feat(asm): emit the amd64 stack-split guard and morestack block

2026-09-14 20:35:55 +02:00
parent 89fa6ea15e
commit 1d8e68c574
5 changed files with 500 additions and 23 deletions
@@ -51,6 +51,7 @@ type sbPatch struct {
 	after  int
 	name   string
 	addend int64
+	kind   RelocKind
 }

 // spadjStep is one stack-adjustment boundary within a function: Value is the
@@ -75,13 +76,18 @@ func assemble(t *ast.Text, link *linkInfo) ([]byte, []sbPatch, map[string]int, [
 		return name
 	}

-	// Layout: iterate jump sizes to a fixed point.
+	// Layout: iterate jump sizes to a fixed point.  The stack-split guard
+	// prefix and the trailing morestack block participate in the iteration:
+	// their conditional branches relax from rel8 to rel32 when the body
+	// outgrows the short form.
 	long := make([]bool, len(t.Body))
 	sizes := make([]int, len(t.Body))
 	offsets := map[string]int{}
 	pcs := make([]int, len(t.Body))
+	var guardJBlong, guardJBElong, moreJMPlong bool
 	for {
-		pos := len(fi.prologue)
+		guard := fi.guardLen(guardJBlong, guardJBElong)
+		pos := guard + len(fi.prologue)
 		for i, stmt := range t.Body {
 			switch s := stmt.(type) {
 			case *ast.Label:
@@ -96,6 +102,7 @@ func assemble(t *ast.Text, link *linkInfo) ([]byte, []sbPatch, map[string]int, [
 				pos += sz
 			}
 		}
+		bodyLen := pos - (guard + len(fi.prologue))
 		// Expand any short jump whose displacement no longer fits rel8.
 		changed := false
 		for i, stmt := range t.Body {
@@ -121,25 +128,75 @@ func assemble(t *ast.Text, link *linkInfo) ([]byte, []sbPatch, map[string]int, [
 				changed = true
 			}
 		}
+		// The guard's conditional branches target the morestack block, which
+		// starts right after the body: the JBE measures from the end of the
+		// guard, so its displacement is the prologue plus the body.
+		if !guardJBElong && !fits8(int64(len(fi.prologue)+bodyLen)) {
+			guardJBElong = true
+			changed = true
+		}
+		if fi.splitClass == 2 && !guardJBlong {
+			// The underflow JB sits before the CMPQ; its displacement spans
+			// the rest of the guard plus the prologue and the body.
+			jbLen := 2
+			if guardJBlong {
+				jbLen = 6
+			}
+			rest := fi.guardLen(guardJBlong, guardJBElong) - (9 + 3 + 7 + jbLen)
+			if !fits8(int64(rest + len(fi.prologue) + bodyLen)) {
+				guardJBlong = true
+				changed = true
+			}
+		}
+		// The morestack JMP returns to the function start, so its
+		// displacement is the negated distance from its own end.
+		if !moreJMPlong {
+			jmpLen := 2
+			if moreJMPlong {
+				jmpLen = 5
+			}
+			if !fits8(-int64(guard + len(fi.prologue) + bodyLen + 5 + jmpLen)) {
+				moreJMPlong = true
+				changed = true
+			}
+		}
 		if !changed {
 			break
 		}
 	}

-	// Pass 2: emit.
-	out := append([]byte(nil), fi.prologue...)
+	// Pass 2: emit.  The guard comes first, then the prologue, the body and
+	// the morestack block.
+	guardLen := fi.guardLen(guardJBlong, guardJBElong)
+	bodyLen := 0
+	{
+		pos := guardLen + len(fi.prologue)
+		for i, stmt := range t.Body {
+			if _, ok := stmt.(*ast.Instr); ok {
+				pos += sizes[i]
+			}
+		}
+		bodyLen = pos - (guardLen + len(fi.prologue))
+	}
+	var out []byte
 	var patches []sbPatch
+	if fi.needSplit {
+		guard, tlsPatch := buildGuard(fi, int32(len(fi.prologue)+bodyLen), int32(fi.guardLen(guardJBlong, guardJBElong)-(9+3+7+2)+len(fi.prologue)+bodyLen))
+		out = append(out, guard...)
+		patches = append(patches, tlsPatch)
+	}
+	out = append(out, fi.prologue...)
 	var steps []spadjStep
 	var lines []LineEntry
 	if fi.useFP {
 		// PUSHQ BP saves the return-address-relative base (+8); the MOVQ
 		// changes nothing; SUBQ $size, SP completes the frame.
 		steps = append(steps,
-			spadjStep{1, 8},
-			spadjStep{len(fi.prologue), 8 + fi.size},
+			spadjStep{guardLen + 1, 8},
+			spadjStep{guardLen + len(fi.prologue), 8 + fi.size},
 		)
 	}
-	pos := len(fi.prologue)
+	pos := guardLen + len(fi.prologue)
 	for i, stmt := range t.Body {
 		s, ok := stmt.(*ast.Instr)
 		if !ok {
@@ -161,11 +218,32 @@ func assemble(t *ast.Text, link *linkInfo) ([]byte, []sbPatch, map[string]int, [
 		if len(code) != sizes[i] {
 			return nil, nil, nil, nil, nil, fmt.Errorf("%s: size mismatch (%d vs %d)", s.Mnemonic.Text, len(code), sizes[i])
 		}
+		if strings.ToUpper(s.Mnemonic.Text) == "CALL" {
+			for k := range ps {
+				ps[k].kind = RelCall
+			}
+		}
 		patches = append(patches, ps...)
 		lines = append(lines, LineEntry{Offset: pos, Line: s.Pos().Line})
 		out = append(out, code...)
 		pos += len(code)
 	}
+	if fi.needSplit {
+		// The morestack block: CALL runtime.morestack_noctxt, then a JMP
+		// back to the function entry.
+		jmpLen := 2
+		if moreJMPlong {
+			jmpLen = 5
+		}
+		jmpDisp := -int64(pos + 5 + jmpLen)
+		suffix, callPatch := buildMoreStack(int32(jmpDisp))
+		callPatch.off += pos
+		callPatch.after = pos + 5
+		patches = append(patches, callPatch)
+		out = append(out, suffix...)
+		pos += len(suffix)
+	}
+	_ = pos
 	return out, patches, offsets, steps, lines, nil
 }

@@ -229,10 +307,31 @@ type frameInfo struct {
 	spAdjust int64 // x-N(SP) becomes (spAdjust - N)(SP)
 	prologue []byte
 	epilogue []byte
+
+	// Stack-split guard state (matching the toolchain's stacksplit): needSplit
+	// is false for NOSPLIT functions and for leaf functions whose frame is
+	// below StackSmall, which the toolchain auto-marks NOSPLIT.
+	needSplit  bool
+	splitClass int // 0: <=StackSmall, 1: <=StackBig, 2: >StackBig
+	framesize  int // the size the guard checks: frame+8 for framed functions
 }

+// Stack-frame size classes from runtime/stack.go.
+const (
+	stackSmall = 128
+	stackBig   = 4096
+)
+
+// sbPatch gains a kind so the emitters can tell CALL and TLS patches from
+// plain PC-relative displacements.
+
 // computeFrame derives the frame layout, matching the Go assembler's default
-// (a frame pointer is used whenever the function has a non-zero frame).
+// (a frame pointer is used whenever the function has a non-zero frame).  It
+// also decides whether the function needs the stack-split guard, mirroring
+// obj6: a NOSPLIT function never splits, and a leaf function whose frame is
+// below StackSmall is auto-marked NOSPLIT.  One deliberate deviation: the
+// toolchain treats zero-argument runtime calls (duffcopy and friends) as
+// leaf-compatible; here any CALL makes the function a non-leaf.
 func computeFrame(t *ast.Text) frameInfo {
 	fi := frameInfo{}
 	if t.Frame != nil && t.Frame.Imm.HasVal {
@@ -247,9 +346,141 @@ func computeFrame(t *ast.Text) frameInfo {
 	} else {
 		fi.fpAdjust = 8 // return address only
 	}
+
+	noSplit := false
+	for _, f := range t.Flags {
+		if strings.EqualFold(f, "NOSPLIT") {
+			noSplit = true
+		}
+	}
+	// The toolchain's autoffset: the frame plus the saved base pointer.
+	framesize := fi.size
+	if framesize > 0 {
+		framesize += 8
+	}
+	switch {
+	case noSplit:
+	case framesize < stackSmall && !hasCall(t):
+		// Auto-NOSPLIT, as the toolchain's leaf search concludes.
+	default:
+		fi.needSplit = true
+		fi.framesize = framesize
+		switch {
+		case framesize <= stackSmall:
+			fi.splitClass = 0
+		case framesize <= stackBig:
+			fi.splitClass = 1
+		default:
+			fi.splitClass = 2
+		}
+	}
 	return fi
 }

+// hasCall reports whether the function body contains a CALL instruction.
+func hasCall(t *ast.Text) bool {
+	for _, stmt := range t.Body {
+		in, ok := stmt.(*ast.Instr)
+		if !ok {
+			continue
+		}
+		if strings.ToUpper(in.Mnemonic.Text) == "CALL" {
+			return true
+		}
+	}
+	return false
+}
+
+// guardLen returns the byte length of the stack-split guard prefix.  The
+// final conditional branch (JBE, and JB in the big class) is 2 bytes in the
+// short form and 6 in the long form.
+func (fi frameInfo) guardLen(jbLong, jbeLong bool) int {
+	if !fi.needSplit {
+		return 0
+	}
+	jb, jbe := 2, 2
+	if jbLong {
+		jb = 6
+	}
+	if jbeLong {
+		jbe = 6
+	}
+	switch fi.splitClass {
+	case 0:
+		return 9 + 4 + jbe
+	case 1:
+		return 9 + 8 + 4 + jbe
+	default:
+		return 9 + 3 + 7 + jb + 4 + jbe
+	}
+}
+
+// moreLen returns the byte length of the trailing morestack block: the CALL
+// (always rel32) plus the JMP back to the function start.
+func moreLen(jmpLong bool) int {
+	jmp := 2
+	if jmpLong {
+		jmp = 5
+	}
+	return 5 + jmp
+}
+
+// buildGuard emits the stack-split guard prefix.  jbeDisp and jbDisp are the
+// already-computed displacements of the conditional branches that jump to the
+// morestack block (unused in classes without them).  The TLS load carries a
+// R_TLS_LE patch site at offset 5.
+func buildGuard(fi frameInfo, jbeDisp, jbDisp int32) ([]byte, sbPatch) {
+	out := []byte{
+		0x64, 0x4c, 0x8b, 0x34, 0x25, // MOVQ FS:0, R14
+		0, 0, 0, 0, // TLS slot offset, filled by the linker
+	}
+	tls := sbPatch{off: 5, after: 9, kind: RelTLSLE}
+	jmp := func(op8, op32 byte, disp int32) []byte {
+		if disp >= -128 && disp <= 127 {
+			return []byte{op8, byte(disp)}
+		}
+		return append([]byte{0x0F, op32}, le32(int64(disp))...)
+	}
+	switch fi.splitClass {
+	case 0:
+		// CMPQ SP, 16(R14)
+		out = append(out, 0x49, 0x3b, 0x66, 0x10)
+		out = append(out, jmp(0x76, 0x86, jbeDisp)...)
+	case 1:
+		// LEAQ -(framesize-StackSmall)(SP), R12; CMPQ R12, 16(R14)
+		out = append(out, 0x4c, 0x8d, 0xa4, 0x24)
+		out = append(out, le32(-int64(fi.framesize-stackSmall))...)
+		out = append(out, 0x4d, 0x3b, 0x66, 0x10)
+		out = append(out, jmp(0x76, 0x86, jbeDisp)...)
+	default:
+		// MOVQ SP, R12; SUBQ $(framesize-StackSmall), R12; JB; CMPQ R12, 16(R14)
+		out = append(out, 0x49, 0x89, 0xe4)
+		out = append(out, 0x49, 0x81, 0xec)
+		out = append(out, le32(int64(fi.framesize-stackSmall))...)
+		out = append(out, jmp(0x72, 0x82, jbDisp)...)
+		out = append(out, 0x4d, 0x3b, 0x66, 0x10)
+		out = append(out, jmp(0x76, 0x86, jbeDisp)...)
+	}
+	return out, tls
+}
+
+// buildMoreStack emits the trailing block: CALL runtime.morestack_noctxt
+// (patched by the linker) and a JMP back to the function start.
+func buildMoreStack(jmpDisp int32) ([]byte, sbPatch) {
+	out := []byte{0xE8, 0, 0, 0, 0}
+	call := sbPatch{off: 1, after: 5, name: "runtime\u00b7morestack_noctxt", kind: RelCall}
+	out = append(out, jmpBytes(jmpDisp)...)
+	return out, call
+}
+
+// jmpBytes encodes a near JMP in the short or long form.
+func jmpBytes(disp int32) []byte {
+	if disp >= -128 && disp <= 127 {
+		return []byte{0xEB, byte(disp)}
+	}
+	return append([]byte{0xE9}, le32(int64(disp))...)
+}
+
 // prologueBytes emits: PUSHQ BP; MOVQ SP, BP; SUBQ $size, SP.
 func prologueBytes(size int) []byte {
 	out := []byte{0x55, 0x48, 0x89, 0xE5} // PUSHQ BP; MOVQ SP, BP
@@ -333,6 +564,23 @@ func encodeInstr(s *ast.Instr, pc int, offsets map[string]int, fi frameInfo, lon
 	var ps []sbPatch
 	var err error
 	if isJumpMnemonic(mnem) {
+		if mnem == "CALL" && isSBCall(s) {
+			// CALL sym(SB): a rel32 call against a static or external
+			// symbol, resolved by the file-level layout or the linker.
+			code, ps, err = encodeSBCall(s, link)
+			if err != nil {
+				return nil, nil, err
+			}
+			for i := range ps {
+				ps[i].kind = RelCall
+			}
+			body := pc + len(prefix)
+			for i := range ps {
+				ps[i].off += body
+				ps[i].after = body + len(code)
+			}
+			return append(prefix, code...), ps, nil
+		}
 		code, err = encodeJump(s, mnem, pc+len(prefix), offsets, long, resolve)
 	} else {
 		code, ps, err = encodeNormal(s, fi, link)
@@ -414,6 +662,33 @@ func encodeJump(s *ast.Instr, mnem string, pc int, offsets map[string]int, long
 	}
 }

+// isSBCall reports whether the CALL operand is a symbol reference.
+func isSBCall(s *ast.Instr) bool {
+	return len(s.Operands) == 1 && s.Operands[0].Kind == ast.OpAddr &&
+		s.Operands[0].Addr.Sym != nil && s.Operands[0].Addr.Sym.Pseudo == "SB"
+}
+
+// encodeSBCall encodes CALL sym(SB) as E8 rel32 with a patch site.
+func encodeSBCall(s *ast.Instr, link *linkInfo) ([]byte, []sbPatch, error) {
+	o, err := operandFromAST(s.Operands[0], 8, frameInfo{}, link)
+	if err != nil {
+		return nil, nil, err
+	}
+	m, ok := o.(sbMem)
+	if !ok {
+		return nil, nil, fmt.Errorf("CALL: unsupported operand")
+	}
+	e := &enc{}
+	if err := e.emit(&instr{opcode: []byte{0xE8}, modrm: -1, sib: -1, disp: le32(0), sb: &sbRef{name: m.name, addend: m.addend}}); err != nil {
+		return nil, nil, err
+	}
+	ps := make([]sbPatch, len(e.patches))
+	for i, p := range e.patches {
+		ps[i] = sbPatch{off: p.off, name: p.name, addend: p.addend, kind: RelCall}
+	}
+	return e.out, ps, nil
+}
+
 // labelName extracts a local-label name from a jump operand.
 func labelName(op *ast.Operand) (string, bool) {
 	if op.Kind == ast.OpAddr && op.Addr.Sym != nil && op.Addr.Sym.Pseudo == "" &&
@@ -43,6 +43,7 @@ const (
 	stInfoShift = 4

 	rX8664PC32    = 2
+	rX8664TPOFF32 = 20
 )

 // elfSym is one symbol-table entry in construction.
@@ -125,11 +126,19 @@ func (img *Image) ELFObject() ([]byte, error) {
 	type elfRela struct {
 		off    uint64
 		sym    int
+		typ    uint32
 		addend int64
 	}
 	var relas []elfRela
 	for _, fn := range img.Funcs {
 		for _, r := range fn.Relocs {
+			var typ uint32 = rX8664PC32
+			if r.Kind == RelTLSLE {
+				// R_X86_64_TPOFF32 resolves to the local-exec TLS offset and
+				// carries no symbol.
+				relas = append(relas, elfRela{off: uint64(fn.Offset + r.Off), sym: 0, typ: rX8664TPOFF32})
+				continue
+			}
 			idx, ok := symIdx[r.Name]
 			if !ok {
 				return nil, fmt.Errorf("relocation references unknown symbol %q", r.Name)
@@ -137,6 +146,7 @@ func (img *Image) ELFObject() ([]byte, error) {
 			relas = append(relas, elfRela{
 				off: uint64(fn.Offset + r.Off),
 				sym: idx,
+				typ: typ,
 				// R_X86_64_PC32 computes S + A − P with P the patch site; the
 				// assembler measures the symbol from the instruction end,
 				// After − Off bytes past the field, so the addend carries
@@ -94,10 +94,12 @@ const (
 )

 // Relocation types (cmd/internal/objabi).
-// R_PCREL and R_ADDR are stable across Go versions.
+// R_ADDR, R_CALL, R_PCREL and R_TLS_LE are stable across Go versions.
 const (
-	relocPCRel = 14 // R_PCREL
 	relocAddr  = 1  // R_ADDR
+	relocCall  = 7  // R_CALL
+	relocPCRel = 14 // R_PCREL
+	relocTLSLE = 15 // R_TLS_LE
 )

 // relocDWTXTADDRU4 returns the R_DWTXTADDR_U4 relocation type for the
@@ -142,8 +144,18 @@ func isGo127OrLater() bool {
 const (
 	pkgIdxNone    = 0x7fffffff
 	pkgIdxSelf    = 0x7ffffffb
+	pkgIdxBuiltin = 0x7ffffffc
 )

+// goobjBuiltinMorestackNoctxt is the index of runtime.morestack_noctxt in
+// cmd/internal/goobj/builtinlist.go of the toolchain the object targets
+// (246 since Go 1.25; the list is append-only).
+const goobjBuiltinMorestackNoctxt = 246
+
+// goobjBuiltinMorestack is the builtin reference the toolchain emits for the
+// stack-guard call.
+var goobjBuiltinMorestack = "runtime\u00b7morestack_noctxt"
+
 const goobjMagic = "\x00go120ld"

 // goSym is one symbol definition under construction.
@@ -184,8 +196,18 @@ func (img *Image) GOObject(pkgPath, srcPath string) ([]byte, error) {
 	if err != nil {
 		return nil, err
 	}
-	// amd64: MinLC 1, R_PCREL for the code relocations.
-	return img.emitGOObject(pkgPath, srcPath, pre, 1, func(Reloc) (uint16, uint8) { return relocPCRel, 4 })
+	// amd64: MinLC 1, R_PCREL for displacements, R_CALL for calls and
+	// R_TLS_LE for the stack-guard TLS load.
+	return img.emitGOObject(pkgPath, srcPath, pre, 1, func(r Reloc) (uint16, uint8) {
+		switch r.Kind {
+		case RelCall:
+			return relocCall, 4
+		case RelTLSLE:
+			return relocTLSLE, 4
+		default:
+			return relocPCRel, 4
+		}
+	})
 }

 // emitGOObject assembles the GOOBJ payload for any architecture.  pre is
@@ -247,7 +269,7 @@ func (img *Image) emitGOObject(pkgPath, srcPath string, pre []byte, minLC int, r
 			// their relocations cover whole AUIPC/pcalau12i pairs, so
 			// zeroing r.Off would erase the opcode/register bits the linker
 			// preserves when it patches only the immediate.
-			if r.Kind != RelPCRel32 {
+			if r.Kind != RelPCRel32 && r.Kind != RelCall {
 				continue
 			}
 			if r.Off >= 0 && r.Off+4 <= len(code) {
@@ -320,6 +342,13 @@ func (img *Image) emitGOObject(pkgPath, srcPath string, pre []byte, minLC int, r
 		)
 	}

+	// Index the non-package TEXT definitions by short name for the internal
+	// call references.
+	textNpIdx := map[string]int{}
+	for i, fn := range img.Funcs {
+		textNpIdx[fn.Name] = fnNpIdx[i]
+	}
+
 	// Resolve external symbol references (cross-package).  Build the
 	// package index table and determine each external symbol's SymIdx
 	// by reading the target package's export data.
@@ -327,12 +356,22 @@ func (img *Image) emitGOObject(pkgPath, srcPath string, pre []byte, minLC int, r
 	var extPkgIdx map[string]int
 	var extSymIdx map[string]int
 	if len(img.Externals) > 0 {
+		// The morestack call is a builtin reference, not a resolved external.
+		var need []string
+		for _, n := range img.Externals {
+			if n == goobjBuiltinMorestack {
+				continue
+			}
+			need = append(need, n)
+		}
+		if len(need) > 0 {
 			var err error
-		extPkgTable, extPkgIdx, extSymIdx, err = resolveExternalSymbols(img.Externals)
+			extPkgTable, extPkgIdx, extSymIdx, err = resolveExternalSymbols(need)
 			if err != nil {
 				return nil, fmt.Errorf("GOOBJ emission: resolving external symbols: %w", err)
 			}
 		}
+	}

 	// Relocations, per defined symbol in definition order (package defs,
 	// then non-package defs).
@@ -342,6 +381,31 @@ func (img *Image) emitGOObject(pkgPath, srcPath string, pre []byte, minLC int, r
 		si := len(defs) + fnNpIdx[i]
 		for _, r := range fn.Relocs {
 			typ, size := relocField(r)
+			if r.Kind == RelTLSLE {
+				// The TLS load has no symbol: {0, 0} is the nil ref.
+				var rec [23]byte
+				binary.LittleEndian.PutUint32(rec[0:], uint32(int32(r.Off)))
+				rec[4] = size
+				binary.LittleEndian.PutUint16(rec[5:], typ)
+				binary.LittleEndian.PutUint64(rec[7:], uint64(r.Addend))
+				binary.LittleEndian.PutUint32(rec[15:], 0)
+				binary.LittleEndian.PutUint32(rec[19:], 0)
+				symRelocs[si] = append(symRelocs[si], rec[:]...)
+				continue
+			}
+			if r.External && r.Kind == RelCall && r.Name == goobjBuiltinMorestack {
+				// The stack-guard morestack call uses the toolchain's
+				// builtin reference.
+				var rec [23]byte
+				binary.LittleEndian.PutUint32(rec[0:], uint32(int32(r.Off)))
+				rec[4] = size
+				binary.LittleEndian.PutUint16(rec[5:], typ)
+				binary.LittleEndian.PutUint64(rec[7:], uint64(r.Addend))
+				binary.LittleEndian.PutUint32(rec[15:], pkgIdxBuiltin)
+				binary.LittleEndian.PutUint32(rec[19:], goobjBuiltinMorestackNoctxt)
+				symRelocs[si] = append(symRelocs[si], rec[:]...)
+				continue
+			}
 			if r.External {
 				// Split package-qualified name: "runtime·morestack" → runtime, morestack.
 				pkg, name := splitQualified(r.Name)
@@ -366,16 +430,24 @@ func (img *Image) emitGOObject(pkgPath, srcPath string, pre []byte, minLC int, r
 				symRelocs[si] = append(symRelocs[si], rec[:]...)
 				continue
 			}
+			pkg := uint32(pkgIdxSelf)
 			di, ok := defIdx[r.Name]
 			if !ok {
+				// A call to a TEXT function of the same file references the
+				// non-package definition table.
+				ni, isText := textNpIdx[r.Name]
+				if !isText || r.Kind != RelCall {
 					return nil, fmt.Errorf("GOOBJ emission: reference to unknown symbol %q", r.Name)
 				}
+				pkg = pkgIdxNone
+				di = ni
+			}
 			var rec [23]byte
 			binary.LittleEndian.PutUint32(rec[0:], uint32(int32(r.Off)))
 			rec[4] = size // field width
 			binary.LittleEndian.PutUint16(rec[5:], typ)
 			binary.LittleEndian.PutUint64(rec[7:], uint64(r.Addend))
-			binary.LittleEndian.PutUint32(rec[15:], pkgIdxSelf)
+			binary.LittleEndian.PutUint32(rec[15:], pkg)
 			binary.LittleEndian.PutUint32(rec[19:], uint32(di))
 			symRelocs[si] = append(symRelocs[si], rec[:]...)
 		}
@@ -0,0 +1,102 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+package asm
+
+import (
+	"bytes"
+	"encoding/hex"
+	"testing"
+
+	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+)
+
+// The expected bytes are pinned from `go tool asm` output (Go 1.27, amd64,
+// verified with go tool objdump): the stack-split guard classes, the morestack
+// block and the auto-NOSPLIT leaf behaviour.
+func TestStackGuardBytes(t *testing.T) {
+	for _, tt := range []struct {
+		name string
+		src  string
+		want string
+	}{
+		{"leafsmall", "TEXT \u00b7leafsmall(SB), $16-0\n\tRET\n",
+			"554889e54883ec104883c4105dc3"},
+		{"leafmed", "TEXT \u00b7leafmed(SB), $256-0\n\tRET\n",
+			"644c8b3425000000004c8da42478ffffff4d3b66107614554889e54881ec000100004881c4000100005dc3e800000000ebce"},
+		{"leafbig", "TEXT \u00b7leafbig(SB), $8192-0\n\tRET\n",
+			"644c8b3425000000004989e44981ec881f0000721a4d3b66107614554889e54881ec002000004881c4002000005dc3e800000000ebca"},
+		{"callsmall", "TEXT \u00b7callsmall(SB), $16-0\n\tCALL \u00b7other(SB)\n\tRET\nTEXT \u00b7other(SB), NOSPLIT, $0\n\tRET\n",
+			"644c8b342500000000493b66107613554889e54883ec10e8000000004883c4105dc3e800000000ebd7"},
+		{"nosplit", "TEXT \u00b7nosplit(SB), NOSPLIT, $16-0\n\tRET\n",
+			"554889e54883ec104883c4105dc3"},
+	} {
+		f, errs := parser.Parse("g_amd64.s", tt.src)
+		if len(errs) > 0 {
+			t.Fatalf("%s: parse: %v", tt.name, errs)
+		}
+		img, err := AssembleFile(f)
+		if err != nil {
+			t.Fatalf("%s: assemble: %v", tt.name, err)
+		}
+		fn := img.Funcs[0]
+		// The toolchain's object leaves every relocation field zero for the
+		// linker, while the gasm image resolves file-internal references, so
+		// the comparison masks the patch sites the way verify's ground truth
+		// does.
+		code := append([]byte(nil), img.Code[fn.Offset:fn.Offset+fn.Size]...)
+		for _, r := range fn.Relocs {
+			for j := r.Off; j < r.Off+4 && j < len(code); j++ {
+				code[j] = 0
+			}
+		}
+		got := hex.EncodeToString(code)
+		if got != tt.want {
+			t.Errorf("%s:\n  got  %s\n  want %s", tt.name, got, tt.want)
+		}
+	}
+}
+
+// TestStackGuardRelocs checks the guard's patch sites: the TLS slot and the
+// morestack call.
+func TestStackGuardRelocs(t *testing.T) {
+	f, errs := parser.Parse("g_amd64.s", "TEXT \u00b7f(SB), $256-0\n\tRET\n")
+	if len(errs) > 0 {
+		t.Fatalf("parse: %v", errs)
+	}
+	img, err := AssembleFile(f)
+	if err != nil {
+		t.Fatalf("assemble: %v", err)
+	}
+	relocs := img.Funcs[0].Relocs
+	if len(relocs) != 2 {
+		t.Fatalf("relocs = %d, want 2", len(relocs))
+	}
+	tls, call := relocs[0], relocs[1]
+	if tls.Kind != RelTLSLE || tls.Off != 5 || tls.Name != "" || tls.External {
+		t.Errorf("tls reloc = %+v, want RelTLSLE at 5 with no symbol", tls)
+	}
+	if call.Kind != RelCall || call.Name != "runtime\u00b7morestack_noctxt" || !call.External {
+		t.Errorf("call reloc = %+v, want RelCall to runtime.morestack_noctxt", call)
+	}
+}
+
+// TestStackGuardGOObj emissions succeed with the guard's TLS and builtin
+// references in play.
+func TestStackGuardGOObj(t *testing.T) {
+	f, errs := parser.Parse("g_amd64.s", "TEXT \u00b7f(SB), $256-0\n\tCALL \u00b7helper(SB)\n\tRET\nTEXT \u00b7helper(SB), NOSPLIT, $0\n\tRET\n")
+	if len(errs) > 0 {
+		t.Fatalf("parse: %v", errs)
+	}
+	img, err := AssembleFile(f)
+	if err != nil {
+		t.Fatalf("assemble: %v", err)
+	}
+	obj, err := img.GOObject("testpkg", "g_amd64.s")
+	if err != nil {
+		t.Fatalf("GOObject: %v", err)
+	}
+	if !bytes.Contains(obj, []byte("go120ld")) {
+		t.Fatal("object lacks the GOOBJ magic")
+	}
+}
@@ -91,6 +91,8 @@ type RelocKind int

 const (
 	RelPCRel32         RelocKind = iota // 32-bit PC-relative (amd64)
+	RelCall                             // R_CALL: CALL to a function symbol (amd64)
+	RelTLSLE                            // R_TLS_LE: local-exec TLS load, no symbol (amd64 guard)
 	RelRISCVPCRELIType                  // R_RISCV_PCREL_ITYPE (AUIPC + I-type pair)
 	RelRISCVPCRELSType                  // R_RISCV_PCREL_STYPE (AUIPC + S-type pair)
 	RelRISCVJal                         // R_RISCV_JAL (J-type call)
@@ -148,6 +150,7 @@ func AssembleFile(f *ast.File) (*Image, error) {
 	link := &linkInfo{symbols: known, allowExternal: true}

 	img := &Image{Symbols: map[string]int{}}
+	textOff := map[string]int{}
 	type asmFunc struct {
 		name    string
 		patches []sbPatch
@@ -185,6 +188,7 @@ func AssembleFile(f *ast.File) (*Image, error) {
 		for _, s := range steps {
 			fl.Spadj = append(fl.Spadj, SpadjStep{PC: s.pc, Value: s.value})
 		}
+		textOff[t.Name.Name] = len(img.Code)
 		img.Funcs = append(img.Funcs, fl)
 		img.Code = append(img.Code, code...)
 		funcs = append(funcs, asmFunc{name: t.Name.Name, patches: patches})
@@ -217,13 +221,27 @@ func AssembleFile(f *ast.File) (*Image, error) {
 		base := img.Funcs[i].Offset
 		code := img.Code[base : base+img.Funcs[i].Size]
 		for _, p := range fn.patches {
-			reloc := Reloc{Off: p.off, After: p.after, Name: p.name, Addend: p.addend}
+			reloc := Reloc{Off: p.off, After: p.after, Name: p.name, Addend: p.addend, Kind: p.kind}
+			if p.kind == RelTLSLE {
+				// The TLS slot has no symbol: the linker fills the offset
+				// from the runtime's TLS layout.
+				img.Funcs[i].Relocs = append(img.Funcs[i].Relocs, reloc)
+				continue
+			}
 			if imgOff, ok := img.Symbols[p.name]; ok {
 				rel := int64(imgOff) + p.addend - int64(base+p.after)
 				if rel < -1<<31 || rel >= 1<<31 {
 					return nil, fmt.Errorf("%s: displacement to %q out of rel32 range", fn.name, p.name)
 				}
 				copy(code[p.off:p.off+4], le32(rel))
+			} else if imgOff, ok := textOff[p.name]; ok {
+				// A CALL to a TEXT function of the same file: resolve the
+				// displacement against the function's layout position.
+				rel := int64(imgOff) + p.addend - int64(base+p.after)
+				if rel < -1<<31 || rel >= 1<<31 {
+					return nil, fmt.Errorf("%s: displacement to %q out of rel32 range", fn.name, p.name)
+				}
+				copy(code[p.off:p.off+4], le32(rel))
 			} else {
 				reloc.External = true
 				externals[p.name] = true