Type Function StackMachine [src]

A stack machine that can decode and run DWARF expressions. Expressions can be decoded for non-native address size and endianness, but can only be executed if the current target matches the configuration.

Prototype

pub fn StackMachine(comptime options: Options) type

Parameters

options: Options

Source

pub fn StackMachine(comptime options: Options) type { const addr_type = switch (options.addr_size) { 2 => u16, 4 => u32, 8 => u64, else => @compileError("Unsupported address size of " ++ options.addr_size), }; const addr_type_signed = switch (options.addr_size) { 2 => i16, 4 => i32, 8 => i64, else => @compileError("Unsupported address size of " ++ options.addr_size), }; return struct { const Self = @This(); const Operand = union(enum) { generic: addr_type, register: u8, type_size: u8, branch_offset: i16, base_register: struct { base_register: u8, offset: i64, }, composite_location: struct { size: u64, offset: i64, }, block: []const u8, register_type: struct { register: u8, type_offset: addr_type, }, const_type: struct { type_offset: addr_type, value_bytes: []const u8, }, deref_type: struct { size: u8, type_offset: addr_type, }, }; const Value = union(enum) { generic: addr_type, // Typed value with a maximum size of a register regval_type: struct { // Offset of DW_TAG_base_type DIE type_offset: addr_type, type_size: u8, value: addr_type, }, // Typed value specified directly in the instruction stream const_type: struct { // Offset of DW_TAG_base_type DIE type_offset: addr_type, // Backed by the instruction stream value_bytes: []const u8, }, pub fn asIntegral(self: Value) !addr_type { return switch (self) { .generic => |v| v, // TODO: For these two prongs, look up the type and assert it's integral? .regval_type => |regval_type| regval_type.value, .const_type => |const_type| { const value: u64 = switch (const_type.value_bytes.len) { 1 => mem.readInt(u8, const_type.value_bytes[0..1], native_endian), 2 => mem.readInt(u16, const_type.value_bytes[0..2], native_endian), 4 => mem.readInt(u32, const_type.value_bytes[0..4], native_endian), 8 => mem.readInt(u64, const_type.value_bytes[0..8], native_endian), else => return error.InvalidIntegralTypeSize, }; return std.math.cast(addr_type, value) orelse error.TruncatedIntegralType; }, }; } }; stack: std.ArrayListUnmanaged(Value) = .empty, pub fn reset(self: *Self) void { self.stack.clearRetainingCapacity(); } pub fn deinit(self: *Self, allocator: std.mem.Allocator) void { self.stack.deinit(allocator); } fn generic(value: anytype) Operand { const int_info = @typeInfo(@TypeOf(value)).int; if (@sizeOf(@TypeOf(value)) > options.addr_size) { return .{ .generic = switch (int_info.signedness) { .signed => @bitCast(@as(addr_type_signed, @truncate(value))), .unsigned => @truncate(value), } }; } else { return .{ .generic = switch (int_info.signedness) { .signed => @bitCast(@as(addr_type_signed, @intCast(value))), .unsigned => @intCast(value), } }; } } pub fn readOperand(stream: *std.io.FixedBufferStream([]const u8), opcode: u8, context: Context) !?Operand { const reader = stream.reader(); return switch (opcode) { OP.addr => generic(try reader.readInt(addr_type, options.endian)), OP.call_ref => switch (context.format) { .@"32" => generic(try reader.readInt(u32, options.endian)), .@"64" => generic(try reader.readInt(u64, options.endian)), }, OP.const1u, OP.pick, => generic(try reader.readByte()), OP.deref_size, OP.xderef_size, => .{ .type_size = try reader.readByte() }, OP.const1s => generic(try reader.readByteSigned()), OP.const2u, OP.call2, => generic(try reader.readInt(u16, options.endian)), OP.call4 => generic(try reader.readInt(u32, options.endian)), OP.const2s => generic(try reader.readInt(i16, options.endian)), OP.bra, OP.skip, => .{ .branch_offset = try reader.readInt(i16, options.endian) }, OP.const4u => generic(try reader.readInt(u32, options.endian)), OP.const4s => generic(try reader.readInt(i32, options.endian)), OP.const8u => generic(try reader.readInt(u64, options.endian)), OP.const8s => generic(try reader.readInt(i64, options.endian)), OP.constu, OP.plus_uconst, OP.addrx, OP.constx, OP.convert, OP.reinterpret, => generic(try leb.readUleb128(u64, reader)), OP.consts, OP.fbreg, => generic(try leb.readIleb128(i64, reader)), OP.lit0...OP.lit31 => |n| generic(n - OP.lit0), OP.reg0...OP.reg31 => |n| .{ .register = n - OP.reg0 }, OP.breg0...OP.breg31 => |n| .{ .base_register = .{ .base_register = n - OP.breg0, .offset = try leb.readIleb128(i64, reader), } }, OP.regx => .{ .register = try leb.readUleb128(u8, reader) }, OP.bregx => blk: { const base_register = try leb.readUleb128(u8, reader); const offset = try leb.readIleb128(i64, reader); break :blk .{ .base_register = .{ .base_register = base_register, .offset = offset, } }; }, OP.regval_type => blk: { const register = try leb.readUleb128(u8, reader); const type_offset = try leb.readUleb128(addr_type, reader); break :blk .{ .register_type = .{ .register = register, .type_offset = type_offset, } }; }, OP.piece => .{ .composite_location = .{ .size = try leb.readUleb128(u8, reader), .offset = 0, }, }, OP.bit_piece => blk: { const size = try leb.readUleb128(u8, reader); const offset = try leb.readIleb128(i64, reader); break :blk .{ .composite_location = .{ .size = size, .offset = offset, } }; }, OP.implicit_value, OP.entry_value => blk: { const size = try leb.readUleb128(u8, reader); if (stream.pos + size > stream.buffer.len) return error.InvalidExpression; const block = stream.buffer[stream.pos..][0..size]; stream.pos += size; break :blk .{ .block = block, }; }, OP.const_type => blk: { const type_offset = try leb.readUleb128(addr_type, reader); const size = try reader.readByte(); if (stream.pos + size > stream.buffer.len) return error.InvalidExpression; const value_bytes = stream.buffer[stream.pos..][0..size]; stream.pos += size; break :blk .{ .const_type = .{ .type_offset = type_offset, .value_bytes = value_bytes, } }; }, OP.deref_type, OP.xderef_type, => .{ .deref_type = .{ .size = try reader.readByte(), .type_offset = try leb.readUleb128(addr_type, reader), }, }, OP.lo_user...OP.hi_user => return error.UnimplementedUserOpcode, else => null, }; } pub fn run( self: *Self, expression: []const u8, allocator: std.mem.Allocator, context: Context, initial_value: ?usize, ) Error!?Value { if (initial_value) |i| try self.stack.append(allocator, .{ .generic = i }); var stream = std.io.fixedBufferStream(expression); while (try self.step(&stream, allocator, context)) {} if (self.stack.items.len == 0) return null; return self.stack.items[self.stack.items.len - 1]; } /// Reads an opcode and its operands from `stream`, then executes it pub fn step( self: *Self, stream: *std.io.FixedBufferStream([]const u8), allocator: std.mem.Allocator, context: Context, ) Error!bool { if (@sizeOf(usize) != @sizeOf(addr_type) or options.endian != native_endian) @compileError("Execution of non-native address sizes / endianness is not supported"); const opcode = try stream.reader().readByte(); if (options.call_frame_context and !isOpcodeValidInCFA(opcode)) return error.InvalidCFAOpcode; const operand = try readOperand(stream, opcode, context); switch (opcode) { // 2.5.1.1: Literal Encodings OP.lit0...OP.lit31, OP.addr, OP.const1u, OP.const2u, OP.const4u, OP.const8u, OP.const1s, OP.const2s, OP.const4s, OP.const8s, OP.constu, OP.consts, => try self.stack.append(allocator, .{ .generic = operand.?.generic }), OP.const_type => { const const_type = operand.?.const_type; try self.stack.append(allocator, .{ .const_type = .{ .type_offset = const_type.type_offset, .value_bytes = const_type.value_bytes, } }); }, OP.addrx, OP.constx, => { if (context.compile_unit == null) return error.IncompleteExpressionContext; if (context.debug_addr == null) return error.IncompleteExpressionContext; const debug_addr_index = operand.?.generic; const offset = context.compile_unit.?.addr_base + debug_addr_index; if (offset >= context.debug_addr.?.len) return error.InvalidExpression; const value = mem.readInt(usize, context.debug_addr.?[offset..][0..@sizeOf(usize)], native_endian); try self.stack.append(allocator, .{ .generic = value }); }, // 2.5.1.2: Register Values OP.fbreg => { if (context.compile_unit == null) return error.IncompleteExpressionContext; if (context.compile_unit.?.frame_base == null) return error.IncompleteExpressionContext; const offset: i64 = @intCast(operand.?.generic); _ = offset; switch (context.compile_unit.?.frame_base.?.*) { .exprloc => { // TODO: Run this expression in a nested stack machine return error.UnimplementedOpcode; }, .loclistx => { // TODO: Read value from .debug_loclists return error.UnimplementedOpcode; }, .sec_offset => { // TODO: Read value from .debug_loclists return error.UnimplementedOpcode; }, else => return error.InvalidFrameBase, } }, OP.breg0...OP.breg31, OP.bregx, => { if (context.thread_context == null) return error.IncompleteExpressionContext; const base_register = operand.?.base_register; var value: i64 = @intCast(mem.readInt(usize, (try abi.regBytes( context.thread_context.?, base_register.base_register, context.reg_context, ))[0..@sizeOf(usize)], native_endian)); value += base_register.offset; try self.stack.append(allocator, .{ .generic = @intCast(value) }); }, OP.regval_type => { const register_type = operand.?.register_type; const value = mem.readInt(usize, (try abi.regBytes( context.thread_context.?, register_type.register, context.reg_context, ))[0..@sizeOf(usize)], native_endian); try self.stack.append(allocator, .{ .regval_type = .{ .type_offset = register_type.type_offset, .type_size = @sizeOf(addr_type), .value = value, }, }); }, // 2.5.1.3: Stack Operations OP.dup => { if (self.stack.items.len == 0) return error.InvalidExpression; try self.stack.append(allocator, self.stack.items[self.stack.items.len - 1]); }, OP.drop => { _ = self.stack.pop(); }, OP.pick, OP.over => { const stack_index = if (opcode == OP.over) 1 else operand.?.generic; if (stack_index >= self.stack.items.len) return error.InvalidExpression; try self.stack.append(allocator, self.stack.items[self.stack.items.len - 1 - stack_index]); }, OP.swap => { if (self.stack.items.len < 2) return error.InvalidExpression; mem.swap(Value, &self.stack.items[self.stack.items.len - 1], &self.stack.items[self.stack.items.len - 2]); }, OP.rot => { if (self.stack.items.len < 3) return error.InvalidExpression; const first = self.stack.items[self.stack.items.len - 1]; self.stack.items[self.stack.items.len - 1] = self.stack.items[self.stack.items.len - 2]; self.stack.items[self.stack.items.len - 2] = self.stack.items[self.stack.items.len - 3]; self.stack.items[self.stack.items.len - 3] = first; }, OP.deref, OP.xderef, OP.deref_size, OP.xderef_size, OP.deref_type, OP.xderef_type, => { if (self.stack.items.len == 0) return error.InvalidExpression; const addr = try self.stack.items[self.stack.items.len - 1].asIntegral(); const addr_space_identifier: ?usize = switch (opcode) { OP.xderef, OP.xderef_size, OP.xderef_type, => blk: { _ = self.stack.pop(); if (self.stack.items.len == 0) return error.InvalidExpression; break :blk try self.stack.items[self.stack.items.len - 1].asIntegral(); }, else => null, }; // Usage of addr_space_identifier in the address calculation is implementation defined. // This code will need to be updated to handle any architectures that utilize this. _ = addr_space_identifier; const size = switch (opcode) { OP.deref, OP.xderef, => @sizeOf(addr_type), OP.deref_size, OP.xderef_size, => operand.?.type_size, OP.deref_type, OP.xderef_type, => operand.?.deref_type.size, else => unreachable, }; if (context.memory_accessor) |memory_accessor| { if (!switch (size) { 1 => memory_accessor.load(u8, addr) != null, 2 => memory_accessor.load(u16, addr) != null, 4 => memory_accessor.load(u32, addr) != null, 8 => memory_accessor.load(u64, addr) != null, else => return error.InvalidExpression, }) return error.InvalidExpression; } const value: addr_type = std.math.cast(addr_type, @as(u64, switch (size) { 1 => @as(*const u8, @ptrFromInt(addr)).*, 2 => @as(*const u16, @ptrFromInt(addr)).*, 4 => @as(*const u32, @ptrFromInt(addr)).*, 8 => @as(*const u64, @ptrFromInt(addr)).*, else => return error.InvalidExpression, })) orelse return error.InvalidExpression; switch (opcode) { OP.deref_type, OP.xderef_type, => { self.stack.items[self.stack.items.len - 1] = .{ .regval_type = .{ .type_offset = operand.?.deref_type.type_offset, .type_size = operand.?.deref_type.size, .value = value, }, }; }, else => { self.stack.items[self.stack.items.len - 1] = .{ .generic = value }; }, } }, OP.push_object_address => { // In sub-expressions, `push_object_address` is not meaningful (as per the // spec), so treat it like a nop if (!context.entry_value_context) { if (context.object_address == null) return error.IncompleteExpressionContext; try self.stack.append(allocator, .{ .generic = @intFromPtr(context.object_address.?) }); } }, OP.form_tls_address => { return error.UnimplementedOpcode; }, OP.call_frame_cfa => { if (context.cfa) |cfa| { try self.stack.append(allocator, .{ .generic = cfa }); } else return error.IncompleteExpressionContext; }, // 2.5.1.4: Arithmetic and Logical Operations OP.abs => { if (self.stack.items.len == 0) return error.InvalidExpression; const value: isize = @bitCast(try self.stack.items[self.stack.items.len - 1].asIntegral()); self.stack.items[self.stack.items.len - 1] = .{ .generic = @abs(value), }; }, OP.@"and" => { if (self.stack.items.len < 2) return error.InvalidExpression; const a = try self.stack.pop().?.asIntegral(); self.stack.items[self.stack.items.len - 1] = .{ .generic = a & try self.stack.items[self.stack.items.len - 1].asIntegral(), }; }, OP.div => { if (self.stack.items.len < 2) return error.InvalidExpression; const a: isize = @bitCast(try self.stack.pop().?.asIntegral()); const b: isize = @bitCast(try self.stack.items[self.stack.items.len - 1].asIntegral()); self.stack.items[self.stack.items.len - 1] = .{ .generic = @bitCast(try std.math.divTrunc(isize, b, a)), }; }, OP.minus => { if (self.stack.items.len < 2) return error.InvalidExpression; const b = try self.stack.pop().?.asIntegral(); self.stack.items[self.stack.items.len - 1] = .{ .generic = try std.math.sub(addr_type, try self.stack.items[self.stack.items.len - 1].asIntegral(), b), }; }, OP.mod => { if (self.stack.items.len < 2) return error.InvalidExpression; const a: isize = @bitCast(try self.stack.pop().?.asIntegral()); const b: isize = @bitCast(try self.stack.items[self.stack.items.len - 1].asIntegral()); self.stack.items[self.stack.items.len - 1] = .{ .generic = @bitCast(@mod(b, a)), }; }, OP.mul => { if (self.stack.items.len < 2) return error.InvalidExpression; const a: isize = @bitCast(try self.stack.pop().?.asIntegral()); const b: isize = @bitCast(try self.stack.items[self.stack.items.len - 1].asIntegral()); self.stack.items[self.stack.items.len - 1] = .{ .generic = @bitCast(@mulWithOverflow(a, b)[0]), }; }, OP.neg => { if (self.stack.items.len == 0) return error.InvalidExpression; self.stack.items[self.stack.items.len - 1] = .{ .generic = @bitCast( try std.math.negate( @as(isize, @bitCast(try self.stack.items[self.stack.items.len - 1].asIntegral())), ), ), }; }, OP.not => { if (self.stack.items.len == 0) return error.InvalidExpression; self.stack.items[self.stack.items.len - 1] = .{ .generic = ~try self.stack.items[self.stack.items.len - 1].asIntegral(), }; }, OP.@"or" => { if (self.stack.items.len < 2) return error.InvalidExpression; const a = try self.stack.pop().?.asIntegral(); self.stack.items[self.stack.items.len - 1] = .{ .generic = a | try self.stack.items[self.stack.items.len - 1].asIntegral(), }; }, OP.plus => { if (self.stack.items.len < 2) return error.InvalidExpression; const b = try self.stack.pop().?.asIntegral(); self.stack.items[self.stack.items.len - 1] = .{ .generic = try std.math.add(addr_type, try self.stack.items[self.stack.items.len - 1].asIntegral(), b), }; }, OP.plus_uconst => { if (self.stack.items.len == 0) return error.InvalidExpression; const constant = operand.?.generic; self.stack.items[self.stack.items.len - 1] = .{ .generic = try std.math.add(addr_type, try self.stack.items[self.stack.items.len - 1].asIntegral(), constant), }; }, OP.shl => { if (self.stack.items.len < 2) return error.InvalidExpression; const a = try self.stack.pop().?.asIntegral(); const b = try self.stack.items[self.stack.items.len - 1].asIntegral(); self.stack.items[self.stack.items.len - 1] = .{ .generic = std.math.shl(usize, b, a), }; }, OP.shr => { if (self.stack.items.len < 2) return error.InvalidExpression; const a = try self.stack.pop().?.asIntegral(); const b = try self.stack.items[self.stack.items.len - 1].asIntegral(); self.stack.items[self.stack.items.len - 1] = .{ .generic = std.math.shr(usize, b, a), }; }, OP.shra => { if (self.stack.items.len < 2) return error.InvalidExpression; const a = try self.stack.pop().?.asIntegral(); const b: isize = @bitCast(try self.stack.items[self.stack.items.len - 1].asIntegral()); self.stack.items[self.stack.items.len - 1] = .{ .generic = @bitCast(std.math.shr(isize, b, a)), }; }, OP.xor => { if (self.stack.items.len < 2) return error.InvalidExpression; const a = try self.stack.pop().?.asIntegral(); self.stack.items[self.stack.items.len - 1] = .{ .generic = a ^ try self.stack.items[self.stack.items.len - 1].asIntegral(), }; }, // 2.5.1.5: Control Flow Operations OP.le, OP.ge, OP.eq, OP.lt, OP.gt, OP.ne, => { if (self.stack.items.len < 2) return error.InvalidExpression; const a = self.stack.pop().?; const b = self.stack.items[self.stack.items.len - 1]; if (a == .generic and b == .generic) { const a_int: isize = @bitCast(a.asIntegral() catch unreachable); const b_int: isize = @bitCast(b.asIntegral() catch unreachable); const result = @intFromBool(switch (opcode) { OP.le => b_int <= a_int, OP.ge => b_int >= a_int, OP.eq => b_int == a_int, OP.lt => b_int < a_int, OP.gt => b_int > a_int, OP.ne => b_int != a_int, else => unreachable, }); self.stack.items[self.stack.items.len - 1] = .{ .generic = result }; } else { // TODO: Load the types referenced by these values, find their comparison operator, and run it return error.UnimplementedTypedComparison; } }, OP.skip, OP.bra => { const branch_offset = operand.?.branch_offset; const condition = if (opcode == OP.bra) blk: { if (self.stack.items.len == 0) return error.InvalidExpression; break :blk try self.stack.pop().?.asIntegral() != 0; } else true; if (condition) { const new_pos = std.math.cast( usize, try std.math.add(isize, @as(isize, @intCast(stream.pos)), branch_offset), ) orelse return error.InvalidExpression; if (new_pos < 0 or new_pos > stream.buffer.len) return error.InvalidExpression; stream.pos = new_pos; } }, OP.call2, OP.call4, OP.call_ref, => { const debug_info_offset = operand.?.generic; _ = debug_info_offset; // TODO: Load a DIE entry at debug_info_offset in a .debug_info section (the spec says that it // can be in a separate exe / shared object from the one containing this expression). // Transfer control to the DW_AT_location attribute, with the current stack as input. return error.UnimplementedExpressionCall; }, // 2.5.1.6: Type Conversions OP.convert => { if (self.stack.items.len == 0) return error.InvalidExpression; const type_offset = operand.?.generic; // TODO: Load the DW_TAG_base_type entries in context.compile_unit and verify both types are the same size const value = self.stack.items[self.stack.items.len - 1]; if (type_offset == 0) { self.stack.items[self.stack.items.len - 1] = .{ .generic = try value.asIntegral() }; } else { // TODO: Load the DW_TAG_base_type entry in context.compile_unit, find a conversion operator // from the old type to the new type, run it. return error.UnimplementedTypeConversion; } }, OP.reinterpret => { if (self.stack.items.len == 0) return error.InvalidExpression; const type_offset = operand.?.generic; // TODO: Load the DW_TAG_base_type entries in context.compile_unit and verify both types are the same size const value = self.stack.items[self.stack.items.len - 1]; if (type_offset == 0) { self.stack.items[self.stack.items.len - 1] = .{ .generic = try value.asIntegral() }; } else { self.stack.items[self.stack.items.len - 1] = switch (value) { .generic => |v| .{ .regval_type = .{ .type_offset = type_offset, .type_size = @sizeOf(addr_type), .value = v, }, }, .regval_type => |r| .{ .regval_type = .{ .type_offset = type_offset, .type_size = r.type_size, .value = r.value, }, }, .const_type => |c| .{ .const_type = .{ .type_offset = type_offset, .value_bytes = c.value_bytes, }, }, }; } }, // 2.5.1.7: Special Operations OP.nop => {}, OP.entry_value => { const block = operand.?.block; if (block.len == 0) return error.InvalidSubExpression; // TODO: The spec states that this sub-expression needs to observe the state (ie. registers) // as it was upon entering the current subprogram. If this isn't being called at the // end of a frame unwind operation, an additional ThreadContext with this state will be needed. if (isOpcodeRegisterLocation(block[0])) { if (context.thread_context == null) return error.IncompleteExpressionContext; var block_stream = std.io.fixedBufferStream(block); const register = (try readOperand(&block_stream, block[0], context)).?.register; const value = mem.readInt(usize, (try abi.regBytes(context.thread_context.?, register, context.reg_context))[0..@sizeOf(usize)], native_endian); try self.stack.append(allocator, .{ .generic = value }); } else { var stack_machine: Self = .{}; defer stack_machine.deinit(allocator); var sub_context = context; sub_context.entry_value_context = true; const result = try stack_machine.run(block, allocator, sub_context, null); try self.stack.append(allocator, result orelse return error.InvalidSubExpression); } }, // These have already been handled by readOperand OP.lo_user...OP.hi_user => unreachable, else => { //std.debug.print("Unknown DWARF expression opcode: {x}\n", .{opcode}); return error.UnknownExpressionOpcode; }, } return stream.pos < stream.buffer.len; } }; }