arm64 replace call code use X17 register instead of X9

asm_bytes.go

@@ -8,10 +8,15 @@ const (
 
 // arm/arm64
 var (
-	armcode   = []byte{0x04, 0xF0, 0x1F, 0xE5} //LDR PC, [PC, #-4]
-	arm64code = []byte{
-		0x49, 0x00, 0x00, 0x58, // LDR X9 [PC+8]
-		0x20, 0x01, 0x1F, 0xD6} // BR X9
+	armReplaceCallCode = []byte{0x04, 0xF0, 0x1F, 0xE5} //LDR PC, [PC, #-4]
+	// X16 and X17 are the IP0 and IP1 intra-procedure-call corruptible registers -
+	// since Go only uses them for the stack prologue and epilogue calculations,
+	// and we should already be clear of that by the time we hit a R_CALLARM64,
+	// so we should be able to safely use them for far jumps
+	armReplace64CALLCode = []byte{
+		0x51, 0x00, 0x00, 0x58, // LDR X17 [PC+8] - read 64 bit address from PC+8 into X17
+		0x20, 0x02, 0x1f, 0xd6, // BR  X17 - jump to address in X17
+	}
 	arm64Bcode   = []byte{0x00, 0x00, 0x00, 0x14} // B [PC+0x0]
 	arm64LDRcode = []byte{0x00, 0x00, 0x40, 0xF9} // LDR XX [XX]
 )

relocate.go

@@ -164,11 +164,11 @@ func (linker *Linker) relocteCALLARM(addr uintptr, loc obj.Reloc, segment *segme
 		}
 		putUint24(segment.codeByte[loc.Offset:], off)
 		if loc.Type == reloctype.R_CALLARM64 {
-			copy(segment.codeByte[segment.codeOff:], arm64code)
-			segment.codeOff += len(arm64code)
+			copy(segment.codeByte[segment.codeOff:], armReplace64CALLCode)
+			segment.codeOff += len(armReplace64CALLCode)
 		} else {
-			copy(segment.codeByte[segment.codeOff:], armcode)
-			segment.codeOff += len(armcode)
+			copy(segment.codeByte[segment.codeOff:], armReplaceCallCode)
+			segment.codeOff += len(armReplaceCallCode)
 		}
 		putAddressAddOffset(byteorder, segment.codeByte, &segment.codeOff, uint64(int(addr)+add))
 	} else {