aboutsummaryrefslogtreecommitdiff
path: root/lib/std/debug
diff options
context:
space:
mode:
Diffstat (limited to 'lib/std/debug')
-rw-r--r--lib/std/debug/Dwarf.zig2709
-rw-r--r--lib/std/debug/Dwarf/abi.zig410
-rw-r--r--lib/std/debug/Dwarf/call_frame.zig687
-rw-r--r--lib/std/debug/Dwarf/expression.zig1638
4 files changed, 5444 insertions, 0 deletions
diff --git a/lib/std/debug/Dwarf.zig b/lib/std/debug/Dwarf.zig
new file mode 100644
index 0000000000..f17fd737a1
--- /dev/null
+++ b/lib/std/debug/Dwarf.zig
@@ -0,0 +1,2709 @@
+//! Implements parsing, decoding, and caching of DWARF information.
+//!
+//! For unopinionated types and bits, see `std.dwarf`.
+
+const builtin = @import("builtin");
+const std = @import("../std.zig");
+const AT = DW.AT;
+const Allocator = std.mem.Allocator;
+const DW = std.dwarf;
+const EH = DW.EH;
+const FORM = DW.FORM;
+const Format = DW.Format;
+const RLE = DW.RLE;
+const StackIterator = std.debug.StackIterator;
+const UT = DW.UT;
+const assert = std.debug.assert;
+const cast = std.math.cast;
+const maxInt = std.math.maxInt;
+const native_endian = builtin.cpu.arch.endian();
+const readInt = std.mem.readInt;
+
+const Dwarf = @This();
+
+pub const expression = @import("Dwarf/expression.zig");
+pub const abi = @import("Dwarf/abi.zig");
+pub const call_frame = @import("Dwarf/call_frame.zig");
+
+endian: std.builtin.Endian,
+sections: SectionArray = null_section_array,
+is_macho: bool,
+
+// Filled later by the initializer
+abbrev_table_list: std.ArrayListUnmanaged(Abbrev.Table) = .{},
+compile_unit_list: std.ArrayListUnmanaged(CompileUnit) = .{},
+func_list: std.ArrayListUnmanaged(Func) = .{},
+
+eh_frame_hdr: ?ExceptionFrameHeader = null,
+// These lookup tables are only used if `eh_frame_hdr` is null
+cie_map: std.AutoArrayHashMapUnmanaged(u64, CommonInformationEntry) = .{},
+// Sorted by start_pc
+fde_list: std.ArrayListUnmanaged(FrameDescriptionEntry) = .{},
+
+pub const Section = struct {
+ data: []const u8,
+ // Module-relative virtual address.
+ // Only set if the section data was loaded from disk.
+ virtual_address: ?usize = null,
+ // If `data` is owned by this Dwarf.
+ owned: bool,
+
+ pub const Id = enum {
+ debug_info,
+ debug_abbrev,
+ debug_str,
+ debug_str_offsets,
+ debug_line,
+ debug_line_str,
+ debug_ranges,
+ debug_loclists,
+ debug_rnglists,
+ debug_addr,
+ debug_names,
+ debug_frame,
+ eh_frame,
+ eh_frame_hdr,
+ };
+
+ // For sections that are not memory mapped by the loader, this is an offset
+ // from `data.ptr` to where the section would have been mapped. Otherwise,
+ // `data` is directly backed by the section and the offset is zero.
+ pub fn virtualOffset(self: Section, base_address: usize) i64 {
+ return if (self.virtual_address) |va|
+ @as(i64, @intCast(base_address + va)) -
+ @as(i64, @intCast(@intFromPtr(self.data.ptr)))
+ else
+ 0;
+ }
+};
+
+pub const Abbrev = struct {
+ code: u64,
+ tag_id: u64,
+ has_children: bool,
+ attrs: []Attr,
+
+ fn deinit(abbrev: *Abbrev, allocator: Allocator) void {
+ allocator.free(abbrev.attrs);
+ abbrev.* = undefined;
+ }
+
+ const Attr = struct {
+ id: u64,
+ form_id: u64,
+ /// Only valid if form_id is .implicit_const
+ payload: i64,
+ };
+
+ const Table = struct {
+ // offset from .debug_abbrev
+ offset: u64,
+ abbrevs: []Abbrev,
+
+ fn deinit(table: *Table, allocator: Allocator) void {
+ for (table.abbrevs) |*abbrev| {
+ abbrev.deinit(allocator);
+ }
+ allocator.free(table.abbrevs);
+ table.* = undefined;
+ }
+
+ fn get(table: *const Table, abbrev_code: u64) ?*const Abbrev {
+ return for (table.abbrevs) |*abbrev| {
+ if (abbrev.code == abbrev_code) break abbrev;
+ } else null;
+ }
+ };
+};
+
+pub const CompileUnit = struct {
+ version: u16,
+ format: Format,
+ die: Die,
+ pc_range: ?PcRange,
+
+ str_offsets_base: usize,
+ addr_base: usize,
+ rnglists_base: usize,
+ loclists_base: usize,
+ frame_base: ?*const FormValue,
+};
+
+pub const FormValue = union(enum) {
+ addr: u64,
+ addrx: usize,
+ block: []const u8,
+ udata: u64,
+ data16: *const [16]u8,
+ sdata: i64,
+ exprloc: []const u8,
+ flag: bool,
+ sec_offset: u64,
+ ref: u64,
+ ref_addr: u64,
+ string: [:0]const u8,
+ strp: u64,
+ strx: usize,
+ line_strp: u64,
+ loclistx: u64,
+ rnglistx: u64,
+
+ fn getString(fv: FormValue, di: Dwarf) ![:0]const u8 {
+ switch (fv) {
+ .string => |s| return s,
+ .strp => |off| return di.getString(off),
+ .line_strp => |off| return di.getLineString(off),
+ else => return badDwarf(),
+ }
+ }
+
+ fn getUInt(fv: FormValue, comptime U: type) !U {
+ return switch (fv) {
+ inline .udata,
+ .sdata,
+ .sec_offset,
+ => |c| cast(U, c) orelse badDwarf(),
+ else => badDwarf(),
+ };
+ }
+};
+
+pub const Die = struct {
+ tag_id: u64,
+ has_children: bool,
+ attrs: []Attr,
+
+ const Attr = struct {
+ id: u64,
+ value: FormValue,
+ };
+
+ fn deinit(self: *Die, allocator: Allocator) void {
+ allocator.free(self.attrs);
+ self.* = undefined;
+ }
+
+ fn getAttr(self: *const Die, id: u64) ?*const FormValue {
+ for (self.attrs) |*attr| {
+ if (attr.id == id) return &attr.value;
+ }
+ return null;
+ }
+
+ fn getAttrAddr(
+ self: *const Die,
+ di: *const Dwarf,
+ id: u64,
+ compile_unit: CompileUnit,
+ ) error{ InvalidDebugInfo, MissingDebugInfo }!u64 {
+ const form_value = self.getAttr(id) orelse return error.MissingDebugInfo;
+ return switch (form_value.*) {
+ .addr => |value| value,
+ .addrx => |index| di.readDebugAddr(compile_unit, index),
+ else => error.InvalidDebugInfo,
+ };
+ }
+
+ fn getAttrSecOffset(self: *const Die, id: u64) !u64 {
+ const form_value = self.getAttr(id) orelse return error.MissingDebugInfo;
+ return form_value.getUInt(u64);
+ }
+
+ fn getAttrUnsignedLe(self: *const Die, id: u64) !u64 {
+ const form_value = self.getAttr(id) orelse return error.MissingDebugInfo;
+ return switch (form_value.*) {
+ .Const => |value| value.asUnsignedLe(),
+ else => error.InvalidDebugInfo,
+ };
+ }
+
+ fn getAttrRef(self: *const Die, id: u64) !u64 {
+ const form_value = self.getAttr(id) orelse return error.MissingDebugInfo;
+ return switch (form_value.*) {
+ .ref => |value| value,
+ else => error.InvalidDebugInfo,
+ };
+ }
+
+ pub fn getAttrString(
+ self: *const Die,
+ di: *Dwarf,
+ id: u64,
+ opt_str: ?[]const u8,
+ compile_unit: CompileUnit,
+ ) error{ InvalidDebugInfo, MissingDebugInfo }![]const u8 {
+ const form_value = self.getAttr(id) orelse return error.MissingDebugInfo;
+ switch (form_value.*) {
+ .string => |value| return value,
+ .strp => |offset| return di.getString(offset),
+ .strx => |index| {
+ const debug_str_offsets = di.section(.debug_str_offsets) orelse return badDwarf();
+ if (compile_unit.str_offsets_base == 0) return badDwarf();
+ switch (compile_unit.format) {
+ .@"32" => {
+ const byte_offset = compile_unit.str_offsets_base + 4 * index;
+ if (byte_offset + 4 > debug_str_offsets.len) return badDwarf();
+ const offset = readInt(u32, debug_str_offsets[byte_offset..][0..4], di.endian);
+ return getStringGeneric(opt_str, offset);
+ },
+ .@"64" => {
+ const byte_offset = compile_unit.str_offsets_base + 8 * index;
+ if (byte_offset + 8 > debug_str_offsets.len) return badDwarf();
+ const offset = readInt(u64, debug_str_offsets[byte_offset..][0..8], di.endian);
+ return getStringGeneric(opt_str, offset);
+ },
+ }
+ },
+ .line_strp => |offset| return di.getLineString(offset),
+ else => return badDwarf(),
+ }
+ }
+};
+
+/// This represents the decoded .eh_frame_hdr header
+pub const ExceptionFrameHeader = struct {
+ eh_frame_ptr: usize,
+ table_enc: u8,
+ fde_count: usize,
+ entries: []const u8,
+
+ pub fn entrySize(table_enc: u8) !u8 {
+ return switch (table_enc & EH.PE.type_mask) {
+ EH.PE.udata2,
+ EH.PE.sdata2,
+ => 4,
+ EH.PE.udata4,
+ EH.PE.sdata4,
+ => 8,
+ EH.PE.udata8,
+ EH.PE.sdata8,
+ => 16,
+ // This is a binary search table, so all entries must be the same length
+ else => return badDwarf(),
+ };
+ }
+
+ fn isValidPtr(
+ self: ExceptionFrameHeader,
+ comptime T: type,
+ ptr: usize,
+ ma: *StackIterator.MemoryAccessor,
+ eh_frame_len: ?usize,
+ ) bool {
+ if (eh_frame_len) |len| {
+ return ptr >= self.eh_frame_ptr and ptr <= self.eh_frame_ptr + len - @sizeOf(T);
+ } else {
+ return ma.load(T, ptr) != null;
+ }
+ }
+
+ /// Find an entry by binary searching the eh_frame_hdr section.
+ ///
+ /// Since the length of the eh_frame section (`eh_frame_len`) may not be known by the caller,
+ /// MemoryAccessor will be used to verify readability of the header entries.
+ /// If `eh_frame_len` is provided, then these checks can be skipped.
+ pub fn findEntry(
+ self: ExceptionFrameHeader,
+ ma: *StackIterator.MemoryAccessor,
+ eh_frame_len: ?usize,
+ eh_frame_hdr_ptr: usize,
+ pc: usize,
+ cie: *CommonInformationEntry,
+ fde: *FrameDescriptionEntry,
+ ) !void {
+ const entry_size = try entrySize(self.table_enc);
+
+ var left: usize = 0;
+ var len: usize = self.fde_count;
+
+ var fbr: FixedBufferReader = .{ .buf = self.entries, .endian = native_endian };
+
+ while (len > 1) {
+ const mid = left + len / 2;
+
+ fbr.pos = mid * entry_size;
+ const pc_begin = try readEhPointer(&fbr, self.table_enc, @sizeOf(usize), .{
+ .pc_rel_base = @intFromPtr(&self.entries[fbr.pos]),
+ .follow_indirect = true,
+ .data_rel_base = eh_frame_hdr_ptr,
+ }) orelse return badDwarf();
+
+ if (pc < pc_begin) {
+ len /= 2;
+ } else {
+ left = mid;
+ if (pc == pc_begin) break;
+ len -= len / 2;
+ }
+ }
+
+ if (len == 0) return badDwarf();
+ fbr.pos = left * entry_size;
+
+ // Read past the pc_begin field of the entry
+ _ = try readEhPointer(&fbr, self.table_enc, @sizeOf(usize), .{
+ .pc_rel_base = @intFromPtr(&self.entries[fbr.pos]),
+ .follow_indirect = true,
+ .data_rel_base = eh_frame_hdr_ptr,
+ }) orelse return badDwarf();
+
+ const fde_ptr = cast(usize, try readEhPointer(&fbr, self.table_enc, @sizeOf(usize), .{
+ .pc_rel_base = @intFromPtr(&self.entries[fbr.pos]),
+ .follow_indirect = true,
+ .data_rel_base = eh_frame_hdr_ptr,
+ }) orelse return badDwarf()) orelse return badDwarf();
+
+ if (fde_ptr < self.eh_frame_ptr) return badDwarf();
+
+ // Even if eh_frame_len is not specified, all ranges accssed are checked via MemoryAccessor
+ const eh_frame = @as([*]const u8, @ptrFromInt(self.eh_frame_ptr))[0 .. eh_frame_len orelse maxInt(u32)];
+
+ const fde_offset = fde_ptr - self.eh_frame_ptr;
+ var eh_frame_fbr: FixedBufferReader = .{
+ .buf = eh_frame,
+ .pos = fde_offset,
+ .endian = native_endian,
+ };
+
+ const fde_entry_header = try EntryHeader.read(&eh_frame_fbr, if (eh_frame_len == null) ma else null, .eh_frame);
+ if (!self.isValidPtr(u8, @intFromPtr(&fde_entry_header.entry_bytes[fde_entry_header.entry_bytes.len - 1]), ma, eh_frame_len)) return badDwarf();
+ if (fde_entry_header.type != .fde) return badDwarf();
+
+ // CIEs always come before FDEs (the offset is a subtraction), so we can assume this memory is readable
+ const cie_offset = fde_entry_header.type.fde;
+ try eh_frame_fbr.seekTo(cie_offset);
+ const cie_entry_header = try EntryHeader.read(&eh_frame_fbr, if (eh_frame_len == null) ma else null, .eh_frame);
+ if (!self.isValidPtr(u8, @intFromPtr(&cie_entry_header.entry_bytes[cie_entry_header.entry_bytes.len - 1]), ma, eh_frame_len)) return badDwarf();
+ if (cie_entry_header.type != .cie) return badDwarf();
+
+ cie.* = try CommonInformationEntry.parse(
+ cie_entry_header.entry_bytes,
+ 0,
+ true,
+ cie_entry_header.format,
+ .eh_frame,
+ cie_entry_header.length_offset,
+ @sizeOf(usize),
+ native_endian,
+ );
+
+ fde.* = try FrameDescriptionEntry.parse(
+ fde_entry_header.entry_bytes,
+ 0,
+ true,
+ cie.*,
+ @sizeOf(usize),
+ native_endian,
+ );
+ }
+};
+
+pub const EntryHeader = struct {
+ /// Offset of the length field in the backing buffer
+ length_offset: usize,
+ format: Format,
+ type: union(enum) {
+ cie,
+ /// Value is the offset of the corresponding CIE
+ fde: u64,
+ terminator,
+ },
+ /// The entry's contents, not including the ID field
+ entry_bytes: []const u8,
+
+ /// The length of the entry including the ID field, but not the length field itself
+ pub fn entryLength(self: EntryHeader) usize {
+ return self.entry_bytes.len + @as(u8, if (self.format == .@"64") 8 else 4);
+ }
+
+ /// Reads a header for either an FDE or a CIE, then advances the fbr to the position after the trailing structure.
+ /// `fbr` must be a FixedBufferReader backed by either the .eh_frame or .debug_frame sections.
+ pub fn read(
+ fbr: *FixedBufferReader,
+ opt_ma: ?*StackIterator.MemoryAccessor,
+ dwarf_section: Section.Id,
+ ) !EntryHeader {
+ assert(dwarf_section == .eh_frame or dwarf_section == .debug_frame);
+
+ const length_offset = fbr.pos;
+ const unit_header = try readUnitHeader(fbr, opt_ma);
+ const unit_length = cast(usize, unit_header.unit_length) orelse return badDwarf();
+ if (unit_length == 0) return .{
+ .length_offset = length_offset,
+ .format = unit_header.format,
+ .type = .terminator,
+ .entry_bytes = &.{},
+ };
+ const start_offset = fbr.pos;
+ const end_offset = start_offset + unit_length;
+ defer fbr.pos = end_offset;
+
+ const id = try if (opt_ma) |ma|
+ fbr.readAddressChecked(unit_header.format, ma)
+ else
+ fbr.readAddress(unit_header.format);
+ const entry_bytes = fbr.buf[fbr.pos..end_offset];
+ const cie_id: u64 = switch (dwarf_section) {
+ .eh_frame => CommonInformationEntry.eh_id,
+ .debug_frame => switch (unit_header.format) {
+ .@"32" => CommonInformationEntry.dwarf32_id,
+ .@"64" => CommonInformationEntry.dwarf64_id,
+ },
+ else => unreachable,
+ };
+
+ return .{
+ .length_offset = length_offset,
+ .format = unit_header.format,
+ .type = if (id == cie_id) .cie else .{ .fde = switch (dwarf_section) {
+ .eh_frame => try std.math.sub(u64, start_offset, id),
+ .debug_frame => id,
+ else => unreachable,
+ } },
+ .entry_bytes = entry_bytes,
+ };
+ }
+};
+
+pub const CommonInformationEntry = struct {
+ // Used in .eh_frame
+ pub const eh_id = 0;
+
+ // Used in .debug_frame (DWARF32)
+ pub const dwarf32_id = maxInt(u32);
+
+ // Used in .debug_frame (DWARF64)
+ pub const dwarf64_id = maxInt(u64);
+
+ // Offset of the length field of this entry in the eh_frame section.
+ // This is the key that FDEs use to reference CIEs.
+ length_offset: u64,
+ version: u8,
+ address_size: u8,
+ format: Format,
+
+ // Only present in version 4
+ segment_selector_size: ?u8,
+
+ code_alignment_factor: u32,
+ data_alignment_factor: i32,
+ return_address_register: u8,
+
+ aug_str: []const u8,
+ aug_data: []const u8,
+ lsda_pointer_enc: u8,
+ personality_enc: ?u8,
+ personality_routine_pointer: ?u64,
+ fde_pointer_enc: u8,
+ initial_instructions: []const u8,
+
+ pub fn isSignalFrame(self: CommonInformationEntry) bool {
+ for (self.aug_str) |c| if (c == 'S') return true;
+ return false;
+ }
+
+ pub fn addressesSignedWithBKey(self: CommonInformationEntry) bool {
+ for (self.aug_str) |c| if (c == 'B') return true;
+ return false;
+ }
+
+ pub fn mteTaggedFrame(self: CommonInformationEntry) bool {
+ for (self.aug_str) |c| if (c == 'G') return true;
+ return false;
+ }
+
+ /// This function expects to read the CIE starting with the version field.
+ /// The returned struct references memory backed by cie_bytes.
+ ///
+ /// See the FrameDescriptionEntry.parse documentation for the description
+ /// of `pc_rel_offset` and `is_runtime`.
+ ///
+ /// `length_offset` specifies the offset of this CIE's length field in the
+ /// .eh_frame / .debug_frame section.
+ pub fn parse(
+ cie_bytes: []const u8,
+ pc_rel_offset: i64,
+ is_runtime: bool,
+ format: Format,
+ dwarf_section: Section.Id,
+ length_offset: u64,
+ addr_size_bytes: u8,
+ endian: std.builtin.Endian,
+ ) !CommonInformationEntry {
+ if (addr_size_bytes > 8) return error.UnsupportedAddrSize;
+
+ var fbr: FixedBufferReader = .{ .buf = cie_bytes, .endian = endian };
+
+ const version = try fbr.readByte();
+ switch (dwarf_section) {
+ .eh_frame => if (version != 1 and version != 3) return error.UnsupportedDwarfVersion,
+ .debug_frame => if (version != 4) return error.UnsupportedDwarfVersion,
+ else => return error.UnsupportedDwarfSection,
+ }
+
+ var has_eh_data = false;
+ var has_aug_data = false;
+
+ var aug_str_len: usize = 0;
+ const aug_str_start = fbr.pos;
+ var aug_byte = try fbr.readByte();
+ while (aug_byte != 0) : (aug_byte = try fbr.readByte()) {
+ switch (aug_byte) {
+ 'z' => {
+ if (aug_str_len != 0) return badDwarf();
+ has_aug_data = true;
+ },
+ 'e' => {
+ if (has_aug_data or aug_str_len != 0) return badDwarf();
+ if (try fbr.readByte() != 'h') return badDwarf();
+ has_eh_data = true;
+ },
+ else => if (has_eh_data) return badDwarf(),
+ }
+
+ aug_str_len += 1;
+ }
+
+ if (has_eh_data) {
+ // legacy data created by older versions of gcc - unsupported here
+ for (0..addr_size_bytes) |_| _ = try fbr.readByte();
+ }
+
+ const address_size = if (version == 4) try fbr.readByte() else addr_size_bytes;
+ const segment_selector_size = if (version == 4) try fbr.readByte() else null;
+
+ const code_alignment_factor = try fbr.readUleb128(u32);
+ const data_alignment_factor = try fbr.readIleb128(i32);
+ const return_address_register = if (version == 1) try fbr.readByte() else try fbr.readUleb128(u8);
+
+ var lsda_pointer_enc: u8 = EH.PE.omit;
+ var personality_enc: ?u8 = null;
+ var personality_routine_pointer: ?u64 = null;
+ var fde_pointer_enc: u8 = EH.PE.absptr;
+
+ var aug_data: []const u8 = &[_]u8{};
+ const aug_str = if (has_aug_data) blk: {
+ const aug_data_len = try fbr.readUleb128(usize);
+ const aug_data_start = fbr.pos;
+ aug_data = cie_bytes[aug_data_start..][0..aug_data_len];
+
+ const aug_str = cie_bytes[aug_str_start..][0..aug_str_len];
+ for (aug_str[1..]) |byte| {
+ switch (byte) {
+ 'L' => {
+ lsda_pointer_enc = try fbr.readByte();
+ },
+ 'P' => {
+ personality_enc = try fbr.readByte();
+ personality_routine_pointer = try readEhPointer(&fbr, personality_enc.?, addr_size_bytes, .{
+ .pc_rel_base = try pcRelBase(@intFromPtr(&cie_bytes[fbr.pos]), pc_rel_offset),
+ .follow_indirect = is_runtime,
+ });
+ },
+ 'R' => {
+ fde_pointer_enc = try fbr.readByte();
+ },
+ 'S', 'B', 'G' => {},
+ else => return badDwarf(),
+ }
+ }
+
+ // aug_data_len can include padding so the CIE ends on an address boundary
+ fbr.pos = aug_data_start + aug_data_len;
+ break :blk aug_str;
+ } else &[_]u8{};
+
+ const initial_instructions = cie_bytes[fbr.pos..];
+ return .{
+ .length_offset = length_offset,
+ .version = version,
+ .address_size = address_size,
+ .format = format,
+ .segment_selector_size = segment_selector_size,
+ .code_alignment_factor = code_alignment_factor,
+ .data_alignment_factor = data_alignment_factor,
+ .return_address_register = return_address_register,
+ .aug_str = aug_str,
+ .aug_data = aug_data,
+ .lsda_pointer_enc = lsda_pointer_enc,
+ .personality_enc = personality_enc,
+ .personality_routine_pointer = personality_routine_pointer,
+ .fde_pointer_enc = fde_pointer_enc,
+ .initial_instructions = initial_instructions,
+ };
+ }
+};
+
+pub const FrameDescriptionEntry = struct {
+ // Offset into eh_frame where the CIE for this FDE is stored
+ cie_length_offset: u64,
+
+ pc_begin: u64,
+ pc_range: u64,
+ lsda_pointer: ?u64,
+ aug_data: []const u8,
+ instructions: []const u8,
+
+ /// This function expects to read the FDE starting at the PC Begin field.
+ /// The returned struct references memory backed by `fde_bytes`.
+ ///
+ /// `pc_rel_offset` specifies an offset to be applied to pc_rel_base values
+ /// used when decoding pointers. This should be set to zero if fde_bytes is
+ /// backed by the memory of a .eh_frame / .debug_frame section in the running executable.
+ /// Otherwise, it should be the relative offset to translate addresses from
+ /// where the section is currently stored in memory, to where it *would* be
+ /// stored at runtime: section base addr - backing data base ptr.
+ ///
+ /// Similarly, `is_runtime` specifies this function is being called on a runtime
+ /// section, and so indirect pointers can be followed.
+ pub fn parse(
+ fde_bytes: []const u8,
+ pc_rel_offset: i64,
+ is_runtime: bool,
+ cie: CommonInformationEntry,
+ addr_size_bytes: u8,
+ endian: std.builtin.Endian,
+ ) !FrameDescriptionEntry {
+ if (addr_size_bytes > 8) return error.InvalidAddrSize;
+
+ var fbr: FixedBufferReader = .{ .buf = fde_bytes, .endian = endian };
+
+ const pc_begin = try readEhPointer(&fbr, cie.fde_pointer_enc, addr_size_bytes, .{
+ .pc_rel_base = try pcRelBase(@intFromPtr(&fde_bytes[fbr.pos]), pc_rel_offset),
+ .follow_indirect = is_runtime,
+ }) orelse return badDwarf();
+
+ const pc_range = try readEhPointer(&fbr, cie.fde_pointer_enc, addr_size_bytes, .{
+ .pc_rel_base = 0,
+ .follow_indirect = false,
+ }) orelse return badDwarf();
+
+ var aug_data: []const u8 = &[_]u8{};
+ const lsda_pointer = if (cie.aug_str.len > 0) blk: {
+ const aug_data_len = try fbr.readUleb128(usize);
+ const aug_data_start = fbr.pos;
+ aug_data = fde_bytes[aug_data_start..][0..aug_data_len];
+
+ const lsda_pointer = if (cie.lsda_pointer_enc != EH.PE.omit)
+ try readEhPointer(&fbr, cie.lsda_pointer_enc, addr_size_bytes, .{
+ .pc_rel_base = try pcRelBase(@intFromPtr(&fde_bytes[fbr.pos]), pc_rel_offset),
+ .follow_indirect = is_runtime,
+ })
+ else
+ null;
+
+ fbr.pos = aug_data_start + aug_data_len;
+ break :blk lsda_pointer;
+ } else null;
+
+ const instructions = fde_bytes[fbr.pos..];
+ return .{
+ .cie_length_offset = cie.length_offset,
+ .pc_begin = pc_begin,
+ .pc_range = pc_range,
+ .lsda_pointer = lsda_pointer,
+ .aug_data = aug_data,
+ .instructions = instructions,
+ };
+ }
+};
+
+pub const UnwindContext = struct {
+ allocator: Allocator,
+ cfa: ?usize,
+ pc: usize,
+ thread_context: *std.debug.ThreadContext,
+ reg_context: abi.RegisterContext,
+ vm: call_frame.VirtualMachine,
+ stack_machine: expression.StackMachine(.{ .call_frame_context = true }),
+
+ pub fn init(
+ allocator: Allocator,
+ thread_context: *const std.debug.ThreadContext,
+ ) !UnwindContext {
+ const pc = abi.stripInstructionPtrAuthCode(
+ (try abi.regValueNative(
+ usize,
+ thread_context,
+ abi.ipRegNum(),
+ null,
+ )).*,
+ );
+
+ const context_copy = try allocator.create(std.debug.ThreadContext);
+ std.debug.copyContext(thread_context, context_copy);
+
+ return .{
+ .allocator = allocator,
+ .cfa = null,
+ .pc = pc,
+ .thread_context = context_copy,
+ .reg_context = undefined,
+ .vm = .{},
+ .stack_machine = .{},
+ };
+ }
+
+ pub fn deinit(self: *UnwindContext) void {
+ self.vm.deinit(self.allocator);
+ self.stack_machine.deinit(self.allocator);
+ self.allocator.destroy(self.thread_context);
+ self.* = undefined;
+ }
+
+ pub fn getFp(self: *const UnwindContext) !usize {
+ return (try abi.regValueNative(usize, self.thread_context, abi.fpRegNum(self.reg_context), self.reg_context)).*;
+ }
+};
+
+const num_sections = std.enums.directEnumArrayLen(Section.Id, 0);
+pub const SectionArray = [num_sections]?Section;
+pub const null_section_array = [_]?Section{null} ** num_sections;
+
+/// Initialize DWARF info. The caller has the responsibility to initialize most
+/// the `Dwarf` fields before calling. `binary_mem` is the raw bytes of the
+/// main binary file (not the secondary debug info file).
+pub fn open(di: *Dwarf, allocator: Allocator) !void {
+ try di.scanAllFunctions(allocator);
+ try di.scanAllCompileUnits(allocator);
+}
+
+const PcRange = struct {
+ start: u64,
+ end: u64,
+};
+
+const Func = struct {
+ pc_range: ?PcRange,
+ name: ?[]const u8,
+};
+
+pub fn section(di: Dwarf, dwarf_section: Section.Id) ?[]const u8 {
+ return if (di.sections[@intFromEnum(dwarf_section)]) |s| s.data else null;
+}
+
+pub fn sectionVirtualOffset(di: Dwarf, dwarf_section: Section.Id, base_address: usize) ?i64 {
+ return if (di.sections[@intFromEnum(dwarf_section)]) |s| s.virtualOffset(base_address) else null;
+}
+
+pub fn deinit(di: *Dwarf, allocator: Allocator) void {
+ for (di.sections) |opt_section| {
+ if (opt_section) |s| if (s.owned) allocator.free(s.data);
+ }
+ for (di.abbrev_table_list.items) |*abbrev| {
+ abbrev.deinit(allocator);
+ }
+ di.abbrev_table_list.deinit(allocator);
+ for (di.compile_unit_list.items) |*cu| {
+ cu.die.deinit(allocator);
+ }
+ di.compile_unit_list.deinit(allocator);
+ di.func_list.deinit(allocator);
+ di.cie_map.deinit(allocator);
+ di.fde_list.deinit(allocator);
+ di.* = undefined;
+}
+
+pub fn getSymbolName(di: *Dwarf, address: u64) ?[]const u8 {
+ for (di.func_list.items) |*func| {
+ if (func.pc_range) |range| {
+ if (address >= range.start and address < range.end) {
+ return func.name;
+ }
+ }
+ }
+
+ return null;
+}
+
+fn scanAllFunctions(di: *Dwarf, allocator: Allocator) !void {
+ var fbr: FixedBufferReader = .{ .buf = di.section(.debug_info).?, .endian = di.endian };
+ var this_unit_offset: u64 = 0;
+
+ while (this_unit_offset < fbr.buf.len) {
+ try fbr.seekTo(this_unit_offset);
+
+ const unit_header = try readUnitHeader(&fbr, null);
+ if (unit_header.unit_length == 0) return;
+ const next_offset = unit_header.header_length + unit_header.unit_length;
+
+ const version = try fbr.readInt(u16);
+ if (version < 2 or version > 5) return badDwarf();
+
+ var address_size: u8 = undefined;
+ var debug_abbrev_offset: u64 = undefined;
+ if (version >= 5) {
+ const unit_type = try fbr.readInt(u8);
+ if (unit_type != DW.UT.compile) return badDwarf();
+ address_size = try fbr.readByte();
+ debug_abbrev_offset = try fbr.readAddress(unit_header.format);
+ } else {
+ debug_abbrev_offset = try fbr.readAddress(unit_header.format);
+ address_size = try fbr.readByte();
+ }
+ if (address_size != @sizeOf(usize)) return badDwarf();
+
+ const abbrev_table = try di.getAbbrevTable(allocator, debug_abbrev_offset);
+
+ var max_attrs: usize = 0;
+ var zig_padding_abbrev_code: u7 = 0;
+ for (abbrev_table.abbrevs) |abbrev| {
+ max_attrs = @max(max_attrs, abbrev.attrs.len);
+ if (cast(u7, abbrev.code)) |code| {
+ if (abbrev.tag_id == DW.TAG.ZIG_padding and
+ !abbrev.has_children and
+ abbrev.attrs.len == 0)
+ {
+ zig_padding_abbrev_code = code;
+ }
+ }
+ }
+ const attrs_buf = try allocator.alloc(Die.Attr, max_attrs * 3);
+ defer allocator.free(attrs_buf);
+ var attrs_bufs: [3][]Die.Attr = undefined;
+ for (&attrs_bufs, 0..) |*buf, index| buf.* = attrs_buf[index * max_attrs ..][0..max_attrs];
+
+ const next_unit_pos = this_unit_offset + next_offset;
+
+ var compile_unit: CompileUnit = .{
+ .version = version,
+ .format = unit_header.format,
+ .die = undefined,
+ .pc_range = null,
+
+ .str_offsets_base = 0,
+ .addr_base = 0,
+ .rnglists_base = 0,
+ .loclists_base = 0,
+ .frame_base = null,
+ };
+
+ while (true) {
+ fbr.pos = std.mem.indexOfNonePos(u8, fbr.buf, fbr.pos, &.{
+ zig_padding_abbrev_code, 0,
+ }) orelse fbr.buf.len;
+ if (fbr.pos >= next_unit_pos) break;
+ var die_obj = (try parseDie(
+ &fbr,
+ attrs_bufs[0],
+ abbrev_table,
+ unit_header.format,
+ )) orelse continue;
+
+ switch (die_obj.tag_id) {
+ DW.TAG.compile_unit => {
+ compile_unit.die = die_obj;
+ compile_unit.die.attrs = attrs_bufs[1][0..die_obj.attrs.len];
+ @memcpy(compile_unit.die.attrs, die_obj.attrs);
+
+ compile_unit.str_offsets_base = if (die_obj.getAttr(AT.str_offsets_base)) |fv| try fv.getUInt(usize) else 0;
+ compile_unit.addr_base = if (die_obj.getAttr(AT.addr_base)) |fv| try fv.getUInt(usize) else 0;
+ compile_unit.rnglists_base = if (die_obj.getAttr(AT.rnglists_base)) |fv| try fv.getUInt(usize) else 0;
+ compile_unit.loclists_base = if (die_obj.getAttr(AT.loclists_base)) |fv| try fv.getUInt(usize) else 0;
+ compile_unit.frame_base = die_obj.getAttr(AT.frame_base);
+ },
+ DW.TAG.subprogram, DW.TAG.inlined_subroutine, DW.TAG.subroutine, DW.TAG.entry_point => {
+ const fn_name = x: {
+ var this_die_obj = die_obj;
+ // Prevent endless loops
+ for (0..3) |_| {
+ if (this_die_obj.getAttr(AT.name)) |_| {
+ break :x try this_die_obj.getAttrString(di, AT.name, di.section(.debug_str), compile_unit);
+ } else if (this_die_obj.getAttr(AT.abstract_origin)) |_| {
+ const after_die_offset = fbr.pos;
+ defer fbr.pos = after_die_offset;
+
+ // Follow the DIE it points to and repeat
+ const ref_offset = try this_die_obj.getAttrRef(AT.abstract_origin);
+ if (ref_offset > next_offset) return badDwarf();
+ try fbr.seekTo(this_unit_offset + ref_offset);
+ this_die_obj = (try parseDie(
+ &fbr,
+ attrs_bufs[2],
+ abbrev_table,
+ unit_header.format,
+ )) orelse return badDwarf();
+ } else if (this_die_obj.getAttr(AT.specification)) |_| {
+ const after_die_offset = fbr.pos;
+ defer fbr.pos = after_die_offset;
+
+ // Follow the DIE it points to and repeat
+ const ref_offset = try this_die_obj.getAttrRef(AT.specification);
+ if (ref_offset > next_offset) return badDwarf();
+ try fbr.seekTo(this_unit_offset + ref_offset);
+ this_die_obj = (try parseDie(
+ &fbr,
+ attrs_bufs[2],
+ abbrev_table,
+ unit_header.format,
+ )) orelse return badDwarf();
+ } else {
+ break :x null;
+ }
+ }
+
+ break :x null;
+ };
+
+ var range_added = if (die_obj.getAttrAddr(di, AT.low_pc, compile_unit)) |low_pc| blk: {
+ if (die_obj.getAttr(AT.high_pc)) |high_pc_value| {
+ const pc_end = switch (high_pc_value.*) {
+ .addr => |value| value,
+ .udata => |offset| low_pc + offset,
+ else => return badDwarf(),
+ };
+
+ try di.func_list.append(allocator, .{
+ .name = fn_name,
+ .pc_range = .{
+ .start = low_pc,
+ .end = pc_end,
+ },
+ });
+
+ break :blk true;
+ }
+
+ break :blk false;
+ } else |err| blk: {
+ if (err != error.MissingDebugInfo) return err;
+ break :blk false;
+ };
+
+ if (die_obj.getAttr(AT.ranges)) |ranges_value| blk: {
+ var iter = DebugRangeIterator.init(ranges_value, di, &compile_unit) catch |err| {
+ if (err != error.MissingDebugInfo) return err;
+ break :blk;
+ };
+
+ while (try iter.next()) |range| {
+ range_added = true;
+ try di.func_list.append(allocator, .{
+ .name = fn_name,
+ .pc_range = .{
+ .start = range.start_addr,
+ .end = range.end_addr,
+ },
+ });
+ }
+ }
+
+ if (fn_name != null and !range_added) {
+ try di.func_list.append(allocator, .{
+ .name = fn_name,
+ .pc_range = null,
+ });
+ }
+ },
+ else => {},
+ }
+ }
+
+ this_unit_offset += next_offset;
+ }
+}
+
+fn scanAllCompileUnits(di: *Dwarf, allocator: Allocator) !void {
+ var fbr: FixedBufferReader = .{ .buf = di.section(.debug_info).?, .endian = di.endian };
+ var this_unit_offset: u64 = 0;
+
+ var attrs_buf = std.ArrayList(Die.Attr).init(allocator);
+ defer attrs_buf.deinit();
+
+ while (this_unit_offset < fbr.buf.len) {
+ try fbr.seekTo(this_unit_offset);
+
+ const unit_header = try readUnitHeader(&fbr, null);
+ if (unit_header.unit_length == 0) return;
+ const next_offset = unit_header.header_length + unit_header.unit_length;
+
+ const version = try fbr.readInt(u16);
+ if (version < 2 or version > 5) return badDwarf();
+
+ var address_size: u8 = undefined;
+ var debug_abbrev_offset: u64 = undefined;
+ if (version >= 5) {
+ const unit_type = try fbr.readInt(u8);
+ if (unit_type != UT.compile) return badDwarf();
+ address_size = try fbr.readByte();
+ debug_abbrev_offset = try fbr.readAddress(unit_header.format);
+ } else {
+ debug_abbrev_offset = try fbr.readAddress(unit_header.format);
+ address_size = try fbr.readByte();
+ }
+ if (address_size != @sizeOf(usize)) return badDwarf();
+
+ const abbrev_table = try di.getAbbrevTable(allocator, debug_abbrev_offset);
+
+ var max_attrs: usize = 0;
+ for (abbrev_table.abbrevs) |abbrev| {
+ max_attrs = @max(max_attrs, abbrev.attrs.len);
+ }
+ try attrs_buf.resize(max_attrs);
+
+ var compile_unit_die = (try parseDie(
+ &fbr,
+ attrs_buf.items,
+ abbrev_table,
+ unit_header.format,
+ )) orelse return badDwarf();
+
+ if (compile_unit_die.tag_id != DW.TAG.compile_unit) return badDwarf();
+
+ compile_unit_die.attrs = try allocator.dupe(Die.Attr, compile_unit_die.attrs);
+
+ var compile_unit: CompileUnit = .{
+ .version = version,
+ .format = unit_header.format,
+ .pc_range = null,
+ .die = compile_unit_die,
+ .str_offsets_base = if (compile_unit_die.getAttr(AT.str_offsets_base)) |fv| try fv.getUInt(usize) else 0,
+ .addr_base = if (compile_unit_die.getAttr(AT.addr_base)) |fv| try fv.getUInt(usize) else 0,
+ .rnglists_base = if (compile_unit_die.getAttr(AT.rnglists_base)) |fv| try fv.getUInt(usize) else 0,
+ .loclists_base = if (compile_unit_die.getAttr(AT.loclists_base)) |fv| try fv.getUInt(usize) else 0,
+ .frame_base = compile_unit_die.getAttr(AT.frame_base),
+ };
+
+ compile_unit.pc_range = x: {
+ if (compile_unit_die.getAttrAddr(di, AT.low_pc, compile_unit)) |low_pc| {
+ if (compile_unit_die.getAttr(AT.high_pc)) |high_pc_value| {
+ const pc_end = switch (high_pc_value.*) {
+ .addr => |value| value,
+ .udata => |offset| low_pc + offset,
+ else => return badDwarf(),
+ };
+ break :x PcRange{
+ .start = low_pc,
+ .end = pc_end,
+ };
+ } else {
+ break :x null;
+ }
+ } else |err| {
+ if (err != error.MissingDebugInfo) return err;
+ break :x null;
+ }
+ };
+
+ try di.compile_unit_list.append(allocator, compile_unit);
+
+ this_unit_offset += next_offset;
+ }
+}
+
+const DebugRangeIterator = struct {
+ base_address: u64,
+ section_type: Section.Id,
+ di: *const Dwarf,
+ compile_unit: *const CompileUnit,
+ fbr: FixedBufferReader,
+
+ pub fn init(ranges_value: *const FormValue, di: *const Dwarf, compile_unit: *const CompileUnit) !@This() {
+ const section_type = if (compile_unit.version >= 5) Section.Id.debug_rnglists else Section.Id.debug_ranges;
+ const debug_ranges = di.section(section_type) orelse return error.MissingDebugInfo;
+
+ const ranges_offset = switch (ranges_value.*) {
+ .sec_offset, .udata => |off| off,
+ .rnglistx => |idx| off: {
+ switch (compile_unit.format) {
+ .@"32" => {
+ const offset_loc = @as(usize, @intCast(compile_unit.rnglists_base + 4 * idx));
+ if (offset_loc + 4 > debug_ranges.len) return badDwarf();
+ const offset = readInt(u32, debug_ranges[offset_loc..][0..4], di.endian);
+ break :off compile_unit.rnglists_base + offset;
+ },
+ .@"64" => {
+ const offset_loc = @as(usize, @intCast(compile_unit.rnglists_base + 8 * idx));
+ if (offset_loc + 8 > debug_ranges.len) return badDwarf();
+ const offset = readInt(u64, debug_ranges[offset_loc..][0..8], di.endian);
+ break :off compile_unit.rnglists_base + offset;
+ },
+ }
+ },
+ else => return badDwarf(),
+ };
+
+ // All the addresses in the list are relative to the value
+ // specified by DW_AT.low_pc or to some other value encoded
+ // in the list itself.
+ // If no starting value is specified use zero.
+ const base_address = compile_unit.die.getAttrAddr(di, AT.low_pc, compile_unit.*) catch |err| switch (err) {
+ error.MissingDebugInfo => 0,
+ else => return err,
+ };
+
+ return .{
+ .base_address = base_address,
+ .section_type = section_type,
+ .di = di,
+ .compile_unit = compile_unit,
+ .fbr = .{
+ .buf = debug_ranges,
+ .pos = cast(usize, ranges_offset) orelse return badDwarf(),
+ .endian = di.endian,
+ },
+ };
+ }
+
+ // Returns the next range in the list, or null if the end was reached.
+ pub fn next(self: *@This()) !?struct { start_addr: u64, end_addr: u64 } {
+ switch (self.section_type) {
+ .debug_rnglists => {
+ const kind = try self.fbr.readByte();
+ switch (kind) {
+ RLE.end_of_list => return null,
+ RLE.base_addressx => {
+ const index = try self.fbr.readUleb128(usize);
+ self.base_address = try self.di.readDebugAddr(self.compile_unit.*, index);
+ return try self.next();
+ },
+ RLE.startx_endx => {
+ const start_index = try self.fbr.readUleb128(usize);
+ const start_addr = try self.di.readDebugAddr(self.compile_unit.*, start_index);
+
+ const end_index = try self.fbr.readUleb128(usize);
+ const end_addr = try self.di.readDebugAddr(self.compile_unit.*, end_index);
+
+ return .{
+ .start_addr = start_addr,
+ .end_addr = end_addr,
+ };
+ },
+ RLE.startx_length => {
+ const start_index = try self.fbr.readUleb128(usize);
+ const start_addr = try self.di.readDebugAddr(self.compile_unit.*, start_index);
+
+ const len = try self.fbr.readUleb128(usize);
+ const end_addr = start_addr + len;
+
+ return .{
+ .start_addr = start_addr,
+ .end_addr = end_addr,
+ };
+ },
+ RLE.offset_pair => {
+ const start_addr = try self.fbr.readUleb128(usize);
+ const end_addr = try self.fbr.readUleb128(usize);
+
+ // This is the only kind that uses the base address
+ return .{
+ .start_addr = self.base_address + start_addr,
+ .end_addr = self.base_address + end_addr,
+ };
+ },
+ RLE.base_address => {
+ self.base_address = try self.fbr.readInt(usize);
+ return try self.next();
+ },
+ RLE.start_end => {
+ const start_addr = try self.fbr.readInt(usize);
+ const end_addr = try self.fbr.readInt(usize);
+
+ return .{
+ .start_addr = start_addr,
+ .end_addr = end_addr,
+ };
+ },
+ RLE.start_length => {
+ const start_addr = try self.fbr.readInt(usize);
+ const len = try self.fbr.readUleb128(usize);
+ const end_addr = start_addr + len;
+
+ return .{
+ .start_addr = start_addr,
+ .end_addr = end_addr,
+ };
+ },
+ else => return badDwarf(),
+ }
+ },
+ .debug_ranges => {
+ const start_addr = try self.fbr.readInt(usize);
+ const end_addr = try self.fbr.readInt(usize);
+ if (start_addr == 0 and end_addr == 0) return null;
+
+ // This entry selects a new value for the base address
+ if (start_addr == maxInt(usize)) {
+ self.base_address = end_addr;
+ return try self.next();
+ }
+
+ return .{
+ .start_addr = self.base_address + start_addr,
+ .end_addr = self.base_address + end_addr,
+ };
+ },
+ else => unreachable,
+ }
+ }
+};
+
+pub fn findCompileUnit(di: *const Dwarf, target_address: u64) !*const CompileUnit {
+ for (di.compile_unit_list.items) |*compile_unit| {
+ if (compile_unit.pc_range) |range| {
+ if (target_address >= range.start and target_address < range.end) return compile_unit;
+ }
+
+ const ranges_value = compile_unit.die.getAttr(AT.ranges) orelse continue;
+ var iter = DebugRangeIterator.init(ranges_value, di, compile_unit) catch continue;
+ while (try iter.next()) |range| {
+ if (target_address >= range.start_addr and target_address < range.end_addr) return compile_unit;
+ }
+ }
+
+ return missingDwarf();
+}
+
+/// Gets an already existing AbbrevTable given the abbrev_offset, or if not found,
+/// seeks in the stream and parses it.
+fn getAbbrevTable(di: *Dwarf, allocator: Allocator, abbrev_offset: u64) !*const Abbrev.Table {
+ for (di.abbrev_table_list.items) |*table| {
+ if (table.offset == abbrev_offset) {
+ return table;
+ }
+ }
+ try di.abbrev_table_list.append(
+ allocator,
+ try di.parseAbbrevTable(allocator, abbrev_offset),
+ );
+ return &di.abbrev_table_list.items[di.abbrev_table_list.items.len - 1];
+}
+
+fn parseAbbrevTable(di: *Dwarf, allocator: Allocator, offset: u64) !Abbrev.Table {
+ var fbr: FixedBufferReader = .{
+ .buf = di.section(.debug_abbrev).?,
+ .pos = cast(usize, offset) orelse return badDwarf(),
+ .endian = di.endian,
+ };
+
+ var abbrevs = std.ArrayList(Abbrev).init(allocator);
+ defer {
+ for (abbrevs.items) |*abbrev| {
+ abbrev.deinit(allocator);
+ }
+ abbrevs.deinit();
+ }
+
+ var attrs = std.ArrayList(Abbrev.Attr).init(allocator);
+ defer attrs.deinit();
+
+ while (true) {
+ const code = try fbr.readUleb128(u64);
+ if (code == 0) break;
+ const tag_id = try fbr.readUleb128(u64);
+ const has_children = (try fbr.readByte()) == DW.CHILDREN.yes;
+
+ while (true) {
+ const attr_id = try fbr.readUleb128(u64);
+ const form_id = try fbr.readUleb128(u64);
+ if (attr_id == 0 and form_id == 0) break;
+ try attrs.append(.{
+ .id = attr_id,
+ .form_id = form_id,
+ .payload = switch (form_id) {
+ FORM.implicit_const => try fbr.readIleb128(i64),
+ else => undefined,
+ },
+ });
+ }
+
+ try abbrevs.append(.{
+ .code = code,
+ .tag_id = tag_id,
+ .has_children = has_children,
+ .attrs = try attrs.toOwnedSlice(),
+ });
+ }
+
+ return .{
+ .offset = offset,
+ .abbrevs = try abbrevs.toOwnedSlice(),
+ };
+}
+
+fn parseDie(
+ fbr: *FixedBufferReader,
+ attrs_buf: []Die.Attr,
+ abbrev_table: *const Abbrev.Table,
+ format: Format,
+) !?Die {
+ const abbrev_code = try fbr.readUleb128(u64);
+ if (abbrev_code == 0) return null;
+ const table_entry = abbrev_table.get(abbrev_code) orelse return badDwarf();
+
+ const attrs = attrs_buf[0..table_entry.attrs.len];
+ for (attrs, table_entry.attrs) |*result_attr, attr| result_attr.* = Die.Attr{
+ .id = attr.id,
+ .value = try parseFormValue(
+ fbr,
+ attr.form_id,
+ format,
+ attr.payload,
+ ),
+ };
+ return .{
+ .tag_id = table_entry.tag_id,
+ .has_children = table_entry.has_children,
+ .attrs = attrs,
+ };
+}
+
+pub fn getLineNumberInfo(
+ di: *Dwarf,
+ allocator: Allocator,
+ compile_unit: CompileUnit,
+ target_address: u64,
+) !std.debug.LineInfo {
+ const compile_unit_cwd = try compile_unit.die.getAttrString(di, AT.comp_dir, di.section(.debug_line_str), compile_unit);
+ const line_info_offset = try compile_unit.die.getAttrSecOffset(AT.stmt_list);
+
+ var fbr: FixedBufferReader = .{ .buf = di.section(.debug_line).?, .endian = di.endian };
+ try fbr.seekTo(line_info_offset);
+
+ const unit_header = try readUnitHeader(&fbr, null);
+ if (unit_header.unit_length == 0) return missingDwarf();
+ const next_offset = unit_header.header_length + unit_header.unit_length;
+
+ const version = try fbr.readInt(u16);
+ if (version < 2) return badDwarf();
+
+ var addr_size: u8 = switch (unit_header.format) {
+ .@"32" => 4,
+ .@"64" => 8,
+ };
+ var seg_size: u8 = 0;
+ if (version >= 5) {
+ addr_size = try fbr.readByte();
+ seg_size = try fbr.readByte();
+ }
+
+ const prologue_length = try fbr.readAddress(unit_header.format);
+ const prog_start_offset = fbr.pos + prologue_length;
+
+ const minimum_instruction_length = try fbr.readByte();
+ if (minimum_instruction_length == 0) return badDwarf();
+
+ if (version >= 4) {
+ // maximum_operations_per_instruction
+ _ = try fbr.readByte();
+ }
+
+ const default_is_stmt = (try fbr.readByte()) != 0;
+ const line_base = try fbr.readByteSigned();
+
+ const line_range = try fbr.readByte();
+ if (line_range == 0) return badDwarf();
+
+ const opcode_base = try fbr.readByte();
+
+ const standard_opcode_lengths = try fbr.readBytes(opcode_base - 1);
+
+ var include_directories = std.ArrayList(FileEntry).init(allocator);
+ defer include_directories.deinit();
+ var file_entries = std.ArrayList(FileEntry).init(allocator);
+ defer file_entries.deinit();
+
+ if (version < 5) {
+ try include_directories.append(.{ .path = compile_unit_cwd });
+
+ while (true) {
+ const dir = try fbr.readBytesTo(0);
+ if (dir.len == 0) break;
+ try include_directories.append(.{ .path = dir });
+ }
+
+ while (true) {
+ const file_name = try fbr.readBytesTo(0);
+ if (file_name.len == 0) break;
+ const dir_index = try fbr.readUleb128(u32);
+ const mtime = try fbr.readUleb128(u64);
+ const size = try fbr.readUleb128(u64);
+ try file_entries.append(.{
+ .path = file_name,
+ .dir_index = dir_index,
+ .mtime = mtime,
+ .size = size,
+ });
+ }
+ } else {
+ const FileEntFmt = struct {
+ content_type_code: u8,
+ form_code: u16,
+ };
+ {
+ var dir_ent_fmt_buf: [10]FileEntFmt = undefined;
+ const directory_entry_format_count = try fbr.readByte();
+ if (directory_entry_format_count > dir_ent_fmt_buf.len) return badDwarf();
+ for (dir_ent_fmt_buf[0..directory_entry_format_count]) |*ent_fmt| {
+ ent_fmt.* = .{
+ .content_type_code = try fbr.readUleb128(u8),
+ .form_code = try fbr.readUleb128(u16),
+ };
+ }
+
+ const directories_count = try fbr.readUleb128(usize);
+ try include_directories.ensureUnusedCapacity(directories_count);
+ {
+ var i: usize = 0;
+ while (i < directories_count) : (i += 1) {
+ var e: FileEntry = .{ .path = &.{} };
+ for (dir_ent_fmt_buf[0..directory_entry_format_count]) |ent_fmt| {
+ const form_value = try parseFormValue(
+ &fbr,
+ ent_fmt.form_code,
+ unit_header.format,
+ null,
+ );
+ switch (ent_fmt.content_type_code) {
+ DW.LNCT.path => e.path = try form_value.getString(di.*),
+ DW.LNCT.directory_index => e.dir_index = try form_value.getUInt(u32),
+ DW.LNCT.timestamp => e.mtime = try form_value.getUInt(u64),
+ DW.LNCT.size => e.size = try form_value.getUInt(u64),
+ DW.LNCT.MD5 => e.md5 = switch (form_value) {
+ .data16 => |data16| data16.*,
+ else => return badDwarf(),
+ },
+ else => continue,
+ }
+ }
+ include_directories.appendAssumeCapacity(e);
+ }
+ }
+ }
+
+ var file_ent_fmt_buf: [10]FileEntFmt = undefined;
+ const file_name_entry_format_count = try fbr.readByte();
+ if (file_name_entry_format_count > file_ent_fmt_buf.len) return badDwarf();
+ for (file_ent_fmt_buf[0..file_name_entry_format_count]) |*ent_fmt| {
+ ent_fmt.* = .{
+ .content_type_code = try fbr.readUleb128(u8),
+ .form_code = try fbr.readUleb128(u16),
+ };
+ }
+
+ const file_names_count = try fbr.readUleb128(usize);
+ try file_entries.ensureUnusedCapacity(file_names_count);
+ {
+ var i: usize = 0;
+ while (i < file_names_count) : (i += 1) {
+ var e: FileEntry = .{ .path = &.{} };
+ for (file_ent_fmt_buf[0..file_name_entry_format_count]) |ent_fmt| {
+ const form_value = try parseFormValue(
+ &fbr,
+ ent_fmt.form_code,
+ unit_header.format,
+ null,
+ );
+ switch (ent_fmt.content_type_code) {
+ DW.LNCT.path => e.path = try form_value.getString(di.*),
+ DW.LNCT.directory_index => e.dir_index = try form_value.getUInt(u32),
+ DW.LNCT.timestamp => e.mtime = try form_value.getUInt(u64),
+ DW.LNCT.size => e.size = try form_value.getUInt(u64),
+ DW.LNCT.MD5 => e.md5 = switch (form_value) {
+ .data16 => |data16| data16.*,
+ else => return badDwarf(),
+ },
+ else => continue,
+ }
+ }
+ file_entries.appendAssumeCapacity(e);
+ }
+ }
+ }
+
+ var prog = LineNumberProgram.init(
+ default_is_stmt,
+ include_directories.items,
+ target_address,
+ version,
+ );
+
+ try fbr.seekTo(prog_start_offset);
+
+ const next_unit_pos = line_info_offset + next_offset;
+
+ while (fbr.pos < next_unit_pos) {
+ const opcode = try fbr.readByte();
+
+ if (opcode == DW.LNS.extended_op) {
+ const op_size = try fbr.readUleb128(u64);
+ if (op_size < 1) return badDwarf();
+ const sub_op = try fbr.readByte();
+ switch (sub_op) {
+ DW.LNE.end_sequence => {
+ prog.end_sequence = true;
+ if (try prog.checkLineMatch(allocator, file_entries.items)) |info| return info;
+ prog.reset();
+ },
+ DW.LNE.set_address => {
+ const addr = try fbr.readInt(usize);
+ prog.address = addr;
+ },
+ DW.LNE.define_file => {
+ const path = try fbr.readBytesTo(0);
+ const dir_index = try fbr.readUleb128(u32);
+ const mtime = try fbr.readUleb128(u64);
+ const size = try fbr.readUleb128(u64);
+ try file_entries.append(.{
+ .path = path,
+ .dir_index = dir_index,
+ .mtime = mtime,
+ .size = size,
+ });
+ },
+ else => try fbr.seekForward(op_size - 1),
+ }
+ } else if (opcode >= opcode_base) {
+ // special opcodes
+ const adjusted_opcode = opcode - opcode_base;
+ const inc_addr = minimum_instruction_length * (adjusted_opcode / line_range);
+ const inc_line = @as(i32, line_base) + @as(i32, adjusted_opcode % line_range);
+ prog.line += inc_line;
+ prog.address += inc_addr;
+ if (try prog.checkLineMatch(allocator, file_entries.items)) |info| return info;
+ prog.basic_block = false;
+ } else {
+ switch (opcode) {
+ DW.LNS.copy => {
+ if (try prog.checkLineMatch(allocator, file_entries.items)) |info| return info;
+ prog.basic_block = false;
+ },
+ DW.LNS.advance_pc => {
+ const arg = try fbr.readUleb128(usize);
+ prog.address += arg * minimum_instruction_length;
+ },
+ DW.LNS.advance_line => {
+ const arg = try fbr.readIleb128(i64);
+ prog.line += arg;
+ },
+ DW.LNS.set_file => {
+ const arg = try fbr.readUleb128(usize);
+ prog.file = arg;
+ },
+ DW.LNS.set_column => {
+ const arg = try fbr.readUleb128(u64);
+ prog.column = arg;
+ },
+ DW.LNS.negate_stmt => {
+ prog.is_stmt = !prog.is_stmt;
+ },
+ DW.LNS.set_basic_block => {
+ prog.basic_block = true;
+ },
+ DW.LNS.const_add_pc => {
+ const inc_addr = minimum_instruction_length * ((255 - opcode_base) / line_range);
+ prog.address += inc_addr;
+ },
+ DW.LNS.fixed_advance_pc => {
+ const arg = try fbr.readInt(u16);
+ prog.address += arg;
+ },
+ DW.LNS.set_prologue_end => {},
+ else => {
+ if (opcode - 1 >= standard_opcode_lengths.len) return badDwarf();
+ try fbr.seekForward(standard_opcode_lengths[opcode - 1]);
+ },
+ }
+ }
+ }
+
+ return missingDwarf();
+}
+
+fn getString(di: Dwarf, offset: u64) ![:0]const u8 {
+ return getStringGeneric(di.section(.debug_str), offset);
+}
+
+fn getLineString(di: Dwarf, offset: u64) ![:0]const u8 {
+ return getStringGeneric(di.section(.debug_line_str), offset);
+}
+
+fn readDebugAddr(di: Dwarf, compile_unit: CompileUnit, index: u64) !u64 {
+ const debug_addr = di.section(.debug_addr) orelse return badDwarf();
+
+ // addr_base points to the first item after the header, however we
+ // need to read the header to know the size of each item. Empirically,
+ // it may disagree with is_64 on the compile unit.
+ // The header is 8 or 12 bytes depending on is_64.
+ if (compile_unit.addr_base < 8) return badDwarf();
+
+ const version = readInt(u16, debug_addr[compile_unit.addr_base - 4 ..][0..2], di.endian);
+ if (version != 5) return badDwarf();
+
+ const addr_size = debug_addr[compile_unit.addr_base - 2];
+ const seg_size = debug_addr[compile_unit.addr_base - 1];
+
+ const byte_offset = @as(usize, @intCast(compile_unit.addr_base + (addr_size + seg_size) * index));
+ if (byte_offset + addr_size > debug_addr.len) return badDwarf();
+ return switch (addr_size) {
+ 1 => debug_addr[byte_offset],
+ 2 => readInt(u16, debug_addr[byte_offset..][0..2], di.endian),
+ 4 => readInt(u32, debug_addr[byte_offset..][0..4], di.endian),
+ 8 => readInt(u64, debug_addr[byte_offset..][0..8], di.endian),
+ else => badDwarf(),
+ };
+}
+
+/// If .eh_frame_hdr is present, then only the header needs to be parsed.
+///
+/// Otherwise, .eh_frame and .debug_frame are scanned and a sorted list
+/// of FDEs is built for binary searching during unwinding.
+pub fn scanAllUnwindInfo(di: *Dwarf, allocator: Allocator, base_address: usize) !void {
+ if (di.section(.eh_frame_hdr)) |eh_frame_hdr| blk: {
+ var fbr: FixedBufferReader = .{ .buf = eh_frame_hdr, .endian = native_endian };
+
+ const version = try fbr.readByte();
+ if (version != 1) break :blk;
+
+ const eh_frame_ptr_enc = try fbr.readByte();
+ if (eh_frame_ptr_enc == EH.PE.omit) break :blk;
+ const fde_count_enc = try fbr.readByte();
+ if (fde_count_enc == EH.PE.omit) break :blk;
+ const table_enc = try fbr.readByte();
+ if (table_enc == EH.PE.omit) break :blk;
+
+ const eh_frame_ptr = cast(usize, try readEhPointer(&fbr, eh_frame_ptr_enc, @sizeOf(usize), .{
+ .pc_rel_base = @intFromPtr(&eh_frame_hdr[fbr.pos]),
+ .follow_indirect = true,
+ }) orelse return badDwarf()) orelse return badDwarf();
+
+ const fde_count = cast(usize, try readEhPointer(&fbr, fde_count_enc, @sizeOf(usize), .{
+ .pc_rel_base = @intFromPtr(&eh_frame_hdr[fbr.pos]),
+ .follow_indirect = true,
+ }) orelse return badDwarf()) orelse return badDwarf();
+
+ const entry_size = try ExceptionFrameHeader.entrySize(table_enc);
+ const entries_len = fde_count * entry_size;
+ if (entries_len > eh_frame_hdr.len - fbr.pos) return badDwarf();
+
+ di.eh_frame_hdr = .{
+ .eh_frame_ptr = eh_frame_ptr,
+ .table_enc = table_enc,
+ .fde_count = fde_count,
+ .entries = eh_frame_hdr[fbr.pos..][0..entries_len],
+ };
+
+ // No need to scan .eh_frame, we have a binary search table already
+ return;
+ }
+
+ const frame_sections = [2]Section.Id{ .eh_frame, .debug_frame };
+ for (frame_sections) |frame_section| {
+ if (di.section(frame_section)) |section_data| {
+ var fbr: FixedBufferReader = .{ .buf = section_data, .endian = di.endian };
+ while (fbr.pos < fbr.buf.len) {
+ const entry_header = try EntryHeader.read(&fbr, null, frame_section);
+ switch (entry_header.type) {
+ .cie => {
+ const cie = try CommonInformationEntry.parse(
+ entry_header.entry_bytes,
+ di.sectionVirtualOffset(frame_section, base_address).?,
+ true,
+ entry_header.format,
+ frame_section,
+ entry_header.length_offset,
+ @sizeOf(usize),
+ di.endian,
+ );
+ try di.cie_map.put(allocator, entry_header.length_offset, cie);
+ },
+ .fde => |cie_offset| {
+ const cie = di.cie_map.get(cie_offset) orelse return badDwarf();
+ const fde = try FrameDescriptionEntry.parse(
+ entry_header.entry_bytes,
+ di.sectionVirtualOffset(frame_section, base_address).?,
+ true,
+ cie,
+ @sizeOf(usize),
+ di.endian,
+ );
+ try di.fde_list.append(allocator, fde);
+ },
+ .terminator => break,
+ }
+ }
+
+ std.mem.sortUnstable(FrameDescriptionEntry, di.fde_list.items, {}, struct {
+ fn lessThan(ctx: void, a: FrameDescriptionEntry, b: FrameDescriptionEntry) bool {
+ _ = ctx;
+ return a.pc_begin < b.pc_begin;
+ }
+ }.lessThan);
+ }
+ }
+}
+
+/// Unwind a stack frame using DWARF unwinding info, updating the register context.
+///
+/// If `.eh_frame_hdr` is available, it will be used to binary search for the FDE.
+/// Otherwise, a linear scan of `.eh_frame` and `.debug_frame` is done to find the FDE.
+///
+/// `explicit_fde_offset` is for cases where the FDE offset is known, such as when __unwind_info
+/// defers unwinding to DWARF. This is an offset into the `.eh_frame` section.
+pub fn unwindFrame(di: *const Dwarf, context: *UnwindContext, ma: *StackIterator.MemoryAccessor, explicit_fde_offset: ?usize) !usize {
+ if (!comptime abi.supportsUnwinding(builtin.target)) return error.UnsupportedCpuArchitecture;
+ if (context.pc == 0) return 0;
+
+ // Find the FDE and CIE
+ var cie: CommonInformationEntry = undefined;
+ var fde: FrameDescriptionEntry = undefined;
+
+ if (explicit_fde_offset) |fde_offset| {
+ const dwarf_section: Section.Id = .eh_frame;
+ const frame_section = di.section(dwarf_section) orelse return error.MissingFDE;
+ if (fde_offset >= frame_section.len) return error.MissingFDE;
+
+ var fbr: FixedBufferReader = .{
+ .buf = frame_section,
+ .pos = fde_offset,
+ .endian = di.endian,
+ };
+
+ const fde_entry_header = try EntryHeader.read(&fbr, null, dwarf_section);
+ if (fde_entry_header.type != .fde) return error.MissingFDE;
+
+ const cie_offset = fde_entry_header.type.fde;
+ try fbr.seekTo(cie_offset);
+
+ fbr.endian = native_endian;
+ const cie_entry_header = try EntryHeader.read(&fbr, null, dwarf_section);
+ if (cie_entry_header.type != .cie) return badDwarf();
+
+ cie = try CommonInformationEntry.parse(
+ cie_entry_header.entry_bytes,
+ 0,
+ true,
+ cie_entry_header.format,
+ dwarf_section,
+ cie_entry_header.length_offset,
+ @sizeOf(usize),
+ native_endian,
+ );
+
+ fde = try FrameDescriptionEntry.parse(
+ fde_entry_header.entry_bytes,
+ 0,
+ true,
+ cie,
+ @sizeOf(usize),
+ native_endian,
+ );
+ } else if (di.eh_frame_hdr) |header| {
+ const eh_frame_len = if (di.section(.eh_frame)) |eh_frame| eh_frame.len else null;
+ try header.findEntry(
+ ma,
+ eh_frame_len,
+ @intFromPtr(di.section(.eh_frame_hdr).?.ptr),
+ context.pc,
+ &cie,
+ &fde,
+ );
+ } else {
+ const index = std.sort.binarySearch(FrameDescriptionEntry, context.pc, di.fde_list.items, {}, struct {
+ pub fn compareFn(_: void, pc: usize, mid_item: FrameDescriptionEntry) std.math.Order {
+ if (pc < mid_item.pc_begin) return .lt;
+
+ const range_end = mid_item.pc_begin + mid_item.pc_range;
+ if (pc < range_end) return .eq;
+
+ return .gt;
+ }
+ }.compareFn);
+
+ fde = if (index) |i| di.fde_list.items[i] else return error.MissingFDE;
+ cie = di.cie_map.get(fde.cie_length_offset) orelse return error.MissingCIE;
+ }
+
+ var expression_context: expression.Context = .{
+ .format = cie.format,
+ .memory_accessor = ma,
+ .compile_unit = di.findCompileUnit(fde.pc_begin) catch null,
+ .thread_context = context.thread_context,
+ .reg_context = context.reg_context,
+ .cfa = context.cfa,
+ };
+
+ context.vm.reset();
+ context.reg_context.eh_frame = cie.version != 4;
+ context.reg_context.is_macho = di.is_macho;
+
+ const row = try context.vm.runToNative(context.allocator, context.pc, cie, fde);
+ context.cfa = switch (row.cfa.rule) {
+ .val_offset => |offset| blk: {
+ const register = row.cfa.register orelse return error.InvalidCFARule;
+ const value = readInt(usize, (try abi.regBytes(context.thread_context, register, context.reg_context))[0..@sizeOf(usize)], native_endian);
+ break :blk try call_frame.applyOffset(value, offset);
+ },
+ .expression => |expr| blk: {
+ context.stack_machine.reset();
+ const value = try context.stack_machine.run(
+ expr,
+ context.allocator,
+ expression_context,
+ context.cfa,
+ );
+
+ if (value) |v| {
+ if (v != .generic) return error.InvalidExpressionValue;
+ break :blk v.generic;
+ } else return error.NoExpressionValue;
+ },
+ else => return error.InvalidCFARule,
+ };
+
+ if (ma.load(usize, context.cfa.?) == null) return error.InvalidCFA;
+ expression_context.cfa = context.cfa;
+
+ // Buffering the modifications is done because copying the thread context is not portable,
+ // some implementations (ie. darwin) use internal pointers to the mcontext.
+ var arena = std.heap.ArenaAllocator.init(context.allocator);
+ defer arena.deinit();
+ const update_allocator = arena.allocator();
+
+ const RegisterUpdate = struct {
+ // Backed by thread_context
+ dest: []u8,
+ // Backed by arena
+ src: []const u8,
+ prev: ?*@This(),
+ };
+
+ var update_tail: ?*RegisterUpdate = null;
+ var has_return_address = true;
+ for (context.vm.rowColumns(row)) |column| {
+ if (column.register) |register| {
+ if (register == cie.return_address_register) {
+ has_return_address = column.rule != .undefined;
+ }
+
+ const dest = try abi.regBytes(context.thread_context, register, context.reg_context);
+ const src = try update_allocator.alloc(u8, dest.len);
+
+ const prev = update_tail;
+ update_tail = try update_allocator.create(RegisterUpdate);
+ update_tail.?.* = .{
+ .dest = dest,
+ .src = src,
+ .prev = prev,
+ };
+
+ try column.resolveValue(
+ context,
+ expression_context,
+ ma,
+ src,
+ );
+ }
+ }
+
+ // On all implemented architectures, the CFA is defined as being the previous frame's SP
+ (try abi.regValueNative(usize, context.thread_context, abi.spRegNum(context.reg_context), context.reg_context)).* = context.cfa.?;
+
+ while (update_tail) |tail| {
+ @memcpy(tail.dest, tail.src);
+ update_tail = tail.prev;
+ }
+
+ if (has_return_address) {
+ context.pc = abi.stripInstructionPtrAuthCode(readInt(usize, (try abi.regBytes(
+ context.thread_context,
+ cie.return_address_register,
+ context.reg_context,
+ ))[0..@sizeOf(usize)], native_endian));
+ } else {
+ context.pc = 0;
+ }
+
+ (try abi.regValueNative(usize, context.thread_context, abi.ipRegNum(), context.reg_context)).* = context.pc;
+
+ // The call instruction will have pushed the address of the instruction that follows the call as the return address.
+ // This next instruction may be past the end of the function if the caller was `noreturn` (ie. the last instruction in
+ // the function was the call). If we were to look up an FDE entry using the return address directly, it could end up
+ // either not finding an FDE at all, or using the next FDE in the program, producing incorrect results. To prevent this,
+ // we subtract one so that the next lookup is guaranteed to land inside the
+ //
+ // The exception to this rule is signal frames, where we return execution would be returned to the instruction
+ // that triggered the handler.
+ const return_address = context.pc;
+ if (context.pc > 0 and !cie.isSignalFrame()) context.pc -= 1;
+
+ return return_address;
+}
+
+fn parseFormValue(
+ fbr: *FixedBufferReader,
+ form_id: u64,
+ format: Format,
+ implicit_const: ?i64,
+) anyerror!FormValue {
+ return switch (form_id) {
+ FORM.addr => .{ .addr = try fbr.readAddress(switch (@bitSizeOf(usize)) {
+ 32 => .@"32",
+ 64 => .@"64",
+ else => @compileError("unsupported @sizeOf(usize)"),
+ }) },
+ FORM.addrx1 => .{ .addrx = try fbr.readInt(u8) },
+ FORM.addrx2 => .{ .addrx = try fbr.readInt(u16) },
+ FORM.addrx3 => .{ .addrx = try fbr.readInt(u24) },
+ FORM.addrx4 => .{ .addrx = try fbr.readInt(u32) },
+ FORM.addrx => .{ .addrx = try fbr.readUleb128(usize) },
+
+ FORM.block1,
+ FORM.block2,
+ FORM.block4,
+ FORM.block,
+ => .{ .block = try fbr.readBytes(switch (form_id) {
+ FORM.block1 => try fbr.readInt(u8),
+ FORM.block2 => try fbr.readInt(u16),
+ FORM.block4 => try fbr.readInt(u32),
+ FORM.block => try fbr.readUleb128(usize),
+ else => unreachable,
+ }) },
+
+ FORM.data1 => .{ .udata = try fbr.readInt(u8) },
+ FORM.data2 => .{ .udata = try fbr.readInt(u16) },
+ FORM.data4 => .{ .udata = try fbr.readInt(u32) },
+ FORM.data8 => .{ .udata = try fbr.readInt(u64) },
+ FORM.data16 => .{ .data16 = (try fbr.readBytes(16))[0..16] },
+ FORM.udata => .{ .udata = try fbr.readUleb128(u64) },
+ FORM.sdata => .{ .sdata = try fbr.readIleb128(i64) },
+ FORM.exprloc => .{ .exprloc = try fbr.readBytes(try fbr.readUleb128(usize)) },
+ FORM.flag => .{ .flag = (try fbr.readByte()) != 0 },
+ FORM.flag_present => .{ .flag = true },
+ FORM.sec_offset => .{ .sec_offset = try fbr.readAddress(format) },
+
+ FORM.ref1 => .{ .ref = try fbr.readInt(u8) },
+ FORM.ref2 => .{ .ref = try fbr.readInt(u16) },
+ FORM.ref4 => .{ .ref = try fbr.readInt(u32) },
+ FORM.ref8 => .{ .ref = try fbr.readInt(u64) },
+ FORM.ref_udata => .{ .ref = try fbr.readUleb128(u64) },
+
+ FORM.ref_addr => .{ .ref_addr = try fbr.readAddress(format) },
+ FORM.ref_sig8 => .{ .ref = try fbr.readInt(u64) },
+
+ FORM.string => .{ .string = try fbr.readBytesTo(0) },
+ FORM.strp => .{ .strp = try fbr.readAddress(format) },
+ FORM.strx1 => .{ .strx = try fbr.readInt(u8) },
+ FORM.strx2 => .{ .strx = try fbr.readInt(u16) },
+ FORM.strx3 => .{ .strx = try fbr.readInt(u24) },
+ FORM.strx4 => .{ .strx = try fbr.readInt(u32) },
+ FORM.strx => .{ .strx = try fbr.readUleb128(usize) },
+ FORM.line_strp => .{ .line_strp = try fbr.readAddress(format) },
+ FORM.indirect => parseFormValue(fbr, try fbr.readUleb128(u64), format, implicit_const),
+ FORM.implicit_const => .{ .sdata = implicit_const orelse return badDwarf() },
+ FORM.loclistx => .{ .loclistx = try fbr.readUleb128(u64) },
+ FORM.rnglistx => .{ .rnglistx = try fbr.readUleb128(u64) },
+ else => {
+ //debug.print("unrecognized form id: {x}\n", .{form_id});
+ return badDwarf();
+ },
+ };
+}
+
+const FileEntry = struct {
+ path: []const u8,
+ dir_index: u32 = 0,
+ mtime: u64 = 0,
+ size: u64 = 0,
+ md5: [16]u8 = [1]u8{0} ** 16,
+};
+
+const LineNumberProgram = struct {
+ address: u64,
+ file: usize,
+ line: i64,
+ column: u64,
+ version: u16,
+ is_stmt: bool,
+ basic_block: bool,
+ end_sequence: bool,
+
+ default_is_stmt: bool,
+ target_address: u64,
+ include_dirs: []const FileEntry,
+
+ prev_valid: bool,
+ prev_address: u64,
+ prev_file: usize,
+ prev_line: i64,
+ prev_column: u64,
+ prev_is_stmt: bool,
+ prev_basic_block: bool,
+ prev_end_sequence: bool,
+
+ // Reset the state machine following the DWARF specification
+ pub fn reset(self: *LineNumberProgram) void {
+ self.address = 0;
+ self.file = 1;
+ self.line = 1;
+ self.column = 0;
+ self.is_stmt = self.default_is_stmt;
+ self.basic_block = false;
+ self.end_sequence = false;
+ // Invalidate all the remaining fields
+ self.prev_valid = false;
+ self.prev_address = 0;
+ self.prev_file = undefined;
+ self.prev_line = undefined;
+ self.prev_column = undefined;
+ self.prev_is_stmt = undefined;
+ self.prev_basic_block = undefined;
+ self.prev_end_sequence = undefined;
+ }
+
+ pub fn init(
+ is_stmt: bool,
+ include_dirs: []const FileEntry,
+ target_address: u64,
+ version: u16,
+ ) LineNumberProgram {
+ return LineNumberProgram{
+ .address = 0,
+ .file = 1,
+ .line = 1,
+ .column = 0,
+ .version = version,
+ .is_stmt = is_stmt,
+ .basic_block = false,
+ .end_sequence = false,
+ .include_dirs = include_dirs,
+ .default_is_stmt = is_stmt,
+ .target_address = target_address,
+ .prev_valid = false,
+ .prev_address = 0,
+ .prev_file = undefined,
+ .prev_line = undefined,
+ .prev_column = undefined,
+ .prev_is_stmt = undefined,
+ .prev_basic_block = undefined,
+ .prev_end_sequence = undefined,
+ };
+ }
+
+ pub fn checkLineMatch(
+ self: *LineNumberProgram,
+ allocator: Allocator,
+ file_entries: []const FileEntry,
+ ) !?std.debug.LineInfo {
+ if (self.prev_valid and
+ self.target_address >= self.prev_address and
+ self.target_address < self.address)
+ {
+ const file_index = if (self.version >= 5) self.prev_file else i: {
+ if (self.prev_file == 0) return missingDwarf();
+ break :i self.prev_file - 1;
+ };
+
+ if (file_index >= file_entries.len) return badDwarf();
+ const file_entry = &file_entries[file_index];
+
+ if (file_entry.dir_index >= self.include_dirs.len) return badDwarf();
+ const dir_name = self.include_dirs[file_entry.dir_index].path;
+
+ const file_name = try std.fs.path.join(allocator, &[_][]const u8{
+ dir_name, file_entry.path,
+ });
+
+ return std.debug.LineInfo{
+ .line = if (self.prev_line >= 0) @as(u64, @intCast(self.prev_line)) else 0,
+ .column = self.prev_column,
+ .file_name = file_name,
+ };
+ }
+
+ self.prev_valid = true;
+ self.prev_address = self.address;
+ self.prev_file = self.file;
+ self.prev_line = self.line;
+ self.prev_column = self.column;
+ self.prev_is_stmt = self.is_stmt;
+ self.prev_basic_block = self.basic_block;
+ self.prev_end_sequence = self.end_sequence;
+ return null;
+ }
+};
+
+const UnitHeader = struct {
+ format: Format,
+ header_length: u4,
+ unit_length: u64,
+};
+fn readUnitHeader(fbr: *FixedBufferReader, opt_ma: ?*StackIterator.MemoryAccessor) !UnitHeader {
+ return switch (try if (opt_ma) |ma| fbr.readIntChecked(u32, ma) else fbr.readInt(u32)) {
+ 0...0xfffffff0 - 1 => |unit_length| .{
+ .format = .@"32",
+ .header_length = 4,
+ .unit_length = unit_length,
+ },
+ 0xfffffff0...0xffffffff - 1 => badDwarf(),
+ 0xffffffff => .{
+ .format = .@"64",
+ .header_length = 12,
+ .unit_length = try if (opt_ma) |ma| fbr.readIntChecked(u64, ma) else fbr.readInt(u64),
+ },
+ };
+}
+
+/// Returns the DWARF register number for an x86_64 register number found in compact unwind info
+fn compactUnwindToDwarfRegNumber(unwind_reg_number: u3) !u8 {
+ return switch (unwind_reg_number) {
+ 1 => 3, // RBX
+ 2 => 12, // R12
+ 3 => 13, // R13
+ 4 => 14, // R14
+ 5 => 15, // R15
+ 6 => 6, // RBP
+ else => error.InvalidUnwindRegisterNumber,
+ };
+}
+
+/// This function is to make it handy to comment out the return and make it
+/// into a crash when working on this file.
+fn badDwarf() error{InvalidDebugInfo} {
+ //if (true) @panic("badDwarf"); // can be handy to uncomment when working on this file
+ return error.InvalidDebugInfo;
+}
+
+fn missingDwarf() error{MissingDebugInfo} {
+ //if (true) @panic("missingDwarf"); // can be handy to uncomment when working on this file
+ return error.MissingDebugInfo;
+}
+
+fn getStringGeneric(opt_str: ?[]const u8, offset: u64) ![:0]const u8 {
+ const str = opt_str orelse return badDwarf();
+ if (offset > str.len) return badDwarf();
+ const casted_offset = cast(usize, offset) orelse return badDwarf();
+ // Valid strings always have a terminating zero byte
+ const last = std.mem.indexOfScalarPos(u8, str, casted_offset, 0) orelse return badDwarf();
+ return str[casted_offset..last :0];
+}
+
+// Reading debug info needs to be fast, even when compiled in debug mode,
+// so avoid using a `std.io.FixedBufferStream` which is too slow.
+pub const FixedBufferReader = struct {
+ buf: []const u8,
+ pos: usize = 0,
+ endian: std.builtin.Endian,
+
+ pub const Error = error{ EndOfBuffer, Overflow, InvalidBuffer };
+
+ fn seekTo(fbr: *FixedBufferReader, pos: u64) Error!void {
+ if (pos > fbr.buf.len) return error.EndOfBuffer;
+ fbr.pos = @intCast(pos);
+ }
+
+ fn seekForward(fbr: *FixedBufferReader, amount: u64) Error!void {
+ if (fbr.buf.len - fbr.pos < amount) return error.EndOfBuffer;
+ fbr.pos += @intCast(amount);
+ }
+
+ pub inline fn readByte(fbr: *FixedBufferReader) Error!u8 {
+ if (fbr.pos >= fbr.buf.len) return error.EndOfBuffer;
+ defer fbr.pos += 1;
+ return fbr.buf[fbr.pos];
+ }
+
+ fn readByteSigned(fbr: *FixedBufferReader) Error!i8 {
+ return @bitCast(try fbr.readByte());
+ }
+
+ fn readInt(fbr: *FixedBufferReader, comptime T: type) Error!T {
+ const size = @divExact(@typeInfo(T).Int.bits, 8);
+ if (fbr.buf.len - fbr.pos < size) return error.EndOfBuffer;
+ defer fbr.pos += size;
+ return std.mem.readInt(T, fbr.buf[fbr.pos..][0..size], fbr.endian);
+ }
+
+ fn readIntChecked(
+ fbr: *FixedBufferReader,
+ comptime T: type,
+ ma: *std.debug.StackIterator.MemoryAccessor,
+ ) Error!T {
+ if (ma.load(T, @intFromPtr(fbr.buf[fbr.pos..].ptr)) == null)
+ return error.InvalidBuffer;
+
+ return fbr.readInt(T);
+ }
+
+ fn readUleb128(fbr: *FixedBufferReader, comptime T: type) Error!T {
+ return std.leb.readUleb128(T, fbr);
+ }
+
+ fn readIleb128(fbr: *FixedBufferReader, comptime T: type) Error!T {
+ return std.leb.readIleb128(T, fbr);
+ }
+
+ fn readAddress(fbr: *FixedBufferReader, format: Format) Error!u64 {
+ return switch (format) {
+ .@"32" => try fbr.readInt(u32),
+ .@"64" => try fbr.readInt(u64),
+ };
+ }
+
+ fn readAddressChecked(
+ fbr: *FixedBufferReader,
+ format: Format,
+ ma: *std.debug.StackIterator.MemoryAccessor,
+ ) Error!u64 {
+ return switch (format) {
+ .@"32" => try fbr.readIntChecked(u32, ma),
+ .@"64" => try fbr.readIntChecked(u64, ma),
+ };
+ }
+
+ fn readBytes(fbr: *FixedBufferReader, len: usize) Error![]const u8 {
+ if (fbr.buf.len - fbr.pos < len) return error.EndOfBuffer;
+ defer fbr.pos += len;
+ return fbr.buf[fbr.pos..][0..len];
+ }
+
+ fn readBytesTo(fbr: *FixedBufferReader, comptime sentinel: u8) Error![:sentinel]const u8 {
+ const end = @call(.always_inline, std.mem.indexOfScalarPos, .{
+ u8,
+ fbr.buf,
+ fbr.pos,
+ sentinel,
+ }) orelse return error.EndOfBuffer;
+ defer fbr.pos = end + 1;
+ return fbr.buf[fbr.pos..end :sentinel];
+ }
+};
+
+/// Unwind a frame using MachO compact unwind info (from __unwind_info).
+/// If the compact encoding can't encode a way to unwind a frame, it will
+/// defer unwinding to DWARF, in which case `.eh_frame` will be used if available.
+pub fn unwindFrameMachO(
+ context: *UnwindContext,
+ ma: *StackIterator.MemoryAccessor,
+ unwind_info: []const u8,
+ eh_frame: ?[]const u8,
+ module_base_address: usize,
+) !usize {
+ const macho = std.macho;
+
+ const header = std.mem.bytesAsValue(
+ macho.unwind_info_section_header,
+ unwind_info[0..@sizeOf(macho.unwind_info_section_header)],
+ );
+ const indices = std.mem.bytesAsSlice(
+ macho.unwind_info_section_header_index_entry,
+ unwind_info[header.indexSectionOffset..][0 .. header.indexCount * @sizeOf(macho.unwind_info_section_header_index_entry)],
+ );
+ if (indices.len == 0) return error.MissingUnwindInfo;
+
+ const mapped_pc = context.pc - module_base_address;
+ const second_level_index = blk: {
+ var left: usize = 0;
+ var len: usize = indices.len;
+
+ while (len > 1) {
+ const mid = left + len / 2;
+ const offset = indices[mid].functionOffset;
+ if (mapped_pc < offset) {
+ len /= 2;
+ } else {
+ left = mid;
+ if (mapped_pc == offset) break;
+ len -= len / 2;
+ }
+ }
+
+ // Last index is a sentinel containing the highest address as its functionOffset
+ if (indices[left].secondLevelPagesSectionOffset == 0) return error.MissingUnwindInfo;
+ break :blk &indices[left];
+ };
+
+ const common_encodings = std.mem.bytesAsSlice(
+ macho.compact_unwind_encoding_t,
+ unwind_info[header.commonEncodingsArraySectionOffset..][0 .. header.commonEncodingsArrayCount * @sizeOf(macho.compact_unwind_encoding_t)],
+ );
+
+ const start_offset = second_level_index.secondLevelPagesSectionOffset;
+ const kind = std.mem.bytesAsValue(
+ macho.UNWIND_SECOND_LEVEL,
+ unwind_info[start_offset..][0..@sizeOf(macho.UNWIND_SECOND_LEVEL)],
+ );
+
+ const entry: struct {
+ function_offset: usize,
+ raw_encoding: u32,
+ } = switch (kind.*) {
+ .REGULAR => blk: {
+ const page_header = std.mem.bytesAsValue(
+ macho.unwind_info_regular_second_level_page_header,
+ unwind_info[start_offset..][0..@sizeOf(macho.unwind_info_regular_second_level_page_header)],
+ );
+
+ const entries = std.mem.bytesAsSlice(
+ macho.unwind_info_regular_second_level_entry,
+ unwind_info[start_offset + page_header.entryPageOffset ..][0 .. page_header.entryCount * @sizeOf(macho.unwind_info_regular_second_level_entry)],
+ );
+ if (entries.len == 0) return error.InvalidUnwindInfo;
+
+ var left: usize = 0;
+ var len: usize = entries.len;
+ while (len > 1) {
+ const mid = left + len / 2;
+ const offset = entries[mid].functionOffset;
+ if (mapped_pc < offset) {
+ len /= 2;
+ } else {
+ left = mid;
+ if (mapped_pc == offset) break;
+ len -= len / 2;
+ }
+ }
+
+ break :blk .{
+ .function_offset = entries[left].functionOffset,
+ .raw_encoding = entries[left].encoding,
+ };
+ },
+ .COMPRESSED => blk: {
+ const page_header = std.mem.bytesAsValue(
+ macho.unwind_info_compressed_second_level_page_header,
+ unwind_info[start_offset..][0..@sizeOf(macho.unwind_info_compressed_second_level_page_header)],
+ );
+
+ const entries = std.mem.bytesAsSlice(
+ macho.UnwindInfoCompressedEntry,
+ unwind_info[start_offset + page_header.entryPageOffset ..][0 .. page_header.entryCount * @sizeOf(macho.UnwindInfoCompressedEntry)],
+ );
+ if (entries.len == 0) return error.InvalidUnwindInfo;
+
+ var left: usize = 0;
+ var len: usize = entries.len;
+ while (len > 1) {
+ const mid = left + len / 2;
+ const offset = second_level_index.functionOffset + entries[mid].funcOffset;
+ if (mapped_pc < offset) {
+ len /= 2;
+ } else {
+ left = mid;
+ if (mapped_pc == offset) break;
+ len -= len / 2;
+ }
+ }
+
+ const entry = entries[left];
+ const function_offset = second_level_index.functionOffset + entry.funcOffset;
+ if (entry.encodingIndex < header.commonEncodingsArrayCount) {
+ if (entry.encodingIndex >= common_encodings.len) return error.InvalidUnwindInfo;
+ break :blk .{
+ .function_offset = function_offset,
+ .raw_encoding = common_encodings[entry.encodingIndex],
+ };
+ } else {
+ const local_index = try std.math.sub(
+ u8,
+ entry.encodingIndex,
+ cast(u8, header.commonEncodingsArrayCount) orelse return error.InvalidUnwindInfo,
+ );
+ const local_encodings = std.mem.bytesAsSlice(
+ macho.compact_unwind_encoding_t,
+ unwind_info[start_offset + page_header.encodingsPageOffset ..][0 .. page_header.encodingsCount * @sizeOf(macho.compact_unwind_encoding_t)],
+ );
+ if (local_index >= local_encodings.len) return error.InvalidUnwindInfo;
+ break :blk .{
+ .function_offset = function_offset,
+ .raw_encoding = local_encodings[local_index],
+ };
+ }
+ },
+ else => return error.InvalidUnwindInfo,
+ };
+
+ if (entry.raw_encoding == 0) return error.NoUnwindInfo;
+ const reg_context = abi.RegisterContext{
+ .eh_frame = false,
+ .is_macho = true,
+ };
+
+ const encoding: macho.CompactUnwindEncoding = @bitCast(entry.raw_encoding);
+ const new_ip = switch (builtin.cpu.arch) {
+ .x86_64 => switch (encoding.mode.x86_64) {
+ .OLD => return error.UnimplementedUnwindEncoding,
+ .RBP_FRAME => blk: {
+ const regs: [5]u3 = .{
+ encoding.value.x86_64.frame.reg0,
+ encoding.value.x86_64.frame.reg1,
+ encoding.value.x86_64.frame.reg2,
+ encoding.value.x86_64.frame.reg3,
+ encoding.value.x86_64.frame.reg4,
+ };
+
+ const frame_offset = encoding.value.x86_64.frame.frame_offset * @sizeOf(usize);
+ var max_reg: usize = 0;
+ inline for (regs, 0..) |reg, i| {
+ if (reg > 0) max_reg = i;
+ }
+
+ const fp = (try abi.regValueNative(usize, context.thread_context, abi.fpRegNum(reg_context), reg_context)).*;
+ const new_sp = fp + 2 * @sizeOf(usize);
+
+ // Verify the stack range we're about to read register values from
+ if (ma.load(usize, new_sp) == null or ma.load(usize, fp - frame_offset + max_reg * @sizeOf(usize)) == null) return error.InvalidUnwindInfo;
+
+ const ip_ptr = fp + @sizeOf(usize);
+ const new_ip = @as(*const usize, @ptrFromInt(ip_ptr)).*;
+ const new_fp = @as(*const usize, @ptrFromInt(fp)).*;
+
+ (try abi.regValueNative(usize, context.thread_context, abi.fpRegNum(reg_context), reg_context)).* = new_fp;
+ (try abi.regValueNative(usize, context.thread_context, abi.spRegNum(reg_context), reg_context)).* = new_sp;
+ (try abi.regValueNative(usize, context.thread_context, abi.ipRegNum(), reg_context)).* = new_ip;
+
+ for (regs, 0..) |reg, i| {
+ if (reg == 0) continue;
+ const addr = fp - frame_offset + i * @sizeOf(usize);
+ const reg_number = try compactUnwindToDwarfRegNumber(reg);
+ (try abi.regValueNative(usize, context.thread_context, reg_number, reg_context)).* = @as(*const usize, @ptrFromInt(addr)).*;
+ }
+
+ break :blk new_ip;
+ },
+ .STACK_IMMD,
+ .STACK_IND,
+ => blk: {
+ const sp = (try abi.regValueNative(usize, context.thread_context, abi.spRegNum(reg_context), reg_context)).*;
+ const stack_size = if (encoding.mode.x86_64 == .STACK_IMMD)
+ @as(usize, encoding.value.x86_64.frameless.stack.direct.stack_size) * @sizeOf(usize)
+ else stack_size: {
+ // In .STACK_IND, the stack size is inferred from the subq instruction at the beginning of the function.
+ const sub_offset_addr =
+ module_base_address +
+ entry.function_offset +
+ encoding.value.x86_64.frameless.stack.indirect.sub_offset;
+ if (ma.load(usize, sub_offset_addr) == null) return error.InvalidUnwindInfo;
+
+ // `sub_offset_addr` points to the offset of the literal within the instruction
+ const sub_operand = @as(*align(1) const u32, @ptrFromInt(sub_offset_addr)).*;
+ break :stack_size sub_operand + @sizeOf(usize) * @as(usize, encoding.value.x86_64.frameless.stack.indirect.stack_adjust);
+ };
+
+ // Decode the Lehmer-coded sequence of registers.
+ // For a description of the encoding see lib/libc/include/any-macos.13-any/mach-o/compact_unwind_encoding.h
+
+ // Decode the variable-based permutation number into its digits. Each digit represents
+ // an index into the list of register numbers that weren't yet used in the sequence at
+ // the time the digit was added.
+ const reg_count = encoding.value.x86_64.frameless.stack_reg_count;
+ const ip_ptr = if (reg_count > 0) reg_blk: {
+ var digits: [6]u3 = undefined;
+ var accumulator: usize = encoding.value.x86_64.frameless.stack_reg_permutation;
+ var base: usize = 2;
+ for (0..reg_count) |i| {
+ const div = accumulator / base;
+ digits[digits.len - 1 - i] = @intCast(accumulator - base * div);
+ accumulator = div;
+ base += 1;
+ }
+
+ const reg_numbers = [_]u3{ 1, 2, 3, 4, 5, 6 };
+ var registers: [reg_numbers.len]u3 = undefined;
+ var used_indices = [_]bool{false} ** reg_numbers.len;
+ for (digits[digits.len - reg_count ..], 0..) |target_unused_index, i| {
+ var unused_count: u8 = 0;
+ const unused_index = for (used_indices, 0..) |used, index| {
+ if (!used) {
+ if (target_unused_index == unused_count) break index;
+ unused_count += 1;
+ }
+ } else unreachable;
+
+ registers[i] = reg_numbers[unused_index];
+ used_indices[unused_index] = true;
+ }
+
+ var reg_addr = sp + stack_size - @sizeOf(usize) * @as(usize, reg_count + 1);
+ if (ma.load(usize, reg_addr) == null) return error.InvalidUnwindInfo;
+ for (0..reg_count) |i| {
+ const reg_number = try compactUnwindToDwarfRegNumber(registers[i]);
+ (try abi.regValueNative(usize, context.thread_context, reg_number, reg_context)).* = @as(*const usize, @ptrFromInt(reg_addr)).*;
+ reg_addr += @sizeOf(usize);
+ }
+
+ break :reg_blk reg_addr;
+ } else sp + stack_size - @sizeOf(usize);
+
+ const new_ip = @as(*const usize, @ptrFromInt(ip_ptr)).*;
+ const new_sp = ip_ptr + @sizeOf(usize);
+ if (ma.load(usize, new_sp) == null) return error.InvalidUnwindInfo;
+
+ (try abi.regValueNative(usize, context.thread_context, abi.spRegNum(reg_context), reg_context)).* = new_sp;
+ (try abi.regValueNative(usize, context.thread_context, abi.ipRegNum(), reg_context)).* = new_ip;
+
+ break :blk new_ip;
+ },
+ .DWARF => {
+ return unwindFrameMachODwarf(context, ma, eh_frame orelse return error.MissingEhFrame, @intCast(encoding.value.x86_64.dwarf));
+ },
+ },
+ .aarch64 => switch (encoding.mode.arm64) {
+ .OLD => return error.UnimplementedUnwindEncoding,
+ .FRAMELESS => blk: {
+ const sp = (try abi.regValueNative(usize, context.thread_context, abi.spRegNum(reg_context), reg_context)).*;
+ const new_sp = sp + encoding.value.arm64.frameless.stack_size * 16;
+ const new_ip = (try abi.regValueNative(usize, context.thread_context, 30, reg_context)).*;
+ if (ma.load(usize, new_sp) == null) return error.InvalidUnwindInfo;
+ (try abi.regValueNative(usize, context.thread_context, abi.spRegNum(reg_context), reg_context)).* = new_sp;
+ break :blk new_ip;
+ },
+ .DWARF => {
+ return unwindFrameMachODwarf(context, ma, eh_frame orelse return error.MissingEhFrame, @intCast(encoding.value.arm64.dwarf));
+ },
+ .FRAME => blk: {
+ const fp = (try abi.regValueNative(usize, context.thread_context, abi.fpRegNum(reg_context), reg_context)).*;
+ const new_sp = fp + 16;
+ const ip_ptr = fp + @sizeOf(usize);
+
+ const num_restored_pairs: usize =
+ @popCount(@as(u5, @bitCast(encoding.value.arm64.frame.x_reg_pairs))) +
+ @popCount(@as(u4, @bitCast(encoding.value.arm64.frame.d_reg_pairs)));
+ const min_reg_addr = fp - num_restored_pairs * 2 * @sizeOf(usize);
+
+ if (ma.load(usize, new_sp) == null or ma.load(usize, min_reg_addr) == null) return error.InvalidUnwindInfo;
+
+ var reg_addr = fp - @sizeOf(usize);
+ inline for (@typeInfo(@TypeOf(encoding.value.arm64.frame.x_reg_pairs)).Struct.fields, 0..) |field, i| {
+ if (@field(encoding.value.arm64.frame.x_reg_pairs, field.name) != 0) {
+ (try abi.regValueNative(usize, context.thread_context, 19 + i, reg_context)).* = @as(*const usize, @ptrFromInt(reg_addr)).*;
+ reg_addr += @sizeOf(usize);
+ (try abi.regValueNative(usize, context.thread_context, 20 + i, reg_context)).* = @as(*const usize, @ptrFromInt(reg_addr)).*;
+ reg_addr += @sizeOf(usize);
+ }
+ }
+
+ inline for (@typeInfo(@TypeOf(encoding.value.arm64.frame.d_reg_pairs)).Struct.fields, 0..) |field, i| {
+ if (@field(encoding.value.arm64.frame.d_reg_pairs, field.name) != 0) {
+ // Only the lower half of the 128-bit V registers are restored during unwinding
+ @memcpy(
+ try abi.regBytes(context.thread_context, 64 + 8 + i, context.reg_context),
+ std.mem.asBytes(@as(*const usize, @ptrFromInt(reg_addr))),
+ );
+ reg_addr += @sizeOf(usize);
+ @memcpy(
+ try abi.regBytes(context.thread_context, 64 + 9 + i, context.reg_context),
+ std.mem.asBytes(@as(*const usize, @ptrFromInt(reg_addr))),
+ );
+ reg_addr += @sizeOf(usize);
+ }
+ }
+
+ const new_ip = @as(*const usize, @ptrFromInt(ip_ptr)).*;
+ const new_fp = @as(*const usize, @ptrFromInt(fp)).*;
+
+ (try abi.regValueNative(usize, context.thread_context, abi.fpRegNum(reg_context), reg_context)).* = new_fp;
+ (try abi.regValueNative(usize, context.thread_context, abi.ipRegNum(), reg_context)).* = new_ip;
+
+ break :blk new_ip;
+ },
+ },
+ else => return error.UnimplementedArch,
+ };
+
+ context.pc = abi.stripInstructionPtrAuthCode(new_ip);
+ if (context.pc > 0) context.pc -= 1;
+ return new_ip;
+}
+
+fn unwindFrameMachODwarf(
+ context: *UnwindContext,
+ ma: *std.debug.StackIterator.MemoryAccessor,
+ eh_frame: []const u8,
+ fde_offset: usize,
+) !usize {
+ var di = Dwarf{
+ .endian = native_endian,
+ .is_macho = true,
+ };
+ defer di.deinit(context.allocator);
+
+ di.sections[@intFromEnum(Section.Id.eh_frame)] = .{
+ .data = eh_frame,
+ .owned = false,
+ };
+
+ return di.unwindFrame(context, ma, fde_offset);
+}
+
+const EhPointerContext = struct {
+ // The address of the pointer field itself
+ pc_rel_base: u64,
+
+ // Whether or not to follow indirect pointers. This should only be
+ // used when decoding pointers at runtime using the current process's
+ // debug info
+ follow_indirect: bool,
+
+ // These relative addressing modes are only used in specific cases, and
+ // might not be available / required in all parsing contexts
+ data_rel_base: ?u64 = null,
+ text_rel_base: ?u64 = null,
+ function_rel_base: ?u64 = null,
+};
+fn readEhPointer(fbr: *FixedBufferReader, enc: u8, addr_size_bytes: u8, ctx: EhPointerContext) !?u64 {
+ if (enc == EH.PE.omit) return null;
+
+ const value: union(enum) {
+ signed: i64,
+ unsigned: u64,
+ } = switch (enc & EH.PE.type_mask) {
+ EH.PE.absptr => .{
+ .unsigned = switch (addr_size_bytes) {
+ 2 => try fbr.readInt(u16),
+ 4 => try fbr.readInt(u32),
+ 8 => try fbr.readInt(u64),
+ else => return error.InvalidAddrSize,
+ },
+ },
+ EH.PE.uleb128 => .{ .unsigned = try fbr.readUleb128(u64) },
+ EH.PE.udata2 => .{ .unsigned = try fbr.readInt(u16) },
+ EH.PE.udata4 => .{ .unsigned = try fbr.readInt(u32) },
+ EH.PE.udata8 => .{ .unsigned = try fbr.readInt(u64) },
+ EH.PE.sleb128 => .{ .signed = try fbr.readIleb128(i64) },
+ EH.PE.sdata2 => .{ .signed = try fbr.readInt(i16) },
+ EH.PE.sdata4 => .{ .signed = try fbr.readInt(i32) },
+ EH.PE.sdata8 => .{ .signed = try fbr.readInt(i64) },
+ else => return badDwarf(),
+ };
+
+ const base = switch (enc & EH.PE.rel_mask) {
+ EH.PE.pcrel => ctx.pc_rel_base,
+ EH.PE.textrel => ctx.text_rel_base orelse return error.PointerBaseNotSpecified,
+ EH.PE.datarel => ctx.data_rel_base orelse return error.PointerBaseNotSpecified,
+ EH.PE.funcrel => ctx.function_rel_base orelse return error.PointerBaseNotSpecified,
+ else => null,
+ };
+
+ const ptr: u64 = if (base) |b| switch (value) {
+ .signed => |s| @intCast(try std.math.add(i64, s, @as(i64, @intCast(b)))),
+ // absptr can actually contain signed values in some cases (aarch64 MachO)
+ .unsigned => |u| u +% b,
+ } else switch (value) {
+ .signed => |s| @as(u64, @intCast(s)),
+ .unsigned => |u| u,
+ };
+
+ if ((enc & EH.PE.indirect) > 0 and ctx.follow_indirect) {
+ if (@sizeOf(usize) != addr_size_bytes) {
+ // See the documentation for `follow_indirect`
+ return error.NonNativeIndirection;
+ }
+
+ const native_ptr = cast(usize, ptr) orelse return error.PointerOverflow;
+ return switch (addr_size_bytes) {
+ 2, 4, 8 => return @as(*const usize, @ptrFromInt(native_ptr)).*,
+ else => return error.UnsupportedAddrSize,
+ };
+ } else {
+ return ptr;
+ }
+}
+
+fn pcRelBase(field_ptr: usize, pc_rel_offset: i64) !usize {
+ if (pc_rel_offset < 0) {
+ return std.math.sub(usize, field_ptr, @as(usize, @intCast(-pc_rel_offset)));
+ } else {
+ return std.math.add(usize, field_ptr, @as(usize, @intCast(pc_rel_offset)));
+ }
+}
diff --git a/lib/std/debug/Dwarf/abi.zig b/lib/std/debug/Dwarf/abi.zig
new file mode 100644
index 0000000000..1a47625ae7
--- /dev/null
+++ b/lib/std/debug/Dwarf/abi.zig
@@ -0,0 +1,410 @@
+const builtin = @import("builtin");
+const std = @import("../../std.zig");
+const mem = std.mem;
+const native_os = builtin.os.tag;
+const posix = std.posix;
+
+pub fn supportsUnwinding(target: std.Target) bool {
+ return switch (target.cpu.arch) {
+ .x86 => switch (target.os.tag) {
+ .linux, .netbsd, .solaris, .illumos => true,
+ else => false,
+ },
+ .x86_64 => switch (target.os.tag) {
+ .linux, .netbsd, .freebsd, .openbsd, .macos, .ios, .solaris, .illumos => true,
+ else => false,
+ },
+ .arm => switch (target.os.tag) {
+ .linux => true,
+ else => false,
+ },
+ .aarch64 => switch (target.os.tag) {
+ .linux, .netbsd, .freebsd, .macos, .ios => true,
+ else => false,
+ },
+ else => false,
+ };
+}
+
+pub fn ipRegNum() u8 {
+ return switch (builtin.cpu.arch) {
+ .x86 => 8,
+ .x86_64 => 16,
+ .arm => 15,
+ .aarch64 => 32,
+ else => unreachable,
+ };
+}
+
+pub fn fpRegNum(reg_context: RegisterContext) u8 {
+ return switch (builtin.cpu.arch) {
+ // GCC on OS X historically did the opposite of ELF for these registers (only in .eh_frame), and that is now the convention for MachO
+ .x86 => if (reg_context.eh_frame and reg_context.is_macho) 4 else 5,
+ .x86_64 => 6,
+ .arm => 11,
+ .aarch64 => 29,
+ else => unreachable,
+ };
+}
+
+pub fn spRegNum(reg_context: RegisterContext) u8 {
+ return switch (builtin.cpu.arch) {
+ .x86 => if (reg_context.eh_frame and reg_context.is_macho) 5 else 4,
+ .x86_64 => 7,
+ .arm => 13,
+ .aarch64 => 31,
+ else => unreachable,
+ };
+}
+
+/// Some platforms use pointer authentication - the upper bits of instruction pointers contain a signature.
+/// This function clears these signature bits to make the pointer usable.
+pub inline fn stripInstructionPtrAuthCode(ptr: usize) usize {
+ if (builtin.cpu.arch == .aarch64) {
+ // `hint 0x07` maps to `xpaclri` (or `nop` if the hardware doesn't support it)
+ // The save / restore is because `xpaclri` operates on x30 (LR)
+ return asm (
+ \\mov x16, x30
+ \\mov x30, x15
+ \\hint 0x07
+ \\mov x15, x30
+ \\mov x30, x16
+ : [ret] "={x15}" (-> usize),
+ : [ptr] "{x15}" (ptr),
+ : "x16"
+ );
+ }
+
+ return ptr;
+}
+
+pub const RegisterContext = struct {
+ eh_frame: bool,
+ is_macho: bool,
+};
+
+pub const AbiError = error{
+ InvalidRegister,
+ UnimplementedArch,
+ UnimplementedOs,
+ RegisterContextRequired,
+ ThreadContextNotSupported,
+};
+
+fn RegValueReturnType(comptime ContextPtrType: type, comptime T: type) type {
+ const reg_bytes_type = comptime RegBytesReturnType(ContextPtrType);
+ const info = @typeInfo(reg_bytes_type).Pointer;
+ return @Type(.{
+ .Pointer = .{
+ .size = .One,
+ .is_const = info.is_const,
+ .is_volatile = info.is_volatile,
+ .is_allowzero = info.is_allowzero,
+ .alignment = info.alignment,
+ .address_space = info.address_space,
+ .child = T,
+ .sentinel = null,
+ },
+ });
+}
+
+/// Returns a pointer to a register stored in a ThreadContext, preserving the pointer attributes of the context.
+pub fn regValueNative(
+ comptime T: type,
+ thread_context_ptr: anytype,
+ reg_number: u8,
+ reg_context: ?RegisterContext,
+) !RegValueReturnType(@TypeOf(thread_context_ptr), T) {
+ const reg_bytes = try regBytes(thread_context_ptr, reg_number, reg_context);
+ if (@sizeOf(T) != reg_bytes.len) return error.IncompatibleRegisterSize;
+ return mem.bytesAsValue(T, reg_bytes[0..@sizeOf(T)]);
+}
+
+fn RegBytesReturnType(comptime ContextPtrType: type) type {
+ const info = @typeInfo(ContextPtrType);
+ if (info != .Pointer or info.Pointer.child != std.debug.ThreadContext) {
+ @compileError("Expected a pointer to std.debug.ThreadContext, got " ++ @typeName(@TypeOf(ContextPtrType)));
+ }
+
+ return if (info.Pointer.is_const) return []const u8 else []u8;
+}
+
+/// Returns a slice containing the backing storage for `reg_number`.
+///
+/// `reg_context` describes in what context the register number is used, as it can have different
+/// meanings depending on the DWARF container. It is only required when getting the stack or
+/// frame pointer register on some architectures.
+pub fn regBytes(
+ thread_context_ptr: anytype,
+ reg_number: u8,
+ reg_context: ?RegisterContext,
+) AbiError!RegBytesReturnType(@TypeOf(thread_context_ptr)) {
+ if (native_os == .windows) {
+ return switch (builtin.cpu.arch) {
+ .x86 => switch (reg_number) {
+ 0 => mem.asBytes(&thread_context_ptr.Eax),
+ 1 => mem.asBytes(&thread_context_ptr.Ecx),
+ 2 => mem.asBytes(&thread_context_ptr.Edx),
+ 3 => mem.asBytes(&thread_context_ptr.Ebx),
+ 4 => mem.asBytes(&thread_context_ptr.Esp),
+ 5 => mem.asBytes(&thread_context_ptr.Ebp),
+ 6 => mem.asBytes(&thread_context_ptr.Esi),
+ 7 => mem.asBytes(&thread_context_ptr.Edi),
+ 8 => mem.asBytes(&thread_context_ptr.Eip),
+ 9 => mem.asBytes(&thread_context_ptr.EFlags),
+ 10 => mem.asBytes(&thread_context_ptr.SegCs),
+ 11 => mem.asBytes(&thread_context_ptr.SegSs),
+ 12 => mem.asBytes(&thread_context_ptr.SegDs),
+ 13 => mem.asBytes(&thread_context_ptr.SegEs),
+ 14 => mem.asBytes(&thread_context_ptr.SegFs),
+ 15 => mem.asBytes(&thread_context_ptr.SegGs),
+ else => error.InvalidRegister,
+ },
+ .x86_64 => switch (reg_number) {
+ 0 => mem.asBytes(&thread_context_ptr.Rax),
+ 1 => mem.asBytes(&thread_context_ptr.Rdx),
+ 2 => mem.asBytes(&thread_context_ptr.Rcx),
+ 3 => mem.asBytes(&thread_context_ptr.Rbx),
+ 4 => mem.asBytes(&thread_context_ptr.Rsi),
+ 5 => mem.asBytes(&thread_context_ptr.Rdi),
+ 6 => mem.asBytes(&thread_context_ptr.Rbp),
+ 7 => mem.asBytes(&thread_context_ptr.Rsp),
+ 8 => mem.asBytes(&thread_context_ptr.R8),
+ 9 => mem.asBytes(&thread_context_ptr.R9),
+ 10 => mem.asBytes(&thread_context_ptr.R10),
+ 11 => mem.asBytes(&thread_context_ptr.R11),
+ 12 => mem.asBytes(&thread_context_ptr.R12),
+ 13 => mem.asBytes(&thread_context_ptr.R13),
+ 14 => mem.asBytes(&thread_context_ptr.R14),
+ 15 => mem.asBytes(&thread_context_ptr.R15),
+ 16 => mem.asBytes(&thread_context_ptr.Rip),
+ else => error.InvalidRegister,
+ },
+ .aarch64 => switch (reg_number) {
+ 0...30 => mem.asBytes(&thread_context_ptr.DUMMYUNIONNAME.X[reg_number]),
+ 31 => mem.asBytes(&thread_context_ptr.Sp),
+ 32 => mem.asBytes(&thread_context_ptr.Pc),
+ else => error.InvalidRegister,
+ },
+ else => error.UnimplementedArch,
+ };
+ }
+
+ if (!std.debug.have_ucontext) return error.ThreadContextNotSupported;
+
+ const ucontext_ptr = thread_context_ptr;
+ return switch (builtin.cpu.arch) {
+ .x86 => switch (native_os) {
+ .linux, .netbsd, .solaris, .illumos => switch (reg_number) {
+ 0 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.EAX]),
+ 1 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.ECX]),
+ 2 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.EDX]),
+ 3 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.EBX]),
+ 4...5 => if (reg_context) |r| bytes: {
+ if (reg_number == 4) {
+ break :bytes if (r.eh_frame and r.is_macho)
+ mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.EBP])
+ else
+ mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.ESP]);
+ } else {
+ break :bytes if (r.eh_frame and r.is_macho)
+ mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.ESP])
+ else
+ mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.EBP]);
+ }
+ } else error.RegisterContextRequired,
+ 6 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.ESI]),
+ 7 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.EDI]),
+ 8 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.EIP]),
+ 9 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.EFL]),
+ 10 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.CS]),
+ 11 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.SS]),
+ 12 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.DS]),
+ 13 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.ES]),
+ 14 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.FS]),
+ 15 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.GS]),
+ 16...23 => error.InvalidRegister, // TODO: Support loading ST0-ST7 from mcontext.fpregs
+ 32...39 => error.InvalidRegister, // TODO: Support loading XMM0-XMM7 from mcontext.fpregs
+ else => error.InvalidRegister,
+ },
+ else => error.UnimplementedOs,
+ },
+ .x86_64 => switch (native_os) {
+ .linux, .solaris, .illumos => switch (reg_number) {
+ 0 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.RAX]),
+ 1 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.RDX]),
+ 2 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.RCX]),
+ 3 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.RBX]),
+ 4 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.RSI]),
+ 5 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.RDI]),
+ 6 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.RBP]),
+ 7 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.RSP]),
+ 8 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.R8]),
+ 9 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.R9]),
+ 10 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.R10]),
+ 11 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.R11]),
+ 12 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.R12]),
+ 13 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.R13]),
+ 14 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.R14]),
+ 15 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.R15]),
+ 16 => mem.asBytes(&ucontext_ptr.mcontext.gregs[posix.REG.RIP]),
+ 17...32 => |i| if (native_os.isSolarish())
+ mem.asBytes(&ucontext_ptr.mcontext.fpregs.chip_state.xmm[i - 17])
+ else
+ mem.asBytes(&ucontext_ptr.mcontext.fpregs.xmm[i - 17]),
+ else => error.InvalidRegister,
+ },
+ .freebsd => switch (reg_number) {
+ 0 => mem.asBytes(&ucontext_ptr.mcontext.rax),
+ 1 => mem.asBytes(&ucontext_ptr.mcontext.rdx),
+ 2 => mem.asBytes(&ucontext_ptr.mcontext.rcx),
+ 3 => mem.asBytes(&ucontext_ptr.mcontext.rbx),
+ 4 => mem.asBytes(&ucontext_ptr.mcontext.rsi),
+ 5 => mem.asBytes(&ucontext_ptr.mcontext.rdi),
+ 6 => mem.asBytes(&ucontext_ptr.mcontext.rbp),
+ 7 => mem.asBytes(&ucontext_ptr.mcontext.rsp),
+ 8 => mem.asBytes(&ucontext_ptr.mcontext.r8),
+ 9 => mem.asBytes(&ucontext_ptr.mcontext.r9),
+ 10 => mem.asBytes(&ucontext_ptr.mcontext.r10),
+ 11 => mem.asBytes(&ucontext_ptr.mcontext.r11),
+ 12 => mem.asBytes(&ucontext_ptr.mcontext.r12),
+ 13 => mem.asBytes(&ucontext_ptr.mcontext.r13),
+ 14 => mem.asBytes(&ucontext_ptr.mcontext.r14),
+ 15 => mem.asBytes(&ucontext_ptr.mcontext.r15),
+ 16 => mem.asBytes(&ucontext_ptr.mcontext.rip),
+ // TODO: Extract xmm state from mcontext.fpstate?
+ else => error.InvalidRegister,
+ },
+ .openbsd => switch (reg_number) {
+ 0 => mem.asBytes(&ucontext_ptr.sc_rax),
+ 1 => mem.asBytes(&ucontext_ptr.sc_rdx),
+ 2 => mem.asBytes(&ucontext_ptr.sc_rcx),
+ 3 => mem.asBytes(&ucontext_ptr.sc_rbx),
+ 4 => mem.asBytes(&ucontext_ptr.sc_rsi),
+ 5 => mem.asBytes(&ucontext_ptr.sc_rdi),
+ 6 => mem.asBytes(&ucontext_ptr.sc_rbp),
+ 7 => mem.asBytes(&ucontext_ptr.sc_rsp),
+ 8 => mem.asBytes(&ucontext_ptr.sc_r8),
+ 9 => mem.asBytes(&ucontext_ptr.sc_r9),
+ 10 => mem.asBytes(&ucontext_ptr.sc_r10),
+ 11 => mem.asBytes(&ucontext_ptr.sc_r11),
+ 12 => mem.asBytes(&ucontext_ptr.sc_r12),
+ 13 => mem.asBytes(&ucontext_ptr.sc_r13),
+ 14 => mem.asBytes(&ucontext_ptr.sc_r14),
+ 15 => mem.asBytes(&ucontext_ptr.sc_r15),
+ 16 => mem.asBytes(&ucontext_ptr.sc_rip),
+ // TODO: Extract xmm state from sc_fpstate?
+ else => error.InvalidRegister,
+ },
+ .macos, .ios => switch (reg_number) {
+ 0 => mem.asBytes(&ucontext_ptr.mcontext.ss.rax),
+ 1 => mem.asBytes(&ucontext_ptr.mcontext.ss.rdx),
+ 2 => mem.asBytes(&ucontext_ptr.mcontext.ss.rcx),
+ 3 => mem.asBytes(&ucontext_ptr.mcontext.ss.rbx),
+ 4 => mem.asBytes(&ucontext_ptr.mcontext.ss.rsi),
+ 5 => mem.asBytes(&ucontext_ptr.mcontext.ss.rdi),
+ 6 => mem.asBytes(&ucontext_ptr.mcontext.ss.rbp),
+ 7 => mem.asBytes(&ucontext_ptr.mcontext.ss.rsp),
+ 8 => mem.asBytes(&ucontext_ptr.mcontext.ss.r8),
+ 9 => mem.asBytes(&ucontext_ptr.mcontext.ss.r9),
+ 10 => mem.asBytes(&ucontext_ptr.mcontext.ss.r10),
+ 11 => mem.asBytes(&ucontext_ptr.mcontext.ss.r11),
+ 12 => mem.asBytes(&ucontext_ptr.mcontext.ss.r12),
+ 13 => mem.asBytes(&ucontext_ptr.mcontext.ss.r13),
+ 14 => mem.asBytes(&ucontext_ptr.mcontext.ss.r14),
+ 15 => mem.asBytes(&ucontext_ptr.mcontext.ss.r15),
+ 16 => mem.asBytes(&ucontext_ptr.mcontext.ss.rip),
+ else => error.InvalidRegister,
+ },
+ else => error.UnimplementedOs,
+ },
+ .arm => switch (native_os) {
+ .linux => switch (reg_number) {
+ 0 => mem.asBytes(&ucontext_ptr.mcontext.arm_r0),
+ 1 => mem.asBytes(&ucontext_ptr.mcontext.arm_r1),
+ 2 => mem.asBytes(&ucontext_ptr.mcontext.arm_r2),
+ 3 => mem.asBytes(&ucontext_ptr.mcontext.arm_r3),
+ 4 => mem.asBytes(&ucontext_ptr.mcontext.arm_r4),
+ 5 => mem.asBytes(&ucontext_ptr.mcontext.arm_r5),
+ 6 => mem.asBytes(&ucontext_ptr.mcontext.arm_r6),
+ 7 => mem.asBytes(&ucontext_ptr.mcontext.arm_r7),
+ 8 => mem.asBytes(&ucontext_ptr.mcontext.arm_r8),
+ 9 => mem.asBytes(&ucontext_ptr.mcontext.arm_r9),
+ 10 => mem.asBytes(&ucontext_ptr.mcontext.arm_r10),
+ 11 => mem.asBytes(&ucontext_ptr.mcontext.arm_fp),
+ 12 => mem.asBytes(&ucontext_ptr.mcontext.arm_ip),
+ 13 => mem.asBytes(&ucontext_ptr.mcontext.arm_sp),
+ 14 => mem.asBytes(&ucontext_ptr.mcontext.arm_lr),
+ 15 => mem.asBytes(&ucontext_ptr.mcontext.arm_pc),
+ // CPSR is not allocated a register number (See: https://github.com/ARM-software/abi-aa/blob/main/aadwarf32/aadwarf32.rst, Section 4.1)
+ else => error.InvalidRegister,
+ },
+ else => error.UnimplementedOs,
+ },
+ .aarch64 => switch (native_os) {
+ .macos, .ios => switch (reg_number) {
+ 0...28 => mem.asBytes(&ucontext_ptr.mcontext.ss.regs[reg_number]),
+ 29 => mem.asBytes(&ucontext_ptr.mcontext.ss.fp),
+ 30 => mem.asBytes(&ucontext_ptr.mcontext.ss.lr),
+ 31 => mem.asBytes(&ucontext_ptr.mcontext.ss.sp),
+ 32 => mem.asBytes(&ucontext_ptr.mcontext.ss.pc),
+
+ // TODO: Find storage for this state
+ //34 => mem.asBytes(&ucontext_ptr.ra_sign_state),
+
+ // V0-V31
+ 64...95 => mem.asBytes(&ucontext_ptr.mcontext.ns.q[reg_number - 64]),
+ else => error.InvalidRegister,
+ },
+ .netbsd => switch (reg_number) {
+ 0...34 => mem.asBytes(&ucontext_ptr.mcontext.gregs[reg_number]),
+ else => error.InvalidRegister,
+ },
+ .freebsd => switch (reg_number) {
+ 0...29 => mem.asBytes(&ucontext_ptr.mcontext.gpregs.x[reg_number]),
+ 30 => mem.asBytes(&ucontext_ptr.mcontext.gpregs.lr),
+ 31 => mem.asBytes(&ucontext_ptr.mcontext.gpregs.sp),
+
+ // TODO: This seems wrong, but it was in the previous debug.zig code for mapping PC, check this
+ 32 => mem.asBytes(&ucontext_ptr.mcontext.gpregs.elr),
+
+ else => error.InvalidRegister,
+ },
+ .openbsd => switch (reg_number) {
+ 0...30 => mem.asBytes(&ucontext_ptr.sc_x[reg_number]),
+ 31 => mem.asBytes(&ucontext_ptr.sc_sp),
+ 32 => mem.asBytes(&ucontext_ptr.sc_lr),
+ 33 => mem.asBytes(&ucontext_ptr.sc_elr),
+ 34 => mem.asBytes(&ucontext_ptr.sc_spsr),
+ else => error.InvalidRegister,
+ },
+ else => switch (reg_number) {
+ 0...30 => mem.asBytes(&ucontext_ptr.mcontext.regs[reg_number]),
+ 31 => mem.asBytes(&ucontext_ptr.mcontext.sp),
+ 32 => mem.asBytes(&ucontext_ptr.mcontext.pc),
+ else => error.InvalidRegister,
+ },
+ },
+ else => error.UnimplementedArch,
+ };
+}
+
+/// Returns the ABI-defined default value this register has in the unwinding table
+/// before running any of the CIE instructions. The DWARF spec defines these as having
+/// the .undefined rule by default, but allows ABI authors to override that.
+pub fn getRegDefaultValue(reg_number: u8, context: *std.debug.Dwarf.UnwindContext, out: []u8) !void {
+ switch (builtin.cpu.arch) {
+ .aarch64 => {
+ // Callee-saved registers are initialized as if they had the .same_value rule
+ if (reg_number >= 19 and reg_number <= 28) {
+ const src = try regBytes(context.thread_context, reg_number, context.reg_context);
+ if (src.len != out.len) return error.RegisterSizeMismatch;
+ @memcpy(out, src);
+ return;
+ }
+ },
+ else => {},
+ }
+
+ @memset(out, undefined);
+}
diff --git a/lib/std/debug/Dwarf/call_frame.zig b/lib/std/debug/Dwarf/call_frame.zig
new file mode 100644
index 0000000000..73e00d3099
--- /dev/null
+++ b/lib/std/debug/Dwarf/call_frame.zig
@@ -0,0 +1,687 @@
+const builtin = @import("builtin");
+const std = @import("../../std.zig");
+const mem = std.mem;
+const debug = std.debug;
+const leb = std.leb;
+const DW = std.dwarf;
+const abi = std.debug.Dwarf.abi;
+const assert = std.debug.assert;
+const native_endian = builtin.cpu.arch.endian();
+
+/// TODO merge with std.dwarf.CFA
+const Opcode = enum(u8) {
+ advance_loc = 0x1 << 6,
+ offset = 0x2 << 6,
+ restore = 0x3 << 6,
+
+ nop = 0x00,
+ set_loc = 0x01,
+ advance_loc1 = 0x02,
+ advance_loc2 = 0x03,
+ advance_loc4 = 0x04,
+ offset_extended = 0x05,
+ restore_extended = 0x06,
+ undefined = 0x07,
+ same_value = 0x08,
+ register = 0x09,
+ remember_state = 0x0a,
+ restore_state = 0x0b,
+ def_cfa = 0x0c,
+ def_cfa_register = 0x0d,
+ def_cfa_offset = 0x0e,
+ def_cfa_expression = 0x0f,
+ expression = 0x10,
+ offset_extended_sf = 0x11,
+ def_cfa_sf = 0x12,
+ def_cfa_offset_sf = 0x13,
+ val_offset = 0x14,
+ val_offset_sf = 0x15,
+ val_expression = 0x16,
+
+ // These opcodes encode an operand in the lower 6 bits of the opcode itself
+ pub const lo_inline = @intFromEnum(Opcode.advance_loc);
+ pub const hi_inline = @intFromEnum(Opcode.restore) | 0b111111;
+
+ // These opcodes are trailed by zero or more operands
+ pub const lo_reserved = @intFromEnum(Opcode.nop);
+ pub const hi_reserved = @intFromEnum(Opcode.val_expression);
+
+ // Vendor-specific opcodes
+ pub const lo_user = 0x1c;
+ pub const hi_user = 0x3f;
+};
+
+fn readBlock(stream: *std.io.FixedBufferStream([]const u8)) ![]const u8 {
+ const reader = stream.reader();
+ const block_len = try leb.readUleb128(usize, reader);
+ if (stream.pos + block_len > stream.buffer.len) return error.InvalidOperand;
+
+ const block = stream.buffer[stream.pos..][0..block_len];
+ reader.context.pos += block_len;
+
+ return block;
+}
+
+pub const Instruction = union(Opcode) {
+ advance_loc: struct {
+ delta: u8,
+ },
+ offset: struct {
+ register: u8,
+ offset: u64,
+ },
+ restore: struct {
+ register: u8,
+ },
+ nop: void,
+ set_loc: struct {
+ address: u64,
+ },
+ advance_loc1: struct {
+ delta: u8,
+ },
+ advance_loc2: struct {
+ delta: u16,
+ },
+ advance_loc4: struct {
+ delta: u32,
+ },
+ offset_extended: struct {
+ register: u8,
+ offset: u64,
+ },
+ restore_extended: struct {
+ register: u8,
+ },
+ undefined: struct {
+ register: u8,
+ },
+ same_value: struct {
+ register: u8,
+ },
+ register: struct {
+ register: u8,
+ target_register: u8,
+ },
+ remember_state: void,
+ restore_state: void,
+ def_cfa: struct {
+ register: u8,
+ offset: u64,
+ },
+ def_cfa_register: struct {
+ register: u8,
+ },
+ def_cfa_offset: struct {
+ offset: u64,
+ },
+ def_cfa_expression: struct {
+ block: []const u8,
+ },
+ expression: struct {
+ register: u8,
+ block: []const u8,
+ },
+ offset_extended_sf: struct {
+ register: u8,
+ offset: i64,
+ },
+ def_cfa_sf: struct {
+ register: u8,
+ offset: i64,
+ },
+ def_cfa_offset_sf: struct {
+ offset: i64,
+ },
+ val_offset: struct {
+ register: u8,
+ offset: u64,
+ },
+ val_offset_sf: struct {
+ register: u8,
+ offset: i64,
+ },
+ val_expression: struct {
+ register: u8,
+ block: []const u8,
+ },
+
+ pub fn read(
+ stream: *std.io.FixedBufferStream([]const u8),
+ addr_size_bytes: u8,
+ endian: std.builtin.Endian,
+ ) !Instruction {
+ const reader = stream.reader();
+ switch (try reader.readByte()) {
+ Opcode.lo_inline...Opcode.hi_inline => |opcode| {
+ const e: Opcode = @enumFromInt(opcode & 0b11000000);
+ const value: u6 = @intCast(opcode & 0b111111);
+ return switch (e) {
+ .advance_loc => .{
+ .advance_loc = .{ .delta = value },
+ },
+ .offset => .{
+ .offset = .{
+ .register = value,
+ .offset = try leb.readUleb128(u64, reader),
+ },
+ },
+ .restore => .{
+ .restore = .{ .register = value },
+ },
+ else => unreachable,
+ };
+ },
+ Opcode.lo_reserved...Opcode.hi_reserved => |opcode| {
+ const e: Opcode = @enumFromInt(opcode);
+ return switch (e) {
+ .advance_loc,
+ .offset,
+ .restore,
+ => unreachable,
+ .nop => .{ .nop = {} },
+ .set_loc => .{
+ .set_loc = .{
+ .address = switch (addr_size_bytes) {
+ 2 => try reader.readInt(u16, endian),
+ 4 => try reader.readInt(u32, endian),
+ 8 => try reader.readInt(u64, endian),
+ else => return error.InvalidAddrSize,
+ },
+ },
+ },
+ .advance_loc1 => .{
+ .advance_loc1 = .{ .delta = try reader.readByte() },
+ },
+ .advance_loc2 => .{
+ .advance_loc2 = .{ .delta = try reader.readInt(u16, endian) },
+ },
+ .advance_loc4 => .{
+ .advance_loc4 = .{ .delta = try reader.readInt(u32, endian) },
+ },
+ .offset_extended => .{
+ .offset_extended = .{
+ .register = try leb.readUleb128(u8, reader),
+ .offset = try leb.readUleb128(u64, reader),
+ },
+ },
+ .restore_extended => .{
+ .restore_extended = .{
+ .register = try leb.readUleb128(u8, reader),
+ },
+ },
+ .undefined => .{
+ .undefined = .{
+ .register = try leb.readUleb128(u8, reader),
+ },
+ },
+ .same_value => .{
+ .same_value = .{
+ .register = try leb.readUleb128(u8, reader),
+ },
+ },
+ .register => .{
+ .register = .{
+ .register = try leb.readUleb128(u8, reader),
+ .target_register = try leb.readUleb128(u8, reader),
+ },
+ },
+ .remember_state => .{ .remember_state = {} },
+ .restore_state => .{ .restore_state = {} },
+ .def_cfa => .{
+ .def_cfa = .{
+ .register = try leb.readUleb128(u8, reader),
+ .offset = try leb.readUleb128(u64, reader),
+ },
+ },
+ .def_cfa_register => .{
+ .def_cfa_register = .{
+ .register = try leb.readUleb128(u8, reader),
+ },
+ },
+ .def_cfa_offset => .{
+ .def_cfa_offset = .{
+ .offset = try leb.readUleb128(u64, reader),
+ },
+ },
+ .def_cfa_expression => .{
+ .def_cfa_expression = .{
+ .block = try readBlock(stream),
+ },
+ },
+ .expression => .{
+ .expression = .{
+ .register = try leb.readUleb128(u8, reader),
+ .block = try readBlock(stream),
+ },
+ },
+ .offset_extended_sf => .{
+ .offset_extended_sf = .{
+ .register = try leb.readUleb128(u8, reader),
+ .offset = try leb.readIleb128(i64, reader),
+ },
+ },
+ .def_cfa_sf => .{
+ .def_cfa_sf = .{
+ .register = try leb.readUleb128(u8, reader),
+ .offset = try leb.readIleb128(i64, reader),
+ },
+ },
+ .def_cfa_offset_sf => .{
+ .def_cfa_offset_sf = .{
+ .offset = try leb.readIleb128(i64, reader),
+ },
+ },
+ .val_offset => .{
+ .val_offset = .{
+ .register = try leb.readUleb128(u8, reader),
+ .offset = try leb.readUleb128(u64, reader),
+ },
+ },
+ .val_offset_sf => .{
+ .val_offset_sf = .{
+ .register = try leb.readUleb128(u8, reader),
+ .offset = try leb.readIleb128(i64, reader),
+ },
+ },
+ .val_expression => .{
+ .val_expression = .{
+ .register = try leb.readUleb128(u8, reader),
+ .block = try readBlock(stream),
+ },
+ },
+ };
+ },
+ Opcode.lo_user...Opcode.hi_user => return error.UnimplementedUserOpcode,
+ else => return error.InvalidOpcode,
+ }
+ }
+};
+
+/// Since register rules are applied (usually) during a panic,
+/// checked addition / subtraction is used so that we can return
+/// an error and fall back to FP-based unwinding.
+pub fn applyOffset(base: usize, offset: i64) !usize {
+ return if (offset >= 0)
+ try std.math.add(usize, base, @as(usize, @intCast(offset)))
+ else
+ try std.math.sub(usize, base, @as(usize, @intCast(-offset)));
+}
+
+/// This is a virtual machine that runs DWARF call frame instructions.
+pub const VirtualMachine = struct {
+ /// See section 6.4.1 of the DWARF5 specification for details on each
+ const RegisterRule = union(enum) {
+ // The spec says that the default rule for each column is the undefined rule.
+ // However, it also allows ABI / compiler authors to specify alternate defaults, so
+ // there is a distinction made here.
+ default: void,
+
+ undefined: void,
+ same_value: void,
+
+ // offset(N)
+ offset: i64,
+
+ // val_offset(N)
+ val_offset: i64,
+
+ // register(R)
+ register: u8,
+
+ // expression(E)
+ expression: []const u8,
+
+ // val_expression(E)
+ val_expression: []const u8,
+
+ // Augmenter-defined rule
+ architectural: void,
+ };
+
+ /// Each row contains unwinding rules for a set of registers.
+ pub const Row = struct {
+ /// Offset from `FrameDescriptionEntry.pc_begin`
+ offset: u64 = 0,
+
+ /// Special-case column that defines the CFA (Canonical Frame Address) rule.
+ /// The register field of this column defines the register that CFA is derived from.
+ cfa: Column = .{},
+
+ /// The register fields in these columns define the register the rule applies to.
+ columns: ColumnRange = .{},
+
+ /// Indicates that the next write to any column in this row needs to copy
+ /// the backing column storage first, as it may be referenced by previous rows.
+ copy_on_write: bool = false,
+ };
+
+ pub const Column = struct {
+ register: ?u8 = null,
+ rule: RegisterRule = .{ .default = {} },
+
+ /// Resolves the register rule and places the result into `out` (see dwarf.abi.regBytes)
+ pub fn resolveValue(
+ self: Column,
+ context: *std.debug.Dwarf.UnwindContext,
+ expression_context: std.debug.Dwarf.expression.Context,
+ ma: *debug.StackIterator.MemoryAccessor,
+ out: []u8,
+ ) !void {
+ switch (self.rule) {
+ .default => {
+ const register = self.register orelse return error.InvalidRegister;
+ try abi.getRegDefaultValue(register, context, out);
+ },
+ .undefined => {
+ @memset(out, undefined);
+ },
+ .same_value => {
+ // TODO: This copy could be eliminated if callers always copy the state then call this function to update it
+ const register = self.register orelse return error.InvalidRegister;
+ const src = try abi.regBytes(context.thread_context, register, context.reg_context);
+ if (src.len != out.len) return error.RegisterSizeMismatch;
+ @memcpy(out, src);
+ },
+ .offset => |offset| {
+ if (context.cfa) |cfa| {
+ const addr = try applyOffset(cfa, offset);
+ if (ma.load(usize, addr) == null) return error.InvalidAddress;
+ const ptr: *const usize = @ptrFromInt(addr);
+ mem.writeInt(usize, out[0..@sizeOf(usize)], ptr.*, native_endian);
+ } else return error.InvalidCFA;
+ },
+ .val_offset => |offset| {
+ if (context.cfa) |cfa| {
+ mem.writeInt(usize, out[0..@sizeOf(usize)], try applyOffset(cfa, offset), native_endian);
+ } else return error.InvalidCFA;
+ },
+ .register => |register| {
+ const src = try abi.regBytes(context.thread_context, register, context.reg_context);
+ if (src.len != out.len) return error.RegisterSizeMismatch;
+ @memcpy(out, try abi.regBytes(context.thread_context, register, context.reg_context));
+ },
+ .expression => |expression| {
+ context.stack_machine.reset();
+ const value = try context.stack_machine.run(expression, context.allocator, expression_context, context.cfa.?);
+ const addr = if (value) |v| blk: {
+ if (v != .generic) return error.InvalidExpressionValue;
+ break :blk v.generic;
+ } else return error.NoExpressionValue;
+
+ if (ma.load(usize, addr) == null) return error.InvalidExpressionAddress;
+ const ptr: *usize = @ptrFromInt(addr);
+ mem.writeInt(usize, out[0..@sizeOf(usize)], ptr.*, native_endian);
+ },
+ .val_expression => |expression| {
+ context.stack_machine.reset();
+ const value = try context.stack_machine.run(expression, context.allocator, expression_context, context.cfa.?);
+ if (value) |v| {
+ if (v != .generic) return error.InvalidExpressionValue;
+ mem.writeInt(usize, out[0..@sizeOf(usize)], v.generic, native_endian);
+ } else return error.NoExpressionValue;
+ },
+ .architectural => return error.UnimplementedRegisterRule,
+ }
+ }
+ };
+
+ const ColumnRange = struct {
+ /// Index into `columns` of the first column in this row.
+ start: usize = undefined,
+ len: u8 = 0,
+ };
+
+ columns: std.ArrayListUnmanaged(Column) = .{},
+ stack: std.ArrayListUnmanaged(ColumnRange) = .{},
+ current_row: Row = .{},
+
+ /// The result of executing the CIE's initial_instructions
+ cie_row: ?Row = null,
+
+ pub fn deinit(self: *VirtualMachine, allocator: std.mem.Allocator) void {
+ self.stack.deinit(allocator);
+ self.columns.deinit(allocator);
+ self.* = undefined;
+ }
+
+ pub fn reset(self: *VirtualMachine) void {
+ self.stack.clearRetainingCapacity();
+ self.columns.clearRetainingCapacity();
+ self.current_row = .{};
+ self.cie_row = null;
+ }
+
+ /// Return a slice backed by the row's non-CFA columns
+ pub fn rowColumns(self: VirtualMachine, row: Row) []Column {
+ if (row.columns.len == 0) return &.{};
+ return self.columns.items[row.columns.start..][0..row.columns.len];
+ }
+
+ /// Either retrieves or adds a column for `register` (non-CFA) in the current row.
+ fn getOrAddColumn(self: *VirtualMachine, allocator: std.mem.Allocator, register: u8) !*Column {
+ for (self.rowColumns(self.current_row)) |*c| {
+ if (c.register == register) return c;
+ }
+
+ if (self.current_row.columns.len == 0) {
+ self.current_row.columns.start = self.columns.items.len;
+ }
+ self.current_row.columns.len += 1;
+
+ const column = try self.columns.addOne(allocator);
+ column.* = .{
+ .register = register,
+ };
+
+ return column;
+ }
+
+ /// Runs the CIE instructions, then the FDE instructions. Execution halts
+ /// once the row that corresponds to `pc` is known, and the row is returned.
+ pub fn runTo(
+ self: *VirtualMachine,
+ allocator: std.mem.Allocator,
+ pc: u64,
+ cie: std.debug.Dwarf.CommonInformationEntry,
+ fde: std.debug.Dwarf.FrameDescriptionEntry,
+ addr_size_bytes: u8,
+ endian: std.builtin.Endian,
+ ) !Row {
+ assert(self.cie_row == null);
+ if (pc < fde.pc_begin or pc >= fde.pc_begin + fde.pc_range) return error.AddressOutOfRange;
+
+ var prev_row: Row = self.current_row;
+
+ var cie_stream = std.io.fixedBufferStream(cie.initial_instructions);
+ var fde_stream = std.io.fixedBufferStream(fde.instructions);
+ var streams = [_]*std.io.FixedBufferStream([]const u8){
+ &cie_stream,
+ &fde_stream,
+ };
+
+ for (&streams, 0..) |stream, i| {
+ while (stream.pos < stream.buffer.len) {
+ const instruction = try std.debug.Dwarf.call_frame.Instruction.read(stream, addr_size_bytes, endian);
+ prev_row = try self.step(allocator, cie, i == 0, instruction);
+ if (pc < fde.pc_begin + self.current_row.offset) return prev_row;
+ }
+ }
+
+ return self.current_row;
+ }
+
+ pub fn runToNative(
+ self: *VirtualMachine,
+ allocator: std.mem.Allocator,
+ pc: u64,
+ cie: std.debug.Dwarf.CommonInformationEntry,
+ fde: std.debug.Dwarf.FrameDescriptionEntry,
+ ) !Row {
+ return self.runTo(allocator, pc, cie, fde, @sizeOf(usize), builtin.target.cpu.arch.endian());
+ }
+
+ fn resolveCopyOnWrite(self: *VirtualMachine, allocator: std.mem.Allocator) !void {
+ if (!self.current_row.copy_on_write) return;
+
+ const new_start = self.columns.items.len;
+ if (self.current_row.columns.len > 0) {
+ try self.columns.ensureUnusedCapacity(allocator, self.current_row.columns.len);
+ self.columns.appendSliceAssumeCapacity(self.rowColumns(self.current_row));
+ self.current_row.columns.start = new_start;
+ }
+ }
+
+ /// Executes a single instruction.
+ /// If this instruction is from the CIE, `is_initial` should be set.
+ /// Returns the value of `current_row` before executing this instruction.
+ pub fn step(
+ self: *VirtualMachine,
+ allocator: std.mem.Allocator,
+ cie: std.debug.Dwarf.CommonInformationEntry,
+ is_initial: bool,
+ instruction: Instruction,
+ ) !Row {
+ // CIE instructions must be run before FDE instructions
+ assert(!is_initial or self.cie_row == null);
+ if (!is_initial and self.cie_row == null) {
+ self.cie_row = self.current_row;
+ self.current_row.copy_on_write = true;
+ }
+
+ const prev_row = self.current_row;
+ switch (instruction) {
+ .set_loc => |i| {
+ if (i.address <= self.current_row.offset) return error.InvalidOperation;
+ // TODO: Check cie.segment_selector_size != 0 for DWARFV4
+ self.current_row.offset = i.address;
+ },
+ inline .advance_loc,
+ .advance_loc1,
+ .advance_loc2,
+ .advance_loc4,
+ => |i| {
+ self.current_row.offset += i.delta * cie.code_alignment_factor;
+ self.current_row.copy_on_write = true;
+ },
+ inline .offset,
+ .offset_extended,
+ .offset_extended_sf,
+ => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ const column = try self.getOrAddColumn(allocator, i.register);
+ column.rule = .{ .offset = @as(i64, @intCast(i.offset)) * cie.data_alignment_factor };
+ },
+ inline .restore,
+ .restore_extended,
+ => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ if (self.cie_row) |cie_row| {
+ const column = try self.getOrAddColumn(allocator, i.register);
+ column.rule = for (self.rowColumns(cie_row)) |cie_column| {
+ if (cie_column.register == i.register) break cie_column.rule;
+ } else .{ .default = {} };
+ } else return error.InvalidOperation;
+ },
+ .nop => {},
+ .undefined => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ const column = try self.getOrAddColumn(allocator, i.register);
+ column.rule = .{ .undefined = {} };
+ },
+ .same_value => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ const column = try self.getOrAddColumn(allocator, i.register);
+ column.rule = .{ .same_value = {} };
+ },
+ .register => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ const column = try self.getOrAddColumn(allocator, i.register);
+ column.rule = .{ .register = i.target_register };
+ },
+ .remember_state => {
+ try self.stack.append(allocator, self.current_row.columns);
+ self.current_row.copy_on_write = true;
+ },
+ .restore_state => {
+ const restored_columns = self.stack.popOrNull() orelse return error.InvalidOperation;
+ self.columns.shrinkRetainingCapacity(self.columns.items.len - self.current_row.columns.len);
+ try self.columns.ensureUnusedCapacity(allocator, restored_columns.len);
+
+ self.current_row.columns.start = self.columns.items.len;
+ self.current_row.columns.len = restored_columns.len;
+ self.columns.appendSliceAssumeCapacity(self.columns.items[restored_columns.start..][0..restored_columns.len]);
+ },
+ .def_cfa => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ self.current_row.cfa = .{
+ .register = i.register,
+ .rule = .{ .val_offset = @intCast(i.offset) },
+ };
+ },
+ .def_cfa_sf => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ self.current_row.cfa = .{
+ .register = i.register,
+ .rule = .{ .val_offset = i.offset * cie.data_alignment_factor },
+ };
+ },
+ .def_cfa_register => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ if (self.current_row.cfa.register == null or self.current_row.cfa.rule != .val_offset) return error.InvalidOperation;
+ self.current_row.cfa.register = i.register;
+ },
+ .def_cfa_offset => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ if (self.current_row.cfa.register == null or self.current_row.cfa.rule != .val_offset) return error.InvalidOperation;
+ self.current_row.cfa.rule = .{
+ .val_offset = @intCast(i.offset),
+ };
+ },
+ .def_cfa_offset_sf => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ if (self.current_row.cfa.register == null or self.current_row.cfa.rule != .val_offset) return error.InvalidOperation;
+ self.current_row.cfa.rule = .{
+ .val_offset = i.offset * cie.data_alignment_factor,
+ };
+ },
+ .def_cfa_expression => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ self.current_row.cfa.register = undefined;
+ self.current_row.cfa.rule = .{
+ .expression = i.block,
+ };
+ },
+ .expression => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ const column = try self.getOrAddColumn(allocator, i.register);
+ column.rule = .{
+ .expression = i.block,
+ };
+ },
+ .val_offset => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ const column = try self.getOrAddColumn(allocator, i.register);
+ column.rule = .{
+ .val_offset = @as(i64, @intCast(i.offset)) * cie.data_alignment_factor,
+ };
+ },
+ .val_offset_sf => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ const column = try self.getOrAddColumn(allocator, i.register);
+ column.rule = .{
+ .val_offset = i.offset * cie.data_alignment_factor,
+ };
+ },
+ .val_expression => |i| {
+ try self.resolveCopyOnWrite(allocator);
+ const column = try self.getOrAddColumn(allocator, i.register);
+ column.rule = .{
+ .val_expression = i.block,
+ };
+ },
+ }
+
+ return prev_row;
+ }
+};
diff --git a/lib/std/debug/Dwarf/expression.zig b/lib/std/debug/Dwarf/expression.zig
new file mode 100644
index 0000000000..6243ea9717
--- /dev/null
+++ b/lib/std/debug/Dwarf/expression.zig
@@ -0,0 +1,1638 @@
+const std = @import("std");
+const builtin = @import("builtin");
+const leb = std.leb;
+const OP = std.dwarf.OP;
+const abi = std.debug.Dwarf.abi;
+const mem = std.mem;
+const assert = std.debug.assert;
+const native_endian = builtin.cpu.arch.endian();
+
+/// Expressions can be evaluated in different contexts, each requiring its own set of inputs.
+/// Callers should specify all the fields relevant to their context. If a field is required
+/// by the expression and it isn't in the context, error.IncompleteExpressionContext is returned.
+pub const Context = struct {
+ /// The dwarf format of the section this expression is in
+ format: std.dwarf.Format = .@"32",
+ /// If specified, any addresses will pass through before being accessed
+ memory_accessor: ?*std.debug.StackIterator.MemoryAccessor = null,
+ /// The compilation unit this expression relates to, if any
+ compile_unit: ?*const std.debug.Dwarf.CompileUnit = null,
+ /// When evaluating a user-presented expression, this is the address of the object being evaluated
+ object_address: ?*const anyopaque = null,
+ /// .debug_addr section
+ debug_addr: ?[]const u8 = null,
+ /// Thread context
+ thread_context: ?*std.debug.ThreadContext = null,
+ reg_context: ?abi.RegisterContext = null,
+ /// Call frame address, if in a CFI context
+ cfa: ?usize = null,
+ /// This expression is a sub-expression from an OP.entry_value instruction
+ entry_value_context: bool = false,
+};
+
+pub const Options = struct {
+ /// The address size of the target architecture
+ addr_size: u8 = @sizeOf(usize),
+ /// Endianness of the target architecture
+ endian: std.builtin.Endian = builtin.target.cpu.arch.endian(),
+ /// Restrict the stack machine to a subset of opcodes used in call frame instructions
+ call_frame_context: bool = false,
+};
+
+// Explicitly defined to support executing sub-expressions
+pub const Error = error{
+ UnimplementedExpressionCall,
+ UnimplementedOpcode,
+ UnimplementedUserOpcode,
+ UnimplementedTypedComparison,
+ UnimplementedTypeConversion,
+
+ UnknownExpressionOpcode,
+
+ IncompleteExpressionContext,
+
+ InvalidCFAOpcode,
+ InvalidExpression,
+ InvalidFrameBase,
+ InvalidIntegralTypeSize,
+ InvalidRegister,
+ InvalidSubExpression,
+ InvalidTypeLength,
+
+ TruncatedIntegralType,
+} || abi.AbiError || error{ EndOfStream, Overflow, OutOfMemory, DivisionByZero };
+
+/// 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.
+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) = .{},
+
+ 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 != comptime builtin.target.cpu.arch.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;
+ }
+ };
+}
+
+pub fn Builder(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),
+ };
+
+ return struct {
+ /// Zero-operand instructions
+ pub fn writeOpcode(writer: anytype, comptime opcode: u8) !void {
+ if (options.call_frame_context and !comptime isOpcodeValidInCFA(opcode)) return error.InvalidCFAOpcode;
+ switch (opcode) {
+ OP.dup,
+ OP.drop,
+ OP.over,
+ OP.swap,
+ OP.rot,
+ OP.deref,
+ OP.xderef,
+ OP.push_object_address,
+ OP.form_tls_address,
+ OP.call_frame_cfa,
+ OP.abs,
+ OP.@"and",
+ OP.div,
+ OP.minus,
+ OP.mod,
+ OP.mul,
+ OP.neg,
+ OP.not,
+ OP.@"or",
+ OP.plus,
+ OP.shl,
+ OP.shr,
+ OP.shra,
+ OP.xor,
+ OP.le,
+ OP.ge,
+ OP.eq,
+ OP.lt,
+ OP.gt,
+ OP.ne,
+ OP.nop,
+ OP.stack_value,
+ => try writer.writeByte(opcode),
+ else => @compileError("This opcode requires operands, use `write<Opcode>()` instead"),
+ }
+ }
+
+ // 2.5.1.1: Literal Encodings
+ pub fn writeLiteral(writer: anytype, literal: u8) !void {
+ switch (literal) {
+ 0...31 => |n| try writer.writeByte(n + OP.lit0),
+ else => return error.InvalidLiteral,
+ }
+ }
+
+ pub fn writeConst(writer: anytype, comptime T: type, value: T) !void {
+ if (@typeInfo(T) != .Int) @compileError("Constants must be integers");
+
+ switch (T) {
+ u8, i8, u16, i16, u32, i32, u64, i64 => {
+ try writer.writeByte(switch (T) {
+ u8 => OP.const1u,
+ i8 => OP.const1s,
+ u16 => OP.const2u,
+ i16 => OP.const2s,
+ u32 => OP.const4u,
+ i32 => OP.const4s,
+ u64 => OP.const8u,
+ i64 => OP.const8s,
+ else => unreachable,
+ });
+
+ try writer.writeInt(T, value, options.endian);
+ },
+ else => switch (@typeInfo(T).Int.signedness) {
+ .unsigned => {
+ try writer.writeByte(OP.constu);
+ try leb.writeUleb128(writer, value);
+ },
+ .signed => {
+ try writer.writeByte(OP.consts);
+ try leb.writeIleb128(writer, value);
+ },
+ },
+ }
+ }
+
+ pub fn writeConstx(writer: anytype, debug_addr_offset: anytype) !void {
+ try writer.writeByte(OP.constx);
+ try leb.writeUleb128(writer, debug_addr_offset);
+ }
+
+ pub fn writeConstType(writer: anytype, die_offset: anytype, value_bytes: []const u8) !void {
+ if (options.call_frame_context) return error.InvalidCFAOpcode;
+ if (value_bytes.len > 0xff) return error.InvalidTypeLength;
+ try writer.writeByte(OP.const_type);
+ try leb.writeUleb128(writer, die_offset);
+ try writer.writeByte(@intCast(value_bytes.len));
+ try writer.writeAll(value_bytes);
+ }
+
+ pub fn writeAddr(writer: anytype, value: addr_type) !void {
+ try writer.writeByte(OP.addr);
+ try writer.writeInt(addr_type, value, options.endian);
+ }
+
+ pub fn writeAddrx(writer: anytype, debug_addr_offset: anytype) !void {
+ if (options.call_frame_context) return error.InvalidCFAOpcode;
+ try writer.writeByte(OP.addrx);
+ try leb.writeUleb128(writer, debug_addr_offset);
+ }
+
+ // 2.5.1.2: Register Values
+ pub fn writeFbreg(writer: anytype, offset: anytype) !void {
+ try writer.writeByte(OP.fbreg);
+ try leb.writeIleb128(writer, offset);
+ }
+
+ pub fn writeBreg(writer: anytype, register: u8, offset: anytype) !void {
+ if (register > 31) return error.InvalidRegister;
+ try writer.writeByte(OP.breg0 + register);
+ try leb.writeIleb128(writer, offset);
+ }
+
+ pub fn writeBregx(writer: anytype, register: anytype, offset: anytype) !void {
+ try writer.writeByte(OP.bregx);
+ try leb.writeUleb128(writer, register);
+ try leb.writeIleb128(writer, offset);
+ }
+
+ pub fn writeRegvalType(writer: anytype, register: anytype, offset: anytype) !void {
+ if (options.call_frame_context) return error.InvalidCFAOpcode;
+ try writer.writeByte(OP.regval_type);
+ try leb.writeUleb128(writer, register);
+ try leb.writeUleb128(writer, offset);
+ }
+
+ // 2.5.1.3: Stack Operations
+ pub fn writePick(writer: anytype, index: u8) !void {
+ try writer.writeByte(OP.pick);
+ try writer.writeByte(index);
+ }
+
+ pub fn writeDerefSize(writer: anytype, size: u8) !void {
+ try writer.writeByte(OP.deref_size);
+ try writer.writeByte(size);
+ }
+
+ pub fn writeXDerefSize(writer: anytype, size: u8) !void {
+ try writer.writeByte(OP.xderef_size);
+ try writer.writeByte(size);
+ }
+
+ pub fn writeDerefType(writer: anytype, size: u8, die_offset: anytype) !void {
+ if (options.call_frame_context) return error.InvalidCFAOpcode;
+ try writer.writeByte(OP.deref_type);
+ try writer.writeByte(size);
+ try leb.writeUleb128(writer, die_offset);
+ }
+
+ pub fn writeXDerefType(writer: anytype, size: u8, die_offset: anytype) !void {
+ try writer.writeByte(OP.xderef_type);
+ try writer.writeByte(size);
+ try leb.writeUleb128(writer, die_offset);
+ }
+
+ // 2.5.1.4: Arithmetic and Logical Operations
+
+ pub fn writePlusUconst(writer: anytype, uint_value: anytype) !void {
+ try writer.writeByte(OP.plus_uconst);
+ try leb.writeUleb128(writer, uint_value);
+ }
+
+ // 2.5.1.5: Control Flow Operations
+
+ pub fn writeSkip(writer: anytype, offset: i16) !void {
+ try writer.writeByte(OP.skip);
+ try writer.writeInt(i16, offset, options.endian);
+ }
+
+ pub fn writeBra(writer: anytype, offset: i16) !void {
+ try writer.writeByte(OP.bra);
+ try writer.writeInt(i16, offset, options.endian);
+ }
+
+ pub fn writeCall(writer: anytype, comptime T: type, offset: T) !void {
+ if (options.call_frame_context) return error.InvalidCFAOpcode;
+ switch (T) {
+ u16 => try writer.writeByte(OP.call2),
+ u32 => try writer.writeByte(OP.call4),
+ else => @compileError("Call operand must be a 2 or 4 byte offset"),
+ }
+
+ try writer.writeInt(T, offset, options.endian);
+ }
+
+ pub fn writeCallRef(writer: anytype, comptime is_64: bool, value: if (is_64) u64 else u32) !void {
+ if (options.call_frame_context) return error.InvalidCFAOpcode;
+ try writer.writeByte(OP.call_ref);
+ try writer.writeInt(if (is_64) u64 else u32, value, options.endian);
+ }
+
+ pub fn writeConvert(writer: anytype, die_offset: anytype) !void {
+ if (options.call_frame_context) return error.InvalidCFAOpcode;
+ try writer.writeByte(OP.convert);
+ try leb.writeUleb128(writer, die_offset);
+ }
+
+ pub fn writeReinterpret(writer: anytype, die_offset: anytype) !void {
+ if (options.call_frame_context) return error.InvalidCFAOpcode;
+ try writer.writeByte(OP.reinterpret);
+ try leb.writeUleb128(writer, die_offset);
+ }
+
+ // 2.5.1.7: Special Operations
+
+ pub fn writeEntryValue(writer: anytype, expression: []const u8) !void {
+ try writer.writeByte(OP.entry_value);
+ try leb.writeUleb128(writer, expression.len);
+ try writer.writeAll(expression);
+ }
+
+ // 2.6: Location Descriptions
+ pub fn writeReg(writer: anytype, register: u8) !void {
+ try writer.writeByte(OP.reg0 + register);
+ }
+
+ pub fn writeRegx(writer: anytype, register: anytype) !void {
+ try writer.writeByte(OP.regx);
+ try leb.writeUleb128(writer, register);
+ }
+
+ pub fn writeImplicitValue(writer: anytype, value_bytes: []const u8) !void {
+ try writer.writeByte(OP.implicit_value);
+ try leb.writeUleb128(writer, value_bytes.len);
+ try writer.writeAll(value_bytes);
+ }
+ };
+}
+
+// Certain opcodes are not allowed in a CFA context, see 6.4.2
+fn isOpcodeValidInCFA(opcode: u8) bool {
+ return switch (opcode) {
+ OP.addrx,
+ OP.call2,
+ OP.call4,
+ OP.call_ref,
+ OP.const_type,
+ OP.constx,
+ OP.convert,
+ OP.deref_type,
+ OP.regval_type,
+ OP.reinterpret,
+ OP.push_object_address,
+ OP.call_frame_cfa,
+ => false,
+ else => true,
+ };
+}
+
+fn isOpcodeRegisterLocation(opcode: u8) bool {
+ return switch (opcode) {
+ OP.reg0...OP.reg31, OP.regx => true,
+ else => false,
+ };
+}
+
+const testing = std.testing;
+test "DWARF expressions" {
+ const allocator = std.testing.allocator;
+
+ const options = Options{};
+ var stack_machine = StackMachine(options){};
+ defer stack_machine.deinit(allocator);
+
+ const b = Builder(options);
+
+ var program = std.ArrayList(u8).init(allocator);
+ defer program.deinit();
+
+ const writer = program.writer();
+
+ // Literals
+ {
+ const context = Context{};
+ for (0..32) |i| {
+ try b.writeLiteral(writer, @intCast(i));
+ }
+
+ _ = try stack_machine.run(program.items, allocator, context, 0);
+
+ for (0..32) |i| {
+ const expected = 31 - i;
+ try testing.expectEqual(expected, stack_machine.stack.popOrNull().?.generic);
+ }
+ }
+
+ // Constants
+ {
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+
+ const input = [_]comptime_int{
+ 1,
+ -1,
+ @as(usize, @truncate(0x0fff)),
+ @as(isize, @truncate(-0x0fff)),
+ @as(usize, @truncate(0x0fffffff)),
+ @as(isize, @truncate(-0x0fffffff)),
+ @as(usize, @truncate(0x0fffffffffffffff)),
+ @as(isize, @truncate(-0x0fffffffffffffff)),
+ @as(usize, @truncate(0x8000000)),
+ @as(isize, @truncate(-0x8000000)),
+ @as(usize, @truncate(0x12345678_12345678)),
+ @as(usize, @truncate(0xffffffff_ffffffff)),
+ @as(usize, @truncate(0xeeeeeeee_eeeeeeee)),
+ };
+
+ try b.writeConst(writer, u8, input[0]);
+ try b.writeConst(writer, i8, input[1]);
+ try b.writeConst(writer, u16, input[2]);
+ try b.writeConst(writer, i16, input[3]);
+ try b.writeConst(writer, u32, input[4]);
+ try b.writeConst(writer, i32, input[5]);
+ try b.writeConst(writer, u64, input[6]);
+ try b.writeConst(writer, i64, input[7]);
+ try b.writeConst(writer, u28, input[8]);
+ try b.writeConst(writer, i28, input[9]);
+ try b.writeAddr(writer, input[10]);
+
+ var mock_compile_unit: std.debug.Dwarf.CompileUnit = undefined;
+ mock_compile_unit.addr_base = 1;
+
+ var mock_debug_addr = std.ArrayList(u8).init(allocator);
+ defer mock_debug_addr.deinit();
+
+ try mock_debug_addr.writer().writeInt(u16, 0, native_endian);
+ try mock_debug_addr.writer().writeInt(usize, input[11], native_endian);
+ try mock_debug_addr.writer().writeInt(usize, input[12], native_endian);
+
+ const context = Context{
+ .compile_unit = &mock_compile_unit,
+ .debug_addr = mock_debug_addr.items,
+ };
+
+ try b.writeConstx(writer, @as(usize, 1));
+ try b.writeAddrx(writer, @as(usize, 1 + @sizeOf(usize)));
+
+ const die_offset: usize = @truncate(0xaabbccdd);
+ const type_bytes: []const u8 = &.{ 1, 2, 3, 4 };
+ try b.writeConstType(writer, die_offset, type_bytes);
+
+ _ = try stack_machine.run(program.items, allocator, context, 0);
+
+ const const_type = stack_machine.stack.popOrNull().?.const_type;
+ try testing.expectEqual(die_offset, const_type.type_offset);
+ try testing.expectEqualSlices(u8, type_bytes, const_type.value_bytes);
+
+ const expected = .{
+ .{ usize, input[12], usize },
+ .{ usize, input[11], usize },
+ .{ usize, input[10], usize },
+ .{ isize, input[9], isize },
+ .{ usize, input[8], usize },
+ .{ isize, input[7], isize },
+ .{ usize, input[6], usize },
+ .{ isize, input[5], isize },
+ .{ usize, input[4], usize },
+ .{ isize, input[3], isize },
+ .{ usize, input[2], usize },
+ .{ isize, input[1], isize },
+ .{ usize, input[0], usize },
+ };
+
+ inline for (expected) |e| {
+ try testing.expectEqual(@as(e[0], e[1]), @as(e[2], @bitCast(stack_machine.stack.popOrNull().?.generic)));
+ }
+ }
+
+ // Register values
+ if (@sizeOf(std.debug.ThreadContext) != 0) {
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+
+ const reg_context = abi.RegisterContext{
+ .eh_frame = true,
+ .is_macho = builtin.os.tag == .macos,
+ };
+ var thread_context: std.debug.ThreadContext = undefined;
+ std.debug.relocateContext(&thread_context);
+ const context = Context{
+ .thread_context = &thread_context,
+ .reg_context = reg_context,
+ };
+
+ // Only test register operations on arch / os that have them implemented
+ if (abi.regBytes(&thread_context, 0, reg_context)) |reg_bytes| {
+
+ // TODO: Test fbreg (once implemented): mock a DIE and point compile_unit.frame_base at it
+
+ mem.writeInt(usize, reg_bytes[0..@sizeOf(usize)], 0xee, native_endian);
+ (try abi.regValueNative(usize, &thread_context, abi.fpRegNum(reg_context), reg_context)).* = 1;
+ (try abi.regValueNative(usize, &thread_context, abi.spRegNum(reg_context), reg_context)).* = 2;
+ (try abi.regValueNative(usize, &thread_context, abi.ipRegNum(), reg_context)).* = 3;
+
+ try b.writeBreg(writer, abi.fpRegNum(reg_context), @as(usize, 100));
+ try b.writeBreg(writer, abi.spRegNum(reg_context), @as(usize, 200));
+ try b.writeBregx(writer, abi.ipRegNum(), @as(usize, 300));
+ try b.writeRegvalType(writer, @as(u8, 0), @as(usize, 400));
+
+ _ = try stack_machine.run(program.items, allocator, context, 0);
+
+ const regval_type = stack_machine.stack.popOrNull().?.regval_type;
+ try testing.expectEqual(@as(usize, 400), regval_type.type_offset);
+ try testing.expectEqual(@as(u8, @sizeOf(usize)), regval_type.type_size);
+ try testing.expectEqual(@as(usize, 0xee), regval_type.value);
+
+ try testing.expectEqual(@as(usize, 303), stack_machine.stack.popOrNull().?.generic);
+ try testing.expectEqual(@as(usize, 202), stack_machine.stack.popOrNull().?.generic);
+ try testing.expectEqual(@as(usize, 101), stack_machine.stack.popOrNull().?.generic);
+ } else |err| {
+ switch (err) {
+ error.UnimplementedArch,
+ error.UnimplementedOs,
+ error.ThreadContextNotSupported,
+ => {},
+ else => return err,
+ }
+ }
+ }
+
+ // Stack operations
+ {
+ var context = Context{};
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u8, 1);
+ try b.writeOpcode(writer, OP.dup);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 1), stack_machine.stack.popOrNull().?.generic);
+ try testing.expectEqual(@as(usize, 1), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u8, 1);
+ try b.writeOpcode(writer, OP.drop);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expect(stack_machine.stack.popOrNull() == null);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u8, 4);
+ try b.writeConst(writer, u8, 5);
+ try b.writeConst(writer, u8, 6);
+ try b.writePick(writer, 2);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 4), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u8, 4);
+ try b.writeConst(writer, u8, 5);
+ try b.writeConst(writer, u8, 6);
+ try b.writeOpcode(writer, OP.over);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 5), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u8, 5);
+ try b.writeConst(writer, u8, 6);
+ try b.writeOpcode(writer, OP.swap);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 5), stack_machine.stack.popOrNull().?.generic);
+ try testing.expectEqual(@as(usize, 6), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u8, 4);
+ try b.writeConst(writer, u8, 5);
+ try b.writeConst(writer, u8, 6);
+ try b.writeOpcode(writer, OP.rot);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 5), stack_machine.stack.popOrNull().?.generic);
+ try testing.expectEqual(@as(usize, 4), stack_machine.stack.popOrNull().?.generic);
+ try testing.expectEqual(@as(usize, 6), stack_machine.stack.popOrNull().?.generic);
+
+ const deref_target: usize = @truncate(0xffeeffee_ffeeffee);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeAddr(writer, @intFromPtr(&deref_target));
+ try b.writeOpcode(writer, OP.deref);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(deref_target, stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeLiteral(writer, 0);
+ try b.writeAddr(writer, @intFromPtr(&deref_target));
+ try b.writeOpcode(writer, OP.xderef);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(deref_target, stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeAddr(writer, @intFromPtr(&deref_target));
+ try b.writeDerefSize(writer, 1);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, @as(*const u8, @ptrCast(&deref_target)).*), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeLiteral(writer, 0);
+ try b.writeAddr(writer, @intFromPtr(&deref_target));
+ try b.writeXDerefSize(writer, 1);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, @as(*const u8, @ptrCast(&deref_target)).*), stack_machine.stack.popOrNull().?.generic);
+
+ const type_offset: usize = @truncate(0xaabbaabb_aabbaabb);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeAddr(writer, @intFromPtr(&deref_target));
+ try b.writeDerefType(writer, 1, type_offset);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ const deref_type = stack_machine.stack.popOrNull().?.regval_type;
+ try testing.expectEqual(type_offset, deref_type.type_offset);
+ try testing.expectEqual(@as(u8, 1), deref_type.type_size);
+ try testing.expectEqual(@as(usize, @as(*const u8, @ptrCast(&deref_target)).*), deref_type.value);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeLiteral(writer, 0);
+ try b.writeAddr(writer, @intFromPtr(&deref_target));
+ try b.writeXDerefType(writer, 1, type_offset);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ const xderef_type = stack_machine.stack.popOrNull().?.regval_type;
+ try testing.expectEqual(type_offset, xderef_type.type_offset);
+ try testing.expectEqual(@as(u8, 1), xderef_type.type_size);
+ try testing.expectEqual(@as(usize, @as(*const u8, @ptrCast(&deref_target)).*), xderef_type.value);
+
+ context.object_address = &deref_target;
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeOpcode(writer, OP.push_object_address);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, @intFromPtr(context.object_address.?)), stack_machine.stack.popOrNull().?.generic);
+
+ // TODO: Test OP.form_tls_address
+
+ context.cfa = @truncate(0xccddccdd_ccddccdd);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeOpcode(writer, OP.call_frame_cfa);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(context.cfa.?, stack_machine.stack.popOrNull().?.generic);
+ }
+
+ // Arithmetic and Logical Operations
+ {
+ const context = Context{};
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, i16, -4096);
+ try b.writeOpcode(writer, OP.abs);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 4096), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u16, 0xff0f);
+ try b.writeConst(writer, u16, 0xf0ff);
+ try b.writeOpcode(writer, OP.@"and");
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 0xf00f), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, i16, -404);
+ try b.writeConst(writer, i16, 100);
+ try b.writeOpcode(writer, OP.div);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(isize, -404 / 100), @as(isize, @bitCast(stack_machine.stack.popOrNull().?.generic)));
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u16, 200);
+ try b.writeConst(writer, u16, 50);
+ try b.writeOpcode(writer, OP.minus);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 150), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u16, 123);
+ try b.writeConst(writer, u16, 100);
+ try b.writeOpcode(writer, OP.mod);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 23), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u16, 0xff);
+ try b.writeConst(writer, u16, 0xee);
+ try b.writeOpcode(writer, OP.mul);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 0xed12), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u16, 5);
+ try b.writeOpcode(writer, OP.neg);
+ try b.writeConst(writer, i16, -6);
+ try b.writeOpcode(writer, OP.neg);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 6), stack_machine.stack.popOrNull().?.generic);
+ try testing.expectEqual(@as(isize, -5), @as(isize, @bitCast(stack_machine.stack.popOrNull().?.generic)));
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u16, 0xff0f);
+ try b.writeOpcode(writer, OP.not);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(~@as(usize, 0xff0f), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u16, 0xff0f);
+ try b.writeConst(writer, u16, 0xf0ff);
+ try b.writeOpcode(writer, OP.@"or");
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 0xffff), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, i16, 402);
+ try b.writeConst(writer, i16, 100);
+ try b.writeOpcode(writer, OP.plus);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 502), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u16, 4096);
+ try b.writePlusUconst(writer, @as(usize, 8192));
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 4096 + 8192), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u16, 0xfff);
+ try b.writeConst(writer, u16, 1);
+ try b.writeOpcode(writer, OP.shl);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 0xfff << 1), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u16, 0xfff);
+ try b.writeConst(writer, u16, 1);
+ try b.writeOpcode(writer, OP.shr);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 0xfff >> 1), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u16, 0xfff);
+ try b.writeConst(writer, u16, 1);
+ try b.writeOpcode(writer, OP.shr);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, @bitCast(@as(isize, 0xfff) >> 1)), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConst(writer, u16, 0xf0ff);
+ try b.writeConst(writer, u16, 0xff0f);
+ try b.writeOpcode(writer, OP.xor);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 0x0ff0), stack_machine.stack.popOrNull().?.generic);
+ }
+
+ // Control Flow Operations
+ {
+ const context = Context{};
+ const expected = .{
+ .{ OP.le, 1, 1, 0 },
+ .{ OP.ge, 1, 0, 1 },
+ .{ OP.eq, 1, 0, 0 },
+ .{ OP.lt, 0, 1, 0 },
+ .{ OP.gt, 0, 0, 1 },
+ .{ OP.ne, 0, 1, 1 },
+ };
+
+ inline for (expected) |e| {
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+
+ try b.writeConst(writer, u16, 0);
+ try b.writeConst(writer, u16, 0);
+ try b.writeOpcode(writer, e[0]);
+ try b.writeConst(writer, u16, 0);
+ try b.writeConst(writer, u16, 1);
+ try b.writeOpcode(writer, e[0]);
+ try b.writeConst(writer, u16, 1);
+ try b.writeConst(writer, u16, 0);
+ try b.writeOpcode(writer, e[0]);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, e[3]), stack_machine.stack.popOrNull().?.generic);
+ try testing.expectEqual(@as(usize, e[2]), stack_machine.stack.popOrNull().?.generic);
+ try testing.expectEqual(@as(usize, e[1]), stack_machine.stack.popOrNull().?.generic);
+ }
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeLiteral(writer, 2);
+ try b.writeSkip(writer, 1);
+ try b.writeLiteral(writer, 3);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 2), stack_machine.stack.popOrNull().?.generic);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeLiteral(writer, 2);
+ try b.writeBra(writer, 1);
+ try b.writeLiteral(writer, 3);
+ try b.writeLiteral(writer, 0);
+ try b.writeBra(writer, 1);
+ try b.writeLiteral(writer, 4);
+ try b.writeLiteral(writer, 5);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 5), stack_machine.stack.popOrNull().?.generic);
+ try testing.expectEqual(@as(usize, 4), stack_machine.stack.popOrNull().?.generic);
+ try testing.expect(stack_machine.stack.popOrNull() == null);
+
+ // TODO: Test call2, call4, call_ref once implemented
+
+ }
+
+ // Type conversions
+ {
+ const context = Context{};
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+
+ // TODO: Test typed OP.convert once implemented
+
+ const value: usize = @truncate(0xffeeffee_ffeeffee);
+ var value_bytes: [options.addr_size]u8 = undefined;
+ mem.writeInt(usize, &value_bytes, value, native_endian);
+
+ // Convert to generic type
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConstType(writer, @as(usize, 0), &value_bytes);
+ try b.writeConvert(writer, @as(usize, 0));
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(value, stack_machine.stack.popOrNull().?.generic);
+
+ // Reinterpret to generic type
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConstType(writer, @as(usize, 0), &value_bytes);
+ try b.writeReinterpret(writer, @as(usize, 0));
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(value, stack_machine.stack.popOrNull().?.generic);
+
+ // Reinterpret to new type
+ const die_offset: usize = 0xffee;
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeConstType(writer, @as(usize, 0), &value_bytes);
+ try b.writeReinterpret(writer, die_offset);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ const const_type = stack_machine.stack.popOrNull().?.const_type;
+ try testing.expectEqual(die_offset, const_type.type_offset);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeLiteral(writer, 0);
+ try b.writeReinterpret(writer, die_offset);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ const regval_type = stack_machine.stack.popOrNull().?.regval_type;
+ try testing.expectEqual(die_offset, regval_type.type_offset);
+ }
+
+ // Special operations
+ {
+ var context = Context{};
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeOpcode(writer, OP.nop);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expect(stack_machine.stack.popOrNull() == null);
+
+ // Sub-expression
+ {
+ var sub_program = std.ArrayList(u8).init(allocator);
+ defer sub_program.deinit();
+ const sub_writer = sub_program.writer();
+ try b.writeLiteral(sub_writer, 3);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeEntryValue(writer, sub_program.items);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 3), stack_machine.stack.popOrNull().?.generic);
+ }
+
+ // Register location description
+ const reg_context = abi.RegisterContext{
+ .eh_frame = true,
+ .is_macho = builtin.os.tag == .macos,
+ };
+ var thread_context: std.debug.ThreadContext = undefined;
+ std.debug.relocateContext(&thread_context);
+ context = Context{
+ .thread_context = &thread_context,
+ .reg_context = reg_context,
+ };
+
+ if (abi.regBytes(&thread_context, 0, reg_context)) |reg_bytes| {
+ mem.writeInt(usize, reg_bytes[0..@sizeOf(usize)], 0xee, native_endian);
+
+ var sub_program = std.ArrayList(u8).init(allocator);
+ defer sub_program.deinit();
+ const sub_writer = sub_program.writer();
+ try b.writeReg(sub_writer, 0);
+
+ stack_machine.reset();
+ program.clearRetainingCapacity();
+ try b.writeEntryValue(writer, sub_program.items);
+ _ = try stack_machine.run(program.items, allocator, context, null);
+ try testing.expectEqual(@as(usize, 0xee), stack_machine.stack.popOrNull().?.generic);
+ } else |err| {
+ switch (err) {
+ error.UnimplementedArch,
+ error.UnimplementedOs,
+ error.ThreadContextNotSupported,
+ => {},
+ else => return err,
+ }
+ }
+ }
+}
generated by cgit v1.2.3 (git 2.39.1) at 2025-12-27 21:11:42 +0000