feat(asm): encode the arm64 local-exec TLS load

Assisted-by: GLM 5.3 Flash

asm/arm64_assemble.go

@@ -30,8 +30,9 @@ import (
 // (the morestack check in the prologue and the call back into the runtime in
 // the epilogue) is not emitted, so the bytes match only for NOSPLIT functions
 // or zero-frame leaves, where the toolchain emits no guard either.
-func assembleARM64(t *ast.Text) ([]byte, map[string]int, []Reloc, []LineEntry, []SpadjStep, []Arm64Literal, error) {
+func assembleARM64(t *ast.Text, tls map[string]bool) ([]byte, map[string]int, []Reloc, []LineEntry, []SpadjStep, []Arm64Literal, error) {
 	fi := arm64ComputeFrame(t)
+	fi.tls = tls
 	prologue := arm64Prologue(fi)
 	guardLen := arm64GuardLen(fi)
 	chain := arm64JumpChain(t)
@@ -1989,6 +1990,24 @@ func encodeARM64Mov(instr *ast.Instr, mnem string, wb string, fi arm64FrameInfo,
 		if rd < 0 {
 			return nil, fmt.Errorf("%s sym(SB): invalid destination register", mnem)
 		}
+		// A TLSBSS symbol's load rides the local-exec model: one MOVZ word
+		// with the R_ARM64_TLS_LE relocation (asm7.go case 69), the width
+		// rows existing for MOVD alone.
+		if fi.tls[src.Addr.Sym.Name] {
+			if mnem != "MOVD" && mnem != "MOV" {
+				return nil, fmt.Errorf("%s: illegal combination: the TLS load exists for MOVD only", mnem)
+			}
+			if relocs != nil {
+				*relocs = append(*relocs, Reloc{
+					Off:    0,
+					After:  4,
+					Name:   src.Addr.Sym.Name,
+					Addend: src.Addr.Sym.Offset,
+					Kind:   RelArm64TLSLE,
+				})
+			}
+			return a64wordLE(a64MoveWide(1, 2, 0, 0, uint32(rd))), nil
+		}
 		return encodeARM64SBLoad(src.Addr.Sym, rd, mnem, relocs)
 	}
 	if dst.Addr.Sym != nil && dst.Addr.Sym.Pseudo == "SB" && isMemOperand(dst) {
@@ -2116,6 +2135,9 @@ func arm64MovSize(mnem string, ops []*ast.Operand, fi arm64FrameInfo) int {
 		if mnem == "FMOVQ" {
 			return 12 // ADRP + ADD + LDR: the unaligned-access fallback
 		}
+		if fi.tls != nil && fi.tls[src.Addr.Sym.Name] {
+			return 4 // the TLS-LE load is one MOVZ word
+		}
 		return 8 // ADRP + LDR
 	case dst.Addr.Sym != nil && dst.Addr.Sym.Pseudo == "SB":
 		if mnem == "FMOVQ" {
@@ -5771,6 +5793,22 @@ func AssembleFileARM64(f *ast.File) (*Image, error) {
 		return nil, err
 	}
 
+	// The TLS symbols: a GLOBL marked TLSBSS makes the toolchain's aclass
+	// key its load off objabi.STLSBSS (asm7.go C_TLSIE), so the loads of
+	// these names encode as the local-exec MOVZ with the TLS_LE relocation.
+	tlsSyms := map[string]bool{}
+	for _, d := range f.Decls {
+		g, ok := d.(*ast.Globl)
+		if !ok || g.Name == nil {
+			continue
+		}
+		for _, fl := range g.Flags {
+			if fl == "TLSBSS" {
+				tlsSyms[g.Name.Name] = true
+			}
+		}
+	}
+
 	img := &Image{Symbols: map[string]int{}, SourcePath: f.Path}
 	var pendingLits []Arm64Literal
 	litSeen := map[string]bool{}
@@ -5779,7 +5817,7 @@ func AssembleFileARM64(f *ast.File) (*Image, error) {
 		if !ok {
 			continue
 		}
-		code, labels, relocs, lines, spadj, lits, err := assembleARM64(t)
+		code, labels, relocs, lines, spadj, lits, err := assembleARM64(t, tlsSyms)
 		if err != nil {
 			return nil, fmt.Errorf("%s: %w", t.Name.Name, err)
 		}