diff --git a/Cargo.toml b/Cargo.toml
index 6f61e5f95c..3f373fdace 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -22,8 +22,7 @@ path = "jit.rs"
 
 [features]
 disasm = ["yjit?/disasm", "zjit?/disasm"]
-# TODO(GH-642) Turning this on trips a btest failure.
-runtime_checks = [] # ["yjit?/runtime_checks", "zjit?/runtime_checks"]
+runtime_checks = ["yjit?/runtime_checks", "zjit?/runtime_checks"]
 yjit = [ "dep:yjit" ]
 zjit = [ "dep:zjit" ]
 
diff --git a/shape.h b/shape.h
index 63d5534d46..a418dc7821 100644
--- a/shape.h
+++ b/shape.h
@@ -48,7 +48,9 @@ enum shape_id_fl_type {
 
 // This masks allows to check if a shape_id contains any ivar.
 // It rely on ROOT_SHAPE_WITH_OBJ_ID==1.
-#define SHAPE_ID_HAS_IVAR_MASK (SHAPE_ID_FL_TOO_COMPLEX | (SHAPE_ID_OFFSET_MASK - 1))
+enum {
+    SHAPE_ID_HAS_IVAR_MASK = SHAPE_ID_FL_TOO_COMPLEX | (SHAPE_ID_OFFSET_MASK - 1),
+};
 
 // The interpreter doesn't care about frozen status or slot size when reading ivars.
 // So we normalize shape_id by clearing these bits to improve cache hits.
diff --git a/yjit/bindgen/src/main.rs b/yjit/bindgen/src/main.rs
index a0446ad17b..7c8ce7bce5 100644
--- a/yjit/bindgen/src/main.rs
+++ b/yjit/bindgen/src/main.rs
@@ -103,6 +103,7 @@ fn main() {
         .allowlist_function("rb_yjit_shape_capacity")
         .allowlist_function("rb_yjit_shape_index")
         .allowlist_var("SHAPE_ID_NUM_BITS")
+        .allowlist_var("SHAPE_ID_HAS_IVAR_MASK")
 
         // From ruby/internal/intern/object.h
         .allowlist_function("rb_obj_is_kind_of")
@@ -228,7 +229,6 @@ fn main() {
         .allowlist_function("rb_obj_as_string_result")
         .allowlist_function("rb_str_byte_substr")
         .allowlist_function("rb_str_substr_two_fixnums")
-        .allowlist_function("rb_str_dup_m")
 
         // From include/ruby/internal/intern/parse.h
         .allowlist_function("rb_backref_get")
diff --git a/yjit/src/codegen.rs b/yjit/src/codegen.rs
index 66932fd46c..b8b15adc8b 100644
--- a/yjit/src/codegen.rs
+++ b/yjit/src/codegen.rs
@@ -6278,9 +6278,14 @@ fn jit_rb_str_dup(
     let recv_opnd = asm.stack_pop(1);
     let recv_opnd = asm.load(recv_opnd);
 
+    let shape_id_offset = unsafe { rb_shape_id_offset() };
+    let shape_opnd = Opnd::mem(64, recv_opnd, shape_id_offset);
+    asm.test(shape_opnd, Opnd::UImm(SHAPE_ID_HAS_IVAR_MASK as u64));
+    asm.jnz(Target::side_exit(Counter::send_str_dup_exivar));
+
     // Call rb_str_dup
     let stack_ret = asm.stack_push(Type::CString);
-    let ret_opnd = asm.ccall(rb_str_dup_m as *const u8, vec![recv_opnd]);
+    let ret_opnd = asm.ccall(rb_str_dup as *const u8, vec![recv_opnd]);
     asm.mov(stack_ret, ret_opnd);
 
     true
diff --git a/yjit/src/cruby_bindings.inc.rs b/yjit/src/cruby_bindings.inc.rs
index e36b6f9f5f..9b871c9e7c 100644
--- a/yjit/src/cruby_bindings.inc.rs
+++ b/yjit/src/cruby_bindings.inc.rs
@@ -689,6 +689,8 @@ pub const VM_ENV_FLAG_ISOLATED: vm_frame_env_flags = 16;
 pub type vm_frame_env_flags = u32;
 pub type attr_index_t = u16;
 pub type shape_id_t = u32;
+pub const SHAPE_ID_HAS_IVAR_MASK: _bindgen_ty_37 = 134742014;
+pub type _bindgen_ty_37 = u32;
 #[repr(C)]
 pub struct rb_cvar_class_tbl_entry {
     pub index: u32,
@@ -1120,7 +1122,6 @@ extern "C" {
     pub fn rb_gvar_set(arg1: ID, arg2: VALUE) -> VALUE;
     pub fn rb_ensure_iv_list_size(obj: VALUE, current_len: u32, newsize: u32);
     pub fn rb_vm_barrier();
-    pub fn rb_str_dup_m(str_: VALUE) -> VALUE;
     pub fn rb_str_byte_substr(str_: VALUE, beg: VALUE, len: VALUE) -> VALUE;
     pub fn rb_str_substr_two_fixnums(
         str_: VALUE,