webkitgtk-2.16.5HEADwebkitgtk-2.16.5master

1 files changed, 371 insertions, 276 deletions

diff --git a/Source/JavaScriptCore/assembler/ARMv7Assembler.h b/Source/JavaScriptCore/assembler/ARMv7Assembler.h
index 5257f32a8..86218ea72 100644
--- a/Source/JavaScriptCore/assembler/ARMv7Assembler.h
+++ b/Source/JavaScriptCore/assembler/ARMv7Assembler.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2009, 2010, 2012, 2013 Apple Inc. All rights reserved.
+ * Copyright (C) 2009, 2010, 2012, 2013, 2014 Apple Inc. All rights reserved.
  * Copyright (C) 2010 University of Szeged
  *
  * Redistribution and use in source and binary forms, with or without
@@ -24,12 +24,12 @@
  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 
  */
 
-#ifndef ARMAssembler_h
-#define ARMAssembler_h
+#pragma once
 
 #if ENABLE(ASSEMBLER) && CPU(ARM_THUMB2)
 
 #include "AssemblerBuffer.h"
+#include "AssemblerCommon.h"
 #include <limits.h>
 #include <wtf/Assertions.h>
 #include <wtf/Vector.h>
@@ -38,23 +38,83 @@
 namespace JSC {
 
 namespace ARMRegisters {
+
+    #define FOR_EACH_CPU_REGISTER(V) \
+        FOR_EACH_CPU_GPREGISTER(V) \
+        FOR_EACH_CPU_SPECIAL_REGISTER(V) \
+        FOR_EACH_CPU_FPREGISTER(V)
+
+    // The following are defined as pairs of the following value:
+    // 1. type of the storage needed to save the register value by the JIT probe.
+    // 2. name of the register.
+    #define FOR_EACH_CPU_GPREGISTER(V) \
+        V(void*, r0) \
+        V(void*, r1) \
+        V(void*, r2) \
+        V(void*, r3) \
+        V(void*, r4) \
+        V(void*, r5) \
+        V(void*, r6) \
+        V(void*, r7) \
+        V(void*, r8) \
+        V(void*, r9) \
+        V(void*, r10) \
+        V(void*, r11) \
+        V(void*, ip) \
+        V(void*, sp) \
+        V(void*, lr) \
+        V(void*, pc)
+
+    #define FOR_EACH_CPU_SPECIAL_REGISTER(V) \
+        V(void*, apsr) \
+        V(void*, fpscr) \
+
+    #define FOR_EACH_CPU_FPREGISTER(V) \
+        V(double, d0) \
+        V(double, d1) \
+        V(double, d2) \
+        V(double, d3) \
+        V(double, d4) \
+        V(double, d5) \
+        V(double, d6) \
+        V(double, d7) \
+        V(double, d8) \
+        V(double, d9) \
+        V(double, d10) \
+        V(double, d11) \
+        V(double, d12) \
+        V(double, d13) \
+        V(double, d14) \
+        V(double, d15) \
+        V(double, d16) \
+        V(double, d17) \
+        V(double, d18) \
+        V(double, d19) \
+        V(double, d20) \
+        V(double, d21) \
+        V(double, d22) \
+        V(double, d23) \
+        V(double, d24) \
+        V(double, d25) \
+        V(double, d26) \
+        V(double, d27) \
+        V(double, d28) \
+        V(double, d29) \
+        V(double, d30) \
+        V(double, d31)
+
     typedef enum {
-        r0,
-        r1,
-        r2,
-        r3,
-        r4,
-        r5,
-        r6,
-        r7, fp = r7,   // frame pointer
-        r8,
-        r9, sb = r9,   // static base
-        r10, sl = r10, // stack limit
-        r11,
-        r12, ip = r12,
-        r13, sp = r13,
-        r14, lr = r14,
-        r15, pc = r15,
+        #define DECLARE_REGISTER(_type, _regName) _regName,
+        FOR_EACH_CPU_GPREGISTER(DECLARE_REGISTER)
+        #undef DECLARE_REGISTER
+
+        fp = r7,   // frame pointer
+        sb = r9,   // static base
+        sl = r10,  // stack limit
+        r12 = ip,
+        r13 = sp,
+        r14 = lr,
+        r15 = pc
     } RegisterID;
 
     typedef enum {
@@ -93,38 +153,9 @@ namespace ARMRegisters {
     } FPSingleRegisterID;
 
     typedef enum {
-        d0,
-        d1,
-        d2,
-        d3,
-        d4,
-        d5,
-        d6,
-        d7,
-        d8,
-        d9,
-        d10,
-        d11,
-        d12,
-        d13,
-        d14,
-        d15,
-        d16,
-        d17,
-        d18,
-        d19,
-        d20,
-        d21,
-        d22,
-        d23,
-        d24,
-        d25,
-        d26,
-        d27,
-        d28,
-        d29,
-        d30,
-        d31,
+        #define DECLARE_REGISTER(_type, _regName) _regName,
+        FOR_EACH_CPU_FPREGISTER(DECLARE_REGISTER)
+        #undef DECLARE_REGISTER
     } FPDoubleRegisterID;
 
     typedef enum {
@@ -174,77 +205,7 @@ namespace ARMRegisters {
         return (FPDoubleRegisterID)(reg >> 1);
     }
 
-#if USE(MASM_PROBE)
-    #define FOR_EACH_CPU_REGISTER(V) \
-        FOR_EACH_CPU_GPREGISTER(V) \
-        FOR_EACH_CPU_SPECIAL_REGISTER(V) \
-        FOR_EACH_CPU_FPREGISTER(V)
-
-    #define FOR_EACH_CPU_GPREGISTER(V) \
-        V(void*, r0) \
-        V(void*, r1) \
-        V(void*, r2) \
-        V(void*, r3) \
-        V(void*, r4) \
-        V(void*, r5) \
-        V(void*, r6) \
-        V(void*, r7) \
-        V(void*, r8) \
-        V(void*, r9) \
-        V(void*, r10) \
-        V(void*, r11) \
-        V(void*, ip) \
-        V(void*, sp) \
-        V(void*, lr) \
-        V(void*, pc)
-
-    #define FOR_EACH_CPU_SPECIAL_REGISTER(V) \
-        V(void*, apsr) \
-        V(void*, fpscr) \
-
-    #define FOR_EACH_CPU_FPREGISTER(V) \
-        V(double, d0) \
-        V(double, d1) \
-        V(double, d2) \
-        V(double, d3) \
-        V(double, d4) \
-        V(double, d5) \
-        V(double, d6) \
-        V(double, d7) \
-        V(double, d8) \
-        V(double, d9) \
-        V(double, d10) \
-        V(double, d11) \
-        V(double, d12) \
-        V(double, d13) \
-        V(double, d14) \
-        V(double, d15) \
-        FOR_EACH_CPU_FPREGISTER_EXTENSION(V)
-
-#if CPU(APPLE_ARMV7S)
-    #define FOR_EACH_CPU_FPREGISTER_EXTENSION(V) \
-        V(double, d16) \
-        V(double, d17) \
-        V(double, d18) \
-        V(double, d19) \
-        V(double, d20) \
-        V(double, d21) \
-        V(double, d22) \
-        V(double, d23) \
-        V(double, d24) \
-        V(double, d25) \
-        V(double, d26) \
-        V(double, d27) \
-        V(double, d28) \
-        V(double, d29) \
-        V(double, d30) \
-        V(double, d31)
-#else
-    #define FOR_EACH_CPU_FPREGISTER_EXTENSION(V) // Nothing to add.
-#endif // CPU(APPLE_ARMV7S)
-
-#endif // USE(MASM_PROBE)
-}
+} // namespace ARMRegisters
 
 class ARMv7Assembler;
 class ARMThumbImmediate {
@@ -492,11 +453,11 @@ public:
     typedef ARMRegisters::FPQuadRegisterID FPQuadRegisterID;
     typedef FPDoubleRegisterID FPRegisterID;
     
-    static RegisterID firstRegister() { return ARMRegisters::r0; }
-    static RegisterID lastRegister() { return ARMRegisters::r13; }
-    
-    static FPRegisterID firstFPRegister() { return ARMRegisters::d0; }
-    static FPRegisterID lastFPRegister() { return ARMRegisters::d31; }
+    static constexpr RegisterID firstRegister() { return ARMRegisters::r0; }
+    static constexpr RegisterID lastRegister() { return ARMRegisters::r13; }
+
+    static constexpr FPRegisterID firstFPRegister() { return ARMRegisters::d0; }
+    static constexpr FPRegisterID lastFPRegister() { return ARMRegisters::d31; }
 
     // (HS, LO, HI, LS) -> (AE, B, A, BE)
     // (VS, VC) -> (O, NO)
@@ -583,6 +544,8 @@ public:
     {
     }
 
+    AssemblerBuffer& buffer() { return m_formatter.m_buffer; }
+
 private:
 
     // ARMv7, Appx-A.6.3
@@ -646,6 +609,8 @@ private:
         OP_ADD_SP_imm_T1    = 0xA800,
         OP_ADD_SP_imm_T2    = 0xB000,
         OP_SUB_SP_imm_T1    = 0xB080,
+        OP_PUSH_T1          = 0xB400,
+        OP_POP_T1           = 0xBC00,
         OP_BKPT             = 0xBE00,
         OP_IT               = 0xBF00,
         OP_NOP_T1           = 0xBF00,
@@ -654,6 +619,8 @@ private:
     typedef enum {
         OP_B_T1         = 0xD000,
         OP_B_T2         = 0xE000,
+        OP_POP_T2       = 0xE8BD,
+        OP_PUSH_T2      = 0xE92D,
         OP_AND_reg_T2   = 0xEA00,
         OP_TST_reg_T2   = 0xEA10,
         OP_ORR_reg_T2   = 0xEA40,
@@ -714,7 +681,7 @@ private:
         OP_MOVT         = 0xF2C0,
         OP_UBFX_T1      = 0xF3C0,
         OP_NOP_T2a      = 0xF3AF,
-        OP_DMB_SY_T2a   = 0xF3BF,
+        OP_DMB_T1a      = 0xF3BF,
         OP_STRB_imm_T3  = 0xF800,
         OP_STRB_reg_T2  = 0xF800,
         OP_LDRB_imm_T3  = 0xF810,
@@ -741,39 +708,40 @@ private:
         OP_ROR_reg_T2   = 0xFA60,
         OP_CLZ          = 0xFAB0,
         OP_SMULL_T1     = 0xFB80,
-#if CPU(APPLE_ARMV7S)
+#if HAVE(ARM_IDIV_INSTRUCTIONS)
         OP_SDIV_T1      = 0xFB90,
         OP_UDIV_T1      = 0xFBB0,
 #endif
     } OpcodeID1;
 
     typedef enum {
-        OP_VADD_T2b     = 0x0A00,
-        OP_VDIVb        = 0x0A00,
-        OP_FLDSb        = 0x0A00,
-        OP_VLDRb        = 0x0A00,
-        OP_VMOV_IMM_T2b = 0x0A00,
-        OP_VMOV_T2b     = 0x0A40,
-        OP_VMUL_T2b     = 0x0A00,
-        OP_FSTSb        = 0x0A00,
-        OP_VSTRb        = 0x0A00,
-        OP_VMOV_StoCb   = 0x0A10,
-        OP_VMOV_CtoSb   = 0x0A10,
-        OP_VMOV_DtoCb   = 0x0A10,
-        OP_VMOV_CtoDb   = 0x0A10,
-        OP_VMRSb        = 0x0A10,
-        OP_VABS_T2b     = 0x0A40,
-        OP_VCMPb        = 0x0A40,
-        OP_VCVT_FPIVFPb = 0x0A40,
-        OP_VNEG_T2b     = 0x0A40,
-        OP_VSUB_T2b     = 0x0A40,
-        OP_VSQRT_T1b    = 0x0A40,
-        OP_VCVTSD_T1b   = 0x0A40,
-        OP_VCVTDS_T1b   = 0x0A40,
-        OP_NOP_T2b      = 0x8000,
-        OP_DMB_SY_T2b   = 0x8F5F,
-        OP_B_T3b        = 0x8000,
-        OP_B_T4b        = 0x9000,
+        OP_VADD_T2b      = 0x0A00,
+        OP_VDIVb         = 0x0A00,
+        OP_FLDSb         = 0x0A00,
+        OP_VLDRb         = 0x0A00,
+        OP_VMOV_IMM_T2b  = 0x0A00,
+        OP_VMOV_T2b      = 0x0A40,
+        OP_VMUL_T2b      = 0x0A00,
+        OP_FSTSb         = 0x0A00,
+        OP_VSTRb         = 0x0A00,
+        OP_VMOV_StoCb    = 0x0A10,
+        OP_VMOV_CtoSb    = 0x0A10,
+        OP_VMOV_DtoCb    = 0x0A10,
+        OP_VMOV_CtoDb    = 0x0A10,
+        OP_VMRSb         = 0x0A10,
+        OP_VABS_T2b      = 0x0A40,
+        OP_VCMPb         = 0x0A40,
+        OP_VCVT_FPIVFPb  = 0x0A40,
+        OP_VNEG_T2b      = 0x0A40,
+        OP_VSUB_T2b      = 0x0A40,
+        OP_VSQRT_T1b     = 0x0A40,
+        OP_VCVTSD_T1b    = 0x0A40,
+        OP_VCVTDS_T1b    = 0x0A40,
+        OP_NOP_T2b       = 0x8000,
+        OP_DMB_SY_T1b    = 0x8F5F,
+        OP_DMB_ISHST_T1b = 0x8F5A,
+        OP_B_T3b         = 0x8000,
+        OP_B_T4b         = 0x9000,
     } OpcodeID2;
 
     struct FourFours {
@@ -799,11 +767,11 @@ private:
     class ARMInstructionFormatter;
 
     // false means else!
-    bool ifThenElseConditionBit(Condition condition, bool isIf)
+    static bool ifThenElseConditionBit(Condition condition, bool isIf)
     {
         return isIf ? (condition & 1) : !(condition & 1);
     }
-    uint8_t ifThenElse(Condition condition, bool inst2if, bool inst3if, bool inst4if)
+    static uint8_t ifThenElse(Condition condition, bool inst2if, bool inst3if, bool inst4if)
     {
         int mask = (ifThenElseConditionBit(condition, inst2if) << 3)
             | (ifThenElseConditionBit(condition, inst3if) << 2)
@@ -812,7 +780,7 @@ private:
         ASSERT((condition != ConditionAL) || !(mask & (mask - 1)));
         return (condition << 4) | mask;
     }
-    uint8_t ifThenElse(Condition condition, bool inst2if, bool inst3if)
+    static uint8_t ifThenElse(Condition condition, bool inst2if, bool inst3if)
     {
         int mask = (ifThenElseConditionBit(condition, inst2if) << 3)
             | (ifThenElseConditionBit(condition, inst3if) << 2)
@@ -820,7 +788,7 @@ private:
         ASSERT((condition != ConditionAL) || !(mask & (mask - 1)));
         return (condition << 4) | mask;
     }
-    uint8_t ifThenElse(Condition condition, bool inst2if)
+    static uint8_t ifThenElse(Condition condition, bool inst2if)
     {
         int mask = (ifThenElseConditionBit(condition, inst2if) << 3)
             | 4;
@@ -828,7 +796,7 @@ private:
         return (condition << 4) | mask;
     }
 
-    uint8_t ifThenElse(Condition condition)
+    static uint8_t ifThenElse(Condition condition)
     {
         int mask = 8;
         return (condition << 4) | mask;
@@ -855,7 +823,7 @@ public:
         ASSERT(rn != ARMRegisters::pc);
         ASSERT(imm.isValid());
 
-        if (rn == ARMRegisters::sp) {
+        if (rn == ARMRegisters::sp && imm.isUInt16()) {
             ASSERT(!(imm.getUInt16() & 3));
             if (!(rd & 8) && imm.isUInt10()) {
                 m_formatter.oneWordOp5Reg3Imm8(OP_ADD_SP_imm_T1, rd, static_cast<uint8_t>(imm.getUInt10() >> 2));
@@ -894,6 +862,11 @@ public:
     // NOTE: In an IT block, add doesn't modify the flags register.
     ALWAYS_INLINE void add(RegisterID rd, RegisterID rn, RegisterID rm)
     {
+        if (rd == ARMRegisters::sp) {
+            mov(rd, rn);
+            rn = rd;
+        }
+
         if (rd == rn)
             m_formatter.oneWordOp8RegReg143(OP_ADD_reg_T2, rm, rd);
         else if (rd == rm)
@@ -1183,9 +1156,10 @@ public:
     {
         ASSERT(rn != ARMRegisters::pc); // LDR (literal)
         ASSERT(imm.isUInt12());
+        ASSERT(!(imm.getUInt12() & 1));
 
         if (!((rt | rn) & 8) && imm.isUInt6())
-            m_formatter.oneWordOp5Imm5Reg3Reg3(OP_LDRH_imm_T1, imm.getUInt6() >> 2, rn, rt);
+            m_formatter.oneWordOp5Imm5Reg3Reg3(OP_LDRH_imm_T1, imm.getUInt6() >> 1, rn, rt);
         else
             m_formatter.twoWordOp12Reg4Reg4Imm12(OP_LDRH_imm_T2, rn, rt, imm.getUInt12());
     }
@@ -1353,11 +1327,14 @@ public:
         uint16_t* address = static_cast<uint16_t*>(instructionStart);
         ARMThumbImmediate lo16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(imm));
         ARMThumbImmediate hi16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(imm >> 16));
-        address[0] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, lo16);
-        address[1] = twoWordOp5i6Imm4Reg4EncodedImmSecond(right, lo16);
-        address[2] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOVT, hi16);
-        address[3] = twoWordOp5i6Imm4Reg4EncodedImmSecond(right, hi16);
-        address[4] = OP_CMP_reg_T2 | left;
+        uint16_t instruction[] = {
+            twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, lo16),
+            twoWordOp5i6Imm4Reg4EncodedImmSecond(right, lo16),
+            twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOVT, hi16),
+            twoWordOp5i6Imm4Reg4EncodedImmSecond(right, hi16),
+            static_cast<uint16_t>(OP_CMP_reg_T2 | left)
+        };
+        performJITMemcpy(address, instruction, sizeof(uint16_t) * 5);
         cacheFlush(address, sizeof(uint16_t) * 5);
     }
 #else
@@ -1368,8 +1345,11 @@ public:
         ASSERT(!BadReg(rd));
         
         uint16_t* address = static_cast<uint16_t*>(instructionStart);
-        address[0] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, imm);
-        address[1] = twoWordOp5i6Imm4Reg4EncodedImmSecond(rd, imm);
+        uint16_t instruction[] = {
+            twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, imm),
+            twoWordOp5i6Imm4Reg4EncodedImmSecond(rd, imm)
+        };
+        performJITMemcpy(address, instruction, sizeof(uint16_t) * 2);
         cacheFlush(address, sizeof(uint16_t) * 2);
     }
 #endif
@@ -1488,9 +1468,49 @@ public:
         m_formatter.twoWordOp12Reg4FourFours(OP_ROR_reg_T2, rn, FourFours(0xf, rd, 0, rm));
     }
 
-#if CPU(APPLE_ARMV7S)
+    ALWAYS_INLINE void pop(RegisterID dest)
+    {
+        if (dest < ARMRegisters::r8)
+            m_formatter.oneWordOp7Imm9(OP_POP_T1, 1 << dest);
+        else {
+            // Load postindexed with writeback.
+            ldr(dest, ARMRegisters::sp, sizeof(void*), false, true);
+        }
+    }
+
+    ALWAYS_INLINE void pop(uint32_t registerList)
+    {
+        ASSERT(WTF::bitCount(registerList) > 1);
+        ASSERT(!((1 << ARMRegisters::pc) & registerList) || !((1 << ARMRegisters::lr) & registerList));
+        ASSERT(!((1 << ARMRegisters::sp) & registerList));
+        m_formatter.twoWordOp16Imm16(OP_POP_T2, registerList);
+    }
+
+    ALWAYS_INLINE void push(RegisterID src)
+    {
+        if (src < ARMRegisters::r8)
+            m_formatter.oneWordOp7Imm9(OP_PUSH_T1, 1 << src);
+        else if (src == ARMRegisters::lr)
+            m_formatter.oneWordOp7Imm9(OP_PUSH_T1, 0x100);
+        else {
+            // Store preindexed with writeback.
+            str(src, ARMRegisters::sp, -sizeof(void*), true, true);
+        }
+    }
+
+    ALWAYS_INLINE void push(uint32_t registerList)
+    {
+        ASSERT(WTF::bitCount(registerList) > 1);
+        ASSERT(!((1 << ARMRegisters::pc) & registerList));
+        ASSERT(!((1 << ARMRegisters::sp) & registerList));
+        m_formatter.twoWordOp16Imm16(OP_PUSH_T2, registerList);
+    }
+
+#if HAVE(ARM_IDIV_INSTRUCTIONS)
+    template<int datasize>
     ALWAYS_INLINE void sdiv(RegisterID rd, RegisterID rn, RegisterID rm)
     {
+        static_assert(datasize == 32, "sdiv datasize must be 32 for armv7s");        
         ASSERT(!BadReg(rd));
         ASSERT(!BadReg(rn));
         ASSERT(!BadReg(rm));
@@ -1635,8 +1655,8 @@ public:
         ASSERT(rn != ARMRegisters::pc);
         ASSERT(imm.isUInt12());
         
-        if (!((rt | rn) & 8) && imm.isUInt7())
-            m_formatter.oneWordOp5Imm5Reg3Reg3(OP_STRH_imm_T1, imm.getUInt7() >> 2, rn, rt);
+        if (!((rt | rn) & 8) && imm.isUInt6())
+            m_formatter.oneWordOp5Imm5Reg3Reg3(OP_STRH_imm_T1, imm.getUInt6() >> 1, rn, rt);
         else
             m_formatter.twoWordOp12Reg4Reg4Imm12(OP_STRH_imm_T2, rn, rt, imm.getUInt12());
     }
@@ -1834,7 +1854,7 @@ public:
         m_formatter.twoWordOp12Reg40Imm3Reg4Imm20Imm5(OP_UBFX_T1, rd, rn, (lsb & 0x1c) << 10, (lsb & 0x3) << 6, (width - 1) & 0x1f);
     }
 
-#if CPU(APPLE_ARMV7S)
+#if HAVE(ARM_IDIV_INSTRUCTIONS)
     ALWAYS_INLINE void udiv(RegisterID rd, RegisterID rn, RegisterID rm)
     {
         ASSERT(!BadReg(rd));
@@ -1984,9 +2004,51 @@ public:
         m_formatter.twoWordOp16Op16(OP_NOP_T2a, OP_NOP_T2b);
     }
     
+    static constexpr int16_t nopPseudo16()
+    {
+        return OP_NOP_T1;
+    }
+
+    static constexpr int32_t nopPseudo32()
+    {
+        return OP_NOP_T2a | (OP_NOP_T2b << 16);
+    }
+
+    static void fillNops(void* base, size_t size, bool isCopyingToExecutableMemory)
+    {
+        RELEASE_ASSERT(!(size % sizeof(int16_t)));
+
+        char* ptr = static_cast<char*>(base);
+        const size_t num32s = size / sizeof(int32_t);
+        for (size_t i = 0; i < num32s; i++) {
+            const int32_t insn = nopPseudo32();
+            if (isCopyingToExecutableMemory)
+                performJITMemcpy(ptr, &insn, sizeof(int32_t));
+            else
+                memcpy(ptr, &insn, sizeof(int32_t));
+            ptr += sizeof(int32_t);
+        }
+
+        const size_t num16s = (size % sizeof(int32_t)) / sizeof(int16_t);
+        ASSERT(num16s == 0 || num16s == 1);
+        ASSERT(num16s * sizeof(int16_t) + num32s * sizeof(int32_t) == size);
+        if (num16s) {
+            const int16_t insn = nopPseudo16();
+            if (isCopyingToExecutableMemory)
+                performJITMemcpy(ptr, &insn, sizeof(int16_t));
+            else
+                memcpy(ptr, &insn, sizeof(int16_t));
+        }
+    }
+
     void dmbSY()
     {
-        m_formatter.twoWordOp16Op16(OP_DMB_SY_T2a, OP_DMB_SY_T2b);
+        m_formatter.twoWordOp16Op16(OP_DMB_T1a, OP_DMB_SY_T1b);
+    }
+
+    void dmbISHST()
+    {
+        m_formatter.twoWordOp16Op16(OP_DMB_T1a, OP_DMB_ISHST_T1b);
     }
 
     AssemblerLabel labelIgnoringWatchpoints()
@@ -2036,14 +2098,7 @@ public:
         return b.m_offset - a.m_offset;
     }
 
-    int executableOffsetFor(int location)
-    {
-        if (!location)
-            return 0;
-        return static_cast<int32_t*>(m_formatter.data())[location / sizeof(int32_t) - 1];
-    }
-    
-    int jumpSizeDelta(JumpType jumpType, JumpLinkType jumpLinkType) { return JUMP_ENUM_SIZE(jumpType) - JUMP_ENUM_SIZE(jumpLinkType); }
+    static int jumpSizeDelta(JumpType jumpType, JumpLinkType jumpLinkType) { return JUMP_ENUM_SIZE(jumpType) - JUMP_ENUM_SIZE(jumpLinkType); }
     
     // Assembler admin methods:
 
@@ -2052,7 +2107,7 @@ public:
         return a.from() < b.from();
     }
 
-    bool canCompact(JumpType jumpType)
+    static bool canCompact(JumpType jumpType)
     {
         // The following cannot be compacted:
         //   JumpFixed: represents custom jump sequence
@@ -2061,7 +2116,7 @@ public:
         return (jumpType == JumpNoCondition) || (jumpType == JumpCondition);
     }
     
-    JumpLinkType computeJumpType(JumpType jumpType, const uint8_t* from, const uint8_t* to)
+    static JumpLinkType computeJumpType(JumpType jumpType, const uint8_t* from, const uint8_t* to)
     {
         if (jumpType == JumpFixed)
             return LinkInvalid;
@@ -2105,51 +2160,43 @@ public:
         return LinkConditionalBX;
     }
     
-    JumpLinkType computeJumpType(LinkRecord& record, const uint8_t* from, const uint8_t* to)
+    static JumpLinkType computeJumpType(LinkRecord& record, const uint8_t* from, const uint8_t* to)
     {
         JumpLinkType linkType = computeJumpType(record.type(), from, to);
         record.setLinkType(linkType);
         return linkType;
     }
     
-    void recordLinkOffsets(int32_t regionStart, int32_t regionEnd, int32_t offset)
-    {
-        int32_t ptr = regionStart / sizeof(int32_t);
-        const int32_t end = regionEnd / sizeof(int32_t);
-        int32_t* offsets = static_cast<int32_t*>(m_formatter.data());
-        while (ptr < end)
-            offsets[ptr++] = offset;
-    }
-    
     Vector<LinkRecord, 0, UnsafeVectorOverflow>& jumpsToLink()
     {
         std::sort(m_jumpsToLink.begin(), m_jumpsToLink.end(), linkRecordSourceComparator);
         return m_jumpsToLink;
     }
 
-    void ALWAYS_INLINE link(LinkRecord& record, uint8_t* from, uint8_t* to)
+    static void ALWAYS_INLINE link(LinkRecord& record, uint8_t* from, const uint8_t* fromInstruction8, uint8_t* to)
     {
+        const uint16_t* fromInstruction = reinterpret_cast_ptr<const uint16_t*>(fromInstruction8);
         switch (record.linkType()) {
         case LinkJumpT1:
-            linkJumpT1(record.condition(), reinterpret_cast_ptr<uint16_t*>(from), to);
+            linkJumpT1(record.condition(), reinterpret_cast_ptr<uint16_t*>(from), fromInstruction, to);
             break;
         case LinkJumpT2:
-            linkJumpT2(reinterpret_cast_ptr<uint16_t*>(from), to);
+            linkJumpT2(reinterpret_cast_ptr<uint16_t*>(from), fromInstruction, to);
             break;
         case LinkJumpT3:
-            linkJumpT3(record.condition(), reinterpret_cast_ptr<uint16_t*>(from), to);
+            linkJumpT3(record.condition(), reinterpret_cast_ptr<uint16_t*>(from), fromInstruction, to);
             break;
         case LinkJumpT4:
-            linkJumpT4(reinterpret_cast_ptr<uint16_t*>(from), to);
+            linkJumpT4(reinterpret_cast_ptr<uint16_t*>(from), fromInstruction, to);
             break;
         case LinkConditionalJumpT4:
-            linkConditionalJumpT4(record.condition(), reinterpret_cast_ptr<uint16_t*>(from), to);
+            linkConditionalJumpT4(record.condition(), reinterpret_cast_ptr<uint16_t*>(from), fromInstruction, to);
             break;
         case LinkConditionalBX:
-            linkConditionalBX(record.condition(), reinterpret_cast_ptr<uint16_t*>(from), to);
+            linkConditionalBX(record.condition(), reinterpret_cast_ptr<uint16_t*>(from), fromInstruction, to);
             break;
         case LinkBX:
-            linkBX(reinterpret_cast_ptr<uint16_t*>(from), to);
+            linkBX(reinterpret_cast_ptr<uint16_t*>(from), fromInstruction, to);
             break;
         default:
             RELEASE_ASSERT_NOT_REACHED();
@@ -2186,7 +2233,7 @@ public:
         ASSERT(from.isSet());
         
         uint16_t* location = reinterpret_cast<uint16_t*>(reinterpret_cast<intptr_t>(code) + from.m_offset);
-        linkJumpAbsolute(location, to);
+        linkJumpAbsolute(location, location, to);
     }
 
     static void linkCall(void* code, AssemblerLabel from, void* to)
@@ -2202,15 +2249,24 @@ public:
         setPointer(reinterpret_cast<char*>(code) + where.m_offset, value, false);
     }
 
+    // The static relink and replace methods can use can use |from| for both
+    // the write and executable address for call and jump patching
+    // as they're modifying existing (linked) code, so the address being
+    // provided is correct for relative address computation.
     static void relinkJump(void* from, void* to)
     {
         ASSERT(!(reinterpret_cast<intptr_t>(from) & 1));
         ASSERT(!(reinterpret_cast<intptr_t>(to) & 1));
 
-        linkJumpAbsolute(reinterpret_cast<uint16_t*>(from), to);
+        linkJumpAbsolute(reinterpret_cast<uint16_t*>(from), reinterpret_cast<uint16_t*>(from), to);
 
         cacheFlush(reinterpret_cast<uint16_t*>(from) - 5, 5 * sizeof(uint16_t));
     }
+
+    static void relinkJumpToNop(void* from)
+    {
+        relinkJump(from, from);
+    }
     
     static void relinkCall(void* from, void* to)
     {
@@ -2246,8 +2302,9 @@ public:
         offset |= (1 << 11);
 
         uint16_t* location = reinterpret_cast<uint16_t*>(where);
-        location[1] &= ~((1 << 12) - 1);
-        location[1] |= offset;
+        uint16_t instruction = location[1] & ~((1 << 12) - 1);
+        instruction |= offset;
+        performJITMemcpy(location + 1, &instruction, sizeof(uint16_t));
         cacheFlush(location, sizeof(uint16_t) * 2);
     }
 
@@ -2271,16 +2328,16 @@ public:
 #if OS(LINUX)
         if (canBeJumpT4(reinterpret_cast<uint16_t*>(instructionStart), to)) {
             uint16_t* ptr = reinterpret_cast<uint16_t*>(instructionStart) + 2;
-            linkJumpT4(ptr, to);
+            linkJumpT4(ptr, ptr, to);
             cacheFlush(ptr - 2, sizeof(uint16_t) * 2);
         } else {
             uint16_t* ptr = reinterpret_cast<uint16_t*>(instructionStart) + 5;
-            linkBX(ptr, to);
+            linkBX(ptr, ptr, to);
             cacheFlush(ptr - 5, sizeof(uint16_t) * 5);
         }
 #else
         uint16_t* ptr = reinterpret_cast<uint16_t*>(instructionStart) + 2;
-        linkJumpT4(ptr, to);
+        linkJumpT4(ptr, ptr, to);
         cacheFlush(ptr - 2, sizeof(uint16_t) * 2);
 #endif
     }
@@ -2293,6 +2350,11 @@ public:
         return 4;
 #endif
     }
+
+    static constexpr ptrdiff_t patchableJumpSize()
+    {
+        return 10;
+    }
     
     static void replaceWithLoad(void* instructionStart)
     {
@@ -2301,14 +2363,17 @@ public:
         switch (ptr[0] & 0xFFF0) {
         case OP_LDR_imm_T3:
             break;
-        case OP_ADD_imm_T3:
+        case OP_ADD_imm_T3: {
             ASSERT(!(ptr[1] & 0xF000));
-            ptr[0] &= 0x000F;
-            ptr[0] |= OP_LDR_imm_T3;
-            ptr[1] |= (ptr[1] & 0x0F00) << 4;
-            ptr[1] &= 0xF0FF;
+            uint16_t instructions[2];
+            instructions[0] = ptr[0] & 0x000F;
+            instructions[0] |= OP_LDR_imm_T3;
+            instructions[1] = ptr[1] | (ptr[1] & 0x0F00) << 4;
+            instructions[1] &= 0xF0FF;
+            performJITMemcpy(ptr, instructions, sizeof(uint16_t) * 2);
             cacheFlush(ptr, sizeof(uint16_t) * 2);
             break;
+        }
         default:
             RELEASE_ASSERT_NOT_REACHED();
         }
@@ -2319,14 +2384,17 @@ public:
         ASSERT(!(bitwise_cast<uintptr_t>(instructionStart) & 1));
         uint16_t* ptr = reinterpret_cast<uint16_t*>(instructionStart);
         switch (ptr[0] & 0xFFF0) {
-        case OP_LDR_imm_T3:
+        case OP_LDR_imm_T3: {
             ASSERT(!(ptr[1] & 0x0F00));
-            ptr[0] &= 0x000F;
-            ptr[0] |= OP_ADD_imm_T3;
-            ptr[1] |= (ptr[1] & 0xF000) >> 4;
-            ptr[1] &= 0x0FFF;
+            uint16_t instructions[2];
+            instructions[0] = ptr[0] & 0x000F;
+            instructions[0] |= OP_ADD_imm_T3;
+            instructions[1] = ptr[1] | (ptr[1] & 0xF000) >> 4;
+            instructions[1] &= 0x0FFF;
+            performJITMemcpy(ptr, instructions, sizeof(uint16_t) * 2);
             cacheFlush(ptr, sizeof(uint16_t) * 2);
             break;
+        }
         case OP_ADD_imm_T3:
             break;
         default:
@@ -2375,8 +2443,6 @@ public:
             linuxPageFlush(current, current + page);
 
         linuxPageFlush(current, end);
-#elif OS(WINCE)
-        CacheRangeFlush(code, size, CACHE_SYNC_ALL);
 #else
 #error "The cacheFlush support is missing on this platform."
 #endif
@@ -2456,11 +2522,13 @@ private:
 
         ARMThumbImmediate lo16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(value));
         ARMThumbImmediate hi16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(value >> 16));
-        location[-4] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, lo16);
-        location[-3] = twoWordOp5i6Imm4Reg4EncodedImmSecond((location[-3] >> 8) & 0xf, lo16);
-        location[-2] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOVT, hi16);
-        location[-1] = twoWordOp5i6Imm4Reg4EncodedImmSecond((location[-1] >> 8) & 0xf, hi16);
+        uint16_t instructions[4];
+        instructions[0] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, lo16);
+        instructions[1] = twoWordOp5i6Imm4Reg4EncodedImmSecond((location[-3] >> 8) & 0xf, lo16);
+        instructions[2] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOVT, hi16);
+        instructions[3] = twoWordOp5i6Imm4Reg4EncodedImmSecond((location[-1] >> 8) & 0xf, hi16);
 
+        performJITMemcpy(location - 4, instructions, 4 * sizeof(uint16_t));
         if (flush)
             cacheFlush(location - 4, 4 * sizeof(uint16_t));
     }
@@ -2488,8 +2556,10 @@ private:
         ASSERT(imm.isValid());
         ASSERT(imm.isUInt7());
         uint16_t* location = reinterpret_cast<uint16_t*>(code);
-        location[0] &= ~((static_cast<uint16_t>(0x7f) >> 2) << 6);
-        location[0] |= (imm.getUInt7() >> 2) << 6;
+        uint16_t instruction;
+        instruction = location[0] & ~((static_cast<uint16_t>(0x7f) >> 2) << 6);
+        instruction |= (imm.getUInt7() >> 2) << 6;
+        performJITMemcpy(location, &instruction, sizeof(uint16_t));
         cacheFlush(location, sizeof(uint16_t));
     }
 
@@ -2498,39 +2568,39 @@ private:
         setInt32(code, reinterpret_cast<uint32_t>(value), flush);
     }
 
-    static bool isB(void* address)
+    static bool isB(const void* address)
     {
-        uint16_t* instruction = static_cast<uint16_t*>(address);
+        const uint16_t* instruction = static_cast<const uint16_t*>(address);
         return ((instruction[0] & 0xf800) == OP_B_T4a) && ((instruction[1] & 0xd000) == OP_B_T4b);
     }
 
-    static bool isBX(void* address)
+    static bool isBX(const void* address)
     {
-        uint16_t* instruction = static_cast<uint16_t*>(address);
+        const uint16_t* instruction = static_cast<const uint16_t*>(address);
         return (instruction[0] & 0xff87) == OP_BX;
     }
 
-    static bool isMOV_imm_T3(void* address)
+    static bool isMOV_imm_T3(const void* address)
     {
-        uint16_t* instruction = static_cast<uint16_t*>(address);
+        const uint16_t* instruction = static_cast<const uint16_t*>(address);
         return ((instruction[0] & 0xFBF0) == OP_MOV_imm_T3) && ((instruction[1] & 0x8000) == 0);
     }
 
-    static bool isMOVT(void* address)
+    static bool isMOVT(const void* address)
     {
-        uint16_t* instruction = static_cast<uint16_t*>(address);
+        const uint16_t* instruction = static_cast<const uint16_t*>(address);
         return ((instruction[0] & 0xFBF0) == OP_MOVT) && ((instruction[1] & 0x8000) == 0);
     }
 
-    static bool isNOP_T1(void* address)
+    static bool isNOP_T1(const void* address)
     {
-        uint16_t* instruction = static_cast<uint16_t*>(address);
+        const uint16_t* instruction = static_cast<const uint16_t*>(address);
         return instruction[0] == OP_NOP_T1;
     }
 
-    static bool isNOP_T2(void* address)
+    static bool isNOP_T2(const void* address)
     {
-        uint16_t* instruction = static_cast<uint16_t*>(address);
+        const uint16_t* instruction = static_cast<const uint16_t*>(address);
         return (instruction[0] == OP_NOP_T2a) && (instruction[1] == OP_NOP_T2b);
     }
 
@@ -2578,7 +2648,7 @@ private:
         return ((relative << 7) >> 7) == relative;
     }
     
-    void linkJumpT1(Condition cond, uint16_t* instruction, void* target)
+    static void linkJumpT1(Condition cond, uint16_t* writeTarget, const uint16_t* instruction, void* target)
     {
         // FIMXE: this should be up in the MacroAssembler layer. :-(        
         ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1));
@@ -2593,10 +2663,11 @@ private:
         
         // All branch offsets should be an even distance.
         ASSERT(!(relative & 1));
-        instruction[-1] = OP_B_T1 | ((cond & 0xf) << 8) | ((relative & 0x1fe) >> 1);
+        uint16_t newInstruction = OP_B_T1 | ((cond & 0xf) << 8) | ((relative & 0x1fe) >> 1);
+        performJITMemcpy(writeTarget - 1, &newInstruction, sizeof(uint16_t));
     }
     
-    static void linkJumpT2(uint16_t* instruction, void* target)
+    static void linkJumpT2(uint16_t* writeTarget, const uint16_t* instruction, void* target)
     {
         // FIMXE: this should be up in the MacroAssembler layer. :-(        
         ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1));
@@ -2611,10 +2682,11 @@ private:
         
         // All branch offsets should be an even distance.
         ASSERT(!(relative & 1));
-        instruction[-1] = OP_B_T2 | ((relative & 0xffe) >> 1);
+        uint16_t newInstruction = OP_B_T2 | ((relative & 0xffe) >> 1);
+        performJITMemcpy(writeTarget - 1, &newInstruction, sizeof(uint16_t));
     }
     
-    void linkJumpT3(Condition cond, uint16_t* instruction, void* target)
+    static void linkJumpT3(Condition cond, uint16_t* writeTarget, const uint16_t* instruction, void* target)
     {
         // FIMXE: this should be up in the MacroAssembler layer. :-(
         ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1));
@@ -2625,11 +2697,13 @@ private:
         
         // All branch offsets should be an even distance.
         ASSERT(!(relative & 1));
-        instruction[-2] = OP_B_T3a | ((relative & 0x100000) >> 10) | ((cond & 0xf) << 6) | ((relative & 0x3f000) >> 12);
-        instruction[-1] = OP_B_T3b | ((relative & 0x80000) >> 8) | ((relative & 0x40000) >> 5) | ((relative & 0xffe) >> 1);
+        uint16_t instructions[2];
+        instructions[0] = OP_B_T3a | ((relative & 0x100000) >> 10) | ((cond & 0xf) << 6) | ((relative & 0x3f000) >> 12);
+        instructions[1] = OP_B_T3b | ((relative & 0x80000) >> 8) | ((relative & 0x40000) >> 5) | ((relative & 0xffe) >> 1);
+        performJITMemcpy(writeTarget - 2, instructions, 2 * sizeof(uint16_t));
     }
     
-    static void linkJumpT4(uint16_t* instruction, void* target)
+    static void linkJumpT4(uint16_t* writeTarget, const uint16_t* instruction, void* target)
     {
         // FIMXE: this should be up in the MacroAssembler layer. :-(        
         ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1));
@@ -2643,47 +2717,55 @@ private:
         
         // All branch offsets should be an even distance.
         ASSERT(!(relative & 1));
-        instruction[-2] = OP_B_T4a | ((relative & 0x1000000) >> 14) | ((relative & 0x3ff000) >> 12);
-        instruction[-1] = OP_B_T4b | ((relative & 0x800000) >> 10) | ((relative & 0x400000) >> 11) | ((relative & 0xffe) >> 1);
+        uint16_t instructions[2];
+        instructions[0] = OP_B_T4a | ((relative & 0x1000000) >> 14) | ((relative & 0x3ff000) >> 12);
+        instructions[1] = OP_B_T4b | ((relative & 0x800000) >> 10) | ((relative & 0x400000) >> 11) | ((relative & 0xffe) >> 1);
+        performJITMemcpy(writeTarget - 2, instructions, 2 * sizeof(uint16_t));
     }
     
-    void linkConditionalJumpT4(Condition cond, uint16_t* instruction, void* target)
+    static void linkConditionalJumpT4(Condition cond, uint16_t* writeTarget, const uint16_t* instruction, void* target)
     {
         // FIMXE: this should be up in the MacroAssembler layer. :-(        
         ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1));
         ASSERT(!(reinterpret_cast<intptr_t>(target) & 1));
         
-        instruction[-3] = ifThenElse(cond) | OP_IT;
-        linkJumpT4(instruction, target);
+        uint16_t newInstruction = ifThenElse(cond) | OP_IT;
+        performJITMemcpy(writeTarget - 3, &newInstruction, sizeof(uint16_t));
+        linkJumpT4(writeTarget, instruction, target);
     }
     
-    static void linkBX(uint16_t* instruction, void* target)
+    static void linkBX(uint16_t* writeTarget, const uint16_t* instruction, void* target)
     {
         // FIMXE: this should be up in the MacroAssembler layer. :-(
-        ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1));
+        ASSERT_UNUSED(instruction, !(reinterpret_cast<intptr_t>(instruction) & 1));
+        ASSERT(!(reinterpret_cast<intptr_t>(writeTarget) & 1));
         ASSERT(!(reinterpret_cast<intptr_t>(target) & 1));
         
         const uint16_t JUMP_TEMPORARY_REGISTER = ARMRegisters::ip;
         ARMThumbImmediate lo16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(reinterpret_cast<uint32_t>(target) + 1));
         ARMThumbImmediate hi16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(reinterpret_cast<uint32_t>(target) >> 16));
-        instruction[-5] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, lo16);
-        instruction[-4] = twoWordOp5i6Imm4Reg4EncodedImmSecond(JUMP_TEMPORARY_REGISTER, lo16);
-        instruction[-3] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOVT, hi16);
-        instruction[-2] = twoWordOp5i6Imm4Reg4EncodedImmSecond(JUMP_TEMPORARY_REGISTER, hi16);
-        instruction[-1] = OP_BX | (JUMP_TEMPORARY_REGISTER << 3);
+        uint16_t instructions[5];
+        instructions[0] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, lo16);
+        instructions[1] = twoWordOp5i6Imm4Reg4EncodedImmSecond(JUMP_TEMPORARY_REGISTER, lo16);
+        instructions[2] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOVT, hi16);
+        instructions[3] = twoWordOp5i6Imm4Reg4EncodedImmSecond(JUMP_TEMPORARY_REGISTER, hi16);
+        instructions[4] = OP_BX | (JUMP_TEMPORARY_REGISTER << 3);
+
+        performJITMemcpy(writeTarget - 5, instructions, 5 * sizeof(uint16_t));
     }
     
-    void linkConditionalBX(Condition cond, uint16_t* instruction, void* target)
+    static void linkConditionalBX(Condition cond, uint16_t* writeTarget, const uint16_t* instruction, void* target)
     {
         // FIMXE: this should be up in the MacroAssembler layer. :-(        
         ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1));
         ASSERT(!(reinterpret_cast<intptr_t>(target) & 1));
         
-        linkBX(instruction, target);
-        instruction[-6] = ifThenElse(cond, true, true) | OP_IT;
+        linkBX(writeTarget, instruction, target);
+        uint16_t newInstruction = ifThenElse(cond, true, true) | OP_IT;
+        performJITMemcpy(writeTarget - 6, &newInstruction, sizeof(uint16_t));
     }
     
-    static void linkJumpAbsolute(uint16_t* instruction, void* target)
+    static void linkJumpAbsolute(uint16_t* writeTarget, const uint16_t* instruction, void* target)
     {
         // FIMXE: this should be up in the MacroAssembler layer. :-(
         ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1));
@@ -2691,26 +2773,31 @@ private:
         
         ASSERT((isMOV_imm_T3(instruction - 5) && isMOVT(instruction - 3) && isBX(instruction - 1))
                || (isNOP_T1(instruction - 5) && isNOP_T2(instruction - 4) && isB(instruction - 2)));
-        
+
         if (canBeJumpT4(instruction, target)) {
             // There may be a better way to fix this, but right now put the NOPs first, since in the
             // case of an conditional branch this will be coming after an ITTT predicating *three*
             // instructions!  Looking backwards to modify the ITTT to an IT is not easy, due to
             // variable wdith encoding - the previous instruction might *look* like an ITTT but
             // actually be the second half of a 2-word op.
-            instruction[-5] = OP_NOP_T1;
-            instruction[-4] = OP_NOP_T2a;
-            instruction[-3] = OP_NOP_T2b;
-            linkJumpT4(instruction, target);
+            uint16_t instructions[3];
+            instructions[0] = OP_NOP_T1;
+            instructions[1] = OP_NOP_T2a;
+            instructions[2] = OP_NOP_T2b;
+            performJITMemcpy(writeTarget - 5, instructions, 3 * sizeof(uint16_t));
+            linkJumpT4(writeTarget, instruction, target);
         } else {
             const uint16_t JUMP_TEMPORARY_REGISTER = ARMRegisters::ip;
             ARMThumbImmediate lo16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(reinterpret_cast<uint32_t>(target) + 1));
             ARMThumbImmediate hi16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(reinterpret_cast<uint32_t>(target) >> 16));
-            instruction[-5] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, lo16);
-            instruction[-4] = twoWordOp5i6Imm4Reg4EncodedImmSecond(JUMP_TEMPORARY_REGISTER, lo16);
-            instruction[-3] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOVT, hi16);
-            instruction[-2] = twoWordOp5i6Imm4Reg4EncodedImmSecond(JUMP_TEMPORARY_REGISTER, hi16);
-            instruction[-1] = OP_BX | (JUMP_TEMPORARY_REGISTER << 3);
+
+            uint16_t instructions[5];
+            instructions[0] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, lo16);
+            instructions[1] = twoWordOp5i6Imm4Reg4EncodedImmSecond(JUMP_TEMPORARY_REGISTER, lo16);
+            instructions[2] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOVT, hi16);
+            instructions[3] = twoWordOp5i6Imm4Reg4EncodedImmSecond(JUMP_TEMPORARY_REGISTER, hi16);
+            instructions[4] = OP_BX | (JUMP_TEMPORARY_REGISTER << 3);
+            performJITMemcpy(writeTarget - 5, instructions, 5 * sizeof(uint16_t));
         }
     }
     
@@ -2753,6 +2840,11 @@ private:
             m_buffer.putShort(op | (reg1 << 6) | (reg2 << 3) | reg3);
         }
 
+        ALWAYS_INLINE void oneWordOp7Imm9(OpcodeID op, uint16_t imm)
+        {
+            m_buffer.putShort(op | imm);
+        }
+
         ALWAYS_INLINE void oneWordOp8Imm8(OpcodeID op, uint8_t imm)
         {
             m_buffer.putShort(op | imm);
@@ -2791,6 +2883,12 @@ private:
             m_buffer.putShort(op2);
         }
 
+        ALWAYS_INLINE void twoWordOp16Imm16(OpcodeID1 op1, uint16_t imm)
+        {
+            m_buffer.putShort(op1);
+            m_buffer.putShort(imm);
+        }
+        
         ALWAYS_INLINE void twoWordOp5i6Imm4Reg4EncodedImm(OpcodeID1 op, int imm4, RegisterID rd, ARMThumbImmediate imm)
         {
             ARMThumbImmediate newImm = imm;
@@ -2851,7 +2949,6 @@ private:
 
         unsigned debugOffset() { return m_buffer.debugOffset(); }
 
-    private:
         AssemblerBuffer m_buffer;
     } m_formatter;
 
@@ -2863,5 +2960,3 @@ private:
 } // namespace JSC
 
 #endif // ENABLE(ASSEMBLER) && CPU(ARM_THUMB2)
-
-#endif // ARMAssembler_h