wasm linker: finish the flush function

This branch is passing type checking now.
This commit is contained in:
Andrew Kelley 2024-12-12 16:56:30 -08:00
parent 264a63b000
commit 968941b535
3 changed files with 128 additions and 103 deletions

View File

@ -1219,6 +1219,12 @@ pub const Cpu = struct {
} else true;
}
pub fn count(set: Set) std.math.IntFittingRange(0, needed_bit_count) {
var sum: usize = 0;
for (set.ints) |x| sum += @popCount(x);
return @intCast(sum);
}
pub fn isEnabled(set: Set, arch_feature_index: Index) bool {
const usize_index = arch_feature_index / @bitSizeOf(usize);
const bit_index: ShiftInt = @intCast(arch_feature_index % @bitSizeOf(usize));

View File

@ -270,8 +270,16 @@ pub const FunctionIndex = enum(u32) {
return &wasm.functions.keys()[@intFromEnum(index)];
}
pub fn toOutputFunctionIndex(index: FunctionIndex, wasm: *const Wasm) OutputFunctionIndex {
return @enumFromInt(wasm.function_imports.entries.len + @intFromEnum(index));
}
pub fn fromIpNav(wasm: *const Wasm, nav_index: InternPool.Nav.Index) ?FunctionIndex {
const i = wasm.functions.getIndex(.fromIpNav(wasm, nav_index)) orelse return null;
return fromResolution(wasm, .fromIpNav(wasm, nav_index));
}
pub fn fromResolution(wasm: *const Wasm, resolution: FunctionImport.Resolution) ?FunctionIndex {
const i = wasm.functions.getIndex(resolution) orelse return null;
return @enumFromInt(i);
}
};
@ -295,11 +303,11 @@ pub const OutputFunctionIndex = enum(u32) {
_,
};
/// Index into `globals`.
/// Index into `Wasm.globals`.
pub const GlobalIndex = enum(u32) {
_,
fn key(index: GlobalIndex, f: *const Flush) *Wasm.GlobalImport.Resolution {
pub fn ptr(index: GlobalIndex, f: *const Flush) *Wasm.GlobalImport.Resolution {
return &f.globals.items[@intFromEnum(index)];
}
@ -1089,7 +1097,7 @@ pub const DataSegment = extern struct {
/// The size in bytes of the data representing the segment within the section.
len: u32,
fn slice(p: DataSegment.Payload, wasm: *const Wasm) []const u8 {
pub fn slice(p: DataSegment.Payload, wasm: *const Wasm) []const u8 {
assert(p.off != p.len);
return wasm.string_bytes.items[p.off..][0..p.len];
}
@ -2365,6 +2373,7 @@ pub fn flushModule(
_ = tid;
const comp = wasm.base.comp;
const use_lld = build_options.have_llvm and comp.config.use_lld;
const diags = &comp.link_diags;
if (wasm.llvm_object) |llvm_object| {
try wasm.base.emitLlvmObject(arena, llvm_object, prog_node);
@ -2392,7 +2401,11 @@ pub fn flushModule(
defer sub_prog_node.end();
wasm.flush_buffer.clear();
return wasm.flush_buffer.finish(wasm, arena);
return wasm.flush_buffer.finish(wasm, arena) catch |err| switch (err) {
error.OutOfMemory => return error.OutOfMemory,
error.LinkFailure => return error.LinkFailure,
else => |e| return diags.fail("failed to flush wasm: {s}", .{@errorName(e)}),
};
}
fn linkWithLLD(wasm: *Wasm, arena: Allocator, tid: Zcu.PerThread.Id, prog_node: std.Progress.Node) !void {

View File

@ -59,7 +59,7 @@ pub fn finish(f: *Flush, wasm: *Wasm, arena: Allocator) !void {
const gpa = comp.gpa;
const import_memory = comp.config.import_memory;
const export_memory = comp.config.export_memory;
const target = comp.root_mod.resolved_target.result;
const target = &comp.root_mod.resolved_target.result;
const is_obj = comp.config.output_mode == .Obj;
const allow_undefined = is_obj or wasm.import_symbols;
const zcu = wasm.base.comp.zcu.?;
@ -523,7 +523,7 @@ pub fn finish(f: *Flush, wasm: *Wasm, arena: Allocator) !void {
try leb.writeUleb128(binary_writer, @as(u32, @intCast(name.len)));
try binary_bytes.appendSlice(gpa, name);
try binary_bytes.append(gpa, @intFromEnum(std.wasm.ExternalKind.function));
try leb.writeUleb128(binary_writer, @as(u32, @intCast(wasm.function_imports.entries.len + @intFromEnum(exp.function_index))));
try leb.writeUleb128(binary_writer, @intFromEnum(exp.function_index));
}
exports_len += wasm.function_exports.items.len;
@ -543,7 +543,7 @@ pub fn finish(f: *Flush, wasm: *Wasm, arena: Allocator) !void {
try leb.writeUleb128(binary_writer, @as(u32, @intCast(name.len)));
try binary_bytes.appendSlice(gpa, name);
try binary_bytes.append(gpa, @intFromEnum(std.wasm.ExternalKind.global));
try leb.writeUleb128(binary_writer, @as(u32, @intCast(wasm.global_imports.entries.len + @intFromEnum(exp.global_index))));
try leb.writeUleb128(binary_writer, @intFromEnum(exp.global_index));
}
exports_len += wasm.global_exports.items.len;
@ -555,38 +555,39 @@ pub fn finish(f: *Flush, wasm: *Wasm, arena: Allocator) !void {
}
}
if (Wasm.FunctionIndex.fromResolution(wasm.entry_resolution)) |entry_index| {
if (Wasm.FunctionIndex.fromResolution(wasm, wasm.entry_resolution)) |entry_index| {
const header_offset = try reserveVecSectionHeader(gpa, binary_bytes);
replaceVecSectionHeader(binary_bytes, header_offset, .start, @intFromEnum(entry_index));
}
// element section (function table)
if (wasm.function_table.count() > 0) {
const header_offset = try reserveVecSectionHeader(gpa, binary_bytes);
if (f.indirect_function_table.count() > 0) {
@panic("TODO");
//const header_offset = try reserveVecSectionHeader(gpa, binary_bytes);
const table_loc = wasm.globals.get(wasm.preloaded_strings.__indirect_function_table).?;
const table_sym = wasm.finalSymbolByLoc(table_loc);
//const table_loc = wasm.globals.get(wasm.preloaded_strings.__indirect_function_table).?;
//const table_sym = wasm.finalSymbolByLoc(table_loc);
const flags: u32 = if (table_sym.index == 0) 0x0 else 0x02; // passive with implicit 0-index table or set table index manually
try leb.writeUleb128(binary_writer, flags);
if (flags == 0x02) {
try leb.writeUleb128(binary_writer, table_sym.index);
}
try emitInit(binary_writer, .{ .i32_const = 1 }); // We start at index 1, so unresolved function pointers are invalid
if (flags == 0x02) {
try leb.writeUleb128(binary_writer, @as(u8, 0)); // represents funcref
}
try leb.writeUleb128(binary_writer, @as(u32, @intCast(wasm.function_table.count())));
var symbol_it = wasm.function_table.keyIterator();
while (symbol_it.next()) |symbol_loc_ptr| {
const sym = wasm.finalSymbolByLoc(symbol_loc_ptr.*);
assert(sym.flags.alive);
assert(sym.index < wasm.functions.count() + wasm.imported_functions_count);
try leb.writeUleb128(binary_writer, sym.index);
}
//const flags: u32 = if (table_sym.index == 0) 0x0 else 0x02; // passive with implicit 0-index table or set table index manually
//try leb.writeUleb128(binary_writer, flags);
//if (flags == 0x02) {
// try leb.writeUleb128(binary_writer, table_sym.index);
//}
//try emitInit(binary_writer, .{ .i32_const = 1 }); // We start at index 1, so unresolved function pointers are invalid
//if (flags == 0x02) {
// try leb.writeUleb128(binary_writer, @as(u8, 0)); // represents funcref
//}
//try leb.writeUleb128(binary_writer, @as(u32, @intCast(f.indirect_function_table.count())));
//var symbol_it = f.indirect_function_table.keyIterator();
//while (symbol_it.next()) |symbol_loc_ptr| {
// const sym = wasm.finalSymbolByLoc(symbol_loc_ptr.*);
// assert(sym.flags.alive);
// assert(sym.index < wasm.functions.count() + wasm.imported_functions_count);
// try leb.writeUleb128(binary_writer, sym.index);
//}
replaceVecSectionHeader(binary_bytes, header_offset, .element, 1);
section_index += 1;
//replaceVecSectionHeader(binary_bytes, header_offset, .element, 1);
//section_index += 1;
}
// When the shared-memory option is enabled, we *must* emit the 'data count' section.
@ -600,7 +601,7 @@ pub fn finish(f: *Flush, wasm: *Wasm, arena: Allocator) !void {
const header_offset = try reserveVecSectionHeader(gpa, binary_bytes);
const start_offset = binary_bytes.items.len - 5; // minus 5 so start offset is 5 to include entry count
for (wasm.functions.keys()) |resolution| switch (resolution.unpack()) {
for (wasm.functions.keys()) |resolution| switch (resolution.unpack(wasm)) {
.unresolved => unreachable,
.__wasm_apply_global_tls_relocs => @panic("TODO lower __wasm_apply_global_tls_relocs"),
.__wasm_call_ctors => @panic("TODO lower __wasm_call_ctors"),
@ -614,10 +615,19 @@ pub fn finish(f: *Flush, wasm: *Wasm, arena: Allocator) !void {
//try leb.writeUleb128(binary_writer, atom.code.len);
//try binary_bytes.appendSlice(gpa, atom.code.slice(wasm));
},
.nav => |i| {
.nav_exe => |i| {
assert(!is_obj);
_ = i;
_ = start_offset;
@panic("TODO lower nav code and apply relocations");
@panic("TODO lower nav exe code and apply relocations");
//try leb.writeUleb128(binary_writer, atom.code.len);
//try binary_bytes.appendSlice(gpa, atom.code.slice(wasm));
},
.nav_obj => |i| {
assert(is_obj);
_ = i;
_ = start_offset;
@panic("TODO lower nav obj code and apply relocations");
//try leb.writeUleb128(binary_writer, atom.code.len);
//try binary_bytes.appendSlice(gpa, atom.code.slice(wasm));
},
@ -636,7 +646,8 @@ pub fn finish(f: *Flush, wasm: *Wasm, arena: Allocator) !void {
var offset: u32 = undefined;
for (segment_indexes, segment_offsets) |segment_index, segment_offset| {
const segment = segment_index.ptr(wasm);
if (segment.size == 0) continue;
const segment_payload = segment.payload.slice(wasm);
if (segment_payload.len == 0) continue;
if (!import_memory and isBss(wasm, segment.name)) {
// It counted for virtual memory but it does not go into the binary.
continue;
@ -657,8 +668,8 @@ pub fn finish(f: *Flush, wasm: *Wasm, arena: Allocator) !void {
try binary_bytes.appendNTimes(gpa, 0, segment_offset - offset);
offset = segment_offset;
try binary_bytes.appendSlice(gpa, segment.payload.slice(wasm));
offset += segment.payload.len;
try binary_bytes.appendSlice(gpa, segment_payload);
offset += @intCast(segment_payload.len);
if (true) @panic("TODO apply data segment relocations");
}
assert(group_index == f.data_segment_groups.items.len);
@ -680,7 +691,7 @@ pub fn finish(f: *Flush, wasm: *Wasm, arena: Allocator) !void {
// try wasm.emitDataRelocations(binary_bytes, data_index, symbol_table);
//}
} else if (comp.config.debug_format != .strip) {
try wasm.emitNameSection(binary_bytes, arena);
try emitNameSection(wasm, binary_bytes, arena);
}
if (comp.config.debug_format != .strip) {
@ -699,14 +710,14 @@ pub fn finish(f: *Flush, wasm: *Wasm, arena: Allocator) !void {
std.fmt.fmtSliceHexLower(id[8..10]),
std.fmt.fmtSliceHexLower(id[10..]),
});
try emitBuildIdSection(binary_bytes, &uuid);
try emitBuildIdSection(gpa, binary_bytes, &uuid);
},
.hexstring => |hs| {
var buffer: [32 * 2]u8 = undefined;
const str = std.fmt.bufPrint(&buffer, "{s}", .{
std.fmt.fmtSliceHexLower(hs.toSlice()),
}) catch unreachable;
try emitBuildIdSection(binary_bytes, str);
try emitBuildIdSection(gpa, binary_bytes, str);
},
else => |mode| {
var err = try diags.addErrorWithNotes(0);
@ -717,9 +728,8 @@ pub fn finish(f: *Flush, wasm: *Wasm, arena: Allocator) !void {
var debug_bytes = std.ArrayList(u8).init(gpa);
defer debug_bytes.deinit();
try emitProducerSection(binary_bytes);
if (!target.cpu.features.isEmpty())
try emitFeaturesSection(binary_bytes, target.cpu.features);
try emitProducerSection(gpa, binary_bytes);
try emitFeaturesSection(gpa, binary_bytes, target);
}
// Finally, write the entire binary into the file.
@ -729,6 +739,10 @@ pub fn finish(f: *Flush, wasm: *Wasm, arena: Allocator) !void {
}
fn emitNameSection(wasm: *Wasm, binary_bytes: *std.ArrayListUnmanaged(u8), arena: Allocator) !void {
if (true) {
std.log.warn("TODO emit name section", .{});
return;
}
const comp = wasm.base.comp;
const gpa = comp.gpa;
const import_memory = comp.config.import_memory;
@ -796,31 +810,15 @@ fn emitNameSection(wasm: *Wasm, binary_bytes: *std.ArrayListUnmanaged(u8), arena
// Data segments are already ordered.
const header_offset = try reserveCustomSectionHeader(gpa, binary_bytes);
const writer = binary_bytes.writer();
defer writeCustomSectionHeader(binary_bytes, header_offset);
const writer = binary_bytes.writer(gpa);
try leb.writeUleb128(writer, @as(u32, @intCast("name".len)));
try writer.writeAll("name");
try emitNameSubsection(wasm, binary_bytes, .function, funcs.keys(), funcs.values());
try emitNameSubsection(wasm, binary_bytes, .global, globals.items(.index), globals.items(.name));
try emitNameSubsection(wasm, binary_bytes, .data_segment, segments.items(.index), segments.items(.name));
try writeCustomSectionHeader(
binary_bytes.items,
header_offset,
@as(u32, @intCast(binary_bytes.items.len - header_offset - 6)),
);
}
fn writeCustomSectionHeader(buffer: []u8, offset: u32, size: u32) !void {
var buf: [1 + 5]u8 = undefined;
buf[0] = 0; // 0 = 'custom' section
leb.writeUnsignedFixed(5, buf[1..6], size);
buffer[offset..][0..buf.len].* = buf;
}
fn reserveCustomSectionHeader(gpa: Allocator, bytes: *std.ArrayListUnmanaged(u8)) Allocator.Error!u32 {
try bytes.appendNTimes(gpa, 0, section_header_size);
return @intCast(bytes.items.len - section_header_size);
}
fn emitNameSubsection(
@ -856,35 +854,40 @@ fn emitNameSubsection(
fn emitFeaturesSection(
gpa: Allocator,
binary_bytes: *std.ArrayListUnmanaged(u8),
features: []const Wasm.Feature,
) !void {
const header_offset = try reserveCustomSectionHeader(gpa, binary_bytes);
target: *const std.Target,
) Allocator.Error!void {
const feature_count = target.cpu.features.count();
if (feature_count == 0) return;
const writer = binary_bytes.writer();
const header_offset = try reserveCustomSectionHeader(gpa, binary_bytes);
defer writeCustomSectionHeader(binary_bytes, header_offset);
const writer = binary_bytes.writer(gpa);
const target_features = "target_features";
try leb.writeUleb128(writer, @as(u32, @intCast(target_features.len)));
try writer.writeAll(target_features);
try leb.writeUleb128(writer, @as(u32, @intCast(features.len)));
for (features) |feature| {
assert(feature.prefix != .invalid);
try leb.writeUleb128(writer, @tagName(feature.prefix)[0]);
const name = @tagName(feature.tag);
try leb.writeUleb128(writer, @as(u32, name.len));
try leb.writeUleb128(writer, @as(u32, @intCast(feature_count)));
var safety_count = feature_count;
for (target.cpu.arch.allFeaturesList(), 0..) |*feature, i| {
if (!std.Target.wasm.featureSetHas(target.cpu.features, @enumFromInt(i))) continue;
safety_count -= 1;
try leb.writeUleb128(writer, @as(u32, '+'));
// Depends on llvm_name for the hyphenated version that matches wasm tooling conventions.
const name = feature.llvm_name.?;
try leb.writeUleb128(writer, @as(u32, @intCast(name.len)));
try writer.writeAll(name);
}
try writeCustomSectionHeader(
binary_bytes.items,
header_offset,
@as(u32, @intCast(binary_bytes.items.len - header_offset - 6)),
);
assert(safety_count == 0);
}
fn emitBuildIdSection(gpa: Allocator, binary_bytes: *std.ArrayListUnmanaged(u8), build_id: []const u8) !void {
const header_offset = try reserveCustomSectionHeader(gpa, binary_bytes);
defer writeCustomSectionHeader(binary_bytes, header_offset);
const writer = binary_bytes.writer();
const writer = binary_bytes.writer(gpa);
const hdr_build_id = "build_id";
try leb.writeUleb128(writer, @as(u32, @intCast(hdr_build_id.len)));
try writer.writeAll(hdr_build_id);
@ -892,18 +895,13 @@ fn emitBuildIdSection(gpa: Allocator, binary_bytes: *std.ArrayListUnmanaged(u8),
try leb.writeUleb128(writer, @as(u32, 1));
try leb.writeUleb128(writer, @as(u32, @intCast(build_id.len)));
try writer.writeAll(build_id);
try writeCustomSectionHeader(
binary_bytes.items,
header_offset,
@as(u32, @intCast(binary_bytes.items.len - header_offset - 6)),
);
}
fn emitProducerSection(gpa: Allocator, binary_bytes: *std.ArrayListUnmanaged(u8)) !void {
const header_offset = try reserveCustomSectionHeader(gpa, binary_bytes);
defer writeCustomSectionHeader(binary_bytes, header_offset);
const writer = binary_bytes.writer();
const writer = binary_bytes.writer(gpa);
const producers = "producers";
try leb.writeUleb128(writer, @as(u32, @intCast(producers.len)));
try writer.writeAll(producers);
@ -947,12 +945,6 @@ fn emitProducerSection(gpa: Allocator, binary_bytes: *std.ArrayListUnmanaged(u8)
try writer.writeAll(build_options.version);
}
}
try writeCustomSectionHeader(
binary_bytes.items,
header_offset,
@as(u32, @intCast(binary_bytes.items.len - header_offset - 6)),
);
}
///// For each relocatable section, emits a custom "relocation.<section_name>" section
@ -965,7 +957,7 @@ fn emitProducerSection(gpa: Allocator, binary_bytes: *std.ArrayListUnmanaged(u8)
// const comp = wasm.base.comp;
// const gpa = comp.gpa;
// const code_index = wasm.code_section_index.unwrap() orelse return;
// const writer = binary_bytes.writer();
// const writer = binary_bytes.writer(gpa);
// const header_offset = try reserveCustomSectionHeader(gpa, binary_bytes);
//
// // write custom section information
@ -1001,8 +993,7 @@ fn emitProducerSection(gpa: Allocator, binary_bytes: *std.ArrayListUnmanaged(u8)
// var buf: [5]u8 = undefined;
// leb.writeUnsignedFixed(5, &buf, count);
// try binary_bytes.insertSlice(reloc_start, &buf);
// const size: u32 = @intCast(binary_bytes.items.len - header_offset - 6);
// try writeCustomSectionHeader(binary_bytes.items, header_offset, size);
// writeCustomSectionHeader(binary_bytes, header_offset);
//}
//fn emitDataRelocations(
@ -1013,7 +1004,7 @@ fn emitProducerSection(gpa: Allocator, binary_bytes: *std.ArrayListUnmanaged(u8)
//) !void {
// const comp = wasm.base.comp;
// const gpa = comp.gpa;
// const writer = binary_bytes.writer();
// const writer = binary_bytes.writer(gpa);
// const header_offset = try reserveCustomSectionHeader(gpa, binary_bytes);
//
// // write custom section information
@ -1052,8 +1043,7 @@ fn emitProducerSection(gpa: Allocator, binary_bytes: *std.ArrayListUnmanaged(u8)
// var buf: [5]u8 = undefined;
// leb.writeUnsignedFixed(5, &buf, count);
// try binary_bytes.insertSlice(reloc_start, &buf);
// const size = @as(u32, @intCast(binary_bytes.items.len - header_offset - 6));
// try writeCustomSectionHeader(binary_bytes.items, header_offset, size);
// writeCustomSectionHeader(binary_bytes, header_offset);
//}
fn isBss(wasm: *Wasm, optional_name: Wasm.OptionalString) bool {
@ -1104,6 +1094,21 @@ fn replaceVecSectionHeader(
bytes.replaceRangeAssumeCapacity(offset, section_header_reserve_size, fbw.getWritten());
}
fn reserveCustomSectionHeader(gpa: Allocator, bytes: *std.ArrayListUnmanaged(u8)) Allocator.Error!u32 {
try bytes.appendNTimes(gpa, 0, section_header_size);
return @intCast(bytes.items.len - section_header_size);
}
fn writeCustomSectionHeader(bytes: *std.ArrayListUnmanaged(u8), offset: u32) void {
const size: u32 = @intCast(bytes.items.len - offset - section_header_size);
var buf: [section_header_size]u8 = undefined;
var fbw = std.io.fixedBufferStream(&buf);
const w = fbw.writer();
w.writeByte(0) catch unreachable; // 0 = 'custom' section
leb.writeUleb128(w, size) catch unreachable;
bytes.replaceRangeAssumeCapacity(offset, section_header_size, fbw.getWritten());
}
fn emitLimits(
gpa: Allocator,
binary_bytes: *std.ArrayListUnmanaged(u8),
@ -1171,7 +1176,7 @@ pub fn emitExpr(wasm: *const Wasm, binary_bytes: *std.ArrayListUnmanaged(u8), ex
//) !void {
// const gpa = wasm.base.comp.gpa;
// const offset = try reserveCustomSectionHeader(gpa, binary_bytes);
// const writer = binary_bytes.writer();
// const writer = binary_bytes.writer(gpa);
// // emit "linking" custom section name
// const section_name = "linking";
// try leb.writeUleb128(writer, section_name.len);
@ -1186,11 +1191,12 @@ pub fn emitExpr(wasm: *const Wasm, binary_bytes: *std.ArrayListUnmanaged(u8), ex
// try wasm.emitSegmentInfo(binary_bytes);
//
// const size: u32 = @intCast(binary_bytes.items.len - offset - 6);
// try writeCustomSectionHeader(binary_bytes.items, offset, size);
// writeCustomSectionHeader(binary_bytes, offset, size);
//}
fn emitSegmentInfo(wasm: *Wasm, binary_bytes: *std.ArrayList(u8)) !void {
const writer = binary_bytes.writer();
const gpa = wasm.base.comp.gpa;
const writer = binary_bytes.writer(gpa);
try leb.writeUleb128(writer, @intFromEnum(Wasm.SubsectionType.segment_info));
const segment_offset = binary_bytes.items.len;
@ -1370,7 +1376,7 @@ fn emitSegmentInfo(wasm: *Wasm, binary_bytes: *std.ArrayList(u8)) !void {
// .TABLE_INDEX_I64,
// .TABLE_INDEX_SLEB,
// .TABLE_INDEX_SLEB64,
// => wasm.function_table.get(.{ .file = atom.file, .index = @enumFromInt(relocation.index) }) orelse 0,
// => wasm.indirect_function_table.get(.{ .file = atom.file, .index = @enumFromInt(relocation.index) }) orelse 0,
//
// .TYPE_INDEX_LEB => unreachable, // handled above
// .GLOBAL_INDEX_I32, .GLOBAL_INDEX_LEB => if (symbol.flags.undefined)