mirror of
https://github.com/TomHarte/CLK.git
synced 2024-11-29 12:50:28 +00:00
ec2d878e3f
I have yet to get any insight whatsoever on the reason for GCC's failure here and won't have access to a suitable test machine for a while so all I have for testing is the arduous CI cycle.
247 lines
8.3 KiB
C++
247 lines
8.3 KiB
C++
//
|
|
// Decoder.hpp
|
|
// Clock Signal
|
|
//
|
|
// Created by Thomas Harte on 01/01/21.
|
|
// Copyright © 2021 Thomas Harte. All rights reserved.
|
|
//
|
|
|
|
#ifndef InstructionSets_x86_Decoder_hpp
|
|
#define InstructionSets_x86_Decoder_hpp
|
|
|
|
#include "Instruction.hpp"
|
|
#include "Model.hpp"
|
|
|
|
#include <cstddef>
|
|
#include <utility>
|
|
|
|
namespace InstructionSet::x86 {
|
|
|
|
/*!
|
|
Implements Intel x86 instruction decoding.
|
|
|
|
This is an experimental implementation; it has not yet undergone significant testing.
|
|
*/
|
|
template <Model model> class Decoder {
|
|
public:
|
|
using InstructionT = Instruction<is_32bit(model)>;
|
|
|
|
/*!
|
|
@returns an @c Instruction plus a size; a positive size indicates successful decoding of
|
|
an instruction that was that many bytes long in total; a negative size specifies the [negatived]
|
|
minimum number of further bytes the caller should ideally collect before calling again. The
|
|
caller is free to call with fewer, but may not get a decoded instruction in response, and the
|
|
decoder may still not be able to complete decoding even if given that number of bytes.
|
|
|
|
Successful decoding is defined to mean that all decoding steps are complete. The output
|
|
may still be an illegal instruction (indicated by Operation::Invalid), if the byte sequence
|
|
supplied cannot form a valid instruction.
|
|
|
|
@discussion although instructions also contain an indicator of their length, on chips prior
|
|
to the 80286 there is no limit to potential instruction length.
|
|
|
|
The 80286 and 80386 have instruction length limits of 10 and 15 bytes respectively, so
|
|
cannot overflow the field.
|
|
*/
|
|
std::pair<int, InstructionT> decode(const uint8_t *source, std::size_t length);
|
|
|
|
/*!
|
|
Enables or disables 32-bit protected mode. Meaningful only if the @c Model supports it.
|
|
*/
|
|
void set_32bit_protected_mode(bool);
|
|
|
|
private:
|
|
enum class Phase {
|
|
/// Captures all prefixes and continues until an instruction byte is encountered.
|
|
Instruction,
|
|
/// Having encountered a 0x0f first instruction byte, waits for the next byte fully to determine the instruction.
|
|
InstructionPageF,
|
|
/// Receives a ModRegRM byte and either populates the source_ and dest_ fields appropriately
|
|
/// or completes decoding of the instruction, as per the instruction format.
|
|
ModRegRM,
|
|
/// Awaits n 80386+-style scale-index-base byte ('SIB'), indicating the form of indirect addressing.
|
|
ScaleIndexBase,
|
|
/// Waits for sufficiently many bytes to pass for the required displacement and operand to be captured.
|
|
/// Cf. displacement_size_ and operand_size_.
|
|
DisplacementOrOperand,
|
|
/// Forms and returns an Instruction, and resets parsing state.
|
|
ReadyToPost
|
|
} phase_ = Phase::Instruction;
|
|
|
|
/// During the ModRegRM phase, format dictates interpretation of the ModRegRM byte.
|
|
///
|
|
/// During the ReadyToPost phase, format determines how transiently-recorded fields
|
|
/// are packaged into an Instruction.
|
|
enum class ModRegRMFormat: uint8_t {
|
|
// Parse the ModRegRM for mode, register and register/memory fields
|
|
// and populate the source_ and destination_ fields appropriately.
|
|
MemReg_Reg,
|
|
Reg_MemReg,
|
|
|
|
// Parse for mode and register/memory fields, populating both
|
|
// source_ and destination_ fields with the single register/memory result.
|
|
MemRegSingleOperand,
|
|
|
|
// Parse for mode and register/memory fields, populating both
|
|
// the destination_ field with the result and setting source_ to Immediate.
|
|
MemRegMOV,
|
|
|
|
// Parse for mode and register/memory fields, populating the
|
|
// source_ field with the result. Fills destination_ with a segment
|
|
// register based on the reg field.
|
|
Seg_MemReg,
|
|
MemReg_Seg,
|
|
|
|
//
|
|
// 'Group 1'
|
|
//
|
|
|
|
// Parse for mode and register/memory fields, populating the
|
|
// destination_ field with the result. Use the 'register' field
|
|
// to pick an operation from the ADD/OR/ADC/SBB/AND/SUB/XOR/CMP group and
|
|
// waits for an operand equal to the operation size.
|
|
MemRegADD_to_CMP,
|
|
|
|
// Acts exactly as MemRegADD_to_CMP but the operand is fixed in size
|
|
// at a single byte, which is sign extended to the operation size.
|
|
MemRegADD_to_CMP_SignExtend,
|
|
|
|
//
|
|
// 'Group 2'
|
|
//
|
|
|
|
// Parse for mode and register/memory fields, populating the
|
|
// destination_ field with the result. Use the 'register' field
|
|
// to pick an operation from the ROL/ROR/RCL/RCR/SAL/SHR/SAR group.
|
|
MemRegROL_to_SAR,
|
|
|
|
//
|
|
// 'Group 3'
|
|
//
|
|
|
|
// Parse for mode and register/memory fields, populating both
|
|
// source_ and destination_ fields with the result. Use the 'register'
|
|
// field to pick an operation from the TEST/NOT/NEG/MUL/IMUL/DIV/IDIV group.
|
|
MemRegTEST_to_IDIV,
|
|
|
|
//
|
|
// 'Group 4'
|
|
//
|
|
|
|
// Parse for mode and register/memory fields, populating the
|
|
// source_ and destination_ fields with the result. Uses the
|
|
// 'register' field to pick INC or DEC.
|
|
MemRegINC_DEC,
|
|
|
|
//
|
|
// 'Group 5'
|
|
//
|
|
|
|
// Parse for mode and register/memory fields, populating the
|
|
// source_ and destination_ fields with the result. Uses the
|
|
// 'register' field to pick from INC/DEC/CALL/JMP/PUSH, altering
|
|
// the source to ::Immediate and setting an operand size if necessary.
|
|
MemRegINC_to_PUSH,
|
|
|
|
//
|
|
// 'Group 6'
|
|
//
|
|
|
|
// Parse for mode and register/memory field, populating both source_
|
|
// and destination_ fields with the result. Uses the 'register' field
|
|
// to pick from SLDT/STR/LLDT/LTR/VERR/VERW.
|
|
MemRegSLDT_to_VERW,
|
|
|
|
//
|
|
// 'Group 7'
|
|
//
|
|
|
|
// Parse for mode and register/memory field, populating both source_
|
|
// and destination_ fields with the result. Uses the 'register' field
|
|
// to pick from SGDT/LGDT/SMSW/LMSW.
|
|
MemRegSGDT_to_LMSW,
|
|
|
|
//
|
|
// 'Group 8'
|
|
//
|
|
|
|
// Parse for mode and register/memory field, populating destination,
|
|
// and prepare to read a single byte as source.
|
|
MemRegBT_to_BTC,
|
|
} modregrm_format_ = ModRegRMFormat::MemReg_Reg;
|
|
|
|
// Ephemeral decoding state.
|
|
Operation operation_ = Operation::Invalid;
|
|
uint32_t consumed_ = 0, operand_bytes_ = 0;
|
|
|
|
// Source and destination locations.
|
|
Source source_ = Source::None;
|
|
Source destination_ = Source::None;
|
|
|
|
// Immediate fields.
|
|
int32_t displacement_ = 0;
|
|
uint32_t operand_ = 0;
|
|
uint64_t inward_data_ = 0;
|
|
int next_inward_data_shift_ = 0;
|
|
|
|
// Indirection style.
|
|
ScaleIndexBase sib_;
|
|
|
|
// Facts about the instruction.
|
|
DataSize displacement_size_ = DataSize::None; // i.e. size of in-stream displacement, if any.
|
|
DataSize operand_size_ = DataSize::None; // i.e. size of in-stream operand, if any.
|
|
DataSize operation_size_ = DataSize::None; // i.e. size of data manipulated by the operation.
|
|
|
|
bool sign_extend_displacement_ = true; // If set then sign extend any displacement up to the address
|
|
// size; otherwise it'll be zero-padded.
|
|
bool sign_extend_operand_ = false; // If set then sign extend the operand up to the operation size;
|
|
// otherwise it'll be zero-padded.
|
|
|
|
// Prefix capture fields.
|
|
Repetition repetition_ = Repetition::None;
|
|
bool lock_ = false;
|
|
Source segment_override_ = Source::None;
|
|
|
|
// 32-bit/16-bit selection.
|
|
AddressSize default_address_size_ = AddressSize::b16;
|
|
DataSize default_data_size_ = DataSize::Word;
|
|
AddressSize address_size_ = AddressSize::b16;
|
|
DataSize data_size_ = DataSize::Word;
|
|
|
|
/// Resets size capture and all fields with default values.
|
|
void reset_parsing() {
|
|
consumed_ = operand_bytes_ = 0;
|
|
displacement_size_ = operand_size_ = operation_size_ = DataSize::None;
|
|
displacement_ = operand_ = 0;
|
|
lock_ = false;
|
|
address_size_ = default_address_size_;
|
|
data_size_ = default_data_size_;
|
|
segment_override_ = Source::None;
|
|
repetition_ = Repetition::None;
|
|
phase_ = Phase::Instruction;
|
|
source_ = destination_ = Source::None;
|
|
sib_ = ScaleIndexBase();
|
|
next_inward_data_shift_ = 0;
|
|
inward_data_ = 0;
|
|
sign_extend_operand_ = false;
|
|
sign_extend_displacement_ = true;
|
|
}
|
|
};
|
|
|
|
// This is a temporary measure; for reasons as-yet unknown, GCC isn't picking up the
|
|
// explicit instantiations of the template above at link time, even though is is
|
|
// unambiguously building and linking in Decoder.cpp.
|
|
//
|
|
// So here's a thin non-templated shim to unblock initial PC Compatible development.
|
|
class Decoder8086 {
|
|
public:
|
|
std::pair<int, Instruction<false>> decode(const uint8_t *source, std::size_t length);
|
|
|
|
private:
|
|
Decoder<Model::i8086> decoder;
|
|
};
|
|
|
|
}
|
|
|
|
#endif /* InstructionSets_x86_Decoder_hpp */
|