cmd/compile, cmd/link: support for DWARF file reference relocations

New relocation flavor R_DWARFFILEREF, to be applied to DWARF attribute
values that correspond to file references (ex: DW_AT_decl_file,
DW_AT_call_file). The LSym for this relocation is the file itself; the
linker replaces the relocation target with the index of the specified
file in the line table's file section.

Note: for testing purposes this patch changes the DWARF function
subprogram DIE abbrev to include DW_AT_decl_file (allowed by DWARF
but not especially useful) so as to have a way to test this
functionality. This attribute will be removed once there are other
file reference attributes (coming as part of inlining support).

Change-Id: Icf676beb60fcc33f06d78e747ef717532daaa3ba
Reviewed-on: https://go-review.googlesource.com/73330
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>

src/cmd/internal/dwarf/dwarf.go

@@ -123,6 +123,7 @@ type Context interface {
 	AddAddress(s Sym, t interface{}, ofs int64)
 	AddSectionOffset(s Sym, size int, t interface{}, ofs int64)
 	AddString(s Sym, v string)
+	AddFileRef(s Sym, f interface{})
 }
 
 // AppendUleb128 appends v to b using DWARF's unsigned LEB128 encoding.
@@ -303,6 +304,7 @@ var abbrevs = [DW_NABRV]dwAbbrev{
 			{DW_AT_low_pc, DW_FORM_addr},
 			{DW_AT_high_pc, DW_FORM_addr},
 			{DW_AT_frame_base, DW_FORM_block1},
+			{DW_AT_decl_file, DW_FORM_data4},
 			{DW_AT_external, DW_FORM_flag},
 		},
 	},
@@ -788,12 +790,14 @@ func PutRanges(ctxt Context, sym Sym, base Sym, ranges []Range) {
 
 // PutFunc writes a DIE for a function to s.
 // It also writes child DIEs for each variable in vars.
-func PutFunc(ctxt Context, info, loc, ranges Sym, name string, external bool, startPC Sym, size int64, scopes []Scope) error {
+func PutFunc(ctxt Context, info, loc, ranges, filesym Sym, name string, external bool, startPC Sym, size int64, scopes []Scope) error {
 	Uleb128put(ctxt, info, DW_ABRV_FUNCTION)
 	putattr(ctxt, info, DW_ABRV_FUNCTION, DW_FORM_string, DW_CLS_STRING, int64(len(name)), name)
 	putattr(ctxt, info, DW_ABRV_FUNCTION, DW_FORM_addr, DW_CLS_ADDRESS, 0, startPC)
 	putattr(ctxt, info, DW_ABRV_FUNCTION, DW_FORM_addr, DW_CLS_ADDRESS, size, startPC)
 	putattr(ctxt, info, DW_ABRV_FUNCTION, DW_FORM_block1, DW_CLS_BLOCK, 1, []byte{DW_OP_call_frame_cfa})
+	// DW_AT_decl_file attribute
+	ctxt.AddFileRef(info, filesym)
 	var ev int64
 	if external {
 		ev = 1

src/cmd/internal/obj/objfile.go

@@ -458,7 +458,14 @@ func (c dwCtxt) AddSectionOffset(s dwarf.Sym, size int, t interface{}, ofs int64
 	rsym := t.(*LSym)
 	ls.WriteAddr(c.Link, ls.Size, size, rsym, ofs)
 	r := &ls.R[len(ls.R)-1]
-	r.Type = objabi.R_DWARFREF
+	r.Type = objabi.R_DWARFSECREF
+}
+func (c dwCtxt) AddFileRef(s dwarf.Sym, f interface{}) {
+	ls := s.(*LSym)
+	rsym := f.(*LSym)
+	ls.WriteAddr(c.Link, ls.Size, 4, rsym, 0)
+	r := &ls.R[len(ls.R)-1]
+	r.Type = objabi.R_DWARFFILEREF
 }
 
 // dwarfSym returns the DWARF symbols for TEXT symbol.
@@ -480,6 +487,19 @@ func (s *LSym) Len() int64 {
 	return s.Size
 }
 
+// fileSymbol returns a symbol corresponding to the source file of the
+// first instruction (prog) of the specified function. This will
+// presumably be the file in which the function is defined.
+func (ctxt *Link) fileSymbol(fn *LSym) *LSym {
+	p := fn.Func.Text
+	if p != nil {
+		f, _ := linkgetlineFromPos(ctxt, p.Pos)
+		fsym := ctxt.Lookup(f)
+		return fsym
+	}
+	return nil
+}
+
 // populateDWARF fills in the DWARF Debugging Information Entries for TEXT symbol s.
 // The DWARFs symbol must already have been initialized in InitTextSym.
 func (ctxt *Link) populateDWARF(curfn interface{}, s *LSym) {
@@ -491,7 +511,9 @@ func (ctxt *Link) populateDWARF(curfn interface{}, s *LSym) {
 	if ctxt.DebugInfo != nil {
 		scopes = ctxt.DebugInfo(s, curfn)
 	}
-	err := dwarf.PutFunc(dwCtxt{ctxt}, info, loc, ranges, s.Name, !s.Static(), s, s.Size, scopes)
+
+	fs := ctxt.fileSymbol(s)
+	err := dwarf.PutFunc(dwCtxt{ctxt}, info, loc, ranges, fs, s.Name, !s.Static(), s, s.Size, scopes)
 	if err != nil {
 		ctxt.Diag("emitting DWARF for %s failed: %v", s.Name, err)
 	}

src/cmd/internal/objabi/reloctype.go

@@ -99,8 +99,15 @@ const (
 	// of a JMP instruction, by encoding the address into the instruction.
 	// The stack nosplit check ignores this since it is not a function call.
 	R_JMPMIPS
-	// R_DWARFREF resolves to the offset of the symbol from its section.
-	R_DWARFREF
+
+	// R_DWARFSECREF resolves to the offset of the symbol from its section.
+	// Target of relocation must be size 4 (in current implementation).
+	R_DWARFSECREF
+
+	// R_DWARFFILEREF resolves to an index into the DWARF .debug_line
+	// file table for the specified file symbol. Must be applied to an
+	// attribute of form DW_FORM_data4.
+	R_DWARFFILEREF
 
 	// Platform dependent relocations. Architectures with fixed width instructions
 	// have the inherent issue that a 32-bit (or 64-bit!) displacement cannot be

src/cmd/internal/objabi/reloctype_string.go

@@ -4,9 +4,9 @@ package objabi
 
 import "fmt"
 
-const _RelocType_name = "R_ADDRR_ADDRPOWERR_ADDRARM64R_ADDRMIPSR_ADDROFFR_WEAKADDROFFR_SIZER_CALLR_CALLARMR_CALLARM64R_CALLINDR_CALLPOWERR_CALLMIPSR_CONSTR_PCRELR_TLS_LER_TLS_IER_GOTOFFR_PLT0R_PLT1R_PLT2R_USEFIELDR_USETYPER_METHODOFFR_POWER_TOCR_GOTPCRELR_JMPMIPSR_DWARFREFR_ARM64_TLS_LER_ARM64_TLS_IER_ARM64_GOTPCRELR_POWER_TLS_LER_POWER_TLS_IER_POWER_TLSR_ADDRPOWER_DSR_ADDRPOWER_GOTR_ADDRPOWER_PCRELR_ADDRPOWER_TOCRELR_ADDRPOWER_TOCREL_DSR_PCRELDBLR_ADDRMIPSUR_ADDRMIPSTLS"
+const _RelocType_name = "R_ADDRR_ADDRPOWERR_ADDRARM64R_ADDRMIPSR_ADDROFFR_WEAKADDROFFR_SIZER_CALLR_CALLARMR_CALLARM64R_CALLINDR_CALLPOWERR_CALLMIPSR_CONSTR_PCRELR_TLS_LER_TLS_IER_GOTOFFR_PLT0R_PLT1R_PLT2R_USEFIELDR_USETYPER_METHODOFFR_POWER_TOCR_GOTPCRELR_JMPMIPSR_DWARFSECREFR_DWARFFILEREFR_ARM64_TLS_LER_ARM64_TLS_IER_ARM64_GOTPCRELR_POWER_TLS_LER_POWER_TLS_IER_POWER_TLSR_ADDRPOWER_DSR_ADDRPOWER_GOTR_ADDRPOWER_PCRELR_ADDRPOWER_TOCRELR_ADDRPOWER_TOCREL_DSR_PCRELDBLR_ADDRMIPSUR_ADDRMIPSTLS"
 
-var _RelocType_index = [...]uint16{0, 6, 17, 28, 38, 47, 60, 66, 72, 81, 92, 101, 112, 122, 129, 136, 144, 152, 160, 166, 172, 178, 188, 197, 208, 219, 229, 238, 248, 262, 276, 292, 306, 320, 331, 345, 360, 377, 395, 416, 426, 437, 450}
+var _RelocType_index = [...]uint16{0, 6, 17, 28, 38, 47, 60, 66, 72, 81, 92, 101, 112, 122, 129, 136, 144, 152, 160, 166, 172, 178, 188, 197, 208, 219, 229, 238, 251, 265, 279, 293, 309, 323, 337, 348, 362, 377, 394, 412, 433, 443, 454, 467}
 
 func (i RelocType) String() string {
 	i -= 1

src/cmd/link/internal/amd64/asm.go

@@ -507,7 +507,7 @@ func pereloc1(arch *sys.Arch, out *ld.OutBuf, s *sym.Symbol, r *sym.Reloc, secto
 	default:
 		return false
 
-	case objabi.R_DWARFREF:
+	case objabi.R_DWARFSECREF:
 		v = ld.IMAGE_REL_AMD64_SECREL
 
 	case objabi.R_ADDR:

src/cmd/link/internal/ld/data.go

@@ -110,6 +110,10 @@ func trampoline(ctxt *Link, s *sym.Symbol) {
 func relocsym(ctxt *Link, s *sym.Symbol) {
 	for ri := int32(0); ri < int32(len(s.R)); ri++ {
 		r := &s.R[ri]
+		if r.Done {
+			// Relocation already processed by an earlier phase.
+			continue
+		}
 		r.Done = true
 		off := r.Off
 		siz := int32(r.Siz)
@@ -145,6 +149,12 @@ func relocsym(ctxt *Link, s *sym.Symbol) {
 		if r.Siz == 0 { // informational relocation - no work to do
 			continue
 		}
+		if r.Type == objabi.R_DWARFFILEREF {
+			// These should have been processed previously during
+			// line table writing.
+			Errorf(s, "orphan R_DWARFFILEREF reloc to %v", r.Sym.Name)
+			continue
+		}
 
 		// We need to be able to reference dynimport symbols when linking against
 		// shared libraries, and Solaris needs it always
@@ -306,7 +316,7 @@ func relocsym(ctxt *Link, s *sym.Symbol) {
 				Errorf(s, "non-pc-relative relocation address for %s is too big: %#x (%#x + %#x)", r.Sym.Name, uint64(o), Symaddr(r.Sym), r.Add)
 				errorexit()
 			}
-		case objabi.R_DWARFREF:
+		case objabi.R_DWARFSECREF:
 			if r.Sym.Sect == nil {
 				Errorf(s, "missing DWARF section for relocation target %s", r.Sym.Name)
 			}
@@ -324,9 +334,9 @@ func relocsym(ctxt *Link, s *sym.Symbol) {
 				}
 
 				// PE code emits IMAGE_REL_I386_SECREL and IMAGE_REL_AMD64_SECREL
-				// for R_DWARFREF relocations, while R_ADDR is replaced with
+				// for R_DWARFSECREF relocations, while R_ADDR is replaced with
 				// IMAGE_REL_I386_DIR32, IMAGE_REL_AMD64_ADDR64 and IMAGE_REL_AMD64_ADDR32.
-				// Do not replace R_DWARFREF with R_ADDR for windows -
+				// Do not replace R_DWARFSECREF with R_ADDR for windows -
 				// let PE code emit correct relocations.
 				if ctxt.HeadType != objabi.Hwindows {
 					r.Type = objabi.R_ADDR

src/cmd/link/internal/ld/dwarf.go

@@ -62,10 +62,14 @@ func (c dwctxt) AddSectionOffset(s dwarf.Sym, size int, t interface{}, ofs int64
 		ls.AddAddrPlus4(t.(*sym.Symbol), 0)
 	}
 	r := &ls.R[len(ls.R)-1]
-	r.Type = objabi.R_DWARFREF
+	r.Type = objabi.R_DWARFSECREF
 	r.Add = ofs
 }
 
+func (c dwctxt) AddFileRef(s dwarf.Sym, f interface{}) {
+	panic("should be used only in the compiler")
+}
+
 var gdbscript string
 
 var dwarfp []*sym.Symbol
@@ -220,7 +224,7 @@ func adddwarfref(ctxt *Link, s *sym.Symbol, t *sym.Symbol, size int) int64 {
 		result = s.AddAddrPlus4(t, 0)
 	}
 	r := &s.R[len(s.R)-1]
-	r.Type = objabi.R_DWARFREF
+	r.Type = objabi.R_DWARFSECREF
 	return result
 }
 
@@ -1030,7 +1034,7 @@ func importInfoSymbol(ctxt *Link, dsym *sym.Symbol) {
 	dsym.Attr |= sym.AttrNotInSymbolTable | sym.AttrReachable
 	dsym.Type = sym.SDWARFINFO
 	for _, r := range dsym.R {
-		if r.Type == objabi.R_DWARFREF && r.Sym.Size == 0 {
+		if r.Type == objabi.R_DWARFSECREF && r.Sym.Size == 0 {
 			if ctxt.BuildMode == BuildModeShared {
 				// These type symbols may not be present in BuildModeShared. Skip.
 				continue
@@ -1184,6 +1188,40 @@ func writelines(ctxt *Link, lib *sym.Library, textp []*sym.Symbol, ls *sym.Symbo
 	ls.SetUint32(ctxt.Arch, unitLengthOffset, uint32(ls.Size-unitstart))
 	ls.SetUint32(ctxt.Arch, headerLengthOffset, uint32(headerend-headerstart))
 
+	// Apply any R_DWARFFILEREF relocations, since we now know the
+	// line table file indices for this compilation unit. Note that
+	// this loop visits only subprogram DIEs: if the compiler is
+	// changed to generate DW_AT_decl_file attributes for other
+	// DIE flavors (ex: variables) then those DIEs would need to
+	// be included below.
+	for fidx := 0; fidx < len(funcs); fidx++ {
+		f := funcs[fidx]
+		for ri := 0; ri < len(f.R); ri++ {
+			r := &f.R[ri]
+			if r.Type != objabi.R_DWARFFILEREF {
+				continue
+			}
+			// Mark relocation as applied (signal to relocsym)
+			r.Done = true
+			idx, ok := fileNums[int(r.Sym.Value)]
+			if ok {
+				if int(int32(idx)) != idx {
+					Errorf(f, "bad R_DWARFFILEREF relocation: file index overflow")
+				}
+				if r.Siz != 4 {
+					Errorf(f, "bad R_DWARFFILEREF relocation: has size %d, expected 4", r.Siz)
+				}
+				if r.Off < 0 || r.Off+4 > int32(len(f.P)) {
+					Errorf(f, "bad R_DWARFFILEREF relocation offset %d + 4 would write past length %d", r.Off, len(s.P))
+					continue
+				}
+				ctxt.Arch.ByteOrder.PutUint32(f.P[r.Off:r.Off+4], uint32(idx))
+			} else {
+				Errorf(f, "R_DWARFFILEREF relocation file missing: %v", r.Sym)
+			}
+		}
+	}
+
 	return dwinfo, funcs
 }
 
@@ -1617,7 +1655,7 @@ func collectlocs(ctxt *Link, syms []*sym.Symbol, units []*compilationUnit) []*sy
 	for _, u := range units {
 		for _, fn := range u.funcDIEs {
 			for _, reloc := range fn.R {
-				if reloc.Type == objabi.R_DWARFREF && strings.HasPrefix(reloc.Sym.Name, dwarf.LocPrefix) {
+				if reloc.Type == objabi.R_DWARFSECREF && strings.HasPrefix(reloc.Sym.Name, dwarf.LocPrefix) {
 					reloc.Sym.Attr |= sym.AttrReachable | sym.AttrNotInSymbolTable
 					syms = append(syms, reloc.Sym)
 					empty = false

src/cmd/link/internal/ld/dwarf_test.go

@@ -299,7 +299,7 @@ func main() {
 	}
 }
 
-func TestVarDeclCoords(t *testing.T) {
+func TestVarDeclCoordsAndSubrogramDeclFile(t *testing.T) {
 	testenv.MustHaveGoBuild(t)
 
 	if runtime.GOOS == "plan9" {
@@ -329,6 +329,7 @@ func main() {
 
 	rdr := d.Reader()
 	var iEntry *dwarf.Entry
+	var pEntry *dwarf.Entry
 	foundMain := false
 	for entry, err := rdr.Next(); entry != nil; entry, err = rdr.Next() {
 		if err != nil {
@@ -336,6 +337,7 @@ func main() {
 		}
 		if entry.Tag == dwarf.TagSubprogram && entry.Val(dwarf.AttrName).(string) == "main.main" {
 			foundMain = true
+			pEntry = entry
 			continue
 		}
 		if !foundMain {
@@ -354,4 +356,9 @@ func main() {
 	if line == nil || line.(int64) != 5 {
 		t.Errorf("DW_AT_decl_line for i is %v, want 5", line)
 	}
+
+	file := pEntry.Val(dwarf.AttrDeclFile)
+	if file == nil || file.(int64) != 1 {
+		t.Errorf("DW_AT_decl_file for main is %v, want 1", file)
+	}
 }

src/cmd/link/internal/x86/asm.go

@@ -471,7 +471,7 @@ func pereloc1(arch *sys.Arch, out *ld.OutBuf, s *sym.Symbol, r *sym.Reloc, secto
 	default:
 		return false
 
-	case objabi.R_DWARFREF:
+	case objabi.R_DWARFSECREF:
 		v = ld.IMAGE_REL_I386_SECREL
 
 	case objabi.R_ADDR: