Commit 53c668d3a9dad430749cbc642d7bf8cf45eefba1 | zig

author	Jakub Konka <kubkon@jakubkonka.com> 2022-01-24 10:33:42+01:00
committer	Jakub Konka <kubkon@jakubkonka.com> 2022-01-25 23:51:19+01:00
log	53c668d3a9dad430749cbc642d7bf8cf45eefba1
tree	38b4ab6279aa9cd396b2a6622451c78116f35dea
parent	05c5bb9edd51e73c0d4a2619817fbff73b82f230

stage2: add naive impl of pointer type in ELF

Augment relocation tracking mechanism to de-duplicate potential
creation of base as well as composite types while unrolling
composite types in the linker - there is still potential for
further space optimisation by moving all type information into
a separate section `.debug_types` and providing references to
entries within that section whenever required (e.g., `ref4` form).
Currently, we duplicate type definitions on a per-decl basis.

Anyhow, with this patch, an example function signature of the following
type:

```zig
fn byPtrPtr(ptr_ptr_x: **u32, ptr_x: *u32) void {
    ptr_ptr_x.* = ptr_x;
}
```

will generate the following `.debug_info` for formal parameters:

```
 <1><1aa>: Abbrev Number: 3 (DW_TAG_subprogram)
    <1ab>   DW_AT_low_pc      : 0x8000197
    <1b3>   DW_AT_high_pc     : 0x2c
    <1b7>   DW_AT_name        : byPtrPtr
 <2><1c0>: Abbrev Number: 7 (DW_TAG_formal_parameter)
    <1c1>   DW_AT_location    : 1 byte block: 55        (DW_OP_reg5 (rdi))
    <1c3>   DW_AT_type        : <0x1df>
    <1c7>   DW_AT_name        : ptr_ptr_x
 <2><1d1>: Abbrev Number: 7 (DW_TAG_formal_parameter)
    <1d2>   DW_AT_location    : 1 byte block: 54        (DW_OP_reg4 (rsi))
    <1d4>   DW_AT_type        : <0x1e4>
    <1d8>   DW_AT_name        : ptr_x
 <2><1de>: Abbrev Number: 0
 <1><1df>: Abbrev Number: 5 (DW_TAG_pointer_type)
    <1e0>   DW_AT_type        : <0x1e4>
 <1><1e4>: Abbrev Number: 5 (DW_TAG_pointer_type)
    <1e5>   DW_AT_type        : <0x1e9>
 <1><1e9>: Abbrev Number: 4 (DW_TAG_base_type)
    <1ea>   DW_AT_encoding    : 7       (unsigned)
    <1eb>   DW_AT_byte_size   : 4
    <1ec>   DW_AT_name        : u32
```

4 files changed, 66 insertions(+), 21 deletions(-)

src
- link.zig
- link
  - Elf.zig
  - MachO.zig
  - MachO
    DebugSymbols.zig

src/link.zig+6-1

 @@ -235,7 +235,12 @@ pub const File = struct {
     };
     /// For DWARF .debug_info.
-    pub const DbgInfoTypeRelocsTable = std.HashMapUnmanaged(Type, DbgInfoTypeReloc, Type.HashContext64, std.hash_map.default_max_load_percentage);
+    pub const DbgInfoTypeRelocsTable = std.ArrayHashMapUnmanaged(
+        Type,
+        DbgInfoTypeReloc,
+        Type.HashContext32,
+        true,
+    );
     /// For DWARF .debug_info.
     pub const DbgInfoTypeReloc = struct {

src/link/Elf.zig+57-14

 @@ -783,8 +783,9 @@ pub const abbrev_compile_unit = 1;
 pub const abbrev_subprogram = 2;
 pub const abbrev_subprogram_retvoid = 3;
 pub const abbrev_base_type = 4;
-pub const abbrev_pad1 = 5;
+pub const abbrev_ptr_type = 5;
-pub const abbrev_parameter = 6;
+pub const abbrev_pad1 = 6;
+pub const abbrev_parameter = 7;
 pub fn flush(self: *Elf, comp: *Compilation) !void {
     if (self.base.options.emit == null) {
 @@ -871,9 +872,21 @@ pub fn flushModule(self: *Elf, comp: *Compilation) !void {
             DW.AT.byte_size,
             DW.FORM.data1,
             DW.AT.name,
-            DW.FORM.string, 0, 0, // table sentinel
+            DW.FORM.string,
-            abbrev_pad1, DW.TAG.unspecified_type, DW.CHILDREN.no, // header
+            0,
-            0,                0, // table sentinel
+            0, // table sentinel
+            abbrev_ptr_type,
+            DW.TAG.pointer_type,
+            DW.CHILDREN.no, // header
+            DW.AT.type,
+            DW.FORM.ref4,
+            0,
+            0, // table sentinel
+            abbrev_pad1,
+            DW.TAG.unspecified_type,
+            DW.CHILDREN.no, // header
+            0,
+            0, // table sentinel
             abbrev_parameter,
             DW.TAG.formal_parameter, DW.CHILDREN.no, // header
             DW.AT.location,          DW.FORM.exprloc,
 @@ -2309,8 +2322,7 @@ pub fn freeDecl(self: *Elf, decl: *Module.Decl) void {
 }
 fn deinitRelocs(gpa: Allocator, table: *File.DbgInfoTypeRelocsTable) void {
-    var it = table.valueIterator();
+    for (table.values()) |*value| {
-    while (it.next()) |value| {
         value.relocs.deinit(gpa);
     }
     table.deinit(gpa);
 @@ -2387,10 +2399,12 @@ fn finishUpdateDecl(
     // the buffer, so we have to do it before computing the offset, and we can't perform the actual
     // relocations yet.
     {
-        var it = dbg_info_type_relocs.iterator();
+        var it: usize = 0;
-        while (it.next()) |entry| {
+        while (it < dbg_info_type_relocs.count()) : (it += 1) {
-            entry.value_ptr.off = @intCast(u32, dbg_info_buffer.items.len);
+            const ty = dbg_info_type_relocs.keys()[it];
-            try self.addDbgInfoType(entry.key_ptr.*, dbg_info_buffer);
+            const value_ptr = dbg_info_type_relocs.getPtr(ty).?;
+            value_ptr.off = @intCast(u32, dbg_info_buffer.items.len);
+            try self.addDbgInfoType(ty, dbg_info_buffer, dbg_info_type_relocs);
         }
     }
 @@ -2401,8 +2415,7 @@ fn finishUpdateDecl(
     {
         // Now that we have the offset assigned we can finally perform type relocations.
-        var it = dbg_info_type_relocs.valueIterator();
+        for (dbg_info_type_relocs.values()) |value| {
-        while (it.next()) |value| {
             for (value.relocs.items) |off| {
                 mem.writeInt(
                     u32,
 @@ -2706,7 +2719,14 @@ pub fn updateDecl(self: *Elf, module: *Module, decl: *Module.Decl) !void {
 }
 /// Asserts the type has codegen bits.
-fn addDbgInfoType(self: *Elf, ty: Type, dbg_info_buffer: *std.ArrayList(u8)) !void {
+fn addDbgInfoType(
+    self: *Elf,
+    ty: Type,
+    dbg_info_buffer: *std.ArrayList(u8),
+    dbg_info_type_relocs: *File.DbgInfoTypeRelocsTable,
+) error{OutOfMemory}!void {
+    var reloc: ?struct { ty: Type, reloc: u32 } = null;
     switch (ty.zigTypeTag()) {
         .Void => unreachable,
         .NoReturn => unreachable,
 @@ -2747,11 +2767,34 @@ fn addDbgInfoType(self: *Elf, ty: Type, dbg_info_buffer: *std.ArrayList(u8)) !vo
                 try dbg_info_buffer.append(abbrev_pad1);
             }
         },
+        .Pointer => blk: {
+            if (ty.isSlice()) {
+                log.debug("TODO implement .debug_info for type '{}'", .{ty});
+                try dbg_info_buffer.append(abbrev_pad1);
+                break :blk;
+            }
+            try dbg_info_buffer.ensureUnusedCapacity(5);
+            dbg_info_buffer.appendAssumeCapacity(abbrev_ptr_type);
+            const index = dbg_info_buffer.items.len;
+            try dbg_info_buffer.resize(index + 4); // DW.AT.type, DW.FORM.ref4
+            reloc = .{ .ty = ty.childType(), .reloc = @intCast(u32, index) };
+        },
         else => {
             log.debug("TODO implement .debug_info for type '{}'", .{ty});
             try dbg_info_buffer.append(abbrev_pad1);
         },
     }
+    if (reloc) |rel| {
+        const gop = try dbg_info_type_relocs.getOrPut(self.base.allocator, rel.ty);
+        if (!gop.found_existing) {
+            gop.value_ptr.* = .{
+                .off = undefined,
+                .relocs = .{},
+            };
+        }
+        try gop.value_ptr.relocs.append(self.base.allocator, rel.reloc);
+    }
 }
 fn updateDeclDebugInfoAllocation(self: *Elf, text_block: *TextBlock, len: u32) !void {

src/link/MachO.zig+2-4

...	@@ -3578,8 +3578,7 @@ pub fn updateFunc(self: MachO, module: Module, func: *Module.Fn, air: Air, liv	...	@@ -3578,8 +3578,7 @@ pub fn updateFunc(self: MachO, module: Module, func: *Module.Fn, air: Air, liv
3578	if (debug_buffers) \|dbg\| {	3578	if (debug_buffers) \|dbg\| {
3579	dbg.dbg_line_buffer.deinit();	3579	dbg.dbg_line_buffer.deinit();
3580	dbg.dbg_info_buffer.deinit();	3580	dbg.dbg_info_buffer.deinit();
3581	var it = dbg.dbg_info_type_relocs.valueIterator();	3581	for (dbg.dbg_info_type_relocs.values()) \|*value\| {
3582	while (it.next()) \|value\| {
3583	value.relocs.deinit(self.base.allocator);	3582	value.relocs.deinit(self.base.allocator);
3584	}	3583	}
3585	dbg.dbg_info_type_relocs.deinit(self.base.allocator);	3584	dbg.dbg_info_type_relocs.deinit(self.base.allocator);
...	@@ -3659,8 +3658,7 @@ pub fn updateDecl(self: MachO, module: Module, decl: *Module.Decl) !void {	...	@@ -3659,8 +3658,7 @@ pub fn updateDecl(self: MachO, module: Module, decl: *Module.Decl) !void {
3659	if (debug_buffers) \|dbg\| {	3658	if (debug_buffers) \|dbg\| {
3660	dbg.dbg_line_buffer.deinit();	3659	dbg.dbg_line_buffer.deinit();
3661	dbg.dbg_info_buffer.deinit();	3660	dbg.dbg_info_buffer.deinit();
3662	var it = dbg.dbg_info_type_relocs.valueIterator();	3661	for (dbg.dbg_info_type_relocs.values()) \|*value\| {
3663	while (it.next()) \|value\| {
3664	value.relocs.deinit(self.base.allocator);	3662	value.relocs.deinit(self.base.allocator);
3665	}	3663	}
3666	dbg.dbg_info_type_relocs.deinit(self.base.allocator);	3664	dbg.dbg_info_type_relocs.deinit(self.base.allocator);

src/link/MachO/DebugSymbols.zig+1-2

...	@@ -1112,8 +1112,7 @@ pub fn commitDeclDebugInfo(	...	@@ -1112,8 +1112,7 @@ pub fn commitDeclDebugInfo(
1112		1112
1113	{	1113	{
1114	// Now that we have the offset assigned we can finally perform type relocations.	1114	// Now that we have the offset assigned we can finally perform type relocations.
1115	var it = dbg_info_type_relocs.valueIterator();	1115	for (dbg_info_type_relocs.values()) \|value\| {
1116	while (it.next()) \|value\| {
1117	for (value.relocs.items) \|off\| {	1116	for (value.relocs.items) \|off\| {
1118	mem.writeIntLittle(	1117	mem.writeIntLittle(
1119	u32,	1118	u32,