/* * Tiny Code Generator for QEMU * * Copyright (c) 2009, 2011 Stefan Weil * * 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. */ /* TODO list: * - See TODO comments in code. */ /* Marker for missing code. */ #define TODO() \ do { \ fprintf(stderr, "TODO %s:%u: %s()\n", \ __FILE__, __LINE__, __func__); \ tcg_abort(); \ } while (0) /* Bitfield n...m (in 32 bit value). */ #define BITS(n, m) (((0xffffffffU << (31 - n)) >> (31 - n + m)) << m) static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op) { switch (op) { case INDEX_op_ld8u_i32: case INDEX_op_ld8s_i32: case INDEX_op_ld16u_i32: case INDEX_op_ld16s_i32: case INDEX_op_ld_i32: case INDEX_op_ld8u_i64: case INDEX_op_ld8s_i64: case INDEX_op_ld16u_i64: case INDEX_op_ld16s_i64: case INDEX_op_ld32u_i64: case INDEX_op_ld32s_i64: case INDEX_op_ld_i64: case INDEX_op_not_i32: case INDEX_op_not_i64: case INDEX_op_neg_i32: case INDEX_op_neg_i64: case INDEX_op_ext8s_i32: case INDEX_op_ext8s_i64: case INDEX_op_ext16s_i32: case INDEX_op_ext16s_i64: case INDEX_op_ext8u_i32: case INDEX_op_ext8u_i64: case INDEX_op_ext16u_i32: case INDEX_op_ext16u_i64: case INDEX_op_ext32s_i64: case INDEX_op_ext32u_i64: case INDEX_op_ext_i32_i64: case INDEX_op_extu_i32_i64: case INDEX_op_bswap16_i32: case INDEX_op_bswap16_i64: case INDEX_op_bswap32_i32: case INDEX_op_bswap32_i64: case INDEX_op_bswap64_i64: return C_O1_I1(r, r); case INDEX_op_st8_i32: case INDEX_op_st16_i32: case INDEX_op_st_i32: case INDEX_op_st8_i64: case INDEX_op_st16_i64: case INDEX_op_st32_i64: case INDEX_op_st_i64: return C_O0_I2(r, r); case INDEX_op_div_i32: case INDEX_op_div_i64: case INDEX_op_divu_i32: case INDEX_op_divu_i64: case INDEX_op_rem_i32: case INDEX_op_rem_i64: case INDEX_op_remu_i32: case INDEX_op_remu_i64: case INDEX_op_add_i32: case INDEX_op_add_i64: case INDEX_op_sub_i32: case INDEX_op_sub_i64: case INDEX_op_mul_i32: case INDEX_op_mul_i64: case INDEX_op_and_i32: case INDEX_op_and_i64: case INDEX_op_andc_i32: case INDEX_op_andc_i64: case INDEX_op_eqv_i32: case INDEX_op_eqv_i64: case INDEX_op_nand_i32: case INDEX_op_nand_i64: case INDEX_op_nor_i32: case INDEX_op_nor_i64: case INDEX_op_or_i32: case INDEX_op_or_i64: case INDEX_op_orc_i32: case INDEX_op_orc_i64: case INDEX_op_xor_i32: case INDEX_op_xor_i64: case INDEX_op_shl_i32: case INDEX_op_shl_i64: case INDEX_op_shr_i32: case INDEX_op_shr_i64: case INDEX_op_sar_i32: case INDEX_op_sar_i64: case INDEX_op_rotl_i32: case INDEX_op_rotl_i64: case INDEX_op_rotr_i32: case INDEX_op_rotr_i64: case INDEX_op_setcond_i32: case INDEX_op_setcond_i64: return C_O1_I2(r, r, r); case INDEX_op_deposit_i32: case INDEX_op_deposit_i64: return C_O1_I2(r, 0, r); case INDEX_op_brcond_i32: case INDEX_op_brcond_i64: return C_O0_I2(r, r); #if TCG_TARGET_REG_BITS == 32 /* TODO: Support R, R, R, R, RI, RI? Will it be faster? */ case INDEX_op_add2_i32: case INDEX_op_sub2_i32: return C_O2_I4(r, r, r, r, r, r); case INDEX_op_brcond2_i32: return C_O0_I4(r, r, r, r); case INDEX_op_mulu2_i32: return C_O2_I2(r, r, r, r); case INDEX_op_setcond2_i32: return C_O1_I4(r, r, r, r, r); #endif case INDEX_op_qemu_ld_i32: return (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS ? C_O1_I1(r, r) : C_O1_I2(r, r, r)); case INDEX_op_qemu_ld_i64: return (TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, r) : TARGET_LONG_BITS <= TCG_TARGET_REG_BITS ? C_O2_I1(r, r, r) : C_O2_I2(r, r, r, r)); case INDEX_op_qemu_st_i32: return (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS ? C_O0_I2(r, r) : C_O0_I3(r, r, r)); case INDEX_op_qemu_st_i64: return (TCG_TARGET_REG_BITS == 64 ? C_O0_I2(r, r) : TARGET_LONG_BITS <= TCG_TARGET_REG_BITS ? C_O0_I3(r, r, r) : C_O0_I4(r, r, r, r)); default: g_assert_not_reached(); } } static const int tcg_target_reg_alloc_order[] = { TCG_REG_R0, TCG_REG_R1, TCG_REG_R2, TCG_REG_R3, TCG_REG_R4, TCG_REG_R5, TCG_REG_R6, TCG_REG_R7, TCG_REG_R8, TCG_REG_R9, TCG_REG_R10, TCG_REG_R11, TCG_REG_R12, TCG_REG_R13, TCG_REG_R14, TCG_REG_R15, }; #if MAX_OPC_PARAM_IARGS != 6 # error Fix needed, number of supported input arguments changed! #endif static const int tcg_target_call_iarg_regs[] = { TCG_REG_R0, TCG_REG_R1, TCG_REG_R2, TCG_REG_R3, TCG_REG_R4, TCG_REG_R5, #if TCG_TARGET_REG_BITS == 32 /* 32 bit hosts need 2 * MAX_OPC_PARAM_IARGS registers. */ TCG_REG_R6, TCG_REG_R7, TCG_REG_R8, TCG_REG_R9, TCG_REG_R10, TCG_REG_R11, #endif }; static const int tcg_target_call_oarg_regs[] = { TCG_REG_R0, #if TCG_TARGET_REG_BITS == 32 TCG_REG_R1 #endif }; #ifdef CONFIG_DEBUG_TCG static const char *const tcg_target_reg_names[TCG_TARGET_NB_REGS] = { "r00", "r01", "r02", "r03", "r04", "r05", "r06", "r07", "r08", "r09", "r10", "r11", "r12", "r13", "r14", "r15", }; #endif static bool patch_reloc(tcg_insn_unit *code_ptr, int type, intptr_t value, intptr_t addend) { /* tcg_out_reloc always uses the same type, addend. */ tcg_debug_assert(type == sizeof(tcg_target_long)); tcg_debug_assert(addend == 0); tcg_debug_assert(value != 0); if (TCG_TARGET_REG_BITS == 32) { tcg_patch32(code_ptr, value); } else { tcg_patch64(code_ptr, value); } return true; } #if defined(CONFIG_DEBUG_TCG_INTERPRETER) /* Show current bytecode. Used by tcg interpreter. */ void tci_disas(uint8_t opc) { const TCGOpDef *def = &tcg_op_defs[opc]; fprintf(stderr, "TCG %s %u, %u, %u\n", def->name, def->nb_oargs, def->nb_iargs, def->nb_cargs); } #endif /* Write value (native size). */ static void tcg_out_i(TCGContext *s, tcg_target_ulong v) { if (TCG_TARGET_REG_BITS == 32) { tcg_out32(s, v); } else { tcg_out64(s, v); } } /* Write opcode. */ static void tcg_out_op_t(TCGContext *s, TCGOpcode op) { tcg_out8(s, op); tcg_out8(s, 0); } /* Write register. */ static void tcg_out_r(TCGContext *s, TCGArg t0) { tcg_debug_assert(t0 < TCG_TARGET_NB_REGS); tcg_out8(s, t0); } /* Write label. */ static void tci_out_label(TCGContext *s, TCGLabel *label) { if (label->has_value) { tcg_out_i(s, label->u.value); tcg_debug_assert(label->u.value); } else { tcg_out_reloc(s, s->code_ptr, sizeof(tcg_target_ulong), label, 0); s->code_ptr += sizeof(tcg_target_ulong); } } static void stack_bounds_check(TCGReg base, target_long offset) { if (base == TCG_REG_CALL_STACK) { tcg_debug_assert(offset < 0); tcg_debug_assert(offset >= -(CPU_TEMP_BUF_NLONGS * sizeof(long))); } } static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg1, intptr_t arg2) { uint8_t *old_code_ptr = s->code_ptr; stack_bounds_check(arg1, arg2); if (type == TCG_TYPE_I32) { tcg_out_op_t(s, INDEX_op_ld_i32); tcg_out_r(s, ret); tcg_out_r(s, arg1); tcg_out32(s, arg2); } else { tcg_debug_assert(type == TCG_TYPE_I64); #if TCG_TARGET_REG_BITS == 64 tcg_out_op_t(s, INDEX_op_ld_i64); tcg_out_r(s, ret); tcg_out_r(s, arg1); tcg_debug_assert(arg2 == (int32_t)arg2); tcg_out32(s, arg2); #else TODO(); #endif } old_code_ptr[1] = s->code_ptr - old_code_ptr; } static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg) { uint8_t *old_code_ptr = s->code_ptr; tcg_debug_assert(ret != arg); #if TCG_TARGET_REG_BITS == 32 tcg_out_op_t(s, INDEX_op_mov_i32); #else tcg_out_op_t(s, INDEX_op_mov_i64); #endif tcg_out_r(s, ret); tcg_out_r(s, arg); old_code_ptr[1] = s->code_ptr - old_code_ptr; return true; } static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg t0, tcg_target_long arg) { uint8_t *old_code_ptr = s->code_ptr; uint32_t arg32 = arg; if (type == TCG_TYPE_I32 || arg == arg32) { tcg_out_op_t(s, INDEX_op_tci_movi_i32); tcg_out_r(s, t0); tcg_out32(s, arg32); } else { tcg_debug_assert(type == TCG_TYPE_I64); #if TCG_TARGET_REG_BITS == 64 tcg_out_op_t(s, INDEX_op_tci_movi_i64); tcg_out_r(s, t0); tcg_out64(s, arg); #else TODO(); #endif } old_code_ptr[1] = s->code_ptr - old_code_ptr; } static inline void tcg_out_call(TCGContext *s, const tcg_insn_unit *arg) { uint8_t *old_code_ptr = s->code_ptr; tcg_out_op_t(s, INDEX_op_call); tcg_out_i(s, (uintptr_t)arg); old_code_ptr[1] = s->code_ptr - old_code_ptr; } #if TCG_TARGET_REG_BITS == 64 # define CASE_32_64(x) \ case glue(glue(INDEX_op_, x), _i64): \ case glue(glue(INDEX_op_, x), _i32): # define CASE_64(x) \ case glue(glue(INDEX_op_, x), _i64): #else # define CASE_32_64(x) \ case glue(glue(INDEX_op_, x), _i32): # define CASE_64(x) #endif static void tcg_out_op(TCGContext *s, TCGOpcode opc, const TCGArg *args, const int *const_args) { uint8_t *old_code_ptr = s->code_ptr; tcg_out_op_t(s, opc); switch (opc) { case INDEX_op_exit_tb: tcg_out64(s, args[0]); break; case INDEX_op_goto_tb: if (s->tb_jmp_insn_offset) { /* Direct jump method. */ /* Align for atomic patching and thread safety */ s->code_ptr = QEMU_ALIGN_PTR_UP(s->code_ptr, 4); s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s); tcg_out32(s, 0); } else { /* Indirect jump method. */ TODO(); } set_jmp_reset_offset(s, args[0]); break; case INDEX_op_br: tci_out_label(s, arg_label(args[0])); break; CASE_32_64(setcond) tcg_out_r(s, args[0]); tcg_out_r(s, args[1]); tcg_out_r(s, args[2]); tcg_out8(s, args[3]); /* condition */ break; #if TCG_TARGET_REG_BITS == 32 case INDEX_op_setcond2_i32: /* setcond2_i32 cond, t0, t1_low, t1_high, t2_low, t2_high */ tcg_out_r(s, args[0]); tcg_out_r(s, args[1]); tcg_out_r(s, args[2]); tcg_out_r(s, args[3]); tcg_out_r(s, args[4]); tcg_out8(s, args[5]); /* condition */ break; #endif CASE_32_64(ld8u) CASE_32_64(ld8s) CASE_32_64(ld16u) CASE_32_64(ld16s) case INDEX_op_ld_i32: CASE_64(ld32u) CASE_64(ld32s) CASE_64(ld) CASE_32_64(st8) CASE_32_64(st16) case INDEX_op_st_i32: CASE_64(st32) CASE_64(st) stack_bounds_check(args[1], args[2]); tcg_out_r(s, args[0]); tcg_out_r(s, args[1]); tcg_debug_assert(args[2] == (int32_t)args[2]); tcg_out32(s, args[2]); break; CASE_32_64(add) CASE_32_64(sub) CASE_32_64(mul) CASE_32_64(and) CASE_32_64(or) CASE_32_64(xor) CASE_32_64(andc) /* Optional (TCG_TARGET_HAS_andc_*). */ CASE_32_64(orc) /* Optional (TCG_TARGET_HAS_orc_*). */ CASE_32_64(eqv) /* Optional (TCG_TARGET_HAS_eqv_*). */ CASE_32_64(nand) /* Optional (TCG_TARGET_HAS_nand_*). */ CASE_32_64(nor) /* Optional (TCG_TARGET_HAS_nor_*). */ CASE_32_64(shl) CASE_32_64(shr) CASE_32_64(sar) CASE_32_64(rotl) /* Optional (TCG_TARGET_HAS_rot_*). */ CASE_32_64(rotr) /* Optional (TCG_TARGET_HAS_rot_*). */ CASE_32_64(div) /* Optional (TCG_TARGET_HAS_div_*). */ CASE_32_64(divu) /* Optional (TCG_TARGET_HAS_div_*). */ CASE_32_64(rem) /* Optional (TCG_TARGET_HAS_div_*). */ CASE_32_64(remu) /* Optional (TCG_TARGET_HAS_div_*). */ tcg_out_r(s, args[0]); tcg_out_r(s, args[1]); tcg_out_r(s, args[2]); break; CASE_32_64(deposit) /* Optional (TCG_TARGET_HAS_deposit_*). */ tcg_out_r(s, args[0]); tcg_out_r(s, args[1]); tcg_out_r(s, args[2]); tcg_debug_assert(args[3] <= UINT8_MAX); tcg_out8(s, args[3]); tcg_debug_assert(args[4] <= UINT8_MAX); tcg_out8(s, args[4]); break; CASE_32_64(brcond) tcg_out_r(s, args[0]); tcg_out_r(s, args[1]); tcg_out8(s, args[2]); /* condition */ tci_out_label(s, arg_label(args[3])); break; CASE_32_64(neg) /* Optional (TCG_TARGET_HAS_neg_*). */ CASE_32_64(not) /* Optional (TCG_TARGET_HAS_not_*). */ CASE_32_64(ext8s) /* Optional (TCG_TARGET_HAS_ext8s_*). */ CASE_32_64(ext8u) /* Optional (TCG_TARGET_HAS_ext8u_*). */ CASE_32_64(ext16s) /* Optional (TCG_TARGET_HAS_ext16s_*). */ CASE_32_64(ext16u) /* Optional (TCG_TARGET_HAS_ext16u_*). */ CASE_64(ext32s) /* Optional (TCG_TARGET_HAS_ext32s_i64). */ CASE_64(ext32u) /* Optional (TCG_TARGET_HAS_ext32u_i64). */ CASE_64(ext_i32) CASE_64(extu_i32) CASE_32_64(bswap16) /* Optional (TCG_TARGET_HAS_bswap16_*). */ CASE_32_64(bswap32) /* Optional (TCG_TARGET_HAS_bswap32_*). */ CASE_64(bswap64) /* Optional (TCG_TARGET_HAS_bswap64_i64). */ tcg_out_r(s, args[0]); tcg_out_r(s, args[1]); break; #if TCG_TARGET_REG_BITS == 32 case INDEX_op_add2_i32: case INDEX_op_sub2_i32: tcg_out_r(s, args[0]); tcg_out_r(s, args[1]); tcg_out_r(s, args[2]); tcg_out_r(s, args[3]); tcg_out_r(s, args[4]); tcg_out_r(s, args[5]); break; case INDEX_op_brcond2_i32: tcg_out_r(s, args[0]); tcg_out_r(s, args[1]); tcg_out_r(s, args[2]); tcg_out_r(s, args[3]); tcg_out8(s, args[4]); /* condition */ tci_out_label(s, arg_label(args[5])); break; case INDEX_op_mulu2_i32: tcg_out_r(s, args[0]); tcg_out_r(s, args[1]); tcg_out_r(s, args[2]); tcg_out_r(s, args[3]); break; #endif case INDEX_op_qemu_ld_i32: case INDEX_op_qemu_st_i32: tcg_out_r(s, *args++); tcg_out_r(s, *args++); if (TARGET_LONG_BITS > TCG_TARGET_REG_BITS) { tcg_out_r(s, *args++); } tcg_out_i(s, *args++); break; case INDEX_op_qemu_ld_i64: case INDEX_op_qemu_st_i64: tcg_out_r(s, *args++); if (TCG_TARGET_REG_BITS == 32) { tcg_out_r(s, *args++); } tcg_out_r(s, *args++); if (TARGET_LONG_BITS > TCG_TARGET_REG_BITS) { tcg_out_r(s, *args++); } tcg_out_i(s, *args++); break; case INDEX_op_mb: break; case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */ case INDEX_op_mov_i64: case INDEX_op_call: /* Always emitted via tcg_out_call. */ default: tcg_abort(); } old_code_ptr[1] = s->code_ptr - old_code_ptr; } static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, TCGReg arg1, intptr_t arg2) { uint8_t *old_code_ptr = s->code_ptr; stack_bounds_check(arg1, arg2); if (type == TCG_TYPE_I32) { tcg_out_op_t(s, INDEX_op_st_i32); tcg_out_r(s, arg); tcg_out_r(s, arg1); tcg_out32(s, arg2); } else { tcg_debug_assert(type == TCG_TYPE_I64); #if TCG_TARGET_REG_BITS == 64 tcg_out_op_t(s, INDEX_op_st_i64); tcg_out_r(s, arg); tcg_out_r(s, arg1); tcg_out32(s, arg2); #else TODO(); #endif } old_code_ptr[1] = s->code_ptr - old_code_ptr; } static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val, TCGReg base, intptr_t ofs) { return false; } /* Test if a constant matches the constraint. */ static int tcg_target_const_match(tcg_target_long val, TCGType type, const TCGArgConstraint *arg_ct) { /* No need to return 0 or 1, 0 or != 0 is good enough. */ return arg_ct->ct & TCG_CT_CONST; } static void tcg_target_init(TCGContext *s) { #if defined(CONFIG_DEBUG_TCG_INTERPRETER) const char *envval = getenv("DEBUG_TCG"); if (envval) { qemu_set_log(strtol(envval, NULL, 0)); } #endif /* The current code uses uint8_t for tcg operations. */ tcg_debug_assert(tcg_op_defs_max <= UINT8_MAX); /* Registers available for 32 bit operations. */ tcg_target_available_regs[TCG_TYPE_I32] = BIT(TCG_TARGET_NB_REGS) - 1; /* Registers available for 64 bit operations. */ tcg_target_available_regs[TCG_TYPE_I64] = BIT(TCG_TARGET_NB_REGS) - 1; /* TODO: Which registers should be set here? */ tcg_target_call_clobber_regs = BIT(TCG_TARGET_NB_REGS) - 1; s->reserved_regs = 0; tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK); /* We use negative offsets from "sp" so that we can distinguish stores that might pretend to be call arguments. */ tcg_set_frame(s, TCG_REG_CALL_STACK, -CPU_TEMP_BUF_NLONGS * sizeof(long), CPU_TEMP_BUF_NLONGS * sizeof(long)); } /* Generate global QEMU prologue and epilogue code. */ static inline void tcg_target_qemu_prologue(TCGContext *s) { }