Add qemu 2.4.0
[kvmfornfv.git] / qemu / tcg / mips / tcg-target.c
diff --git a/qemu/tcg/mips/tcg-target.c b/qemu/tcg/mips/tcg-target.c
new file mode 100644 (file)
index 0000000..e97980d
--- /dev/null
@@ -0,0 +1,1824 @@
+/*
+ * Tiny Code Generator for QEMU
+ *
+ * Copyright (c) 2008-2009 Arnaud Patard <arnaud.patard@rtp-net.org>
+ * Copyright (c) 2009 Aurelien Jarno <aurelien@aurel32.net>
+ * Based on i386/tcg-target.c - Copyright (c) 2008 Fabrice Bellard
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+#include "tcg-be-ldst.h"
+
+#ifdef HOST_WORDS_BIGENDIAN
+# define MIPS_BE  1
+#else
+# define MIPS_BE  0
+#endif
+
+#define LO_OFF    (MIPS_BE * 4)
+#define HI_OFF    (4 - LO_OFF)
+
+#ifndef NDEBUG
+static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
+    "zero",
+    "at",
+    "v0",
+    "v1",
+    "a0",
+    "a1",
+    "a2",
+    "a3",
+    "t0",
+    "t1",
+    "t2",
+    "t3",
+    "t4",
+    "t5",
+    "t6",
+    "t7",
+    "s0",
+    "s1",
+    "s2",
+    "s3",
+    "s4",
+    "s5",
+    "s6",
+    "s7",
+    "t8",
+    "t9",
+    "k0",
+    "k1",
+    "gp",
+    "sp",
+    "s8",
+    "ra",
+};
+#endif
+
+#define TCG_TMP0  TCG_REG_AT
+#define TCG_TMP1  TCG_REG_T9
+
+/* check if we really need so many registers :P */
+static const TCGReg tcg_target_reg_alloc_order[] = {
+    /* Call saved registers.  */
+    TCG_REG_S0,
+    TCG_REG_S1,
+    TCG_REG_S2,
+    TCG_REG_S3,
+    TCG_REG_S4,
+    TCG_REG_S5,
+    TCG_REG_S6,
+    TCG_REG_S7,
+    TCG_REG_S8,
+
+    /* Call clobbered registers.  */
+    TCG_REG_T0,
+    TCG_REG_T1,
+    TCG_REG_T2,
+    TCG_REG_T3,
+    TCG_REG_T4,
+    TCG_REG_T5,
+    TCG_REG_T6,
+    TCG_REG_T7,
+    TCG_REG_T8,
+    TCG_REG_T9,
+    TCG_REG_V1,
+    TCG_REG_V0,
+
+    /* Argument registers, opposite order of allocation.  */
+    TCG_REG_A3,
+    TCG_REG_A2,
+    TCG_REG_A1,
+    TCG_REG_A0,
+};
+
+static const TCGReg tcg_target_call_iarg_regs[4] = {
+    TCG_REG_A0,
+    TCG_REG_A1,
+    TCG_REG_A2,
+    TCG_REG_A3
+};
+
+static const TCGReg tcg_target_call_oarg_regs[2] = {
+    TCG_REG_V0,
+    TCG_REG_V1
+};
+
+static tcg_insn_unit *tb_ret_addr;
+
+static inline uint32_t reloc_pc16_val(tcg_insn_unit *pc, tcg_insn_unit *target)
+{
+    /* Let the compiler perform the right-shift as part of the arithmetic.  */
+    ptrdiff_t disp = target - (pc + 1);
+    assert(disp == (int16_t)disp);
+    return disp & 0xffff;
+}
+
+static inline void reloc_pc16(tcg_insn_unit *pc, tcg_insn_unit *target)
+{
+    *pc = deposit32(*pc, 0, 16, reloc_pc16_val(pc, target));
+}
+
+static inline uint32_t reloc_26_val(tcg_insn_unit *pc, tcg_insn_unit *target)
+{
+    assert((((uintptr_t)pc ^ (uintptr_t)target) & 0xf0000000) == 0);
+    return ((uintptr_t)target >> 2) & 0x3ffffff;
+}
+
+static inline void reloc_26(tcg_insn_unit *pc, tcg_insn_unit *target)
+{
+    *pc = deposit32(*pc, 0, 26, reloc_26_val(pc, target));
+}
+
+static void patch_reloc(tcg_insn_unit *code_ptr, int type,
+                        intptr_t value, intptr_t addend)
+{
+    assert(type == R_MIPS_PC16);
+    assert(addend == 0);
+    reloc_pc16(code_ptr, (tcg_insn_unit *)value);
+}
+
+#define TCG_CT_CONST_ZERO 0x100
+#define TCG_CT_CONST_U16  0x200    /* Unsigned 16-bit: 0 - 0xffff.  */
+#define TCG_CT_CONST_S16  0x400    /* Signed 16-bit: -32768 - 32767 */
+#define TCG_CT_CONST_P2M1 0x800    /* Power of 2 minus 1.  */
+#define TCG_CT_CONST_N16  0x1000   /* "Negatable" 16-bit: -32767 - 32767 */
+
+static inline bool is_p2m1(tcg_target_long val)
+{
+    return val && ((val + 1) & val) == 0;
+}
+
+/* parse target specific constraints */
+static int target_parse_constraint(TCGArgConstraint *ct, const char **pct_str)
+{
+    const char *ct_str;
+
+    ct_str = *pct_str;
+    switch(ct_str[0]) {
+    case 'r':
+        ct->ct |= TCG_CT_REG;
+        tcg_regset_set(ct->u.regs, 0xffffffff);
+        break;
+    case 'L': /* qemu_ld output arg constraint */
+        ct->ct |= TCG_CT_REG;
+        tcg_regset_set(ct->u.regs, 0xffffffff);
+        tcg_regset_reset_reg(ct->u.regs, TCG_REG_V0);
+        break;
+    case 'l': /* qemu_ld input arg constraint */
+        ct->ct |= TCG_CT_REG;
+        tcg_regset_set(ct->u.regs, 0xffffffff);
+        tcg_regset_reset_reg(ct->u.regs, TCG_REG_A0);
+#if defined(CONFIG_SOFTMMU)
+        if (TARGET_LONG_BITS == 64) {
+            tcg_regset_reset_reg(ct->u.regs, TCG_REG_A2);
+        }
+#endif
+        break;
+    case 'S': /* qemu_st constraint */
+        ct->ct |= TCG_CT_REG;
+        tcg_regset_set(ct->u.regs, 0xffffffff);
+        tcg_regset_reset_reg(ct->u.regs, TCG_REG_A0);
+#if defined(CONFIG_SOFTMMU)
+        if (TARGET_LONG_BITS == 32) {
+            tcg_regset_reset_reg(ct->u.regs, TCG_REG_A1);
+        } else {
+            tcg_regset_reset_reg(ct->u.regs, TCG_REG_A2);
+            tcg_regset_reset_reg(ct->u.regs, TCG_REG_A3);
+        }
+#endif
+        break;
+    case 'I':
+        ct->ct |= TCG_CT_CONST_U16;
+        break;
+    case 'J':
+        ct->ct |= TCG_CT_CONST_S16;
+        break;
+    case 'K':
+        ct->ct |= TCG_CT_CONST_P2M1;
+        break;
+    case 'N':
+        ct->ct |= TCG_CT_CONST_N16;
+        break;
+    case 'Z':
+        /* We are cheating a bit here, using the fact that the register
+           ZERO is also the register number 0. Hence there is no need
+           to check for const_args in each instruction. */
+        ct->ct |= TCG_CT_CONST_ZERO;
+        break;
+    default:
+        return -1;
+    }
+    ct_str++;
+    *pct_str = ct_str;
+    return 0;
+}
+
+/* test if a constant matches the constraint */
+static inline int tcg_target_const_match(tcg_target_long val, TCGType type,
+                                         const TCGArgConstraint *arg_ct)
+{
+    int ct;
+    ct = arg_ct->ct;
+    if (ct & TCG_CT_CONST) {
+        return 1;
+    } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
+        return 1;
+    } else if ((ct & TCG_CT_CONST_U16) && val == (uint16_t)val) {
+        return 1;
+    } else if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
+        return 1;
+    } else if ((ct & TCG_CT_CONST_N16) && val >= -32767 && val <= 32767) {
+        return 1;
+    } else if ((ct & TCG_CT_CONST_P2M1)
+               && use_mips32r2_instructions && is_p2m1(val)) {
+        return 1;
+    }
+    return 0;
+}
+
+/* instruction opcodes */
+typedef enum {
+    OPC_J        = 0x02 << 26,
+    OPC_JAL      = 0x03 << 26,
+    OPC_BEQ      = 0x04 << 26,
+    OPC_BNE      = 0x05 << 26,
+    OPC_BLEZ     = 0x06 << 26,
+    OPC_BGTZ     = 0x07 << 26,
+    OPC_ADDIU    = 0x09 << 26,
+    OPC_SLTI     = 0x0A << 26,
+    OPC_SLTIU    = 0x0B << 26,
+    OPC_ANDI     = 0x0C << 26,
+    OPC_ORI      = 0x0D << 26,
+    OPC_XORI     = 0x0E << 26,
+    OPC_LUI      = 0x0F << 26,
+    OPC_LB       = 0x20 << 26,
+    OPC_LH       = 0x21 << 26,
+    OPC_LW       = 0x23 << 26,
+    OPC_LBU      = 0x24 << 26,
+    OPC_LHU      = 0x25 << 26,
+    OPC_LWU      = 0x27 << 26,
+    OPC_SB       = 0x28 << 26,
+    OPC_SH       = 0x29 << 26,
+    OPC_SW       = 0x2B << 26,
+
+    OPC_SPECIAL  = 0x00 << 26,
+    OPC_SLL      = OPC_SPECIAL | 0x00,
+    OPC_SRL      = OPC_SPECIAL | 0x02,
+    OPC_ROTR     = OPC_SPECIAL | (0x01 << 21) | 0x02,
+    OPC_SRA      = OPC_SPECIAL | 0x03,
+    OPC_SLLV     = OPC_SPECIAL | 0x04,
+    OPC_SRLV     = OPC_SPECIAL | 0x06,
+    OPC_ROTRV    = OPC_SPECIAL | (0x01 <<  6) | 0x06,
+    OPC_SRAV     = OPC_SPECIAL | 0x07,
+    OPC_JR       = OPC_SPECIAL | 0x08,
+    OPC_JALR     = OPC_SPECIAL | 0x09,
+    OPC_MOVZ     = OPC_SPECIAL | 0x0A,
+    OPC_MOVN     = OPC_SPECIAL | 0x0B,
+    OPC_MFHI     = OPC_SPECIAL | 0x10,
+    OPC_MFLO     = OPC_SPECIAL | 0x12,
+    OPC_MULT     = OPC_SPECIAL | 0x18,
+    OPC_MULTU    = OPC_SPECIAL | 0x19,
+    OPC_DIV      = OPC_SPECIAL | 0x1A,
+    OPC_DIVU     = OPC_SPECIAL | 0x1B,
+    OPC_ADDU     = OPC_SPECIAL | 0x21,
+    OPC_SUBU     = OPC_SPECIAL | 0x23,
+    OPC_AND      = OPC_SPECIAL | 0x24,
+    OPC_OR       = OPC_SPECIAL | 0x25,
+    OPC_XOR      = OPC_SPECIAL | 0x26,
+    OPC_NOR      = OPC_SPECIAL | 0x27,
+    OPC_SLT      = OPC_SPECIAL | 0x2A,
+    OPC_SLTU     = OPC_SPECIAL | 0x2B,
+
+    OPC_REGIMM   = 0x01 << 26,
+    OPC_BLTZ     = OPC_REGIMM | (0x00 << 16),
+    OPC_BGEZ     = OPC_REGIMM | (0x01 << 16),
+
+    OPC_SPECIAL2 = 0x1c << 26,
+    OPC_MUL      = OPC_SPECIAL2 | 0x002,
+
+    OPC_SPECIAL3 = 0x1f << 26,
+    OPC_EXT      = OPC_SPECIAL3 | 0x000,
+    OPC_INS      = OPC_SPECIAL3 | 0x004,
+    OPC_WSBH     = OPC_SPECIAL3 | 0x0a0,
+    OPC_SEB      = OPC_SPECIAL3 | 0x420,
+    OPC_SEH      = OPC_SPECIAL3 | 0x620,
+} MIPSInsn;
+
+/*
+ * Type reg
+ */
+static inline void tcg_out_opc_reg(TCGContext *s, MIPSInsn opc,
+                                   TCGReg rd, TCGReg rs, TCGReg rt)
+{
+    int32_t inst;
+
+    inst = opc;
+    inst |= (rs & 0x1F) << 21;
+    inst |= (rt & 0x1F) << 16;
+    inst |= (rd & 0x1F) << 11;
+    tcg_out32(s, inst);
+}
+
+/*
+ * Type immediate
+ */
+static inline void tcg_out_opc_imm(TCGContext *s, MIPSInsn opc,
+                                   TCGReg rt, TCGReg rs, TCGArg imm)
+{
+    int32_t inst;
+
+    inst = opc;
+    inst |= (rs & 0x1F) << 21;
+    inst |= (rt & 0x1F) << 16;
+    inst |= (imm & 0xffff);
+    tcg_out32(s, inst);
+}
+
+/*
+ * Type bitfield
+ */
+static inline void tcg_out_opc_bf(TCGContext *s, MIPSInsn opc, TCGReg rt,
+                                  TCGReg rs, int msb, int lsb)
+{
+    int32_t inst;
+
+    inst = opc;
+    inst |= (rs & 0x1F) << 21;
+    inst |= (rt & 0x1F) << 16;
+    inst |= (msb & 0x1F) << 11;
+    inst |= (lsb & 0x1F) << 6;
+    tcg_out32(s, inst);
+}
+
+/*
+ * Type branch
+ */
+static inline void tcg_out_opc_br(TCGContext *s, MIPSInsn opc,
+                                  TCGReg rt, TCGReg rs)
+{
+    /* We pay attention here to not modify the branch target by reading
+       the existing value and using it again. This ensure that caches and
+       memory are kept coherent during retranslation. */
+    uint16_t offset = (uint16_t)*s->code_ptr;
+
+    tcg_out_opc_imm(s, opc, rt, rs, offset);
+}
+
+/*
+ * Type sa
+ */
+static inline void tcg_out_opc_sa(TCGContext *s, MIPSInsn opc,
+                                  TCGReg rd, TCGReg rt, TCGArg sa)
+{
+    int32_t inst;
+
+    inst = opc;
+    inst |= (rt & 0x1F) << 16;
+    inst |= (rd & 0x1F) << 11;
+    inst |= (sa & 0x1F) <<  6;
+    tcg_out32(s, inst);
+
+}
+
+/*
+ * Type jump.
+ * Returns true if the branch was in range and the insn was emitted.
+ */
+static bool tcg_out_opc_jmp(TCGContext *s, MIPSInsn opc, void *target)
+{
+    uintptr_t dest = (uintptr_t)target;
+    uintptr_t from = (uintptr_t)s->code_ptr + 4;
+    int32_t inst;
+
+    /* The pc-region branch happens within the 256MB region of
+       the delay slot (thus the +4).  */
+    if ((from ^ dest) & -(1 << 28)) {
+        return false;
+    }
+    assert((dest & 3) == 0);
+
+    inst = opc;
+    inst |= (dest >> 2) & 0x3ffffff;
+    tcg_out32(s, inst);
+    return true;
+}
+
+static inline void tcg_out_nop(TCGContext *s)
+{
+    tcg_out32(s, 0);
+}
+
+static inline void tcg_out_mov(TCGContext *s, TCGType type,
+                               TCGReg ret, TCGReg arg)
+{
+    /* Simple reg-reg move, optimising out the 'do nothing' case */
+    if (ret != arg) {
+        tcg_out_opc_reg(s, OPC_ADDU, ret, arg, TCG_REG_ZERO);
+    }
+}
+
+static inline void tcg_out_movi(TCGContext *s, TCGType type,
+                                TCGReg reg, tcg_target_long arg)
+{
+    if (arg == (int16_t)arg) {
+        tcg_out_opc_imm(s, OPC_ADDIU, reg, TCG_REG_ZERO, arg);
+    } else if (arg == (uint16_t)arg) {
+        tcg_out_opc_imm(s, OPC_ORI, reg, TCG_REG_ZERO, arg);
+    } else {
+        tcg_out_opc_imm(s, OPC_LUI, reg, TCG_REG_ZERO, arg >> 16);
+        if (arg & 0xffff) {
+            tcg_out_opc_imm(s, OPC_ORI, reg, reg, arg & 0xffff);
+        }
+    }
+}
+
+static inline void tcg_out_bswap16(TCGContext *s, TCGReg ret, TCGReg arg)
+{
+    if (use_mips32r2_instructions) {
+        tcg_out_opc_reg(s, OPC_WSBH, ret, 0, arg);
+    } else {
+        /* ret and arg can't be register at */
+        if (ret == TCG_TMP0 || arg == TCG_TMP0) {
+            tcg_abort();
+        }
+
+        tcg_out_opc_sa(s, OPC_SRL, TCG_TMP0, arg, 8);
+        tcg_out_opc_sa(s, OPC_SLL, ret, arg, 8);
+        tcg_out_opc_imm(s, OPC_ANDI, ret, ret, 0xff00);
+        tcg_out_opc_reg(s, OPC_OR, ret, ret, TCG_TMP0);
+    }
+}
+
+static inline void tcg_out_bswap16s(TCGContext *s, TCGReg ret, TCGReg arg)
+{
+    if (use_mips32r2_instructions) {
+        tcg_out_opc_reg(s, OPC_WSBH, ret, 0, arg);
+        tcg_out_opc_reg(s, OPC_SEH, ret, 0, ret);
+    } else {
+        /* ret and arg can't be register at */
+        if (ret == TCG_TMP0 || arg == TCG_TMP0) {
+            tcg_abort();
+        }
+
+        tcg_out_opc_sa(s, OPC_SRL, TCG_TMP0, arg, 8);
+        tcg_out_opc_sa(s, OPC_SLL, ret, arg, 24);
+        tcg_out_opc_sa(s, OPC_SRA, ret, ret, 16);
+        tcg_out_opc_reg(s, OPC_OR, ret, ret, TCG_TMP0);
+    }
+}
+
+static inline void tcg_out_bswap32(TCGContext *s, TCGReg ret, TCGReg arg)
+{
+    if (use_mips32r2_instructions) {
+        tcg_out_opc_reg(s, OPC_WSBH, ret, 0, arg);
+        tcg_out_opc_sa(s, OPC_ROTR, ret, ret, 16);
+    } else {
+        /* ret and arg must be different and can't be register at */
+        if (ret == arg || ret == TCG_TMP0 || arg == TCG_TMP0) {
+            tcg_abort();
+        }
+
+        tcg_out_opc_sa(s, OPC_SLL, ret, arg, 24);
+
+        tcg_out_opc_sa(s, OPC_SRL, TCG_TMP0, arg, 24);
+        tcg_out_opc_reg(s, OPC_OR, ret, ret, TCG_TMP0);
+
+        tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP0, arg, 0xff00);
+        tcg_out_opc_sa(s, OPC_SLL, TCG_TMP0, TCG_TMP0, 8);
+        tcg_out_opc_reg(s, OPC_OR, ret, ret, TCG_TMP0);
+
+        tcg_out_opc_sa(s, OPC_SRL, TCG_TMP0, arg, 8);
+        tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP0, TCG_TMP0, 0xff00);
+        tcg_out_opc_reg(s, OPC_OR, ret, ret, TCG_TMP0);
+    }
+}
+
+static inline void tcg_out_ext8s(TCGContext *s, TCGReg ret, TCGReg arg)
+{
+    if (use_mips32r2_instructions) {
+        tcg_out_opc_reg(s, OPC_SEB, ret, 0, arg);
+    } else {
+        tcg_out_opc_sa(s, OPC_SLL, ret, arg, 24);
+        tcg_out_opc_sa(s, OPC_SRA, ret, ret, 24);
+    }
+}
+
+static inline void tcg_out_ext16s(TCGContext *s, TCGReg ret, TCGReg arg)
+{
+    if (use_mips32r2_instructions) {
+        tcg_out_opc_reg(s, OPC_SEH, ret, 0, arg);
+    } else {
+        tcg_out_opc_sa(s, OPC_SLL, ret, arg, 16);
+        tcg_out_opc_sa(s, OPC_SRA, ret, ret, 16);
+    }
+}
+
+static void tcg_out_ldst(TCGContext *s, MIPSInsn opc, TCGReg data,
+                         TCGReg addr, intptr_t ofs)
+{
+    int16_t lo = ofs;
+    if (ofs != lo) {
+        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs - lo);
+        if (addr != TCG_REG_ZERO) {
+            tcg_out_opc_reg(s, OPC_ADDU, TCG_TMP0, TCG_TMP0, addr);
+        }
+        addr = TCG_TMP0;
+    }
+    tcg_out_opc_imm(s, opc, data, addr, lo);
+}
+
+static inline void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg,
+                              TCGReg arg1, intptr_t arg2)
+{
+    tcg_out_ldst(s, OPC_LW, arg, arg1, arg2);
+}
+
+static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
+                              TCGReg arg1, intptr_t arg2)
+{
+    tcg_out_ldst(s, OPC_SW, arg, arg1, arg2);
+}
+
+static inline void tcg_out_addi(TCGContext *s, TCGReg reg, TCGArg val)
+{
+    if (val == (int16_t)val) {
+        tcg_out_opc_imm(s, OPC_ADDIU, reg, reg, val);
+    } else {
+        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, val);
+        tcg_out_opc_reg(s, OPC_ADDU, reg, reg, TCG_TMP0);
+    }
+}
+
+/* Bit 0 set if inversion required; bit 1 set if swapping required.  */
+#define MIPS_CMP_INV  1
+#define MIPS_CMP_SWAP 2
+
+static const uint8_t mips_cmp_map[16] = {
+    [TCG_COND_LT]  = 0,
+    [TCG_COND_LTU] = 0,
+    [TCG_COND_GE]  = MIPS_CMP_INV,
+    [TCG_COND_GEU] = MIPS_CMP_INV,
+    [TCG_COND_LE]  = MIPS_CMP_INV | MIPS_CMP_SWAP,
+    [TCG_COND_LEU] = MIPS_CMP_INV | MIPS_CMP_SWAP,
+    [TCG_COND_GT]  = MIPS_CMP_SWAP,
+    [TCG_COND_GTU] = MIPS_CMP_SWAP,
+};
+
+static void tcg_out_setcond(TCGContext *s, TCGCond cond, TCGReg ret,
+                            TCGReg arg1, TCGReg arg2)
+{
+    MIPSInsn s_opc = OPC_SLTU;
+    int cmp_map;
+
+    switch (cond) {
+    case TCG_COND_EQ:
+        if (arg2 != 0) {
+            tcg_out_opc_reg(s, OPC_XOR, ret, arg1, arg2);
+            arg1 = ret;
+        }
+        tcg_out_opc_imm(s, OPC_SLTIU, ret, arg1, 1);
+        break;
+
+    case TCG_COND_NE:
+        if (arg2 != 0) {
+            tcg_out_opc_reg(s, OPC_XOR, ret, arg1, arg2);
+            arg1 = ret;
+        }
+        tcg_out_opc_reg(s, OPC_SLTU, ret, TCG_REG_ZERO, arg1);
+        break;
+
+    case TCG_COND_LT:
+    case TCG_COND_GE:
+    case TCG_COND_LE:
+    case TCG_COND_GT:
+        s_opc = OPC_SLT;
+        /* FALLTHRU */
+
+    case TCG_COND_LTU:
+    case TCG_COND_GEU:
+    case TCG_COND_LEU:
+    case TCG_COND_GTU:
+        cmp_map = mips_cmp_map[cond];
+        if (cmp_map & MIPS_CMP_SWAP) {
+            TCGReg t = arg1;
+            arg1 = arg2;
+            arg2 = t;
+        }
+        tcg_out_opc_reg(s, s_opc, ret, arg1, arg2);
+        if (cmp_map & MIPS_CMP_INV) {
+            tcg_out_opc_imm(s, OPC_XORI, ret, ret, 1);
+        }
+        break;
+
+     default:
+         tcg_abort();
+         break;
+     }
+}
+
+static void tcg_out_brcond(TCGContext *s, TCGCond cond, TCGReg arg1,
+                           TCGReg arg2, TCGLabel *l)
+{
+    static const MIPSInsn b_zero[16] = {
+        [TCG_COND_LT] = OPC_BLTZ,
+        [TCG_COND_GT] = OPC_BGTZ,
+        [TCG_COND_LE] = OPC_BLEZ,
+        [TCG_COND_GE] = OPC_BGEZ,
+    };
+
+    MIPSInsn s_opc = OPC_SLTU;
+    MIPSInsn b_opc;
+    int cmp_map;
+
+    switch (cond) {
+    case TCG_COND_EQ:
+        b_opc = OPC_BEQ;
+        break;
+    case TCG_COND_NE:
+        b_opc = OPC_BNE;
+        break;
+
+    case TCG_COND_LT:
+    case TCG_COND_GT:
+    case TCG_COND_LE:
+    case TCG_COND_GE:
+        if (arg2 == 0) {
+            b_opc = b_zero[cond];
+            arg2 = arg1;
+            arg1 = 0;
+            break;
+        }
+        s_opc = OPC_SLT;
+        /* FALLTHRU */
+
+    case TCG_COND_LTU:
+    case TCG_COND_GTU:
+    case TCG_COND_LEU:
+    case TCG_COND_GEU:
+        cmp_map = mips_cmp_map[cond];
+        if (cmp_map & MIPS_CMP_SWAP) {
+            TCGReg t = arg1;
+            arg1 = arg2;
+            arg2 = t;
+        }
+        tcg_out_opc_reg(s, s_opc, TCG_TMP0, arg1, arg2);
+        b_opc = (cmp_map & MIPS_CMP_INV ? OPC_BEQ : OPC_BNE);
+        arg1 = TCG_TMP0;
+        arg2 = TCG_REG_ZERO;
+        break;
+
+    default:
+        tcg_abort();
+        break;
+    }
+
+    tcg_out_opc_br(s, b_opc, arg1, arg2);
+    if (l->has_value) {
+        reloc_pc16(s->code_ptr - 1, l->u.value_ptr);
+    } else {
+        tcg_out_reloc(s, s->code_ptr - 1, R_MIPS_PC16, l, 0);
+    }
+    tcg_out_nop(s);
+}
+
+static TCGReg tcg_out_reduce_eq2(TCGContext *s, TCGReg tmp0, TCGReg tmp1,
+                                 TCGReg al, TCGReg ah,
+                                 TCGReg bl, TCGReg bh)
+{
+    /* Merge highpart comparison into AH.  */
+    if (bh != 0) {
+        if (ah != 0) {
+            tcg_out_opc_reg(s, OPC_XOR, tmp0, ah, bh);
+            ah = tmp0;
+        } else {
+            ah = bh;
+        }
+    }
+    /* Merge lowpart comparison into AL.  */
+    if (bl != 0) {
+        if (al != 0) {
+            tcg_out_opc_reg(s, OPC_XOR, tmp1, al, bl);
+            al = tmp1;
+        } else {
+            al = bl;
+        }
+    }
+    /* Merge high and low part comparisons into AL.  */
+    if (ah != 0) {
+        if (al != 0) {
+            tcg_out_opc_reg(s, OPC_OR, tmp0, ah, al);
+            al = tmp0;
+        } else {
+            al = ah;
+        }
+    }
+    return al;
+}
+
+static void tcg_out_setcond2(TCGContext *s, TCGCond cond, TCGReg ret,
+                             TCGReg al, TCGReg ah, TCGReg bl, TCGReg bh)
+{
+    TCGReg tmp0 = TCG_TMP0;
+    TCGReg tmp1 = ret;
+
+    assert(ret != TCG_TMP0);
+    if (ret == ah || ret == bh) {
+        assert(ret != TCG_TMP1);
+        tmp1 = TCG_TMP1;
+    }
+
+    switch (cond) {
+    case TCG_COND_EQ:
+    case TCG_COND_NE:
+        tmp1 = tcg_out_reduce_eq2(s, tmp0, tmp1, al, ah, bl, bh);
+        tcg_out_setcond(s, cond, ret, tmp1, TCG_REG_ZERO);
+        break;
+
+    default:
+        tcg_out_setcond(s, TCG_COND_EQ, tmp0, ah, bh);
+        tcg_out_setcond(s, tcg_unsigned_cond(cond), tmp1, al, bl);
+        tcg_out_opc_reg(s, OPC_AND, tmp1, tmp1, tmp0);
+        tcg_out_setcond(s, tcg_high_cond(cond), tmp0, ah, bh);
+        tcg_out_opc_reg(s, OPC_OR, ret, tmp1, tmp0);
+        break;
+    }
+}
+
+static void tcg_out_brcond2(TCGContext *s, TCGCond cond, TCGReg al, TCGReg ah,
+                            TCGReg bl, TCGReg bh, TCGLabel *l)
+{
+    TCGCond b_cond = TCG_COND_NE;
+    TCGReg tmp = TCG_TMP1;
+
+    /* With branches, we emit between 4 and 9 insns with 2 or 3 branches.
+       With setcond, we emit between 3 and 10 insns and only 1 branch,
+       which ought to get better branch prediction.  */
+     switch (cond) {
+     case TCG_COND_EQ:
+     case TCG_COND_NE:
+        b_cond = cond;
+        tmp = tcg_out_reduce_eq2(s, TCG_TMP0, TCG_TMP1, al, ah, bl, bh);
+        break;
+
+    default:
+        /* Minimize code size by preferring a compare not requiring INV.  */
+        if (mips_cmp_map[cond] & MIPS_CMP_INV) {
+            cond = tcg_invert_cond(cond);
+            b_cond = TCG_COND_EQ;
+        }
+        tcg_out_setcond2(s, cond, tmp, al, ah, bl, bh);
+        break;
+    }
+
+    tcg_out_brcond(s, b_cond, tmp, TCG_REG_ZERO, l);
+}
+
+static void tcg_out_movcond(TCGContext *s, TCGCond cond, TCGReg ret,
+                            TCGReg c1, TCGReg c2, TCGReg v)
+{
+    MIPSInsn m_opc = OPC_MOVN;
+
+    switch (cond) {
+    case TCG_COND_EQ:
+        m_opc = OPC_MOVZ;
+        /* FALLTHRU */
+    case TCG_COND_NE:
+        if (c2 != 0) {
+            tcg_out_opc_reg(s, OPC_XOR, TCG_TMP0, c1, c2);
+            c1 = TCG_TMP0;
+        }
+        break;
+
+    default:
+        /* Minimize code size by preferring a compare not requiring INV.  */
+        if (mips_cmp_map[cond] & MIPS_CMP_INV) {
+            cond = tcg_invert_cond(cond);
+            m_opc = OPC_MOVZ;
+        }
+        tcg_out_setcond(s, cond, TCG_TMP0, c1, c2);
+        c1 = TCG_TMP0;
+        break;
+    }
+
+    tcg_out_opc_reg(s, m_opc, ret, v, c1);
+}
+
+static void tcg_out_call_int(TCGContext *s, tcg_insn_unit *arg, bool tail)
+{
+    /* Note that the ABI requires the called function's address to be
+       loaded into T9, even if a direct branch is in range.  */
+    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T9, (uintptr_t)arg);
+
+    /* But do try a direct branch, allowing the cpu better insn prefetch.  */
+    if (tail) {
+        if (!tcg_out_opc_jmp(s, OPC_J, arg)) {
+            tcg_out_opc_reg(s, OPC_JR, 0, TCG_REG_T9, 0);
+        }
+    } else {
+        if (!tcg_out_opc_jmp(s, OPC_JAL, arg)) {
+            tcg_out_opc_reg(s, OPC_JALR, TCG_REG_RA, TCG_REG_T9, 0);
+        }
+    }
+}
+
+static void tcg_out_call(TCGContext *s, tcg_insn_unit *arg)
+{
+    tcg_out_call_int(s, arg, false);
+    tcg_out_nop(s);
+}
+
+#if defined(CONFIG_SOFTMMU)
+static void * const qemu_ld_helpers[16] = {
+    [MO_UB]   = helper_ret_ldub_mmu,
+    [MO_SB]   = helper_ret_ldsb_mmu,
+    [MO_LEUW] = helper_le_lduw_mmu,
+    [MO_LESW] = helper_le_ldsw_mmu,
+    [MO_LEUL] = helper_le_ldul_mmu,
+    [MO_LEQ]  = helper_le_ldq_mmu,
+    [MO_BEUW] = helper_be_lduw_mmu,
+    [MO_BESW] = helper_be_ldsw_mmu,
+    [MO_BEUL] = helper_be_ldul_mmu,
+    [MO_BEQ]  = helper_be_ldq_mmu,
+};
+
+static void * const qemu_st_helpers[16] = {
+    [MO_UB]   = helper_ret_stb_mmu,
+    [MO_LEUW] = helper_le_stw_mmu,
+    [MO_LEUL] = helper_le_stl_mmu,
+    [MO_LEQ]  = helper_le_stq_mmu,
+    [MO_BEUW] = helper_be_stw_mmu,
+    [MO_BEUL] = helper_be_stl_mmu,
+    [MO_BEQ]  = helper_be_stq_mmu,
+};
+
+/* Helper routines for marshalling helper function arguments into
+ * the correct registers and stack.
+ * I is where we want to put this argument, and is updated and returned
+ * for the next call. ARG is the argument itself.
+ *
+ * We provide routines for arguments which are: immediate, 32 bit
+ * value in register, 16 and 8 bit values in register (which must be zero
+ * extended before use) and 64 bit value in a lo:hi register pair.
+ */
+
+static int tcg_out_call_iarg_reg(TCGContext *s, int i, TCGReg arg)
+{
+    if (i < ARRAY_SIZE(tcg_target_call_iarg_regs)) {
+        tcg_out_mov(s, TCG_TYPE_REG, tcg_target_call_iarg_regs[i], arg);
+    } else {
+        tcg_out_st(s, TCG_TYPE_REG, arg, TCG_REG_SP, 4 * i);
+    }
+    return i + 1;
+}
+
+static int tcg_out_call_iarg_reg8(TCGContext *s, int i, TCGReg arg)
+{
+    TCGReg tmp = TCG_TMP0;
+    if (i < ARRAY_SIZE(tcg_target_call_iarg_regs)) {
+        tmp = tcg_target_call_iarg_regs[i];
+    }
+    tcg_out_opc_imm(s, OPC_ANDI, tmp, arg, 0xff);
+    return tcg_out_call_iarg_reg(s, i, tmp);
+}
+
+static int tcg_out_call_iarg_reg16(TCGContext *s, int i, TCGReg arg)
+{
+    TCGReg tmp = TCG_TMP0;
+    if (i < ARRAY_SIZE(tcg_target_call_iarg_regs)) {
+        tmp = tcg_target_call_iarg_regs[i];
+    }
+    tcg_out_opc_imm(s, OPC_ANDI, tmp, arg, 0xffff);
+    return tcg_out_call_iarg_reg(s, i, tmp);
+}
+
+static int tcg_out_call_iarg_imm(TCGContext *s, int i, TCGArg arg)
+{
+    TCGReg tmp = TCG_TMP0;
+    if (arg == 0) {
+        tmp = TCG_REG_ZERO;
+    } else {
+        if (i < ARRAY_SIZE(tcg_target_call_iarg_regs)) {
+            tmp = tcg_target_call_iarg_regs[i];
+        }
+        tcg_out_movi(s, TCG_TYPE_REG, tmp, arg);
+    }
+    return tcg_out_call_iarg_reg(s, i, tmp);
+}
+
+static int tcg_out_call_iarg_reg2(TCGContext *s, int i, TCGReg al, TCGReg ah)
+{
+    i = (i + 1) & ~1;
+    i = tcg_out_call_iarg_reg(s, i, (MIPS_BE ? ah : al));
+    i = tcg_out_call_iarg_reg(s, i, (MIPS_BE ? al : ah));
+    return i;
+}
+
+/* Perform the tlb comparison operation.  The complete host address is
+   placed in BASE.  Clobbers AT, T0, A0.  */
+static void tcg_out_tlb_load(TCGContext *s, TCGReg base, TCGReg addrl,
+                             TCGReg addrh, int mem_index, TCGMemOp s_bits,
+                             tcg_insn_unit *label_ptr[2], bool is_load)
+{
+    int cmp_off
+        = (is_load
+           ? offsetof(CPUArchState, tlb_table[mem_index][0].addr_read)
+           : offsetof(CPUArchState, tlb_table[mem_index][0].addr_write));
+    int add_off = offsetof(CPUArchState, tlb_table[mem_index][0].addend);
+
+    tcg_out_opc_sa(s, OPC_SRL, TCG_REG_A0, addrl,
+                   TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
+    tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_A0, TCG_REG_A0,
+                    (CPU_TLB_SIZE - 1) << CPU_TLB_ENTRY_BITS);
+    tcg_out_opc_reg(s, OPC_ADDU, TCG_REG_A0, TCG_REG_A0, TCG_AREG0);
+
+    /* Compensate for very large offsets.  */
+    if (add_off >= 0x8000) {
+        /* Most target env are smaller than 32k; none are larger than 64k.
+           Simplify the logic here merely to offset by 0x7ff0, giving us a
+           range just shy of 64k.  Check this assumption.  */
+        QEMU_BUILD_BUG_ON(offsetof(CPUArchState,
+                                   tlb_table[NB_MMU_MODES - 1][1])
+                          > 0x7ff0 + 0x7fff);
+        tcg_out_opc_imm(s, OPC_ADDIU, TCG_REG_A0, TCG_REG_A0, 0x7ff0);
+        cmp_off -= 0x7ff0;
+        add_off -= 0x7ff0;
+    }
+
+    /* Load the tlb comparator.  */
+    if (TARGET_LONG_BITS == 64) {
+        tcg_out_opc_imm(s, OPC_LW, TCG_TMP0, TCG_REG_A0, cmp_off + LO_OFF);
+        tcg_out_opc_imm(s, OPC_LW, base, TCG_REG_A0, cmp_off + HI_OFF);
+    } else {
+        tcg_out_opc_imm(s, OPC_LW, TCG_TMP0, TCG_REG_A0, cmp_off);
+    }
+
+    /* Mask the page bits, keeping the alignment bits to compare against.
+       In between, load the tlb addend for the fast path.  */
+    tcg_out_movi(s, TCG_TYPE_I32, TCG_TMP1,
+                 TARGET_PAGE_MASK | ((1 << s_bits) - 1));
+    tcg_out_opc_imm(s, OPC_LW, TCG_REG_A0, TCG_REG_A0, add_off);
+    tcg_out_opc_reg(s, OPC_AND, TCG_TMP1, TCG_TMP1, addrl);
+
+    label_ptr[0] = s->code_ptr;
+    tcg_out_opc_br(s, OPC_BNE, TCG_TMP1, TCG_TMP0);
+
+    if (TARGET_LONG_BITS == 64) {
+        /* delay slot */
+        tcg_out_nop(s);
+
+        label_ptr[1] = s->code_ptr;
+        tcg_out_opc_br(s, OPC_BNE, addrh, base);
+    }
+
+    /* delay slot */
+    tcg_out_opc_reg(s, OPC_ADDU, base, TCG_REG_A0, addrl);
+}
+
+static void add_qemu_ldst_label(TCGContext *s, int is_ld, TCGMemOpIdx oi,
+                                TCGReg datalo, TCGReg datahi,
+                                TCGReg addrlo, TCGReg addrhi,
+                                void *raddr, tcg_insn_unit *label_ptr[2])
+{
+    TCGLabelQemuLdst *label = new_ldst_label(s);
+
+    label->is_ld = is_ld;
+    label->oi = oi;
+    label->datalo_reg = datalo;
+    label->datahi_reg = datahi;
+    label->addrlo_reg = addrlo;
+    label->addrhi_reg = addrhi;
+    label->raddr = raddr;
+    label->label_ptr[0] = label_ptr[0];
+    if (TARGET_LONG_BITS == 64) {
+        label->label_ptr[1] = label_ptr[1];
+    }
+}
+
+static void tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
+{
+    TCGMemOpIdx oi = l->oi;
+    TCGMemOp opc = get_memop(oi);
+    TCGReg v0;
+    int i;
+
+    /* resolve label address */
+    reloc_pc16(l->label_ptr[0], s->code_ptr);
+    if (TARGET_LONG_BITS == 64) {
+        reloc_pc16(l->label_ptr[1], s->code_ptr);
+    }
+
+    i = 1;
+    if (TARGET_LONG_BITS == 64) {
+        i = tcg_out_call_iarg_reg2(s, i, l->addrlo_reg, l->addrhi_reg);
+    } else {
+        i = tcg_out_call_iarg_reg(s, i, l->addrlo_reg);
+    }
+    i = tcg_out_call_iarg_imm(s, i, oi);
+    i = tcg_out_call_iarg_imm(s, i, (intptr_t)l->raddr);
+    tcg_out_call_int(s, qemu_ld_helpers[opc & (MO_BSWAP | MO_SSIZE)], false);
+    /* delay slot */
+    tcg_out_mov(s, TCG_TYPE_PTR, tcg_target_call_iarg_regs[0], TCG_AREG0);
+
+    v0 = l->datalo_reg;
+    if ((opc & MO_SIZE) == MO_64) {
+        /* We eliminated V0 from the possible output registers, so it
+           cannot be clobbered here.  So we must move V1 first.  */
+        if (MIPS_BE) {
+            tcg_out_mov(s, TCG_TYPE_I32, v0, TCG_REG_V1);
+            v0 = l->datahi_reg;
+        } else {
+            tcg_out_mov(s, TCG_TYPE_I32, l->datahi_reg, TCG_REG_V1);
+        }
+    }
+
+    reloc_pc16(s->code_ptr, l->raddr);
+    tcg_out_opc_br(s, OPC_BEQ, TCG_REG_ZERO, TCG_REG_ZERO);
+    /* delay slot */
+    tcg_out_mov(s, TCG_TYPE_REG, v0, TCG_REG_V0);
+}
+
+static void tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
+{
+    TCGMemOpIdx oi = l->oi;
+    TCGMemOp opc = get_memop(oi);
+    TCGMemOp s_bits = opc & MO_SIZE;
+    int i;
+
+    /* resolve label address */
+    reloc_pc16(l->label_ptr[0], s->code_ptr);
+    if (TARGET_LONG_BITS == 64) {
+        reloc_pc16(l->label_ptr[1], s->code_ptr);
+    }
+
+    i = 1;
+    if (TARGET_LONG_BITS == 64) {
+        i = tcg_out_call_iarg_reg2(s, i, l->addrlo_reg, l->addrhi_reg);
+    } else {
+        i = tcg_out_call_iarg_reg(s, i, l->addrlo_reg);
+    }
+    switch (s_bits) {
+    case MO_8:
+        i = tcg_out_call_iarg_reg8(s, i, l->datalo_reg);
+        break;
+    case MO_16:
+        i = tcg_out_call_iarg_reg16(s, i, l->datalo_reg);
+        break;
+    case MO_32:
+        i = tcg_out_call_iarg_reg(s, i, l->datalo_reg);
+        break;
+    case MO_64:
+        i = tcg_out_call_iarg_reg2(s, i, l->datalo_reg, l->datahi_reg);
+        break;
+    default:
+        tcg_abort();
+    }
+    i = tcg_out_call_iarg_imm(s, i, oi);
+
+    /* Tail call to the store helper.  Thus force the return address
+       computation to take place in the return address register.  */
+    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_RA, (intptr_t)l->raddr);
+    i = tcg_out_call_iarg_reg(s, i, TCG_REG_RA);
+    tcg_out_call_int(s, qemu_st_helpers[opc & (MO_BSWAP | MO_SIZE)], true);
+    /* delay slot */
+    tcg_out_mov(s, TCG_TYPE_PTR, tcg_target_call_iarg_regs[0], TCG_AREG0);
+}
+#endif
+
+static void tcg_out_qemu_ld_direct(TCGContext *s, TCGReg datalo, TCGReg datahi,
+                                   TCGReg base, TCGMemOp opc)
+{
+    switch (opc & (MO_SSIZE | MO_BSWAP)) {
+    case MO_UB:
+        tcg_out_opc_imm(s, OPC_LBU, datalo, base, 0);
+        break;
+    case MO_SB:
+        tcg_out_opc_imm(s, OPC_LB, datalo, base, 0);
+        break;
+    case MO_UW | MO_BSWAP:
+        tcg_out_opc_imm(s, OPC_LHU, TCG_TMP1, base, 0);
+        tcg_out_bswap16(s, datalo, TCG_TMP1);
+        break;
+    case MO_UW:
+        tcg_out_opc_imm(s, OPC_LHU, datalo, base, 0);
+        break;
+    case MO_SW | MO_BSWAP:
+        tcg_out_opc_imm(s, OPC_LHU, TCG_TMP1, base, 0);
+        tcg_out_bswap16s(s, datalo, TCG_TMP1);
+        break;
+    case MO_SW:
+        tcg_out_opc_imm(s, OPC_LH, datalo, base, 0);
+        break;
+    case MO_UL | MO_BSWAP:
+        tcg_out_opc_imm(s, OPC_LW, TCG_TMP1, base, 0);
+        tcg_out_bswap32(s, datalo, TCG_TMP1);
+        break;
+    case MO_UL:
+        tcg_out_opc_imm(s, OPC_LW, datalo, base, 0);
+        break;
+    case MO_Q | MO_BSWAP:
+        tcg_out_opc_imm(s, OPC_LW, TCG_TMP1, base, HI_OFF);
+        tcg_out_bswap32(s, datalo, TCG_TMP1);
+        tcg_out_opc_imm(s, OPC_LW, TCG_TMP1, base, LO_OFF);
+        tcg_out_bswap32(s, datahi, TCG_TMP1);
+        break;
+    case MO_Q:
+        tcg_out_opc_imm(s, OPC_LW, datalo, base, LO_OFF);
+        tcg_out_opc_imm(s, OPC_LW, datahi, base, HI_OFF);
+        break;
+    default:
+        tcg_abort();
+    }
+}
+
+static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64)
+{
+    TCGReg addr_regl, addr_regh __attribute__((unused));
+    TCGReg data_regl, data_regh;
+    TCGMemOpIdx oi;
+    TCGMemOp opc;
+#if defined(CONFIG_SOFTMMU)
+    tcg_insn_unit *label_ptr[2];
+    int mem_index;
+    TCGMemOp s_bits;
+#endif
+    /* Note that we've eliminated V0 from the output registers,
+       so we won't overwrite the base register during loading.  */
+    TCGReg base = TCG_REG_V0;
+
+    data_regl = *args++;
+    data_regh = (is_64 ? *args++ : 0);
+    addr_regl = *args++;
+    addr_regh = (TARGET_LONG_BITS == 64 ? *args++ : 0);
+    oi = *args++;
+    opc = get_memop(oi);
+
+#if defined(CONFIG_SOFTMMU)
+    mem_index = get_mmuidx(oi);
+    s_bits = opc & MO_SIZE;
+
+    tcg_out_tlb_load(s, base, addr_regl, addr_regh, mem_index,
+                     s_bits, label_ptr, 1);
+    tcg_out_qemu_ld_direct(s, data_regl, data_regh, base, opc);
+    add_qemu_ldst_label(s, 1, oi, data_regl, data_regh, addr_regl, addr_regh,
+                        s->code_ptr, label_ptr);
+#else
+    if (GUEST_BASE == 0 && data_regl != addr_regl) {
+        base = addr_regl;
+    } else if (GUEST_BASE == (int16_t)GUEST_BASE) {
+        tcg_out_opc_imm(s, OPC_ADDIU, base, addr_regl, GUEST_BASE);
+    } else {
+        tcg_out_movi(s, TCG_TYPE_PTR, base, GUEST_BASE);
+        tcg_out_opc_reg(s, OPC_ADDU, base, base, addr_regl);
+    }
+    tcg_out_qemu_ld_direct(s, data_regl, data_regh, base, opc);
+#endif
+}
+
+static void tcg_out_qemu_st_direct(TCGContext *s, TCGReg datalo, TCGReg datahi,
+                                   TCGReg base, TCGMemOp opc)
+{
+    switch (opc & (MO_SIZE | MO_BSWAP)) {
+    case MO_8:
+        tcg_out_opc_imm(s, OPC_SB, datalo, base, 0);
+        break;
+
+    case MO_16 | MO_BSWAP:
+        tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP1, datalo, 0xffff);
+        tcg_out_bswap16(s, TCG_TMP1, TCG_TMP1);
+        datalo = TCG_TMP1;
+        /* FALLTHRU */
+    case MO_16:
+        tcg_out_opc_imm(s, OPC_SH, datalo, base, 0);
+        break;
+
+    case MO_32 | MO_BSWAP:
+        tcg_out_bswap32(s, TCG_TMP1, datalo);
+        datalo = TCG_TMP1;
+        /* FALLTHRU */
+    case MO_32:
+        tcg_out_opc_imm(s, OPC_SW, datalo, base, 0);
+        break;
+
+    case MO_64 | MO_BSWAP:
+        tcg_out_bswap32(s, TCG_TMP1, datalo);
+        tcg_out_opc_imm(s, OPC_SW, TCG_TMP1, base, HI_OFF);
+        tcg_out_bswap32(s, TCG_TMP1, datahi);
+        tcg_out_opc_imm(s, OPC_SW, TCG_TMP1, base, LO_OFF);
+        break;
+    case MO_64:
+        tcg_out_opc_imm(s, OPC_SW, datalo, base, LO_OFF);
+        tcg_out_opc_imm(s, OPC_SW, datahi, base, HI_OFF);
+        break;
+
+    default:
+        tcg_abort();
+    }
+}
+
+static void tcg_out_addsub2(TCGContext *s, TCGReg rl, TCGReg rh, TCGReg al,
+                            TCGReg ah, TCGArg bl, TCGArg bh, bool cbl,
+                            bool cbh, bool is_sub)
+{
+    TCGReg th = TCG_TMP1;
+
+    /* If we have a negative constant such that negating it would
+       make the high part zero, we can (usually) eliminate one insn.  */
+    if (cbl && cbh && bh == -1 && bl != 0) {
+        bl = -bl;
+        bh = 0;
+        is_sub = !is_sub;
+    }
+
+    /* By operating on the high part first, we get to use the final
+       carry operation to move back from the temporary.  */
+    if (!cbh) {
+        tcg_out_opc_reg(s, (is_sub ? OPC_SUBU : OPC_ADDU), th, ah, bh);
+    } else if (bh != 0 || ah == rl) {
+        tcg_out_opc_imm(s, OPC_ADDIU, th, ah, (is_sub ? -bh : bh));
+    } else {
+        th = ah;
+    }
+
+    /* Note that tcg optimization should eliminate the bl == 0 case.  */
+    if (is_sub) {
+        if (cbl) {
+            tcg_out_opc_imm(s, OPC_SLTIU, TCG_TMP0, al, bl);
+            tcg_out_opc_imm(s, OPC_ADDIU, rl, al, -bl);
+        } else {
+            tcg_out_opc_reg(s, OPC_SLTU, TCG_TMP0, al, bl);
+            tcg_out_opc_reg(s, OPC_SUBU, rl, al, bl);
+        }
+        tcg_out_opc_reg(s, OPC_SUBU, rh, th, TCG_TMP0);
+    } else {
+        if (cbl) {
+            tcg_out_opc_imm(s, OPC_ADDIU, rl, al, bl);
+            tcg_out_opc_imm(s, OPC_SLTIU, TCG_TMP0, rl, bl);
+        } else if (rl == al && rl == bl) {
+            tcg_out_opc_sa(s, OPC_SRL, TCG_TMP0, al, 31);
+            tcg_out_opc_reg(s, OPC_ADDU, rl, al, bl);
+        } else {
+            tcg_out_opc_reg(s, OPC_ADDU, rl, al, bl);
+            tcg_out_opc_reg(s, OPC_SLTU, TCG_TMP0, rl, (rl == bl ? al : bl));
+        }
+        tcg_out_opc_reg(s, OPC_ADDU, rh, th, TCG_TMP0);
+    }
+}
+
+static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64)
+{
+    TCGReg addr_regl, addr_regh __attribute__((unused));
+    TCGReg data_regl, data_regh, base;
+    TCGMemOpIdx oi;
+    TCGMemOp opc;
+#if defined(CONFIG_SOFTMMU)
+    tcg_insn_unit *label_ptr[2];
+    int mem_index;
+    TCGMemOp s_bits;
+#endif
+
+    data_regl = *args++;
+    data_regh = (is_64 ? *args++ : 0);
+    addr_regl = *args++;
+    addr_regh = (TARGET_LONG_BITS == 64 ? *args++ : 0);
+    oi = *args++;
+    opc = get_memop(oi);
+
+#if defined(CONFIG_SOFTMMU)
+    mem_index = get_mmuidx(oi);
+    s_bits = opc & 3;
+
+    /* Note that we eliminated the helper's address argument,
+       so we can reuse that for the base.  */
+    base = (TARGET_LONG_BITS == 32 ? TCG_REG_A1 : TCG_REG_A2);
+    tcg_out_tlb_load(s, base, addr_regl, addr_regh, mem_index,
+                     s_bits, label_ptr, 0);
+    tcg_out_qemu_st_direct(s, data_regl, data_regh, base, opc);
+    add_qemu_ldst_label(s, 0, oi, data_regl, data_regh, addr_regl, addr_regh,
+                        s->code_ptr, label_ptr);
+#else
+    if (GUEST_BASE == 0) {
+        base = addr_regl;
+    } else {
+        base = TCG_REG_A0;
+        if (GUEST_BASE == (int16_t)GUEST_BASE) {
+            tcg_out_opc_imm(s, OPC_ADDIU, base, addr_regl, GUEST_BASE);
+        } else {
+            tcg_out_movi(s, TCG_TYPE_PTR, base, GUEST_BASE);
+            tcg_out_opc_reg(s, OPC_ADDU, base, base, addr_regl);
+        }
+    }
+    tcg_out_qemu_st_direct(s, data_regl, data_regh, base, opc);
+#endif
+}
+
+static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
+                              const TCGArg *args, const int *const_args)
+{
+    MIPSInsn i1, i2;
+    TCGArg a0, a1, a2;
+    int c2;
+
+    a0 = args[0];
+    a1 = args[1];
+    a2 = args[2];
+    c2 = const_args[2];
+
+    switch (opc) {
+    case INDEX_op_exit_tb:
+        {
+            TCGReg b0 = TCG_REG_ZERO;
+
+            if (a0 & ~0xffff) {
+                tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_V0, a0 & ~0xffff);
+                b0 = TCG_REG_V0;
+            }
+            if (!tcg_out_opc_jmp(s, OPC_J, tb_ret_addr)) {
+                tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0,
+                             (uintptr_t)tb_ret_addr);
+                tcg_out_opc_reg(s, OPC_JR, 0, TCG_TMP0, 0);
+            }
+            tcg_out_opc_imm(s, OPC_ORI, TCG_REG_V0, b0, a0 & 0xffff);
+        }
+        break;
+    case INDEX_op_goto_tb:
+        if (s->tb_jmp_offset) {
+            /* direct jump method */
+            s->tb_jmp_offset[a0] = tcg_current_code_size(s);
+            /* Avoid clobbering the address during retranslation.  */
+            tcg_out32(s, OPC_J | (*(uint32_t *)s->code_ptr & 0x3ffffff));
+        } else {
+            /* indirect jump method */
+            tcg_out_ld(s, TCG_TYPE_PTR, TCG_TMP0, TCG_REG_ZERO,
+                       (uintptr_t)(s->tb_next + a0));
+            tcg_out_opc_reg(s, OPC_JR, 0, TCG_TMP0, 0);
+        }
+        tcg_out_nop(s);
+        s->tb_next_offset[a0] = tcg_current_code_size(s);
+        break;
+    case INDEX_op_br:
+        tcg_out_brcond(s, TCG_COND_EQ, TCG_REG_ZERO, TCG_REG_ZERO,
+                       arg_label(a0));
+        break;
+
+    case INDEX_op_ld8u_i32:
+        i1 = OPC_LBU;
+        goto do_ldst;
+    case INDEX_op_ld8s_i32:
+        i1 = OPC_LB;
+        goto do_ldst;
+    case INDEX_op_ld16u_i32:
+        i1 = OPC_LHU;
+        goto do_ldst;
+    case INDEX_op_ld16s_i32:
+        i1 = OPC_LH;
+        goto do_ldst;
+    case INDEX_op_ld_i32:
+        i1 = OPC_LW;
+        goto do_ldst;
+    case INDEX_op_st8_i32:
+        i1 = OPC_SB;
+        goto do_ldst;
+    case INDEX_op_st16_i32:
+        i1 = OPC_SH;
+        goto do_ldst;
+    case INDEX_op_st_i32:
+        i1 = OPC_SW;
+    do_ldst:
+        tcg_out_ldst(s, i1, a0, a1, a2);
+        break;
+
+    case INDEX_op_add_i32:
+        i1 = OPC_ADDU, i2 = OPC_ADDIU;
+        goto do_binary;
+    case INDEX_op_or_i32:
+        i1 = OPC_OR, i2 = OPC_ORI;
+        goto do_binary;
+    case INDEX_op_xor_i32:
+        i1 = OPC_XOR, i2 = OPC_XORI;
+    do_binary:
+        if (c2) {
+            tcg_out_opc_imm(s, i2, a0, a1, a2);
+            break;
+        }
+    do_binaryv:
+        tcg_out_opc_reg(s, i1, a0, a1, a2);
+        break;
+
+    case INDEX_op_sub_i32:
+        if (c2) {
+            tcg_out_opc_imm(s, OPC_ADDIU, a0, a1, -a2);
+            break;
+        }
+        i1 = OPC_SUBU;
+        goto do_binary;
+    case INDEX_op_and_i32:
+        if (c2 && a2 != (uint16_t)a2) {
+            int msb = ctz32(~a2) - 1;
+            assert(use_mips32r2_instructions);
+            assert(is_p2m1(a2));
+            tcg_out_opc_bf(s, OPC_EXT, a0, a1, msb, 0);
+            break;
+        }
+        i1 = OPC_AND, i2 = OPC_ANDI;
+        goto do_binary;
+    case INDEX_op_nor_i32:
+        i1 = OPC_NOR;
+        goto do_binaryv;
+
+    case INDEX_op_mul_i32:
+        if (use_mips32_instructions) {
+            tcg_out_opc_reg(s, OPC_MUL, a0, a1, a2);
+            break;
+        }
+        i1 = OPC_MULT, i2 = OPC_MFLO;
+        goto do_hilo1;
+    case INDEX_op_mulsh_i32:
+        i1 = OPC_MULT, i2 = OPC_MFHI;
+        goto do_hilo1;
+    case INDEX_op_muluh_i32:
+        i1 = OPC_MULTU, i2 = OPC_MFHI;
+        goto do_hilo1;
+    case INDEX_op_div_i32:
+        i1 = OPC_DIV, i2 = OPC_MFLO;
+        goto do_hilo1;
+    case INDEX_op_divu_i32:
+        i1 = OPC_DIVU, i2 = OPC_MFLO;
+        goto do_hilo1;
+    case INDEX_op_rem_i32:
+        i1 = OPC_DIV, i2 = OPC_MFHI;
+        goto do_hilo1;
+    case INDEX_op_remu_i32:
+        i1 = OPC_DIVU, i2 = OPC_MFHI;
+    do_hilo1:
+        tcg_out_opc_reg(s, i1, 0, a1, a2);
+        tcg_out_opc_reg(s, i2, a0, 0, 0);
+        break;
+
+    case INDEX_op_muls2_i32:
+        i1 = OPC_MULT;
+        goto do_hilo2;
+    case INDEX_op_mulu2_i32:
+        i1 = OPC_MULTU;
+    do_hilo2:
+        tcg_out_opc_reg(s, i1, 0, a2, args[3]);
+        tcg_out_opc_reg(s, OPC_MFLO, a0, 0, 0);
+        tcg_out_opc_reg(s, OPC_MFHI, a1, 0, 0);
+        break;
+
+    case INDEX_op_not_i32:
+        i1 = OPC_NOR;
+        goto do_unary;
+    case INDEX_op_bswap16_i32:
+        i1 = OPC_WSBH;
+        goto do_unary;
+    case INDEX_op_ext8s_i32:
+        i1 = OPC_SEB;
+        goto do_unary;
+    case INDEX_op_ext16s_i32:
+        i1 = OPC_SEH;
+    do_unary:
+        tcg_out_opc_reg(s, i1, a0, TCG_REG_ZERO, a1);
+        break;
+
+    case INDEX_op_sar_i32:
+        i1 = OPC_SRAV, i2 = OPC_SRA;
+        goto do_shift;
+    case INDEX_op_shl_i32:
+        i1 = OPC_SLLV, i2 = OPC_SLL;
+        goto do_shift;
+    case INDEX_op_shr_i32:
+        i1 = OPC_SRLV, i2 = OPC_SRL;
+        goto do_shift;
+    case INDEX_op_rotr_i32:
+        i1 = OPC_ROTRV, i2 = OPC_ROTR;
+    do_shift:
+        if (c2) {
+            tcg_out_opc_sa(s, i2, a0, a1, a2);
+        } else {
+            tcg_out_opc_reg(s, i1, a0, a2, a1);
+        }
+        break;
+    case INDEX_op_rotl_i32:
+        if (c2) {
+            tcg_out_opc_sa(s, OPC_ROTR, a0, a1, 32 - a2);
+        } else {
+            tcg_out_opc_reg(s, OPC_SUBU, TCG_TMP0, TCG_REG_ZERO, a2);
+            tcg_out_opc_reg(s, OPC_ROTRV, a0, TCG_TMP0, a1);
+        }
+        break;
+
+    case INDEX_op_bswap32_i32:
+        tcg_out_opc_reg(s, OPC_WSBH, a0, 0, a1);
+        tcg_out_opc_sa(s, OPC_ROTR, a0, a0, 16);
+        break;
+
+    case INDEX_op_deposit_i32:
+        tcg_out_opc_bf(s, OPC_INS, a0, a2, args[3] + args[4] - 1, args[3]);
+        break;
+
+    case INDEX_op_brcond_i32:
+        tcg_out_brcond(s, a2, a0, a1, arg_label(args[3]));
+        break;
+    case INDEX_op_brcond2_i32:
+        tcg_out_brcond2(s, args[4], a0, a1, a2, args[3], arg_label(args[5]));
+        break;
+
+    case INDEX_op_movcond_i32:
+        tcg_out_movcond(s, args[5], a0, a1, a2, args[3]);
+        break;
+
+    case INDEX_op_setcond_i32:
+        tcg_out_setcond(s, args[3], a0, a1, a2);
+        break;
+    case INDEX_op_setcond2_i32:
+        tcg_out_setcond2(s, args[5], a0, a1, a2, args[3], args[4]);
+        break;
+
+    case INDEX_op_qemu_ld_i32:
+        tcg_out_qemu_ld(s, args, false);
+        break;
+    case INDEX_op_qemu_ld_i64:
+        tcg_out_qemu_ld(s, args, true);
+        break;
+    case INDEX_op_qemu_st_i32:
+        tcg_out_qemu_st(s, args, false);
+        break;
+    case INDEX_op_qemu_st_i64:
+        tcg_out_qemu_st(s, args, true);
+        break;
+
+    case INDEX_op_add2_i32:
+        tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5],
+                        const_args[4], const_args[5], false);
+        break;
+    case INDEX_op_sub2_i32:
+        tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], args[5],
+                        const_args[4], const_args[5], true);
+        break;
+
+    case INDEX_op_mov_i32:  /* Always emitted via tcg_out_mov.  */
+    case INDEX_op_movi_i32: /* Always emitted via tcg_out_movi.  */
+    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
+    default:
+        tcg_abort();
+    }
+}
+
+static const TCGTargetOpDef mips_op_defs[] = {
+    { INDEX_op_exit_tb, { } },
+    { INDEX_op_goto_tb, { } },
+    { INDEX_op_br, { } },
+
+    { INDEX_op_ld8u_i32, { "r", "r" } },
+    { INDEX_op_ld8s_i32, { "r", "r" } },
+    { INDEX_op_ld16u_i32, { "r", "r" } },
+    { INDEX_op_ld16s_i32, { "r", "r" } },
+    { INDEX_op_ld_i32, { "r", "r" } },
+    { INDEX_op_st8_i32, { "rZ", "r" } },
+    { INDEX_op_st16_i32, { "rZ", "r" } },
+    { INDEX_op_st_i32, { "rZ", "r" } },
+
+    { INDEX_op_add_i32, { "r", "rZ", "rJ" } },
+    { INDEX_op_mul_i32, { "r", "rZ", "rZ" } },
+    { INDEX_op_muls2_i32, { "r", "r", "rZ", "rZ" } },
+    { INDEX_op_mulu2_i32, { "r", "r", "rZ", "rZ" } },
+    { INDEX_op_mulsh_i32, { "r", "rZ", "rZ" } },
+    { INDEX_op_muluh_i32, { "r", "rZ", "rZ" } },
+    { INDEX_op_div_i32, { "r", "rZ", "rZ" } },
+    { INDEX_op_divu_i32, { "r", "rZ", "rZ" } },
+    { INDEX_op_rem_i32, { "r", "rZ", "rZ" } },
+    { INDEX_op_remu_i32, { "r", "rZ", "rZ" } },
+    { INDEX_op_sub_i32, { "r", "rZ", "rN" } },
+
+    { INDEX_op_and_i32, { "r", "rZ", "rIK" } },
+    { INDEX_op_nor_i32, { "r", "rZ", "rZ" } },
+    { INDEX_op_not_i32, { "r", "rZ" } },
+    { INDEX_op_or_i32, { "r", "rZ", "rIZ" } },
+    { INDEX_op_xor_i32, { "r", "rZ", "rIZ" } },
+
+    { INDEX_op_shl_i32, { "r", "rZ", "ri" } },
+    { INDEX_op_shr_i32, { "r", "rZ", "ri" } },
+    { INDEX_op_sar_i32, { "r", "rZ", "ri" } },
+    { INDEX_op_rotr_i32, { "r", "rZ", "ri" } },
+    { INDEX_op_rotl_i32, { "r", "rZ", "ri" } },
+
+    { INDEX_op_bswap16_i32, { "r", "r" } },
+    { INDEX_op_bswap32_i32, { "r", "r" } },
+
+    { INDEX_op_ext8s_i32, { "r", "rZ" } },
+    { INDEX_op_ext16s_i32, { "r", "rZ" } },
+
+    { INDEX_op_deposit_i32, { "r", "0", "rZ" } },
+
+    { INDEX_op_brcond_i32, { "rZ", "rZ" } },
+    { INDEX_op_movcond_i32, { "r", "rZ", "rZ", "rZ", "0" } },
+    { INDEX_op_setcond_i32, { "r", "rZ", "rZ" } },
+    { INDEX_op_setcond2_i32, { "r", "rZ", "rZ", "rZ", "rZ" } },
+
+    { INDEX_op_add2_i32, { "r", "r", "rZ", "rZ", "rN", "rN" } },
+    { INDEX_op_sub2_i32, { "r", "r", "rZ", "rZ", "rN", "rN" } },
+    { INDEX_op_brcond2_i32, { "rZ", "rZ", "rZ", "rZ" } },
+
+#if TARGET_LONG_BITS == 32
+    { INDEX_op_qemu_ld_i32, { "L", "lZ" } },
+    { INDEX_op_qemu_st_i32, { "SZ", "SZ" } },
+    { INDEX_op_qemu_ld_i64, { "L", "L", "lZ" } },
+    { INDEX_op_qemu_st_i64, { "SZ", "SZ", "SZ" } },
+#else
+    { INDEX_op_qemu_ld_i32, { "L", "lZ", "lZ" } },
+    { INDEX_op_qemu_st_i32, { "SZ", "SZ", "SZ" } },
+    { INDEX_op_qemu_ld_i64, { "L", "L", "lZ", "lZ" } },
+    { INDEX_op_qemu_st_i64, { "SZ", "SZ", "SZ", "SZ" } },
+#endif
+    { -1 },
+};
+
+static int tcg_target_callee_save_regs[] = {
+    TCG_REG_S0,       /* used for the global env (TCG_AREG0) */
+    TCG_REG_S1,
+    TCG_REG_S2,
+    TCG_REG_S3,
+    TCG_REG_S4,
+    TCG_REG_S5,
+    TCG_REG_S6,
+    TCG_REG_S7,
+    TCG_REG_S8,
+    TCG_REG_RA,       /* should be last for ABI compliance */
+};
+
+/* The Linux kernel doesn't provide any information about the available
+   instruction set. Probe it using a signal handler. */
+
+#include <signal.h>
+
+#ifndef use_movnz_instructions
+bool use_movnz_instructions = false;
+#endif
+
+#ifndef use_mips32_instructions
+bool use_mips32_instructions = false;
+#endif
+
+#ifndef use_mips32r2_instructions
+bool use_mips32r2_instructions = false;
+#endif
+
+static volatile sig_atomic_t got_sigill;
+
+static void sigill_handler(int signo, siginfo_t *si, void *data)
+{
+    /* Skip the faulty instruction */
+    ucontext_t *uc = (ucontext_t *)data;
+    uc->uc_mcontext.pc += 4;
+
+    got_sigill = 1;
+}
+
+static void tcg_target_detect_isa(void)
+{
+    struct sigaction sa_old, sa_new;
+
+    memset(&sa_new, 0, sizeof(sa_new));
+    sa_new.sa_flags = SA_SIGINFO;
+    sa_new.sa_sigaction = sigill_handler;
+    sigaction(SIGILL, &sa_new, &sa_old);
+
+    /* Probe for movn/movz, necessary to implement movcond. */
+#ifndef use_movnz_instructions
+    got_sigill = 0;
+    asm volatile(".set push\n"
+                 ".set mips32\n"
+                 "movn $zero, $zero, $zero\n"
+                 "movz $zero, $zero, $zero\n"
+                 ".set pop\n"
+                 : : : );
+    use_movnz_instructions = !got_sigill;
+#endif
+
+    /* Probe for MIPS32 instructions. As no subsetting is allowed
+       by the specification, it is only necessary to probe for one
+       of the instructions. */
+#ifndef use_mips32_instructions
+    got_sigill = 0;
+    asm volatile(".set push\n"
+                 ".set mips32\n"
+                 "mul $zero, $zero\n"
+                 ".set pop\n"
+                 : : : );
+    use_mips32_instructions = !got_sigill;
+#endif
+
+    /* Probe for MIPS32r2 instructions if MIPS32 instructions are
+       available. As no subsetting is allowed by the specification,
+       it is only necessary to probe for one of the instructions. */
+#ifndef use_mips32r2_instructions
+    if (use_mips32_instructions) {
+        got_sigill = 0;
+        asm volatile(".set push\n"
+                     ".set mips32r2\n"
+                     "seb $zero, $zero\n"
+                     ".set pop\n"
+                     : : : );
+        use_mips32r2_instructions = !got_sigill;
+    }
+#endif
+
+    sigaction(SIGILL, &sa_old, NULL);
+}
+
+/* Generate global QEMU prologue and epilogue code */
+static void tcg_target_qemu_prologue(TCGContext *s)
+{
+    int i, frame_size;
+
+    /* reserve some stack space, also for TCG temps. */
+    frame_size = ARRAY_SIZE(tcg_target_callee_save_regs) * 4
+                 + TCG_STATIC_CALL_ARGS_SIZE
+                 + CPU_TEMP_BUF_NLONGS * sizeof(long);
+    frame_size = (frame_size + TCG_TARGET_STACK_ALIGN - 1) &
+                 ~(TCG_TARGET_STACK_ALIGN - 1);
+    tcg_set_frame(s, TCG_REG_SP, ARRAY_SIZE(tcg_target_callee_save_regs) * 4
+                  + TCG_STATIC_CALL_ARGS_SIZE,
+                  CPU_TEMP_BUF_NLONGS * sizeof(long));
+
+    /* TB prologue */
+    tcg_out_addi(s, TCG_REG_SP, -frame_size);
+    for(i = 0 ; i < ARRAY_SIZE(tcg_target_callee_save_regs) ; i++) {
+        tcg_out_st(s, TCG_TYPE_I32, tcg_target_callee_save_regs[i],
+                   TCG_REG_SP, TCG_STATIC_CALL_ARGS_SIZE + i * 4);
+    }
+
+    /* Call generated code */
+    tcg_out_opc_reg(s, OPC_JR, 0, tcg_target_call_iarg_regs[1], 0);
+    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
+    tb_ret_addr = s->code_ptr;
+
+    /* TB epilogue */
+    for(i = 0 ; i < ARRAY_SIZE(tcg_target_callee_save_regs) ; i++) {
+        tcg_out_ld(s, TCG_TYPE_I32, tcg_target_callee_save_regs[i],
+                   TCG_REG_SP, TCG_STATIC_CALL_ARGS_SIZE + i * 4);
+    }
+
+    tcg_out_opc_reg(s, OPC_JR, 0, TCG_REG_RA, 0);
+    tcg_out_addi(s, TCG_REG_SP, frame_size);
+}
+
+static void tcg_target_init(TCGContext *s)
+{
+    tcg_target_detect_isa();
+    tcg_regset_set(tcg_target_available_regs[TCG_TYPE_I32], 0xffffffff);
+    tcg_regset_set(tcg_target_call_clobber_regs,
+                   (1 << TCG_REG_V0) |
+                   (1 << TCG_REG_V1) |
+                   (1 << TCG_REG_A0) |
+                   (1 << TCG_REG_A1) |
+                   (1 << TCG_REG_A2) |
+                   (1 << TCG_REG_A3) |
+                   (1 << TCG_REG_T0) |
+                   (1 << TCG_REG_T1) |
+                   (1 << TCG_REG_T2) |
+                   (1 << TCG_REG_T3) |
+                   (1 << TCG_REG_T4) |
+                   (1 << TCG_REG_T5) |
+                   (1 << TCG_REG_T6) |
+                   (1 << TCG_REG_T7) |
+                   (1 << TCG_REG_T8) |
+                   (1 << TCG_REG_T9));
+
+    tcg_regset_clear(s->reserved_regs);
+    tcg_regset_set_reg(s->reserved_regs, TCG_REG_ZERO); /* zero register */
+    tcg_regset_set_reg(s->reserved_regs, TCG_REG_K0);   /* kernel use only */
+    tcg_regset_set_reg(s->reserved_regs, TCG_REG_K1);   /* kernel use only */
+    tcg_regset_set_reg(s->reserved_regs, TCG_TMP0);     /* internal use */
+    tcg_regset_set_reg(s->reserved_regs, TCG_TMP1);     /* internal use */
+    tcg_regset_set_reg(s->reserved_regs, TCG_REG_RA);   /* return address */
+    tcg_regset_set_reg(s->reserved_regs, TCG_REG_SP);   /* stack pointer */
+    tcg_regset_set_reg(s->reserved_regs, TCG_REG_GP);   /* global pointer */
+
+    tcg_add_target_add_op_defs(mips_op_defs);
+}
+
+void tb_set_jmp_target1(uintptr_t jmp_addr, uintptr_t addr)
+{
+    uint32_t *ptr = (uint32_t *)jmp_addr;
+    *ptr = deposit32(*ptr, 0, 26, addr >> 2);
+    flush_icache_range(jmp_addr, jmp_addr + 4);
+}