feat(asm): emit the arm64 stack-split guard and morestack block

asm/arm64_assemble.go

@@ -26,6 +26,7 @@ import (
 func assembleARM64(t *ast.Text) ([]byte, map[string]int, []Reloc, []LineEntry, []SpadjStep, error) {
 	fi := arm64ComputeFrame(t)
 	prologue := arm64Prologue(fi)
+	guardLen := arm64GuardLen(fi)
 	chain := arm64JumpChain(t)
 	resolve := func(name string) string {
 		if r, ok := chain[name]; ok {
@@ -38,14 +39,14 @@ func assembleARM64(t *ast.Text) ([]byte, map[string]int, []Reloc, []LineEntry, [
 	var spadj []SpadjStep
 
 	// The prologue (3 instructions when a small frame, 4 for large)
-	// raises the SP delta by autosize.
+	// raises the SP delta by autosize.  The guard prefix shifts its PC.
 	if fi.autosize != 0 {
-		spadj = append(spadj, SpadjStep{PC: arm64PrologueSpadjPC(fi), Value: fi.autosize})
+		spadj = append(spadj, SpadjStep{PC: guardLen + arm64PrologueSpadjPC(fi), Value: fi.autosize})
 	}
 
 	// Pass 1: label offsets from the instruction sizes.
 	offsets := map[string]int{}
-	pos := len(prologue)
+	pos := guardLen + len(prologue)
 	for _, stmt := range t.Body {
 		switch s := stmt.(type) {
 		case *ast.Label:
@@ -55,9 +56,25 @@ func assembleARM64(t *ast.Text) ([]byte, map[string]int, []Reloc, []LineEntry, [
 		}
 	}
 
-	// Pass 2: encode.  Relocation offsets are recorded function-relative.
-	out := append([]byte(nil), prologue...)
-	pc := len(prologue)
+	// Pass 2: encode.  The guard prefix precedes the prologue; its branches
+	// target the morestack block at the end of the function, whose position
+	// the first pass has settled.
+	bodyLen := 0
+	{
+		p := guardLen + len(prologue)
+		for _, stmt := range t.Body {
+			if in, ok := stmt.(*ast.Instr); ok {
+				p += arm64InstrSize(in, fi)
+			}
+		}
+		bodyLen = p - (guardLen + len(prologue))
+	}
+	var out []byte
+	if fi.needSplit {
+		out = append(out, arm64GuardBytes(fi, guardLen+len(prologue)+bodyLen)...)
+	}
+	out = append(out, prologue...)
+	pc := guardLen + len(prologue)
 	preCount := len(relocs)
 	var lines []LineEntry
 	for _, stmt := range t.Body {
@@ -87,6 +104,12 @@ func assembleARM64(t *ast.Text) ([]byte, map[string]int, []Reloc, []LineEntry, [
 		out = append(out, code...)
 		pc += len(code)
 	}
+	if fi.needSplit {
+		block, blReloc := arm64MoreStackBlock(pc)
+		out = append(out, block...)
+		relocs = append(relocs, blReloc)
+		pc += len(block)
+	}
 	return out, offsets, relocs, lines, spadj, nil
 }