Commit ed7073c630

Jakub Konka <kubkon@jakubkonka.com>
2024-05-24 15:34:30
link/macho: fix perf bug in DWARF parsing
1 parent fb88cfd
src/link/MachO/dwarf.zig
@@ -0,0 +1,285 @@
+pub const InfoReader = struct {
+    bytes: []const u8,
+    strtab: []const u8,
+    pos: usize = 0,
+
+    pub fn readCompileUnitHeader(p: *InfoReader) !CompileUnitHeader {
+        var length: u64 = try p.readInt(u32);
+        const is_64bit = length == 0xffffffff;
+        if (is_64bit) {
+            length = try p.readInt(u64);
+        }
+        const dw_fmt: DwarfFormat = if (is_64bit) .dwarf64 else .dwarf32;
+        return .{
+            .format = dw_fmt,
+            .length = length,
+            .version = try p.readInt(u16),
+            .debug_abbrev_offset = try p.readOffset(dw_fmt),
+            .address_size = try p.readByte(),
+        };
+    }
+
+    pub fn seekToDie(p: *InfoReader, code: Code, cuh: CompileUnitHeader, abbrev_reader: *AbbrevReader) !void {
+        const cuh_length = math.cast(usize, cuh.length) orelse return error.Overflow;
+        const end_pos = p.pos + switch (cuh.format) {
+            .dwarf32 => @as(usize, 4),
+            .dwarf64 => 12,
+        } + cuh_length;
+        while (p.pos < end_pos) {
+            const di_code = try p.readULEB128(u64);
+            if (di_code == 0) return error.Eof;
+            if (di_code == code) return;
+
+            while (try abbrev_reader.readAttr()) |attr| switch (attr.at) {
+                dwarf.FORM.sec_offset,
+                dwarf.FORM.ref_addr,
+                => {
+                    _ = try p.readOffset(cuh.format);
+                },
+
+                dwarf.FORM.addr => {
+                    _ = try p.readNBytes(cuh.address_size);
+                },
+
+                dwarf.FORM.block1,
+                dwarf.FORM.block2,
+                dwarf.FORM.block4,
+                dwarf.FORM.block,
+                => {
+                    _ = try p.readBlock(attr.form);
+                },
+
+                dwarf.FORM.exprloc => {
+                    _ = try p.readExprLoc();
+                },
+
+                dwarf.FORM.flag_present => {},
+
+                dwarf.FORM.data1,
+                dwarf.FORM.ref1,
+                dwarf.FORM.flag,
+                dwarf.FORM.data2,
+                dwarf.FORM.ref2,
+                dwarf.FORM.data4,
+                dwarf.FORM.ref4,
+                dwarf.FORM.data8,
+                dwarf.FORM.ref8,
+                dwarf.FORM.ref_sig8,
+                dwarf.FORM.udata,
+                dwarf.FORM.ref_udata,
+                dwarf.FORM.sdata,
+                => {
+                    _ = try p.readConstant(attr.form);
+                },
+
+                dwarf.FORM.strp,
+                dwarf.FORM.string,
+                => {
+                    _ = try p.readString(attr.form, cuh);
+                },
+
+                else => {
+                    // TODO better errors
+                    log.err("unhandled DW_FORM_* value with identifier {x}", .{attr.form});
+                    return error.UnhandledDwFormValue;
+                },
+            };
+        }
+    }
+
+    pub fn readBlock(p: *InfoReader, form: Form) ![]const u8 {
+        const len: u64 = switch (form) {
+            dwarf.FORM.block1 => try p.readByte(),
+            dwarf.FORM.block2 => try p.readInt(u16),
+            dwarf.FORM.block4 => try p.readInt(u32),
+            dwarf.FORM.block => try p.readULEB128(u64),
+            else => unreachable,
+        };
+        return p.readNBytes(len);
+    }
+
+    pub fn readExprLoc(p: *InfoReader) ![]const u8 {
+        const len: u64 = try p.readULEB128(u64);
+        return p.readNBytes(len);
+    }
+
+    pub fn readConstant(p: *InfoReader, form: Form) !u64 {
+        return switch (form) {
+            dwarf.FORM.data1, dwarf.FORM.ref1, dwarf.FORM.flag => try p.readByte(),
+            dwarf.FORM.data2, dwarf.FORM.ref2 => try p.readInt(u16),
+            dwarf.FORM.data4, dwarf.FORM.ref4 => try p.readInt(u32),
+            dwarf.FORM.data8, dwarf.FORM.ref8, dwarf.FORM.ref_sig8 => try p.readInt(u64),
+            dwarf.FORM.udata, dwarf.FORM.ref_udata => try p.readULEB128(u64),
+            dwarf.FORM.sdata => @bitCast(try p.readILEB128(i64)),
+            else => return error.UnhandledConstantForm,
+        };
+    }
+
+    pub fn readString(p: *InfoReader, form: Form, cuh: CompileUnitHeader) ![:0]const u8 {
+        switch (form) {
+            dwarf.FORM.strp => {
+                const off = try p.readOffset(cuh.format);
+                return mem.sliceTo(@as([*:0]const u8, @ptrCast(p.strtab.ptr + off)), 0);
+            },
+            dwarf.FORM.string => {
+                const start = p.pos;
+                while (p.pos < p.bytes.len) : (p.pos += 1) {
+                    if (p.bytes[p.pos] == 0) break;
+                }
+                if (p.bytes[p.pos] != 0) return error.Eof;
+                return p.bytes[start..p.pos :0];
+            },
+            else => unreachable,
+        }
+    }
+
+    pub fn readByte(p: *InfoReader) !u8 {
+        if (p.pos + 1 > p.bytes.len) return error.Eof;
+        defer p.pos += 1;
+        return p.bytes[p.pos];
+    }
+
+    pub fn readNBytes(p: *InfoReader, num: u64) ![]const u8 {
+        const num_usize = math.cast(usize, num) orelse return error.Overflow;
+        if (p.pos + num_usize > p.bytes.len) return error.Eof;
+        defer p.pos += num_usize;
+        return p.bytes[p.pos..][0..num_usize];
+    }
+
+    pub fn readInt(p: *InfoReader, comptime Int: type) !Int {
+        if (p.pos + @sizeOf(Int) > p.bytes.len) return error.Eof;
+        defer p.pos += @sizeOf(Int);
+        return mem.readInt(Int, p.bytes[p.pos..][0..@sizeOf(Int)], .little);
+    }
+
+    pub fn readOffset(p: *InfoReader, dw_fmt: DwarfFormat) !u64 {
+        return switch (dw_fmt) {
+            .dwarf32 => try p.readInt(u32),
+            .dwarf64 => try p.readInt(u64),
+        };
+    }
+
+    pub fn readULEB128(p: *InfoReader, comptime Type: type) !Type {
+        var stream = std.io.fixedBufferStream(p.bytes[p.pos..]);
+        var creader = std.io.countingReader(stream.reader());
+        const value: Type = try leb.readULEB128(Type, creader.reader());
+        p.pos += creader.bytes_read;
+        return value;
+    }
+
+    pub fn readILEB128(p: *InfoReader, comptime Type: type) !Type {
+        var stream = std.io.fixedBufferStream(p.bytes[p.pos..]);
+        var creader = std.io.countingReader(stream.reader());
+        const value: Type = try leb.readILEB128(Type, creader.reader());
+        p.pos += creader.bytes_read;
+        return value;
+    }
+
+    pub fn seekTo(p: *InfoReader, off: u64) !void {
+        p.pos = math.cast(usize, off) orelse return error.Overflow;
+    }
+};
+
+pub const AbbrevReader = struct {
+    bytes: []const u8,
+    pos: usize = 0,
+
+    pub fn hasMore(p: AbbrevReader) bool {
+        return p.pos < p.bytes.len;
+    }
+
+    pub fn readDecl(p: *AbbrevReader) !?AbbrevDecl {
+        const pos = p.pos;
+        const code = try p.readULEB128(Code);
+        if (code == 0) return null;
+
+        const tag = try p.readULEB128(Tag);
+        const has_children = (try p.readByte()) > 0;
+        return .{
+            .code = code,
+            .pos = pos,
+            .len = p.pos - pos,
+            .tag = tag,
+            .has_children = has_children,
+        };
+    }
+
+    pub fn readAttr(p: *AbbrevReader) !?AbbrevAttr {
+        const pos = p.pos;
+        const at = try p.readULEB128(At);
+        const form = try p.readULEB128(Form);
+        return if (at == 0 and form == 0) null else .{
+            .at = at,
+            .form = form,
+            .pos = pos,
+            .len = p.pos - pos,
+        };
+    }
+
+    pub fn readByte(p: *AbbrevReader) !u8 {
+        if (p.pos + 1 > p.bytes.len) return error.Eof;
+        defer p.pos += 1;
+        return p.bytes[p.pos];
+    }
+
+    pub fn readULEB128(p: *AbbrevReader, comptime Type: type) !Type {
+        var stream = std.io.fixedBufferStream(p.bytes[p.pos..]);
+        var creader = std.io.countingReader(stream.reader());
+        const value: Type = try leb.readULEB128(Type, creader.reader());
+        p.pos += creader.bytes_read;
+        return value;
+    }
+
+    pub fn seekTo(p: *AbbrevReader, off: u64) !void {
+        p.pos = math.cast(usize, off) orelse return error.Overflow;
+    }
+};
+
+const AbbrevDecl = struct {
+    code: Code,
+    pos: usize,
+    len: usize,
+    tag: Tag,
+    has_children: bool,
+};
+
+const AbbrevAttr = struct {
+    at: At,
+    form: Form,
+    pos: usize,
+    len: usize,
+};
+
+const CompileUnitHeader = struct {
+    format: DwarfFormat,
+    length: u64,
+    version: u16,
+    debug_abbrev_offset: u64,
+    address_size: u8,
+};
+
+const Die = struct {
+    pos: usize,
+    len: usize,
+};
+
+const DwarfFormat = enum {
+    dwarf32,
+    dwarf64,
+};
+
+const dwarf = std.dwarf;
+const leb = std.leb;
+const log = std.log.scoped(.link);
+const math = std.math;
+const mem = std.mem;
+const std = @import("std");
+
+const At = u64;
+const Code = u64;
+const Form = u64;
+const Tag = u64;
+
+pub const AT = dwarf.AT;
+pub const FORM = dwarf.FORM;
+pub const TAG = dwarf.TAG;
src/link/MachO/DwarfInfo.zig
@@ -1,490 +0,0 @@
-/// Abbreviation table indexed by offset in the .debug_abbrev bytestream
-abbrev_tables: std.AutoArrayHashMapUnmanaged(u64, AbbrevTable) = .{},
-/// List of compile units as they appear in the .debug_info bytestream
-compile_units: std.ArrayListUnmanaged(CompileUnit) = .{},
-/// Debug info string table
-strtab: std.ArrayListUnmanaged(u8) = .{},
-/// Debug info data
-di_data: std.ArrayListUnmanaged(u8) = .{},
-
-pub fn init(dw: *DwarfInfo, allocator: Allocator, di: DebugInfo) !void {
-    try dw.strtab.ensureTotalCapacityPrecise(allocator, di.debug_str.len);
-    dw.strtab.appendSliceAssumeCapacity(di.debug_str);
-    try dw.parseAbbrevTables(allocator, di);
-    try dw.parseCompileUnits(allocator, di);
-}
-
-pub fn deinit(dw: *DwarfInfo, allocator: Allocator) void {
-    dw.abbrev_tables.deinit(allocator);
-    for (dw.compile_units.items) |*cu| {
-        cu.deinit(allocator);
-    }
-    dw.compile_units.deinit(allocator);
-    dw.strtab.deinit(allocator);
-    dw.di_data.deinit(allocator);
-}
-
-fn appendDiData(dw: *DwarfInfo, allocator: Allocator, values: []const u8) error{OutOfMemory}!u32 {
-    const index: u32 = @intCast(dw.di_data.items.len);
-    try dw.di_data.ensureUnusedCapacity(allocator, values.len);
-    dw.di_data.appendSliceAssumeCapacity(values);
-    return index;
-}
-
-fn getString(dw: DwarfInfo, off: usize) [:0]const u8 {
-    assert(off < dw.strtab.items.len);
-    return mem.sliceTo(@as([*:0]const u8, @ptrCast(dw.strtab.items.ptr + off)), 0);
-}
-
-fn parseAbbrevTables(dw: *DwarfInfo, allocator: Allocator, di: DebugInfo) !void {
-    const tracy = trace(@src());
-    defer tracy.end();
-
-    const debug_abbrev = di.debug_abbrev;
-    var stream = std.io.fixedBufferStream(debug_abbrev);
-    var creader = std.io.countingReader(stream.reader());
-    const reader = creader.reader();
-
-    while (true) {
-        if (creader.bytes_read >= debug_abbrev.len) break;
-
-        try dw.abbrev_tables.ensureUnusedCapacity(allocator, 1);
-        const table_gop = dw.abbrev_tables.getOrPutAssumeCapacity(@intCast(creader.bytes_read));
-        assert(!table_gop.found_existing);
-        const table = table_gop.value_ptr;
-        table.* = .{};
-
-        while (true) {
-            const code = try leb.readULEB128(Code, reader);
-            if (code == 0) break;
-
-            try table.decls.ensureUnusedCapacity(allocator, 1);
-            const decl_gop = table.decls.getOrPutAssumeCapacity(code);
-            assert(!decl_gop.found_existing);
-            const decl = decl_gop.value_ptr;
-            decl.* = .{
-                .code = code,
-                .tag = undefined,
-                .children = false,
-            };
-            decl.tag = try leb.readULEB128(Tag, reader);
-            decl.children = (try reader.readByte()) > 0;
-
-            while (true) {
-                const at = try leb.readULEB128(At, reader);
-                const form = try leb.readULEB128(Form, reader);
-                if (at == 0 and form == 0) break;
-
-                try decl.attrs.ensureUnusedCapacity(allocator, 1);
-                const attr_gop = decl.attrs.getOrPutAssumeCapacity(at);
-                assert(!attr_gop.found_existing);
-                const attr = attr_gop.value_ptr;
-                attr.* = .{
-                    .at = at,
-                    .form = form,
-                };
-            }
-        }
-    }
-}
-
-fn parseCompileUnits(dw: *DwarfInfo, allocator: Allocator, di: DebugInfo) !void {
-    const tracy = trace(@src());
-    defer tracy.end();
-
-    const debug_info = di.debug_info;
-    var stream = std.io.fixedBufferStream(debug_info);
-    var creader = std.io.countingReader(stream.reader());
-    const reader = creader.reader();
-
-    while (true) {
-        if (creader.bytes_read == debug_info.len) break;
-
-        const cu = try dw.compile_units.addOne(allocator);
-        cu.* = .{
-            .header = undefined,
-            .pos = creader.bytes_read,
-        };
-
-        var length: u64 = try reader.readInt(u32, .little);
-        const is_64bit = length == 0xffffffff;
-        if (is_64bit) {
-            length = try reader.readInt(u64, .little);
-        }
-        cu.header.format = if (is_64bit) .dwarf64 else .dwarf32;
-        cu.header.length = length;
-        cu.header.version = try reader.readInt(u16, .little);
-        cu.header.debug_abbrev_offset = try readOffset(cu.header.format, reader);
-        cu.header.address_size = try reader.readInt(u8, .little);
-
-        const table = dw.abbrev_tables.get(cu.header.debug_abbrev_offset).?;
-        try dw.parseDie(allocator, cu, table, di, null, &creader);
-    }
-}
-
-fn parseDie(
-    dw: *DwarfInfo,
-    allocator: Allocator,
-    cu: *CompileUnit,
-    table: AbbrevTable,
-    di: DebugInfo,
-    parent: ?u32,
-    creader: anytype,
-) anyerror!void {
-    const tracy = trace(@src());
-    defer tracy.end();
-
-    while (creader.bytes_read < cu.nextCompileUnitOffset()) {
-        const die = try cu.addDie(allocator);
-        cu.diePtr(die).* = .{ .code = undefined };
-        if (parent) |p| {
-            try cu.diePtr(p).children.append(allocator, die);
-        } else {
-            try cu.children.append(allocator, die);
-        }
-
-        const code = try leb.readULEB128(Code, creader.reader());
-        cu.diePtr(die).code = code;
-
-        if (code == 0) {
-            if (parent == null) continue;
-            return; // Close scope
-        }
-
-        const decl = table.decls.get(code) orelse return error.MalformedDwarf; // TODO better errors
-        const data = di.debug_info;
-        try cu.diePtr(die).values.ensureTotalCapacityPrecise(allocator, decl.attrs.values().len);
-
-        for (decl.attrs.values()) |attr| {
-            const start = std.math.cast(usize, creader.bytes_read) orelse return error.Overflow;
-            try advanceByFormSize(cu, attr.form, creader);
-            const end = std.math.cast(usize, creader.bytes_read) orelse return error.Overflow;
-            const index = try dw.appendDiData(allocator, data[start..end]);
-            cu.diePtr(die).values.appendAssumeCapacity(.{ .index = index, .len = @intCast(end - start) });
-        }
-
-        if (decl.children) {
-            // Open scope
-            try dw.parseDie(allocator, cu, table, di, die, creader);
-        }
-    }
-}
-
-fn advanceByFormSize(cu: *CompileUnit, form: Form, creader: anytype) !void {
-    const tracy = trace(@src());
-    defer tracy.end();
-
-    const reader = creader.reader();
-    switch (form) {
-        dwarf.FORM.strp,
-        dwarf.FORM.sec_offset,
-        dwarf.FORM.ref_addr,
-        => {
-            _ = try readOffset(cu.header.format, reader);
-        },
-
-        dwarf.FORM.addr => try reader.skipBytes(cu.header.address_size, .{}),
-
-        dwarf.FORM.block1,
-        dwarf.FORM.block2,
-        dwarf.FORM.block4,
-        dwarf.FORM.block,
-        => {
-            const len: u64 = switch (form) {
-                dwarf.FORM.block1 => try reader.readInt(u8, .little),
-                dwarf.FORM.block2 => try reader.readInt(u16, .little),
-                dwarf.FORM.block4 => try reader.readInt(u32, .little),
-                dwarf.FORM.block => try leb.readULEB128(u64, reader),
-                else => unreachable,
-            };
-            var i: u64 = 0;
-            while (i < len) : (i += 1) {
-                _ = try reader.readByte();
-            }
-        },
-
-        dwarf.FORM.exprloc => {
-            const len = try leb.readULEB128(u64, reader);
-            var i: u64 = 0;
-            while (i < len) : (i += 1) {
-                _ = try reader.readByte();
-            }
-        },
-        dwarf.FORM.flag_present => {},
-
-        dwarf.FORM.data1,
-        dwarf.FORM.ref1,
-        dwarf.FORM.flag,
-        => try reader.skipBytes(1, .{}),
-
-        dwarf.FORM.data2,
-        dwarf.FORM.ref2,
-        => try reader.skipBytes(2, .{}),
-
-        dwarf.FORM.data4,
-        dwarf.FORM.ref4,
-        => try reader.skipBytes(4, .{}),
-
-        dwarf.FORM.data8,
-        dwarf.FORM.ref8,
-        dwarf.FORM.ref_sig8,
-        => try reader.skipBytes(8, .{}),
-
-        dwarf.FORM.udata,
-        dwarf.FORM.ref_udata,
-        => {
-            _ = try leb.readULEB128(u64, reader);
-        },
-
-        dwarf.FORM.sdata => {
-            _ = try leb.readILEB128(i64, reader);
-        },
-
-        dwarf.FORM.string => {
-            while (true) {
-                const byte = try reader.readByte();
-                if (byte == 0x0) break;
-            }
-        },
-
-        else => {
-            // TODO better errors
-            log.err("unhandled DW_FORM_* value with identifier {x}", .{form});
-            return error.UnhandledDwFormValue;
-        },
-    }
-}
-
-fn readOffset(format: Format, reader: anytype) !u64 {
-    return switch (format) {
-        .dwarf32 => try reader.readInt(u32, .little),
-        .dwarf64 => try reader.readInt(u64, .little),
-    };
-}
-
-pub const AbbrevTable = struct {
-    /// Table of abbreviation declarations indexed by their assigned code value
-    decls: std.AutoArrayHashMapUnmanaged(Code, Decl) = .{},
-
-    pub fn deinit(table: *AbbrevTable, gpa: Allocator) void {
-        for (table.decls.values()) |*decl| {
-            decl.deinit(gpa);
-        }
-        table.decls.deinit(gpa);
-    }
-};
-
-pub const Decl = struct {
-    code: Code,
-    tag: Tag,
-    children: bool,
-
-    /// Table of attributes indexed by their AT value
-    attrs: std.AutoArrayHashMapUnmanaged(At, Attr) = .{},
-
-    pub fn deinit(decl: *Decl, gpa: Allocator) void {
-        decl.attrs.deinit(gpa);
-    }
-};
-
-pub const Attr = struct {
-    at: At,
-    form: Form,
-};
-
-pub const At = u64;
-pub const Code = u64;
-pub const Form = u64;
-pub const Tag = u64;
-
-pub const CompileUnitHeader = struct {
-    format: Format,
-    length: u64,
-    version: u16,
-    debug_abbrev_offset: u64,
-    address_size: u8,
-};
-
-pub const CompileUnit = struct {
-    header: CompileUnitHeader,
-    pos: u64,
-    dies: std.ArrayListUnmanaged(Die) = .{},
-    children: std.ArrayListUnmanaged(Die.Index) = .{},
-
-    pub fn deinit(cu: *CompileUnit, gpa: Allocator) void {
-        for (cu.dies.items) |*die| {
-            die.deinit(gpa);
-        }
-        cu.dies.deinit(gpa);
-        cu.children.deinit(gpa);
-    }
-
-    pub fn addDie(cu: *CompileUnit, gpa: Allocator) !Die.Index {
-        const index = @as(Die.Index, @intCast(cu.dies.items.len));
-        _ = try cu.dies.addOne(gpa);
-        return index;
-    }
-
-    pub fn diePtr(cu: *CompileUnit, index: Die.Index) *Die {
-        return &cu.dies.items[index];
-    }
-
-    pub fn getCompileDir(cu: CompileUnit, ctx: DwarfInfo) error{Overflow}!?[:0]const u8 {
-        assert(cu.dies.items.len > 0);
-        const die = cu.dies.items[0];
-        const res = die.find(dwarf.AT.comp_dir, cu, ctx) orelse return null;
-        return res.getString(cu.header.format, ctx);
-    }
-
-    pub fn getSourceFile(cu: CompileUnit, ctx: DwarfInfo) error{Overflow}!?[:0]const u8 {
-        assert(cu.dies.items.len > 0);
-        const die = cu.dies.items[0];
-        const res = die.find(dwarf.AT.name, cu, ctx) orelse return null;
-        return res.getString(cu.header.format, ctx);
-    }
-
-    pub fn nextCompileUnitOffset(cu: CompileUnit) u64 {
-        return cu.pos + switch (cu.header.format) {
-            .dwarf32 => @as(u64, 4),
-            .dwarf64 => 12,
-        } + cu.header.length;
-    }
-};
-
-pub const Die = struct {
-    code: Code,
-    values: std.ArrayListUnmanaged(struct { index: u32, len: u32 }) = .{},
-    children: std.ArrayListUnmanaged(Die.Index) = .{},
-
-    pub fn deinit(die: *Die, gpa: Allocator) void {
-        die.values.deinit(gpa);
-        die.children.deinit(gpa);
-    }
-
-    pub fn find(die: Die, at: At, cu: CompileUnit, ctx: DwarfInfo) ?DieValue {
-        const table = ctx.abbrev_tables.get(cu.header.debug_abbrev_offset) orelse return null;
-        const decl = table.decls.get(die.code).?;
-        const index = decl.attrs.getIndex(at) orelse return null;
-        const attr = decl.attrs.values()[index];
-        const value = die.values.items[index];
-        return .{ .attr = attr, .bytes = ctx.di_data.items[value.index..][0..value.len] };
-    }
-
-    pub const Index = u32;
-};
-
-pub const DieValue = struct {
-    attr: Attr,
-    bytes: []const u8,
-
-    pub fn getFlag(value: DieValue) ?bool {
-        return switch (value.attr.form) {
-            dwarf.FORM.flag => value.bytes[0] == 1,
-            dwarf.FORM.flag_present => true,
-            else => null,
-        };
-    }
-
-    pub fn getString(value: DieValue, format: Format, ctx: DwarfInfo) error{Overflow}!?[:0]const u8 {
-        switch (value.attr.form) {
-            dwarf.FORM.string => {
-                return mem.sliceTo(@as([*:0]const u8, @ptrCast(value.bytes.ptr)), 0);
-            },
-            dwarf.FORM.strp => {
-                const off = switch (format) {
-                    .dwarf64 => mem.readInt(u64, value.bytes[0..8], .little),
-                    .dwarf32 => mem.readInt(u32, value.bytes[0..4], .little),
-                };
-                const off_u = std.math.cast(usize, off) orelse return error.Overflow;
-                return ctx.getString(off_u);
-            },
-            else => return null,
-        }
-    }
-
-    pub fn getSecOffset(value: DieValue, format: Format) ?u64 {
-        return switch (value.attr.form) {
-            dwarf.FORM.sec_offset => switch (format) {
-                .dwarf32 => mem.readInt(u32, value.bytes[0..4], .little),
-                .dwarf64 => mem.readInt(u64, value.bytes[0..8], .little),
-            },
-            else => null,
-        };
-    }
-
-    pub fn getConstant(value: DieValue) !?i128 {
-        var stream = std.io.fixedBufferStream(value.bytes);
-        const reader = stream.reader();
-        return switch (value.attr.form) {
-            dwarf.FORM.data1 => value.bytes[0],
-            dwarf.FORM.data2 => mem.readInt(u16, value.bytes[0..2], .little),
-            dwarf.FORM.data4 => mem.readInt(u32, value.bytes[0..4], .little),
-            dwarf.FORM.data8 => mem.readInt(u64, value.bytes[0..8], .little),
-            dwarf.FORM.udata => try leb.readULEB128(u64, reader),
-            dwarf.FORM.sdata => try leb.readILEB128(i64, reader),
-            else => null,
-        };
-    }
-
-    pub fn getReference(value: DieValue, format: Format) !?u64 {
-        var stream = std.io.fixedBufferStream(value.bytes);
-        const reader = stream.reader();
-        return switch (value.attr.form) {
-            dwarf.FORM.ref1 => value.bytes[0],
-            dwarf.FORM.ref2 => mem.readInt(u16, value.bytes[0..2], .little),
-            dwarf.FORM.ref4 => mem.readInt(u32, value.bytes[0..4], .little),
-            dwarf.FORM.ref8 => mem.readInt(u64, value.bytes[0..8], .little),
-            dwarf.FORM.ref_udata => try leb.readULEB128(u64, reader),
-            dwarf.FORM.ref_addr => switch (format) {
-                .dwarf32 => mem.readInt(u32, value.bytes[0..4], .little),
-                .dwarf64 => mem.readInt(u64, value.bytes[0..8], .little),
-            },
-            else => null,
-        };
-    }
-
-    pub fn getAddr(value: DieValue, header: CompileUnitHeader) ?u64 {
-        return switch (value.attr.form) {
-            dwarf.FORM.addr => switch (header.address_size) {
-                1 => value.bytes[0],
-                2 => mem.readInt(u16, value.bytes[0..2], .little),
-                4 => mem.readInt(u32, value.bytes[0..4], .little),
-                8 => mem.readInt(u64, value.bytes[0..8], .little),
-                else => null,
-            },
-            else => null,
-        };
-    }
-
-    pub fn getExprloc(value: DieValue) !?[]const u8 {
-        if (value.attr.form != dwarf.FORM.exprloc) return null;
-        var stream = std.io.fixedBufferStream(value.bytes);
-        var creader = std.io.countingReader(stream.reader());
-        const reader = creader.reader();
-        const expr_len = try leb.readULEB128(u64, reader);
-        return value.bytes[creader.bytes_read..][0..expr_len];
-    }
-};
-
-pub const Format = enum {
-    dwarf32,
-    dwarf64,
-};
-
-const DebugInfo = struct {
-    debug_info: []const u8,
-    debug_abbrev: []const u8,
-    debug_str: []const u8,
-};
-
-const assert = std.debug.assert;
-const dwarf = std.dwarf;
-const leb = std.leb;
-const log = std.log.scoped(.link);
-const mem = std.mem;
-const std = @import("std");
-const trace = @import("../../tracy.zig").trace;
-
-const Allocator = mem.Allocator;
-const DwarfInfo = @This();
-const MachO = @import("../MachO.zig");
src/link/MachO/Object.zig
@@ -13,7 +13,7 @@ symbols: std.ArrayListUnmanaged(Symbol.Index) = .{},
 atoms: std.ArrayListUnmanaged(Atom.Index) = .{},
 
 platform: ?MachO.Platform = null,
-dwarf_info: ?DwarfInfo = null,
+compile_unit: ?CompileUnit = null,
 stab_files: std.ArrayListUnmanaged(StabFile) = .{},
 
 eh_frame_sect_index: ?u8 = null,
@@ -31,12 +31,6 @@ dynamic_relocs: MachO.DynamicRelocs = .{},
 output_symtab_ctx: MachO.SymtabCtx = .{},
 output_ar_state: Archive.ArState = .{},
 
-const InArchive = struct {
-    path: []const u8,
-    offset: u64,
-    size: u32,
-};
-
 pub fn isObject(path: []const u8) !bool {
     const file = try std.fs.cwd().openFile(path, .{});
     defer file.close();
@@ -60,7 +54,6 @@ pub fn deinit(self: *Object, allocator: Allocator) void {
     self.fdes.deinit(allocator);
     self.eh_frame_data.deinit(allocator);
     self.unwind_records.deinit(allocator);
-    if (self.dwarf_info) |*dw| dw.deinit(allocator);
     for (self.stab_files.items) |*sf| {
         sf.stabs.deinit(allocator);
     }
@@ -251,8 +244,6 @@ pub fn parse(self: *Object, macho_file: *MachO) !void {
         // }
     }
 
-    try self.initDwarfInfo(macho_file);
-
     for (self.atoms.items) |atom_index| {
         const atom = macho_file.getAtom(atom_index).?;
         const isec = atom.getInputSection(macho_file);
@@ -1214,7 +1205,7 @@ fn parseUnwindRecords(self: *Object, macho_file: *MachO) !void {
 /// and record that so that we can emit symbol stabs.
 /// TODO in the future, we want parse debug info and debug line sections so that
 /// we can provide nice error locations to the user.
-fn initDwarfInfo(self: *Object, macho_file: *MachO) !void {
+pub fn parseDebugInfo(self: *Object, macho_file: *MachO) !void {
     const tracy = trace(@src());
     defer tracy.end();
 
@@ -1240,17 +1231,107 @@ fn initDwarfInfo(self: *Object, macho_file: *MachO) !void {
     const debug_str = if (debug_str_index) |index| try self.getSectionData(@intCast(index), macho_file) else &[0]u8{};
     defer gpa.free(debug_str);
 
-    var dwarf_info = DwarfInfo{};
-    errdefer dwarf_info.deinit(gpa);
-    dwarf_info.init(gpa, .{
+    self.compile_unit = self.findCompileUnit(.{
+        .gpa = gpa,
         .debug_info = debug_info,
         .debug_abbrev = debug_abbrev,
         .debug_str = debug_str,
-    }) catch {
-        try macho_file.reportParseError2(self.index, "invalid __DWARF info found", .{});
-        return error.MalformedObject;
+    }) catch null; // TODO figure out what errors are fatal, and when we silently fail
+}
+
+fn findCompileUnit(self: *Object, args: struct {
+    gpa: Allocator,
+    debug_info: []const u8,
+    debug_abbrev: []const u8,
+    debug_str: []const u8,
+}) !CompileUnit {
+    var cu_wip: struct {
+        comp_dir: ?[:0]const u8 = null,
+        tu_name: ?[:0]const u8 = null,
+    } = .{};
+
+    const gpa = args.gpa;
+    var info_reader = dwarf.InfoReader{ .bytes = args.debug_info, .strtab = args.debug_str };
+    var abbrev_reader = dwarf.AbbrevReader{ .bytes = args.debug_abbrev };
+
+    const cuh = try info_reader.readCompileUnitHeader();
+    try abbrev_reader.seekTo(cuh.debug_abbrev_offset);
+
+    const cu_decl = (try abbrev_reader.readDecl()) orelse return error.Eof;
+    if (cu_decl.tag != dwarf.TAG.compile_unit) return error.UnexpectedTag;
+
+    try info_reader.seekToDie(cu_decl.code, cuh, &abbrev_reader);
+
+    while (try abbrev_reader.readAttr()) |attr| switch (attr.at) {
+        dwarf.AT.name => {
+            cu_wip.tu_name = try info_reader.readString(attr.form, cuh);
+        },
+        dwarf.AT.comp_dir => {
+            cu_wip.comp_dir = try info_reader.readString(attr.form, cuh);
+        },
+        else => switch (attr.form) {
+            dwarf.FORM.sec_offset,
+            dwarf.FORM.ref_addr,
+            => {
+                _ = try info_reader.readOffset(cuh.format);
+            },
+
+            dwarf.FORM.addr => {
+                _ = try info_reader.readNBytes(cuh.address_size);
+            },
+
+            dwarf.FORM.block1,
+            dwarf.FORM.block2,
+            dwarf.FORM.block4,
+            dwarf.FORM.block,
+            => {
+                _ = try info_reader.readBlock(attr.form);
+            },
+
+            dwarf.FORM.exprloc => {
+                _ = try info_reader.readExprLoc();
+            },
+
+            dwarf.FORM.flag_present => {},
+
+            dwarf.FORM.data1,
+            dwarf.FORM.ref1,
+            dwarf.FORM.flag,
+            dwarf.FORM.data2,
+            dwarf.FORM.ref2,
+            dwarf.FORM.data4,
+            dwarf.FORM.ref4,
+            dwarf.FORM.data8,
+            dwarf.FORM.ref8,
+            dwarf.FORM.ref_sig8,
+            dwarf.FORM.udata,
+            dwarf.FORM.ref_udata,
+            dwarf.FORM.sdata,
+            => {
+                _ = try info_reader.readConstant(attr.form);
+            },
+
+            dwarf.FORM.strp,
+            dwarf.FORM.string,
+            => {
+                _ = try info_reader.readString(attr.form, cuh);
+            },
+
+            else => {
+                // TODO actual errors?
+                log.err("unhandled DW_FORM_* value with identifier {x}", .{attr.form});
+                return error.UnhandledForm;
+            },
+        },
+    };
+
+    if (cu_wip.comp_dir == null) return error.MissingCompDir;
+    if (cu_wip.tu_name == null) return error.MissingTuName;
+
+    return .{
+        .comp_dir = try self.addString(gpa, cu_wip.comp_dir.?),
+        .tu_name = try self.addString(gpa, cu_wip.tu_name.?),
     };
-    self.dwarf_info = dwarf_info;
 }
 
 pub fn resolveSymbols(self: *Object, macho_file: *MachO) void {
@@ -1591,10 +1672,9 @@ pub fn calcSymtabSize(self: *Object, macho_file: *MachO) !void {
 }
 
 pub fn calcStabsSize(self: *Object, macho_file: *MachO) error{Overflow}!void {
-    if (self.dwarf_info) |dw| {
-        const cu = dw.compile_units.items[0];
-        const comp_dir = try cu.getCompileDir(dw) orelse return;
-        const tu_name = try cu.getSourceFile(dw) orelse return;
+    if (self.compile_unit) |cu| {
+        const comp_dir = cu.getCompDir(self);
+        const tu_name = cu.getTuName(self);
 
         self.output_symtab_ctx.nstabs += 4; // N_SO, N_SO, N_OSO, N_SO
         self.output_symtab_ctx.strsize += @as(u32, @intCast(comp_dir.len + 1)); // comp_dir
@@ -1709,10 +1789,9 @@ pub fn writeStabs(self: *const Object, macho_file: *MachO, ctx: anytype) error{O
 
     var index = self.output_symtab_ctx.istab;
 
-    if (self.dwarf_info) |dw| {
-        const cu = dw.compile_units.items[0];
-        const comp_dir = try cu.getCompileDir(dw) orelse return;
-        const tu_name = try cu.getSourceFile(dw) orelse return;
+    if (self.compile_unit) |cu| {
+        const comp_dir = cu.getCompDir(self);
+        const tu_name = cu.getTuName(self);
 
         // Open scope
         // N_SO comp_dir
@@ -1958,10 +2037,7 @@ pub fn hasEhFrameRecords(self: Object) bool {
 }
 
 pub fn hasDebugInfo(self: Object) bool {
-    if (self.dwarf_info) |dw| {
-        return dw.compile_units.items.len > 0;
-    }
-    return self.hasSymbolStabs();
+    return self.compile_unit != null or self.hasSymbolStabs();
 }
 
 fn hasSymbolStabs(self: Object) bool {
@@ -2194,6 +2270,25 @@ const StabFile = struct {
     };
 };
 
+const CompileUnit = struct {
+    comp_dir: u32,
+    tu_name: u32,
+
+    fn getCompDir(cu: CompileUnit, object: *const Object) [:0]const u8 {
+        return object.getString(cu.comp_dir);
+    }
+
+    fn getTuName(cu: CompileUnit, object: *const Object) [:0]const u8 {
+        return object.getString(cu.tu_name);
+    }
+};
+
+const InArchive = struct {
+    path: []const u8,
+    offset: u64,
+    size: u32,
+};
+
 const x86_64 = struct {
     fn parseRelocs(
         self: *const Object,
@@ -2548,6 +2643,7 @@ const aarch64 = struct {
 };
 
 const assert = std.debug.assert;
+const dwarf = @import("dwarf.zig");
 const eh_frame = @import("eh_frame.zig");
 const log = std.log.scoped(.link);
 const macho = std.macho;
@@ -2560,7 +2656,6 @@ const Allocator = mem.Allocator;
 const Archive = @import("Archive.zig");
 const Atom = @import("Atom.zig");
 const Cie = eh_frame.Cie;
-const DwarfInfo = @import("DwarfInfo.zig");
 const Fde = eh_frame.Fde;
 const File = @import("file.zig").File;
 const LoadCommandIterator = macho.LoadCommandIterator;
src/link/MachO/relocatable.zig
@@ -46,6 +46,7 @@ pub fn flushObject(macho_file: *MachO, comp: *Compilation, module_obj_path: ?[]c
 
     try macho_file.addUndefinedGlobals();
     try macho_file.resolveSymbols();
+    try macho_file.parseDebugInfo();
     try macho_file.dedupLiterals();
     markExports(macho_file);
     claimUnresolved(macho_file);
src/link/MachO.zig
@@ -535,6 +535,7 @@ pub fn flushModule(self: *MachO, arena: Allocator, prog_node: *std.Progress.Node
 
     try self.addUndefinedGlobals();
     try self.resolveSymbols();
+    try self.parseDebugInfo();
     try self.resolveSyntheticSymbols();
 
     try self.convertTentativeDefinitions();
@@ -1409,6 +1410,12 @@ fn markLive(self: *MachO) void {
     }
 }
 
+pub fn parseDebugInfo(self: *MachO) !void {
+    for (self.objects.items) |index| {
+        try self.getFile(index).?.object.parseDebugInfo(self);
+    }
+}
+
 fn resolveSyntheticSymbols(self: *MachO) !void {
     const internal = self.getInternalObject() orelse return;
 
@@ -4840,7 +4847,6 @@ const Cache = std.Build.Cache;
 const CodeSignature = @import("MachO/CodeSignature.zig");
 const Compilation = @import("../Compilation.zig");
 pub const DebugSymbols = @import("MachO/DebugSymbols.zig");
-const DwarfInfo = @import("MachO/DwarfInfo.zig");
 const Dylib = @import("MachO/Dylib.zig");
 const ExportTrieSection = synthetic.ExportTrieSection;
 const File = @import("MachO/file.zig").File;
CMakeLists.txt
@@ -613,7 +613,6 @@ set(ZIG_STAGE2_SOURCES
     src/link/MachO/Atom.zig
     src/link/MachO/CodeSignature.zig
     src/link/MachO/DebugSymbols.zig
-    src/link/MachO/DwarfInfo.zig
     src/link/MachO/Dylib.zig
     src/link/MachO/InternalObject.zig
     src/link/MachO/Object.zig
@@ -625,6 +624,7 @@ set(ZIG_STAGE2_SOURCES
     src/link/MachO/dyld_info/Rebase.zig
     src/link/MachO/dyld_info/Trie.zig
     src/link/MachO/dyld_info/bind.zig
+    src/link/MachO/dwarf.zig
     src/link/MachO/eh_frame.zig
     src/link/MachO/fat.zig
     src/link/MachO/file.zig