def parse_instructions(self) -> None:
for i, i_name in enumerate(self.hexArch["!instanceof"]["HInst"]):
llvm_instruction = self.hexArch[i_name]
if llvm_instruction is None:
log(
"Could not find instruction with name: {} in json file.".
format(i_name),
LogLevel.ERROR,
)
continue
if llvm_instruction["isPseudo"]:
log(
"Pseudo instruction passed. Name: {}".format(i_name),
LogLevel.VERBOSE,
)
continue
log("{} | Parse {}".format(i, i_name), LogLevel.VERBOSE)
self.llvm_instructions[i_name] = llvm_instruction
if llvm_instruction["Type"]["def"] == "TypeSUBINSN":
self.sub_instruction_names.append(i_name)
self.sub_instructions[i_name] = SubInstruction(
llvm_instruction)
# log(i_name, LogLevel.DEBUG)
else:
self.normal_instruction_names.append(i_name)
self.normal_instructions[i_name] = Instruction(
llvm_instruction)
log("Parsed {} normal instructions.".format(
len(self.normal_instructions)))
log("Parsed {} sub-instructions.".format(len(self.sub_instructions)))
def build_hexagon_analysis_h(
path: str = "./rizin/librz/analysis/arch/hexagon/hexagon_analysis.h",
) -> None:
with open(path, "w+") as f:
f.write(get_generation_warning_c_code())
with open("handwritten/hexagon_analysis_h/include.h") as include:
set_pos_after_license(include)
f.writelines(include.readlines())
log("hexagon_analysis.h written to {}".format(path))
def parse_hardware_registers(self) -> None:
cc = 0
cr = 0
for reg_class_name in HexagonArchInfo.REG_CLASS_NAMES:
# LLVM fake register class; VectRegRev = reverse double register: V0:1 instead of V1:0
if reg_class_name == "UsrBits" or reg_class_name == "VectRegRev":
continue
# Register class which holds all new register of an arch version. Irrelevant for us at the moment.
if reg_class_name == "V65Regs" or reg_class_name == "V62Regs":
continue
self.hardware_regs[reg_class_name] = dict()
reg_class: dict = self.hexArch[reg_class_name]
# Use "Alignment" although a "Size" attribute exists. But Double Regs set that to 0.
size: int = (reg_class["Alignment"]
if reg_class["Size"] == 0 else reg_class["Size"])
reg_names = list()
for a in reg_class["MemberList"]["args"]:
arg = a[0]
if "def" in arg:
# VTMP = LLVM fake register
if arg["def"] == "VTMP":
continue
if reg_class_name == "CtrRegs" and arg["def"] == "C8":
# For whatever reason this C8 occurs twice, but as USR reg.
# We better use the USR reg as it lists c8 als alternative name for itself.
continue
reg_names.append(arg["def"])
elif "sequence" in arg["printable"]:
reg_names = reg_names + unfold_llvm_sequence(
arg["printable"])
# Remove registers whichs tart with WR; WR register are reverse double vector regs: V0:1 instead of V1:0
# TODO This is not nice. Isn't there a simpler way?
reg_names = [
name for name in reg_names
if not re.search(r"WR\d{1,2}", name)
]
for name in reg_names:
llvm_reg = self.hexArch[name]
reg = HardwareRegister(
llvm_reg_class=reg_class_name,
name=name,
llvm_object=llvm_reg,
size=size,
)
self.hardware_regs[reg_class_name][name] = reg
cr += 1
# log("Added reg: {}::{} with hw encoding: {}".format(name, reg_class_name,
# reg.hw_encoding), LogLevel.DEBUG)
cc += 1
log("Parsed {} hardware registers of {} different register classes.".
format(cr, cc))
def copy_tests() -> None:
with open("handwritten/analysis-tests/hexagon") as f:
with open("./rizin/test/db/analysis/hexagon", "w+") as g:
set_pos_after_license(g)
g.writelines(f.readlines())
with open("handwritten/asm-tests/hexagon") as f:
with open("./rizin/test/db/asm/hexagon", "w+") as g:
set_pos_after_license(g)
g.writelines(f.readlines())
log("Copied test files to ./rizin/test/db/")
def get_alias_profile(self) -> str:
"""Returns the alias profile of register. A0 = R0, SP = R29 PC = pc etc."""
indent = PluginInfo.LINE_INDENT
p = "\n" + '"=PC{}pc\\n"'.format(indent) + "\n"
p += '"=SP{}r29\\n"'.format(indent) + "\n"
p += '"=BP{}r30\\n"'.format(indent) + "\n"
p += '"=LR{}r31\\n"'.format(indent) + "\n"
p += '"=SR{}usr\\n"'.format(indent) + "\n"
arg_regs = ""
ret_regs = ""
arguments = HexagonArchInfo.CC_REGS["GPR_args"]
returns = HexagonArchInfo.CC_REGS["GPR_ret"]
general_ps = list(self.hardware_regs["IntRegs"].values()) + list(
self.hardware_regs["DoubleRegs"].values())
gpr: HardwareRegister
for gpr in general_ps:
try:
i = arguments.index(gpr.name)
except ValueError:
continue
if i > 9 and gpr.name in HexagonArchInfo.CC_REGS["GPR_args"]:
log(
"Can not add register {} as argument reg to the register profile. "
.format(gpr.name) +
"Rizin only supports 10 argument registers. Check rz_reg.h if this changed.",
LogLevel.WARNING,
)
if gpr.name in HexagonArchInfo.CC_REGS["GPR_args"]:
arg_regs += '"=A{}{}{}\\n"'.format(i, indent,
gpr.asm_name) + "\n"
for gpr in general_ps:
try:
i = returns.index(gpr.name)
except ValueError:
continue
if i > 3 and gpr.name in HexagonArchInfo.CC_REGS["GPR_ret"]:
log(
"Can not add register {} as return reg to the register profile. "
.format(gpr.name) +
"Rizin only supports 4 return registers. Check rz_reg.h if this changed.",
LogLevel.WARNING,
)
if gpr.name in HexagonArchInfo.CC_REGS["GPR_ret"]:
ret_regs += '"=R{}{}{}\\n"'.format(i, indent,
gpr.asm_name) + "\n"
p += arg_regs + ret_regs + "\n"
return p
def add_license_header() -> None:
log("Add license headers")
for subdir, dirs, files in os.walk("rizin/"):
for file in files:
if file == "hexagon": # Tests
continue
p = os.path.join(subdir, file)
with open(p, "r+") as f:
content = f.read()
f.seek(0, 0)
f.write(get_license() + "\n" + content)
def build_hexagon_h(
self,
path: str = "./rizin/librz/asm/arch/hexagon/hexagon.h") -> None:
indent = PluginInfo.LINE_INDENT
general_prefix = PluginInfo.GENERAL_ENUM_PREFIX
with open(path, "w+") as dest:
dest.write(get_generation_warning_c_code())
dest.write("\n")
dest.write(get_include_guard("hexagon.h"))
with open("handwritten/hexagon_h/typedefs.h") as typedefs:
set_pos_after_license(typedefs)
dest.writelines(typedefs.readlines())
reg_class: str
for reg_class in self.hardware_regs:
dest.write("\ntypedef enum {\n")
hw_reg: HardwareRegister
for hw_reg in sorted(self.hardware_regs[reg_class].values(),
key=lambda x: x.hw_encoding):
alias = ",".join(hw_reg.alias)
dest.write("{}{} = {},{}\n".format(
indent,
hw_reg.enum_name,
hw_reg.hw_encoding,
" // " + alias if alias != "" else "",
))
dest.write("}} {}{}; // {}\n".format(
general_prefix,
HardwareRegister.register_class_name_to_upper(reg_class),
reg_class,
))
with open("handwritten/hexagon_h/macros.h") as macros:
set_pos_after_license(macros)
dest.writelines(macros.readlines())
dest.write("\n")
if len(self.reg_resolve_decl) == 0:
raise ImplementationException(
"Register resolve declarations missing"
"(They get generated together with hexagon.c)."
"Please generate hexagon.c before hexagon.h")
for decl in self.reg_resolve_decl:
dest.write(decl + "\n")
with open("handwritten/hexagon_h/declarations.h") as decl:
set_pos_after_license(decl)
dest.writelines(decl.readlines())
dest.write("#endif")
log("hexagon.h written to: {}".format(path))
def build_asm_hexagon_c(
path: str = "rizin/librz/asm/p/asm_hexagon.c") -> None:
with open(path, "w+") as f:
f.write(get_generation_warning_c_code())
with open("handwritten/asm_hexagon_c/include.c") as include:
set_pos_after_license(include)
f.writelines(include.readlines())
with open("handwritten/asm_hexagon_c/initialization.c") as init:
set_pos_after_license(init)
f.writelines(init.readlines())
log("asm_hexagon.c written to {}".format(path))
def build_hexagon_disas_c(
self,
path: str = "./rizin/librz/asm/arch/hexagon/hexagon_disas.c"
) -> None:
# TODO Clean up this method
indent = PluginInfo.LINE_INDENT
var = PluginInfo.HEX_INSTR_VAR_SYNTAX
with open(path, "w+") as dest:
dest.write(get_generation_warning_c_code())
with open("handwritten/hexagon_disas_c/include.c") as include:
set_pos_after_license(include)
dest.writelines(include.readlines())
header = (
"int hexagon_disasm_instruction(ut32 hi_u32, HexInsn *hi, ut32 addr) {\n"
+ '{}char signed_imm[16] = "";\n'.format(indent) +
"{}// DUPLEXES\n".format(indent) +
"{}if ((({} >> 14) & 0x3) == 0) {{\n".format(indent, var) +
"{}switch (((({} >> 29) & 0xF) << 1) | (({} >> 13) & 1)) {{\n".
format(indent * 2, var, var))
dest.write(header)
# Duplexes
for c in range(0xF): # Class 0xf is reserved yet.
dest.write("{}case 0x{:x}:\n".format(indent * 3, c))
for d_instr in self.duplex_instructions.values():
if d_instr.encoding.get_i_class() == c:
dest.write(
indent_code_block(
d_instr.get_instruction_init_in_c(), 4))
dest.write("{}break;\n".format(indent * 4))
# Normal instructions
# Brackets for switch, if
dest.write("{}}}\n{}}}\n{}else {{\n".format(
indent * 2, indent, indent))
dest.write("{}switch (({} >> 28) & 0xF) {{\n".format(
indent * 2, var))
for c in range(0x10):
dest.write("{}case 0x{:x}:\n".format(indent * 3, c))
for instr in self.normal_instructions.values():
if instr.encoding.get_i_class() == c:
dest.write(
indent_code_block(
instr.get_instruction_init_in_c(), 4))
dest.write("{}break;\n".format(indent * 4))
# Closing brackets for switch, else, function
dest.write("{}}}\n{}}}\n{}return 4;\n}}".format(
indent * 2, indent, indent))
log("Hexagon instruction disassembler code written to: {}".format(
path))
def build_hexagon_insn_enum_h(
self,
path: str = "./rizin/librz/asm/arch/hexagon/hexagon_insn.h"
) -> None:
with open(path, "w+") as dest:
dest.write(get_generation_warning_c_code())
dest.write("\n")
dest.write(get_include_guard("hexagon_insn.h"))
dest.write("\n")
dest.write("enum HEX_INS {\n")
for name in self.normal_instruction_names + self.duplex_instructions_names:
dest.write(PluginInfo.LINE_INDENT +
PluginInfo.INSTR_ENUM_PREFIX + name.upper() + ",\n")
dest.write("};\n\n")
dest.write("#endif")
log("Hexagon instruction enum written to: {}".format(path))
def build_hexagon_c(
self,
path: str = "./rizin/librz/asm/arch/hexagon/hexagon.c") -> None:
indent = PluginInfo.LINE_INDENT
with open(path, "w+") as dest:
dest.write(get_generation_warning_c_code())
with open("handwritten/hexagon_c/include.c") as include:
set_pos_after_license(include)
dest.writelines(include.readlines())
dest.write("\n")
reg_class: str
for reg_class in self.hardware_regs:
func_name = HardwareRegister.get_func_name_of_class(reg_class)
function = "char* {}(int opcode_reg)".format(func_name)
self.reg_resolve_decl.append(function + ";")
dest.write("\n{} {{\n".format(function))
parsing_code = HardwareRegister.get_parse_code_reg_bits(
reg_class, "opcode_reg")
parsing_code = indent_code_block(parsing_code, 1)
if parsing_code != "":
dest.write("{}\n".format(parsing_code))
dest.write("{}switch (opcode_reg) {{\n".format(indent))
dest.write(
'{}default:\n{}rz_warn_if_reached();\n{}return "<err>";\n'.
format(indent * 2, indent * 3, indent * 3))
hw_reg: HardwareRegister
for hw_reg in self.hardware_regs[reg_class].values():
dest.write('{}case {}:\n{}return "{}";\n'.format(
indent * 2,
hw_reg.enum_name,
indent * 3,
hw_reg.asm_name.upper(),
))
dest.write("{}}}\n}}\n".format(indent))
with open("handwritten/hexagon_c/functions.c") as func:
set_pos_after_license(func)
dest.writelines(func.readlines())
dest.write("\n")
log("hexagon.c written to: {}".format(path))
def fulfill_constraints(low: SubInstruction, high: SubInstruction) -> bool:
"""Sub-Instructions in a duplex instr have to fulfill some constraints, all of which are checked here.
Src: llvm-project/llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCDuplexInfo.cpp::isOrderedDuplexPair
"""
if not HexagonArchInfo.duplex_constrains_info_shown:
log(
"Checked Duplex constraints:"
+ "\n\t- Max. one extendable sub instruction per duplex."
+ "\n\t- Extendable instruction in slot 1 (high)."
+ "\n\t- Same sub type: smaller instruction in slot 1 (high)."
+ "\n\t- SL2_jumpr31 never in slot 1 (high)."
+ "\n\t- S2_allocframe never in slot 1 (high).",
LogLevel.INFO,
)
HexagonArchInfo.duplex_constrains_info_shown = True
# Max. one extendable sub instruction per duplex.
# Extendable instruction in slot 1 (high)
if low.has_extendable_imm:
# log("low: {}, high: {} rejected because: low is extendable.".format(
# low.name, high.name), LogLevel.DEBUG)
return False
# Same sub namespace (A, S1, S2, L1, L2...): smaller instruction in slot 1 (high)
if low.namespace == high.namespace:
if low.encoding.num_representation < high.encoding.num_representation:
# log("low: {}, high: {} rejected because: type is same but numerically smaller value is high.".format(
# low.name, high.name), LogLevel.DEBUG)
return False
# SL2_jumpr31[...] never in slot 1 (high)
if "SL2_jumpr31" in high.name:
# log("low: {}, high: {} rejected because: SL2_jumpr31 is in slot 1.".format(
# low.name, high.name), LogLevel.DEBUG)
return False
# S2_allocframe never in slot 1 (high).
if "SS2_allocframe" in high.name:
# log("low: {}, high: {} rejected because: S2_allocframe is in slot 1.".format(
# low.name, high.name), LogLevel.DEBUG)
return False
return True
def generate_duplex_instructions(self) -> None:
sub_instr_pairs = itertools.product(self.sub_instructions.values(),
self.sub_instructions.values())
for pair in sub_instr_pairs:
low_instr = pair[0]
high_instr = pair[1]
i_class = DuplexInstruction.get_duplex_i_class_of_instr_pair(
low=low_instr, high=high_instr)
if (i_class != DuplexIClass.INVALID
and DuplexInstruction.fulfill_constraints(
low_instr, high_instr)):
llvm_dup_instr = self.hexArch[i_class.name]
dup_instr = DuplexInstruction(llvm_duplex_instr=llvm_dup_instr,
low=low_instr,
high=high_instr)
self.duplex_instructions[dup_instr.name] = dup_instr
self.duplex_instructions_names.append(dup_instr.name)
# log("Duplex instruction generated: {}".format(dup_instr.name), LogLevel.DEBUG)
log("Generated {} duplex instructions.".format(
len(self.duplex_instructions)))
def __init__(self, hexagon_target_json_path: str, test_mode=False):
self.hexagon_target_json_path = hexagon_target_json_path
with open(self.hexagon_target_json_path) as file:
self.hexArch = json.load(file)
log("LLVM Hexagon target dump successfully loaded.")
# Save types
HexagonArchInfo.IMMEDIATE_TYPES = self.hexArch["!instanceof"][
"Operand"]
HexagonArchInfo.REG_CLASS_NAMES = self.hexArch["!instanceof"][
"RegisterClass"]
HexagonArchInfo.LLVM_FAKE_REGS = self.hexArch["!instanceof"][
"HexagonFakeReg"]
HexagonArchInfo.ALL_REG_NAMES = self.hexArch["!instanceof"][
"DwarfRegNum"]
HexagonArchInfo.CALLEE_SAVED_REGS = [
name[0]["def"]
for name in self.hexArch["HexagonCSR"]["SaveList"]["args"]
]
HexagonArchInfo.CC_REGS = self.get_cc_regs()
# RIZIN SPECIFIC
# Name of the function which parses the encoded register index bits.
self.reg_resolve_decl = list()
self.parse_hardware_registers()
self.parse_instructions()
self.generate_duplex_instructions()
self.check_insn_syntax_length()
if not test_mode:
self.generate_asm_code()
self.generate_analysis_code()
self.generate_decompiler_code()
self.add_license_header()
self.apply_clang_format()
self.print_todos()
def parse_imm_type(self, llvm_imm_type: str) -> None:
"""Parse immediate types like: u4_2Imm. This method sets all kinds of flags, the scale and total width."""
type_letter = re.search(r"^([a-z]+)\d{1,2}", llvm_imm_type)
if not type_letter:
raise ImplementationException(
"Unhandled immediate type: {}".format(llvm_imm_type))
else:
type_letter = type_letter.group(1)
if type_letter == "s":
self.is_signed = True
elif type_letter == "u":
self.is_signed = False
# Address used in "call" and "jmp" instructions (a for call, b for jmp) is relative to PC.
elif type_letter == "a" or type_letter == "b":
self.is_signed = True
self.is_pc_relative = True
# Constant value -1
elif type_letter == "n":
self.is_signed = True
self.is_constant = True
self.encoding_width = None # Is not encoded in the llvm instruction
width = re.search(r"[a-z](\d+)", llvm_imm_type)
if not width:
raise ImplementationException(
"Unhandled immediate type: {}".format(llvm_imm_type))
else:
self.total_width = width.group(1)
self.is_extendable = False
# log("Parsed imm type: {}, width: {}".format(imm_type, self.total_width), LogLevel.DEBUG)
return
else:
raise ImplementationException(
"Unhandled immediate type: {}".format(llvm_imm_type))
# Value before _ represents number of encoded bits.
result = re.search(r"[a-z](\d+)\_", llvm_imm_type)
if result:
self.encoding_width = int(result.group(1))
else:
raise ImplementationException(
"Could not parse encoding width of immediate type: {}".format(
llvm_imm_type))
# Value after the _ represents tells how often the immediate has to be shifted.
result = re.search(r"\_(\d+)Imm", llvm_imm_type)
if result:
self.scale = int(result.group(1))
else:
raise ImplementationException(
"Could not find parse scale of immediate type: {}".format(
llvm_imm_type))
self.total_width = self.encoding_width + self.scale
mx = HexagonArchInfo.MAX_IMM_LEN
nw = int((self.total_width - mx) / 4)
if self.total_width > mx:
log(
"Rizins hexagon_disas.c assumes that immediate values are not larger than {}bit.\n"
.format(mx) + "\tImmediate type: {} is {}bits long.\n".format(
self.llvm_type, self.total_width) +
"\tPlease increase the buffer hexagon_disas.c::signed_imm by at least {} bits."
.format(nw),
LogLevel.WARNING,
)
# The extended immediate should have always the op_type of a 32/64bit wide immediate.
if self.is_extendable and not (self.total_width == 32
or self.total_width == 64):
raise UnexpectedException(
"Extendable immediate is not 32 or 64bits long!\n" +
"imm: {}".format(self.llvm_syntax))
def apply_clang_format() -> None:
for subdir, dirs, files in os.walk("rizin/librz/"):
for file in files:
os.system("./clang-format.py -f " + os.path.join(subdir, file))
log("clang-format applied.")
def parse_instruction(self) -> None:
"""Parses all operands of the instruction which are encoded."""
# TODO A lot of duplicate code with Instruction::parse:instruction()
# Operand names seen during parsing the encoding. Twin operands (Operands which appear in high and low instr.)
# were renamed.
all_ops = deepcopy(
self.high_instr.llvm_in_out_operands + self.low_instr.llvm_in_out_operands
)
self.llvm_filtered_operands = self.remove_invisible_in_out_regs(
self.llvm_syntax, all_ops
)
self.operand_indices = self.get_syntax_operand_indices(
self.llvm_syntax, self.llvm_filtered_operands
)
# Update syntax indices
if self.has_new_non_predicate:
op_name = self.llvm_in_out_operands[self.new_operand_index][1]
self.new_operand_index = self.operand_indices[op_name]
# log("{}\n new: {}".format(self.llvm_syntax, self.new_operand_index), LogLevel.DEBUG)
if self.has_extendable_imm:
op_name = self.llvm_in_out_operands[self.ext_operand_index][1]
self.ext_operand_index = self.operand_indices[op_name]
# log("{}\n ext: {}".format(self.llvm_syntax, self.ext_operand_index), LogLevel.DEBUG)
if len(self.llvm_filtered_operands) > PluginInfo.MAX_OPERANDS:
warning = "{} instruction struct can only hold {} operands. This instruction has {} operands.".format(
PluginInfo.FRAMEWORK_NAME,
PluginInfo.MAX_OPERANDS,
len(self.llvm_filtered_operands),
)
raise ImplementationException(warning)
for in_out_operand in self.llvm_filtered_operands:
op_name = in_out_operand[1]
op_type = in_out_operand[0]["def"]
index = self.operand_indices[op_name]
# Parse register operand
if Operand.get_operand_type(op_type) is OperandType.REGISTER:
# Indices of new values (stored in "opNewValue") are only for non predicates.
is_new_value = (
self.new_operand_index == index and self.has_new_non_predicate
)
operand = Register(op_name, op_type, is_new_value, index)
# Whether the predicate registers holds a new value is denoted in "isPredicatedNew".
if self.predicate_info.new_value and operand.is_predicate:
operand.is_new_value = True
# Parse immediate operands
elif Operand.get_operand_type(op_type) is OperandType.IMMEDIATE:
extendable = self.has_extendable_imm and self.ext_operand_index == index
if self.extendable_alignment > 0:
log(str(self.extendable_alignment), op_type)
operand = Immediate(
op_name, op_type, extendable, self.extendable_alignment, index
)
else:
raise ImplementationException(
"Unknown operand type: {}, op_name: {}".format(op_type, op_name)
)
# Use lower() because we can get RX16in and Rx16in but constraints are always Rx16in.
if op_name.lower() in self.constraints.lower():
operand.is_in_out_operand = True
operand.is_out_operand = True
operand.is_in_operand = True
elif in_out_operand in self.llvm_in_operands:
operand.is_in_operand = True
elif in_out_operand in self.llvm_out_operands:
operand.is_out_operand = True
# Add opcode extraction code
if (
operand.type == OperandType.IMMEDIATE and operand.is_constant
): # Constants have no parsing code.
pass
else:
if (
operand.is_in_out_operand and op_name[-2:] == "in"
): # In/Out Register
mask = self.encoding.operand_masks[op_name[:-2]] # Ends with "in"
else:
mask = self.encoding.operand_masks[op_name]
operand.opcode_mask = mask
operand.add_code_for_opcode_parsing(Operand.make_sparse_mask(mask))
# On the fly check whether the new values have been assigned correctly.
if op_name + ".new" in self.llvm_syntax:
if not operand.is_new_value:
raise ImplementationException(
"Register has new value in syntax but not as object."
+ "It has been parsed incorrectly! Are the indices correctly set?"
+ "Affected instruction: {}".format(self.llvm_syntax)
)
# log("Add operand: {}".format(op_name), LogLevel.DEBUG)
# TODO This uses the llvm name as key. Maybe use normalized name? Rs16 -> Rs?
self.operands[op_name] = operand
def build_hexagon_analysis_c(
self,
path: str = "./rizin/librz/analysis/arch/hexagon/hexagon_analysis.c"
) -> None:
indent = PluginInfo.LINE_INDENT
with open(path, "w+") as f:
f.write(get_generation_warning_c_code())
with open("handwritten/hexagon_analysis_c/include.c") as include:
set_pos_after_license(include)
f.writelines(include.readlines())
f.write("\n")
f.write(
"int hexagon_analysis_instruction(HexInsn *hi, RzAnalysisOp *op) {\n"
)
f.write("{}static ut32 hw_loop0_start = 0;\n".format(indent))
f.write("{}static ut32 hw_loop1_start = 0;\n\n".format(indent))
f.write("{}switch (hi->instruction) {{\n".format(indent))
default = "{}default:\n".format(indent * 2)
default += (
"{}if (is_endloop01_instr(hi) && hi->pkt_info.last_insn) {{\n".
format(indent * 3) +
"{}op->type = RZ_ANALYSIS_OP_TYPE_CJMP;\n".format(indent * 4) +
"{}op->fail = hw_loop0_start;\n".format(indent * 4) +
"{}op->jump = hw_loop1_start;\n".format(indent * 4) +
"{}hw_loop1_start = 0;\n".format(indent * 4) +
"{}hw_loop0_start = 0;\n{}}}\n".format(indent * 4, indent * 3))
default += (
"{}else if (is_endloop0_instr(hi) && hi->pkt_info.last_insn) {{\n"
.format(indent * 3) +
"{}op->type = RZ_ANALYSIS_OP_TYPE_CJMP;\n".format(indent * 4) +
"{}op->jump = hw_loop0_start;\n".format(indent * 4) +
"{}hw_loop0_start = 0;\n{}}}\n".format(indent * 4, indent * 3))
default += (
"{}else if (is_endloop1_instr(hi) && hi->pkt_info.last_insn) {{\n"
.format(indent * 3) +
"{}op->type = RZ_ANALYSIS_OP_TYPE_CJMP;\n".format(indent * 4) +
"{}op->jump = hw_loop1_start;\n".format(indent * 4) +
"{}hw_loop1_start = 0;\n{}}}\n".format(indent * 4, indent * 3))
default += "{}break;\n".format(indent * 3)
f.write(default)
all_instr = [i for i in self.normal_instructions.values()]
for i in self.duplex_instructions.values():
all_instr.append(i)
i: Instruction
for i in all_instr:
if i.has_jump_target:
f.write("{}case {}:\n".format(indent * 2, i.plugin_name))
f.write("{}// {}\n".format(indent * 3, i.syntax))
f.write("{}{}\n".format(indent * 3, i.get_rizin_op_type()))
if i.has_imm_jmp_target():
index = i.get_jmp_operand_syntax_index()
if index < 0:
raise ImplementationException(
"Not PC relative operand given. But the jump needs one."
"{}".format(i.llvm_syntax))
f.write(
"{}op->jump = op->addr + (st32) hi->ops[{}].op.imm;\n"
.format(indent * 3, index))
if i.predicated:
f.write(
"{}op->fail = op->addr + op->size;\n".format(
indent * 3))
if i.is_loop_begin:
f.write("{}if (is_loop0_begin(hi)) {{\n".format(
indent * 3) +
"{}hw_loop0_start = op->jump;\n{}}}\n".
format(indent * 4, indent * 3))
f.write("{}else if (is_loop1_begin(hi)) {{\n".
format(indent * 3) +
"{}hw_loop1_start = op->jump;\n{}}}\n".
format(indent * 4, indent * 3))
f.write("{}break;\n".format(indent * 3, indent * 2))
f.write("{i}}}\n{i}return op->size;\n}}".format(i=indent))
log("hexagon_analysis.c written to: {}".format(path))
def print_todos() -> None:
log(
"CtrRegs64: C21:20, C23:22, C25:24, C27:C26, C29:28 recognition are not yet implemented by LLVM.\n"
"\tSo at the moment we don't have them either.",
LogLevel.TODO,
)
log(
"HVX: HVX register classes and instructions can, at the moment, not be tested properly.\n"
"\tThe public SDK does not support them yet, so we can't compile reasonable test cases for them.",
LogLevel.TODO,
)
log(
"All not generated files have to be moved to handwritten_files",
LogLevel.TODO,
)
log(
"A lot of instruction types can be set in analyse like NOP etc.",
LogLevel.TODO,
)
log("Syntax highlighting is weird.", LogLevel.TODO)
log("Analysis tests are missing.", LogLevel.TODO)
log("No UTF-8 free version yet.", LogLevel.TODO)
log(
"The support for 512bit Hardware registers was removed in V66 (replaced by 1024bit once).\n\t"
"But LLVM Vx registers still have a size of 512bit.\n\t"
"We multiply the size of all vector register by 2.",
LogLevel.TODO,
)
log(
"Argument and return registers for the HVX calling convention are not described in the register profile.",
LogLevel.TODO,
)