[dev.link] cmd/link: use more compact representation for external relocations

Currently, for external relocations, the ExtReloc structure
contains all the fields of the relocation. In fact, many of the
fields are the same with the original relocation. So, instead, we
can just use an index to reference the original relocation and
not expand the fields.

There is one place where we modify relocation type: changing
R_DWARFSECTREF to R_ADDR. Get away with it by changing
downstreams.

It also makes it easier to retrieve the reloc variant.

This reduces some allocation. Linking cmd/compile with external
linking,

name           old alloc/op   new alloc/op   delta
Reloc_GC         34.1MB ± 0%    22.7MB ± 0%  -33.30%  (p=0.000 n=5+4)

Change-Id: Id08a89ed2aee705296886d3b95014b806a0d55cf
Reviewed-on: https://go-review.googlesource.com/c/go/+/231217
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Jeremy Faller <jeremy@golang.org>

10 files changed, 55 insertions, 67 deletions

diff --git a/src/cmd/link/internal/amd64/asm.go b/src/cmd/link/internal/amd64/asm.go
index e1daaec2c7..e2c33b8001 100644
--- a/src/cmd/link/internal/amd64/asm.go
+++ b/src/cmd/link/internal/amd64/asm.go
@@ -400,7 +400,7 @@ func elfreloc1(ctxt *ld.Link, r *sym.Reloc, sectoff int64) bool {
 	switch r.Type {
 	default:
 		return false
-	case objabi.R_ADDR:
+	case objabi.R_ADDR, objabi.R_DWARFSECREF:
 		if r.Siz == 4 {
 			ctxt.Out.Write64(uint64(elf.R_X86_64_32) | uint64(elfsym)<<32)
 		} else if r.Siz == 8 {
diff --git a/src/cmd/link/internal/arm/asm.go b/src/cmd/link/internal/arm/asm.go
index c95de0de2f..a2024bcede 100644
--- a/src/cmd/link/internal/arm/asm.go
+++ b/src/cmd/link/internal/arm/asm.go
@@ -256,7 +256,7 @@ func elfreloc1(ctxt *ld.Link, r *sym.Reloc, sectoff int64) bool {
 	switch r.Type {
 	default:
 		return false
-	case objabi.R_ADDR:
+	case objabi.R_ADDR, objabi.R_DWARFSECREF:
 		if r.Siz == 4 {
 			ctxt.Out.Write32(uint32(elf.R_ARM_ABS32) | uint32(elfsym)<<8)
 		} else {
diff --git a/src/cmd/link/internal/arm64/asm.go b/src/cmd/link/internal/arm64/asm.go
index 2d12fc2f59..dc3e45d6c0 100644
--- a/src/cmd/link/internal/arm64/asm.go
+++ b/src/cmd/link/internal/arm64/asm.go
@@ -332,7 +332,7 @@ func elfreloc1(ctxt *ld.Link, r *sym.Reloc, sectoff int64) bool {
 	switch r.Type {
 	default:
 		return false
-	case objabi.R_ADDR:
+	case objabi.R_ADDR, objabi.R_DWARFSECREF:
 		switch r.Siz {
 		case 4:
 			ctxt.Out.Write64(uint64(elf.R_AARCH64_ABS32) | uint64(elfsym)<<32)
diff --git a/src/cmd/link/internal/ld/data.go b/src/cmd/link/internal/ld/data.go
index af1b335db7..102fcabe47 100644
--- a/src/cmd/link/internal/ld/data.go
+++ b/src/cmd/link/internal/ld/data.go
@@ -224,11 +224,7 @@ func (st *relocSymState) relocsym(s loader.Sym, P []byte) {
 		var rr loader.ExtReloc
 		needExtReloc := false // will set to true below in case it is needed
 		if target.IsExternal() {
-			rr.Sym = rs
-			rr.Type = rt
-			rr.Off = off
-			rr.Siz = uint8(siz)
-			rr.Add = r.Add()
+			rr.Idx = ri
 		}
 
 		// TODO(mundaym): remove this special case - see issue 14218.
@@ -282,14 +278,14 @@ func (st *relocSymState) relocsym(s loader.Sym, P []byte) {
 		case objabi.R_TLS_LE:
 			if target.IsExternal() && target.IsElf() {
 				needExtReloc = true
-				if rr.Sym == 0 {
-					rr.Sym = syms.Tlsg2
+				rr.Xsym = rs
+				if rr.Xsym == 0 {
+					rr.Xsym = syms.Tlsg2
 				}
-				rr.Xsym = rr.Sym
-				rr.Xadd = rr.Add
+				rr.Xadd = r.Add()
 				o = 0
 				if !target.IsAMD64() {
-					o = rr.Add
+					o = r.Add()
 				}
 				break
 			}
@@ -313,14 +309,14 @@ func (st *relocSymState) relocsym(s loader.Sym, P []byte) {
 		case objabi.R_TLS_IE:
 			if target.IsExternal() && target.IsElf() {
 				needExtReloc = true
-				if rr.Sym == 0 {
-					rr.Sym = syms.Tlsg2
+				rr.Xsym = rs
+				if rr.Xsym == 0 {
+					rr.Xsym = syms.Tlsg2
 				}
-				rr.Xsym = rr.Sym
-				rr.Xadd = rr.Add
+				rr.Xadd = r.Add()
 				o = 0
 				if !target.IsAMD64() {
-					o = rr.Add
+					o = r.Add()
 				}
 				break
 			}
@@ -342,7 +338,7 @@ func (st *relocSymState) relocsym(s loader.Sym, P []byte) {
 				// set up addend for eventual relocation via outer symbol.
 				rs := rs
 				rs, off := foldSubSymbolOffset(ldr, rs)
-				rr.Xadd = rr.Add + off
+				rr.Xadd = r.Add() + off
 				rst := ldr.SymType(rs)
 				if rst != sym.SHOSTOBJ && rst != sym.SDYNIMPORT && rst != sym.SUNDEFEXT && ldr.SymSect(rs) == nil {
 					st.err.Errorf(s, "missing section for relocation target %s", ldr.SymName(rs))
@@ -361,7 +357,7 @@ func (st *relocSymState) relocsym(s loader.Sym, P []byte) {
 				} else if target.IsWindows() {
 					// nothing to do
 				} else if target.IsAIX() {
-					o = ldr.SymValue(rr.Sym) + rr.Add
+					o = ldr.SymValue(rs) + r.Add()
 				} else {
 					st.err.Errorf(s, "unhandled pcrel relocation to %s on %v", ldr.SymName(rs), target.HeadType)
 				}
@@ -413,17 +409,8 @@ func (st *relocSymState) relocsym(s loader.Sym, P []byte) {
 					needExtReloc = false
 				}
 
-				// PE code emits IMAGE_REL_I386_SECREL and IMAGE_REL_AMD64_SECREL
-				// 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_DWARFSECREF with R_ADDR for windows -
-				// let PE code emit correct relocations.
-				if !target.IsWindows() {
-					rr.Type = objabi.R_ADDR
-				}
-
-				rr.Xsym = loader.Sym(ldr.SymSect(rr.Sym).Sym2)
-				rr.Xadd = rr.Add + ldr.SymValue(rr.Sym) - int64(ldr.SymSect(rr.Sym).Vaddr)
+				rr.Xsym = loader.Sym(ldr.SymSect(rs).Sym2)
+				rr.Xadd = r.Add() + ldr.SymValue(rs) - int64(ldr.SymSect(rs).Vaddr)
 
 				o = rr.Xadd
 				if target.IsElf() && target.IsAMD64() {
@@ -455,12 +442,12 @@ func (st *relocSymState) relocsym(s loader.Sym, P []byte) {
 		case objabi.R_GOTPCREL:
 			if target.IsDynlinkingGo() && target.IsDarwin() && rs != 0 && rst != sym.SCONST {
 				needExtReloc = true
-				rr.Xadd = rr.Add
-				rr.Xadd -= int64(rr.Siz) // relative to address after the relocated chunk
-				rr.Xsym = rr.Sym
+				rr.Xadd = r.Add()
+				rr.Xadd -= int64(siz) // relative to address after the relocated chunk
+				rr.Xsym = rs
 
 				o = rr.Xadd
-				o += int64(rr.Siz)
+				o += int64(siz)
 				break
 			}
 			fallthrough
@@ -470,9 +457,9 @@ func (st *relocSymState) relocsym(s loader.Sym, P []byte) {
 				needExtReloc = true
 				rr.Xadd = 0
 				if target.IsElf() {
-					rr.Xadd -= int64(rr.Siz)
+					rr.Xadd -= int64(siz)
 				}
-				rr.Xsym = rr.Sym
+				rr.Xsym = rs
 				o = 0
 				break
 			}
@@ -482,8 +469,8 @@ func (st *relocSymState) relocsym(s loader.Sym, P []byte) {
 				// set up addend for eventual relocation via outer symbol.
 				rs := rs
 				rs, off := foldSubSymbolOffset(ldr, rs)
-				rr.Xadd = rr.Add + off
-				rr.Xadd -= int64(rr.Siz) // relative to address after the relocated chunk
+				rr.Xadd = r.Add() + off
+				rr.Xadd -= int64(siz) // relative to address after the relocated chunk
 				rst := ldr.SymType(rs)
 				if rst != sym.SHOSTOBJ && rst != sym.SDYNIMPORT && ldr.SymSect(rs) == nil {
 					st.err.Errorf(s, "missing section for relocation target %s", ldr.SymName(rs))
@@ -496,25 +483,25 @@ func (st *relocSymState) relocsym(s loader.Sym, P []byte) {
 						o = 0
 					}
 				} else if target.IsDarwin() {
-					if rr.Type == objabi.R_CALL {
+					if rt == objabi.R_CALL {
 						if target.IsExternal() && rst == sym.SDYNIMPORT {
 							if target.IsAMD64() {
 								// AMD64 dynamic relocations are relative to the end of the relocation.
-								o += int64(rr.Siz)
+								o += int64(siz)
 							}
 						} else {
 							if rst != sym.SHOSTOBJ {
 								o += int64(uint64(ldr.SymValue(rs)) - ldr.SymSect(rs).Vaddr)
 							}
-							o -= int64(rr.Off) // relative to section offset, not symbol
+							o -= int64(off) // relative to section offset, not symbol
 						}
 					} else {
-						o += int64(rr.Siz)
+						o += int64(siz)
 					}
 				} else if target.IsWindows() && target.IsAMD64() { // only amd64 needs PCREL
 					// PE/COFF's PC32 relocation uses the address after the relocated
 					// bytes as the base. Compensate by skewing the addend.
-					o += int64(rr.Siz)
+					o += int64(siz)
 				} else {
 					st.err.Errorf(s, "unhandled pcrel relocation to %s on %v", ldr.SymName(rs), target.HeadType)
 				}
@@ -539,8 +526,8 @@ func (st *relocSymState) relocsym(s loader.Sym, P []byte) {
 				st.err.Errorf(s, "find XCOFF R_REF with internal linking")
 			}
 			needExtReloc = true
-			rr.Xsym = rr.Sym
-			rr.Xadd = rr.Add
+			rr.Xsym = rs
+			rr.Xadd = r.Add()
 
 			// This isn't a real relocation so it must not update
 			// its offset value.
diff --git a/src/cmd/link/internal/loader/loader.go b/src/cmd/link/internal/loader/loader.go
index a7b65e3580..749995bf8e 100644
--- a/src/cmd/link/internal/loader/loader.go
+++ b/src/cmd/link/internal/loader/loader.go
@@ -51,11 +51,7 @@ type Reloc struct {
 
 // ExtReloc contains the payload for an external relocation.
 type ExtReloc struct {
-	Off  int32            // offset to rewrite
-	Siz  uint8            // number of bytes to rewrite: 0, 1, 2, or 4
-	Type objabi.RelocType // the relocation type
-	Sym  Sym              // global index of symbol the reloc addresses
-	Add  int64            // addend
+	Idx  int // index of the original relocation
 	Xsym Sym
 	Xadd int64
 }
@@ -2763,25 +2759,30 @@ func (l *Loader) convertExtRelocs(dst *sym.Symbol, src Sym) {
 	if int(src) >= len(l.extRelocs) {
 		return
 	}
-	relocs := l.extRelocs[src]
-	if len(relocs) == 0 {
+	extRelocs := l.extRelocs[src]
+	if len(extRelocs) == 0 {
 		return
 	}
 	if len(dst.R) != 0 {
 		panic("bad")
 	}
-	dst.R = make([]sym.Reloc, len(relocs))
+	dst.R = make([]sym.Reloc, len(extRelocs))
+	relocs := l.Relocs(src)
 	for i := range dst.R {
-		sr := &relocs[i]
+		er := &extRelocs[i]
+		sr := relocs.At2(er.Idx)
 		r := &dst.R[i]
 		r.InitExt()
-		r.Off = sr.Off
-		r.Siz = sr.Siz
-		r.Type = sr.Type
-		r.Sym = l.Syms[sr.Sym]
-		r.Add = sr.Add
-		r.Xsym = l.Syms[sr.Xsym]
-		r.Xadd = sr.Xadd
+		r.Off = sr.Off()
+		r.Siz = sr.Siz()
+		r.Type = sr.Type()
+		r.Sym = l.Syms[l.ResolveABIAlias(sr.Sym())]
+		r.Add = sr.Add()
+		r.Xsym = l.Syms[er.Xsym]
+		r.Xadd = er.Xadd
+		if rv := l.RelocVariant(src, er.Idx); rv != 0 {
+			r.Variant = rv
+		}
 	}
 }
 
diff --git a/src/cmd/link/internal/mips/asm.go b/src/cmd/link/internal/mips/asm.go
index 53032a7e33..a366e80ea3 100644
--- a/src/cmd/link/internal/mips/asm.go
+++ b/src/cmd/link/internal/mips/asm.go
@@ -58,7 +58,7 @@ func elfreloc1(ctxt *ld.Link, r *sym.Reloc, sectoff int64) bool {
 	switch r.Type {
 	default:
 		return false
-	case objabi.R_ADDR:
+	case objabi.R_ADDR, objabi.R_DWARFSECREF:
 		if r.Siz != 4 {
 			return false
 		}
diff --git a/src/cmd/link/internal/mips64/asm.go b/src/cmd/link/internal/mips64/asm.go
index 33f8b33509..e69db29809 100644
--- a/src/cmd/link/internal/mips64/asm.go
+++ b/src/cmd/link/internal/mips64/asm.go
@@ -69,7 +69,7 @@ func elfreloc1(ctxt *ld.Link, r *sym.Reloc, sectoff int64) bool {
 	switch r.Type {
 	default:
 		return false
-	case objabi.R_ADDR:
+	case objabi.R_ADDR, objabi.R_DWARFSECREF:
 		switch r.Siz {
 		case 4:
 			ctxt.Out.Write8(uint8(elf.R_MIPS_32))
diff --git a/src/cmd/link/internal/ppc64/asm.go b/src/cmd/link/internal/ppc64/asm.go
index 7a5dc565e6..4dc50eab79 100644
--- a/src/cmd/link/internal/ppc64/asm.go
+++ b/src/cmd/link/internal/ppc64/asm.go
@@ -457,7 +457,7 @@ func elfreloc1(ctxt *ld.Link, r *sym.Reloc, sectoff int64) bool {
 	switch r.Type {
 	default:
 		return false
-	case objabi.R_ADDR:
+	case objabi.R_ADDR, objabi.R_DWARFSECREF:
 		switch r.Siz {
 		case 4:
 			ctxt.Out.Write64(uint64(elf.R_PPC64_ADDR32) | uint64(elfsym)<<32)
diff --git a/src/cmd/link/internal/s390x/asm.go b/src/cmd/link/internal/s390x/asm.go
index 4def25868a..bec7705be5 100644
--- a/src/cmd/link/internal/s390x/asm.go
+++ b/src/cmd/link/internal/s390x/asm.go
@@ -246,7 +246,7 @@ func elfreloc1(ctxt *ld.Link, r *sym.Reloc, sectoff int64) bool {
 		case 4:
 			ctxt.Out.Write64(uint64(elf.R_390_TLS_IEENT) | uint64(elfsym)<<32)
 		}
-	case objabi.R_ADDR:
+	case objabi.R_ADDR, objabi.R_DWARFSECREF:
 		switch r.Siz {
 		default:
 			return false
diff --git a/src/cmd/link/internal/x86/asm.go b/src/cmd/link/internal/x86/asm.go
index df704a444d..21ea5780cb 100644
--- a/src/cmd/link/internal/x86/asm.go
+++ b/src/cmd/link/internal/x86/asm.go
@@ -347,7 +347,7 @@ func elfreloc1(ctxt *ld.Link, r *sym.Reloc, sectoff int64) bool {
 	switch r.Type {
 	default:
 		return false
-	case objabi.R_ADDR:
+	case objabi.R_ADDR, objabi.R_DWARFSECREF:
 		if r.Siz == 4 {
 			ctxt.Out.Write32(uint32(elf.R_386_32) | uint32(elfsym)<<8)
 		} else {