cmd/link: support linker for linux/loong64

The basic arch-specific hooks are implemented, which
are used for internal and external linker.

Contributors to the loong64 port are:
  Weining Lu <luweining@loongson.cn>
  Lei Wang <wanglei@loongson.cn>
  Lingqin Gong <gonglingqin@loongson.cn>
  Xiaolin Zhao <zhaoxiaolin@loongson.cn>
  Meidan Li <limeidan@loongson.cn>
  Xiaojuan Zhai <zhaixiaojuan@loongson.cn>
  Qiyuan Pu <puqiyuan@loongson.cn>
  Guoqi Chen <chenguoqi@loongson.cn>

This port has been updated to Go 1.15.6:
  https://github.com/loongson/go

Updates #46229

Change-Id: I4680eb0635dd0fa3d6ea8348a2488da9c7e33d3b
Reviewed-on: https://go-review.googlesource.com/c/go/+/349514
Reviewed-by: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@google.com>
Auto-Submit: Ian Lance Taylor <iant@google.com>
Run-TryBot: Ian Lance Taylor <iant@google.com>

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

@@ -199,7 +199,7 @@ func mustLinkExternal(ctxt *Link) (res bool, reason string) {
 
 	// Internally linking cgo is incomplete on some architectures.
 	// https://golang.org/issue/14449
-	if iscgo && ctxt.Arch.InFamily(sys.MIPS64, sys.MIPS, sys.RISCV64) {
+	if iscgo && ctxt.Arch.InFamily(sys.Loong64, sys.MIPS64, sys.MIPS, sys.RISCV64) {
 		return true, buildcfg.GOARCH + " does not support internal cgo"
 	}
 	if iscgo && (buildcfg.GOOS == "android" || buildcfg.GOOS == "dragonfly") {

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

@@ -208,7 +208,7 @@ we write section and prog headers.
 func Elfinit(ctxt *Link) {
 	ctxt.IsELF = true
 
-	if ctxt.Arch.InFamily(sys.AMD64, sys.ARM64, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) {
+	if ctxt.Arch.InFamily(sys.AMD64, sys.ARM64, sys.Loong64, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) {
 		elfRelType = ".rela"
 	} else {
 		elfRelType = ".rel"
@@ -223,10 +223,13 @@ func Elfinit(ctxt *Link) {
 			ehdr.Flags = 2 /* Version 2 ABI */
 		}
 		fallthrough
-	case sys.AMD64, sys.ARM64, sys.MIPS64, sys.RISCV64:
+	case sys.AMD64, sys.ARM64, sys.Loong64, sys.MIPS64, sys.RISCV64:
 		if ctxt.Arch.Family == sys.MIPS64 {
 			ehdr.Flags = 0x20000004 /* MIPS 3 CPIC */
 		}
+		if ctxt.Arch.Family == sys.Loong64 {
+			ehdr.Flags = 0x3 /* LoongArch lp64d */
+		}
 		if ctxt.Arch.Family == sys.RISCV64 {
 			ehdr.Flags = 0x4 /* RISCV Float ABI Double */
 		}
@@ -1655,6 +1658,8 @@ func asmbElf(ctxt *Link) {
 		Exitf("unknown architecture in asmbelf: %v", ctxt.Arch.Family)
 	case sys.MIPS, sys.MIPS64:
 		eh.Machine = uint16(elf.EM_MIPS)
+	case sys.Loong64:
+		eh.Machine = uint16(elf.EM_LOONGARCH)
 	case sys.ARM:
 		eh.Machine = uint16(elf.EM_ARM)
 	case sys.AMD64:

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

@@ -1899,6 +1899,8 @@ func hostlinkArchArgs(arch *sys.Arch) []string {
 		if buildcfg.GOOS == "darwin" {
 			return []string{"-arch", "arm64"}
 		}
+	case sys.Loong64:
+		return []string{"-mabi=lp64d"}
 	case sys.MIPS64:
 		return []string{"-mabi=64"}
 	case sys.MIPS:

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

@@ -132,7 +132,7 @@ func computeDeferReturn(ctxt *Link, deferReturnSym, s loader.Sym) uint32 {
 				switch target.Arch.Family {
 				case sys.AMD64, sys.I386:
 					deferreturn--
-				case sys.ARM, sys.ARM64, sys.MIPS, sys.MIPS64, sys.PPC64, sys.RISCV64:
+				case sys.ARM, sys.ARM64, sys.Loong64, sys.MIPS, sys.MIPS64, sys.PPC64, sys.RISCV64:
 					// no change
 				case sys.S390X:
 					deferreturn -= 2

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

@@ -112,6 +112,10 @@ func (t *Target) IsMIPS64() bool {
 	return t.Arch.Family == sys.MIPS64
 }
 
+func (t *Target) IsLOONG64() bool {
+	return t.Arch.Family == sys.Loong64
+}
+
 func (t *Target) IsPPC64() bool {
 	return t.Arch.Family == sys.PPC64
 }

src/cmd/link/internal/loadelf/ldelf.go

@@ -346,6 +346,10 @@ func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, f *bio.Reader,
 		if mach != elf.EM_MIPS || class != elf.ELFCLASS64 {
 			return errorf("elf object but not mips64")
 		}
+	case sys.Loong64:
+		if mach != elf.EM_LOONGARCH || class != elf.ELFCLASS64 {
+			return errorf("elf object but not loong64")
+		}
 
 	case sys.ARM:
 		if e != binary.LittleEndian || mach != elf.EM_ARM || class != elf.ELFCLASS32 {
@@ -956,6 +960,7 @@ func relSize(arch *sys.Arch, pn string, elftype uint32) (uint8, uint8, error) {
 		ARM     = uint32(sys.ARM)
 		ARM64   = uint32(sys.ARM64)
 		I386    = uint32(sys.I386)
+		LOONG64 = uint32(sys.Loong64)
 		MIPS    = uint32(sys.MIPS)
 		MIPS64  = uint32(sys.MIPS64)
 		PPC64   = uint32(sys.PPC64)
@@ -991,6 +996,15 @@ func relSize(arch *sys.Arch, pn string, elftype uint32) (uint8, uint8, error) {
 		MIPS64 | uint32(elf.R_MIPS_GOT_DISP)<<16:
 		return 4, 4, nil
 
+	case LOONG64 | uint32(elf.R_LARCH_SOP_PUSH_PCREL)<<16,
+		LOONG64 | uint32(elf.R_LARCH_SOP_PUSH_GPREL)<<16,
+		LOONG64 | uint32(elf.R_LARCH_SOP_PUSH_ABSOLUTE)<<16,
+		LOONG64 | uint32(elf.R_LARCH_MARK_LA)<<16,
+		LOONG64 | uint32(elf.R_LARCH_SOP_POP_32_S_0_10_10_16_S2)<<16,
+		LOONG64 | uint32(elf.R_LARCH_64)<<16,
+		LOONG64 | uint32(elf.R_LARCH_MARK_PCREL)<<16:
+		return 4, 4, nil
+
 	case S390X | uint32(elf.R_390_8)<<16:
 		return 1, 1, nil
 

src/cmd/link/internal/loong64/asm.go

@@ -0,0 +1,240 @@
+// Copyright 2022 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package loong64
+
+import (
+	"cmd/internal/objabi"
+	"cmd/internal/sys"
+	"cmd/link/internal/ld"
+	"cmd/link/internal/loader"
+	"cmd/link/internal/sym"
+	"debug/elf"
+	"log"
+)
+
+func gentext(ctxt *ld.Link, ldr *loader.Loader) {}
+
+func adddynrel(target *ld.Target, ldr *loader.Loader, syms *ld.ArchSyms, s loader.Sym, r loader.Reloc, rIdx int) bool {
+	log.Fatalf("adddynrel not implemented")
+	return false
+}
+
+func elfreloc1(ctxt *ld.Link, out *ld.OutBuf, ldr *loader.Loader, s loader.Sym, r loader.ExtReloc, ri int, sectoff int64) bool {
+	// loong64 ELF relocation (endian neutral)
+	//		offset     uint64
+	//		symreloc   uint64  // The high 32-bit is the symbol, the low 32-bit is the relocation type.
+	//		addend     int64
+
+	elfsym := ld.ElfSymForReloc(ctxt, r.Xsym)
+	switch r.Type {
+	default:
+		return false
+	case objabi.R_ADDR, objabi.R_DWARFSECREF:
+		switch r.Size {
+		case 4:
+			out.Write64(uint64(sectoff))
+			out.Write64(uint64(elf.R_LARCH_32) | uint64(elfsym)<<32)
+			out.Write64(uint64(r.Xadd))
+		case 8:
+			out.Write64(uint64(sectoff))
+			out.Write64(uint64(elf.R_LARCH_64) | uint64(elfsym)<<32)
+			out.Write64(uint64(r.Xadd))
+		default:
+			return false
+		}
+	case objabi.R_ADDRLOONG64TLS:
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_PUSH_TLS_TPREL) | uint64(elfsym)<<32)
+		out.Write64(uint64(r.Xadd))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_PUSH_ABSOLUTE))
+		out.Write64(uint64(0xfff))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_AND))
+		out.Write64(uint64(0x0))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_POP_32_U_10_12))
+		out.Write64(uint64(0x0))
+
+	case objabi.R_ADDRLOONG64TLSU:
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_PUSH_TLS_TPREL) | uint64(elfsym)<<32)
+		out.Write64(uint64(r.Xadd))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_PUSH_ABSOLUTE))
+		out.Write64(uint64(0xc))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_SR))
+		out.Write64(uint64(0x0))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_POP_32_S_5_20) | uint64(0)<<32)
+		out.Write64(uint64(0x0))
+
+	case objabi.R_CALLLOONG64:
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_PUSH_PLT_PCREL) | uint64(elfsym)<<32)
+		out.Write64(uint64(r.Xadd))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_POP_32_S_0_10_10_16_S2))
+		out.Write64(uint64(0x0))
+	// The pcaddu12i + addi.d instructions is used to obtain address of a symbol on Loong64.
+	// The low 12-bit of the symbol address need to be added. The addi.d instruction have
+	// signed 12-bit immediate operand. The 0x800 (addr+U12 <=> addr+0x800+S12) is introduced
+	// to do sign extending from 12 bits. The 0x804 is 0x800 + 4, 4 is instruction bit
+	// width on Loong64 and is used to correct the PC of the addi.d instruction.
+	case objabi.R_ADDRLOONG64:
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_PUSH_PCREL) | uint64(elfsym)<<32)
+		out.Write64(uint64(r.Xadd + 0x4))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_PUSH_PCREL) | uint64(elfsym)<<32)
+		out.Write64(uint64(r.Xadd + 0x804))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_PUSH_ABSOLUTE))
+		out.Write64(uint64(0xc))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_SR))
+		out.Write64(uint64(0x0))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_PUSH_ABSOLUTE))
+		out.Write64(uint64(0xc))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_SL))
+		out.Write64(uint64(0x0))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_SUB))
+		out.Write64(uint64(0x0))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_POP_32_S_10_12))
+		out.Write64(uint64(0x0))
+
+	case objabi.R_ADDRLOONG64U:
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_PUSH_PCREL) | uint64(elfsym)<<32)
+		out.Write64(uint64(r.Xadd + 0x800))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_PUSH_ABSOLUTE))
+		out.Write64(uint64(0xc))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_SR))
+		out.Write64(uint64(0x0))
+
+		out.Write64(uint64(sectoff))
+		out.Write64(uint64(elf.R_LARCH_SOP_POP_32_S_5_20) | uint64(0)<<32)
+		out.Write64(uint64(0x0))
+	}
+
+	return true
+}
+
+func elfsetupplt(ctxt *ld.Link, plt, gotplt *loader.SymbolBuilder, dynamic loader.Sym) {
+	return
+}
+
+func machoreloc1(*sys.Arch, *ld.OutBuf, *loader.Loader, loader.Sym, loader.ExtReloc, int64) bool {
+	return false
+}
+
+func archreloc(target *ld.Target, ldr *loader.Loader, syms *ld.ArchSyms, r loader.Reloc, s loader.Sym, val int64) (o int64, nExtReloc int, ok bool) {
+	rs := r.Sym()
+	if target.IsExternal() {
+		nExtReloc := 0
+		switch r.Type() {
+		default:
+			return val, 0, false
+		case objabi.R_ADDRLOONG64,
+			objabi.R_ADDRLOONG64U:
+			// set up addend for eventual relocation via outer symbol.
+			rs, _ := ld.FoldSubSymbolOffset(ldr, rs)
+			rst := ldr.SymType(rs)
+			if rst != sym.SHOSTOBJ && rst != sym.SDYNIMPORT && ldr.SymSect(rs) == nil {
+				ldr.Errorf(s, "missing section for %s", ldr.SymName(rs))
+			}
+			nExtReloc = 8 // need 8 ELF relocations. see elfreloc1
+			if r.Type() == objabi.R_ADDRLOONG64U {
+				nExtReloc = 4
+			}
+			return val, nExtReloc, true
+		case objabi.R_ADDRLOONG64TLS,
+			objabi.R_ADDRLOONG64TLSU,
+			objabi.R_CALLLOONG64,
+			objabi.R_JMPLOONG64:
+			nExtReloc = 4
+			if r.Type() == objabi.R_CALLLOONG64 || r.Type() == objabi.R_JMPLOONG64 {
+				nExtReloc = 2
+			}
+			return val, nExtReloc, true
+		}
+	}
+
+	const isOk = true
+	const noExtReloc = 0
+
+	switch r.Type() {
+	case objabi.R_CONST:
+		return r.Add(), noExtReloc, isOk
+	case objabi.R_GOTOFF:
+		return ldr.SymValue(r.Sym()) + r.Add() - ldr.SymValue(syms.GOT), noExtReloc, isOk
+	case objabi.R_ADDRLOONG64,
+		objabi.R_ADDRLOONG64U:
+		pc := ldr.SymValue(s) + int64(r.Off())
+		t := ldr.SymAddr(rs) + r.Add() - pc
+		if r.Type() == objabi.R_ADDRLOONG64 {
+			return int64(val&0xffc003ff | (((t + 4 - ((t + 4 + 1<<11) >> 12 << 12)) << 10) & 0x3ffc00)), noExtReloc, isOk
+		}
+		return int64(val&0xfe00001f | (((t + 1<<11) >> 12 << 5) & 0x1ffffe0)), noExtReloc, isOk
+	case objabi.R_ADDRLOONG64TLS,
+		objabi.R_ADDRLOONG64TLSU:
+		t := ldr.SymAddr(rs) + r.Add()
+		if r.Type() == objabi.R_ADDRLOONG64TLS {
+			return int64(val&0xffc003ff | ((t & 0xfff) << 10)), noExtReloc, isOk
+		}
+		return int64(val&0xfe00001f | (((t) >> 12 << 5) & 0x1ffffe0)), noExtReloc, isOk
+	case objabi.R_CALLLOONG64,
+		objabi.R_JMPLOONG64:
+		pc := ldr.SymValue(s) + int64(r.Off())
+		t := ldr.SymAddr(rs) + r.Add() - pc
+		return int64(val&0xfc000000 | (((t >> 2) & 0xffff) << 10) | (((t >> 2) & 0x3ff0000) >> 16)), noExtReloc, isOk
+	}
+
+	return val, 0, false
+}
+
+func archrelocvariant(*ld.Target, *loader.Loader, loader.Reloc, sym.RelocVariant, loader.Sym, int64, []byte) int64 {
+	return -1
+}
+
+func extreloc(target *ld.Target, ldr *loader.Loader, r loader.Reloc, s loader.Sym) (loader.ExtReloc, bool) {
+	switch r.Type() {
+	case objabi.R_ADDRLOONG64,
+		objabi.R_ADDRLOONG64U:
+		return ld.ExtrelocViaOuterSym(ldr, r, s), true
+
+	case objabi.R_ADDRLOONG64TLS,
+		objabi.R_ADDRLOONG64TLSU,
+		objabi.R_CONST,
+		objabi.R_GOTOFF,
+		objabi.R_CALLLOONG64,
+		objabi.R_JMPLOONG64:
+		return ld.ExtrelocSimple(ldr, r), true
+	}
+	return loader.ExtReloc{}, false
+}

src/cmd/link/internal/loong64/l.go

@@ -0,0 +1,17 @@
+// Copyright 2022 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package loong64
+
+const (
+	maxAlign  = 32 // max data alignment
+	minAlign  = 1  // min data alignment
+	funcAlign = 8
+)
+
+/* Used by ../../internal/ld/dwarf.go */
+const (
+	dwarfRegSP = 3
+	dwarfRegLR = 1
+)

src/cmd/link/internal/loong64/obj.go

@@ -0,0 +1,58 @@
+// Copyright 2022 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package loong64
+
+import (
+	"cmd/internal/objabi"
+	"cmd/internal/sys"
+	"cmd/link/internal/ld"
+)
+
+func Init() (*sys.Arch, ld.Arch) {
+	arch := sys.ArchLoong64
+
+	theArch := ld.Arch{
+		Funcalign:        funcAlign,
+		Maxalign:         maxAlign,
+		Minalign:         minAlign,
+		Dwarfregsp:       dwarfRegSP,
+		Dwarfreglr:       dwarfRegLR,
+		Adddynrel:        adddynrel,
+		Archinit:         archinit,
+		Archreloc:        archreloc,
+		Archrelocvariant: archrelocvariant,
+		Extreloc:         extreloc,
+		Elfreloc1:        elfreloc1,
+		ElfrelocSize:     24,
+		Elfsetupplt:      elfsetupplt,
+		Machoreloc1:      machoreloc1,
+		Gentext:          gentext,
+
+		Linuxdynld:     "/lib64/ld.so.1",
+		Freebsddynld:   "XXX",
+		Openbsddynld:   "XXX",
+		Netbsddynld:    "XXX",
+		Dragonflydynld: "XXX",
+		Solarisdynld:   "XXX",
+	}
+
+	return arch, theArch
+}
+
+func archinit(ctxt *ld.Link) {
+	switch ctxt.HeadType {
+	default:
+		ld.Exitf("unknown -H option: %v", ctxt.HeadType)
+	case objabi.Hlinux: /* loong64 elf */
+		ld.Elfinit(ctxt)
+		ld.HEADR = ld.ELFRESERVE
+		if *ld.FlagTextAddr == -1 {
+			*ld.FlagTextAddr = 0x10000 + int64(ld.HEADR)
+		}
+		if *ld.FlagRound == -1 {
+			*ld.FlagRound = 0x10000
+		}
+	}
+}

src/cmd/link/internal/sym/reloc.go

@@ -59,6 +59,8 @@ func RelocName(arch *sys.Arch, r objabi.RelocType) string {
 			return elf.R_AARCH64(nr).String()
 		case sys.I386:
 			return elf.R_386(nr).String()
+		case sys.Loong64:
+			return elf.R_LARCH(nr).String()
 		case sys.MIPS, sys.MIPS64:
 			return elf.R_MIPS(nr).String()
 		case sys.PPC64:

src/cmd/link/link_test.go

@@ -175,6 +175,8 @@ func TestIssue33979(t *testing.T) {
 
 	// Skip test on platforms that do not support cgo internal linking.
 	switch runtime.GOARCH {
+	case "loong64":
+		t.Skipf("Skipping on %s/%s", runtime.GOOS, runtime.GOARCH)
 	case "mips", "mipsle", "mips64", "mips64le":
 		t.Skipf("Skipping on %s/%s", runtime.GOOS, runtime.GOARCH)
 	}

src/cmd/link/main.go

@@ -10,6 +10,7 @@ import (
 	"cmd/link/internal/arm"
 	"cmd/link/internal/arm64"
 	"cmd/link/internal/ld"
+	"cmd/link/internal/loong64"
 	"cmd/link/internal/mips"
 	"cmd/link/internal/mips64"
 	"cmd/link/internal/ppc64"
@@ -53,6 +54,8 @@ func main() {
 		arch, theArch = arm.Init()
 	case "arm64":
 		arch, theArch = arm64.Init()
+	case "loong64":
+		arch, theArch = loong64.Init()
 	case "mips", "mipsle":
 		arch, theArch = mips.Init()
 	case "mips64", "mips64le":