Files
zig/src/link/Wasm/ZigObject.zig
Luuk de Gram 0a030d6598 wasm: Use File.Index for symbol locations
Rather than using the optional, we now directly use `File.Index` which
can already represent an unknown file due to its `.null` value. This
means we do not pay for the memory cost.

This type of index is now used for:
- SymbolLoc
- Key of the functions map
- InitFunc

Now we can simply pass things like atom.file, object.file, loc.file etc
whenever we need to access its representing object file which makes it
a lot easier.
2024-02-29 15:23:03 +01:00

1214 lines
50 KiB
Zig

//! ZigObject encapsulates the state of the incrementally compiled Zig module.
//! It stores the associated input local and global symbols, allocated atoms,
//! and any relocations that may have been emitted.
//! Think about this as fake in-memory Object file for the Zig module.
path: []const u8,
/// Index within the list of relocatable objects of the linker driver.
index: File.Index,
/// List of all `Decl` that are currently alive.
/// Each index maps to the corresponding `Atom.Index`.
decls: std.AutoHashMapUnmanaged(InternPool.DeclIndex, Atom.Index) = .{},
/// List of function type signatures for this Zig module.
func_types: std.ArrayListUnmanaged(std.wasm.Type) = .{},
/// List of `std.wasm.Func`. Each entry contains the function signature,
/// rather than the actual body.
functions: std.ArrayListUnmanaged(std.wasm.Func) = .{},
/// Map of symbol locations, represented by its `types.Import`.
imports: std.AutoHashMapUnmanaged(u32, types.Import) = .{},
/// List of WebAssembly globals.
globals: std.ArrayListUnmanaged(std.wasm.Global) = .{},
/// Mapping between an `Atom` and its type index representing the Wasm
/// type of the function signature.
atom_types: std.AutoHashMapUnmanaged(Atom.Index, u32) = .{},
/// List of all symbols generated by Zig code.
symbols: std.ArrayListUnmanaged(Symbol) = .{},
/// Map from symbol name offset to their index into the `symbols` list.
global_syms: std.AutoHashMapUnmanaged(u32, u32) = .{},
/// List of symbol indexes which are free to be used.
symbols_free_list: std.ArrayListUnmanaged(u32) = .{},
/// Extra metadata about the linking section, such as alignment of segments and their name.
segment_info: std.ArrayListUnmanaged(types.Segment) = .{},
/// File encapsulated string table, used to deduplicate strings within the generated file.
string_table: StringTable = .{},
/// Map for storing anonymous declarations. Each anonymous decl maps to its Atom's index.
anon_decls: std.AutoArrayHashMapUnmanaged(InternPool.Index, Atom.Index) = .{},
/// Represents the symbol index of the error name table
/// When this is `null`, no code references an error using runtime `@errorName`.
/// During initializion, a symbol with corresponding atom will be created that is
/// used to perform relocations to the pointer of this table.
/// The actual table is populated during `flush`.
error_table_symbol: ?u32 = null,
/// Amount of functions in the `import` sections.
imported_functions_count: u32 = 0,
/// Amount of globals in the `import` section.
imported_globals_count: u32 = 0,
/// Symbol index representing the stack pointer. This will be set upon initializion
/// of a new `ZigObject`. Codegen will make calls into this to create relocations for
/// this symbol each time the stack pointer is moved.
stack_pointer_sym: u32,
/// Debug information for the Zig module.
dwarf: ?Dwarf = null,
// Debug section atoms. These are only set when the current compilation
// unit contains Zig code. The lifetime of these atoms are extended
// until the end of the compiler's lifetime. Meaning they're not freed
// during `flush()` in incremental-mode.
debug_info_atom: ?Atom.Index = null,
debug_line_atom: ?Atom.Index = null,
debug_loc_atom: ?Atom.Index = null,
debug_ranges_atom: ?Atom.Index = null,
debug_abbrev_atom: ?Atom.Index = null,
debug_str_atom: ?Atom.Index = null,
debug_pubnames_atom: ?Atom.Index = null,
debug_pubtypes_atom: ?Atom.Index = null,
/// The index of the segment representing the custom '.debug_info' section.
debug_info_index: ?u32 = null,
/// The index of the segment representing the custom '.debug_line' section.
debug_line_index: ?u32 = null,
/// The index of the segment representing the custom '.debug_loc' section.
debug_loc_index: ?u32 = null,
/// The index of the segment representing the custom '.debug_ranges' section.
debug_ranges_index: ?u32 = null,
/// The index of the segment representing the custom '.debug_pubnames' section.
debug_pubnames_index: ?u32 = null,
/// The index of the segment representing the custom '.debug_pubtypes' section.
debug_pubtypes_index: ?u32 = null,
/// The index of the segment representing the custom '.debug_pubtypes' section.
debug_str_index: ?u32 = null,
/// The index of the segment representing the custom '.debug_pubtypes' section.
debug_abbrev_index: ?u32 = null,
/// Initializes the `ZigObject` with initial symbols.
pub fn init(zig_object: *ZigObject, wasm_file: *Wasm) !void {
// Initialize an undefined global with the name __stack_pointer. Codegen will use
// this to generate relocations when moving the stack pointer. This symbol will be
// resolved automatically by the final linking stage.
try zig_object.createStackPointer(wasm_file);
// TODO: Initialize debug information when we reimplement Dwarf support.
}
fn createStackPointer(zig_object: *ZigObject, wasm_file: *Wasm) !void {
const gpa = wasm_file.base.comp.gpa;
const sym_index = try zig_object.getGlobalSymbol(gpa, "__stack_pointer");
const sym = zig_object.symbol(sym_index);
sym.index = zig_object.imported_globals_count;
sym.tag = .global;
const is_wasm32 = wasm_file.base.comp.root_mod.resolved_target.result.cpu.arch == .wasm32;
try zig_object.imports.putNoClobber(gpa, sym_index, .{
.name = sym.name,
.module_name = try zig_object.string_table.insert(gpa, wasm_file.host_name),
.kind = .{ .global = .{ .valtype = if (is_wasm32) .i32 else .i64, .mutable = true } },
});
zig_object.imported_globals_count += 1;
zig_object.stack_pointer_sym = sym_index;
}
fn symbol(zig_object: *const ZigObject, index: u32) *Symbol {
return &zig_object.symbols.items[index];
}
/// Frees and invalidates all memory of the incrementally compiled Zig module.
/// It is illegal behavior to access the `ZigObject` after calling `deinit`.
pub fn deinit(zig_object: *ZigObject, wasm_file: *Wasm) void {
const gpa = wasm_file.base.comp.gpa;
for (zig_object.segment_info.items) |segment_info| {
gpa.free(segment_info.name);
}
// For decls and anon decls we free the memory of its atoms.
// The memory of atoms parsed from object files is managed by
// the object file itself, and therefore we can skip those.
{
var it = zig_object.decls.valueIterator();
while (it.next()) |atom_index_ptr| {
const atom = wasm_file.getAtomPtr(atom_index_ptr.*);
for (atom.locals.items) |local_index| {
const local_atom = wasm_file.getAtomPtr(local_index);
local_atom.deinit(gpa);
}
atom.deinit(gpa);
}
}
{
for (zig_object.anon_decls.values()) |atom_index| {
const atom = wasm_file.getAtomPtr(atom_index);
for (atom.locals.items) |local_index| {
const local_atom = wasm_file.getAtomPtr(local_index);
local_atom.deinit(gpa);
}
atom.deinit(gpa);
}
}
zig_object.decls.deinit(gpa);
zig_object.anon_decls.deinit(gpa);
zig_object.symbols.deinit(gpa);
zig_object.symbols_free_list.deinit(gpa);
zig_object.segment_info.deinit(gpa);
zig_object.string_table.deinit(gpa);
if (zig_object.dwarf) |*dwarf| {
dwarf.deinit();
}
gpa.free(zig_object.path);
zig_object.* = undefined;
}
/// Allocates a new symbol and returns its index.
/// Will re-use slots when a symbol was freed at an earlier stage.
pub fn allocateSymbol(zig_object: *ZigObject, gpa: std.mem.Allocator) !u32 {
try zig_object.symbols.ensureUnusedCapacity(gpa, 1);
const sym: Symbol = .{
.name = std.math.maxInt(u32), // will be set after updateDecl as well as during atom creation for decls
.flags = @intFromEnum(Symbol.Flag.WASM_SYM_BINDING_LOCAL),
.tag = .undefined, // will be set after updateDecl
.index = std.math.maxInt(u32), // will be set during atom parsing
.virtual_address = std.math.maxInt(u32), // will be set during atom allocation
};
if (zig_object.symbols_free_list.popOrNull()) |index| {
zig_object.symbols.items[index] = sym;
return index;
}
const index = @as(u32, @intCast(zig_object.symbols.items.len));
zig_object.symbols.appendAssumeCapacity(sym);
return index;
}
// Generate code for the Decl, storing it in memory to be later written to
// the file on flush().
pub fn updateDecl(
zig_object: *ZigObject,
wasm_file: *Wasm,
mod: *Module,
decl_index: InternPool.DeclIndex,
) !void {
const decl = mod.declPtr(decl_index);
if (decl.val.getFunction(mod)) |_| {
return;
} else if (decl.val.getExternFunc(mod)) |_| {
return;
}
const gpa = wasm_file.base.comp.gpa;
const atom_index = try zig_object.getOrCreateAtomForDecl(wasm_file, decl_index);
const atom = wasm_file.getAtomPtr(atom_index);
atom.clear();
if (decl.isExtern(mod)) {
const variable = decl.getOwnedVariable(mod).?;
const name = mod.intern_pool.stringToSlice(decl.name);
const lib_name = mod.intern_pool.stringToSliceUnwrap(variable.lib_name);
return zig_object.addOrUpdateImport(wasm_file, name, atom.sym_index, lib_name, null);
}
const val = if (decl.val.getVariable(mod)) |variable| Value.fromInterned(variable.init) else decl.val;
var code_writer = std.ArrayList(u8).init(gpa);
defer code_writer.deinit();
const res = try codegen.generateSymbol(
&wasm_file.base,
decl.srcLoc(mod),
.{ .ty = decl.ty, .val = val },
&code_writer,
.none,
.{ .parent_atom_index = atom.sym_index },
);
const code = switch (res) {
.ok => code_writer.items,
.fail => |em| {
decl.analysis = .codegen_failure;
try mod.failed_decls.put(mod.gpa, decl_index, em);
return;
},
};
return zig_object.finishUpdateDecl(wasm_file, decl_index, code, .data);
}
pub fn updateFunc(
zig_object: *ZigObject,
wasm_file: *Wasm,
mod: *Module,
func_index: InternPool.Index,
air: Air,
liveness: Liveness,
) !void {
const gpa = wasm_file.base.comp.gpa;
const func = mod.funcInfo(func_index);
const decl_index = func.owner_decl;
const decl = mod.declPtr(decl_index);
const atom_index = try zig_object.getOrCreateAtomForDecl(wasm_file, decl_index);
const atom = wasm_file.getAtomPtr(atom_index);
atom.clear();
var code_writer = std.ArrayList(u8).init(gpa);
defer code_writer.deinit();
const result = try codegen.generateFunction(
&wasm_file.base,
decl.srcLoc(mod),
func_index,
air,
liveness,
&code_writer,
.none,
);
const code = switch (result) {
.ok => code_writer.items,
.fail => |em| {
decl.analysis = .codegen_failure;
try mod.failed_decls.put(mod.gpa, decl_index, em);
return;
},
};
return zig_object.finishUpdateDecl(wasm_file, decl_index, code, .function);
}
fn finishUpdateDecl(
zig_object: *ZigObject,
wasm_file: *Wasm,
decl_index: InternPool.DeclIndex,
code: []const u8,
symbol_tag: Symbol.Tag,
) !void {
const gpa = wasm_file.base.comp.gpa;
const mod = wasm_file.base.comp.module.?;
const decl = mod.declPtr(decl_index);
const atom_index = zig_object.decls.get(decl_index).?;
const atom = wasm_file.getAtomPtr(atom_index);
const sym = zig_object.symbol(atom.getSymbolIndex().?);
const full_name = mod.intern_pool.stringToSlice(try decl.getFullyQualifiedName(mod));
sym.name = try zig_object.string_table.insert(gpa, full_name);
sym.tag = symbol_tag;
try atom.code.appendSlice(gpa, code);
try wasm_file.resolved_symbols.put(gpa, atom.symbolLoc(), {});
atom.size = @intCast(code.len);
if (code.len == 0) return;
atom.alignment = decl.getAlignment(mod);
}
/// For a given `InternPool.DeclIndex` returns its corresponding `Atom.Index`.
/// When the index was not found, a new `Atom` will be created, and its index will be returned.
/// The newly created Atom is empty with default fields as specified by `Atom.empty`.
pub fn getOrCreateAtomForDecl(zig_object: *ZigObject, wasm_file: *Wasm, decl_index: InternPool.DeclIndex) !Atom.Index {
const gpa = wasm_file.base.comp.gpa;
const gop = try zig_object.decls.getOrPut(gpa, decl_index);
if (!gop.found_existing) {
const sym_index = try zig_object.allocateSymbol(gpa);
gop.value_ptr.* = try wasm_file.createAtom(sym_index, zig_object.index);
const mod = wasm_file.base.comp.module.?;
const decl = mod.declPtr(decl_index);
const full_name = mod.intern_pool.stringToSlice(try decl.getFullyQualifiedName(mod));
const sym = zig_object.symbol(sym_index);
sym.name = try zig_object.string_table.insert(gpa, full_name);
}
return gop.value_ptr.*;
}
pub fn lowerAnonDecl(
zig_object: *ZigObject,
wasm_file: *Wasm,
decl_val: InternPool.Index,
explicit_alignment: InternPool.Alignment,
src_loc: Module.SrcLoc,
) !codegen.Result {
const gpa = wasm_file.base.comp.gpa;
const gop = try zig_object.anon_decls.getOrPut(gpa, decl_val);
if (!gop.found_existing) {
const mod = wasm_file.base.comp.module.?;
const ty = Type.fromInterned(mod.intern_pool.typeOf(decl_val));
const tv: TypedValue = .{ .ty = ty, .val = Value.fromInterned(decl_val) };
var name_buf: [32]u8 = undefined;
const name = std.fmt.bufPrint(&name_buf, "__anon_{d}", .{
@intFromEnum(decl_val),
}) catch unreachable;
switch (try zig_object.lowerConst(wasm_file, name, tv, src_loc)) {
.ok => |atom_index| zig_object.anon_decls.values()[gop.index] = atom_index,
.fail => |em| return .{ .fail = em },
}
}
const atom = wasm_file.getAtomPtr(zig_object.anon_decls.values()[gop.index]);
atom.alignment = switch (atom.alignment) {
.none => explicit_alignment,
else => switch (explicit_alignment) {
.none => atom.alignment,
else => atom.alignment.maxStrict(explicit_alignment),
},
};
return .ok;
}
/// Lowers a constant typed value to a local symbol and atom.
/// Returns the symbol index of the local
/// The given `decl` is the parent decl whom owns the constant.
pub fn lowerUnnamedConst(zig_object: *ZigObject, wasm_file: *Wasm, tv: TypedValue, decl_index: InternPool.DeclIndex) !u32 {
const gpa = wasm_file.base.comp.gpa;
const mod = wasm_file.base.comp.module.?;
std.debug.assert(tv.ty.zigTypeTag(mod) != .Fn); // cannot create local symbols for functions
const decl = mod.declPtr(decl_index);
const parent_atom_index = try zig_object.getOrCreateAtomForDecl(wasm_file, decl_index);
const parent_atom = wasm_file.getAtom(parent_atom_index);
const local_index = parent_atom.locals.items.len;
const fqn = mod.intern_pool.stringToSlice(try decl.getFullyQualifiedName(mod));
const name = try std.fmt.allocPrintZ(gpa, "__unnamed_{s}_{d}", .{
fqn, local_index,
});
defer gpa.free(name);
switch (try zig_object.lowerConst(wasm_file, name, tv, decl.srcLoc(mod))) {
.ok => |atom_index| {
try wasm_file.getAtomPtr(parent_atom_index).locals.append(gpa, atom_index);
return wasm_file.getAtom(atom_index).getSymbolIndex().?;
},
.fail => |em| {
decl.analysis = .codegen_failure;
try mod.failed_decls.put(mod.gpa, decl_index, em);
return error.CodegenFail;
},
}
}
const LowerConstResult = union(enum) {
ok: Atom.Index,
fail: *Module.ErrorMsg,
};
fn lowerConst(zig_object: *ZigObject, wasm_file: *Wasm, name: []const u8, tv: TypedValue, src_loc: Module.SrcLoc) !LowerConstResult {
const gpa = wasm_file.base.comp.gpa;
const mod = wasm_file.base.comp.module.?;
// Create and initialize a new local symbol and atom
const sym_index = try zig_object.allocateSymbol(gpa);
const atom_index = try wasm_file.createAtom(sym_index, zig_object.index);
var value_bytes = std.ArrayList(u8).init(gpa);
defer value_bytes.deinit();
const code = code: {
const atom = wasm_file.getAtomPtr(atom_index);
atom.alignment = tv.ty.abiAlignment(mod);
zig_object.symbols.items[sym_index] = .{
.name = try zig_object.string_table.insert(gpa, name),
.flags = @intFromEnum(Symbol.Flag.WASM_SYM_BINDING_LOCAL),
.tag = .data,
.index = undefined,
.virtual_address = undefined,
};
const result = try codegen.generateSymbol(
&wasm_file.base,
src_loc,
tv,
&value_bytes,
.none,
.{
.parent_atom_index = atom.sym_index,
.addend = null,
},
);
break :code switch (result) {
.ok => value_bytes.items,
.fail => |em| {
return .{ .fail = em };
},
};
};
const atom = wasm_file.getAtomPtr(atom_index);
atom.size = @intCast(code.len);
try atom.code.appendSlice(gpa, code);
return .{ .ok = atom_index };
}
/// Returns the symbol index of the error name table.
///
/// When the symbol does not yet exist, it will create a new one instead.
pub fn getErrorTableSymbol(zig_object: *ZigObject, wasm_file: *Wasm) !u32 {
if (zig_object.error_table_symbol) |sym| {
return sym;
}
// no error was referenced yet, so create a new symbol and atom for it
// and then return said symbol's index. The final table will be populated
// during `flush` when we know all possible error names.
const gpa = wasm_file.base.comp.gpa;
const sym_index = try zig_object.allocateSymbol(gpa);
const atom_index = try wasm_file.createAtom(sym_index, zig_object.index);
const atom = wasm_file.getAtomPtr(atom_index);
const slice_ty = Type.slice_const_u8_sentinel_0;
const mod = wasm_file.base.comp.module.?;
atom.alignment = slice_ty.abiAlignment(mod);
const sym_name = try zig_object.string_table.insert(gpa, "__zig_err_name_table");
const sym = zig_object.symbol(sym_index);
sym.* = .{
.name = sym_name,
.tag = .data,
.flags = @intFromEnum(Symbol.Flag.WASM_SYM_BINDING_LOCAL),
.index = 0,
.virtual_address = undefined,
};
// TODO: can we remove this?
// sym.mark();
log.debug("Error name table was created with symbol index: ({d})", .{sym_index});
zig_object.error_table_symbol = sym_index;
return sym_index;
}
/// Populates the error name table, when `error_table_symbol` is not null.
///
/// This creates a table that consists of pointers and length to each error name.
/// The table is what is being pointed to within the runtime bodies that are generated.
fn populateErrorNameTable(zig_object: *ZigObject, wasm_file: *Wasm) !void {
const symbol_index = zig_object.error_table_symbol orelse return;
const gpa = wasm_file.base.comp.gpa;
const atom_index = wasm_file.symbol_atom.get(.{ .file = zig_object.index, .index = symbol_index }).?;
// Rather than creating a symbol for each individual error name,
// we create a symbol for the entire region of error names. We then calculate
// the pointers into the list using addends which are appended to the relocation.
const names_sym_index = try zig_object.allocateSymbol(gpa);
const names_atom_index = try wasm_file.createAtom(names_sym_index, zig_object.index);
const names_atom = wasm_file.getAtomPtr(names_atom_index);
names_atom.alignment = .@"1";
const sym_name = try zig_object.string_table.insert(gpa, "__zig_err_names");
const names_symbol = &zig_object.symbols.items[names_sym_index];
names_symbol.* = .{
.name = sym_name,
.tag = .data,
.flags = @intFromEnum(Symbol.Flag.WASM_SYM_BINDING_LOCAL),
.index = 0,
.virtual_address = undefined,
};
names_symbol.mark();
log.debug("Populating error names", .{});
// Addend for each relocation to the table
var addend: u32 = 0;
const mod = wasm_file.base.comp.module.?;
for (mod.global_error_set.keys()) |error_name_nts| {
const atom = wasm_file.getAtomPtr(atom_index);
const error_name = mod.intern_pool.stringToSlice(error_name_nts);
const len = @as(u32, @intCast(error_name.len + 1)); // names are 0-termianted
const slice_ty = Type.slice_const_u8_sentinel_0;
const offset = @as(u32, @intCast(atom.code.items.len));
// first we create the data for the slice of the name
try atom.code.appendNTimes(gpa, 0, 4); // ptr to name, will be relocated
try atom.code.writer(gpa).writeInt(u32, len - 1, .little);
// create relocation to the error name
try atom.relocs.append(gpa, .{
.index = names_atom.sym_index,
.relocation_type = .R_WASM_MEMORY_ADDR_I32,
.offset = offset,
.addend = @as(i32, @intCast(addend)),
});
atom.size += @as(u32, @intCast(slice_ty.abiSize(mod)));
addend += len;
// as we updated the error name table, we now store the actual name within the names atom
try names_atom.code.ensureUnusedCapacity(gpa, len);
names_atom.code.appendSliceAssumeCapacity(error_name);
names_atom.code.appendAssumeCapacity(0);
log.debug("Populated error name: '{s}'", .{error_name});
}
names_atom.size = addend;
// link the atoms with the rest of the binary so they can be allocated
// and relocations will be performed.
try wasm_file.parseAtom(atom_index, .{ .data = .read_only });
try wasm_file.parseAtom(names_atom_index, .{ .data = .read_only });
}
/// Either creates a new import, or updates one if existing.
/// When `type_index` is non-null, we assume an external function.
/// In all other cases, a data-symbol will be created instead.
pub fn addOrUpdateImport(
zig_object: *ZigObject,
wasm_file: *Wasm,
/// Name of the import
name: []const u8,
/// Symbol index that is external
symbol_index: u32,
/// Optional library name (i.e. `extern "c" fn foo() void`
lib_name: ?[:0]const u8,
/// The index of the type that represents the function signature
/// when the extern is a function. When this is null, a data-symbol
/// is asserted instead.
type_index: ?u32,
) !void {
const gpa = wasm_file.base.comp.gpa;
std.debug.assert(symbol_index != 0);
// For the import name, we use the decl's name, rather than the fully qualified name
// Also mangle the name when the lib name is set and not equal to "C" so imports with the same
// name but different module can be resolved correctly.
const mangle_name = lib_name != null and
!std.mem.eql(u8, lib_name.?, "c");
const full_name = if (mangle_name) full_name: {
break :full_name try std.fmt.allocPrint(gpa, "{s}|{s}", .{ name, lib_name.? });
} else name;
defer if (mangle_name) gpa.free(full_name);
const decl_name_index = try zig_object.string_table.insert(gpa, full_name);
const sym: *Symbol = &zig_object.symbols.items[symbol_index];
sym.setUndefined(true);
sym.setGlobal(true);
sym.name = decl_name_index;
if (mangle_name) {
// we specified a specific name for the symbol that does not match the import name
sym.setFlag(.WASM_SYM_EXPLICIT_NAME);
}
if (type_index) |ty_index| {
const gop = try zig_object.imports.getOrPut(gpa, symbol_index);
const module_name = if (lib_name) |l_name| blk: {
break :blk l_name;
} else wasm_file.host_name;
if (!gop.found_existing) {
gop.value_ptr.* = .{
.module_name = try zig_object.string_table.insert(gpa, module_name),
.name = try zig_object.string_table.insert(gpa, name),
.kind = .{ .function = ty_index },
};
zig_object.imported_functions_count += 1;
}
}
}
/// Returns the symbol index from a symbol of which its flag is set global,
/// such as an exported or imported symbol.
/// If the symbol does not yet exist, creates a new one symbol instead
/// and then returns the index to it.
pub fn getGlobalSymbol(zig_object: *ZigObject, gpa: std.mem.Allocator, name: []const u8) !u32 {
const name_index = try zig_object.string_table.insert(gpa, name);
const gop = try zig_object.global_syms.getOrPut(gpa, name_index);
if (gop.found_existing) {
return gop.value_ptr.*;
}
var sym: Symbol = .{
.name = name_index,
.flags = 0,
.index = undefined, // index to type will be set after merging symbols
.tag = .function,
.virtual_address = std.math.maxInt(u32),
};
sym.setGlobal(true);
sym.setUndefined(true);
const sym_index = if (zig_object.symbols_free_list.popOrNull()) |index| index else blk: {
const index: u32 = @intCast(zig_object.symbols.items.len);
try zig_object.symbols.ensureUnusedCapacity(gpa, 1);
zig_object.symbols.items.len += 1;
break :blk index;
};
zig_object.symbols.items[sym_index] = sym;
gop.value_ptr.* = sym_index;
return sym_index;
}
/// For a given decl, find the given symbol index's atom, and create a relocation for the type.
/// Returns the given pointer address
pub fn getDeclVAddr(
zig_object: *ZigObject,
wasm_file: *Wasm,
decl_index: InternPool.DeclIndex,
reloc_info: link.File.RelocInfo,
) !u64 {
const target = wasm_file.base.comp.root_mod.resolved_target.result;
const gpa = wasm_file.base.comp.gpa;
const mod = wasm_file.base.comp.module.?;
const decl = mod.declPtr(decl_index);
const target_atom_index = try zig_object.getOrCreateAtomForDecl(wasm_file, decl_index);
const target_symbol_index = wasm_file.getAtom(target_atom_index).sym_index;
std.debug.assert(reloc_info.parent_atom_index != 0);
const atom_index = wasm_file.symbol_atom.get(.{ .file = zig_object.index, .index = reloc_info.parent_atom_index }).?;
const atom = wasm_file.getAtomPtr(atom_index);
const is_wasm32 = target.cpu.arch == .wasm32;
if (decl.ty.zigTypeTag(mod) == .Fn) {
std.debug.assert(reloc_info.addend == 0); // addend not allowed for function relocations
try atom.relocs.append(gpa, .{
.index = target_symbol_index,
.offset = @intCast(reloc_info.offset),
.relocation_type = if (is_wasm32) .R_WASM_TABLE_INDEX_I32 else .R_WASM_TABLE_INDEX_I64,
});
} else {
try atom.relocs.append(gpa, .{
.index = target_symbol_index,
.offset = @intCast(reloc_info.offset),
.relocation_type = if (is_wasm32) .R_WASM_MEMORY_ADDR_I32 else .R_WASM_MEMORY_ADDR_I64,
.addend = @intCast(reloc_info.addend),
});
}
// we do not know the final address at this point,
// as atom allocation will determine the address and relocations
// will calculate and rewrite this. Therefore, we simply return the symbol index
// that was targeted.
return target_symbol_index;
}
pub fn getAnonDeclVAddr(
zig_object: *ZigObject,
wasm_file: *Wasm,
decl_val: InternPool.Index,
reloc_info: link.File.RelocInfo,
) !u64 {
const gpa = wasm_file.base.comp.gpa;
const target = wasm_file.base.comp.root_mod.resolved_target.result;
const atom_index = zig_object.anon_decls.get(decl_val).?;
const target_symbol_index = wasm_file.getAtom(atom_index).getSymbolIndex().?;
const parent_atom_index = wasm_file.symbol_atom.get(.{ .file = zig_object.index, .index = reloc_info.parent_atom_index }).?;
const parent_atom = wasm_file.getAtomPtr(parent_atom_index);
const is_wasm32 = target.cpu.arch == .wasm32;
const mod = wasm_file.base.comp.module.?;
const ty = Type.fromInterned(mod.intern_pool.typeOf(decl_val));
if (ty.zigTypeTag(mod) == .Fn) {
std.debug.assert(reloc_info.addend == 0); // addend not allowed for function relocations
try parent_atom.relocs.append(gpa, .{
.index = target_symbol_index,
.offset = @intCast(reloc_info.offset),
.relocation_type = if (is_wasm32) .R_WASM_TABLE_INDEX_I32 else .R_WASM_TABLE_INDEX_I64,
});
} else {
try parent_atom.relocs.append(gpa, .{
.index = target_symbol_index,
.offset = @intCast(reloc_info.offset),
.relocation_type = if (is_wasm32) .R_WASM_MEMORY_ADDR_I32 else .R_WASM_MEMORY_ADDR_I64,
.addend = @intCast(reloc_info.addend),
});
}
// we do not know the final address at this point,
// as atom allocation will determine the address and relocations
// will calculate and rewrite this. Therefore, we simply return the symbol index
// that was targeted.
return target_symbol_index;
}
pub fn deleteDeclExport(
zig_object: *ZigObject,
wasm_file: *Wasm,
decl_index: InternPool.DeclIndex,
) void {
const atom_index = zig_object.decls.get(decl_index) orelse return;
const sym_index = wasm_file.getAtom(atom_index).sym_index;
const loc: Wasm.SymbolLoc = .{ .file = zig_object.index, .index = sym_index };
const sym = loc.getSymbol(wasm_file);
std.debug.assert(zig_object.global_syms.remove(sym.name));
}
pub fn updateExports(
zig_object: *ZigObject,
wasm_file: *Wasm,
mod: *Module,
exported: Module.Exported,
exports: []const *Module.Export,
) !void {
const decl_index = switch (exported) {
.decl_index => |i| i,
.value => |val| {
_ = val;
@panic("TODO: implement Wasm linker code for exporting a constant value");
},
};
const decl = mod.declPtr(decl_index);
const atom_index = try zig_object.getOrCreateAtomForDecl(wasm_file, decl_index);
const atom = wasm_file.getAtom(atom_index);
const atom_sym = atom.symbolLoc().getSymbol(wasm_file).*;
const gpa = mod.gpa;
for (exports) |exp| {
if (mod.intern_pool.stringToSliceUnwrap(exp.opts.section)) |section| {
try mod.failed_exports.putNoClobber(gpa, exp, try Module.ErrorMsg.create(
gpa,
decl.srcLoc(mod),
"Unimplemented: ExportOptions.section '{s}'",
.{section},
));
continue;
}
const exported_decl_index = switch (exp.exported) {
.value => {
try mod.failed_exports.putNoClobber(gpa, exp, try Module.ErrorMsg.create(
gpa,
decl.srcLoc(mod),
"Unimplemented: exporting a named constant value",
.{},
));
continue;
},
.decl_index => |i| i,
};
const exported_atom_index = try zig_object.getOrCreateAtomForDecl(wasm_file, exported_decl_index);
const exported_atom = wasm_file.getAtom(exported_atom_index);
// const export_name = try zig_object.string_table.put(gpa, mod.intern_pool.stringToSlice(exp.opts.name));
const sym_loc = exported_atom.symbolLoc();
const sym = sym_loc.getSymbol(wasm_file);
sym.setGlobal(true);
sym.setUndefined(false);
sym.index = atom_sym.index;
sym.tag = atom_sym.tag;
sym.name = atom_sym.name;
switch (exp.opts.linkage) {
.Internal => {
sym.setFlag(.WASM_SYM_VISIBILITY_HIDDEN);
sym.setFlag(.WASM_SYM_BINDING_WEAK);
},
.Weak => {
sym.setFlag(.WASM_SYM_BINDING_WEAK);
},
.Strong => {}, // symbols are strong by default
.LinkOnce => {
try mod.failed_exports.putNoClobber(gpa, exp, try Module.ErrorMsg.create(
gpa,
decl.srcLoc(mod),
"Unimplemented: LinkOnce",
.{},
));
continue;
},
}
// TODO: Revisit this
// if (zig_object.global_syms.get(export_name)) |existing_loc| {
// if (existing_loc.index == atom.sym_index) continue;
// const existing_sym: Symbol = existing_loc.getSymbol(wasm_file).*;
// if (!existing_sym.isUndefined()) blk: {
// if (symbol.isWeak()) {
// try wasm_file.discarded.put(gpa, existing_loc, sym_loc);
// continue; // to-be-exported symbol is weak, so we keep the existing symbol
// }
// // new symbol is not weak while existing is, replace existing symbol
// if (existing_sym.isWeak()) {
// break :blk;
// }
// // When both the to-be-exported symbol and the already existing symbol
// // are strong symbols, we have a linker error.
// // In the other case we replace one with the other.
// try mod.failed_exports.put(gpa, exp, try Module.ErrorMsg.create(
// gpa,
// decl.srcLoc(mod),
// \\LinkError: symbol '{}' defined multiple times
// \\ first definition in '{s}'
// \\ next definition in '{s}'
// ,
// .{ exp.opts.name.fmt(&mod.intern_pool), wasm_file.name, wasm_file.name },
// ));
// continue;
// }
// // in this case the existing symbol must be replaced either because it's weak or undefined.
// try wasm.discarded.put(gpa, existing_loc, sym_loc);
// _ = wasm.imports.remove(existing_loc);
// _ = wasm.undefs.swapRemove(existing_sym.name);
// }
// // Ensure the symbol will be exported using the given name
// if (!mod.intern_pool.stringEqlSlice(exp.opts.name, sym_loc.getName(wasm))) {
// try wasm.export_names.put(gpa, sym_loc, export_name);
// }
// try wasm.globals.put(
// gpa,
// export_name,
// sym_loc,
// );
}
}
pub fn freeDecl(zig_object: *ZigObject, wasm_file: *Wasm, decl_index: InternPool.DeclIndex) void {
const gpa = wasm_file.base.comp.gpa;
const mod = wasm_file.base.comp.module.?;
const decl = mod.declPtr(decl_index);
const atom_index = zig_object.decls.get(decl_index).?;
const atom = wasm_file.getAtomPtr(atom_index);
zig_object.symbols_free_list.append(gpa, atom.sym_index) catch {};
_ = zig_object.decls.remove(decl_index);
zig_object.symbols.items[atom.sym_index].tag = .dead;
for (atom.locals.items) |local_atom_index| {
const local_atom = wasm_file.getAtom(local_atom_index);
const local_symbol = &zig_object.symbols.items[local_atom.sym_index];
local_symbol.tag = .dead; // also for any local symbol
zig_object.symbols_free_list.append(gpa, local_atom.sym_index) catch {};
std.denug.assert(wasm_file.symbol_atom.remove(local_atom.symbolLoc()));
}
if (decl.isExtern(mod)) {
_ = zig_object.imports.remove(atom.getSymbolIndex().?);
}
_ = wasm_file.symbol_atom.remove(atom.symbolLoc());
// if (wasm.dwarf) |*dwarf| {
// dwarf.freeDecl(decl_index);
// }
if (atom.next) |next_atom_index| {
const next_atom = wasm_file.getAtomPtr(next_atom_index);
next_atom.prev = atom.prev;
atom.next = null;
}
if (atom.prev) |prev_index| {
const prev_atom = wasm_file.getAtomPtr(prev_index);
prev_atom.next = atom.next;
atom.prev = null;
}
}
fn getTypeIndex(zig_object: *const ZigObject, func_type: std.wasm.Type) ?u32 {
var index: u32 = 0;
while (index < zig_object.func_types.items.len) : (index += 1) {
if (zig_object.func_types.items[index].eql(func_type)) return index;
}
return null;
}
/// Searches for a matching function signature. When no matching signature is found,
/// a new entry will be made. The value returned is the index of the type within `wasm.func_types`.
pub fn putOrGetFuncType(zig_object: *ZigObject, gpa: std.mem.Allocator, func_type: std.wasm.Type) !u32 {
if (zig_object.getTypeIndex(func_type)) |index| {
return index;
}
// functype does not exist.
const index: u32 = @intCast(zig_object.func_types.items.len);
const params = try gpa.dupe(std.wasm.Valtype, func_type.params);
errdefer gpa.free(params);
const returns = try gpa.dupe(std.wasm.Valtype, func_type.returns);
errdefer gpa.free(returns);
try zig_object.func_types.append(gpa, .{
.params = params,
.returns = returns,
});
return index;
}
/// Kind represents the type of an Atom, which is only
/// used to parse a decl into an Atom to define in which section
/// or segment it should be placed.
const Kind = union(enum) {
/// Represents the segment the data symbol should
/// be inserted into.
/// TODO: Add TLS segments
data: enum {
read_only,
uninitialized,
initialized,
},
function: void,
/// Returns the segment name the data kind represents.
/// Asserts `kind` has its active tag set to `data`.
fn segmentName(kind: Kind) []const u8 {
switch (kind.data) {
.read_only => return ".rodata.",
.uninitialized => return ".bss.",
.initialized => return ".data.",
}
}
};
/// Parses an Atom and inserts its metadata into the corresponding sections.
pub fn parseAtom(zig_object: *ZigObject, wasm_file: *Wasm, atom_index: Atom.Index, kind: Kind) !void {
// TODO: Revisit
_ = zig_object;
_ = wasm_file;
_ = atom_index;
_ = kind;
// const comp = wasm.base.comp;
// const gpa = comp.gpa;
// const shared_memory = comp.config.shared_memory;
// const import_memory = comp.config.import_memory;
// const atom = wasm.getAtomPtr(atom_index);
// const symbol = (SymbolLoc{ .file = null, .index = atom.sym_index }).getSymbol(wasm);
// const do_garbage_collect = wasm.base.gc_sections;
// if (symbol.isDead() and do_garbage_collect) {
// // Prevent unreferenced symbols from being parsed.
// return;
// }
// const final_index: u32 = switch (kind) {
// .function => result: {
// const index: u32 = @intCast(wasm.functions.count() + wasm.imported_functions_count);
// const type_index = wasm.atom_types.get(atom_index).?;
// try wasm.functions.putNoClobber(
// gpa,
// .{ .file = null, .index = index },
// .{ .func = .{ .type_index = type_index }, .sym_index = atom.sym_index },
// );
// symbol.tag = .function;
// symbol.index = index;
// if (wasm.code_section_index == null) {
// wasm.code_section_index = @intCast(wasm.segments.items.len);
// try wasm.segments.append(gpa, .{
// .alignment = atom.alignment,
// .size = atom.size,
// .offset = 0,
// .flags = 0,
// });
// }
// break :result wasm.code_section_index.?;
// },
// .data => result: {
// const segment_name = try std.mem.concat(gpa, u8, &.{
// kind.segmentName(),
// wasm.string_table.get(symbol.name),
// });
// errdefer gpa.free(segment_name);
// const segment_info: types.Segment = .{
// .name = segment_name,
// .alignment = atom.alignment,
// .flags = 0,
// };
// symbol.tag = .data;
// // when creating an object file, or importing memory and the data belongs in the .bss segment
// // we set the entire region of it to zeroes.
// // We do not have to do this when exporting the memory (the default) because the runtime
// // will do it for us, and we do not emit the bss segment at all.
// if ((wasm.base.comp.config.output_mode == .Obj or import_memory) and kind.data == .uninitialized) {
// @memset(atom.code.items, 0);
// }
// const should_merge = wasm.base.comp.config.output_mode != .Obj;
// const gop = try wasm.data_segments.getOrPut(gpa, segment_info.outputName(should_merge));
// if (gop.found_existing) {
// const index = gop.value_ptr.*;
// wasm.segments.items[index].size += atom.size;
// symbol.index = @intCast(wasm.segment_info.getIndex(index).?);
// // segment info already exists, so free its memory
// gpa.free(segment_name);
// break :result index;
// } else {
// const index: u32 = @intCast(wasm.segments.items.len);
// var flags: u32 = 0;
// if (shared_memory) {
// flags |= @intFromEnum(Segment.Flag.WASM_DATA_SEGMENT_IS_PASSIVE);
// }
// try wasm.segments.append(gpa, .{
// .alignment = atom.alignment,
// .size = 0,
// .offset = 0,
// .flags = flags,
// });
// gop.value_ptr.* = index;
// const info_index: u32 = @intCast(wasm.segment_info.count());
// try wasm.segment_info.put(gpa, index, segment_info);
// symbol.index = info_index;
// break :result index;
// }
// },
// };
// const segment: *Segment = &wasm.segments.items[final_index];
// segment.alignment = segment.alignment.max(atom.alignment);
// try wasm.appendAtomAtIndex(final_index, atom_index);
}
/// Generates an atom containing the global error set' size.
/// This will only be generated if the symbol exists.
fn setupErrorsLen(zig_object: *ZigObject, wasm_file: *Wasm) !void {
const gpa = wasm_file.base.comp.gpa;
const loc = zig_object.findGlobalSymbol("__zig_errors_len") orelse return;
const errors_len = wasm_file.base.comp.module.?.global_error_set.count();
// overwrite existing atom if it already exists (maybe the error set has increased)
// if not, allcoate a new atom.
const atom_index = if (wasm_file.symbol_atom.get(loc)) |index| blk: {
const atom = wasm_file.getAtomPtr(index);
if (atom.next) |next_atom_index| {
const next_atom = wasm_file.getAtomPtr(next_atom_index);
next_atom.prev = atom.prev;
atom.next = null;
}
if (atom.prev) |prev_index| {
const prev_atom = wasm_file.getAtomPtr(prev_index);
prev_atom.next = atom.next;
atom.prev = null;
}
atom.deinit(gpa);
break :blk index;
} else new_atom: {
const atom_index: Atom.Index = @intCast(wasm_file.managed_atoms.items.len);
try wasm_file.symbol_atom.put(gpa, loc, atom_index);
try wasm_file.managed_atoms.append(gpa, undefined);
break :new_atom atom_index;
};
const atom = wasm_file.getAtomPtr(atom_index);
atom.* = Atom.empty;
atom.sym_index = loc.index;
atom.size = 2;
try atom.code.writer(gpa).writeInt(u16, @intCast(errors_len), .little);
// try wasm.parseAtom(atom_index, .{ .data = .read_only });
}
/// Initializes symbols and atoms for the debug sections
/// Initialization is only done when compiling Zig code.
/// When Zig is invoked as a linker instead, the atoms
/// and symbols come from the object files instead.
pub fn initDebugSections(zig_object: *ZigObject) !void {
if (zig_object.dwarf == null) return; // not compiling Zig code, so no need to pre-initialize debug sections
std.debug.assert(zig_object.debug_info_index == null);
// this will create an Atom and set the index for us.
zig_object.debug_info_atom = try zig_object.createDebugSectionForIndex(&zig_object.debug_info_index, ".debug_info");
zig_object.debug_line_atom = try zig_object.createDebugSectionForIndex(&zig_object.debug_line_index, ".debug_line");
zig_object.debug_loc_atom = try zig_object.createDebugSectionForIndex(&zig_object.debug_loc_index, ".debug_loc");
zig_object.debug_abbrev_atom = try zig_object.createDebugSectionForIndex(&zig_object.debug_abbrev_index, ".debug_abbrev");
zig_object.debug_ranges_atom = try zig_object.createDebugSectionForIndex(&zig_object.debug_ranges_index, ".debug_ranges");
zig_object.debug_str_atom = try zig_object.createDebugSectionForIndex(&zig_object.debug_str_index, ".debug_str");
zig_object.debug_pubnames_atom = try zig_object.createDebugSectionForIndex(&zig_object.debug_pubnames_index, ".debug_pubnames");
zig_object.debug_pubtypes_atom = try zig_object.createDebugSectionForIndex(&zig_object.debug_pubtypes_index, ".debug_pubtypes");
}
/// From a given index variable, creates a new debug section.
/// This initializes the index, appends a new segment,
/// and finally, creates a managed `Atom`.
pub fn createDebugSectionForIndex(zig_object: *ZigObject, wasm_file: *Wasm, index: *?u32, name: []const u8) !Atom.Index {
const gpa = wasm_file.base.comp.gpa;
const new_index: u32 = @intCast(zig_object.segments.items.len);
index.* = new_index;
try zig_object.appendDummySegment();
const sym_index = try zig_object.allocateSymbol(gpa);
const atom_index = try wasm_file.createAtom(sym_index, zig_object.index);
const atom = wasm_file.getAtomPtr(atom_index);
zig_object.symbols.items[sym_index] = .{
.tag = .section,
.name = try zig_object.string_table.put(gpa, name),
.index = 0,
.flags = @intFromEnum(Symbol.Flag.WASM_SYM_BINDING_LOCAL),
};
atom.alignment = .@"1"; // debug sections are always 1-byte-aligned
return atom_index;
}
pub fn updateDeclLineNumber(zig_object: *ZigObject, mod: *Module, decl_index: InternPool.DeclIndex) !void {
if (zig_object.dwarf) |*dw| {
const decl = mod.declPtr(decl_index);
const decl_name = mod.intern_pool.stringToSlice(try decl.getFullyQualifiedName(mod));
log.debug("updateDeclLineNumber {s}{*}", .{ decl_name, decl });
try dw.updateDeclLineNumber(mod, decl_index);
}
}
/// Allocates debug atoms into their respective debug sections
/// to merge them with maybe-existing debug atoms from object files.
fn allocateDebugAtoms(zig_object: *ZigObject) !void {
if (zig_object.dwarf == null) return;
const allocAtom = struct {
fn f(ctx: *ZigObject, maybe_index: *?u32, atom_index: Atom.Index) !void {
const index = maybe_index.* orelse idx: {
const index = @as(u32, @intCast(ctx.segments.items.len));
try ctx.appendDummySegment();
maybe_index.* = index;
break :idx index;
};
const atom = ctx.getAtomPtr(atom_index);
atom.size = @as(u32, @intCast(atom.code.items.len));
ctx.symbols.items[atom.sym_index].index = index;
try ctx.appendAtomAtIndex(index, atom_index);
}
}.f;
try allocAtom(zig_object, &zig_object.debug_info_index, zig_object.debug_info_atom.?);
try allocAtom(zig_object, &zig_object.debug_line_index, zig_object.debug_line_atom.?);
try allocAtom(zig_object, &zig_object.debug_loc_index, zig_object.debug_loc_atom.?);
try allocAtom(zig_object, &zig_object.debug_str_index, zig_object.debug_str_atom.?);
try allocAtom(zig_object, &zig_object.debug_ranges_index, zig_object.debug_ranges_atom.?);
try allocAtom(zig_object, &zig_object.debug_abbrev_index, zig_object.debug_abbrev_atom.?);
try allocAtom(zig_object, &zig_object.debug_pubnames_index, zig_object.debug_pubnames_atom.?);
try allocAtom(zig_object, &zig_object.debug_pubtypes_index, zig_object.debug_pubtypes_atom.?);
}
/// For the given `decl_index`, stores the corresponding type representing the function signature.
/// Asserts declaration has an associated `Atom`.
/// Returns the index into the list of types.
pub fn storeDeclType(zig_object: *ZigObject, gpa: std.mem.Allocator, decl_index: InternPool.DeclIndex, func_type: std.wasm.Type) !u32 {
const atom_index = zig_object.decls.get(decl_index).?;
const index = try zig_object.putOrGetFuncType(gpa, func_type);
try zig_object.atom_types.put(gpa, atom_index, index);
return index;
}
/// The symbols in ZigObject are already represented by an atom as we need to store its data.
/// So rather than creating a new Atom and returning its index, we use this oppertunity to scan
/// its relocations and create any GOT symbols or function table indexes it may require.
pub fn parseSymbolIntoAtom(zig_object: *ZigObject, wasm_file: *Wasm, index: u32) !Atom.Index {
const gpa = wasm_file.base.comp.gpa;
const loc: Wasm.SymbolLoc = .{ .file = zig_object.index, .index = index };
const final_index = try wasm_file.getMatchingSegment(zig_object.index, index);
const atom_index = wasm_file.symbol_atom.get(loc).?;
try wasm_file.appendAtomAtIndex(final_index, atom_index);
const atom = wasm_file.getAtom(atom_index);
for (atom.relocs.items) |reloc| {
switch (reloc.relocation_type) {
.R_WASM_TABLE_INDEX_I32,
.R_WASM_TABLE_INDEX_I64,
.R_WASM_TABLE_INDEX_SLEB,
.R_WASM_TABLE_INDEX_SLEB64,
=> {
try wasm_file.function_table.put(gpa, loc, 0);
},
.R_WASM_GLOBAL_INDEX_I32,
.R_WASM_GLOBAL_INDEX_LEB,
=> {
const sym = zig_object.symbol(reloc.index);
if (sym.tag != .global) {
try wasm_file.got_symbols.append(gpa, loc);
}
},
else => {},
}
}
return atom_index;
}
const build_options = @import("build_options");
const builtin = @import("builtin");
const codegen = @import("../../codegen.zig");
const link = @import("../../link.zig");
const log = std.log.scoped(.zig_object);
const std = @import("std");
const types = @import("types.zig");
const Air = @import("../../Air.zig");
const Atom = @import("Atom.zig");
const Dwarf = @import("../Dwarf.zig");
const File = @import("file.zig").File;
const InternPool = @import("../../InternPool.zig");
const Liveness = @import("../../Liveness.zig");
const Module = @import("../../Module.zig");
const StringTable = @import("../StringTable.zig");
const Symbol = @import("Symbol.zig");
const Type = @import("../../type.zig").Type;
const TypedValue = @import("../../TypedValue.zig");
const Value = @import("../../value.zig").Value;
const Wasm = @import("../Wasm.zig");
const ZigObject = @This();