fix(asm): encode amd64 frame adjustments above 127 bytes with imm32

asm/assemble.go

@@ -22,6 +22,11 @@ import (
 // FP/SP frame-relative operands, and local-label jumps.  SB (global symbol)
 // operands require relocations and are not yet supported; the SIMD (VEX/AVX2)
 // integer and shuffle/extract/permute/move set is in.
+//
+// Like the other architectures, the stack-growth guard (the morestack check
+// in the prologue and the call back into the runtime in the epilogue) is not
+// emitted: the bytes match go tool asm only for NOSPLIT functions or
+// zero-frame leaves, where the toolchain emits no guard either.
 func Assemble(t *ast.Text) ([]byte, map[string]int, error) {
 	code, _, labels, _, _, err := assemble(t, nil)
 	return code, labels, err
@@ -31,7 +36,7 @@ func Assemble(t *ast.Text) ([]byte, map[string]int, error) {
 // the set of static symbols a GLOBL in the same file defines.  A nil link
 // rejects SB operands outright (single-function assembly cannot resolve
 // them).  When allowExternal is set, a reference to a symbol no GLOBL in the
-// file defines is recorded as an external relocation instead of failing —
+// file defines is recorded as an external relocation instead of failing
 // the object-file emitters resolve it at link time.
 type linkInfo struct {
 	symbols       map[string]bool
@@ -258,14 +263,11 @@ func epilogueBytes(size int) []byte {
 }
 
 func subSP(size int) []byte { // SUBQ $size, SP
+	// imm8 holds -128..127; anything larger takes the imm32 form, exactly as
+	// the Go assembler encodes it (verified for 8, 128, 200 and 255).
 	if size >= -128 && size <= 127 {
 		return []byte{0x48, 0x83, 0xEC, byte(int8(size))}
 	}
-	// 128..255 do not fit SUB's unsigned imm8, but the Go assembler
-	// switches to ADDQ $-size, SP whose sign-extended imm8 does.
-	if size >= -255 && size <= 255 {
-		return []byte{0x48, 0x83, 0xC4, byte(int8(-size))}
-	}
 	return append([]byte{0x48, 0x81, 0xEC}, le32(int64(size))...)
 }
 

asm/assemble_test.go

@@ -4,6 +4,7 @@
 package asm
 
 import (
+	"bytes"
 	"strings"
 	"testing"
 
@@ -351,3 +352,26 @@ TEXT ·pf(SB), NOSPLIT, $0
 		t.Errorf("PREFETCHT0 bytes: got %s, want 0f 18 0b", hex)
 	}
 }
+
+// TestSubSPEncodings pins the prologue SUB against the bytes go tool asm
+// emits for SUBQ $size, SP: imm8 for -128..127, the imm32 form for anything
+// larger.  The intermediate 129..255 range used to encode an ADD with a
+// truncated immediate, moving SP the wrong way.
+func TestSubSPEncodings(t *testing.T) {
+	for _, tt := range []struct {
+		size int
+		want []byte
+	}{
+		{8, []byte{0x48, 0x83, 0xEC, 0x08}},
+		{127, []byte{0x48, 0x83, 0xEC, 0x7F}},
+		{128, []byte{0x48, 0x81, 0xEC, 0x80, 0x00, 0x00, 0x00}},
+		{200, []byte{0x48, 0x81, 0xEC, 0xC8, 0x00, 0x00, 0x00}},
+		{255, []byte{0x48, 0x81, 0xEC, 0xFF, 0x00, 0x00, 0x00}},
+		{4096, []byte{0x48, 0x81, 0xEC, 0x00, 0x10, 0x00, 0x00}},
+	} {
+		got := subSP(tt.size)
+		if !bytes.Equal(got, tt.want) {
+			t.Errorf("subSP(%d) = %x, want %x", tt.size, got, tt.want)
+		}
+	}
+}