllocator::ReleaseAllocatorWorkaround(std::move(Allocator));
+}
+
+TEST_CASE("Allocator - Non-Fit") {
+  const auto RegionSize = 128 * 1024 * 1024;
+  fextl::vector<FEXCore::Allocator::MemoryRegion> MemoryRegions {};
+  for (size_t i = 0; i < 2; ++i) {
+    auto Ptr = mmap(nullptr, RegionSize, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+    MemoryRegions.emplace_back(FEXCore::Allocator::MemoryRegion {
+      .Ptr = Ptr,
+      .Size = RegionSize,
+    });
+  }
+
+  auto Allocator = Alloc::OSAllocator::Create64BitAllocatorWithRegions(MemoryRegions);
+  auto Base = Allocator->Mmap(nullptr, RegionSize / 4, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+  REQUIRE(!HasSyscallError(Base));
+
+  // Try to allocate within the whole VMA size minus a small amount.
+  // FEX had a bug where if the allocation fit within a VMA region, it would try and allocate past the end without checking.
+  // Only occurred when `MAP_FIXED` was used.
+  auto NewBase = Allocator->Mmap(Base, RegionSize - (4096 * 64), PROT_NONE, MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+  // Must either fit in the VMA region, or fail.
+  // - If it matches previous allocation, then it fit in the VMA region.
+  //   - This can happen if FEX's allocator gains support for VMA merging.
+  // - If it errors, then it doesn't fit in the VMA region.
+  REQUIRE((NewBase == Base || HasSyscallError(NewBase)));
+
+  Alloc::OSAllocator::ReleaseAllocatorWorkaround(std::move(Allocator));
+}

FEXCore/unittests/APITests/FlexBitSet.cpp

@@ -3,6 +3,7 @@
 #include <catch2/generators/catch_generators_range.hpp>
 
 #include "Utils/Allocator/FlexBitSet.h"
+#include <sys/mman.h>
 
 TEST_CASE("FlexBitSet - Sizing") {
   // Ensure that FlexBitSet sizing is correct.
@@ -40,3 +41,25 @@ TEST_CASE("FlexBitSet - Sizing") {
   CHECK(FEXCore::FlexBitSet<uint32_t>::SizeInBits(sizeof(uint32_t) * 8) == sizeof(uint32_t) * 8);
   CHECK(FEXCore::FlexBitSet<uint64_t>::SizeInBits(sizeof(uint64_t) * 8) == sizeof(uint64_t) * 8);
 }
+
+TEST_CASE("FlexBitSet - Limit") {
+  // Ensure that the FlexBitSet doesn't read past the limits, and returns correct indexes.
+  const auto Size = 4096 * 3;
+  auto Ptr = mmap(nullptr, Size, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+  auto PtrMiddle = reinterpret_cast<void*>(reinterpret_cast<uintptr_t>(Ptr) + 4096);
+  REQUIRE(mprotect(PtrMiddle, 4096, PROT_READ | PROT_WRITE) != -1);
+
+  using ElementType = uint8_t;
+  const size_t NumElements = 4096 * 8;
+  auto FlexBit = reinterpret_cast<FEXCore::FlexBitSet<ElementType>*>(PtrMiddle);
+
+  for (size_t i = 0; i < NumElements; ++i) {
+    auto Result = FlexBit->ForwardScanForRange<true>(i, 1, NumElements);
+    CHECK(Result.FoundElement == i);
+  }
+
+  for (size_t i = 0; i < NumElements; ++i) {
+    auto Result = FlexBit->BackwardScanForRange<true>(i, 1, 0);
+    CHECK(Result.FoundElement == i);
+  }
+}

FEXCore/unittests/Emitter/ALU_Tests.cpp

@@ -9,17 +9,17 @@ using namespace ARMEmitter;
 TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    adr(Reg::r30, &Label);
+    (void)adr(Reg::r30, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x10fffffe);
   }
 
   {
     ForwardLabel Label;
-    adr(Reg::r30, &Label);
-    Bind(&Label);
+    (void)adr(Reg::r30, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x1000003e);
@@ -27,17 +27,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    adr(Reg::r30, &Label);
+    (void)adr(Reg::r30, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x10fffffe);
   }
 
   {
     BiDirectionalLabel Label;
-    adr(Reg::r30, &Label);
-    Bind(&Label);
+    (void)adr(Reg::r30, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x1000003e);
@@ -45,42 +45,42 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    adrp(Reg::r30, &Label);
+    (void)adrp(Reg::r30, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x9000001e);
   }
 
   {
     ForwardLabel Label;
-    adrp(Reg::r30, &Label);
+    (void)adrp(Reg::r30, &Label);
     // Move label a page away
     for (size_t i = 0; i < 1023; ++i) {
       nop();
     }
-    Bind(&Label);
+    (void)Bind(&Label);
 
     CHECK(DisassembleEncoding(0) == 0xb000001e);
   }
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    adrp(Reg::r30, &Label);
+    (void)adrp(Reg::r30, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x9000001e);
   }
 
   {
     BiDirectionalLabel Label;
-    adrp(Reg::r30, &Label);
+    (void)adrp(Reg::r30, &Label);
     // Move label a page away
     for (size_t i = 0; i < 1023; ++i) {
       nop();
     }
-    Bind(&Label);
+    (void)Bind(&Label);
 
     CHECK(DisassembleEncoding(0) == 0xb000001e);
   }
@@ -88,17 +88,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   {
     // Will generate adr.
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
-    LongAddressGen(Reg::r30, &Label);
+    (void)LongAddressGen(Reg::r30, &Label);
     CHECK(DisassembleEncoding(1) == 0x10fffffe);
   }
   {
     // Will generate nop + adr.
     ForwardLabel Label;
-    LongAddressGen(Reg::r30, &Label);
-    Bind(&Label);
+    (void)LongAddressGen(Reg::r30, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xd503201f);
@@ -107,9 +107,9 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   {
     // Will generate adr.
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    LongAddressGen(Reg::r30, &Label);
+    (void)LongAddressGen(Reg::r30, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x10fffffe);
   }
@@ -117,8 +117,8 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   {
     // Will generate nop + adr.
     BiDirectionalLabel Label;
-    LongAddressGen(Reg::r30, &Label);
-    Bind(&Label);
+    (void)LongAddressGen(Reg::r30, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xd503201f);
@@ -128,7 +128,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   {
     // Will generate adrp.
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     // Move adrp 1MB away.
@@ -136,7 +136,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
       nop();
     }
 
-    LongAddressGen(Reg::r30, &Label);
+    (void)LongAddressGen(Reg::r30, &Label);
     nop();
     CHECK(DisassembleEncoding(262145) == 0x90fff81e);
     CHECK(DisassembleEncoding(262146) == 0xd503201f);
@@ -145,14 +145,14 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   {
     // Will generate nop + adrp.
     ForwardLabel Label;
-    LongAddressGen(Reg::r30, &Label);
+    (void)LongAddressGen(Reg::r30, &Label);
 
     // Move label 1MB away, plus a page, and then aligned to a page.
     for (size_t i = 0; i < ((1 * 1024 * 1024 + 4096) / 4 - 2); ++i) {
       nop();
     }
 
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xd503201f);
@@ -162,14 +162,14 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   {
     // Will generate adrp + add.
     ForwardLabel Label;
-    LongAddressGen(Reg::r30, &Label);
+    (void)LongAddressGen(Reg::r30, &Label);
 
     // Move label 1MB away, plus a page, plus one instruction.
     for (size_t i = 0; i < ((1 * 1024 * 1024 + 4096) / 4 - 1); ++i) {
       nop();
     }
 
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb000081e);
@@ -180,7 +180,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   {
     // Will generate adrp.
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     // Move adrp 1MB away.
@@ -188,7 +188,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
       nop();
     }
 
-    LongAddressGen(Reg::r30, &Label);
+    (void)LongAddressGen(Reg::r30, &Label);
     nop();
     CHECK(DisassembleEncoding(262145) == 0x90fff81e);
     CHECK(DisassembleEncoding(262146) == 0xd503201f);
@@ -197,14 +197,14 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   {
     // Will generate nop + adrp.
     BiDirectionalLabel Label;
-    LongAddressGen(Reg::r30, &Label);
+    (void)LongAddressGen(Reg::r30, &Label);
 
     // Move label 1MB away, plus a page, and then aligned to a page.
     for (size_t i = 0; i < ((1 * 1024 * 1024 + 4096) / 4 - 2); ++i) {
       nop();
     }
 
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xd503201f);
@@ -214,14 +214,14 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: PC relative") {
   {
     // Will generate adrp + add.
     BiDirectionalLabel Label;
-    LongAddressGen(Reg::r30, &Label);
+    (void)LongAddressGen(Reg::r30, &Label);
 
     // Move label 1MB away, plus a page, plus one instruction.
     for (size_t i = 0; i < ((1 * 1024 * 1024 + 4096) / 4 - 1); ++i) {
       nop();
     }
 
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb000081e);

FEXCore/unittests/Emitter/Branch_Tests.cpp

@@ -9,17 +9,17 @@ using namespace ARMEmitter;
 TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Conditional branch immediate") {
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    b(Condition::CC_PL, &Label);
+    (void)b(Condition::CC_PL, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x54ffffe5);
   }
 
   {
     ForwardLabel Label;
-    b(Condition::CC_PL, &Label);
-    Bind(&Label);
+    (void)b(Condition::CC_PL, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x54000025);
@@ -27,17 +27,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Conditional branch immediat
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    b(Condition::CC_PL, &Label);
+    (void)b(Condition::CC_PL, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x54ffffe5);
   }
 
   {
     BiDirectionalLabel Label;
-    b(Condition::CC_PL, &Label);
-    Bind(&Label);
+    (void)b(Condition::CC_PL, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x54000025);
@@ -46,17 +46,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Conditional branch immediat
 TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Branch consistent conditional") {
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    bc(Condition::CC_PL, &Label);
+    (void)bc(Condition::CC_PL, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x54fffff5);
   }
 
   {
     ForwardLabel Label;
-    bc(Condition::CC_PL, &Label);
-    Bind(&Label);
+    (void)bc(Condition::CC_PL, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x54000035);
@@ -64,17 +64,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Branch consistent condition
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    bc(Condition::CC_PL, &Label);
+    (void)bc(Condition::CC_PL, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x54fffff5);
   }
 
   {
     BiDirectionalLabel Label;
-    bc(Condition::CC_PL, &Label);
-    Bind(&Label);
+    (void)bc(Condition::CC_PL, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x54000035);
@@ -89,17 +89,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Unconditional branch regist
 TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Unconditional branch immediate") {
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    b(&Label);
+    (void)b(&Label);
 
     CHECK(DisassembleEncoding(1) == 0x17ffffff);
   }
 
   {
     ForwardLabel Label;
-    b(&Label);
-    Bind(&Label);
+    (void)b(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x14000001);
@@ -107,17 +107,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Unconditional branch immedi
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    b(&Label);
+    (void)b(&Label);
 
     CHECK(DisassembleEncoding(1) == 0x17ffffff);
   }
 
   {
     BiDirectionalLabel Label;
-    b(&Label);
-    Bind(&Label);
+    (void)b(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x14000001);
@@ -125,17 +125,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Unconditional branch immedi
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    bl(&Label);
+    (void)bl(&Label);
 
     CHECK(DisassembleEncoding(1) == 0x97ffffff);
   }
 
   {
     ForwardLabel Label;
-    bl(&Label);
-    Bind(&Label);
+    (void)bl(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x94000001);
@@ -143,17 +143,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Unconditional branch immedi
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    bl(&Label);
+    (void)bl(&Label);
 
     CHECK(DisassembleEncoding(1) == 0x97ffffff);
   }
 
   {
     BiDirectionalLabel Label;
-    bl(&Label);
-    Bind(&Label);
+    (void)bl(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x94000001);
@@ -162,17 +162,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Unconditional branch immedi
 TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    cbz(Size::i32Bit, Reg::r29, &Label);
+    (void)cbz(Size::i32Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x34fffffd);
   }
 
   {
     ForwardLabel Label;
-    cbz(Size::i32Bit, Reg::r29, &Label);
-    Bind(&Label);
+    (void)cbz(Size::i32Bit, Reg::r29, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3400003d);
@@ -180,17 +180,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    cbz(Size::i32Bit, Reg::r29, &Label);
+    (void)cbz(Size::i32Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x34fffffd);
   }
 
   {
     BiDirectionalLabel Label;
-    cbz(Size::i32Bit, Reg::r29, &Label);
-    Bind(&Label);
+    (void)cbz(Size::i32Bit, Reg::r29, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3400003d);
@@ -198,17 +198,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    cbz(Size::i64Bit, Reg::r29, &Label);
+    (void)cbz(Size::i64Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb4fffffd);
   }
 
   {
     ForwardLabel Label;
-    cbz(Size::i64Bit, Reg::r29, &Label);
-    Bind(&Label);
+    (void)cbz(Size::i64Bit, Reg::r29, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb400003d);
@@ -216,17 +216,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    cbz(Size::i64Bit, Reg::r29, &Label);
+    (void)cbz(Size::i64Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb4fffffd);
   }
 
   {
     BiDirectionalLabel Label;
-    cbz(Size::i64Bit, Reg::r29, &Label);
-    Bind(&Label);
+    (void)cbz(Size::i64Bit, Reg::r29, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb400003d);
@@ -234,17 +234,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    cbnz(Size::i32Bit, Reg::r29, &Label);
+    (void)cbnz(Size::i32Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x35fffffd);
   }
 
   {
     ForwardLabel Label;
-    cbnz(Size::i32Bit, Reg::r29, &Label);
-    Bind(&Label);
+    (void)cbnz(Size::i32Bit, Reg::r29, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3500003d);
@@ -252,17 +252,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    cbnz(Size::i32Bit, Reg::r29, &Label);
+    (void)cbnz(Size::i32Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x35fffffd);
   }
 
   {
     BiDirectionalLabel Label;
-    cbnz(Size::i32Bit, Reg::r29, &Label);
-    Bind(&Label);
+    (void)cbnz(Size::i32Bit, Reg::r29, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3500003d);
@@ -270,17 +270,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    cbnz(Size::i64Bit, Reg::r29, &Label);
+    (void)cbnz(Size::i64Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb5fffffd);
   }
 
   {
     ForwardLabel Label;
-    cbnz(Size::i64Bit, Reg::r29, &Label);
-    Bind(&Label);
+    (void)cbnz(Size::i64Bit, Reg::r29, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb500003d);
@@ -288,17 +288,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    cbnz(Size::i64Bit, Reg::r29, &Label);
+    (void)cbnz(Size::i64Bit, Reg::r29, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb5fffffd);
   }
 
   {
     BiDirectionalLabel Label;
-    cbnz(Size::i64Bit, Reg::r29, &Label);
-    Bind(&Label);
+    (void)cbnz(Size::i64Bit, Reg::r29, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb500003d);
@@ -307,17 +307,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Compare and branch") {
 TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate") {
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    tbz(Reg::r29, 0, &Label);
+    (void)tbz(Reg::r29, 0, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x3607fffd);
   }
 
   {
     ForwardLabel Label;
-    tbz(Reg::r29, 0, &Label);
-    Bind(&Label);
+    (void)tbz(Reg::r29, 0, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3600003d);
@@ -325,17 +325,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    tbz(Reg::r29, 0, &Label);
+    (void)tbz(Reg::r29, 0, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x3607fffd);
   }
 
   {
     BiDirectionalLabel Label;
-    tbz(Reg::r29, 0, &Label);
-    Bind(&Label);
+    (void)tbz(Reg::r29, 0, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3600003d);
@@ -343,17 +343,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    tbz(Reg::r29, 63, &Label);
+    (void)tbz(Reg::r29, 63, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb6fffffd);
   }
 
   {
     ForwardLabel Label;
-    tbz(Reg::r29, 63, &Label);
-    Bind(&Label);
+    (void)tbz(Reg::r29, 63, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb6f8003d);
@@ -361,17 +361,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    tbz(Reg::r29, 63, &Label);
+    (void)tbz(Reg::r29, 63, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb6fffffd);
   }
 
   {
     BiDirectionalLabel Label;
-    tbz(Reg::r29, 63, &Label);
-    Bind(&Label);
+    (void)tbz(Reg::r29, 63, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb6f8003d);
@@ -379,17 +379,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    tbnz(Reg::r29, 0, &Label);
+    (void)tbnz(Reg::r29, 0, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x3707fffd);
   }
 
   {
     ForwardLabel Label;
-    tbnz(Reg::r29, 0, &Label);
-    Bind(&Label);
+    (void)tbnz(Reg::r29, 0, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3700003d);
@@ -397,17 +397,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    tbnz(Reg::r29, 0, &Label);
+    (void)tbnz(Reg::r29, 0, &Label);
 
     CHECK(DisassembleEncoding(1) == 0x3707fffd);
   }
 
   {
     BiDirectionalLabel Label;
-    tbnz(Reg::r29, 0, &Label);
-    Bind(&Label);
+    (void)tbnz(Reg::r29, 0, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x3700003d);
@@ -415,17 +415,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    tbnz(Reg::r29, 63, &Label);
+    (void)tbnz(Reg::r29, 63, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb7fffffd);
   }
 
   {
     ForwardLabel Label;
-    tbnz(Reg::r29, 63, &Label);
-    Bind(&Label);
+    (void)tbnz(Reg::r29, 63, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb7f8003d);
@@ -433,17 +433,17 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Branch: Test and branch immediate")
 
   {
     BiDirectionalLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
-    tbnz(Reg::r29, 63, &Label);
+    (void)tbnz(Reg::r29, 63, &Label);
 
     CHECK(DisassembleEncoding(1) == 0xb7fffffd);
   }
 
   {
     BiDirectionalLabel Label;
-    tbnz(Reg::r29, 63, &Label);
-    Bind(&Label);
+    (void)tbnz(Reg::r29, 63, &Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xb7f8003d);

FEXCore/unittests/Emitter/Loadstore_Tests.cpp

@@ -1323,7 +1323,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: LDAPR/STLR unscaled imme
 TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal") {
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
     ldr(WReg::w30, &Label);
 
@@ -1332,7 +1332,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
     ldr(SReg::s30, &Label);
 
@@ -1341,7 +1341,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
     ldr(XReg::x30, &Label);
 
@@ -1350,7 +1350,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
     ldr(DReg::d30, &Label);
 
@@ -1359,7 +1359,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
     ldrsw(XReg::x30, &Label);
 
@@ -1368,7 +1368,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
     ldr(QReg::q30, &Label);
 
@@ -1377,7 +1377,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
 
   {
     BackwardLabel Label;
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
     prfm(Prefetch::PLDL1KEEP, &Label);
 
@@ -1387,7 +1387,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     ldr(WReg::w30, &Label);
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x1800003e);
@@ -1396,7 +1396,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     ldr(SReg::s30, &Label);
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x1c00003e);
@@ -1405,7 +1405,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     ldr(XReg::x30, &Label);
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x5800003e);
@@ -1414,7 +1414,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     ldr(DReg::d30, &Label);
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x5c00003e);
@@ -1423,7 +1423,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     ldrsw(XReg::x30, &Label);
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x9800003e);
@@ -1432,7 +1432,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     ldr(QReg::q30, &Label);
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0x9c00003e);
@@ -1441,7 +1441,7 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: Loadstore: Load register literal")
   {
     ForwardLabel Label;
     prfm(Prefetch::PLDL1KEEP, &Label);
-    Bind(&Label);
+    (void)Bind(&Label);
     dc32(0);
 
     CHECK(DisassembleEncoding(0) == 0xd8000020);

Scripts/InstallFEX.py

@@ -81,11 +81,7 @@ def IsSupportedDistro():
 
     # We only support Ubuntu
     if Distro[0] == "ubuntu":
-        # We only support what is available in ppa:fex-emu/fex
-        return Distro[1] == "22.04" or \
-            Distro[1] == "24.04" or \
-            Distro[1] == "24.10" or \
-            Distro[1] == "25.04"
+        return Distro[1] in {"22.04", "24.04", "24.10", "25.04", "25.10"}
 
     return False
 

Scripts/aarch64_fit_native.py

@@ -81,6 +81,8 @@ BigCoreIDs = {
             [ ["apple-a13", "0.0"], # If we aren't on 12.0+
               ["apple-a14", "12.0"], # Only exists in 12.0+
             ],
+        # QEmu HVF 10.2+
+        tuple([0x61, 0]): "apple-a13", # Can't determine variant, choose lowest.
 }
 
 LittleCoreIDs = {

Source/Common/Config.cpp

@@ -277,15 +277,16 @@ void EnvLoader::Load() {
 #include <FEXCore/Config/ConfigOptions.inl>
 
     if (Value) {
-      EnvMap.insert_or_assign(Key, *Value);
+      EnvMap.insert_or_assign(Key, std::move(*Value));
     } else {
       EnvMap.insert_or_assign(Key, Value_View);
     }
   }
 
-  auto GetVar = [](EnvMapType& EnvMap, const std::string_view id) -> std::optional<std::string_view> {
-    if (EnvMap.find(id) != EnvMap.end()) {
-      return EnvMap.at(id);
+  auto GetVar = [](const EnvMapType& EnvMap, std::string_view id) -> std::optional<std::string_view> {
+    const auto EnvEntry = EnvMap.find(id);
+    if (EnvEntry != EnvMap.end()) {
+      return EnvEntry->second;
     }
 
     // If envp[] was empty, search using std::getenv()
@@ -570,7 +571,7 @@ fextl::string GetConfigDirectory(bool Global, const PortableInformation& Portabl
     return fextl::fmt::format("{}/fex-emu/", PortableInfo.InterpreterPath);
   } else if (PortableInfo.IsPortable && ConfigOverride && !Global) {
     fextl::string AppConfigStr = ConfigOverride;
-    if (PortableInfo.IsPortable && FHU::Filesystem::IsRelative(AppConfigStr)) {
+    if (FHU::Filesystem::IsRelative(AppConfigStr)) {
       AppConfigStr = PortableInfo.InterpreterPath + AppConfigStr;
     }
 

Source/Common/FEXServerClient.cpp

@@ -131,7 +131,7 @@ fextl::string GetServerMountFolder() {
 fextl::string GetServerSocketName() {
   FEX_CONFIG_OPT(ServerSocketPath, SERVERSOCKETPATH);
   if (ServerSocketPath().empty()) {
-    return fextl::fmt::format("{}.FEXServer.Socket", ::geteuid());
+    return fextl::fmt::format("{}.FEXServer.Socket", ::getuid());
   }
   return ServerSocketPath;
 }
@@ -148,7 +148,7 @@ fextl::string GetServerSocketPath() {
   auto Folder = GetTempFolder();
 
   if (name.empty()) {
-    return fextl::fmt::format("{}/{}.FEXServer.Socket", Folder, ::geteuid());
+    return fextl::fmt::format("{}/{}.FEXServer.Socket", Folder, ::getuid());
   } else {
     return fextl::fmt::format("{}/{}", Folder, name);
   }
@@ -300,6 +300,14 @@ int ConnectToAndStartServer(std::string_view InterpreterPath) {
       while (poll(&PollFD, 1, -1) == -1 && errno == EINTR)
         ;
 
+      // Check if child signaled an error
+      uint64_t error = 0;
+      ssize_t bytes_read = read(fds[0], &error, sizeof(error));
+      close(fds[0]);
+      if (bytes_read > 0 && error != 0) {
+        return -1;
+      }
+
       for (size_t i = 0; i < 5; ++i) {
         ServerFD = ConnectToServer(ConnectionOption::Default);
 

Source/Common/FileMappingBaseAddress.h

@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+#pragma once
+
+#include <FEXCore/Utils/TypeDefines.h>
+
+#include <cstdint>
+#include <optional>
+#include <span>
+
+#include <elf.h>
+
+namespace FEXCore {
+
+/**
+ * Infers the base virtual address from a file mapping (as described by parameters to a single
+ * call to mmap()).
+ *
+ * The file offset of any given mapping need not match its virtual address offset from the base
+ * mapping (file offset = 0). Instead, this function searches the corresponding ELF program headers
+ * for an entry that generated the given file mapping.
+ */
+inline std::optional<uint64_t>
+InferMappingBaseAddress(std::span<const Elf64_Phdr> ProgramHeaders, uint64_t Addr, uint64_t Size, uint64_t FileOffset, int AccessFlags) {
+  for (auto& phdr : ProgramHeaders) {
+    if (phdr.p_type != PT_LOAD) {
+      // Skip headers that don't trigger memory mappings
+      continue;
+    }
+
+    if ((phdr.p_flags & (PF_X | PF_W | PF_R)) != (AccessFlags & (PF_X | PF_W | PF_R))) {
+      continue;
+    }
+
+    // The mapped file offset must be included at the start of the section header
+    auto SegmentStartOffset = phdr.p_offset - (phdr.p_vaddr & 0xfff);
+    if (FileOffset >= SegmentStartOffset && FileOffset < SegmentStartOffset + phdr.p_filesz &&
+        (FileOffset & Utils::FEX_PAGE_MASK) == (phdr.p_offset & Utils::FEX_PAGE_MASK)) {
+      // Compute VA offset relative to the base mapping
+      return Addr - (phdr.p_vaddr - (phdr.p_offset & 0xfff)) + (ProgramHeaders[0].p_vaddr - (ProgramHeaders[0].p_offset & 0xfff)) -
+             (FileOffset - SegmentStartOffset);
+    }
+  }
+
+  return std::nullopt;
+}
+} // namespace FEXCore

Source/Common/HostFeatures.cpp

@@ -598,16 +598,6 @@ FEXCore::HostFeatures FetchHostFeatures(FEX::CPUFeatures& Features, bool Support
 #endif
   HostFeatures.SupportsPreserveAllABI = FEX_HAS_PRESERVE_ALL_ATTR;
 
-  if (!Is64BitMode()) {
-    ///< Always disable AVX and AVX2 in 32-bit mode.
-    // When AVX256 is enabled, signal frames start using significantly more stack space.
-    //   - 16bytes * 16 registers = 256 bytes for XMM registers.
-    //   - 32bytes * 16 registers = 512 bytes for YMM registers.
-    // There are known game failures on real x86 hardware where a 32-bit game is running up against the wall on stack space on non-AVX
-    // hardware and then explodes when run on AVX hardware. This is to guard against that.
-    HostFeatures.SupportsAVX = false;
-  }
-
   OverrideFeatures(&HostFeatures, ForceSVEWidth());
   return HostFeatures;
 }

Source/Common/JSONPool.cpp

@@ -2,22 +2,19 @@
 #include "Common/JSONPool.h"
 
 namespace FEX::JSON {
-json_t* PoolInit(jsonPool_t* Pool);
-json_t* PoolAlloc(jsonPool_t* Pool);
+static json_t* PoolInit(jsonPool_t* Pool) {
+  auto* alloc = static_cast<JsonAllocator*>(Pool);
+  return &*alloc->json_objects.emplace(alloc->json_objects.end());
+}
+
+static json_t* PoolAlloc(jsonPool_t* Pool) {
+  auto* alloc = static_cast<JsonAllocator*>(Pool);
+  return &*alloc->json_objects.emplace(alloc->json_objects.end());
+}
 
 JsonAllocator::JsonAllocator()
   : jsonPool_t {
-      .init = FEX::JSON::PoolInit,
-      .alloc = FEX::JSON::PoolAlloc,
+      .init = PoolInit,
+      .alloc = PoolAlloc,
     } {}
-
-json_t* PoolInit(jsonPool_t* Pool) {
-  JsonAllocator* alloc = static_cast<JsonAllocator*>(Pool);
-  return &*alloc->json_objects.emplace(alloc->json_objects.end());
-}
-
-json_t* PoolAlloc(jsonPool_t* Pool) {
-  JsonAllocator* alloc = static_cast<JsonAllocator*>(Pool);
-  return &*alloc->json_objects.emplace(alloc->json_objects.end());
-}
 } // namespace FEX::JSON

Source/Common/SHMStats.cpp

@@ -13,10 +13,11 @@ void StatAllocBase::SaveHeader(FEXCore::SHMStats::AppType AppType) {
   Head = reinterpret_cast<FEXCore::SHMStats::ThreadStatsHeader*>(Base);
   Head->Size.store(CurrentSize, std::memory_order_relaxed);
   Head->Version = FEXCore::SHMStats::STATS_VERSION;
+  Head->app_type = AppType;
+  Head->ThreadStatsSize = sizeof(FEXCore::SHMStats::ThreadStats);
 
   std::string_view GitString = GIT_DESCRIBE_STRING;
   strncpy(Head->fex_version, GitString.data(), std::min(GitString.size(), sizeof(Head->fex_version)));
-  Head->app_type = AppType;
 
   Stats = reinterpret_cast<FEXCore::SHMStats::ThreadStats*>(reinterpret_cast<uint64_t>(Base) + sizeof(FEXCore::SHMStats::ThreadStatsHeader));
 

Source/Tools/CodeSizeValidation/Main.cpp

@@ -447,18 +447,6 @@ public:
     return 0;
   }
 
-  FEXCore::HLE::SyscallABI GetSyscallABI(uint64_t Syscall) override {
-    if (Syscall == 0) {
-      // Claim syscall 0 is simple for instcountci inline tests.
-      return FEXCore::HLE::SyscallABI {
-        .NumArgs = 0,
-        .HasReturn = true,
-        .HostSyscallNumber = 0, // Just map to host syscall zero, it isn't going to get called.
-      };
-    }
-    return {0, false, -1};
-  }
-
   // These are no-ops implementations of the SyscallHandler API
   std::optional<FEXCore::ExecutableFileSectionInfo>
   LookupExecutableFileSection(FEXCore::Core::InternalThreadState& Thread, uint64_t GuestAddr) override {

Source/Tools/CommonTools/DummyHandlers.h

@@ -16,11 +16,6 @@ public:
     return 0;
   }
 
-  FEXCore::HLE::SyscallABI GetSyscallABI(uint64_t Syscall) override {
-    // Don't do anything
-    return {0, false, 0};
-  }
-
   // These are no-ops implementations of the SyscallHandler API
   std::optional<FEXCore::ExecutableFileSectionInfo> LookupExecutableFileSection(FEXCore::Core::InternalThreadState&, uint64_t) override {
     return std::nullopt;

Source/Tools/CommonTools/Linux/Utils/ELFContainer.cpp

@@ -209,7 +209,7 @@ bool ELFContainer::LoadELF_32() {
   DynamicProgram = Header._32.e_type != ET_EXEC;
 
   // Default BRK size
-  BRKSize = 4096;
+  BRKSize = FEXCore::Utils::FEX_PAGE_SIZE;
 
   return true;
 }
@@ -344,7 +344,7 @@ void ELFContainer::CalculateMemoryLayouts() {
   }
 
   // Calculate BRK
-  MaxPhysAddr = FEXCore::AlignUp(MaxPhysAddr, 4096);
+  MaxPhysAddr = FEXCore::AlignUp(MaxPhysAddr, FEXCore::Utils::FEX_PAGE_SIZE);
   BRKBase = MaxPhysAddr;
   MaxPhysAddr += BRKSize;
 

Source/Tools/CommonTools/Linux/Utils/ELFParser.h

@@ -182,17 +182,17 @@ struct ELFParser {
     } else {
       phdrs.resize(ehdr.e_phnum);
 
-      if (pread(fd, &phdrs[0], sizeof(Elf64_Phdr) * ehdr.e_phnum, ehdr.e_phoff) == -1) {
+      if (pread(fd, phdrs.data(), sizeof(Elf64_Phdr) * ehdr.e_phnum, ehdr.e_phoff) == -1) {
         LogMan::Msg::EFmt("Failed to read phdr64 from '{}'", fd);
         return false;
       }
     }
 
-    for (auto phdr : phdrs) {
+    for (const auto& phdr : phdrs) {
       if (phdr.p_type == PT_INTERP) {
         InterpreterElf.resize(phdr.p_filesz);
 
-        if (pread(fd, &InterpreterElf[0], phdr.p_filesz, phdr.p_offset) == -1) {
+        if (pread(fd, InterpreterElf.data(), phdr.p_filesz, phdr.p_offset) == -1) {
           LogMan::Msg::EFmt("Failed to read interpreter from '{}'", fd);
           return false;
         }
@@ -202,10 +202,9 @@ struct ELFParser {
     return true;
   }
 
-  ptrdiff_t FileToVA(off_t FileOffset) {
-    for (auto phdr : phdrs) {
+  ptrdiff_t FileToVA(off_t FileOffset) const {
+    for (const auto& phdr : phdrs) {
       if (phdr.p_offset <= FileOffset && (phdr.p_offset + phdr.p_filesz) > FileOffset) {
-
         auto SectionFileOffset = FileOffset - phdr.p_offset;
 
         if (SectionFileOffset < phdr.p_memsz) {
@@ -217,10 +216,9 @@ struct ELFParser {
     return {};
   }
 
-  off_t VAToFile(ptrdiff_t VAOffset) {
-    for (auto phdr : phdrs) {
+  off_t VAToFile(ptrdiff_t VAOffset) const {
+    for (const auto& phdr : phdrs) {
       if (phdr.p_vaddr <= VAOffset && (phdr.p_vaddr + phdr.p_memsz) > VAOffset) {
-
         auto SectionVAOffset = VAOffset - phdr.p_vaddr;
 
         if (SectionVAOffset < phdr.p_filesz) {

Source/Tools/FEXConfig/main.qml

@@ -507,13 +507,8 @@ ApplicationWindow {
 
                     ButtonGroup {
                         id: tsoButtonGroup
-                        buttons: [tso1, tso2, tso3]
-                        // Trying to be too clever here will trigger property binding loops,
-                        // so require both TSOEnabled and ParanoidTSO to be listed in the config.
-                        // If they are not, the state will be displayed as undetermined.
-                        checkedButton: !(ConfigModel.has("TSOEnabled", refreshCache) && (ConfigModel.has("ParanoidTSO", refreshCache))) ? null
-                                        : ConfigModel.getBool("ParanoidTSO", refreshCache) ? tso3
-                                        : ConfigModel.getBool("TSOEnabled", refreshCache) ? tso2 : tso1
+                        buttons: [tso1, tso2]
+                        checkedButton: ConfigModel.getBool("TSOEnabled", refreshCache) ? tso2 : tso1
 
                         property int pendingItemChange: -1
 
@@ -524,8 +519,6 @@ ApplicationWindow {
 
                             var newIndex = pendingItemChange
                             var TSOEnabled = newIndex === 1
-                            var ParanoidTSO = newIndex === 2
-                            ConfigModel.setBool("ParanoidTSO", ParanoidTSO)
                             ConfigModel.setBool("TSOEnabled", TSOEnabled)
 
                             pendingItemChange = -1;
@@ -543,8 +536,6 @@ ApplicationWindow {
 
                             var newIndex = pendingItemChange
                             var TSOEnabled = newIndex === 1
-                            var ParanoidTSO = newIndex === 2
-                            ConfigModel.setBool("ParanoidTSO", ParanoidTSO)
                             ConfigModel.setBool("TSOEnabled", TSOEnabled)
 
                             pendingItemChange = -1;
@@ -585,12 +576,6 @@ ApplicationWindow {
                                 }
                             }
                         }
-
-                        RadioButton {
-                            id: tso3
-                            text: qsTr("Overly accurate (paranoid TSO)")
-                            onToggled: tsoButtonGroup.onClickedButton(2)
-                        }
                     }
 
                     ConfigCheckBox {
@@ -721,22 +706,10 @@ ApplicationWindow {
             }
 
             ConfigCheckBox {
-                id: x87ReducedPrecisionCheckbox
                 text: qsTr("Reduced x87 precision")
                 config: "X87ReducedPrecision"
             }
 
-            ConfigCheckBox {
-                text: qsTr("Strict reduced x87 precision")
-                config: "X87StrictReducedPrecision"
-                enabled: x87ReducedPrecisionCheckbox.checked
-            }
-
-            ConfigCheckBox {
-                text: qsTr("Unsafe local flags optimization")
-                config: "ABILocalFlags"
-            }
-
             ConfigCheckBox {
                 text: qsTr("Disable JIT optimization passes")
                 config: "O0"

Source/Tools/LinuxEmulation/ArchHelpers/MContext.cpp

@@ -0,0 +1,51 @@
+// SPDX-License-Identifier: MIT
+#include "ArchHelpers/MContext.h"
+
+namespace FEX::ArchHelpers::Context {
+#ifdef _M_ARM_64
+std::string_view GetESRName(uint64_t ESR) {
+  switch ((ESR & ESR1_EC) >> 26) {
+  case 0b000'000: return "Unknown";
+  case 0b000'001: return "Trapped WF*";
+  case 0b000'011: return "Trapped MCR/MRC";
+  case 0b000'100: return "Trapped MCRR/MRRC";
+  case 0b000'101: return "Trapped MCR/MRC (coproc==0b1110)";
+  case 0b000'110: return "Trapped LDC/STC";
+  case 0b000'111: return "Trapped SME;SVE,ASIMD,FP";
+  case 0b001'010: return "Trapped non-covered instruction";
+  case 0b001'100: return "Trapped MRRC (coproc==0b1110)";
+  case 0b001'101: return "Branch target exception";
+  case 0b001'110: return "Illegal Execution State";
+  case 0b010'001: return "AArch32 SVC";
+  case 0b010'100: return "Trapped MSRR/MRRS/System instruction";
+  case 0b010'101: return "AArch64 SVC";
+  case 0b011'000: return "Trapped MSR/MRS/System instruction";
+  case 0b011'001: return "Trapped SVE from ZEN";
+  case 0b011'011: return "TSTART Exception";
+  case 0b011'100: return "PAC Exception";
+  case 0b011'101: return "Trapped SME from SMEN";
+  case 0b100'000: return "Instruction abort";
+  case 0b100'001: return "Instruction abort w/o change to exception level";
+  case 0b100'010: return "PC Alignment fault";
+  case 0b100'100: return "Data abort";
+  case 0b100'101: return "Data abort w/o change to exception level";
+  case 0b100'110: return "SP Alignment fault";
+  case 0b100'111: return "Memory operation exception";
+  case 0b101'000: return "AArch32 Trapped FP Exception";
+  case 0b101'100: return "AArch64 Trapped FP Exception";
+  case 0b101'101: return "GCS exception";
+  case 0b101'111: return "SError exception";
+  case 0b110'000: return "BP Exception";
+  case 0b110'001: return "BP Exception w/o change to exception level";
+  case 0b110'010: return "Software step Exception";
+  case 0b110'011: return "Software step Exception w/o change to exception level";
+  case 0b110'100: return "Watchpoint Exception";
+  case 0b110'101: return "Watchpoit Exception w/o change to exception level";
+  case 0b111'000: return "AArch32 BKPT";
+  case 0b111'100: return "AArch64 BRK";
+  case 0b111'101: return "Profiling Exception";
+  default: return "Reserved";
+  }
+}
+#endif
+} // namespace FEX::ArchHelpers::Context

Source/Tools/LinuxEmulation/ArchHelpers/MContext.h

@@ -203,9 +203,12 @@ constexpr static uint64_t ESR1_DataAbort_Level_EL2 = 0b01;
 constexpr static uint64_t ESR1_DataAbort_Level_EL1 = 0b10;
 constexpr static uint64_t ESR1_DataAbort_Level_EL0 = 0b11;
 
+std::string_view GetESRName(uint64_t ESR);
+
 static inline uint32_t GetProtectFlags(void* ucontext) {
   uint64_t ESR = GetArmESR(ucontext);
-  LOGMAN_THROW_A_FMT((ESR & ESR1_EC) == ESR1_EC_DataAbort, "Unknown ESR1 EC type: 0x{:x} != 0x{:x}", ESR & ESR1_EC, ESR1_EC_DataAbort);
+  LOGMAN_THROW_A_FMT((ESR & ESR1_EC) == ESR1_EC_DataAbort, "Unknown ESR1 EC type: 0x{:x} != 0x{:x}. Received '{}'", ESR & ESR1_EC,
+                     ESR1_EC_DataAbort, GetESRName(ESR));
 
   uint32_t ProtectFlags {};
   if ((ESR & ESR1_DataAbort_Level) == ESR1_DataAbort_Level_EL0) {

Source/Tools/LinuxEmulation/CMakeLists.txt

@@ -3,6 +3,7 @@ add_compile_options(-fno-operator-names)
 set (SRCS
   VDSO_Emulation.cpp
   Thunks.cpp
+  ArchHelpers/MContext.cpp
   GdbServer/Info.cpp
   LinuxSyscalls/GdbServer.cpp
   LinuxSyscalls/EmulatedFiles/EmulatedFiles.cpp

Source/Tools/LinuxEmulation/LinuxSyscalls/FileManagement.cpp

@@ -123,14 +123,14 @@ void FileManager::LoadThunkDatabase(fextl::unordered_map<fextl::string, ThunkDBO
         } else if (ItemName == "Depends") {
           jsonType_t PropertyType = json_getType(LibraryItem);
           if (PropertyType == JSON_TEXT) {
-            DBObject->second.Depends.insert(json_getValue(LibraryItem));
+            DBObject->second.Depends.emplace(json_getValue(LibraryItem));
           } else if (PropertyType == JSON_ARRAY) {
             for (const json_t* Depend = json_getChild(LibraryItem); Depend != nullptr; Depend = json_getSibling(Depend)) {
-              DBObject->second.Depends.insert(json_getValue(Depend));
+              DBObject->second.Depends.emplace(json_getValue(Depend));
             }
           }
         } else if (ItemName == "Overlay") {
-          auto AddWithReplacement = [HomeDirectory, &PathPrefixes](ThunkDBObject& DBObject, fextl::string LibraryItem) {
+          auto AddWithReplacement = [HomeDirectory, &PathPrefixes](ThunkDBObject& DBObject, std::string_view LibraryItem) {
             // Walk through template string and fill in prefixes from right to left
 
             using namespace std::string_view_literals;
@@ -144,8 +144,8 @@ void FileManager::LoadThunkDatabase(fextl::unordered_map<fextl::string, ThunkDBO
             // Sort offsets in descending order to enable safe in-place replacement
             std::sort(std::begin(PrefixPositions), std::end(PrefixPositions), std::greater<> {});
 
-            for (auto& LibPrefix : PathPrefixes) {
-              fextl::string Replacement = LibraryItem;
+            for (const auto& LibPrefix : PathPrefixes) {
+              fextl::string Replacement(LibraryItem);
               for (auto PrefixPos : PrefixPositions) {
                 if (PrefixPos == fextl::string::npos) {
                   continue;

Source/Tools/LinuxEmulation/LinuxSyscalls/Seccomp/BPFEmitter.cpp

@@ -6,19 +6,59 @@ $end_info$
 */
 
 #include "LinuxSyscalls/Seccomp/BPFEmitter.h"
-#include "LinuxSyscalls/Seccomp/SeccompEmulator.h"
+
+#include <FEXCore/Utils/AllocatorHooks.h>
+#include <FEXCore/Utils/LogManager.h>
 
 #include <linux/bpf_common.h>
 #include <linux/filter.h>
 #include <linux/seccomp.h>
 
+namespace FEX::HLE {
+
+#define EMIT_INST(x)               \
+  do {                             \
+    if constexpr (CalculateSize) { \
+      OpSize += 4;                 \
+    } else {                       \
+      x;                           \
+    }                              \
+  } while (0)
+
+#define RETURN_ERROR(x)                                                                \
+  if constexpr (CalculateSize) {                                                       \
+    return ~0ULL;                                                                      \
+  } else {                                                                             \
+    static_assert(x == -EINVAL, "Early return error evaluation only supports EINVAL"); \
+    return x;                                                                          \
+  }
+
+#define RETURN_SUCCESS()           \
+  do {                             \
+    if constexpr (CalculateSize) { \
+      return OpSize;               \
+    } else {                       \
+      return 0;                    \
+    }                              \
+  } while (0)
+
 #define VALIDATE(cond)      \
   do {                      \
     if (!(cond)) {          \
       RETURN_ERROR(-EINVAL) \
     }                       \
   } while (0)
-namespace FEX::HLE {
+
+using SizeErrorCheck = decltype([](uint64_t Result) -> bool { return Result == ~0ULL; });
+using EmissionErrorCheck = decltype([](uint64_t Result) { return Result != 0; });
+
+// Register selection comes from function signature.
+constexpr auto REG_A = ARMEmitter::WReg::w0;
+constexpr auto REG_X = ARMEmitter::WReg::w1;
+constexpr auto REG_TMP = ARMEmitter::WReg::w2;
+constexpr auto REG_TMP2 = ARMEmitter::WReg::w3;
+constexpr auto REG_SECCOMP_DATA = ARMEmitter::XReg::x4;
+
 template<bool CalculateSize>
 uint64_t BPFEmitter::HandleLoad(uint32_t BPFIP, const sock_filter* Inst) {
   VALIDATE(BPF_SIZE(Inst->code) == BPF_W);
@@ -158,13 +198,13 @@ uint64_t BPFEmitter::HandleJmp(uint32_t BPFIP, uint32_t NumInst, const sock_filt
     // Must not jump past the end.
     VALIDATE(Target < NumInst);
 
-    fextl::unordered_map<uint32_t, ARMEmitter::ForwardLabel>::iterator TargetLabel {};
+    JumpLabelIterator TargetLabel {};
 
     if constexpr (!CalculateSize) {
       TargetLabel = JumpLabels.try_emplace(Target, ARMEmitter::ForwardLabel {}).first;
     }
 
-    EMIT_INST(b(&TargetLabel->second));
+    EMIT_INST((void)b(&TargetLabel->second));
     break;
   }
   case BPF_JEQ:
@@ -200,16 +240,16 @@ uint64_t BPFEmitter::HandleJmp(uint32_t BPFIP, uint32_t NumInst, const sock_filt
       RETURN_ERROR(-EINVAL);
     }
 
-    fextl::unordered_map<uint32_t, ARMEmitter::ForwardLabel>::iterator TargetTrueLabel {};
-    fextl::unordered_map<uint32_t, ARMEmitter::ForwardLabel>::iterator TargetFalseLabel {};
+    JumpLabelIterator TargetTrueLabel {};
+    JumpLabelIterator TargetFalseLabel {};
 
     if constexpr (!CalculateSize) {
       TargetTrueLabel = JumpLabels.try_emplace(TargetTrue, ARMEmitter::ForwardLabel {}).first;
       TargetFalseLabel = JumpLabels.try_emplace(TargetFalse, ARMEmitter::ForwardLabel {}).first;
     }
 
-    EMIT_INST(b(CompareResultOp, &TargetTrueLabel->second));
-    EMIT_INST(b(&TargetFalseLabel->second));
+    EMIT_INST((void)b(CompareResultOp, &TargetTrueLabel->second));
+    EMIT_INST((void)b(&TargetFalseLabel->second));
     break;
   }
   default: RETURN_ERROR(-EINVAL); // Unknown jump type
@@ -263,7 +303,7 @@ uint64_t BPFEmitter::HandleEmission(uint32_t flags, const sock_fprog* prog) {
     if constexpr (!CalculateSize) {
       auto jump_label = JumpLabels.find(i);
       if (jump_label != JumpLabels.end()) {
-        Bind(&jump_label->second);
+        (void)Bind(&jump_label->second);
       }
     }
 
@@ -344,6 +384,8 @@ uint64_t BPFEmitter::JITFilter(uint32_t flags, const sock_fprog* prog) {
 
   SetBuffer((uint8_t*)FEXCore::Allocator::mmap(nullptr, FuncSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0), FuncSize);
 
+  FEXCore::Allocator::VirtualName("FEXMem_Misc", reinterpret_cast<void*>(GetBufferBase()), FuncSize);
+
   const auto CodeBegin = GetCursorAddress<uint8_t*>();
 
   uint64_t Result = HandleEmission<false, EmissionErrorCheck>(flags, prog);
@@ -359,7 +401,7 @@ uint64_t BPFEmitter::JITFilter(uint32_t flags, const sock_fprog* prog) {
   // Emit the constant pool.
   Align();
   for (auto& Const : ConstPool) {
-    Bind(&Const.second);
+    (void)Bind(&Const.second);
     dc32(Const.first);
   }
 
@@ -369,7 +411,7 @@ uint64_t BPFEmitter::JITFilter(uint32_t flags, const sock_fprog* prog) {
 
   if constexpr (false) {
     // Useful for debugging seccomp filters.
-    LogMan::Msg::DFmt("JITFilter: disas 0x{:x},+{}", (uint64_t)CodeBegin, CodeOnlySize);
+    LogMan::Msg::DFmt("JITFilter: disas 0x{:x},+{}", fmt::ptr(CodeBegin), CodeOnlySize);
   }
 
   ConstPool.clear();

Source/Tools/LinuxEmulation/LinuxSyscalls/Seccomp/BPFEmitter.h

@@ -49,47 +49,14 @@ private:
   template<bool CalculateSize>
   uint64_t HandleMisc(uint32_t BPFIP, const sock_filter* Inst);
 
-#define EMIT_INST(x)               \
-  do {                             \
-    if constexpr (CalculateSize) { \
-      OpSize += 4;                 \
-    } else {                       \
-      x;                           \
-    }                              \
-  } while (0)
-
-#define RETURN_ERROR(x)                                                                \
-  if constexpr (CalculateSize) {                                                       \
-    return ~0ULL;                                                                      \
-  } else {                                                                             \
-    static_assert(x == -EINVAL, "Early return error evaluation only supports EINVAL"); \
-    return x;                                                                          \
-  }
-
-#define RETURN_SUCCESS()           \
-  do {                             \
-    if constexpr (CalculateSize) { \
-      return OpSize;               \
-    } else {                       \
-      return 0;                    \
-    }                              \
-  } while (0)
-
-  using SizeErrorCheck = decltype([](uint64_t Result) -> bool { return Result == ~0ULL; });
-  using EmissionErrorCheck = decltype([](uint64_t Result) { return Result != 0; });
-
   template<bool CalculateSize, class Pred>
   uint64_t HandleEmission(uint32_t flags, const sock_fprog* prog);
 
-  // Register selection comes from function signature.
-  constexpr static auto REG_A = ARMEmitter::WReg::w0;
-  constexpr static auto REG_X = ARMEmitter::WReg::w1;
-  constexpr static auto REG_TMP = ARMEmitter::WReg::w2;
-  constexpr static auto REG_TMP2 = ARMEmitter::WReg::w3;
-  constexpr static auto REG_SECCOMP_DATA = ARMEmitter::XReg::x4;
   fextl::unordered_map<uint32_t, ARMEmitter::ForwardLabel> JumpLabels;
   fextl::unordered_map<uint32_t, ARMEmitter::ForwardLabel> ConstPool;
 
+  using JumpLabelIterator = decltype(JumpLabels)::iterator;
+
   void* Func {};
   size_t FuncSize {};
 };

Source/Tools/LinuxEmulation/LinuxSyscalls/Seccomp/SeccompEmulator.cpp

@@ -300,6 +300,8 @@ void SeccompEmulator::DeserializeFilters(FEXCore::Core::CpuStateFrame* Frame, in
 
     ::mprotect(Ptr, SFilter.CodeSize, PROT_READ | PROT_EXEC);
 
+    FEXCore::Allocator::VirtualName("FEXMem_Misc", reinterpret_cast<void*>(Ptr), SFilter.CodeSize);
+
     auto& it =
       Filters.emplace_back(SeccompFilterInfo {(SeccompFilterFunc)Ptr, 1, SFilter.CodeSize, SFilter.FilterInstructions, SFilter.ShouldLog});
     TotalFilterInstructions += SFilter.FilterInstructions;

Source/Tools/LinuxEmulation/LinuxSyscalls/SignalDelegator.cpp

@@ -579,80 +579,77 @@ void SignalDelegator::HandleGuestSignal(FEX::HLE::ThreadStateObject* ThreadObjec
   ucontext_t* _context = (ucontext_t*)UContext;
   auto SigInfo = *static_cast<siginfo_t*>(Info);
 
-  constexpr bool SupportDeferredSignals = true;
-  if (SupportDeferredSignals) {
-    auto MustDeferSignal = (Thread->CurrentFrame->State.DeferredSignalRefCount.Load() != 0);
+  auto MustDeferSignal = (Thread->CurrentFrame->State.DeferredSignalRefCount.Load() != 0);
 
-    if (Signal == SIGSEGV && SigInfo.si_code == SEGV_ACCERR && SigInfo.si_addr == reinterpret_cast<void*>(&Thread->InterruptFaultPage)) {
-      if (!MustDeferSignal) {
-        // We just reached the end of the outermost signal-deferring section and faulted to check for pending signals.
-        // Pull a signal frame off the stack.
+  if (Signal == SIGSEGV && SigInfo.si_code == SEGV_ACCERR && SigInfo.si_addr == reinterpret_cast<void*>(&Thread->InterruptFaultPage)) {
+    if (!MustDeferSignal) {
+      // We just reached the end of the outermost signal-deferring section and faulted to check for pending signals.
+      // Pull a signal frame off the stack.
 
-        mprotect(reinterpret_cast<void*>(&Thread->InterruptFaultPage), sizeof(Thread->InterruptFaultPage), PROT_READ | PROT_WRITE);
+      mprotect(reinterpret_cast<void*>(&Thread->InterruptFaultPage), sizeof(Thread->InterruptFaultPage), PROT_READ | PROT_WRITE);
 
-        if (ThreadObject->SignalInfo.DeferredSignalFrames.empty()) {
-          // No signals to defer. Just set the fault page back to RW and continue execution.
-          // This occurs as a minor race condition between the refcount decrement and the access to the fault page.
-          return;
-        }
-
-        const auto& Top = ThreadObject->SignalInfo.DeferredSignalFrames.back();
-        Signal = Top.Signal;
-        SigInfo = Top.Info;
-        // sig mask has been updated at the defer time, recover the original mask
-        memcpy(&_context->uc_sigmask, &Top.SigMask, sizeof(uint64_t));
-        ThreadObject->SignalInfo.DeferredSignalFrames.pop_back();
-
-        // Until we re-protect the page to PROT_NONE, FEX will now *permanently* defer signals and /not/ check them.
-        //
-        // In order to return /back/ to a sane state, we wait for the rt_sigreturn to happen.
-        // rt_sigreturn will check if there are any more deferred signals to handle
-        // - If there are deferred signals
-        //   - mprotect back to PROT_NONE
-        //   - sigreturn will trampoline out to the previous fault address check, SIGSEGV and restart
-        // - If there are *no* deferred signals
-        //  - No need to mprotect, it is already RW
-      } else {
-#ifdef _M_ARM_64
-        // If RefCount != 0 then that means we hit an access with nested signal-deferring sections.
-        // Increment the PC past the `str zr, [x1]` to continue code execution until we reach the outermost section.
-        ArchHelpers::Context::SetPc(UContext, ArchHelpers::Context::GetPc(UContext) + 4);
+      if (ThreadObject->SignalInfo.DeferredSignalFrames.empty()) {
+        // No signals to defer. Just set the fault page back to RW and continue execution.
+        // This occurs as a minor race condition between the refcount decrement and the access to the fault page.
         return;
-#else
-        // X86 should always be doing a refcount compare and branch since we can't guarantee instruction size.
-        // ARM64 just always does the access to reduce branching overhead.
-        ERROR_AND_DIE_FMT("X86 shouldn't hit this InterruptFaultPage");
-#endif
       }
-    } else if (FaultSafeUserMemAccess::TryHandleSafeFault(Signal, SigInfo, UContext)) {
-      ERROR_AND_DIE_FMT("Received invalid data to syscall. Crashing now!");
+
+      const auto& Top = ThreadObject->SignalInfo.DeferredSignalFrames.back();
+      Signal = Top.Signal;
+      SigInfo = Top.Info;
+      // sig mask has been updated at the defer time, recover the original mask
+      memcpy(&_context->uc_sigmask, &Top.SigMask, sizeof(uint64_t));
+      ThreadObject->SignalInfo.DeferredSignalFrames.pop_back();
+
+      // Until we re-protect the page to PROT_NONE, FEX will now *permanently* defer signals and /not/ check them.
+      //
+      // In order to return /back/ to a sane state, we wait for the rt_sigreturn to happen.
+      // rt_sigreturn will check if there are any more deferred signals to handle
+      // - If there are deferred signals
+      //   - mprotect back to PROT_NONE
+      //   - sigreturn will trampoline out to the previous fault address check, SIGSEGV and restart
+      // - If there are *no* deferred signals
+      //  - No need to mprotect, it is already RW
     } else {
-      if (IsAsyncSignal(&SigInfo, Signal) && MustDeferSignal) {
-        // If the signal is asynchronous (as determined by si_code) and FEX is in a state of needing
-        // to defer the signal, then add the signal to the thread's signal queue.
-        LOGMAN_THROW_A_FMT(ThreadObject->SignalInfo.DeferredSignalFrames.size() != ThreadObject->SignalInfo.DeferredSignalFrames.capacity(),
-                           "Deferred signals vector hit "
-                           "capacity size. This will "
-                           "likely crash! Asserting now!");
+#ifdef _M_ARM_64
+      // If RefCount != 0 then that means we hit an access with nested signal-deferring sections.
+      // Increment the PC past the `str zr, [x1]` to continue code execution until we reach the outermost section.
+      ArchHelpers::Context::SetPc(UContext, ArchHelpers::Context::GetPc(UContext) + 4);
+      return;
+#else
+      // X86 should always be doing a refcount compare and branch since we can't guarantee instruction size.
+      // ARM64 just always does the access to reduce branching overhead.
+      ERROR_AND_DIE_FMT("X86 shouldn't hit this InterruptFaultPage");
+#endif
+    }
+  } else if (FaultSafeUserMemAccess::TryHandleSafeFault(Signal, SigInfo, UContext)) {
+    ERROR_AND_DIE_FMT("Received invalid data to syscall. Crashing now!");
+  } else {
+    if (IsAsyncSignal(&SigInfo, Signal) && MustDeferSignal) {
+      // If the signal is asynchronous (as determined by si_code) and FEX is in a state of needing
+      // to defer the signal, then add the signal to the thread's signal queue.
+      LOGMAN_THROW_A_FMT(ThreadObject->SignalInfo.DeferredSignalFrames.size() != ThreadObject->SignalInfo.DeferredSignalFrames.capacity(),
+                         "Deferred signals vector hit "
+                         "capacity size. This will "
+                         "likely crash! Asserting now!");
 
-        ThreadObject->SignalInfo.DeferredSignalFrames.emplace_back(ThreadStateObject::DeferredSignalState {
-          .Info = SigInfo,
-          .Signal = Signal,
-          .SigMask = _context->uc_sigmask.__val[0],
-        });
+      ThreadObject->SignalInfo.DeferredSignalFrames.emplace_back(ThreadStateObject::DeferredSignalState {
+        .Info = SigInfo,
+        .Signal = Signal,
+        .SigMask = _context->uc_sigmask.__val[0],
+      });
 
-        uint64_t NewMask = GetNewSigMask(Signal);
+      uint64_t NewMask = GetNewSigMask(Signal);
 
-        // Update our host signal mask so we don't hit race conditions with signals
-        // This allows us to maintain the expected signal mask through the guest signal handling and then all the way back again
-        memcpy(&_context->uc_sigmask, &NewMask, sizeof(uint64_t));
+      // Update our host signal mask so we don't hit race conditions with signals
+      // This allows us to maintain the expected signal mask through the guest signal handling and then all the way back again
+      memcpy(&_context->uc_sigmask, &NewMask, sizeof(uint64_t));
 
-        // Now update the faulting page permissions so it will fault on write.
-        mprotect(reinterpret_cast<void*>(&Thread->InterruptFaultPage), sizeof(Thread->InterruptFaultPage), PROT_NONE);
+      // Now update the faulting page permissions so it will fault on write.
+      mprotect(reinterpret_cast<void*>(&Thread->InterruptFaultPage), sizeof(Thread->InterruptFaultPage), PROT_NONE);
 
-        // Postpone the remainder of signal handling logic until we process the SIGSEGV triggered by writing to InterruptFaultPage.
-        return;
-      }
+      // Postpone the remainder of signal handling logic until we process the SIGSEGV triggered by writing to InterruptFaultPage.
+      return;
     }
   }
 
@@ -873,9 +870,7 @@ SignalDelegator::SignalDelegator(FEXCore::Context::Context* _CTX, const std::str
   HostHandlers[SIGKILL].Installed = true;
   HostHandlers[SIGSTOP].Installed = true;
 
-  if (ParanoidTSO()) {
-    UnalignedHandlerType = FEXCore::ArchHelpers::Arm64::UnalignedHandlerType::Paranoid;
-  } else if (HalfBarrierTSOEnabled()) {
+  if (HalfBarrierTSOEnabled()) {
     UnalignedHandlerType = FEXCore::ArchHelpers::Arm64::UnalignedHandlerType::HalfBarrier;
   } else {
     UnalignedHandlerType = FEXCore::ArchHelpers::Arm64::UnalignedHandlerType::NonAtomic;
@@ -995,6 +990,7 @@ void SignalDelegator::RegisterTLSState(FEX::HLE::ThreadStateObject* Thread) {
   // Set up our signal alternative stack
   // This is per thread rather than per signal
   Thread->SignalInfo.AltStackPtr = FEXCore::Allocator::mmap(nullptr, SIGSTKSZ * 16, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+  FEXCore::Allocator::VirtualName("FEXMem_Misc", reinterpret_cast<void*>(Thread->SignalInfo.AltStackPtr), SIGSTKSZ * 16);
   stack_t altstack {};
   altstack.ss_sp = reinterpret_cast<void*>(reinterpret_cast<uint64_t>(Thread->SignalInfo.AltStackPtr) + 8);
   altstack.ss_size = SIGSTKSZ * 16 - 8;
@@ -1005,7 +1001,7 @@ void SignalDelegator::RegisterTLSState(FEX::HLE::ThreadStateObject* Thread) {
   memcpy(Thread->SignalInfo.AltStackPtr, &Thread, sizeof(void*));
 
   // Protect the first page of the alt-stack for overflow protection.
-  mprotect(Thread->SignalInfo.AltStackPtr, 4096, PROT_READ);
+  mprotect(Thread->SignalInfo.AltStackPtr, FEXCore::Utils::FEX_PAGE_SIZE, PROT_READ);
 
   // Register the alt stack
   const int Result = sigaltstack(&altstack, nullptr);

Source/Tools/LinuxEmulation/LinuxSyscalls/SignalDelegator.h

@@ -165,7 +165,6 @@ private:
 
   FEX_CONFIG_OPT(Is64BitMode, IS64BIT_MODE);
   const fextl::string ApplicationName;
-  FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
   FEX_CONFIG_OPT(HalfBarrierTSOEnabled, HALFBARRIERTSOENABLED);
 
   FEXCore::ArchHelpers::Arm64::UnalignedHandlerType UnalignedHandlerType {FEXCore::ArchHelpers::Arm64::UnalignedHandlerType::HalfBarrier};

Source/Tools/LinuxEmulation/LinuxSyscalls/Syscalls.cpp

@@ -707,7 +707,7 @@ uint64_t SyscallHandler::HandleBRK(FEXCore::Core::CpuStateFrame* Frame, void* Ad
       DataSpaceMappedSize = 0;
     } else {
       uint64_t NewSize = NewEnd - DataSpace;
-      uint64_t NewSizeAligned = FEXCore::AlignUp(NewSize, 4096);
+      uint64_t NewSizeAligned = FEXCore::AlignUp(NewSize, FEXCore::Utils::FEX_PAGE_SIZE);
 
       if (NewSizeAligned < DataSpaceMappedSize) {
         // If we are shrinking the brk then munmap the ranges
@@ -721,7 +721,7 @@ uint64_t SyscallHandler::HandleBRK(FEXCore::Core::CpuStateFrame* Frame, void* Ad
 
         DataSpaceMappedSize = NewSizeAligned;
       } else if (NewSize > DataSpaceMappedSize) {
-        uint64_t AllocateNewSize = FEXCore::AlignUp(NewSize, 4096) - DataSpaceMappedSize;
+        uint64_t AllocateNewSize = FEXCore::AlignUp(NewSize, FEXCore::Utils::FEX_PAGE_SIZE) - DataSpaceMappedSize;
         if (!Is64BitMode() && (DataSpace + DataSpaceMappedSize + AllocateNewSize > 0x1'0000'0000ULL)) {
           // If we are 32bit and we tried going about the 32bit limit then out of memory
           return DataSpace + DataSpaceSize;

Source/Tools/LinuxEmulation/LinuxSyscalls/Syscalls.h

@@ -138,9 +138,7 @@ public:
       SyscallPtrArg5 Ptr5;
       SyscallPtrArg6 Ptr6;
     };
-    int32_t HostSyscallNumber;
     uint8_t NumArgs;
-    FEXCore::IR::SyscallFlags Flags;
 #ifdef DEBUG_STRACE
     fextl::string StraceFmt;
 #endif
@@ -150,32 +148,14 @@ public:
     return &Definitions.at(Syscall);
   }
 
-  FEXCore::HLE::SyscallABI GetSyscallABI(uint64_t Syscall) override {
-    if (NeedsSeccomp) {
-      // Override ABI if seccomp is enabled.
-      return {FEXCore::HLE::SyscallArguments::MAX_ARGS, true, -1};
-    }
-    auto& Def = Definitions.at(Syscall);
-    return {Def.NumArgs, true, Def.HostSyscallNumber};
-  }
-
-  FEXCore::IR::SyscallFlags GetSyscallFlags(uint64_t Syscall) const override {
-    if (NeedsSeccomp) {
-      // Override flags if seccomp is enabled.
-      return FEXCore::IR::SyscallFlags::DEFAULT;
-    }
-    auto& Def = Definitions.at(Syscall);
-    return Def.Flags;
-  }
-
-  virtual void RegisterSyscall_32(int SyscallNumber, int32_t HostSyscallNumber, FEXCore::IR::SyscallFlags Flags,
+  virtual void RegisterSyscall_32(int SyscallNumber,
 #ifdef DEBUG_STRACE
                                   const fextl::string& TraceFormatString,
 #endif
                                   void* SyscallHandler, int ArgumentCount) {
   }
 
-  virtual void RegisterSyscall_64(int SyscallNumber, int32_t HostSyscallNumber, FEXCore::IR::SyscallFlags Flags,
+  virtual void RegisterSyscall_64(int SyscallNumber,
 #ifdef DEBUG_STRACE
                                   const fextl::string& TraceFormatString,
 #endif
@@ -661,14 +641,8 @@ inline static uint64_t futimesat_compat(int dirfd, const char* pathname, const T
 } // namespace FEX::HLE
 
 // Registers syscall for both 32bit and 64bit
-#define REGISTER_SYSCALL_IMPL(name, lambda) REGISTER_SYSCALL_IMPL_INTERNAL(name, ~0, FEXCore::IR::SyscallFlags::DEFAULT, lambda)
-
-#define REGISTER_SYSCALL_IMPL_FLAGS(name, flags, lambda) REGISTER_SYSCALL_IMPL_INTERNAL(name, ~0, flags, lambda)
-
-#define REGISTER_SYSCALL_IMPL_PASS_FLAGS(name, flags, lambda) REGISTER_SYSCALL_IMPL_INTERNAL(name, SYSCALL_DEF(name), flags, lambda)
-
-#define REGISTER_SYSCALL_IMPL_INTERNAL(name, number, flags, lambda)                                                 \
-  do {                                                                                                              \
-    FEX::HLE::x64::RegisterSyscall(Handler, FEX::HLE::x64::SYSCALL_x64_##name, (number), (flags), #name, (lambda)); \
-    FEX::HLE::x32::RegisterSyscall(Handler, FEX::HLE::x32::SYSCALL_x86_##name, (number), (flags), #name, (lambda)); \
+#define REGISTER_SYSCALL_IMPL(name, lambda)                                                      \
+  do {                                                                                           \
+    FEX::HLE::x64::RegisterSyscall(Handler, FEX::HLE::x64::SYSCALL_x64_##name, #name, (lambda)); \
+    FEX::HLE::x32::RegisterSyscall(Handler, FEX::HLE::x32::SYSCALL_x86_##name, #name, (lambda)); \
   } while (false)

Source/Tools/LinuxEmulation/LinuxSyscalls/Syscalls/EPoll.cpp

@@ -19,10 +19,9 @@ namespace FEX::HLE {
 void RegisterEpoll(FEX::HLE::SyscallHandler* Handler) {
   using namespace FEXCore::IR;
 
-  REGISTER_SYSCALL_IMPL_FLAGS(epoll_create, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int size) -> uint64_t {
-                                uint64_t Result = epoll_create(size);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(epoll_create, [](FEXCore::Core::CpuStateFrame* Frame, int size) -> uint64_t {
+    uint64_t Result = epoll_create(size);
+    SYSCALL_ERRNO();
+  });
 }
 } // namespace FEX::HLE

Source/Tools/LinuxEmulation/LinuxSyscalls/Syscalls/FD.cpp

@@ -31,127 +31,110 @@ $end_info$
 namespace FEX::HLE {
 void RegisterFD(FEX::HLE::SyscallHandler* Handler) {
   using namespace FEXCore::IR;
-  REGISTER_SYSCALL_IMPL_FLAGS(poll, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, struct pollfd* fds, nfds_t nfds, int timeout) -> uint64_t {
-                                if (nfds) {
-                                  // fds is allowed to be garbage if nfds is zero.
-                                  FaultSafeUserMemAccess::VerifyIsWritable(fds, sizeof(struct pollfd) * nfds);
-                                }
-                                uint64_t Result = ::poll(fds, nfds, timeout);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(poll, [](FEXCore::Core::CpuStateFrame* Frame, struct pollfd* fds, nfds_t nfds, int timeout) -> uint64_t {
+    if (nfds) {
+      // fds is allowed to be garbage if nfds is zero.
+      FaultSafeUserMemAccess::VerifyIsWritable(fds, sizeof(struct pollfd) * nfds);
+    }
+    uint64_t Result = ::poll(fds, nfds, timeout);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(open, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, int flags, uint32_t mode) -> uint64_t {
-                                flags = FEX::HLE::RemapFromX86Flags(flags);
-                                uint64_t Result = FEX::HLE::_SyscallHandler->FM.Open(pathname, flags, mode);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(open, [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, int flags, uint32_t mode) -> uint64_t {
+    flags = FEX::HLE::RemapFromX86Flags(flags);
+    uint64_t Result = FEX::HLE::_SyscallHandler->FM.Open(pathname, flags, mode);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(close, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int fd) -> uint64_t {
-                                uint64_t Result = FEX::HLE::_SyscallHandler->FM.Close(fd);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(close, [](FEXCore::Core::CpuStateFrame* Frame, int fd) -> uint64_t {
+    uint64_t Result = FEX::HLE::_SyscallHandler->FM.Close(fd);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(chown, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, uid_t owner, gid_t group) -> uint64_t {
-                                uint64_t Result = ::chown(pathname, owner, group);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(chown, [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, uid_t owner, gid_t group) -> uint64_t {
+    uint64_t Result = ::chown(pathname, owner, group);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(lchown, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, uid_t owner, gid_t group) -> uint64_t {
-                                uint64_t Result = ::lchown(pathname, owner, group);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(lchown, [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, uid_t owner, gid_t group) -> uint64_t {
+    uint64_t Result = ::lchown(pathname, owner, group);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(access, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, int mode) -> uint64_t {
-                                uint64_t Result = FEX::HLE::_SyscallHandler->FM.Access(pathname, mode);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(access, [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, int mode) -> uint64_t {
+    uint64_t Result = FEX::HLE::_SyscallHandler->FM.Access(pathname, mode);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(pipe, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int pipefd[2]) -> uint64_t {
-                                uint64_t Result = ::pipe(pipefd);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(pipe, [](FEXCore::Core::CpuStateFrame* Frame, int pipefd[2]) -> uint64_t {
+    uint64_t Result = ::pipe(pipefd);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(dup3, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int oldfd, int newfd, int flags) -> uint64_t {
-                                flags = FEX::HLE::RemapFromX86Flags(flags);
-                                uint64_t Result = ::dup3(oldfd, newfd, flags);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(dup3, [](FEXCore::Core::CpuStateFrame* Frame, int oldfd, int newfd, int flags) -> uint64_t {
+    flags = FEX::HLE::RemapFromX86Flags(flags);
+    uint64_t Result = ::dup3(oldfd, newfd, flags);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(inotify_init, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame) -> uint64_t {
-                                uint64_t Result = ::inotify_init();
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(inotify_init, [](FEXCore::Core::CpuStateFrame* Frame) -> uint64_t {
+    uint64_t Result = ::inotify_init();
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(openat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int dirfs, const char* pathname, int flags, uint32_t mode) -> uint64_t {
-                                flags = FEX::HLE::RemapFromX86Flags(flags);
-                                uint64_t Result = FEX::HLE::_SyscallHandler->FM.Openat(dirfs, pathname, flags, mode);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(openat, [](FEXCore::Core::CpuStateFrame* Frame, int dirfs, const char* pathname, int flags, uint32_t mode) -> uint64_t {
+    flags = FEX::HLE::RemapFromX86Flags(flags);
+    uint64_t Result = FEX::HLE::_SyscallHandler->FM.Openat(dirfs, pathname, flags, mode);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(readlinkat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int dirfd, const char* pathname, char* buf, size_t bufsiz) -> uint64_t {
-                                uint64_t Result = FEX::HLE::_SyscallHandler->FM.Readlinkat(dirfd, pathname, buf, bufsiz);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(readlinkat, [](FEXCore::Core::CpuStateFrame* Frame, int dirfd, const char* pathname, char* buf, size_t bufsiz) -> uint64_t {
+    uint64_t Result = FEX::HLE::_SyscallHandler->FM.Readlinkat(dirfd, pathname, buf, bufsiz);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(faccessat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int dirfd, const char* pathname, int mode) -> uint64_t {
-                                uint64_t Result = FEX::HLE::_SyscallHandler->FM.FAccessat(dirfd, pathname, mode);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(faccessat, [](FEXCore::Core::CpuStateFrame* Frame, int dirfd, const char* pathname, int mode) -> uint64_t {
+    uint64_t Result = FEX::HLE::_SyscallHandler->FM.FAccessat(dirfd, pathname, mode);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(faccessat2, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int dirfd, const char* pathname, int mode, int flags) -> uint64_t {
-                                uint64_t Result = FEX::HLE::_SyscallHandler->FM.FAccessat2(dirfd, pathname, mode, flags);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(faccessat2, [](FEXCore::Core::CpuStateFrame* Frame, int dirfd, const char* pathname, int mode, int flags) -> uint64_t {
+    uint64_t Result = FEX::HLE::_SyscallHandler->FM.FAccessat2(dirfd, pathname, mode, flags);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(openat2, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int dirfs, const char* pathname, struct open_how* how, size_t usize) -> uint64_t {
-                                open_how HostHow {};
-                                size_t HostSize = std::min(sizeof(open_how), usize);
-                                memcpy(&HostHow, how, HostSize);
+  REGISTER_SYSCALL_IMPL(
+    openat2, [](FEXCore::Core::CpuStateFrame* Frame, int dirfs, const char* pathname, struct open_how* how, size_t usize) -> uint64_t {
+      open_how HostHow {};
+      size_t HostSize = std::min(sizeof(open_how), usize);
+      memcpy(&HostHow, how, HostSize);
 
-                                HostHow.flags = FEX::HLE::RemapFromX86Flags(HostHow.flags);
-                                uint64_t Result = FEX::HLE::_SyscallHandler->FM.Openat2(dirfs, pathname, &HostHow, HostSize);
-                                SYSCALL_ERRNO();
-                              });
+      HostHow.flags = FEX::HLE::RemapFromX86Flags(HostHow.flags);
+      uint64_t Result = FEX::HLE::_SyscallHandler->FM.Openat2(dirfs, pathname, &HostHow, HostSize);
+      SYSCALL_ERRNO();
+    });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(eventfd, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, uint32_t count) -> uint64_t {
-                                uint64_t Result = ::syscall(SYSCALL_DEF(eventfd2), count, 0);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(eventfd, [](FEXCore::Core::CpuStateFrame* Frame, uint32_t count) -> uint64_t {
+    uint64_t Result = ::syscall(SYSCALL_DEF(eventfd2), count, 0);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(pipe2, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int pipefd[2], int flags) -> uint64_t {
-                                flags = FEX::HLE::RemapFromX86Flags(flags);
-                                uint64_t Result = ::pipe2(pipefd, flags);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(pipe2, [](FEXCore::Core::CpuStateFrame* Frame, int pipefd[2], int flags) -> uint64_t {
+    flags = FEX::HLE::RemapFromX86Flags(flags);
+    uint64_t Result = ::pipe2(pipefd, flags);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(
-    statx, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-    [](FEXCore::Core::CpuStateFrame* Frame, int dirfd, const char* pathname, int flags, uint32_t mask, struct statx* statxbuf) -> uint64_t {
+  REGISTER_SYSCALL_IMPL(
+    statx, [](FEXCore::Core::CpuStateFrame* Frame, int dirfd, const char* pathname, int flags, uint32_t mask, struct statx* statxbuf) -> uint64_t {
       // Flags don't need remapped
       uint64_t Result = FEX::HLE::_SyscallHandler->FM.Statx(dirfd, pathname, flags, mask, statxbuf);
       SYSCALL_ERRNO();
     });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(close_range, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, unsigned int first, unsigned int last, unsigned int flags) -> uint64_t {
-                                uint64_t Result = FEX::HLE::_SyscallHandler->FM.CloseRange(first, last, flags);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(close_range, [](FEXCore::Core::CpuStateFrame* Frame, unsigned int first, unsigned int last, unsigned int flags) -> uint64_t {
+    uint64_t Result = FEX::HLE::_SyscallHandler->FM.CloseRange(first, last, flags);
+    SYSCALL_ERRNO();
+  });
 }
 } // namespace FEX::HLE

Source/Tools/LinuxEmulation/LinuxSyscalls/Syscalls/FS.cpp

@@ -25,65 +25,55 @@ namespace FEX::HLE {
 void RegisterFS(FEX::HLE::SyscallHandler* Handler) {
   using namespace FEXCore::IR;
 
-  REGISTER_SYSCALL_IMPL_FLAGS(rename, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* oldpath, const char* newpath) -> uint64_t {
-                                uint64_t Result = ::rename(oldpath, newpath);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(rename, [](FEXCore::Core::CpuStateFrame* Frame, const char* oldpath, const char* newpath) -> uint64_t {
+    uint64_t Result = ::rename(oldpath, newpath);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(mkdir, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, mode_t mode) -> uint64_t {
-                                uint64_t Result = ::mkdir(pathname, mode);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(mkdir, [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, mode_t mode) -> uint64_t {
+    uint64_t Result = ::mkdir(pathname, mode);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(rmdir, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname) -> uint64_t {
-                                uint64_t Result = ::rmdir(pathname);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(rmdir, [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname) -> uint64_t {
+    uint64_t Result = ::rmdir(pathname);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(link, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* oldpath, const char* newpath) -> uint64_t {
-                                uint64_t Result = ::link(oldpath, newpath);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(link, [](FEXCore::Core::CpuStateFrame* Frame, const char* oldpath, const char* newpath) -> uint64_t {
+    uint64_t Result = ::link(oldpath, newpath);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(unlink, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname) -> uint64_t {
-                                uint64_t Result = ::unlink(pathname);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(unlink, [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname) -> uint64_t {
+    uint64_t Result = ::unlink(pathname);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(symlink, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* target, const char* linkpath) -> uint64_t {
-                                uint64_t Result = ::symlink(target, linkpath);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(symlink, [](FEXCore::Core::CpuStateFrame* Frame, const char* target, const char* linkpath) -> uint64_t {
+    uint64_t Result = ::symlink(target, linkpath);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(readlink, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, char* buf, size_t bufsiz) -> uint64_t {
-                                uint64_t Result = FEX::HLE::_SyscallHandler->FM.Readlink(pathname, buf, bufsiz);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(readlink, [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, char* buf, size_t bufsiz) -> uint64_t {
+    uint64_t Result = FEX::HLE::_SyscallHandler->FM.Readlink(pathname, buf, bufsiz);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(chmod, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, mode_t mode) -> uint64_t {
-                                uint64_t Result = ::chmod(pathname, mode);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(chmod, [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, mode_t mode) -> uint64_t {
+    uint64_t Result = ::chmod(pathname, mode);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(mknod, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, mode_t mode, dev_t dev) -> uint64_t {
-                                uint64_t Result = FEX::HLE::_SyscallHandler->FM.Mknod(pathname, mode, dev);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(mknod, [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, mode_t mode, dev_t dev) -> uint64_t {
+    uint64_t Result = FEX::HLE::_SyscallHandler->FM.Mknod(pathname, mode, dev);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(creat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, mode_t mode) -> uint64_t {
-                                uint64_t Result = ::creat(pathname, mode);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(creat, [](FEXCore::Core::CpuStateFrame* Frame, const char* pathname, mode_t mode) -> uint64_t {
+    uint64_t Result = ::creat(pathname, mode);
+    SYSCALL_ERRNO();
+  });
 
   REGISTER_SYSCALL_IMPL(
     setxattr, [](FEXCore::Core::CpuStateFrame* Frame, const char* path, const char* name, const void* value, size_t size, int flags) -> uint64_t {

Source/Tools/LinuxEmulation/LinuxSyscalls/Syscalls/IO.cpp

@@ -19,16 +19,14 @@ namespace FEX::HLE {
 void RegisterIO(FEX::HLE::SyscallHandler* Handler) {
   using namespace FEXCore::IR;
 
-  REGISTER_SYSCALL_IMPL_FLAGS(iopl, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int level) -> uint64_t {
-                                // Just claim we don't have permission
-                                return -EPERM;
-                              });
+  REGISTER_SYSCALL_IMPL(iopl, [](FEXCore::Core::CpuStateFrame* Frame, int level) -> uint64_t {
+    // Just claim we don't have permission
+    return -EPERM;
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(ioperm, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, unsigned long from, unsigned long num, int turn_on) -> uint64_t {
-                                // ioperm not available on our architecture
-                                return -EPERM;
-                              });
+  REGISTER_SYSCALL_IMPL(ioperm, [](FEXCore::Core::CpuStateFrame* Frame, unsigned long from, unsigned long num, int turn_on) -> uint64_t {
+    // ioperm not available on our architecture
+    return -EPERM;
+  });
 }
 } // namespace FEX::HLE

Source/Tools/LinuxEmulation/LinuxSyscalls/Syscalls/Info.cpp

@@ -37,68 +37,65 @@ using cap_user_data_t = void*;
 void RegisterInfo(FEX::HLE::SyscallHandler* Handler) {
   using namespace FEXCore::IR;
 
-  REGISTER_SYSCALL_IMPL_FLAGS(
-    uname, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY, [](FEXCore::Core::CpuStateFrame* Frame, struct utsname* buf) -> uint64_t {
-      auto Thread = FEX::HLE::ThreadManager::GetStateObjectFromCPUState(Frame);
+  REGISTER_SYSCALL_IMPL(uname, [](FEXCore::Core::CpuStateFrame* Frame, struct utsname* buf) -> uint64_t {
+    auto Thread = FEX::HLE::ThreadManager::GetStateObjectFromCPUState(Frame);
 
-      struct utsname Local {};
-      if (::uname(&Local) == 0) {
-        memcpy(buf->nodename, Local.nodename, sizeof(Local.nodename));
-        static_assert(sizeof(Local.nodename) <= sizeof(buf->nodename));
-        memcpy(buf->domainname, Local.domainname, sizeof(Local.domainname));
-        static_assert(sizeof(Local.domainname) <= sizeof(buf->domainname));
-      } else {
-        strcpy(buf->nodename, "FEXCore");
-        LogMan::Msg::EFmt("Couldn't determine host nodename. Defaulting to '{}'", buf->nodename);
-      }
-      strcpy(buf->sysname, "Linux");
-      uint32_t GuestVersion = FEX::HLE::_SyscallHandler->GetGuestKernelVersion();
-      if (Thread->persona & UNAME26) {
-        // Kernel version converts from 6.x.y to 2.6.60+x.
-        GuestVersion = FEX::HLE::SyscallHandler::KernelVersion(2, 6, 60 + FEX::HLE::SyscallHandler::KernelMinor(GuestVersion));
-      }
-      snprintf(buf->release, sizeof(buf->release), "%d.%d.%d", FEX::HLE::SyscallHandler::KernelMajor(GuestVersion),
-               FEX::HLE::SyscallHandler::KernelMinor(GuestVersion), FEX::HLE::SyscallHandler::KernelPatch(GuestVersion));
+    struct utsname Local {};
+    if (::uname(&Local) == 0) {
+      memcpy(buf->nodename, Local.nodename, sizeof(Local.nodename));
+      static_assert(sizeof(Local.nodename) <= sizeof(buf->nodename));
+      memcpy(buf->domainname, Local.domainname, sizeof(Local.domainname));
+      static_assert(sizeof(Local.domainname) <= sizeof(buf->domainname));
+    } else {
+      strcpy(buf->nodename, "FEXCore");
+      LogMan::Msg::EFmt("Couldn't determine host nodename. Defaulting to '{}'", buf->nodename);
+    }
+    strcpy(buf->sysname, "Linux");
+    uint32_t GuestVersion = FEX::HLE::_SyscallHandler->GetGuestKernelVersion();
+    if (Thread->persona & UNAME26) {
+      // Kernel version converts from 6.x.y to 2.6.60+x.
+      GuestVersion = FEX::HLE::SyscallHandler::KernelVersion(2, 6, 60 + FEX::HLE::SyscallHandler::KernelMinor(GuestVersion));
+    }
+    snprintf(buf->release, sizeof(buf->release), "%d.%d.%d", FEX::HLE::SyscallHandler::KernelMajor(GuestVersion),
+             FEX::HLE::SyscallHandler::KernelMinor(GuestVersion), FEX::HLE::SyscallHandler::KernelPatch(GuestVersion));
 
-      const char version[] = "#" GIT_DESCRIBE_STRING " SMP " __DATE__ " " __TIME__;
-      strcpy(buf->version, version);
-      static_assert(sizeof(version) <= sizeof(buf->version), "uname version define became too large!");
-      if (Thread->persona & PER_LINUX32) {
-        // Tell the guest that we are a 32bit kernel
-        strcpy(buf->machine, "i686");
-      } else {
-        // Tell the guest that we are a 64bit kernel
-        strcpy(buf->machine, "x86_64");
-      }
-      return 0;
-    });
+    const char version[] = "#" GIT_DESCRIBE_STRING " SMP " __DATE__ " " __TIME__;
+    strcpy(buf->version, version);
+    static_assert(sizeof(version) <= sizeof(buf->version), "uname version define became too large!");
+    if (Thread->persona & PER_LINUX32) {
+      // Tell the guest that we are a 32bit kernel
+      strcpy(buf->machine, "i686");
+    } else {
+      // Tell the guest that we are a 64bit kernel
+      strcpy(buf->machine, "x86_64");
+    }
+    return 0;
+  });
 
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(personality, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   [](FEXCore::Core::CpuStateFrame* Frame, uint32_t persona) -> uint64_t {
-                                     auto Thread = FEX::HLE::ThreadManager::GetStateObjectFromCPUState(Frame);
+  REGISTER_SYSCALL_IMPL(personality, [](FEXCore::Core::CpuStateFrame* Frame, uint32_t persona) -> uint64_t {
+    auto Thread = FEX::HLE::ThreadManager::GetStateObjectFromCPUState(Frame);
 
-                                     if (persona == ~0U) {
-                                       // Special case, only queries the persona.
-                                       return Thread->persona;
-                                     }
+    if (persona == ~0U) {
+      // Special case, only queries the persona.
+      return Thread->persona;
+    }
 
-                                     // Mask off `PER_LINUX32` because AArch64 doesn't support it.
-                                     uint32_t NewPersona = persona & ~PER_LINUX32;
+    // Mask off `PER_LINUX32` because AArch64 doesn't support it.
+    uint32_t NewPersona = persona & ~PER_LINUX32;
 
-                                     // This syscall can not physically fail with PER_LINUX32 masked off.
-                                     // It also can not fail on a real x86 kernel.
-                                     (void)::syscall(SYSCALL_DEF(personality), NewPersona);
+    // This syscall can not physically fail with PER_LINUX32 masked off.
+    // It also can not fail on a real x86 kernel.
+    (void)::syscall(SYSCALL_DEF(personality), NewPersona);
 
-                                     // Return the old persona while setting the new one.
-                                     auto OldPersona = Thread->persona;
-                                     Thread->persona = persona;
-                                     return OldPersona;
-                                   });
+    // Return the old persona while setting the new one.
+    auto OldPersona = Thread->persona;
+    Thread->persona = persona;
+    return OldPersona;
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(seccomp, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, unsigned int operation, unsigned int flags, void* args) -> uint64_t {
-                                return FEX::HLE::_SyscallHandler->SeccompEmulator.Handle(Frame, operation, flags, args);
-                              });
+  REGISTER_SYSCALL_IMPL(seccomp, [](FEXCore::Core::CpuStateFrame* Frame, unsigned int operation, unsigned int flags, void* args) -> uint64_t {
+    return FEX::HLE::_SyscallHandler->SeccompEmulator.Handle(Frame, operation, flags, args);
+  });
   REGISTER_SYSCALL_IMPL(
     ptrace, [](FEXCore::Core::CpuStateFrame* Frame, int /*enum __ptrace_request*/ request, pid_t pid, void* addr, void* data) -> uint64_t {
       uint64_t Result {};

Source/Tools/LinuxEmulation/LinuxSyscalls/Syscalls/Memory.cpp

@@ -22,20 +22,18 @@ namespace FEX::HLE {
 void RegisterMemory(FEX::HLE::SyscallHandler* Handler) {
   using namespace FEXCore::IR;
 
-  REGISTER_SYSCALL_IMPL_FLAGS(brk, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, void* addr) -> uint64_t {
-                                uint64_t Result = FEX::HLE::_SyscallHandler->HandleBRK(Frame, addr);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(brk, [](FEXCore::Core::CpuStateFrame* Frame, void* addr) -> uint64_t {
+    uint64_t Result = FEX::HLE::_SyscallHandler->HandleBRK(Frame, addr);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(madvise, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, void* addr, size_t length, int32_t advice) -> uint64_t {
-                                uint64_t Result = ::madvise(addr, length, advice);
+  REGISTER_SYSCALL_IMPL(madvise, [](FEXCore::Core::CpuStateFrame* Frame, void* addr, size_t length, int32_t advice) -> uint64_t {
+    uint64_t Result = ::madvise(addr, length, advice);
 
-                                if (Result != -1) {
-                                  FEX::HLE::_SyscallHandler->TrackMadvise(Frame->Thread, (uintptr_t)addr, length, advice);
-                                }
-                                SYSCALL_ERRNO();
-                              });
+    if (Result != -1) {
+      FEX::HLE::_SyscallHandler->TrackMadvise(Frame->Thread, (uintptr_t)addr, length, advice);
+    }
+    SYSCALL_ERRNO();
+  });
 }
 } // namespace FEX::HLE

Source/Tools/LinuxEmulation/LinuxSyscalls/Syscalls/Passthrough.cpp

@@ -213,349 +213,187 @@ uint64_t SyscallPassthrough7(FEXCore::Core::CpuStateFrame* Frame, uint64_t arg1,
 
 void RegisterCommon(FEX::HLE::SyscallHandler* Handler) {
   using namespace FEXCore::IR;
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(read, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY, SyscallPassthrough3<SYSCALL_DEF(read)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(write, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(write)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(lseek, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(lseek)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(sched_yield, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough0<SYSCALL_DEF(sched_yield)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(msync, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(msync)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(mincore, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(mincore)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(shmget, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(shmget)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(shmctl, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(shmctl)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getpid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough0<SYSCALL_DEF(getpid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(socket, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(socket)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(connect, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(connect)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(sendto, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough6<SYSCALL_DEF(sendto)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(recvfrom, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough6<SYSCALL_DEF(recvfrom)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(shutdown, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(shutdown)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(bind, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY, SyscallPassthrough3<SYSCALL_DEF(bind)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(listen, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(listen)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getsockname, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(getsockname)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getpeername, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(getpeername)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(socketpair, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(socketpair)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(kill, SyscallFlags::DEFAULT, SyscallPassthrough2<SYSCALL_DEF(kill)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(semget, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(semget)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(msgget, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(msgget)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(msgsnd, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(msgsnd)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(msgrcv, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(msgrcv)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(msgctl, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(msgctl)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(flock, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(flock)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fsync, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(fsync)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fdatasync, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(fdatasync)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(truncate, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(truncate)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(ftruncate, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(ftruncate)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getcwd, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(getcwd)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(chdir, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(chdir)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fchdir, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(fchdir)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fchmod, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(fchmod)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fchown, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(fchown)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(umask, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(umask)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough0<SYSCALL_DEF(getuid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(syslog, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(syslog)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getgid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough0<SYSCALL_DEF(getgid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(setuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(setuid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(setgid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(setgid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(geteuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough0<SYSCALL_DEF(geteuid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getegid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough0<SYSCALL_DEF(getegid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(setpgid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(setpgid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getppid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough0<SYSCALL_DEF(getppid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(setsid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough0<SYSCALL_DEF(setsid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(setreuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(setreuid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(setregid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(setregid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getgroups, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(getgroups)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(setgroups, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(setgroups)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(setresuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(setresuid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getresuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(getresuid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(setresgid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(setresgid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getresgid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(getresgid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getpgid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(getpgid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(setfsuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(setfsuid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(setfsgid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(setfsgid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getsid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(getsid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(capget, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(capget)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(capset, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(capset)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getpriority, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(getpriority)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(setpriority, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(setpriority)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(sched_setparam, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(sched_setparam)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(sched_getparam, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(sched_getparam)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(sched_setscheduler, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(sched_setscheduler)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(sched_getscheduler, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(sched_getscheduler)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(sched_get_priority_max, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(sched_get_priority_max)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(sched_get_priority_min, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(sched_get_priority_min)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(mlock, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(mlock)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(munlock, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(munlock)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(pivot_root, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(pivot_root)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(chroot, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(chroot)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(sync, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY, SyscallPassthrough0<SYSCALL_DEF(sync)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(acct, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY, SyscallPassthrough1<SYSCALL_DEF(acct)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(mount, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(mount)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(umount2, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(umount2)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(swapon, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(swapon)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(swapoff, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(swapoff)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(gettid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough0<SYSCALL_DEF(gettid)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fsetxattr, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(fsetxattr)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fgetxattr, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(fgetxattr)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(flistxattr, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(flistxattr)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fremovexattr, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(fremovexattr)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(tkill, SyscallFlags::DEFAULT, SyscallPassthrough2<SYSCALL_DEF(tkill)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(sched_setaffinity, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(sched_setaffinity)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(sched_getaffinity, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(sched_getaffinity)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(io_setup, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(io_setup)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(io_destroy, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(io_destroy)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(io_submit, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(io_submit)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(io_cancel, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(io_cancel)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(remap_file_pages, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(remap_file_pages)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(timer_getoverrun, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(timer_getoverrun)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(timer_delete, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(timer_delete)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(tgkill, SyscallFlags::DEFAULT, SyscallPassthrough3<SYSCALL_DEF(tgkill)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(mbind, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough6<SYSCALL_DEF(mbind)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(set_mempolicy, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(set_mempolicy)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(get_mempolicy, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(get_mempolicy)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(mq_unlink, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(mq_unlink)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(add_key, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(add_key)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(request_key, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(request_key)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(keyctl, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(keyctl)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(ioprio_set, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(ioprio_set)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(ioprio_get, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(ioprio_get)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(inotify_add_watch, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(inotify_add_watch)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(inotify_rm_watch, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(inotify_rm_watch)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(migrate_pages, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(migrate_pages)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(mkdirat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(mkdirat)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(mknodat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(mknodat)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fchownat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(fchownat)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(unlinkat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(unlinkat)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(renameat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(renameat)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(linkat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(linkat)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(symlinkat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(symlinkat)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fchmodat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(fchmodat)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(unshare, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(unshare)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(splice, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough6<SYSCALL_DEF(splice)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(tee, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY, SyscallPassthrough4<SYSCALL_DEF(tee)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(move_pages, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough6<SYSCALL_DEF(move_pages)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(timerfd_create, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(timerfd_create)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(accept4, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(accept4)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(eventfd2, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(eventfd2)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(epoll_create1, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(epoll_create1)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(inotify_init1, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(inotify_init1)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fanotify_init, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(fanotify_init)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fanotify_mark, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(fanotify_mark)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(prlimit_64, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(prlimit_64)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(name_to_handle_at, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(name_to_handle_at)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(open_by_handle_at, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(open_by_handle_at)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(syncfs, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(syncfs)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(setns, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(setns)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getcpu, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(getcpu)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(kcmp, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY, SyscallPassthrough5<SYSCALL_DEF(kcmp)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(sched_setattr, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(sched_setattr)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(sched_getattr, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(sched_getattr)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(renameat2, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(renameat2)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(getrandom, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(getrandom)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(memfd_create, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(memfd_create)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(membarrier, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(membarrier)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(mlock2, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(mlock2)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(copy_file_range, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough6<SYSCALL_DEF(copy_file_range)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(pkey_mprotect, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(pkey_mprotect)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(pkey_alloc, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(pkey_alloc)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(pkey_free, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(pkey_free)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(io_uring_setup, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(io_uring_setup)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(io_uring_enter, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough6<SYSCALL_DEF(io_uring_enter)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(io_uring_register, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(io_uring_register)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(open_tree, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(open_tree)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(move_mount, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(move_mount)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fsopen, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(fsopen)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fsconfig, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(fsconfig)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fsmount, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(fsmount)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(fspick, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(fspick)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(pidfd_open, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(pidfd_open)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(pidfd_getfd, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(pidfd_getfd)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(mount_setattr, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough5<SYSCALL_DEF(mount_setattr)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(quotactl_fd, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(quotactl_fd)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(landlock_create_ruleset, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough3<SYSCALL_DEF(landlock_create_ruleset)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(landlock_add_rule, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough4<SYSCALL_DEF(landlock_add_rule)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(landlock_restrict_self, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(landlock_restrict_self)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(memfd_secret, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough1<SYSCALL_DEF(memfd_secret)>);
-  REGISTER_SYSCALL_IMPL_PASS_FLAGS(process_mrelease, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                   SyscallPassthrough2<SYSCALL_DEF(process_mrelease)>);
+  REGISTER_SYSCALL_IMPL(read, SyscallPassthrough3<SYSCALL_DEF(read)>);
+  REGISTER_SYSCALL_IMPL(write, SyscallPassthrough3<SYSCALL_DEF(write)>);
+  REGISTER_SYSCALL_IMPL(lseek, SyscallPassthrough3<SYSCALL_DEF(lseek)>);
+  REGISTER_SYSCALL_IMPL(sched_yield, SyscallPassthrough0<SYSCALL_DEF(sched_yield)>);
+  REGISTER_SYSCALL_IMPL(msync, SyscallPassthrough3<SYSCALL_DEF(msync)>);
+  REGISTER_SYSCALL_IMPL(mincore, SyscallPassthrough3<SYSCALL_DEF(mincore)>);
+  REGISTER_SYSCALL_IMPL(shmget, SyscallPassthrough3<SYSCALL_DEF(shmget)>);
+  REGISTER_SYSCALL_IMPL(shmctl, SyscallPassthrough3<SYSCALL_DEF(shmctl)>);
+  REGISTER_SYSCALL_IMPL(getpid, SyscallPassthrough0<SYSCALL_DEF(getpid)>);
+  REGISTER_SYSCALL_IMPL(socket, SyscallPassthrough3<SYSCALL_DEF(socket)>);
+  REGISTER_SYSCALL_IMPL(connect, SyscallPassthrough3<SYSCALL_DEF(connect)>);
+  REGISTER_SYSCALL_IMPL(sendto, SyscallPassthrough6<SYSCALL_DEF(sendto)>);
+  REGISTER_SYSCALL_IMPL(recvfrom, SyscallPassthrough6<SYSCALL_DEF(recvfrom)>);
+  REGISTER_SYSCALL_IMPL(shutdown, SyscallPassthrough2<SYSCALL_DEF(shutdown)>);
+  REGISTER_SYSCALL_IMPL(bind, SyscallPassthrough3<SYSCALL_DEF(bind)>);
+  REGISTER_SYSCALL_IMPL(listen, SyscallPassthrough2<SYSCALL_DEF(listen)>);
+  REGISTER_SYSCALL_IMPL(getsockname, SyscallPassthrough3<SYSCALL_DEF(getsockname)>);
+  REGISTER_SYSCALL_IMPL(getpeername, SyscallPassthrough3<SYSCALL_DEF(getpeername)>);
+  REGISTER_SYSCALL_IMPL(socketpair, SyscallPassthrough4<SYSCALL_DEF(socketpair)>);
+  REGISTER_SYSCALL_IMPL(kill, SyscallPassthrough2<SYSCALL_DEF(kill)>);
+  REGISTER_SYSCALL_IMPL(semget, SyscallPassthrough3<SYSCALL_DEF(semget)>);
+  REGISTER_SYSCALL_IMPL(msgget, SyscallPassthrough2<SYSCALL_DEF(msgget)>);
+  REGISTER_SYSCALL_IMPL(msgsnd, SyscallPassthrough4<SYSCALL_DEF(msgsnd)>);
+  REGISTER_SYSCALL_IMPL(msgrcv, SyscallPassthrough5<SYSCALL_DEF(msgrcv)>);
+  REGISTER_SYSCALL_IMPL(msgctl, SyscallPassthrough3<SYSCALL_DEF(msgctl)>);
+  REGISTER_SYSCALL_IMPL(flock, SyscallPassthrough2<SYSCALL_DEF(flock)>);
+  REGISTER_SYSCALL_IMPL(fsync, SyscallPassthrough1<SYSCALL_DEF(fsync)>);
+  REGISTER_SYSCALL_IMPL(fdatasync, SyscallPassthrough1<SYSCALL_DEF(fdatasync)>);
+  REGISTER_SYSCALL_IMPL(truncate, SyscallPassthrough2<SYSCALL_DEF(truncate)>);
+  REGISTER_SYSCALL_IMPL(ftruncate, SyscallPassthrough2<SYSCALL_DEF(ftruncate)>);
+  REGISTER_SYSCALL_IMPL(getcwd, SyscallPassthrough2<SYSCALL_DEF(getcwd)>);
+  REGISTER_SYSCALL_IMPL(chdir, SyscallPassthrough1<SYSCALL_DEF(chdir)>);
+  REGISTER_SYSCALL_IMPL(fchdir, SyscallPassthrough1<SYSCALL_DEF(fchdir)>);
+  REGISTER_SYSCALL_IMPL(fchmod, SyscallPassthrough2<SYSCALL_DEF(fchmod)>);
+  REGISTER_SYSCALL_IMPL(fchown, SyscallPassthrough3<SYSCALL_DEF(fchown)>);
+  REGISTER_SYSCALL_IMPL(umask, SyscallPassthrough1<SYSCALL_DEF(umask)>);
+  REGISTER_SYSCALL_IMPL(getuid, SyscallPassthrough0<SYSCALL_DEF(getuid)>);
+  REGISTER_SYSCALL_IMPL(syslog, SyscallPassthrough3<SYSCALL_DEF(syslog)>);
+  REGISTER_SYSCALL_IMPL(getgid, SyscallPassthrough0<SYSCALL_DEF(getgid)>);
+  REGISTER_SYSCALL_IMPL(setuid, SyscallPassthrough1<SYSCALL_DEF(setuid)>);
+  REGISTER_SYSCALL_IMPL(setgid, SyscallPassthrough1<SYSCALL_DEF(setgid)>);
+  REGISTER_SYSCALL_IMPL(geteuid, SyscallPassthrough0<SYSCALL_DEF(geteuid)>);
+  REGISTER_SYSCALL_IMPL(getegid, SyscallPassthrough0<SYSCALL_DEF(getegid)>);
+  REGISTER_SYSCALL_IMPL(setpgid, SyscallPassthrough2<SYSCALL_DEF(setpgid)>);
+  REGISTER_SYSCALL_IMPL(getppid, SyscallPassthrough0<SYSCALL_DEF(getppid)>);
+  REGISTER_SYSCALL_IMPL(setsid, SyscallPassthrough0<SYSCALL_DEF(setsid)>);
+  REGISTER_SYSCALL_IMPL(setreuid, SyscallPassthrough2<SYSCALL_DEF(setreuid)>);
+  REGISTER_SYSCALL_IMPL(setregid, SyscallPassthrough2<SYSCALL_DEF(setregid)>);
+  REGISTER_SYSCALL_IMPL(getgroups, SyscallPassthrough2<SYSCALL_DEF(getgroups)>);
+  REGISTER_SYSCALL_IMPL(setgroups, SyscallPassthrough2<SYSCALL_DEF(setgroups)>);
+  REGISTER_SYSCALL_IMPL(setresuid, SyscallPassthrough3<SYSCALL_DEF(setresuid)>);
+  REGISTER_SYSCALL_IMPL(getresuid, SyscallPassthrough3<SYSCALL_DEF(getresuid)>);
+  REGISTER_SYSCALL_IMPL(setresgid, SyscallPassthrough3<SYSCALL_DEF(setresgid)>);
+  REGISTER_SYSCALL_IMPL(getresgid, SyscallPassthrough3<SYSCALL_DEF(getresgid)>);
+  REGISTER_SYSCALL_IMPL(getpgid, SyscallPassthrough1<SYSCALL_DEF(getpgid)>);
+  REGISTER_SYSCALL_IMPL(setfsuid, SyscallPassthrough1<SYSCALL_DEF(setfsuid)>);
+  REGISTER_SYSCALL_IMPL(setfsgid, SyscallPassthrough1<SYSCALL_DEF(setfsgid)>);
+  REGISTER_SYSCALL_IMPL(getsid, SyscallPassthrough1<SYSCALL_DEF(getsid)>);
+  REGISTER_SYSCALL_IMPL(capget, SyscallPassthrough2<SYSCALL_DEF(capget)>);
+  REGISTER_SYSCALL_IMPL(capset, SyscallPassthrough2<SYSCALL_DEF(capset)>);
+  REGISTER_SYSCALL_IMPL(getpriority, SyscallPassthrough2<SYSCALL_DEF(getpriority)>);
+  REGISTER_SYSCALL_IMPL(setpriority, SyscallPassthrough3<SYSCALL_DEF(setpriority)>);
+  REGISTER_SYSCALL_IMPL(sched_setparam, SyscallPassthrough2<SYSCALL_DEF(sched_setparam)>);
+  REGISTER_SYSCALL_IMPL(sched_getparam, SyscallPassthrough2<SYSCALL_DEF(sched_getparam)>);
+  REGISTER_SYSCALL_IMPL(sched_setscheduler, SyscallPassthrough3<SYSCALL_DEF(sched_setscheduler)>);
+  REGISTER_SYSCALL_IMPL(sched_getscheduler, SyscallPassthrough1<SYSCALL_DEF(sched_getscheduler)>);
+  REGISTER_SYSCALL_IMPL(sched_get_priority_max, SyscallPassthrough1<SYSCALL_DEF(sched_get_priority_max)>);
+  REGISTER_SYSCALL_IMPL(sched_get_priority_min, SyscallPassthrough1<SYSCALL_DEF(sched_get_priority_min)>);
+  REGISTER_SYSCALL_IMPL(mlock, SyscallPassthrough2<SYSCALL_DEF(mlock)>);
+  REGISTER_SYSCALL_IMPL(munlock, SyscallPassthrough2<SYSCALL_DEF(munlock)>);
+  REGISTER_SYSCALL_IMPL(pivot_root, SyscallPassthrough2<SYSCALL_DEF(pivot_root)>);
+  REGISTER_SYSCALL_IMPL(chroot, SyscallPassthrough1<SYSCALL_DEF(chroot)>);
+  REGISTER_SYSCALL_IMPL(sync, SyscallPassthrough0<SYSCALL_DEF(sync)>);
+  REGISTER_SYSCALL_IMPL(acct, SyscallPassthrough1<SYSCALL_DEF(acct)>);
+  REGISTER_SYSCALL_IMPL(mount, SyscallPassthrough5<SYSCALL_DEF(mount)>);
+  REGISTER_SYSCALL_IMPL(umount2, SyscallPassthrough2<SYSCALL_DEF(umount2)>);
+  REGISTER_SYSCALL_IMPL(swapon, SyscallPassthrough2<SYSCALL_DEF(swapon)>);
+  REGISTER_SYSCALL_IMPL(swapoff, SyscallPassthrough1<SYSCALL_DEF(swapoff)>);
+  REGISTER_SYSCALL_IMPL(gettid, SyscallPassthrough0<SYSCALL_DEF(gettid)>);
+  REGISTER_SYSCALL_IMPL(fsetxattr, SyscallPassthrough5<SYSCALL_DEF(fsetxattr)>);
+  REGISTER_SYSCALL_IMPL(fgetxattr, SyscallPassthrough4<SYSCALL_DEF(fgetxattr)>);
+  REGISTER_SYSCALL_IMPL(flistxattr, SyscallPassthrough3<SYSCALL_DEF(flistxattr)>);
+  REGISTER_SYSCALL_IMPL(fremovexattr, SyscallPassthrough2<SYSCALL_DEF(fremovexattr)>);
+  REGISTER_SYSCALL_IMPL(tkill, SyscallPassthrough2<SYSCALL_DEF(tkill)>);
+  REGISTER_SYSCALL_IMPL(sched_setaffinity, SyscallPassthrough3<SYSCALL_DEF(sched_setaffinity)>);
+  REGISTER_SYSCALL_IMPL(sched_getaffinity, SyscallPassthrough3<SYSCALL_DEF(sched_getaffinity)>);
+  REGISTER_SYSCALL_IMPL(io_setup, SyscallPassthrough2<SYSCALL_DEF(io_setup)>);
+  REGISTER_SYSCALL_IMPL(io_destroy, SyscallPassthrough1<SYSCALL_DEF(io_destroy)>);
+  REGISTER_SYSCALL_IMPL(io_submit, SyscallPassthrough3<SYSCALL_DEF(io_submit)>);
+  REGISTER_SYSCALL_IMPL(io_cancel, SyscallPassthrough3<SYSCALL_DEF(io_cancel)>);
+  REGISTER_SYSCALL_IMPL(remap_file_pages, SyscallPassthrough5<SYSCALL_DEF(remap_file_pages)>);
+  REGISTER_SYSCALL_IMPL(timer_getoverrun, SyscallPassthrough1<SYSCALL_DEF(timer_getoverrun)>);
+  REGISTER_SYSCALL_IMPL(timer_delete, SyscallPassthrough1<SYSCALL_DEF(timer_delete)>);
+  REGISTER_SYSCALL_IMPL(tgkill, SyscallPassthrough3<SYSCALL_DEF(tgkill)>);
+  REGISTER_SYSCALL_IMPL(mbind, SyscallPassthrough6<SYSCALL_DEF(mbind)>);
+  REGISTER_SYSCALL_IMPL(set_mempolicy, SyscallPassthrough3<SYSCALL_DEF(set_mempolicy)>);
+  REGISTER_SYSCALL_IMPL(get_mempolicy, SyscallPassthrough5<SYSCALL_DEF(get_mempolicy)>);
+  REGISTER_SYSCALL_IMPL(mq_unlink, SyscallPassthrough1<SYSCALL_DEF(mq_unlink)>);
+  REGISTER_SYSCALL_IMPL(add_key, SyscallPassthrough5<SYSCALL_DEF(add_key)>);
+  REGISTER_SYSCALL_IMPL(request_key, SyscallPassthrough4<SYSCALL_DEF(request_key)>);
+  REGISTER_SYSCALL_IMPL(keyctl, SyscallPassthrough5<SYSCALL_DEF(keyctl)>);
+  REGISTER_SYSCALL_IMPL(ioprio_set, SyscallPassthrough2<SYSCALL_DEF(ioprio_set)>);
+  REGISTER_SYSCALL_IMPL(ioprio_get, SyscallPassthrough3<SYSCALL_DEF(ioprio_get)>);
+  REGISTER_SYSCALL_IMPL(inotify_add_watch, SyscallPassthrough3<SYSCALL_DEF(inotify_add_watch)>);
+  REGISTER_SYSCALL_IMPL(inotify_rm_watch, SyscallPassthrough2<SYSCALL_DEF(inotify_rm_watch)>);
+  REGISTER_SYSCALL_IMPL(migrate_pages, SyscallPassthrough4<SYSCALL_DEF(migrate_pages)>);
+  REGISTER_SYSCALL_IMPL(mkdirat, SyscallPassthrough3<SYSCALL_DEF(mkdirat)>);
+  REGISTER_SYSCALL_IMPL(mknodat, SyscallPassthrough4<SYSCALL_DEF(mknodat)>);
+  REGISTER_SYSCALL_IMPL(fchownat, SyscallPassthrough5<SYSCALL_DEF(fchownat)>);
+  REGISTER_SYSCALL_IMPL(unlinkat, SyscallPassthrough3<SYSCALL_DEF(unlinkat)>);
+  REGISTER_SYSCALL_IMPL(renameat, SyscallPassthrough4<SYSCALL_DEF(renameat)>);
+  REGISTER_SYSCALL_IMPL(linkat, SyscallPassthrough5<SYSCALL_DEF(linkat)>);
+  REGISTER_SYSCALL_IMPL(symlinkat, SyscallPassthrough3<SYSCALL_DEF(symlinkat)>);
+  REGISTER_SYSCALL_IMPL(fchmodat, SyscallPassthrough3<SYSCALL_DEF(fchmodat)>);
+  REGISTER_SYSCALL_IMPL(unshare, SyscallPassthrough1<SYSCALL_DEF(unshare)>);
+  REGISTER_SYSCALL_IMPL(splice, SyscallPassthrough6<SYSCALL_DEF(splice)>);
+  REGISTER_SYSCALL_IMPL(tee, SyscallPassthrough4<SYSCALL_DEF(tee)>);
+  REGISTER_SYSCALL_IMPL(move_pages, SyscallPassthrough6<SYSCALL_DEF(move_pages)>);
+  REGISTER_SYSCALL_IMPL(timerfd_create, SyscallPassthrough2<SYSCALL_DEF(timerfd_create)>);
+  REGISTER_SYSCALL_IMPL(accept4, SyscallPassthrough4<SYSCALL_DEF(accept4)>);
+  REGISTER_SYSCALL_IMPL(eventfd2, SyscallPassthrough2<SYSCALL_DEF(eventfd2)>);
+  REGISTER_SYSCALL_IMPL(epoll_create1, SyscallPassthrough1<SYSCALL_DEF(epoll_create1)>);
+  REGISTER_SYSCALL_IMPL(inotify_init1, SyscallPassthrough1<SYSCALL_DEF(inotify_init1)>);
+  REGISTER_SYSCALL_IMPL(fanotify_init, SyscallPassthrough2<SYSCALL_DEF(fanotify_init)>);
+  REGISTER_SYSCALL_IMPL(fanotify_mark, SyscallPassthrough5<SYSCALL_DEF(fanotify_mark)>);
+  REGISTER_SYSCALL_IMPL(prlimit_64, SyscallPassthrough4<SYSCALL_DEF(prlimit_64)>);
+  REGISTER_SYSCALL_IMPL(name_to_handle_at, SyscallPassthrough5<SYSCALL_DEF(name_to_handle_at)>);
+  REGISTER_SYSCALL_IMPL(open_by_handle_at, SyscallPassthrough3<SYSCALL_DEF(open_by_handle_at)>);
+  REGISTER_SYSCALL_IMPL(syncfs, SyscallPassthrough1<SYSCALL_DEF(syncfs)>);
+  REGISTER_SYSCALL_IMPL(setns, SyscallPassthrough2<SYSCALL_DEF(setns)>);
+  REGISTER_SYSCALL_IMPL(getcpu, SyscallPassthrough3<SYSCALL_DEF(getcpu)>);
+  REGISTER_SYSCALL_IMPL(kcmp, SyscallPassthrough5<SYSCALL_DEF(kcmp)>);
+  REGISTER_SYSCALL_IMPL(sched_setattr, SyscallPassthrough3<SYSCALL_DEF(sched_setattr)>);
+  REGISTER_SYSCALL_IMPL(sched_getattr, SyscallPassthrough4<SYSCALL_DEF(sched_getattr)>);
+  REGISTER_SYSCALL_IMPL(renameat2, SyscallPassthrough5<SYSCALL_DEF(renameat2)>);
+  REGISTER_SYSCALL_IMPL(getrandom, SyscallPassthrough3<SYSCALL_DEF(getrandom)>);
+  REGISTER_SYSCALL_IMPL(memfd_create, SyscallPassthrough2<SYSCALL_DEF(memfd_create)>);
+  REGISTER_SYSCALL_IMPL(membarrier, SyscallPassthrough2<SYSCALL_DEF(membarrier)>);
+  REGISTER_SYSCALL_IMPL(mlock2, SyscallPassthrough3<SYSCALL_DEF(mlock2)>);
+  REGISTER_SYSCALL_IMPL(copy_file_range, SyscallPassthrough6<SYSCALL_DEF(copy_file_range)>);
+  REGISTER_SYSCALL_IMPL(pkey_mprotect, SyscallPassthrough4<SYSCALL_DEF(pkey_mprotect)>);
+  REGISTER_SYSCALL_IMPL(pkey_alloc, SyscallPassthrough2<SYSCALL_DEF(pkey_alloc)>);
+  REGISTER_SYSCALL_IMPL(pkey_free, SyscallPassthrough1<SYSCALL_DEF(pkey_free)>);
+  REGISTER_SYSCALL_IMPL(io_uring_setup, SyscallPassthrough2<SYSCALL_DEF(io_uring_setup)>);
+  REGISTER_SYSCALL_IMPL(io_uring_enter, SyscallPassthrough6<SYSCALL_DEF(io_uring_enter)>);
+  REGISTER_SYSCALL_IMPL(io_uring_register, SyscallPassthrough4<SYSCALL_DEF(io_uring_register)>);
+  REGISTER_SYSCALL_IMPL(open_tree, SyscallPassthrough3<SYSCALL_DEF(open_tree)>);
+  REGISTER_SYSCALL_IMPL(move_mount, SyscallPassthrough5<SYSCALL_DEF(move_mount)>);
+  REGISTER_SYSCALL_IMPL(fsopen, SyscallPassthrough3<SYSCALL_DEF(fsopen)>);
+  REGISTER_SYSCALL_IMPL(fsconfig, SyscallPassthrough5<SYSCALL_DEF(fsconfig)>);
+  REGISTER_SYSCALL_IMPL(fsmount, SyscallPassthrough3<SYSCALL_DEF(fsmount)>);
+  REGISTER_SYSCALL_IMPL(fspick, SyscallPassthrough3<SYSCALL_DEF(fspick)>);
+  REGISTER_SYSCALL_IMPL(pidfd_open, SyscallPassthrough2<SYSCALL_DEF(pidfd_open)>);
+  REGISTER_SYSCALL_IMPL(pidfd_getfd, SyscallPassthrough3<SYSCALL_DEF(pidfd_getfd)>);
+  REGISTER_SYSCALL_IMPL(mount_setattr, SyscallPassthrough5<SYSCALL_DEF(mount_setattr)>);
+  REGISTER_SYSCALL_IMPL(quotactl_fd, SyscallPassthrough4<SYSCALL_DEF(quotactl_fd)>);
+  REGISTER_SYSCALL_IMPL(landlock_create_ruleset, SyscallPassthrough3<SYSCALL_DEF(landlock_create_ruleset)>);
+  REGISTER_SYSCALL_IMPL(landlock_add_rule, SyscallPassthrough4<SYSCALL_DEF(landlock_add_rule)>);
+  REGISTER_SYSCALL_IMPL(landlock_restrict_self, SyscallPassthrough2<SYSCALL_DEF(landlock_restrict_self)>);
+  REGISTER_SYSCALL_IMPL(memfd_secret, SyscallPassthrough1<SYSCALL_DEF(memfd_secret)>);
+  REGISTER_SYSCALL_IMPL(process_mrelease, SyscallPassthrough2<SYSCALL_DEF(process_mrelease)>);
   if (Handler->IsHostKernelVersionAtLeast(5, 16, 0)) {
-    REGISTER_SYSCALL_IMPL_PASS_FLAGS(futex_waitv, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                     SyscallPassthrough5<SYSCALL_DEF(futex_waitv)>);
+    REGISTER_SYSCALL_IMPL(futex_waitv, SyscallPassthrough5<SYSCALL_DEF(futex_waitv)>);
   } else {
     REGISTER_SYSCALL_IMPL(futex_waitv, UnimplementedSyscallSafe);
   }
   if (Handler->IsHostKernelVersionAtLeast(5, 17, 0)) {
-    REGISTER_SYSCALL_IMPL_PASS_FLAGS(set_mempolicy_home_node, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                     SyscallPassthrough4<SYSCALL_DEF(set_mempolicy_home_node)>);
+    REGISTER_SYSCALL_IMPL(set_mempolicy_home_node, SyscallPassthrough4<SYSCALL_DEF(set_mempolicy_home_node)>);
   } else {
     REGISTER_SYSCALL_IMPL(set_mempolicy_home_node, UnimplementedSyscallSafe);
   }
 
   if (Handler->IsHostKernelVersionAtLeast(6, 8, 0)) {
-    REGISTER_SYSCALL_IMPL_PASS_FLAGS(futex_wake, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                     SyscallPassthrough4<SYSCALL_DEF(futex_wake)>);
-    REGISTER_SYSCALL_IMPL_PASS_FLAGS(futex_wait, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                     SyscallPassthrough6<SYSCALL_DEF(futex_wait)>);
-    REGISTER_SYSCALL_IMPL_PASS_FLAGS(futex_requeue, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                     SyscallPassthrough4<SYSCALL_DEF(futex_requeue)>);
-    REGISTER_SYSCALL_IMPL_PASS_FLAGS(statmount, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                     SyscallPassthrough4<SYSCALL_DEF(statmount)>);
-    REGISTER_SYSCALL_IMPL_PASS_FLAGS(listmount, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                     SyscallPassthrough4<SYSCALL_DEF(listmount)>);
-    REGISTER_SYSCALL_IMPL_PASS_FLAGS(lsm_get_self_attr, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                     SyscallPassthrough4<SYSCALL_DEF(lsm_get_self_attr)>);
-    REGISTER_SYSCALL_IMPL_PASS_FLAGS(lsm_set_self_attr, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                     SyscallPassthrough4<SYSCALL_DEF(lsm_set_self_attr)>);
-    REGISTER_SYSCALL_IMPL_PASS_FLAGS(lsm_list_modules, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                     SyscallPassthrough3<SYSCALL_DEF(lsm_list_modules)>);
+    REGISTER_SYSCALL_IMPL(futex_wake, SyscallPassthrough4<SYSCALL_DEF(futex_wake)>);
+    REGISTER_SYSCALL_IMPL(futex_wait, SyscallPassthrough6<SYSCALL_DEF(futex_wait)>);
+    REGISTER_SYSCALL_IMPL(futex_requeue, SyscallPassthrough4<SYSCALL_DEF(futex_requeue)>);
+    REGISTER_SYSCALL_IMPL(statmount, SyscallPassthrough4<SYSCALL_DEF(statmount)>);
+    REGISTER_SYSCALL_IMPL(listmount, SyscallPassthrough4<SYSCALL_DEF(listmount)>);
+    REGISTER_SYSCALL_IMPL(lsm_get_self_attr, SyscallPassthrough4<SYSCALL_DEF(lsm_get_self_attr)>);
+    REGISTER_SYSCALL_IMPL(lsm_set_self_attr, SyscallPassthrough4<SYSCALL_DEF(lsm_set_self_attr)>);
+    REGISTER_SYSCALL_IMPL(lsm_list_modules, SyscallPassthrough3<SYSCALL_DEF(lsm_list_modules)>);
   } else {
     REGISTER_SYSCALL_IMPL(futex_wake, UnimplementedSyscallSafe);
     REGISTER_SYSCALL_IMPL(futex_wait, UnimplementedSyscallSafe);
@@ -567,8 +405,7 @@ void RegisterCommon(FEX::HLE::SyscallHandler* Handler) {
     REGISTER_SYSCALL_IMPL(lsm_list_modules, UnimplementedSyscallSafe);
   }
   if (Handler->IsHostKernelVersionAtLeast(6, 10, 0)) {
-    REGISTER_SYSCALL_IMPL_PASS_FLAGS(mseal, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                     SyscallPassthrough3<SYSCALL_DEF(mseal)>);
+    REGISTER_SYSCALL_IMPL(mseal, SyscallPassthrough3<SYSCALL_DEF(mseal)>);
   } else {
     REGISTER_SYSCALL_IMPL(mseal, UnimplementedSyscallSafe);
   }
@@ -578,157 +415,84 @@ namespace x64 {
   void RegisterPassthrough(FEX::HLE::SyscallHandler* Handler) {
     using namespace FEXCore::IR;
     RegisterCommon(Handler);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(ioctl, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough3<SYSCALL_DEF(ioctl)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(pread_64, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(pread_64)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(pwrite_64, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(pwrite_64)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(readv, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough3<SYSCALL_DEF(readv)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(writev, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough3<SYSCALL_DEF(writev)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(dup, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough1<SYSCALL_DEF(dup)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(nanosleep, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(nanosleep)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(getitimer, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(getitimer)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(setitimer, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough3<SYSCALL_DEF(setitimer)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(sendfile, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(sendfile)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(accept, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough3<SYSCALL_DEF(accept)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(sendmsg, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough3<SYSCALL_DEF(sendmsg)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(recvmsg, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough3<SYSCALL_DEF(recvmsg)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(setsockopt, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough5<SYSCALL_DEF(setsockopt)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(getsockopt, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough5<SYSCALL_DEF(getsockopt)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(wait4, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(wait4)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(semop, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough3<SYSCALL_DEF(semop)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(gettimeofday, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(gettimeofday)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(getrlimit, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(getrlimit)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(getrusage, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(getrusage)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(sysinfo, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough1<SYSCALL_DEF(sysinfo)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(times, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough1<SYSCALL_DEF(times)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(rt_sigqueueinfo, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough3<SYSCALL_DEF(rt_sigqueueinfo)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(fstatfs, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(fstatfs)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(sched_rr_get_interval, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(sched_rr_get_interval)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(mlockall, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough1<SYSCALL_DEF(mlockall)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(munlockall, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough0<SYSCALL_DEF(munlockall)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(adjtimex, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough1<SYSCALL_DEF(adjtimex)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(setrlimit, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(setrlimit)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(settimeofday, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(settimeofday)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(readahead, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough3<SYSCALL_DEF(readahead)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(futex, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough6<SYSCALL_DEF(futex)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(io_getevents, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough5<SYSCALL_DEF(io_getevents)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(semtimedop, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(semtimedop)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(timer_create, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough3<SYSCALL_DEF(timer_create)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(timer_settime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(timer_settime)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(timer_gettime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(timer_gettime)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(clock_settime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(clock_settime)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(clock_gettime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(clock_gettime)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(clock_getres, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(clock_getres)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(clock_nanosleep, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(clock_nanosleep)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(mq_open, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(mq_open)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(mq_timedsend, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough5<SYSCALL_DEF(mq_timedsend)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(mq_timedreceive, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough5<SYSCALL_DEF(mq_timedreceive)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(mq_notify, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(mq_notify)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(mq_getsetattr, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough3<SYSCALL_DEF(mq_getsetattr)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(waitid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough5<SYSCALL_DEF(waitid)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(pselect6, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough6<SYSCALL_DEF(pselect6)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(ppoll, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough5<SYSCALL_DEF(ppoll)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(set_robust_list, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(set_robust_list)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(get_robust_list, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough3<SYSCALL_DEF(get_robust_list)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(sync_file_range, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(sync_file_range)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(vmsplice, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(vmsplice)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(utimensat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(utimensat)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(fallocate, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(fallocate)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(timerfd_settime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(timerfd_settime)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(timerfd_gettime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(timerfd_gettime)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(preadv, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough5<SYSCALL_DEF(preadv)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(pwritev, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough5<SYSCALL_DEF(pwritev)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(rt_tgsigqueueinfo, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(rt_tgsigqueueinfo)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(recvmmsg, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough5<SYSCALL_DEF(recvmmsg)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(clock_adjtime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough2<SYSCALL_DEF(clock_adjtime)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(sendmmsg, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(sendmmsg)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(process_vm_readv, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough6<SYSCALL_DEF(process_vm_readv)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(process_vm_writev, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough6<SYSCALL_DEF(process_vm_writev)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(preadv2, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough6<SYSCALL_DEF(preadv2)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(pwritev2, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough6<SYSCALL_DEF(pwritev2)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(io_pgetevents, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough6<SYSCALL_DEF(io_pgetevents)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(pidfd_send_signal, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(pidfd_send_signal)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(process_madvise, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough5<SYSCALL_DEF(process_madvise)>);
-    REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(fadvise64, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                         SyscallPassthrough4<SYSCALL_DEF(fadvise64)>);
+    REGISTER_SYSCALL_IMPL_X64(ioctl, SyscallPassthrough3<SYSCALL_DEF(ioctl)>);
+    REGISTER_SYSCALL_IMPL_X64(pread_64, SyscallPassthrough4<SYSCALL_DEF(pread_64)>);
+    REGISTER_SYSCALL_IMPL_X64(pwrite_64, SyscallPassthrough4<SYSCALL_DEF(pwrite_64)>);
+    REGISTER_SYSCALL_IMPL_X64(readv, SyscallPassthrough3<SYSCALL_DEF(readv)>);
+    REGISTER_SYSCALL_IMPL_X64(writev, SyscallPassthrough3<SYSCALL_DEF(writev)>);
+    REGISTER_SYSCALL_IMPL_X64(dup, SyscallPassthrough1<SYSCALL_DEF(dup)>);
+    REGISTER_SYSCALL_IMPL_X64(nanosleep, SyscallPassthrough2<SYSCALL_DEF(nanosleep)>);
+    REGISTER_SYSCALL_IMPL_X64(getitimer, SyscallPassthrough2<SYSCALL_DEF(getitimer)>);
+    REGISTER_SYSCALL_IMPL_X64(setitimer, SyscallPassthrough3<SYSCALL_DEF(setitimer)>);
+    REGISTER_SYSCALL_IMPL_X64(sendfile, SyscallPassthrough4<SYSCALL_DEF(sendfile)>);
+    REGISTER_SYSCALL_IMPL_X64(accept, SyscallPassthrough3<SYSCALL_DEF(accept)>);
+    REGISTER_SYSCALL_IMPL_X64(sendmsg, SyscallPassthrough3<SYSCALL_DEF(sendmsg)>);
+    REGISTER_SYSCALL_IMPL_X64(recvmsg, SyscallPassthrough3<SYSCALL_DEF(recvmsg)>);
+    REGISTER_SYSCALL_IMPL_X64(setsockopt, SyscallPassthrough5<SYSCALL_DEF(setsockopt)>);
+    REGISTER_SYSCALL_IMPL_X64(getsockopt, SyscallPassthrough5<SYSCALL_DEF(getsockopt)>);
+    REGISTER_SYSCALL_IMPL_X64(wait4, SyscallPassthrough4<SYSCALL_DEF(wait4)>);
+    REGISTER_SYSCALL_IMPL_X64(semop, SyscallPassthrough3<SYSCALL_DEF(semop)>);
+    REGISTER_SYSCALL_IMPL_X64(gettimeofday, SyscallPassthrough2<SYSCALL_DEF(gettimeofday)>);
+    REGISTER_SYSCALL_IMPL_X64(getrlimit, SyscallPassthrough2<SYSCALL_DEF(getrlimit)>);
+    REGISTER_SYSCALL_IMPL_X64(getrusage, SyscallPassthrough2<SYSCALL_DEF(getrusage)>);
+    REGISTER_SYSCALL_IMPL_X64(sysinfo, SyscallPassthrough1<SYSCALL_DEF(sysinfo)>);
+    REGISTER_SYSCALL_IMPL_X64(times, SyscallPassthrough1<SYSCALL_DEF(times)>);
+    REGISTER_SYSCALL_IMPL_X64(rt_sigqueueinfo, SyscallPassthrough3<SYSCALL_DEF(rt_sigqueueinfo)>);
+    REGISTER_SYSCALL_IMPL_X64(fstatfs, SyscallPassthrough2<SYSCALL_DEF(fstatfs)>);
+    REGISTER_SYSCALL_IMPL_X64(sched_rr_get_interval, SyscallPassthrough2<SYSCALL_DEF(sched_rr_get_interval)>);
+    REGISTER_SYSCALL_IMPL_X64(mlockall, SyscallPassthrough1<SYSCALL_DEF(mlockall)>);
+    REGISTER_SYSCALL_IMPL_X64(munlockall, SyscallPassthrough0<SYSCALL_DEF(munlockall)>);
+    REGISTER_SYSCALL_IMPL_X64(adjtimex, SyscallPassthrough1<SYSCALL_DEF(adjtimex)>);
+    REGISTER_SYSCALL_IMPL_X64(setrlimit, SyscallPassthrough2<SYSCALL_DEF(setrlimit)>);
+    REGISTER_SYSCALL_IMPL_X64(settimeofday, SyscallPassthrough2<SYSCALL_DEF(settimeofday)>);
+    REGISTER_SYSCALL_IMPL_X64(readahead, SyscallPassthrough3<SYSCALL_DEF(readahead)>);
+    REGISTER_SYSCALL_IMPL_X64(futex, SyscallPassthrough6<SYSCALL_DEF(futex)>);
+    REGISTER_SYSCALL_IMPL_X64(io_getevents, SyscallPassthrough5<SYSCALL_DEF(io_getevents)>);
+    REGISTER_SYSCALL_IMPL_X64(semtimedop, SyscallPassthrough4<SYSCALL_DEF(semtimedop)>);
+    REGISTER_SYSCALL_IMPL_X64(timer_create, SyscallPassthrough3<SYSCALL_DEF(timer_create)>);
+    REGISTER_SYSCALL_IMPL_X64(timer_settime, SyscallPassthrough4<SYSCALL_DEF(timer_settime)>);
+    REGISTER_SYSCALL_IMPL_X64(timer_gettime, SyscallPassthrough2<SYSCALL_DEF(timer_gettime)>);
+    REGISTER_SYSCALL_IMPL_X64(clock_settime, SyscallPassthrough2<SYSCALL_DEF(clock_settime)>);
+    REGISTER_SYSCALL_IMPL_X64(clock_gettime, SyscallPassthrough2<SYSCALL_DEF(clock_gettime)>);
+    REGISTER_SYSCALL_IMPL_X64(clock_getres, SyscallPassthrough2<SYSCALL_DEF(clock_getres)>);
+    REGISTER_SYSCALL_IMPL_X64(clock_nanosleep, SyscallPassthrough4<SYSCALL_DEF(clock_nanosleep)>);
+    REGISTER_SYSCALL_IMPL_X64(mq_open, SyscallPassthrough4<SYSCALL_DEF(mq_open)>);
+    REGISTER_SYSCALL_IMPL_X64(mq_timedsend, SyscallPassthrough5<SYSCALL_DEF(mq_timedsend)>);
+    REGISTER_SYSCALL_IMPL_X64(mq_timedreceive, SyscallPassthrough5<SYSCALL_DEF(mq_timedreceive)>);
+    REGISTER_SYSCALL_IMPL_X64(mq_notify, SyscallPassthrough2<SYSCALL_DEF(mq_notify)>);
+    REGISTER_SYSCALL_IMPL_X64(mq_getsetattr, SyscallPassthrough3<SYSCALL_DEF(mq_getsetattr)>);
+    REGISTER_SYSCALL_IMPL_X64(waitid, SyscallPassthrough5<SYSCALL_DEF(waitid)>);
+    REGISTER_SYSCALL_IMPL_X64(pselect6, SyscallPassthrough6<SYSCALL_DEF(pselect6)>);
+    REGISTER_SYSCALL_IMPL_X64(ppoll, SyscallPassthrough5<SYSCALL_DEF(ppoll)>);
+    REGISTER_SYSCALL_IMPL_X64(set_robust_list, SyscallPassthrough2<SYSCALL_DEF(set_robust_list)>);
+    REGISTER_SYSCALL_IMPL_X64(get_robust_list, SyscallPassthrough3<SYSCALL_DEF(get_robust_list)>);
+    REGISTER_SYSCALL_IMPL_X64(sync_file_range, SyscallPassthrough4<SYSCALL_DEF(sync_file_range)>);
+    REGISTER_SYSCALL_IMPL_X64(vmsplice, SyscallPassthrough4<SYSCALL_DEF(vmsplice)>);
+    REGISTER_SYSCALL_IMPL_X64(utimensat, SyscallPassthrough4<SYSCALL_DEF(utimensat)>);
+    REGISTER_SYSCALL_IMPL_X64(fallocate, SyscallPassthrough4<SYSCALL_DEF(fallocate)>);
+    REGISTER_SYSCALL_IMPL_X64(timerfd_settime, SyscallPassthrough4<SYSCALL_DEF(timerfd_settime)>);
+    REGISTER_SYSCALL_IMPL_X64(timerfd_gettime, SyscallPassthrough2<SYSCALL_DEF(timerfd_gettime)>);
+    REGISTER_SYSCALL_IMPL_X64(preadv, SyscallPassthrough5<SYSCALL_DEF(preadv)>);
+    REGISTER_SYSCALL_IMPL_X64(pwritev, SyscallPassthrough5<SYSCALL_DEF(pwritev)>);
+    REGISTER_SYSCALL_IMPL_X64(rt_tgsigqueueinfo, SyscallPassthrough4<SYSCALL_DEF(rt_tgsigqueueinfo)>);
+    REGISTER_SYSCALL_IMPL_X64(recvmmsg, SyscallPassthrough5<SYSCALL_DEF(recvmmsg)>);
+    REGISTER_SYSCALL_IMPL_X64(clock_adjtime, SyscallPassthrough2<SYSCALL_DEF(clock_adjtime)>);
+    REGISTER_SYSCALL_IMPL_X64(sendmmsg, SyscallPassthrough4<SYSCALL_DEF(sendmmsg)>);
+    REGISTER_SYSCALL_IMPL_X64(process_vm_readv, SyscallPassthrough6<SYSCALL_DEF(process_vm_readv)>);
+    REGISTER_SYSCALL_IMPL_X64(process_vm_writev, SyscallPassthrough6<SYSCALL_DEF(process_vm_writev)>);
+    REGISTER_SYSCALL_IMPL_X64(preadv2, SyscallPassthrough6<SYSCALL_DEF(preadv2)>);
+    REGISTER_SYSCALL_IMPL_X64(pwritev2, SyscallPassthrough6<SYSCALL_DEF(pwritev2)>);
+    REGISTER_SYSCALL_IMPL_X64(io_pgetevents, SyscallPassthrough6<SYSCALL_DEF(io_pgetevents)>);
+    REGISTER_SYSCALL_IMPL_X64(pidfd_send_signal, SyscallPassthrough4<SYSCALL_DEF(pidfd_send_signal)>);
+    REGISTER_SYSCALL_IMPL_X64(process_madvise, SyscallPassthrough5<SYSCALL_DEF(process_madvise)>);
+    REGISTER_SYSCALL_IMPL_X64(fadvise64, SyscallPassthrough4<SYSCALL_DEF(fadvise64)>);
     if (Handler->IsHostKernelVersionAtLeast(6, 5, 0)) {
-      REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(cachestat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                           SyscallPassthrough4<SYSCALL_DEF(cachestat)>);
+      REGISTER_SYSCALL_IMPL_X64(cachestat, SyscallPassthrough4<SYSCALL_DEF(cachestat)>);
     } else {
       REGISTER_SYSCALL_IMPL_X64(cachestat, UnimplementedSyscallSafe);
     }
     if (Handler->IsHostKernelVersionAtLeast(6, 6, 0)) {
-      REGISTER_SYSCALL_IMPL_X64_PASS_FLAGS(fchmodat2, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                           SyscallPassthrough4<SYSCALL_DEF(fchmodat2)>);
+      REGISTER_SYSCALL_IMPL_X64(fchmodat2, SyscallPassthrough4<SYSCALL_DEF(fchmodat2)>);
     } else {
       REGISTER_SYSCALL_IMPL_X64(fchmodat2, UnimplementedSyscallSafe);
     }
@@ -739,79 +503,41 @@ namespace x32 {
   void RegisterPassthrough(FEX::HLE::SyscallHandler* Handler) {
     using namespace FEXCore::IR;
     RegisterCommon(Handler);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(getuid32, getuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough0<SYSCALL_DEF(getuid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(getgid32, getgid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough0<SYSCALL_DEF(getgid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(geteuid32, geteuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough0<SYSCALL_DEF(geteuid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(getegid32, getegid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough0<SYSCALL_DEF(getegid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(setreuid32, setreuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough2<SYSCALL_DEF(setreuid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(setregid32, setregid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough2<SYSCALL_DEF(setregid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(getgroups32, getgroups, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough2<SYSCALL_DEF(getgroups)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(setgroups32, setgroups, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough2<SYSCALL_DEF(setgroups)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(fchown32, fchown, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough3<SYSCALL_DEF(fchown)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(setresuid32, setresuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough3<SYSCALL_DEF(setresuid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(getresuid32, getresuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough3<SYSCALL_DEF(getresuid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(setresgid32, setresgid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough3<SYSCALL_DEF(setresgid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(getresgid32, getresgid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough3<SYSCALL_DEF(getresgid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(setuid32, setuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough1<SYSCALL_DEF(setuid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(setgid32, setgid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough1<SYSCALL_DEF(setgid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(setfsuid32, setfsuid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough1<SYSCALL_DEF(setfsuid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(setfsgid32, setfsgid, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough1<SYSCALL_DEF(setfsgid)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(sendfile64, sendfile, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough4<SYSCALL_DEF(sendfile)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(clock_gettime64, clock_gettime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough2<SYSCALL_DEF(clock_gettime)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(clock_settime64, clock_settime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough2<SYSCALL_DEF(clock_settime)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(clock_adjtime64, clock_adjtime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough2<SYSCALL_DEF(clock_adjtime)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(clock_getres_time64, clock_getres, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough2<SYSCALL_DEF(clock_getres)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(clock_nanosleep_time64, clock_nanosleep,
-                                                SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough4<SYSCALL_DEF(clock_nanosleep)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(timer_gettime64, timer_gettime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough2<SYSCALL_DEF(timer_gettime)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(timer_settime64, timer_settime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough4<SYSCALL_DEF(timer_settime)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(timerfd_gettime64, timerfd_gettime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough2<SYSCALL_DEF(timerfd_gettime)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(timerfd_settime64, timerfd_settime, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough4<SYSCALL_DEF(timerfd_settime)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(utimensat_time64, utimensat, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough4<SYSCALL_DEF(utimensat)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(ppoll_time64, ppoll, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough5<SYSCALL_DEF(ppoll)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(io_pgetevents_time64, io_pgetevents, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough6<SYSCALL_DEF(io_pgetevents)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(mq_timedsend_time64, mq_timedsend, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough5<SYSCALL_DEF(mq_timedsend)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(mq_timedreceive_time64, mq_timedreceive,
-                                                SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough5<SYSCALL_DEF(mq_timedreceive)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(semtimedop_time64, semtimedop, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough4<SYSCALL_DEF(semtimedop)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(futex_time64, futex, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough6<SYSCALL_DEF(futex)>);
-    REGISTER_SYSCALL_IMPL_X32_PASS_MANUAL_FLAGS(sched_rr_get_interval_time64, sched_rr_get_interval,
-                                                SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                                                SyscallPassthrough2<SYSCALL_DEF(sched_rr_get_interval)>);
+    REGISTER_SYSCALL_IMPL_X32(getuid32, SyscallPassthrough0<SYSCALL_DEF(getuid)>);
+    REGISTER_SYSCALL_IMPL_X32(getgid32, SyscallPassthrough0<SYSCALL_DEF(getgid)>);
+    REGISTER_SYSCALL_IMPL_X32(geteuid32, SyscallPassthrough0<SYSCALL_DEF(geteuid)>);
+    REGISTER_SYSCALL_IMPL_X32(getegid32, SyscallPassthrough0<SYSCALL_DEF(getegid)>);
+    REGISTER_SYSCALL_IMPL_X32(setreuid32, SyscallPassthrough2<SYSCALL_DEF(setreuid)>);
+    REGISTER_SYSCALL_IMPL_X32(setregid32, SyscallPassthrough2<SYSCALL_DEF(setregid)>);
+    REGISTER_SYSCALL_IMPL_X32(getgroups32, SyscallPassthrough2<SYSCALL_DEF(getgroups)>);
+    REGISTER_SYSCALL_IMPL_X32(setgroups32, SyscallPassthrough2<SYSCALL_DEF(setgroups)>);
+    REGISTER_SYSCALL_IMPL_X32(fchown32, SyscallPassthrough3<SYSCALL_DEF(fchown)>);
+    REGISTER_SYSCALL_IMPL_X32(setresuid32, SyscallPassthrough3<SYSCALL_DEF(setresuid)>);
+    REGISTER_SYSCALL_IMPL_X32(getresuid32, SyscallPassthrough3<SYSCALL_DEF(getresuid)>);
+    REGISTER_SYSCALL_IMPL_X32(setresgid32, SyscallPassthrough3<SYSCALL_DEF(setresgid)>);
+    REGISTER_SYSCALL_IMPL_X32(getresgid32, SyscallPassthrough3<SYSCALL_DEF(getresgid)>);
+    REGISTER_SYSCALL_IMPL_X32(setuid32, SyscallPassthrough1<SYSCALL_DEF(setuid)>);
+    REGISTER_SYSCALL_IMPL_X32(setgid32, SyscallPassthrough1<SYSCALL_DEF(setgid)>);
+    REGISTER_SYSCALL_IMPL_X32(setfsuid32, SyscallPassthrough1<SYSCALL_DEF(setfsuid)>);
+    REGISTER_SYSCALL_IMPL_X32(setfsgid32, SyscallPassthrough1<SYSCALL_DEF(setfsgid)>);
+    REGISTER_SYSCALL_IMPL_X32(sendfile64, SyscallPassthrough4<SYSCALL_DEF(sendfile)>);
+    REGISTER_SYSCALL_IMPL_X32(clock_gettime64, SyscallPassthrough2<SYSCALL_DEF(clock_gettime)>);
+    REGISTER_SYSCALL_IMPL_X32(clock_settime64, SyscallPassthrough2<SYSCALL_DEF(clock_settime)>);
+    REGISTER_SYSCALL_IMPL_X32(clock_adjtime64, SyscallPassthrough2<SYSCALL_DEF(clock_adjtime)>);
+    REGISTER_SYSCALL_IMPL_X32(clock_getres_time64, SyscallPassthrough2<SYSCALL_DEF(clock_getres)>);
+    REGISTER_SYSCALL_IMPL_X32(clock_nanosleep_time64, SyscallPassthrough4<SYSCALL_DEF(clock_nanosleep)>);
+    REGISTER_SYSCALL_IMPL_X32(timer_gettime64, SyscallPassthrough2<SYSCALL_DEF(timer_gettime)>);
+    REGISTER_SYSCALL_IMPL_X32(timer_settime64, SyscallPassthrough4<SYSCALL_DEF(timer_settime)>);
+    REGISTER_SYSCALL_IMPL_X32(timerfd_gettime64, SyscallPassthrough2<SYSCALL_DEF(timerfd_gettime)>);
+    REGISTER_SYSCALL_IMPL_X32(timerfd_settime64, SyscallPassthrough4<SYSCALL_DEF(timerfd_settime)>);
+    REGISTER_SYSCALL_IMPL_X32(utimensat_time64, SyscallPassthrough4<SYSCALL_DEF(utimensat)>);
+    REGISTER_SYSCALL_IMPL_X32(ppoll_time64, SyscallPassthrough5<SYSCALL_DEF(ppoll)>);
+    REGISTER_SYSCALL_IMPL_X32(io_pgetevents_time64, SyscallPassthrough6<SYSCALL_DEF(io_pgetevents)>);
+    REGISTER_SYSCALL_IMPL_X32(mq_timedsend_time64, SyscallPassthrough5<SYSCALL_DEF(mq_timedsend)>);
+    REGISTER_SYSCALL_IMPL_X32(mq_timedreceive_time64, SyscallPassthrough5<SYSCALL_DEF(mq_timedreceive)>);
+    REGISTER_SYSCALL_IMPL_X32(semtimedop_time64, SyscallPassthrough4<SYSCALL_DEF(semtimedop)>);
+    REGISTER_SYSCALL_IMPL_X32(futex_time64, SyscallPassthrough6<SYSCALL_DEF(futex)>);
+    REGISTER_SYSCALL_IMPL_X32(sched_rr_get_interval_time64, SyscallPassthrough2<SYSCALL_DEF(sched_rr_get_interval)>);
   }
 } // namespace x32
 } // namespace FEX::HLE

Source/Tools/LinuxEmulation/LinuxSyscalls/Syscalls/Thread.cpp

@@ -398,127 +398,124 @@ void RegisterThread(FEX::HLE::SyscallHandler* Handler) {
     FEX_UNREACHABLE;
   });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(fork, SyscallFlags::DEFAULT, ([](FEXCore::Core::CpuStateFrame* Frame) -> uint64_t {
-                                FEX::HLE::clone3_args args {.Type = TypeOfClone::TYPE_CLONE2,
-                                                            .args = {
-                                                              .flags = 0,
-                                                              .pidfd = 0,
-                                                              .child_tid = 0,
-                                                              .parent_tid = 0,
-                                                              .exit_signal = SIGCHLD,
-                                                              .stack = 0,
-                                                              .stack_size = 0,
-                                                              .tls = 0,
-                                                              .set_tid = 0,
-                                                              .set_tid_size = 0,
-                                                              .cgroup = 0,
-                                                            }};
+  REGISTER_SYSCALL_IMPL(fork, ([](FEXCore::Core::CpuStateFrame* Frame) -> uint64_t {
+                          FEX::HLE::clone3_args args {.Type = TypeOfClone::TYPE_CLONE2,
+                                                      .args = {
+                                                        .flags = 0,
+                                                        .pidfd = 0,
+                                                        .child_tid = 0,
+                                                        .parent_tid = 0,
+                                                        .exit_signal = SIGCHLD,
+                                                        .stack = 0,
+                                                        .stack_size = 0,
+                                                        .tls = 0,
+                                                        .set_tid = 0,
+                                                        .set_tid_size = 0,
+                                                        .cgroup = 0,
+                                                      }};
 
-                                return ForkGuest(Frame->Thread, Frame, &args);
-                              }));
+                          return ForkGuest(Frame->Thread, Frame, &args);
+                        }));
 
-  REGISTER_SYSCALL_IMPL_FLAGS(vfork, SyscallFlags::DEFAULT, ([](FEXCore::Core::CpuStateFrame* Frame) -> uint64_t {
-                                FEX::HLE::clone3_args args {.Type = TypeOfClone::TYPE_CLONE2,
-                                                            .args = {
-                                                              .flags = CLONE_VFORK,
-                                                              .pidfd = 0,
-                                                              .child_tid = 0,
-                                                              .parent_tid = 0,
-                                                              .exit_signal = SIGCHLD,
-                                                              .stack = 0,
-                                                              .stack_size = 0,
-                                                              .tls = 0,
-                                                              .set_tid = 0,
-                                                              .set_tid_size = 0,
-                                                              .cgroup = 0,
-                                                            }};
+  REGISTER_SYSCALL_IMPL(vfork, ([](FEXCore::Core::CpuStateFrame* Frame) -> uint64_t {
+                          FEX::HLE::clone3_args args {.Type = TypeOfClone::TYPE_CLONE2,
+                                                      .args = {
+                                                        .flags = CLONE_VFORK,
+                                                        .pidfd = 0,
+                                                        .child_tid = 0,
+                                                        .parent_tid = 0,
+                                                        .exit_signal = SIGCHLD,
+                                                        .stack = 0,
+                                                        .stack_size = 0,
+                                                        .tls = 0,
+                                                        .set_tid = 0,
+                                                        .set_tid_size = 0,
+                                                        .cgroup = 0,
+                                                      }};
 
-                                return ForkGuest(Frame->Thread, Frame, &args);
-                              }));
+                          return ForkGuest(Frame->Thread, Frame, &args);
+                        }));
 
-  REGISTER_SYSCALL_IMPL_FLAGS(getpgrp, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame) -> uint64_t {
-                                uint64_t Result = ::getpgrp();
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(getpgrp, [](FEXCore::Core::CpuStateFrame* Frame) -> uint64_t {
+    uint64_t Result = ::getpgrp();
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(clone3, SyscallFlags::DEFAULT,
-                              ([](FEXCore::Core::CpuStateFrame* Frame, FEX::HLE::kernel_clone3_args* cl_args, size_t size) -> uint64_t {
-                                FEX::HLE::clone3_args args {};
-                                args.Type = TypeOfClone::TYPE_CLONE3;
-                                memcpy(&args.args, cl_args, std::min(sizeof(FEX::HLE::kernel_clone3_args), size));
-                                return CloneHandler(Frame, &args);
-                              }));
+  REGISTER_SYSCALL_IMPL(clone3, ([](FEXCore::Core::CpuStateFrame* Frame, FEX::HLE::kernel_clone3_args* cl_args, size_t size) -> uint64_t {
+                          FEX::HLE::clone3_args args {};
+                          args.Type = TypeOfClone::TYPE_CLONE3;
+                          memcpy(&args.args, cl_args, std::min(sizeof(FEX::HLE::kernel_clone3_args), size));
+                          return CloneHandler(Frame, &args);
+                        }));
 
-  REGISTER_SYSCALL_IMPL_FLAGS(exit, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY | SyscallFlags::NORETURN,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int status) -> uint64_t {
-                                // TLS/DTV teardown is something FEX can't control. Disable glibc checking when we leave a pthread.
-                                // Since this thread is hard stopping, we can't track the TLS/DTV teardown in FEX's thread handling.
-                                FEXCore::Allocator::YesIKnowImNotSupposedToUseTheGlibcAllocator::HardDisable();
-                                auto ThreadObject = FEX::HLE::ThreadManager::GetStateObjectFromCPUState(Frame);
+  REGISTER_SYSCALL_IMPL(exit, [](FEXCore::Core::CpuStateFrame* Frame, int status) -> uint64_t {
+    // TLS/DTV teardown is something FEX can't control. Disable glibc checking when we leave a pthread.
+    // Since this thread is hard stopping, we can't track the TLS/DTV teardown in FEX's thread handling.
+    FEXCore::Allocator::YesIKnowImNotSupposedToUseTheGlibcAllocator::HardDisable();
+    auto ThreadObject = FEX::HLE::ThreadManager::GetStateObjectFromCPUState(Frame);
 
-                                if (ThreadObject->ThreadInfo.clear_child_tid) {
-                                  std::atomic<uint32_t>* Addr = reinterpret_cast<std::atomic<uint32_t>*>(ThreadObject->ThreadInfo.clear_child_tid);
-                                  Addr->store(0);
-                                  syscall(SYSCALL_DEF(futex), ThreadObject->ThreadInfo.clear_child_tid, FUTEX_WAKE, ~0ULL, 0, 0, 0);
-                                }
+    if (ThreadObject->ThreadInfo.clear_child_tid) {
+      auto Addr = std::atomic_ref<int32_t>(*ThreadObject->ThreadInfo.clear_child_tid);
+      Addr.store(0);
+      syscall(SYSCALL_DEF(futex), ThreadObject->ThreadInfo.clear_child_tid, FUTEX_WAKE, ~0ULL, 0, 0, 0);
+    }
 
-                                ThreadObject->StatusCode = status;
+    ThreadObject->StatusCode = status;
 
-                                FEX::HLE::_SyscallHandler->UninstallTLSState(ThreadObject);
+    FEX::HLE::_SyscallHandler->UninstallTLSState(ThreadObject);
 
-                                if (ThreadObject->ExecutionThread) {
-                                  // If this is a pthread based execution thread, then there is more work to be done.
-                                  // Delegate final deletion and cleanup to the pthreads Thread management.
-                                  FEX::LinuxEmulation::Threads::LongjumpDeallocateAndExit(ThreadObject, status);
-                                } else {
-                                  FEX::HLE::_SyscallHandler->TM.DestroyThread(ThreadObject, true);
-                                  FEX::LinuxEmulation::Threads::DeallocateStackObjectAndExit(nullptr, status);
-                                }
-                                // This will never be reached
-                                std::terminate();
-                              });
+    if (ThreadObject->ExecutionThread) {
+      // If this is a pthread based execution thread, then there is more work to be done.
+      // Delegate final deletion and cleanup to the pthreads Thread management.
+      FEX::LinuxEmulation::Threads::LongjumpDeallocateAndExit(ThreadObject, status);
+    } else {
+      FEX::HLE::_SyscallHandler->TM.DestroyThread(ThreadObject, true);
+      FEX::LinuxEmulation::Threads::DeallocateStackObjectAndExit(nullptr, status);
+    }
+    // This will never be reached
+    std::terminate();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(prctl, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int option, unsigned long arg2, unsigned long arg3,
-                                 unsigned long arg4, unsigned long arg5) -> uint64_t {
-                                uint64_t Result {};
+  REGISTER_SYSCALL_IMPL(prctl,
+                        [](FEXCore::Core::CpuStateFrame* Frame, int option, unsigned long arg2, unsigned long arg3, unsigned long arg4,
+                           unsigned long arg5) -> uint64_t {
+                          uint64_t Result {};
 #ifndef PR_GET_AUXV
 #define PR_GET_AUXV 0x41555856
 #endif
-                                switch (option) {
-                                case PR_SET_SECCOMP: {
-                                  uint32_t Operation {};
-                                  if (arg2 == SECCOMP_MODE_STRICT) Operation = SECCOMP_SET_MODE_STRICT;
-                                  if (arg2 == SECCOMP_MODE_FILTER) Operation = SECCOMP_SET_MODE_FILTER;
+                          switch (option) {
+                          case PR_SET_SECCOMP: {
+                            uint32_t Operation {};
+                            if (arg2 == SECCOMP_MODE_STRICT) Operation = SECCOMP_SET_MODE_STRICT;
+                            if (arg2 == SECCOMP_MODE_FILTER) Operation = SECCOMP_SET_MODE_FILTER;
 
-                                  return FEX::HLE::_SyscallHandler->SeccompEmulator.Handle(Frame, Operation, 0, reinterpret_cast<void*>(arg3));
-                                }
-                                case PR_GET_SECCOMP: return FEX::HLE::_SyscallHandler->SeccompEmulator.GetSeccomp(Frame);
-                                case PR_GET_AUXV: {
-                                  if (arg4 || arg5) {
-                                    return -EINVAL;
-                                  }
+                            return FEX::HLE::_SyscallHandler->SeccompEmulator.Handle(Frame, Operation, 0, reinterpret_cast<void*>(arg3));
+                          }
+                          case PR_GET_SECCOMP: return FEX::HLE::_SyscallHandler->SeccompEmulator.GetSeccomp(Frame);
+                          case PR_GET_AUXV: {
+                            if (arg4 || arg5) {
+                              return -EINVAL;
+                            }
 
-                                  void* addr = reinterpret_cast<void*>(arg2);
-                                  size_t UserSize = reinterpret_cast<size_t>(arg3);
+                            void* addr = reinterpret_cast<void*>(arg2);
+                            size_t UserSize = reinterpret_cast<size_t>(arg3);
 
-                                  const auto auxv = FEX::HLE::_SyscallHandler->GetCodeLoader()->GetAuxv();
-                                  const auto auxvBase = auxv.address;
-                                  const auto auxvSize = auxv.size;
-                                  size_t MinSize = std::min(auxvSize, UserSize);
+                            const auto auxv = FEX::HLE::_SyscallHandler->GetCodeLoader()->GetAuxv();
+                            const auto auxvBase = auxv.address;
+                            const auto auxvSize = auxv.size;
+                            size_t MinSize = std::min(auxvSize, UserSize);
 
-                                  memcpy(addr, reinterpret_cast<void*>(auxvBase), MinSize);
+                            memcpy(addr, reinterpret_cast<void*>(auxvBase), MinSize);
 
-                                  // Returns the size of auxv without truncation.
-                                  return auxvSize;
-                                }
-                                default: Result = ::prctl(option, arg2, arg3, arg4, arg5); break;
-                                }
-                                SYSCALL_ERRNO();
-                              });
+                            // Returns the size of auxv without truncation.
+                            return auxvSize;
+                          }
+                          default: Result = ::prctl(option, arg2, arg3, arg4, arg5); break;
+                          }
+                          SYSCALL_ERRNO();
+                        });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(arch_prctl, SyscallFlags::DEFAULT, [](FEXCore::Core::CpuStateFrame* Frame, int code, unsigned long addr) -> uint64_t {
+  REGISTER_SYSCALL_IMPL(arch_prctl, [](FEXCore::Core::CpuStateFrame* Frame, int code, unsigned long addr) -> uint64_t {
     uint64_t Result {};
     switch (code) {
     case 0x1001: // ARCH_SET_GS
@@ -562,22 +559,20 @@ void RegisterThread(FEX::HLE::SyscallHandler* Handler) {
     SYSCALL_ERRNO();
   });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(set_tid_address, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int* tidptr) -> uint64_t {
-                                auto ThreadObject = FEX::HLE::ThreadManager::GetStateObjectFromCPUState(Frame);
-                                ThreadObject->ThreadInfo.clear_child_tid = tidptr;
-                                return ThreadObject->ThreadInfo.TID;
-                              });
+  REGISTER_SYSCALL_IMPL(set_tid_address, [](FEXCore::Core::CpuStateFrame* Frame, int* tidptr) -> uint64_t {
+    auto ThreadObject = FEX::HLE::ThreadManager::GetStateObjectFromCPUState(Frame);
+    ThreadObject->ThreadInfo.clear_child_tid = tidptr;
+    return ThreadObject->ThreadInfo.TID;
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(exit_group, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY | SyscallFlags::NORETURN,
-                              [](FEXCore::Core::CpuStateFrame* Frame, int status) -> uint64_t {
-                                // Save telemetry if we're exiting.
-                                FEX::HLE::_SyscallHandler->GetSignalDelegator()->SaveTelemetry();
-                                FEX::HLE::_SyscallHandler->TM.CleanupForExit();
+  REGISTER_SYSCALL_IMPL(exit_group, [](FEXCore::Core::CpuStateFrame* Frame, int status) -> uint64_t {
+    // Save telemetry if we're exiting.
+    FEX::HLE::_SyscallHandler->GetSignalDelegator()->SaveTelemetry();
+    FEX::HLE::_SyscallHandler->TM.CleanupForExit();
 
-                                syscall(SYSCALL_DEF(exit_group), status);
-                                // This will never be reached
-                                std::terminate();
-                              });
+    syscall(SYSCALL_DEF(exit_group), status);
+    // This will never be reached
+    std::terminate();
+  });
 }
 } // namespace FEX::HLE

Source/Tools/LinuxEmulation/LinuxSyscalls/Syscalls/Timer.cpp

@@ -24,16 +24,14 @@ namespace FEX::HLE {
 void RegisterTimer(FEX::HLE::SyscallHandler* Handler) {
   using namespace FEXCore::IR;
 
-  REGISTER_SYSCALL_IMPL_FLAGS(alarm, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame, unsigned int seconds) -> uint64_t {
-                                uint64_t Result = ::alarm(seconds);
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(alarm, [](FEXCore::Core::CpuStateFrame* Frame, unsigned int seconds) -> uint64_t {
+    uint64_t Result = ::alarm(seconds);
+    SYSCALL_ERRNO();
+  });
 
-  REGISTER_SYSCALL_IMPL_FLAGS(pause, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
-                              [](FEXCore::Core::CpuStateFrame* Frame) -> uint64_t {
-                                uint64_t Result = ::pause();
-                                SYSCALL_ERRNO();
-                              });
+  REGISTER_SYSCALL_IMPL(pause, [](FEXCore::Core::CpuStateFrame* Frame) -> uint64_t {
+    uint64_t Result = ::pause();
+    SYSCALL_ERRNO();
+  });
 }
 } // namespace FEX::HLE

Source/Tools/LinuxEmulation/LinuxSyscalls/SyscallsSMCTracking.cpp

@@ -8,6 +8,7 @@ $end_info$
 */
 
 #include "Common/FDUtils.h"
+#include "Common/FileMappingBaseAddress.h"
 
 #include <filesystem>
 #include <sys/shm.h>
@@ -23,6 +24,7 @@ $end_info$
 #include <FEXCore/Utils/SignalScopeGuards.h>
 #include <FEXCore/Utils/TypeDefines.h>
 #include <FEXHeaderUtils/Filesystem.h>
+#include <Linux/Utils/ELFParser.h>
 
 namespace FEX::HLE {
 // SMC interactions
@@ -165,22 +167,20 @@ void SyscallHandler::MarkGuestExecutableRange(FEXCore::Core::InternalThreadState
 }
 
 void SyscallHandler::InvalidateGuestCodeRange(FEXCore::Core::InternalThreadState* Thread, uint64_t Start, uint64_t Length) {
-  FEX::HLE::_SyscallHandler->InvalidateCodeRangeIfNecessary(Thread, Start, Length);
+  InvalidateCodeRangeIfNecessary(Thread, Start, Length);
 }
 
 std::optional<FEXCore::ExecutableFileSectionInfo>
 SyscallHandler::LookupExecutableFileSection(FEXCore::Core::InternalThreadState& Thread, uint64_t GuestAddr) {
   auto lk = FEXCore::GuardSignalDeferringSection<std::shared_lock>(VMATracking.Mutex, &Thread);
 
-  // Get the first mapping after GuestAddr, or end
-  // GuestAddr is inclusive
-  // If the write spans two pages, they will be flushed one at a time (generating two faults)
-  auto Entry = VMATracking.FindVMAEntry(GuestAddr);
-  if (Entry == VMATracking.VMAs.end() || !Entry->second.Resource) {
+  auto EntryIt = VMATracking.FindVMAEntry(GuestAddr);
+  if (EntryIt == VMATracking.VMAs.end() || !EntryIt->second.Resource) {
     return std::nullopt;
   }
 
-  return FEXCore::ExecutableFileSectionInfo {*Entry->second.Resource->MappedFile, Entry->second.Base - Entry->second.Offset};
+  auto& [MappingBaseAddr, Entry] = *EntryIt;
+  return FEXCore::ExecutableFileSectionInfo {*Entry.Resource->MappedFile, Entry.Resource->FirstVMA->Base};
 }
 
 FEXCore::HLE::ExecutableRangeInfo SyscallHandler::QueryGuestExecutableRange(FEXCore::Core::InternalThreadState* Thread, uint64_t Address) {
@@ -195,6 +195,22 @@ FEXCore::HLE::ExecutableRangeInfo SyscallHandler::QueryGuestExecutableRange(FEXC
   return {Entry->first, Entry->second.Length, Entry->second.Prot.Writable};
 }
 
+static fextl::vector<Elf64_Phdr> ReadELFHeaders(int FD, std::span<std::byte> HeaderData = {}) {
+  std::string_view ELFMagic = ELFMAG;
+  if (HeaderData.data()) {
+    if (HeaderData.size_bytes() < ELFMagic.size() || std::memcmp(ELFMagic.data(), HeaderData.data(), ELFMagic.size()) != 0) {
+      // Not an ELF file
+      return {};
+    }
+  } else {
+    // Read from FD in case the caller didn't have a mapped header available
+  }
+
+  ELFParser Parser;
+  Parser.ReadElf(dup(FD));
+  return std::move(Parser.phdrs);
+}
+
 void* SyscallHandler::GuestMmap(bool Is64Bit, FEXCore::Core::InternalThreadState* Thread, void* addr, size_t length, int prot, int flags,
                                 int fd, off_t offset) {
   LOGMAN_THROW_A_FMT(Is64Bit || (length >> 32) == 0, "values must fit to 32 bits");
@@ -223,10 +239,10 @@ void* SyscallHandler::GuestMmap(bool Is64Bit, FEXCore::Core::InternalThreadState
       }
     }
 
-    LateMetadata = FEX::HLE::_SyscallHandler->TrackMmap(Thread, Result, length, prot, flags, fd, offset);
+    LateMetadata = TrackMmap(Thread, Result, length, prot, flags, fd, offset);
   }
 
-  FEX::HLE::_SyscallHandler->InvalidateCodeRangeIfNecessary(Thread, Result, Size);
+  InvalidateCodeRangeIfNecessary(Thread, Result, Size);
 
   if (LateMetadata) {
     auto CodeInvalidationlk = GuardSignalDeferringSectionWithFallback(CTX->GetCodeInvalidationMutex(), Thread);
@@ -250,7 +266,7 @@ uint64_t SyscallHandler::GuestMunmap(bool Is64Bit, FEXCore::Core::InternalThread
     auto lk = FEXCore::GuardSignalDeferringSectionWithFallback(VMATracking.Mutex, Thread);
 
     if (reinterpret_cast<uintptr_t>(addr) < 0x1'0000'0000ULL) {
-      Result = FEX::HLE::_SyscallHandler->Get32BitAllocator()->Munmap(addr, length);
+      Result = Get32BitAllocator()->Munmap(addr, length);
       if (FEX::HLE::HasSyscallError(Result)) {
         return Result;
       }
@@ -260,9 +276,9 @@ uint64_t SyscallHandler::GuestMunmap(bool Is64Bit, FEXCore::Core::InternalThread
         return -errno;
       }
     }
-    FEX::HLE::_SyscallHandler->TrackMunmap(Thread, addr, length);
+    TrackMunmap(Thread, addr, length);
   }
-  FEX::HLE::_SyscallHandler->InvalidateCodeRangeIfNecessary(Thread, reinterpret_cast<uint64_t>(addr), Size);
+  InvalidateCodeRangeIfNecessary(Thread, reinterpret_cast<uint64_t>(addr), Size);
 
   if (length) {
     auto CodeInvalidationlk = GuardSignalDeferringSectionWithFallback(CTX->GetCodeInvalidationMutex(), Thread);
@@ -277,23 +293,22 @@ uint64_t SyscallHandler::GuestMremap(bool Is64Bit, FEXCore::Core::InternalThread
   uint64_t Result {};
 
   {
-    auto lk = FEXCore::GuardSignalDeferringSection(FEX::HLE::_SyscallHandler->VMATracking.Mutex, Thread);
+    auto lk = FEXCore::GuardSignalDeferringSection(VMATracking.Mutex, Thread);
     if (Is64Bit) {
       Result = reinterpret_cast<uint64_t>(::mremap(old_address, old_size, new_size, flags, new_address));
       if (Result == -1) {
         return -errno;
       }
     } else {
-      Result =
-        reinterpret_cast<uint64_t>(FEX::HLE::_SyscallHandler->Get32BitAllocator()->Mremap(old_address, old_size, new_size, flags, new_address));
+      Result = reinterpret_cast<uint64_t>(Get32BitAllocator()->Mremap(old_address, old_size, new_size, flags, new_address));
       if (FEX::HLE::HasSyscallError(Result)) {
         return Result;
       }
     }
-    FEX::HLE::_SyscallHandler->TrackMremap(Thread, reinterpret_cast<uint64_t>(old_address), old_size, new_size, flags, Result);
+    TrackMremap(Thread, reinterpret_cast<uint64_t>(old_address), old_size, new_size, flags, Result);
   }
 
-  FEX::HLE::_SyscallHandler->InvalidateCodeRangeIfNecessaryOnRemap(Thread, reinterpret_cast<uint64_t>(old_address), Result, old_size, new_size);
+  InvalidateCodeRangeIfNecessaryOnRemap(Thread, reinterpret_cast<uint64_t>(old_address), Result, old_size, new_size);
   return Result;
 }
 
@@ -301,17 +316,16 @@ uint64_t SyscallHandler::GuestMprotect(FEXCore::Core::InternalThreadState* Threa
   uint64_t Result {};
 
   {
-    auto lk = FEXCore::GuardSignalDeferringSection(FEX::HLE::_SyscallHandler->VMATracking.Mutex, Thread);
+    auto lk = FEXCore::GuardSignalDeferringSection(VMATracking.Mutex, Thread);
     Result = ::mprotect(addr, len, prot);
     if (Result == -1) {
       return -errno;
     }
 
-    FEX::HLE::_SyscallHandler->TrackMprotect(Thread, addr, len, prot);
+    TrackMprotect(Thread, addr, len, prot);
   }
 
-
-  FEX::HLE::_SyscallHandler->InvalidateCodeRangeIfNecessary(Thread, reinterpret_cast<uint64_t>(addr), len);
+  InvalidateCodeRangeIfNecessary(Thread, reinterpret_cast<uint64_t>(addr), len);
   return Result;
 }
 
@@ -320,7 +334,7 @@ uint64_t SyscallHandler::GuestShmat(bool Is64Bit, FEXCore::Core::InternalThreadS
   uint64_t Length {};
 
   {
-    auto lk = FEXCore::GuardSignalDeferringSection(FEX::HLE::_SyscallHandler->VMATracking.Mutex, Thread);
+    auto lk = FEXCore::GuardSignalDeferringSection(VMATracking.Mutex, Thread);
     if (Is64Bit) {
       Result = reinterpret_cast<uint64_t>(::shmat(shmid, shmaddr, shmflg));
       if (Result == -1) {
@@ -328,7 +342,7 @@ uint64_t SyscallHandler::GuestShmat(bool Is64Bit, FEXCore::Core::InternalThreadS
       }
     } else {
       uint32_t Addr;
-      Result = FEX::HLE::_SyscallHandler->Get32BitAllocator()->Shmat(shmid, shmaddr, shmflg, &Addr);
+      Result = Get32BitAllocator()->Shmat(shmid, shmaddr, shmflg, &Addr);
       if (FEX::HLE::HasSyscallError(Result)) {
         return Result;
       }
@@ -341,10 +355,10 @@ uint64_t SyscallHandler::GuestShmat(bool Is64Bit, FEXCore::Core::InternalThreadS
     LOGMAN_THROW_A_FMT(res != -1, "shmctl IPC_STAT failed");
 
     Length = stat.shm_segsz;
-    FEX::HLE::_SyscallHandler->TrackShmat(Thread, shmid, Result, shmflg, Length);
+    TrackShmat(Thread, shmid, Result, shmflg, Length);
   }
 
-  FEX::HLE::_SyscallHandler->InvalidateCodeRangeIfNecessary(Thread, Result, Length);
+  InvalidateCodeRangeIfNecessary(Thread, Result, Length);
   return Result;
 }
 
@@ -352,23 +366,23 @@ uint64_t SyscallHandler::GuestShmdt(bool Is64Bit, FEXCore::Core::InternalThreadS
   uint64_t Result {};
   uint64_t Length {};
   {
-    auto lk = FEXCore::GuardSignalDeferringSection(FEX::HLE::_SyscallHandler->VMATracking.Mutex, Thread);
+    auto lk = FEXCore::GuardSignalDeferringSection(VMATracking.Mutex, Thread);
     if (Is64Bit) {
       Result = ::shmdt(shmaddr);
       if (Result == -1) {
         return -errno;
       }
     } else {
-      Result = FEX::HLE::_SyscallHandler->Get32BitAllocator()->Shmdt(shmaddr);
+      Result = Get32BitAllocator()->Shmdt(shmaddr);
       if (FEX::HLE::HasSyscallError(Result)) {
         return Result;
       }
     }
 
-    Length = FEX::HLE::_SyscallHandler->TrackShmdt(Thread, reinterpret_cast<uintptr_t>(shmaddr));
+    Length = TrackShmdt(Thread, reinterpret_cast<uintptr_t>(shmaddr));
   }
 
-  FEX::HLE::_SyscallHandler->InvalidateCodeRangeIfNecessary(Thread, reinterpret_cast<uintptr_t>(shmaddr), Length);
+  InvalidateCodeRangeIfNecessary(Thread, reinterpret_cast<uintptr_t>(shmaddr), Length);
   return Result;
 }
 
@@ -385,19 +399,56 @@ std::optional<SyscallHandler::LateApplyExtendedVolatileMetadata> SyscallHandler:
   if (!(flags & MAP_ANONYMOUS)) {
     struct stat64 buf;
     fstat64(fd, &buf);
-    VMATracking::MRID mrid {buf.st_dev, buf.st_ino};
+
+    const VMATracking::MRID mrid {buf.st_dev, buf.st_ino};
 
     char Tmp[PATH_MAX];
     auto PathLength = FEX::get_fdpath(fd, Tmp);
 
-    if (PathLength != -1) {
-      auto [Iter, Inserted] = VMATracking.InsertMappedResource(mrid, {nullptr, nullptr, 0});
-      Resource = &Iter->second;
+    auto [ResourceIt, ResourceEnd] = VMATracking.FindResources(mrid);
+    bool Inserted = false;
+    const bool MappedELFHeaderAgain = ResourceIt != ResourceEnd && offset == 0 && !ResourceIt->second.ProgramHeaders.empty();
+    if (ResourceIt == ResourceEnd || MappedELFHeaderAgain) {
+      // Create a new MappedResource for previously unseen file and for re-mappings of an ELF header
+      ResourceIt = VMATracking.InsertMappedResource(mrid, {nullptr, nullptr, 0});
+      ResourceIt->second.Iterator = ResourceIt;
+      Inserted = true;
+    }
+    Resource = &ResourceIt->second;
 
+    // Only handle FDs that are backed by regular files that are executable
+    if (PathLength != -1 && S_ISREG(buf.st_mode) && (buf.st_mode & S_IXUSR)) {
+      // ELF files that are mapped multiple times get a separate MappedResource for each base virtual address
       if (Inserted) {
         Resource->MappedFile = fextl::make_unique<FEXCore::ExecutableFileInfo>();
         Resource->MappedFile->Filename = fextl::string(Tmp, PathLength);
-        Resource->Iterator = Iter;
+
+        // Read ELF headers if applicable.
+        // For performance, skip ELF checks if we're not mapping the file header
+        bool CheckForElfFile = (offset == 0);
+#if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
+        CheckForElfFile = true;
+#endif
+        if (CheckForElfFile) {
+          Resource->ProgramHeaders = ReadELFHeaders(fd, std::span {reinterpret_cast<std::byte*>(addr), length});
+          LOGMAN_THROW_A_FMT(Resource->ProgramHeaders.empty() || offset == 0, "Expected file offset 0 for the first mapping of an ELF "
+                                                                              "file");
+        }
+      } else if (ResourceIt->second.ProgramHeaders.empty()) {
+        // Not an ELF file, so we don't need to distinguish between different base addresses
+      } else {
+        // Mapped a non-header section of an ELF file.
+        // Look up the corresponding MappedResource using the expected base address.
+
+        ResourceIt = std::find_if(ResourceIt, ResourceEnd, [&](const VMATracking::MappedResource::ContainerType::value_type& ResourcePair) {
+          auto& Resource = ResourcePair.second;
+          auto ExpectedBase = FEXCore::InferMappingBaseAddress(
+            Resource.ProgramHeaders, addr, Size, offset,
+            (ProtMapping.Executable ? PF_X : 0) | (ProtMapping.Writable ? PF_W : 0) | (ProtMapping.Readable ? PF_R : 0));
+          return ExpectedBase == Resource.FirstVMA->Base;
+        });
+        LOGMAN_THROW_A_FMT(ResourceIt != ResourceEnd, "ERROR: Could not find base for file mapping at {:#x} (offset {:#x})", addr, offset);
+        Resource = &ResourceIt->second;
       }
 
       const fextl::string Filename = FHU::Filesystem::GetFilename(Resource->MappedFile->Filename);
@@ -418,12 +469,12 @@ std::optional<SyscallHandler::LateApplyExtendedVolatileMetadata> SyscallHandler:
   } else if (flags & MAP_SHARED) {
     VMATracking::MRID mrid {VMATracking::SpecialDev::Anon, AnonSharedId++};
 
-    auto [Iter, Inserted] = VMATracking.InsertMappedResource(mrid, {nullptr, nullptr, 0});
-    LOGMAN_THROW_A_FMT(Inserted == true, "VMA tracking error");
+    auto [Iter, IterEnd] = VMATracking.FindResources(mrid);
+    LOGMAN_THROW_A_FMT(Iter == IterEnd, "VMA tracking error");
+
+    Iter = VMATracking.InsertMappedResource(mrid, {nullptr, nullptr, 0});
     Resource = &Iter->second;
     Resource->Iterator = Iter;
-  } else {
-    Resource = nullptr;
   }
 
   VMATracking.TrackVMARange(CTX, Resource, addr, offset, Size, VMATracking::VMAFlags::fromFlags(flags), ProtMapping);
@@ -479,11 +530,12 @@ void SyscallHandler::TrackMremap(FEXCore::Core::InternalThreadState* Thread, uin
 void SyscallHandler::TrackShmat(FEXCore::Core::InternalThreadState* Thread, int shmid, uint64_t shmaddr, int shmflg, uint64_t Length) {
   VMATracking::MRID mrid {VMATracking::SpecialDev::SHM, static_cast<uint64_t>(shmid)};
 
-  auto [Iter, Inserted] = VMATracking.InsertMappedResource(mrid, {nullptr, nullptr, Length});
-  auto Resource = &Iter->second;
-  if (Inserted) {
-    Resource->Iterator = Iter;
+  auto [Iter, IterEnd] = VMATracking.FindResources(mrid);
+  if (Iter == IterEnd) {
+    Iter = VMATracking.InsertMappedResource(mrid, {nullptr, nullptr, Length});
+    Iter->second.Iterator = Iter;
   }
+  auto Resource = &Iter->second;
   VMATracking.TrackVMARange(CTX, Resource, shmaddr, 0, Length, VMATracking::VMAFlags::fromFlags(MAP_SHARED), VMATracking::VMAProt::fromSHM(shmflg));
 }
 

Source/Tools/LinuxEmulation/LinuxSyscalls/SyscallsVMATracking.cpp

@@ -35,7 +35,7 @@ auto VMAFlags::fromFlags(int Flags) -> VMAFlags {
 }
 
 /// List Operations ///
-inline void VMATracking::ListCheckVMALinks(VMAEntry* VMA) {
+static inline void ListCheckVMALinks(const VMAEntry* VMA) {
   if (VMA) {
     LOGMAN_THROW_A_FMT(VMA->ResourceNextVMA != VMA, "VMA tracking error");
     LOGMAN_THROW_A_FMT(VMA->ResourcePrevVMA != VMA, "VMA tracking error");
@@ -44,7 +44,7 @@ inline void VMATracking::ListCheckVMALinks(VMAEntry* VMA) {
 
 // Removes a VMA from corresponding MappedResource list
 // Returns true if list is empty
-bool VMATracking::ListRemove(VMAEntry* VMA) {
+static bool ListRemove(VMAEntry* VMA) {
   LOGMAN_THROW_A_FMT(VMA->Resource != nullptr, "VMA tracking error");
 
   // if it has prev, make prev to next
@@ -78,7 +78,7 @@ bool VMATracking::ListRemove(VMAEntry* VMA) {
 
 // Replaces a VMA in corresponding MappedResource list
 // Requires NewVMA->Resource, NewVMA->ResourcePrevVMA and NewVMA->ResourceNextVMA to be already setup
-void VMATracking::ListReplace(VMAEntry* VMA, VMAEntry* NewVMA) {
+static void ListReplace(VMAEntry* VMA, VMAEntry* NewVMA) {
   LOGMAN_THROW_A_FMT(VMA->Resource != nullptr, "VMA tracking error");
 
   LOGMAN_THROW_A_FMT(VMA->Resource == NewVMA->Resource, "VMA tracking error");
@@ -107,7 +107,7 @@ void VMATracking::ListReplace(VMAEntry* VMA, VMAEntry* NewVMA) {
 
 // Inserts a VMA in corresponding MappedResource list
 // Requires NewVMA->Resource, NewVMA->ResourcePrevVMA and NewVMA->ResourceNextVMA to be already setup
-void VMATracking::ListInsertAfter(VMAEntry* AfterVMA, VMAEntry* NewVMA) {
+static void ListInsertAfter(VMAEntry* AfterVMA, VMAEntry* NewVMA) {
   LOGMAN_THROW_A_FMT(NewVMA->Resource != nullptr, "VMA tracking error");
 
   LOGMAN_THROW_A_FMT(AfterVMA->Resource == NewVMA->Resource, "VMA tracking error");
@@ -128,7 +128,7 @@ void VMATracking::ListInsertAfter(VMAEntry* AfterVMA, VMAEntry* NewVMA) {
 
 // Prepends a VMA
 // Requires NewVMA->Resource, NewVMA->ResourcePrevVMA and NewVMA->ResourceNextVMA to be already setup
-void VMATracking::ListPrepend(MappedResource* Resource, VMAEntry* NewVMA) {
+static void ListPrepend(MappedResource* Resource, VMAEntry* NewVMA) {
   LOGMAN_THROW_A_FMT(Resource != nullptr, "VMA tracking error");
 
   LOGMAN_THROW_A_FMT(NewVMA->Resource == Resource, "VMA tracking error");