/* aarch64-dis.c -- AArch64 disassembler. Copyright 2009, 2010, 2011, 2012 Free Software Foundation, Inc. Contributed by ARM Ltd. This file is part of the GNU opcodes library. This library is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3, or (at your option) any later version. It is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program; see the file COPYING3. If not, see <http://www.gnu.org/licenses/>. */ #include "sysdep.h" #include "bfd_stdint.h" #include "dis-asm.h" #include "libiberty.h" #include "opintl.h" #include "aarch64-dis.h" #if !defined(EMBEDDED_ENV) #define SYMTAB_AVAILABLE 1 #include "elf-bfd.h" #include "elf/aarch64.h" #endif #define ERR_OK 0 #define ERR_UND -1 #define ERR_UNP -3 #define ERR_NYI -5 #define INSNLEN 4 /* Cached mapping symbol state. */ enum map_type { MAP_INSN, MAP_DATA }; static enum map_type last_type; static int last_mapping_sym = -1; static bfd_vma last_mapping_addr = 0; /* Other options */ static int no_aliases = 0; /* If set disassemble as most general inst. */ static void set_default_aarch64_dis_options (struct disassemble_info *info ATTRIBUTE_UNUSED) { } static void parse_aarch64_dis_option (const char *option, unsigned int len ATTRIBUTE_UNUSED) { /* Try to match options that are simple flags */ if (CONST_STRNEQ (option, "no-aliases")) { no_aliases = 1; return; } if (CONST_STRNEQ (option, "aliases")) { no_aliases = 0; return; } #ifdef DEBUG_AARCH64 if (CONST_STRNEQ (option, "debug_dump")) { debug_dump = 1; return; } #endif /* DEBUG_AARCH64 */ /* Invalid option. */ fprintf (stderr, _("Unrecognised disassembler option: %s\n"), option); } static void parse_aarch64_dis_options (const char *options) { const char *option_end; if (options == NULL) return; while (*options != '\0') { /* Skip empty options. */ if (*options == ',') { options++; continue; } /* We know that *options is neither NUL or a comma. */ option_end = options + 1; while (*option_end != ',' && *option_end != '\0') option_end++; parse_aarch64_dis_option (options, option_end - options); /* Go on to the next one. If option_end points to a comma, it will be skipped above. */ options = option_end; } } /* Functions doing the instruction disassembling. */ /* The unnamed arguments consist of the number of fields and information about these fields where the VALUE will be extracted from CODE and returned. MASK can be zero or the base mask of the opcode. N.B. the fields are required to be in such an order than the most signficant field for VALUE comes the first, e.g. the <index> in SQDMLAL <Va><d>, <Vb><n>, <Vm>.<Ts>[<index>] is encoded in H:L:M in some cases, the the fields H:L:M should be passed in the order of H, L, M. */ static inline aarch64_insn extract_fields (aarch64_insn code, aarch64_insn mask, ...) { uint32_t num; const aarch64_field *field; enum aarch64_field_kind kind; va_list va; va_start (va, mask); num = va_arg (va, uint32_t); assert (num <= 5); aarch64_insn value = 0x0; while (num--) { kind = va_arg (va, enum aarch64_field_kind); field = &fields[kind]; value <<= field->width; value |= extract_field (kind, code, mask); } return value; } /* Sign-extend bit I of VALUE. */ static inline int32_t sign_extend (aarch64_insn value, unsigned i) { uint32_t ret = value; assert (i < 32); if ((value >> i) & 0x1) { uint32_t val = (uint32_t)(-1) << i; ret = ret | val; } return (int32_t) ret; } /* N.B. the following inline helpfer functions create a dependency on the order of operand qualifier enumerators. */ /* Given VALUE, return qualifier for a general purpose register. */ static inline enum aarch64_opnd_qualifier get_greg_qualifier_from_value (aarch64_insn value) { enum aarch64_opnd_qualifier qualifier = AARCH64_OPND_QLF_W + value; assert (value <= 0x1 && aarch64_get_qualifier_standard_value (qualifier) == value); return qualifier; } /* Given VALUE, return qualifier for a vector register. */ static inline enum aarch64_opnd_qualifier get_vreg_qualifier_from_value (aarch64_insn value) { enum aarch64_opnd_qualifier qualifier = AARCH64_OPND_QLF_V_8B + value; assert (value <= 0x8 && aarch64_get_qualifier_standard_value (qualifier) == value); return qualifier; } /* Given VALUE, return qualifier for an FP or AdvSIMD scalar register. */ static inline enum aarch64_opnd_qualifier get_sreg_qualifier_from_value (aarch64_insn value) { enum aarch64_opnd_qualifier qualifier = AARCH64_OPND_QLF_S_B + value; assert (value <= 0x4 && aarch64_get_qualifier_standard_value (qualifier) == value); return qualifier; } /* Given the instruction in *INST which is probably half way through the decoding and our caller wants to know the expected qualifier for operand I. Return such a qualifier if we can establish it; otherwise return AARCH64_OPND_QLF_NIL. */ static aarch64_opnd_qualifier_t get_expected_qualifier (const aarch64_inst *inst, int i) { aarch64_opnd_qualifier_seq_t qualifiers; /* Should not be called if the qualifier is known. */ assert (inst->operands[i].qualifier == AARCH64_OPND_QLF_NIL); if (aarch64_find_best_match (inst, inst->opcode->qualifiers_list, i, qualifiers)) return qualifiers[i]; else return AARCH64_OPND_QLF_NIL; } /* Operand extractors. */ int aarch64_ext_regno (const aarch64_operand *self, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { info->reg.regno = extract_field (self->fields[0], code, 0); return 1; } /* e.g. IC <ic_op>{, <Xt>}. */ int aarch64_ext_regrt_sysins (const aarch64_operand *self, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { info->reg.regno = extract_field (self->fields[0], code, 0); assert (info->idx == 1 && (aarch64_get_operand_class (inst->operands[0].type) == AARCH64_OPND_CLASS_SYSTEM)); /* This will make the constraint checking happy and more importantly will help the disassembler determine whether this operand is optional or not. */ info->present = inst->operands[0].sysins_op->has_xt; return 1; } /* e.g. SQDMLAL <Va><d>, <Vb><n>, <Vm>.<Ts>[<index>]. */ int aarch64_ext_reglane (const aarch64_operand *self, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { /* regno */ info->reglane.regno = extract_field (self->fields[0], code, inst->opcode->mask); /* Index and/or type. */ if (inst->opcode->iclass == asisdone || inst->opcode->iclass == asimdins) { if (info->type == AARCH64_OPND_En && inst->opcode->operands[0] == AARCH64_OPND_Ed) { unsigned shift; /* index2 for e.g. INS <Vd>.<Ts>[<index1>], <Vn>.<Ts>[<index2>]. */ assert (info->idx == 1); /* Vn */ aarch64_insn value = extract_field (FLD_imm4, code, 0); /* Depend on AARCH64_OPND_Ed to determine the qualifier. */ info->qualifier = get_expected_qualifier (inst, info->idx); shift = get_logsz (aarch64_get_qualifier_esize (info->qualifier)); info->reglane.index = value >> shift; } else { /* index and type for e.g. DUP <V><d>, <Vn>.<T>[<index>]. imm5<3:0> <V> 0000 RESERVED xxx1 B xx10 H x100 S 1000 D */ int pos = -1; aarch64_insn value = extract_field (FLD_imm5, code, 0); while (++pos <= 3 && (value & 0x1) == 0) value >>= 1; if (pos > 3) return 0; info->qualifier = get_sreg_qualifier_from_value (pos); info->reglane.index = (unsigned) (value >> 1); } } else { /* Index only for e.g. SQDMLAL <Va><d>, <Vb><n>, <Vm>.<Ts>[<index>] or SQDMLAL <Va><d>, <Vb><n>, <Vm>.<Ts>[<index>]. */ /* Need information in other operand(s) to help decoding. */ info->qualifier = get_expected_qualifier (inst, info->idx); switch (info->qualifier) { case AARCH64_OPND_QLF_S_H: /* h:l:m */ info->reglane.index = extract_fields (code, 0, 3, FLD_H, FLD_L, FLD_M); info->reglane.regno &= 0xf; break; case AARCH64_OPND_QLF_S_S: /* h:l */ info->reglane.index = extract_fields (code, 0, 2, FLD_H, FLD_L); break; case AARCH64_OPND_QLF_S_D: /* H */ info->reglane.index = extract_field (FLD_H, code, 0); break; default: return 0; } } return 1; } int aarch64_ext_reglist (const aarch64_operand *self, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { /* R */ info->reglist.first_regno = extract_field (self->fields[0], code, 0); /* len */ info->reglist.num_regs = extract_field (FLD_len, code, 0) + 1; return 1; } /* Decode Rt and opcode fields of Vt in AdvSIMD load/store instructions. */ int aarch64_ext_ldst_reglist (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst) { aarch64_insn value; /* Number of elements in each structure to be loaded/stored. */ unsigned expected_num = get_opcode_dependent_value (inst->opcode); struct { unsigned is_reserved; unsigned num_regs; unsigned num_elements; } data [] = { {0, 4, 4}, {1, 4, 4}, {0, 4, 1}, {0, 4, 2}, {0, 3, 3}, {1, 3, 3}, {0, 3, 1}, {0, 1, 1}, {0, 2, 2}, {1, 2, 2}, {0, 2, 1}, }; /* Rt */ info->reglist.first_regno = extract_field (FLD_Rt, code, 0); /* opcode */ value = extract_field (FLD_opcode, code, 0); if (expected_num != data[value].num_elements || data[value].is_reserved) return 0; info->reglist.num_regs = data[value].num_regs; return 1; } /* Decode Rt and S fields of Vt in AdvSIMD load single structure to all lanes instructions. */ int aarch64_ext_ldst_reglist_r (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst) { aarch64_insn value; /* Rt */ info->reglist.first_regno = extract_field (FLD_Rt, code, 0); /* S */ value = extract_field (FLD_S, code, 0); /* Number of registers is equal to the number of elements in each structure to be loaded/stored. */ info->reglist.num_regs = get_opcode_dependent_value (inst->opcode); assert (info->reglist.num_regs >= 1 && info->reglist.num_regs <= 4); /* Except when it is LD1R. */ if (info->reglist.num_regs == 1 && value == (aarch64_insn) 1) info->reglist.num_regs = 2; return 1; } /* Decode Q, opcode<2:1>, S, size and Rt fields of Vt in AdvSIMD load/store single element instructions. */ int aarch64_ext_ldst_elemlist (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { aarch64_field field = {0, 0}; aarch64_insn QSsize; /* fields Q:S:size. */ aarch64_insn opcodeh2; /* opcode<2:1> */ /* Rt */ info->reglist.first_regno = extract_field (FLD_Rt, code, 0); /* Decode the index, opcode<2:1> and size. */ gen_sub_field (FLD_asisdlso_opcode, 1, 2, &field); opcodeh2 = extract_field_2 (&field, code, 0); QSsize = extract_fields (code, 0, 3, FLD_Q, FLD_S, FLD_vldst_size); switch (opcodeh2) { case 0x0: info->qualifier = AARCH64_OPND_QLF_S_B; /* Index encoded in "Q:S:size". */ info->reglist.index = QSsize; break; case 0x1: info->qualifier = AARCH64_OPND_QLF_S_H; /* Index encoded in "Q:S:size<1>". */ info->reglist.index = QSsize >> 1; break; case 0x2: if ((QSsize & 0x1) == 0) { info->qualifier = AARCH64_OPND_QLF_S_S; /* Index encoded in "Q:S". */ info->reglist.index = QSsize >> 2; } else { info->qualifier = AARCH64_OPND_QLF_S_D; /* Index encoded in "Q". */ info->reglist.index = QSsize >> 3; if (extract_field (FLD_S, code, 0)) /* UND */ return 0; } break; default: return 0; } info->reglist.has_index = 1; info->reglist.num_regs = 0; /* Number of registers is equal to the number of elements in each structure to be loaded/stored. */ info->reglist.num_regs = get_opcode_dependent_value (inst->opcode); assert (info->reglist.num_regs >= 1 && info->reglist.num_regs <= 4); return 1; } /* Decode fields immh:immb and/or Q for e.g. SSHR <Vd>.<T>, <Vn>.<T>, #<shift> or SSHR <V><d>, <V><n>, #<shift>. */ int aarch64_ext_advsimd_imm_shift (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst) { int pos; aarch64_insn Q, imm, immh; enum aarch64_insn_class iclass = inst->opcode->iclass; immh = extract_field (FLD_immh, code, 0); if (immh == 0) return 0; imm = extract_fields (code, 0, 2, FLD_immh, FLD_immb); pos = 4; /* Get highest set bit in immh. */ while (--pos >= 0 && (immh & 0x8) == 0) immh <<= 1; assert ((iclass == asimdshf || iclass == asisdshf) && (info->type == AARCH64_OPND_IMM_VLSR || info->type == AARCH64_OPND_IMM_VLSL)); if (iclass == asimdshf) { Q = extract_field (FLD_Q, code, 0); /* immh Q <T> 0000 x SEE AdvSIMD modified immediate 0001 0 8B 0001 1 16B 001x 0 4H 001x 1 8H 01xx 0 2S 01xx 1 4S 1xxx 0 RESERVED 1xxx 1 2D */ info->qualifier = get_vreg_qualifier_from_value ((pos << 1) | (int) Q); } else info->qualifier = get_sreg_qualifier_from_value (pos); if (info->type == AARCH64_OPND_IMM_VLSR) /* immh <shift> 0000 SEE AdvSIMD modified immediate 0001 (16-UInt(immh:immb)) 001x (32-UInt(immh:immb)) 01xx (64-UInt(immh:immb)) 1xxx (128-UInt(immh:immb)) */ info->imm.value = (16 << pos) - imm; else /* immh:immb immh <shift> 0000 SEE AdvSIMD modified immediate 0001 (UInt(immh:immb)-8) 001x (UInt(immh:immb)-16) 01xx (UInt(immh:immb)-32) 1xxx (UInt(immh:immb)-64) */ info->imm.value = imm - (8 << pos); return 1; } /* Decode shift immediate for e.g. sshr (imm). */ int aarch64_ext_shll_imm (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { int64_t imm; aarch64_insn val; val = extract_field (FLD_size, code, 0); switch (val) { case 0: imm = 8; break; case 1: imm = 16; break; case 2: imm = 32; break; default: return 0; } info->imm.value = imm; return 1; } /* Decode imm for e.g. BFM <Wd>, <Wn>, #<immr>, #<imms>. value in the field(s) will be extracted as unsigned immediate value. */ int aarch64_ext_imm (const aarch64_operand *self, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { int64_t imm; /* Maximum of two fields to extract. */ assert (self->fields[2] == FLD_NIL); if (self->fields[1] == FLD_NIL) imm = extract_field (self->fields[0], code, 0); else /* e.g. TBZ b5:b40. */ imm = extract_fields (code, 0, 2, self->fields[0], self->fields[1]); if (info->type == AARCH64_OPND_FPIMM) info->imm.is_fp = 1; if (operand_need_sign_extension (self)) imm = sign_extend (imm, get_operand_fields_width (self) - 1); if (operand_need_shift_by_two (self)) imm <<= 2; if (info->type == AARCH64_OPND_ADDR_ADRP) imm <<= 12; info->imm.value = imm; return 1; } /* Decode imm and its shifter for e.g. MOVZ <Wd>, #<imm16>{, LSL #<shift>}. */ int aarch64_ext_imm_half (const aarch64_operand *self, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { aarch64_ext_imm (self, info, code, inst); info->shifter.kind = AARCH64_MOD_LSL; info->shifter.amount = extract_field (FLD_hw, code, 0) << 4; return 1; } /* Decode cmode and "a:b:c:d:e:f:g:h" for e.g. MOVI <Vd>.<T>, #<imm8> {, LSL #<amount>}. */ int aarch64_ext_advsimd_imm_modified (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { uint64_t imm; enum aarch64_opnd_qualifier opnd0_qualifier = inst->operands[0].qualifier; aarch64_field field = {0, 0}; assert (info->idx == 1); if (info->type == AARCH64_OPND_SIMD_FPIMM) info->imm.is_fp = 1; /* a:b:c:d:e:f:g:h */ imm = extract_fields (code, 0, 2, FLD_abc, FLD_defgh); if (!info->imm.is_fp && aarch64_get_qualifier_esize (opnd0_qualifier) == 8) { /* Either MOVI <Dd>, #<imm> or MOVI <Vd>.2D, #<imm>. <imm> is a 64-bit immediate 'aaaaaaaabbbbbbbbccccccccddddddddeeeeeeeeffffffffgggggggghhhhhhhh', encoded in "a:b:c:d:e:f:g:h". */ int i; unsigned abcdefgh = imm; for (imm = 0ull, i = 0; i < 8; i++) if (((abcdefgh >> i) & 0x1) != 0) imm |= 0xffull << (8 * i); } info->imm.value = imm; /* cmode */ info->qualifier = get_expected_qualifier (inst, info->idx); switch (info->qualifier) { case AARCH64_OPND_QLF_NIL: /* no shift */ info->shifter.kind = AARCH64_MOD_NONE; return 1; case AARCH64_OPND_QLF_LSL: /* shift zeros */ info->shifter.kind = AARCH64_MOD_LSL; switch (aarch64_get_qualifier_esize (opnd0_qualifier)) { case 4: gen_sub_field (FLD_cmode, 1, 2, &field); break; /* per word */ case 2: gen_sub_field (FLD_cmode, 1, 1, &field); break; /* per half */ case 1: gen_sub_field (FLD_cmode, 1, 0, &field); break; /* per byte */ default: assert (0); return 0; } /* 00: 0; 01: 8; 10:16; 11:24. */ info->shifter.amount = extract_field_2 (&field, code, 0) << 3; break; case AARCH64_OPND_QLF_MSL: /* shift ones */ info->shifter.kind = AARCH64_MOD_MSL; gen_sub_field (FLD_cmode, 0, 1, &field); /* per word */ info->shifter.amount = extract_field_2 (&field, code, 0) ? 16 : 8; break; default: assert (0); return 0; } return 1; } /* Decode scale for e.g. SCVTF <Dd>, <Wn>, #<fbits>. */ int aarch64_ext_fbits (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { info->imm.value = 64- extract_field (FLD_scale, code, 0); return 1; } /* Decode arithmetic immediate for e.g. SUBS <Wd>, <Wn|WSP>, #<imm> {, <shift>}. */ int aarch64_ext_aimm (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { aarch64_insn value; info->shifter.kind = AARCH64_MOD_LSL; /* shift */ value = extract_field (FLD_shift, code, 0); if (value >= 2) return 0; info->shifter.amount = value ? 12 : 0; /* imm12 (unsigned) */ info->imm.value = extract_field (FLD_imm12, code, 0); return 1; } /* Decode logical immediate for e.g. ORR <Wd|WSP>, <Wn>, #<imm>. */ int aarch64_ext_limm (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { uint64_t imm, mask; uint32_t sf; uint32_t N, R, S; unsigned simd_size; aarch64_insn value; value = extract_fields (code, 0, 3, FLD_N, FLD_immr, FLD_imms); assert (inst->operands[0].qualifier == AARCH64_OPND_QLF_W || inst->operands[0].qualifier == AARCH64_OPND_QLF_X); sf = aarch64_get_qualifier_esize (inst->operands[0].qualifier) != 4; /* value is N:immr:imms. */ S = value & 0x3f; R = (value >> 6) & 0x3f; N = (value >> 12) & 0x1; if (sf == 0 && N == 1) return 0; /* The immediate value is S+1 bits to 1, left rotated by SIMDsize - R (in other words, right rotated by R), then replicated. */ if (N != 0) { simd_size = 64; mask = 0xffffffffffffffffull; } else { switch (S) { case 0x00 ... 0x1f: /* 0xxxxx */ simd_size = 32; break; case 0x20 ... 0x2f: /* 10xxxx */ simd_size = 16; S &= 0xf; break; case 0x30 ... 0x37: /* 110xxx */ simd_size = 8; S &= 0x7; break; case 0x38 ... 0x3b: /* 1110xx */ simd_size = 4; S &= 0x3; break; case 0x3c ... 0x3d: /* 11110x */ simd_size = 2; S &= 0x1; break; default: return 0; } mask = (1ull << simd_size) - 1; /* Top bits are IGNORED. */ R &= simd_size - 1; } /* NOTE: if S = simd_size - 1 we get 0xf..f which is rejected. */ if (S == simd_size - 1) return 0; /* S+1 consecutive bits to 1. */ /* NOTE: S can't be 63 due to detection above. */ imm = (1ull << (S + 1)) - 1; /* Rotate to the left by simd_size - R. */ if (R != 0) imm = ((imm << (simd_size - R)) & mask) | (imm >> R); /* Replicate the value according to SIMD size. */ switch (simd_size) { case 2: imm = (imm << 2) | imm; case 4: imm = (imm << 4) | imm; case 8: imm = (imm << 8) | imm; case 16: imm = (imm << 16) | imm; case 32: imm = (imm << 32) | imm; case 64: break; default: assert (0); return 0; } info->imm.value = sf ? imm : imm & 0xffffffff; return 1; } /* Decode Ft for e.g. STR <Qt>, [<Xn|SP>, <R><m>{, <extend> {<amount>}}] or LDP <Qt1>, <Qt2>, [<Xn|SP>], #<imm>. */ int aarch64_ext_ft (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, const aarch64_insn code, const aarch64_inst *inst) { aarch64_insn value; /* Rt */ info->reg.regno = extract_field (FLD_Rt, code, 0); /* size */ value = extract_field (FLD_ldst_size, code, 0); if (inst->opcode->iclass == ldstpair_indexed || inst->opcode->iclass == ldstnapair_offs || inst->opcode->iclass == ldstpair_off || inst->opcode->iclass == loadlit) { enum aarch64_opnd_qualifier qualifier; switch (value) { case 0: qualifier = AARCH64_OPND_QLF_S_S; break; case 1: qualifier = AARCH64_OPND_QLF_S_D; break; case 2: qualifier = AARCH64_OPND_QLF_S_Q; break; default: return 0; } info->qualifier = qualifier; } else { /* opc1:size */ value = extract_fields (code, 0, 2, FLD_opc1, FLD_ldst_size); if (value > 0x4) return 0; info->qualifier = get_sreg_qualifier_from_value (value); } return 1; } /* Decode the address operand for e.g. STXRB <Ws>, <Wt>, [<Xn|SP>{,#0}]. */ int aarch64_ext_addr_simple (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { /* Rn */ info->addr.base_regno = extract_field (FLD_Rn, code, 0); return 1; } /* Decode the address operand for e.g. STR <Qt>, [<Xn|SP>, <R><m>{, <extend> {<amount>}}]. */ int aarch64_ext_addr_regoff (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, aarch64_insn code, const aarch64_inst *inst) { aarch64_insn S, value; /* Rn */ info->addr.base_regno = extract_field (FLD_Rn, code, 0); /* Rm */ info->addr.offset.regno = extract_field (FLD_Rm, code, 0); /* option */ value = extract_field (FLD_option, code, 0); info->shifter.kind = aarch64_get_operand_modifier_from_value (value, TRUE /* extend_p */); /* Fix-up the shifter kind; although the table-driven approach is efficient, it is slightly inflexible, thus needing this fix-up. */ if (info->shifter.kind == AARCH64_MOD_UXTX) info->shifter.kind = AARCH64_MOD_LSL; /* S */ S = extract_field (FLD_S, code, 0); if (S == 0) { info->shifter.amount = 0; info->shifter.amount_present = 0; } else { int size; /* Need information in other operand(s) to help achieve the decoding from 'S' field. */ info->qualifier = get_expected_qualifier (inst, info->idx); /* Get the size of the data element that is accessed, which may be different from that of the source register size, e.g. in strb/ldrb. */ size = aarch64_get_qualifier_esize (info->qualifier); info->shifter.amount = get_logsz (size); info->shifter.amount_present = 1; } return 1; } /* Decode the address operand for e.g. LDRSW <Xt>, [<Xn|SP>], #<simm>. */ int aarch64_ext_addr_simm (const aarch64_operand *self, aarch64_opnd_info *info, aarch64_insn code, const aarch64_inst *inst) { aarch64_insn imm; info->qualifier = get_expected_qualifier (inst, info->idx); /* Rn */ info->addr.base_regno = extract_field (FLD_Rn, code, 0); /* simm (imm9 or imm7) */ imm = extract_field (self->fields[0], code, 0); info->addr.offset.imm = sign_extend (imm, fields[self->fields[0]].width - 1); if (self->fields[0] == FLD_imm7) /* scaled immediate in ld/st pair instructions. */ info->addr.offset.imm *= aarch64_get_qualifier_esize (info->qualifier); /* qualifier */ if (inst->opcode->iclass == ldst_unscaled || inst->opcode->iclass == ldstnapair_offs || inst->opcode->iclass == ldstpair_off || inst->opcode->iclass == ldst_unpriv) info->addr.writeback = 0; else { /* pre/post- index */ info->addr.writeback = 1; if (extract_field (self->fields[1], code, 0) == 1) info->addr.preind = 1; else info->addr.postind = 1; } return 1; } /* Decode the address operand for e.g. LDRSW <Xt>, [<Xn|SP>{, #<simm>}]. */ int aarch64_ext_addr_uimm12 (const aarch64_operand *self, aarch64_opnd_info *info, aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { int shift; info->qualifier = get_expected_qualifier (inst, info->idx); shift = get_logsz (aarch64_get_qualifier_esize (info->qualifier)); /* Rn */ info->addr.base_regno = extract_field (self->fields[0], code, 0); /* uimm12 */ info->addr.offset.imm = extract_field (self->fields[1], code, 0) << shift; return 1; } /* Decode the address operand for e.g. LD1 {<Vt>.<T>, <Vt2>.<T>, <Vt3>.<T>}, [<Xn|SP>], <Xm|#<amount>>. */ int aarch64_ext_simd_addr_post (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, aarch64_insn code, const aarch64_inst *inst) { /* The opcode dependent area stores the number of elements in each structure to be loaded/stored. */ int is_ld1r = get_opcode_dependent_value (inst->opcode) == 1; /* Rn */ info->addr.base_regno = extract_field (FLD_Rn, code, 0); /* Rm | #<amount> */ info->addr.offset.regno = extract_field (FLD_Rm, code, 0); if (info->addr.offset.regno == 31) { if (inst->opcode->operands[0] == AARCH64_OPND_LVt_AL) /* Special handling of loading single structure to all lane. */ info->addr.offset.imm = (is_ld1r ? 1 : inst->operands[0].reglist.num_regs) * aarch64_get_qualifier_esize (inst->operands[0].qualifier); else info->addr.offset.imm = inst->operands[0].reglist.num_regs * aarch64_get_qualifier_esize (inst->operands[0].qualifier) * aarch64_get_qualifier_nelem (inst->operands[0].qualifier); } else info->addr.offset.is_reg = 1; info->addr.writeback = 1; return 1; } /* Decode the condition operand for e.g. CSEL <Xd>, <Xn>, <Xm>, <cond>. */ int aarch64_ext_cond (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { aarch64_insn value; /* cond */ value = extract_field (FLD_cond, code, 0); info->cond = get_cond_from_value (value); return 1; } /* Decode the system register operand for e.g. MRS <Xt>, <systemreg>. */ int aarch64_ext_sysreg (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { /* op0:op1:CRn:CRm:op2 */ info->sysreg = extract_fields (code, 0, 5, FLD_op0, FLD_op1, FLD_CRn, FLD_CRm, FLD_op2); return 1; } /* Decode the PSTATE field operand for e.g. MSR <pstatefield>, #<imm>. */ int aarch64_ext_pstatefield (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { int i; /* op1:op2 */ info->pstatefield = extract_fields (code, 0, 2, FLD_op1, FLD_op2); for (i = 0; aarch64_pstatefields[i].name != NULL; ++i) if (aarch64_pstatefields[i].value == (aarch64_insn)info->pstatefield) return 1; /* Reserved value in <pstatefield>. */ return 0; } /* Decode the system instruction op operand for e.g. AT <at_op>, <Xt>. */ int aarch64_ext_sysins_op (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { int i; aarch64_insn value; const aarch64_sys_ins_reg *sysins_ops; /* op0:op1:CRn:CRm:op2 */ value = extract_fields (code, 0, 5, FLD_op0, FLD_op1, FLD_CRn, FLD_CRm, FLD_op2); switch (info->type) { case AARCH64_OPND_SYSREG_AT: sysins_ops = aarch64_sys_regs_at; break; case AARCH64_OPND_SYSREG_DC: sysins_ops = aarch64_sys_regs_dc; break; case AARCH64_OPND_SYSREG_IC: sysins_ops = aarch64_sys_regs_ic; break; case AARCH64_OPND_SYSREG_TLBI: sysins_ops = aarch64_sys_regs_tlbi; break; default: assert (0); return 0; } for (i = 0; sysins_ops[i].template != NULL; ++i) if (sysins_ops[i].value == value) { info->sysins_op = sysins_ops + i; DEBUG_TRACE ("%s found value: %x, has_xt: %d, i: %d.", info->sysins_op->template, (unsigned)info->sysins_op->value, info->sysins_op->has_xt, i); return 1; } return 0; } /* Decode the memory barrier option operand for e.g. DMB <option>|#<imm>. */ int aarch64_ext_barrier (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { /* CRm */ info->barrier = aarch64_barrier_options + extract_field (FLD_CRm, code, 0); return 1; } /* Decode the prefetch operation option operand for e.g. PRFM <prfop>, [<Xn|SP>{, #<pimm>}]. */ int aarch64_ext_prfop (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { /* prfop in Rt */ info->prfop = aarch64_prfops + extract_field (FLD_Rt, code, 0); return 1; } /* Decode the extended register operand for e.g. STR <Qt>, [<Xn|SP>, <R><m>{, <extend> {<amount>}}]. */ int aarch64_ext_reg_extended (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { aarch64_insn value; /* Rm */ info->reg.regno = extract_field (FLD_Rm, code, 0); /* option */ value = extract_field (FLD_option, code, 0); info->shifter.kind = aarch64_get_operand_modifier_from_value (value, TRUE /* extend_p */); /* imm3 */ info->shifter.amount = extract_field (FLD_imm3, code, 0); /* This makes the constraint checking happy. */ info->shifter.operator_present = 1; /* Assume inst->operands[0].qualifier has been resolved. */ assert (inst->operands[0].qualifier != AARCH64_OPND_QLF_NIL); info->qualifier = AARCH64_OPND_QLF_W; if (inst->operands[0].qualifier == AARCH64_OPND_QLF_X && (info->shifter.kind == AARCH64_MOD_UXTX || info->shifter.kind == AARCH64_MOD_SXTX)) info->qualifier = AARCH64_OPND_QLF_X; return 1; } /* Decode the shifted register operand for e.g. SUBS <Xd>, <Xn>, <Xm> {, <shift> #<amount>}. */ int aarch64_ext_reg_shifted (const aarch64_operand *self ATTRIBUTE_UNUSED, aarch64_opnd_info *info, aarch64_insn code, const aarch64_inst *inst ATTRIBUTE_UNUSED) { aarch64_insn value; /* Rm */ info->reg.regno = extract_field (FLD_Rm, code, 0); /* shift */ value = extract_field (FLD_shift, code, 0); info->shifter.kind = aarch64_get_operand_modifier_from_value (value, FALSE /* extend_p */); if (info->shifter.kind == AARCH64_MOD_ROR && inst->opcode->iclass != log_shift) /* ROR is not available for the shifted register operand in arithmetic instructions. */ return 0; /* imm6 */ info->shifter.amount = extract_field (FLD_imm6, code, 0); /* This makes the constraint checking happy. */ info->shifter.operator_present = 1; return 1; } /* Bitfields that are commonly used to encode certain operands' information may be partially used as part of the base opcode in some instructions. For example, the bit 1 of the field 'size' in FCVTXN <Vb><d>, <Va><n> is actually part of the base opcode, while only size<0> is available for encoding the register type. Another example is the AdvSIMD instruction ORR (register), in which the field 'size' is also used for the base opcode, leaving only the field 'Q' available to encode the vector register arrangement specifier '8B' or '16B'. This function tries to deduce the qualifier from the value of partially constrained field(s). Given the VALUE of such a field or fields, the qualifiers CANDIDATES and the MASK (indicating which bits are valid for operand encoding), the function returns the matching qualifier or AARCH64_OPND_QLF_NIL if nothing matches. N.B. CANDIDATES is a group of possible qualifiers that are valid for one operand; it has a maximum of AARCH64_MAX_QLF_SEQ_NUM qualifiers and may end with AARCH64_OPND_QLF_NIL. */ static enum aarch64_opnd_qualifier get_qualifier_from_partial_encoding (aarch64_insn value, const enum aarch64_opnd_qualifier* \ candidates, aarch64_insn mask) { int i; DEBUG_TRACE ("enter with value: %d, mask: %d", (int)value, (int)mask); for (i = 0; i < AARCH64_MAX_QLF_SEQ_NUM; ++i) { aarch64_insn standard_value; if (candidates[i] == AARCH64_OPND_QLF_NIL) break; standard_value = aarch64_get_qualifier_standard_value (candidates[i]); if ((standard_value & mask) == (value & mask)) return candidates[i]; } return AARCH64_OPND_QLF_NIL; } /* Given a list of qualifier sequences, return all possible valid qualifiers for operand IDX in QUALIFIERS. Assume QUALIFIERS is an array whose length is large enough. */ static void get_operand_possible_qualifiers (int idx, const aarch64_opnd_qualifier_seq_t *list, enum aarch64_opnd_qualifier *qualifiers) { int i; for (i = 0; i < AARCH64_MAX_QLF_SEQ_NUM; ++i) if ((qualifiers[i] = list[i][idx]) == AARCH64_OPND_QLF_NIL) break; } /* Decode the size Q field for e.g. SHADD. We tag one operand with the qualifer according to the code; whether the qualifier is valid for this opcode or not, it is the duty of the semantic checking. */ static int decode_sizeq (aarch64_inst *inst) { int idx; enum aarch64_opnd_qualifier qualifier; aarch64_insn code; aarch64_insn value, mask; enum aarch64_field_kind fld_sz; enum aarch64_opnd_qualifier candidates[AARCH64_MAX_QLF_SEQ_NUM]; if (inst->opcode->iclass == asisdlse || inst->opcode->iclass == asisdlsep || inst->opcode->iclass == asisdlso || inst->opcode->iclass == asisdlsop) fld_sz = FLD_vldst_size; else fld_sz = FLD_size; code = inst->value; value = extract_fields (code, inst->opcode->mask, 2, fld_sz, FLD_Q); /* Obtain the info that which bits of fields Q and size are actually available for operand encoding. Opcodes like FMAXNM and FMLA have size[1] unavailable. */ mask = extract_fields (~inst->opcode->mask, 0, 2, fld_sz, FLD_Q); /* The index of the operand we are going to tag a qualifier and the qualifer itself are reasoned from the value of the size and Q fields and the possible valid qualifier lists. */ idx = aarch64_select_operand_for_sizeq_field_coding (inst->opcode); DEBUG_TRACE ("key idx: %d", idx); /* For most related instruciton, size:Q are fully available for operand encoding. */ if (mask == 0x7) { inst->operands[idx].qualifier = get_vreg_qualifier_from_value (value); return 1; } get_operand_possible_qualifiers (idx, inst->opcode->qualifiers_list, candidates); #ifdef DEBUG_AARCH64 if (debug_dump) { int i; for (i = 0; candidates[i] != AARCH64_OPND_QLF_NIL && i < AARCH64_MAX_QLF_SEQ_NUM; ++i) DEBUG_TRACE ("qualifier %d: %s", i, aarch64_get_qualifier_name(candidates[i])); DEBUG_TRACE ("%d, %d", (int)value, (int)mask); } #endif /* DEBUG_AARCH64 */ qualifier = get_qualifier_from_partial_encoding (value, candidates, mask); if (qualifier == AARCH64_OPND_QLF_NIL) return 0; inst->operands[idx].qualifier = qualifier; return 1; } /* Decode size[0]:Q, i.e. bit 22 and bit 30, for e.g. FCVTN<Q> <Vd>.<Tb>, <Vn>.<Ta>. */ static int decode_asimd_fcvt (aarch64_inst *inst) { aarch64_field field = {0, 0}; aarch64_insn value; enum aarch64_opnd_qualifier qualifier; gen_sub_field (FLD_size, 0, 1, &field); value = extract_field_2 (&field, inst->value, 0); qualifier = value == 0 ? AARCH64_OPND_QLF_V_4S : AARCH64_OPND_QLF_V_2D; switch (inst->opcode->op) { case OP_FCVTN: case OP_FCVTN2: /* FCVTN<Q> <Vd>.<Tb>, <Vn>.<Ta>. */ inst->operands[1].qualifier = qualifier; break; case OP_FCVTL: case OP_FCVTL2: /* FCVTL<Q> <Vd>.<Ta>, <Vn>.<Tb>. */ inst->operands[0].qualifier = qualifier; break; default: assert (0); return 0; } return 1; } /* Decode size[0], i.e. bit 22, for e.g. FCVTXN <Vb><d>, <Va><n>. */ static int decode_asisd_fcvtxn (aarch64_inst *inst) { aarch64_field field = {0, 0}; gen_sub_field (FLD_size, 0, 1, &field); if (!extract_field_2 (&field, inst->value, 0)) return 0; inst->operands[0].qualifier = AARCH64_OPND_QLF_S_S; return 1; } /* Decode the 'opc' field for e.g. FCVT <Dd>, <Sn>. */ static int decode_fcvt (aarch64_inst *inst) { enum aarch64_opnd_qualifier qualifier; aarch64_insn value; const aarch64_field field = {15, 2}; /* opc dstsize */ value = extract_field_2 (&field, inst->value, 0); switch (value) { case 0: qualifier = AARCH64_OPND_QLF_S_S; break; case 1: qualifier = AARCH64_OPND_QLF_S_D; break; case 3: qualifier = AARCH64_OPND_QLF_S_H; break; default: return 0; } inst->operands[0].qualifier = qualifier; return 1; } /* Do miscellaneous decodings that are not common enough to be driven by flags. */ static int do_misc_decoding (aarch64_inst *inst) { switch (inst->opcode->op) { case OP_FCVT: return decode_fcvt (inst); case OP_FCVTN: case OP_FCVTN2: case OP_FCVTL: case OP_FCVTL2: return decode_asimd_fcvt (inst); case OP_FCVTXN_S: return decode_asisd_fcvtxn (inst); default: return 0; } } /* Opcodes that have fields shared by multiple operands are usually flagged with flags. In this function, we detect such flags, decode the related field(s) and store the information in one of the related operands. The 'one' operand is not any operand but one of the operands that can accommadate all the information that has been decoded. */ static int do_special_decoding (aarch64_inst *inst) { int idx; aarch64_insn value; /* Condition for truly conditional executed instructions, e.g. b.cond. */ if (inst->opcode->flags & F_COND) { value = extract_field (FLD_cond2, inst->value, 0); inst->cond = get_cond_from_value (value); } /* 'sf' field. */ if (inst->opcode->flags & F_SF) { idx = select_operand_for_sf_field_coding (inst->opcode); value = extract_field (FLD_sf, inst->value, 0); inst->operands[idx].qualifier = get_greg_qualifier_from_value (value); if ((inst->opcode->flags & F_N) && extract_field (FLD_N, inst->value, 0) != value) return 0; } /* size:Q fields. */ if (inst->opcode->flags & F_SIZEQ) return decode_sizeq (inst); if (inst->opcode->flags & F_FPTYPE) { idx = select_operand_for_fptype_field_coding (inst->opcode); value = extract_field (FLD_type, inst->value, 0); switch (value) { case 0: inst->operands[idx].qualifier = AARCH64_OPND_QLF_S_S; break; case 1: inst->operands[idx].qualifier = AARCH64_OPND_QLF_S_D; break; case 3: inst->operands[idx].qualifier = AARCH64_OPND_QLF_S_H; break; default: return 0; } } if (inst->opcode->flags & F_SSIZE) { /* N.B. some opcodes like FCMGT <V><d>, <V><n>, #0 have the size[1] as part of the base opcode. */ aarch64_insn mask; enum aarch64_opnd_qualifier candidates[AARCH64_MAX_QLF_SEQ_NUM]; idx = select_operand_for_scalar_size_field_coding (inst->opcode); value = extract_field (FLD_size, inst->value, inst->opcode->mask); mask = extract_field (FLD_size, ~inst->opcode->mask, 0); /* For most related instruciton, the 'size' field is fully available for operand encoding. */ if (mask == 0x3) inst->operands[idx].qualifier = get_sreg_qualifier_from_value (value); else { get_operand_possible_qualifiers (idx, inst->opcode->qualifiers_list, candidates); inst->operands[idx].qualifier = get_qualifier_from_partial_encoding (value, candidates, mask); } } if (inst->opcode->flags & F_T) { /* Num of consecutive '0's on the right side of imm5<3:0>. */ int num = 0; unsigned val, Q; assert (aarch64_get_operand_class (inst->opcode->operands[0]) == AARCH64_OPND_CLASS_SIMD_REG); /* imm5<3:0> q <t> 0000 x reserved xxx1 0 8b xxx1 1 16b xx10 0 4h xx10 1 8h x100 0 2s x100 1 4s 1000 0 reserved 1000 1 2d */ val = extract_field (FLD_imm5, inst->value, 0); while ((val & 0x1) == 0 && ++num <= 3) val >>= 1; if (num > 3) return 0; Q = (unsigned) extract_field (FLD_Q, inst->value, inst->opcode->mask); inst->operands[0].qualifier = get_vreg_qualifier_from_value ((num << 1) | Q); } if (inst->opcode->flags & F_GPRSIZE_IN_Q) { /* Use Rt to encode in the case of e.g. STXP <Ws>, <Xt1>, <Xt2>, [<Xn|SP>{,#0}]. */ idx = aarch64_operand_index (inst->opcode->operands, AARCH64_OPND_Rt); if (idx == -1) { /* Otherwise use the result operand, which has to be a integer register. */ assert (aarch64_get_operand_class (inst->opcode->operands[0]) == AARCH64_OPND_CLASS_INT_REG); idx = 0; } assert (idx == 0 || idx == 1); value = extract_field (FLD_Q, inst->value, 0); inst->operands[idx].qualifier = get_greg_qualifier_from_value (value); } if (inst->opcode->flags & F_LDS_SIZE) { aarch64_field field = {0, 0}; assert (aarch64_get_operand_class (inst->opcode->operands[0]) == AARCH64_OPND_CLASS_INT_REG); gen_sub_field (FLD_opc, 0, 1, &field); value = extract_field_2 (&field, inst->value, 0); inst->operands[0].qualifier = value ? AARCH64_OPND_QLF_W : AARCH64_OPND_QLF_X; } /* Miscellaneous decoding; done as the last step. */ if (inst->opcode->flags & F_MISC) return do_misc_decoding (inst); return 1; } /* Converters converting a real opcode instruction to its alias form. */ /* ROR <Wd>, <Ws>, #<shift> is equivalent to: EXTR <Wd>, <Ws>, <Ws>, #<shift>. */ static int convert_extr_to_ror (aarch64_inst *inst) { if (inst->operands[1].reg.regno == inst->operands[2].reg.regno) { copy_operand_info (inst, 2, 3); inst->operands[3].type = AARCH64_OPND_NIL; return 1; } return 0; } /* Convert UBFM <Xd>, <Xn>, #<shift>, #63. to LSR <Xd>, <Xn>, #<shift>. */ static int convert_bfm_to_sr (aarch64_inst *inst) { int64_t imms, val; imms = inst->operands[3].imm.value; val = inst->operands[2].qualifier == AARCH64_OPND_QLF_imm_0_31 ? 31 : 63; if (imms == val) { inst->operands[3].type = AARCH64_OPND_NIL; return 1; } return 0; } /* Convert MOV to ORR. */ static int convert_orr_to_mov (aarch64_inst *inst) { /* MOV <Vd>.<T>, <Vn>.<T> is equivalent to: ORR <Vd>.<T>, <Vn>.<T>, <Vn>.<T>. */ if (inst->operands[1].reg.regno == inst->operands[2].reg.regno) { inst->operands[2].type = AARCH64_OPND_NIL; return 1; } return 0; } /* When <imms> >= <immr>, the instruction written: SBFX <Xd>, <Xn>, #<lsb>, #<width> is equivalent to: SBFM <Xd>, <Xn>, #<lsb>, #(<lsb>+<width>-1). */ static int convert_bfm_to_bfx (aarch64_inst *inst) { int64_t immr, imms; immr = inst->operands[2].imm.value; imms = inst->operands[3].imm.value; if (imms >= immr) { int64_t lsb = immr; inst->operands[2].imm.value = lsb; inst->operands[3].imm.value = imms + 1 - lsb; /* The two opcodes have different qualifiers for the immediate operands; reset to help the checking. */ reset_operand_qualifier (inst, 2); reset_operand_qualifier (inst, 3); return 1; } return 0; } /* When <imms> < <immr>, the instruction written: SBFIZ <Xd>, <Xn>, #<lsb>, #<width> is equivalent to: SBFM <Xd>, <Xn>, #((64-<lsb>)&0x3f), #(<width>-1). */ static int convert_bfm_to_bfi (aarch64_inst *inst) { int64_t immr, imms, val; immr = inst->operands[2].imm.value; imms = inst->operands[3].imm.value; val = inst->operands[2].qualifier == AARCH64_OPND_QLF_imm_0_31 ? 32 : 64; if (imms < immr) { inst->operands[2].imm.value = (val - immr) & (val - 1); inst->operands[3].imm.value = imms + 1; /* The two opcodes have different qualifiers for the immediate operands; reset to help the checking. */ reset_operand_qualifier (inst, 2); reset_operand_qualifier (inst, 3); return 1; } return 0; } /* The instruction written: LSL <Xd>, <Xn>, #<shift> is equivalent to: UBFM <Xd>, <Xn>, #((64-<shift>)&0x3f), #(63-<shift>). */ static int convert_ubfm_to_lsl (aarch64_inst *inst) { int64_t immr = inst->operands[2].imm.value; int64_t imms = inst->operands[3].imm.value; int64_t val = inst->operands[2].qualifier == AARCH64_OPND_QLF_imm_0_31 ? 31 : 63; if ((immr == 0 && imms == val) || immr == imms + 1) { inst->operands[3].type = AARCH64_OPND_NIL; inst->operands[2].imm.value = val - imms; return 1; } return 0; } /* CINC <Wd>, <Wn>, <cond> is equivalent to: CSINC <Wd>, <Wn>, <Wn>, invert(<cond>). */ static int convert_from_csel (aarch64_inst *inst) { if (inst->operands[1].reg.regno == inst->operands[2].reg.regno) { copy_operand_info (inst, 2, 3); inst->operands[2].cond = get_inverted_cond (inst->operands[3].cond); inst->operands[3].type = AARCH64_OPND_NIL; return 1; } return 0; } /* CSET <Wd>, <cond> is equivalent to: CSINC <Wd>, WZR, WZR, invert(<cond>). */ static int convert_csinc_to_cset (aarch64_inst *inst) { if (inst->operands[1].reg.regno == 0x1f && inst->operands[2].reg.regno == 0x1f) { copy_operand_info (inst, 1, 3); inst->operands[1].cond = get_inverted_cond (inst->operands[3].cond); inst->operands[3].type = AARCH64_OPND_NIL; inst->operands[2].type = AARCH64_OPND_NIL; return 1; } return 0; } /* MOV <Wd>, #<imm> is equivalent to: MOVZ <Wd>, #<imm16>, LSL #<shift>. A disassembler may output ORR, MOVZ and MOVN as a MOV mnemonic, except when ORR has an immediate that could be generated by a MOVZ or MOVN instruction, or where a MOVN has an immediate that could be encoded by MOVZ, or where MOVZ/MOVN #0 have a shift amount other than LSL #0, in which case the machine-instruction mnemonic must be used. */ static int convert_movewide_to_mov (aarch64_inst *inst) { uint64_t value = inst->operands[1].imm.value; /* MOVZ/MOVN #0 have a shift amount other than LSL #0. */ if (value == 0 && inst->operands[1].shifter.amount != 0) return 0; inst->operands[1].type = AARCH64_OPND_IMM_MOV; inst->operands[1].shifter.kind = AARCH64_MOD_NONE; value <<= inst->operands[1].shifter.amount; /* As an alias convertor, it has to be clear that the INST->OPCODE is the opcode of the real instruction. */ if (inst->opcode->op == OP_MOVN) { int is32 = inst->operands[0].qualifier == AARCH64_OPND_QLF_W; value = ~value; /* A MOVN has an immediate that could be encoded by MOVZ. */ if (aarch64_wide_constant_p (value, is32, NULL) == TRUE) return 0; } inst->operands[1].imm.value = value; inst->operands[1].shifter.amount = 0; return 1; } /* MOV <Wd>, #<imm> is equivalent to: ORR <Wd>, WZR, #<imm>. A disassembler may output ORR, MOVZ and MOVN as a MOV mnemonic, except when ORR has an immediate that could be generated by a MOVZ or MOVN instruction, or where a MOVN has an immediate that could be encoded by MOVZ, or where MOVZ/MOVN #0 have a shift amount other than LSL #0, in which case the machine-instruction mnemonic must be used. */ static int convert_movebitmask_to_mov (aarch64_inst *inst) { int is32; uint64_t value; /* Should have been assured by the base opcode value. */ assert (inst->operands[1].reg.regno == 0x1f); copy_operand_info (inst, 1, 2); is32 = inst->operands[0].qualifier == AARCH64_OPND_QLF_W; inst->operands[1].type = AARCH64_OPND_IMM_MOV; value = inst->operands[1].imm.value; /* ORR has an immediate that could be generated by a MOVZ or MOVN instruction. */ if (inst->operands[0].reg.regno != 0x1f && (aarch64_wide_constant_p (value, is32, NULL) == TRUE || aarch64_wide_constant_p (~value, is32, NULL) == TRUE)) return 0; inst->operands[2].type = AARCH64_OPND_NIL; return 1; } /* Some alias opcodes are disassembled by being converted from their real-form. N.B. INST->OPCODE is the real opcode rather than the alias. */ static int convert_to_alias (aarch64_inst *inst, const aarch64_opcode *alias) { switch (alias->op) { case OP_ASR_IMM: case OP_LSR_IMM: return convert_bfm_to_sr (inst); case OP_LSL_IMM: return convert_ubfm_to_lsl (inst); case OP_CINC: case OP_CINV: case OP_CNEG: return convert_from_csel (inst); case OP_CSET: case OP_CSETM: return convert_csinc_to_cset (inst); case OP_UBFX: case OP_BFXIL: case OP_SBFX: return convert_bfm_to_bfx (inst); case OP_SBFIZ: case OP_BFI: case OP_UBFIZ: return convert_bfm_to_bfi (inst); case OP_MOV_V: return convert_orr_to_mov (inst); case OP_MOV_IMM_WIDE: case OP_MOV_IMM_WIDEN: return convert_movewide_to_mov (inst); case OP_MOV_IMM_LOG: return convert_movebitmask_to_mov (inst); case OP_ROR_IMM: return convert_extr_to_ror (inst); default: return 0; } } static int aarch64_opcode_decode (const aarch64_opcode *, const aarch64_insn, aarch64_inst *, int); /* Given the instruction information in *INST, check if the instruction has any alias form that can be used to represent *INST. If the answer is yes, update *INST to be in the form of the determined alias. */ /* In the opcode description table, the following flags are used in opcode entries to help establish the relations between the real and alias opcodes: F_ALIAS: opcode is an alias F_HAS_ALIAS: opcode has alias(es) F_P1 F_P2 F_P3: Disassembly preference priority 1-3 (the larger the higher). If nothing is specified, it is the priority 0 by default, i.e. the lowest priority. Although the relation between the machine and the alias instructions are not explicitly described, it can be easily determined from the base opcode values, masks and the flags F_ALIAS and F_HAS_ALIAS in their opcode description entries: The mask of an alias opcode must be equal to or a super-set (i.e. more constrained) of that of the aliased opcode; so is the base opcode value. if (opcode_has_alias (real) && alias_opcode_p (opcode) && (opcode->mask & real->mask) == real->mask && (real->mask & opcode->opcode) == (real->mask & real->opcode)) then OPCODE is an alias of, and only of, the REAL instruction The alias relationship is forced flat-structured to keep related algorithm simple; an opcode entry cannot be flagged with both F_ALIAS and F_HAS_ALIAS. During the disassembling, the decoding decision tree (in opcodes/aarch64-dis-2.c) always returns an machine instruction opcode entry; if the decoding of such a machine instruction succeeds (and -Mno-aliases is not specified), the disassembler will check whether there is any alias instruction exists for this real instruction. If there is, the disassembler will try to disassemble the 32-bit binary again using the alias's rule, or try to convert the IR to the form of the alias. In the case of the multiple aliases, the aliases are tried one by one from the highest priority (currently the flag F_P3) to the lowest priority (no priority flag), and the first succeeds first adopted. You may ask why there is a need for the conversion of IR from one form to another in handling certain aliases. This is because on one hand it avoids adding more operand code to handle unusual encoding/decoding; on other hand, during the disassembling, the conversion is an effective approach to check the condition of an alias (as an alias may be adopted only if certain conditions are met). In order to speed up the alias opcode lookup, aarch64-gen has preprocessed aarch64_opcode_table and generated aarch64_find_alias_opcode and aarch64_find_next_alias_opcode (in opcodes/aarch64-dis-2.c) to help. */ static void determine_disassembling_preference (struct aarch64_inst *inst) { const aarch64_opcode *opcode; const aarch64_opcode *alias; opcode = inst->opcode; /* This opcode does not have an alias, so use itself. */ if (opcode_has_alias (opcode) == FALSE) return; alias = aarch64_find_alias_opcode (opcode); assert (alias); #ifdef DEBUG_AARCH64 if (debug_dump) { const aarch64_opcode *tmp = alias; printf ("#### LIST orderd: "); while (tmp) { printf ("%s, ", tmp->name); tmp = aarch64_find_next_alias_opcode (tmp); } printf ("\n"); } #endif /* DEBUG_AARCH64 */ for (; alias; alias = aarch64_find_next_alias_opcode (alias)) { DEBUG_TRACE ("try %s", alias->name); assert (alias_opcode_p (alias)); /* An alias can be a pseudo opcode which will never be used in the disassembly, e.g. BIC logical immediate is such a pseudo opcode aliasing AND. */ if (pseudo_opcode_p (alias)) { DEBUG_TRACE ("skip pseudo %s", alias->name); continue; } if ((inst->value & alias->mask) != alias->opcode) { DEBUG_TRACE ("skip %s as base opcode not match", alias->name); continue; } /* No need to do any complicated transformation on operands, if the alias opcode does not have any operand. */ if (aarch64_num_of_operands (alias) == 0 && alias->opcode == inst->value) { DEBUG_TRACE ("succeed with 0-operand opcode %s", alias->name); aarch64_replace_opcode (inst, alias); return; } if (alias->flags & F_CONV) { aarch64_inst copy; memcpy (©, inst, sizeof (aarch64_inst)); /* ALIAS is the preference as long as the instruction can be successfully converted to the form of ALIAS. */ if (convert_to_alias (©, alias) == 1) { aarch64_replace_opcode (©, alias); assert (aarch64_match_operands_constraint (©, NULL)); DEBUG_TRACE ("succeed with %s via conversion", alias->name); memcpy (inst, ©, sizeof (aarch64_inst)); return; } } else { /* Directly decode the alias opcode. */ aarch64_inst temp; memset (&temp, '\0', sizeof (aarch64_inst)); if (aarch64_opcode_decode (alias, inst->value, &temp, 1) == 1) { DEBUG_TRACE ("succeed with %s via direct decoding", alias->name); memcpy (inst, &temp, sizeof (aarch64_inst)); return; } } } } /* Decode the CODE according to OPCODE; fill INST. Return 0 if the decoding fails, which meanes that CODE is not an instruction of OPCODE; otherwise return 1. If OPCODE has alias(es) and NOALIASES_P is 0, an alias opcode may be determined and used to disassemble CODE; this is done just before the return. */ static int aarch64_opcode_decode (const aarch64_opcode *opcode, const aarch64_insn code, aarch64_inst *inst, int noaliases_p) { int i; DEBUG_TRACE ("enter with %s", opcode->name); assert (opcode && inst); /* Check the base opcode. */ if ((code & opcode->mask) != (opcode->opcode & opcode->mask)) { DEBUG_TRACE ("base opcode match FAIL"); goto decode_fail; } /* Clear inst. */ memset (inst, '\0', sizeof (aarch64_inst)); inst->opcode = opcode; inst->value = code; /* Assign operand codes and indexes. */ for (i = 0; i < AARCH64_MAX_OPND_NUM; ++i) { if (opcode->operands[i] == AARCH64_OPND_NIL) break; inst->operands[i].type = opcode->operands[i]; inst->operands[i].idx = i; } /* Call the opcode decoder indicated by flags. */ if (opcode_has_special_coder (opcode) && do_special_decoding (inst) == 0) { DEBUG_TRACE ("opcode flag-based decoder FAIL"); goto decode_fail; } /* Call operand decoders. */ for (i = 0; i < AARCH64_MAX_OPND_NUM; ++i) { const aarch64_operand *opnd; enum aarch64_opnd type; type = opcode->operands[i]; if (type == AARCH64_OPND_NIL) break; opnd = &aarch64_operands[type]; if (operand_has_extractor (opnd) && (! aarch64_extract_operand (opnd, &inst->operands[i], code, inst))) { DEBUG_TRACE ("operand decoder FAIL at operand %d", i); goto decode_fail; } } /* Match the qualifiers. */ if (aarch64_match_operands_constraint (inst, NULL) == 1) { /* Arriving here, the CODE has been determined as a valid instruction of OPCODE and *INST has been filled with information of this OPCODE instruction. Before the return, check if the instruction has any alias and should be disassembled in the form of its alias instead. If the answer is yes, *INST will be updated. */ if (!noaliases_p) determine_disassembling_preference (inst); DEBUG_TRACE ("SUCCESS"); return 1; } else { DEBUG_TRACE ("constraint matching FAIL"); } decode_fail: return 0; } /* This does some user-friendly fix-up to *INST. It is currently focus on the adjustment of qualifiers to help the printed instruction recognized/understood more easily. */ static void user_friendly_fixup (aarch64_inst *inst) { switch (inst->opcode->iclass) { case testbranch: /* TBNZ Xn|Wn, #uimm6, label Test and Branch Not Zero: conditionally jumps to label if bit number uimm6 in register Xn is not zero. The bit number implies the width of the register, which may be written and should be disassembled as Wn if uimm is less than 32. Limited to a branch offset range of +/- 32KiB. */ if (inst->operands[1].imm.value < 32) inst->operands[0].qualifier = AARCH64_OPND_QLF_W; break; default: break; } } /* Decode INSN and fill in *INST the instruction information. */ static int disas_aarch64_insn (uint64_t pc ATTRIBUTE_UNUSED, uint32_t insn, aarch64_inst *inst) { const aarch64_opcode *opcode = aarch64_opcode_lookup (insn); #ifdef DEBUG_AARCH64 if (debug_dump) { const aarch64_opcode *tmp = opcode; printf ("\n"); DEBUG_TRACE ("opcode lookup:"); while (tmp != NULL) { aarch64_verbose (" %s", tmp->name); tmp = aarch64_find_next_opcode (tmp); } } #endif /* DEBUG_AARCH64 */ /* A list of opcodes may have been found, as aarch64_opcode_lookup cannot distinguish some opcodes, e.g. SSHR and MOVI, which almost share the same opcode field and value, apart from the difference that one of them has an extra field as part of the opcode, but such a field is used for operand encoding in other opcode(s) ('immh' in the case of the example). */ while (opcode != NULL) { /* But only one opcode can be decoded successfully for, as the decoding routine will check the constraint carefully. */ if (aarch64_opcode_decode (opcode, insn, inst, no_aliases) == 1) return ERR_OK; opcode = aarch64_find_next_opcode (opcode); } return ERR_UND; } /* Print operands. */ static void print_operands (bfd_vma pc, const aarch64_opcode *opcode, const aarch64_opnd_info *opnds, struct disassemble_info *info) { int i, pcrel_p, num_printed; for (i = 0, num_printed = 0; i < AARCH64_MAX_OPND_NUM; ++i) { const size_t size = 128; char str[size]; /* We regard the opcode operand info more, however we also look into the inst->operands to support the disassembling of the optional operand. The two operand code should be the same in all cases, apart from when the operand can be optional. */ if (opcode->operands[i] == AARCH64_OPND_NIL || opnds[i].type == AARCH64_OPND_NIL) break; /* Generate the operand string in STR. */ aarch64_print_operand (str, size, pc, opcode, opnds, i, &pcrel_p, &info->target); /* Print the delimiter (taking account of omitted operand(s)). */ if (str[0] != '\0') (*info->fprintf_func) (info->stream, "%s", num_printed++ == 0 ? "\t" : ", "); /* Print the operand. */ if (pcrel_p) (*info->print_address_func) (info->target, info); else (*info->fprintf_func) (info->stream, "%s", str); } } /* Print the instruction mnemonic name. */ static void print_mnemonic_name (const aarch64_inst *inst, struct disassemble_info *info) { if (inst->opcode->flags & F_COND) { /* For instructions that are truly conditionally executed, e.g. b.cond, prepare the full mnemonic name with the corresponding condition suffix. */ char name[8], *ptr; size_t len; ptr = strchr (inst->opcode->name, '.'); assert (ptr && inst->cond); len = ptr - inst->opcode->name; assert (len < 8); strncpy (name, inst->opcode->name, len); name [len] = '\0'; (*info->fprintf_func) (info->stream, "%s.%s", name, inst->cond->names[0]); } else (*info->fprintf_func) (info->stream, "%s", inst->opcode->name); } /* Print the instruction according to *INST. */ static void print_aarch64_insn (bfd_vma pc, const aarch64_inst *inst, struct disassemble_info *info) { print_mnemonic_name (inst, info); print_operands (pc, inst->opcode, inst->operands, info); } /* Entry-point of the instruction disassembler and printer. */ static void print_insn_aarch64_word (bfd_vma pc, uint32_t word, struct disassemble_info *info) { static const char *err_msg[6] = { [ERR_OK] = "_", [-ERR_UND] = "undefined", [-ERR_UNP] = "unpredictable", [-ERR_NYI] = "NYI" }; int ret; aarch64_inst inst; info->insn_info_valid = 1; info->branch_delay_insns = 0; info->data_size = 0; info->target = 0; info->target2 = 0; if (info->flags & INSN_HAS_RELOC) /* If the instruction has a reloc associated with it, then the offset field in the instruction will actually be the addend for the reloc. (If we are using REL type relocs). In such cases, we can ignore the pc when computing addresses, since the addend is not currently pc-relative. */ pc = 0; ret = disas_aarch64_insn (pc, word, &inst); if (((word >> 21) & 0x3ff) == 1) { /* RESERVED for ALES. */ assert (ret != ERR_OK); ret = ERR_NYI; } switch (ret) { case ERR_UND: case ERR_UNP: case ERR_NYI: /* Handle undefined instructions. */ info->insn_type = dis_noninsn; (*info->fprintf_func) (info->stream,".inst\t0x%08x ; %s", word, err_msg[-ret]); break; case ERR_OK: user_friendly_fixup (&inst); print_aarch64_insn (pc, &inst, info); break; default: abort (); } } /* Disallow mapping symbols ($x, $d etc) from being displayed in symbol relative addresses. */ bfd_boolean aarch64_symbol_is_valid (asymbol * sym, struct disassemble_info * info ATTRIBUTE_UNUSED) { const char * name; if (sym == NULL) return FALSE; name = bfd_asymbol_name (sym); return name && (name[0] != '$' || (name[1] != 'x' && name[1] != 'd') || (name[2] != '\0' && name[2] != '.')); } /* Print data bytes on INFO->STREAM. */ static void print_insn_data (bfd_vma pc ATTRIBUTE_UNUSED, uint32_t word, struct disassemble_info *info) { switch (info->bytes_per_chunk) { case 1: info->fprintf_func (info->stream, ".byte\t0x%02x", word); break; case 2: info->fprintf_func (info->stream, ".short\t0x%04x", word); break; case 4: info->fprintf_func (info->stream, ".word\t0x%08x", word); break; default: abort (); } } /* Try to infer the code or data type from a symbol. Returns nonzero if *MAP_TYPE was set. */ static int get_sym_code_type (struct disassemble_info *info, int n, enum map_type *map_type) { elf_symbol_type *es; unsigned int type; const char *name; es = *(elf_symbol_type **)(info->symtab + n); type = ELF_ST_TYPE (es->internal_elf_sym.st_info); /* If the symbol has function type then use that. */ if (type == STT_FUNC) { *map_type = MAP_INSN; return TRUE; } /* Check for mapping symbols. */ name = bfd_asymbol_name(info->symtab[n]); if (name[0] == '$' && (name[1] == 'x' || name[1] == 'd') && (name[2] == '\0' || name[2] == '.')) { *map_type = (name[1] == 'x' ? MAP_INSN : MAP_DATA); return TRUE; } return FALSE; } /* Entry-point of the AArch64 disassembler. */ int print_insn_aarch64 (bfd_vma pc, struct disassemble_info *info) { bfd_byte buffer[INSNLEN]; int status; void (*printer) (bfd_vma, uint32_t, struct disassemble_info *); bfd_boolean found = FALSE; unsigned int size = 4; unsigned long data; if (info->disassembler_options) { set_default_aarch64_dis_options (info); parse_aarch64_dis_options (info->disassembler_options); /* To avoid repeated parsing of these options, we remove them here. */ info->disassembler_options = NULL; } /* Aarch64 instructions are always little-endian */ info->endian_code = BFD_ENDIAN_LITTLE; /* First check the full symtab for a mapping symbol, even if there are no usable non-mapping symbols for this address. */ if (info->symtab_size != 0 && bfd_asymbol_flavour (*info->symtab) == bfd_target_elf_flavour) { enum map_type type = MAP_INSN; int last_sym = -1; bfd_vma addr; int n; if (pc <= last_mapping_addr) last_mapping_sym = -1; /* Start scanning at the start of the function, or wherever we finished last time. */ n = info->symtab_pos + 1; if (n < last_mapping_sym) n = last_mapping_sym; /* Scan up to the location being disassembled. */ for (; n < info->symtab_size; n++) { addr = bfd_asymbol_value (info->symtab[n]); if (addr > pc) break; if ((info->section == NULL || info->section == info->symtab[n]->section) && get_sym_code_type (info, n, &type)) { last_sym = n; found = TRUE; } } if (!found) { n = info->symtab_pos; if (n < last_mapping_sym) n = last_mapping_sym; /* No mapping symbol found at this address. Look backwards for a preceeding one. */ for (; n >= 0; n--) { if (get_sym_code_type (info, n, &type)) { last_sym = n; found = TRUE; break; } } } last_mapping_sym = last_sym; last_type = type; /* Look a little bit ahead to see if we should print out less than four bytes of data. If there's a symbol, mapping or otherwise, after two bytes then don't print more. */ if (last_type == MAP_DATA) { size = 4 - (pc & 3); for (n = last_sym + 1; n < info->symtab_size; n++) { addr = bfd_asymbol_value (info->symtab[n]); if (addr > pc) { if (addr - pc < size) size = addr - pc; break; } } /* If the next symbol is after three bytes, we need to print only part of the data, so that we can use either .byte or .short. */ if (size == 3) size = (pc & 1) ? 1 : 2; } } if (last_type == MAP_DATA) { /* size was set above. */ info->bytes_per_chunk = size; info->display_endian = info->endian; printer = print_insn_data; } else { info->bytes_per_chunk = size = INSNLEN; info->display_endian = info->endian_code; printer = print_insn_aarch64_word; } status = (*info->read_memory_func) (pc, buffer, size, info); if (status != 0) { (*info->memory_error_func) (status, pc, info); return -1; } data = bfd_get_bits (buffer, size * 8, info->display_endian == BFD_ENDIAN_BIG); (*printer) (pc, data, info); return size; } void print_aarch64_disassembler_options (FILE *stream) { fprintf (stream, _("\n\ The following AARCH64 specific disassembler options are supported for use\n\ with the -M switch (multiple options should be separated by commas):\n")); fprintf (stream, _("\n\ no-aliases Don't print instruction aliases.\n")); fprintf (stream, _("\n\ aliases Do print instruction aliases.\n")); #ifdef DEBUG_AARCH64 fprintf (stream, _("\n\ debug_dump Temp switch for debug trace.\n")); #endif /* DEBUG_AARCH64 */ fprintf (stream, _("\n")); }