diff --git a/crates/napi/src/bindgen_runtime/js_values/arraybuffer.rs b/crates/napi/src/bindgen_runtime/js_values/arraybuffer.rs
index c7363b4c..663360a3 100644
--- a/crates/napi/src/bindgen_runtime/js_values/arraybuffer.rs
+++ b/crates/napi/src/bindgen_runtime/js_values/arraybuffer.rs
@@ -7,8 +7,6 @@ use std::sync::{
   Arc,
 };
 
-#[cfg(feature = "napi4")]
-use crate::bindgen_prelude::{CUSTOM_GC_TSFN, CUSTOM_GC_TSFN_CLOSED, MAIN_THREAD_ID};
 pub use crate::js_values::TypedArrayType;
 use crate::{check_status, sys, Error, Result, Status};
 
@@ -66,31 +64,6 @@ macro_rules! impl_typed_array {
       fn drop(&mut self) {
         if Arc::strong_count(&self.drop_in_vm) == 1 {
           if let Some((ref_, env)) = self.raw {
-            #[cfg(feature = "napi4")]
-            {
-              if CUSTOM_GC_TSFN_CLOSED.load(std::sync::atomic::Ordering::SeqCst) {
-                return;
-              }
-              if !MAIN_THREAD_ID
-                .get()
-                .map(|id| &std::thread::current().id() == id)
-                .unwrap_or(false)
-              {
-                let status = unsafe {
-                  sys::napi_call_threadsafe_function(
-                    CUSTOM_GC_TSFN.load(std::sync::atomic::Ordering::SeqCst),
-                    ref_ as *mut c_void,
-                    1,
-                  )
-                };
-                assert!(
-                  status == sys::Status::napi_ok,
-                  "Call custom GC in ArrayBuffer::drop failed {:?}",
-                  Status::from(status)
-                );
-                return;
-              }
-            }
             crate::check_status_or_throw!(
               env,
               unsafe { sys::napi_reference_unref(env, ref_, &mut 0) },
diff --git a/crates/napi/src/bindgen_runtime/js_values/bigint.rs b/crates/napi/src/bindgen_runtime/js_values/bigint.rs
index 27100318..67256488 100644
--- a/crates/napi/src/bindgen_runtime/js_values/bigint.rs
+++ b/crates/napi/src/bindgen_runtime/js_values/bigint.rs
@@ -52,7 +52,7 @@ impl FromNapiValue for BigInt {
         ptr::null_mut(),
       )
     })?;
-    let mut words: Vec<u64> = Vec::with_capacity(word_count as usize);
+    let mut words: Vec<u64> = Vec::with_capacity(word_count);
     let mut sign_bit = 0;
 
     unsafe {
@@ -64,7 +64,7 @@ impl FromNapiValue for BigInt {
         words.as_mut_ptr(),
       ))?;
 
-      words.set_len(word_count as usize);
+      words.set_len(word_count);
     }
     if word_count == 0 {
       words = vec![0];
@@ -155,7 +155,7 @@ impl ToNapiValue for BigInt {
 impl ToNapiValue for i128 {
   unsafe fn to_napi_value(env: sys::napi_env, val: Self) -> crate::Result<sys::napi_value> {
     let mut raw_value = ptr::null_mut();
-    let sign_bit = if val > 0 { 0 } else { 1 };
+    let sign_bit = i32::from(val <= 0);
     let words = &val as *const i128 as *const u64;
     check_status!(unsafe {
       sys::napi_create_bigint_words(env, sign_bit, 2, words, &mut raw_value)
diff --git a/crates/napi/src/bindgen_runtime/js_values/buffer.rs b/crates/napi/src/bindgen_runtime/js_values/buffer.rs
index c39218d6..c3f98677 100644
--- a/crates/napi/src/bindgen_runtime/js_values/buffer.rs
+++ b/crates/napi/src/bindgen_runtime/js_values/buffer.rs
@@ -1,4 +1,4 @@
-#[cfg(debug_assertions)]
+#[cfg(all(debug_assertions, not(windows)))]
 use std::collections::HashSet;
 use std::ffi::c_void;
 use std::mem;
@@ -6,12 +6,12 @@ use std::ops::{Deref, DerefMut};
 use std::ptr::{self, NonNull};
 use std::slice;
 use std::sync::Arc;
-#[cfg(debug_assertions)]
+#[cfg(all(debug_assertions, not(windows)))]
 use std::sync::Mutex;
 
 use crate::{bindgen_prelude::*, check_status, sys, Result, ValueType};
 
-#[cfg(debug_assertions)]
+#[cfg(all(debug_assertions, not(windows)))]
 thread_local! {
   pub (crate) static BUFFER_DATA: Mutex<HashSet<*mut u8>> = Default::default();
 }
@@ -32,31 +32,6 @@ impl Drop for Buffer {
   fn drop(&mut self) {
     if Arc::strong_count(&self.ref_count) == 1 {
       if let Some((ref_, env)) = self.raw {
-        #[cfg(feature = "napi4")]
-        {
-          if CUSTOM_GC_TSFN_CLOSED.load(std::sync::atomic::Ordering::SeqCst) {
-            return;
-          }
-          if !MAIN_THREAD_ID
-            .get()
-            .map(|id| &std::thread::current().id() == id)
-            .unwrap_or(false)
-          {
-            let status = unsafe {
-              sys::napi_call_threadsafe_function(
-                CUSTOM_GC_TSFN.load(std::sync::atomic::Ordering::SeqCst),
-                ref_ as *mut c_void,
-                1,
-              )
-            };
-            assert!(
-              status == sys::Status::napi_ok,
-              "Call custom GC in Buffer::drop failed {:?}",
-              Status::from(status)
-            );
-            return;
-          }
-        }
         let mut ref_count = 0;
         check_status_or_throw!(
           env,
@@ -94,7 +69,7 @@ impl Clone for Buffer {
 impl From<Vec<u8>> for Buffer {
   fn from(mut data: Vec<u8>) -> Self {
     let inner_ptr = data.as_mut_ptr();
-    #[cfg(debug_assertions)]
+    #[cfg(all(debug_assertions, not(windows)))]
     {
       let is_existed = BUFFER_DATA.with(|buffer_data| {
         let buffer = buffer_data.lock().expect("Unlock buffer data failed");
diff --git a/crates/napi/src/bindgen_runtime/js_values/promise.rs b/crates/napi/src/bindgen_runtime/js_values/promise.rs
index 9beea5d0..f6168d1f 100644
--- a/crates/napi/src/bindgen_runtime/js_values/promise.rs
+++ b/crates/napi/src/bindgen_runtime/js_values/promise.rs
@@ -77,7 +77,7 @@ impl<T: FromNapiValue> ValidateNapiValue for Promise<T> {
   }
 }
 
-unsafe impl<T: FromNapiValue> Send for Promise<T> {}
+unsafe impl<T: FromNapiValue + Send> Send for Promise<T> {}
 
 impl<T: FromNapiValue> FromNapiValue for Promise<T> {
   unsafe fn from_napi_value(env: sys::napi_env, napi_val: sys::napi_value) -> crate::Result<Self> {
diff --git a/crates/napi/src/bindgen_runtime/js_values/value_ref.rs b/crates/napi/src/bindgen_runtime/js_values/value_ref.rs
index 48a88d82..0c911680 100644
--- a/crates/napi/src/bindgen_runtime/js_values/value_ref.rs
+++ b/crates/napi/src/bindgen_runtime/js_values/value_ref.rs
@@ -6,7 +6,7 @@ use std::rc::{Rc, Weak};
 use once_cell::sync::Lazy;
 
 use crate::{
-  bindgen_runtime::{PersistedSingleThreadHashMap, ToNapiValue},
+  bindgen_runtime::{PersistedPerInstanceHashMap, ToNapiValue},
   check_status, Env, Error, Result, Status,
 };
 
@@ -16,7 +16,7 @@ type RefInformation = (
   *const Cell<*mut dyn FnOnce()>,
 );
 
-pub(crate) static REFERENCE_MAP: Lazy<PersistedSingleThreadHashMap<*mut c_void, RefInformation>> =
+pub(crate) static REFERENCE_MAP: Lazy<PersistedPerInstanceHashMap<*mut c_void, RefInformation>> =
   Lazy::new(Default::default);
 
 /// ### Experimental feature
@@ -59,6 +59,7 @@ impl<T> Drop for Reference<T> {
 
 impl<T: 'static> Reference<T> {
   #[doc(hidden)]
+  #[allow(clippy::not_unsafe_ptr_arg_deref)]
   pub fn add_ref(env: crate::sys::napi_env, t: *mut c_void, value: RefInformation) {
     REFERENCE_MAP.borrow_mut(|map| {
       if let Some((_, previous_ref, previous_rc)) = map.insert(t, value) {
diff --git a/crates/napi/src/bindgen_runtime/mod.rs b/crates/napi/src/bindgen_runtime/mod.rs
index 436e8389..2019727b 100644
--- a/crates/napi/src/bindgen_runtime/mod.rs
+++ b/crates/napi/src/bindgen_runtime/mod.rs
@@ -76,7 +76,7 @@ pub unsafe extern "C" fn drop_buffer(
   #[allow(unused)] finalize_data: *mut c_void,
   finalize_hint: *mut c_void,
 ) {
-  #[cfg(debug_assertions)]
+  #[cfg(all(debug_assertions, not(windows)))]
   {
     js_values::BUFFER_DATA.with(|buffer_data| {
       let mut buffer = buffer_data.lock().expect("Unlock Buffer data failed");
diff --git a/crates/napi/src/bindgen_runtime/module_register.rs b/crates/napi/src/bindgen_runtime/module_register.rs
index f0d0024e..429ebd67 100644
--- a/crates/napi/src/bindgen_runtime/module_register.rs
+++ b/crates/napi/src/bindgen_runtime/module_register.rs
@@ -1,21 +1,11 @@
 use std::collections::{HashMap, HashSet};
-#[cfg(feature = "napi4")]
-use std::ffi::c_void;
 use std::ffi::CStr;
-use std::mem;
-#[cfg(feature = "napi4")]
-use std::os::raw::c_char;
 use std::ptr;
-use std::sync::{
-  atomic::{AtomicBool, AtomicPtr, Ordering},
-  Mutex,
-};
-#[cfg(feature = "napi4")]
+use std::sync::atomic::AtomicUsize;
+use std::sync::atomic::{AtomicBool, AtomicPtr, Ordering};
 use std::thread::ThreadId;
 
 use once_cell::sync::Lazy;
-#[cfg(feature = "napi4")]
-use once_cell::sync::OnceCell;
 
 use crate::{
   check_status, check_status_or_throw, sys, Env, JsError, JsFunction, Property, Result, Value,
@@ -26,120 +16,131 @@ pub type ExportRegisterCallback = unsafe fn(sys::napi_env) -> Result<sys::napi_v
 pub type ModuleExportsCallback =
   unsafe fn(env: sys::napi_env, exports: sys::napi_value) -> Result<()>;
 
-#[cfg(feature = "napi4")]
-pub(crate) static CUSTOM_GC_TSFN: AtomicPtr<sys::napi_threadsafe_function__> =
-  AtomicPtr::new(ptr::null_mut());
-#[cfg(feature = "napi4")]
-// CustomGC ThreadsafeFunction may be deleted during the process exit.
-// And there may still some Buffer alive after that.
-pub(crate) static CUSTOM_GC_TSFN_CLOSED: AtomicBool = AtomicBool::new(false);
-#[cfg(feature = "napi4")]
-pub(crate) static MAIN_THREAD_ID: OnceCell<ThreadId> = OnceCell::new();
-
-struct PersistedSingleThreadVec<T> {
-  inner: Mutex<Vec<T>>,
+struct PersistedPerInstanceVec<T> {
+  inner: AtomicPtr<T>,
+  length: AtomicUsize,
 }
 
-impl<T> Default for PersistedSingleThreadVec<T> {
+impl<T> Default for PersistedPerInstanceVec<T> {
   fn default() -> Self {
-    PersistedSingleThreadVec {
-      inner: Mutex::new(Vec::new()),
-    }
+    let mut vec: Vec<T> = Vec::with_capacity(1);
+    let ret = Self {
+      inner: AtomicPtr::new(vec.as_mut_ptr()),
+      length: AtomicUsize::new(0),
+    };
+    std::mem::forget(vec);
+    ret
   }
 }
 
-impl<T> PersistedSingleThreadVec<T> {
+impl<T> PersistedPerInstanceVec<T> {
   #[allow(clippy::mut_from_ref)]
   fn borrow_mut<F>(&self, f: F)
   where
     F: FnOnce(&mut [T]),
   {
-    let mut locked = self
-      .inner
-      .lock()
-      .expect("Acquire persisted thread vec lock failed");
-    f(&mut locked);
+    let length = self.length.load(Ordering::Relaxed);
+    if length == 0 {
+      f(&mut []);
+    } else {
+      let inner = self.inner.load(Ordering::Relaxed);
+      let mut temp = unsafe { Vec::from_raw_parts(inner, length, length) };
+      f(temp.as_mut_slice());
+      // Inner Vec has been reallocated, so we need to update the pointer
+      if temp.as_mut_ptr() != inner {
+        self.inner.store(temp.as_mut_ptr(), Ordering::Relaxed);
+      }
+      self.length.store(temp.len(), Ordering::Relaxed);
+      std::mem::forget(temp);
+    }
   }
 
   fn push(&self, item: T) {
-    let mut locked = self
-      .inner
-      .lock()
-      .expect("Acquire persisted thread vec lock failed");
-    locked.push(item);
+    let length = self.length.load(Ordering::Relaxed);
+    let inner = self.inner.load(Ordering::Relaxed);
+    let mut temp = unsafe { Vec::from_raw_parts(inner, length, length) };
+    temp.push(item);
+    // Inner Vec has been reallocated, so we need to update the pointer
+    if temp.as_mut_ptr() != inner {
+      self.inner.store(temp.as_mut_ptr(), Ordering::Relaxed);
+    }
+    std::mem::forget(temp);
+
+    self.length.fetch_add(1, Ordering::Relaxed);
   }
 }
 
-unsafe impl<T> Send for PersistedSingleThreadVec<T> {}
-unsafe impl<T> Sync for PersistedSingleThreadVec<T> {}
+unsafe impl<T: Send> Send for PersistedPerInstanceVec<T> {}
+unsafe impl<T: Sync> Sync for PersistedPerInstanceVec<T> {}
 
-pub(crate) struct PersistedSingleThreadHashMap<K, V>(Mutex<HashMap<K, V>>);
+pub(crate) struct PersistedPerInstanceHashMap<K, V>(*mut HashMap<K, V>);
+
+impl<K, V> PersistedPerInstanceHashMap<K, V> {
+  pub(crate) fn from_hashmap(hashmap: HashMap<K, V>) -> Self {
+    Self(Box::into_raw(Box::new(hashmap)))
+  }
 
-impl<K, V> PersistedSingleThreadHashMap<K, V> {
   #[allow(clippy::mut_from_ref)]
   pub(crate) fn borrow_mut<F, R>(&self, f: F) -> R
   where
     F: FnOnce(&mut HashMap<K, V>) -> R,
   {
-    let mut lock = self
-      .0
-      .lock()
-      .expect("Acquire persisted thread hash map lock failed");
-    f(&mut *lock)
+    f(unsafe { Box::leak(Box::from_raw(self.0)) })
   }
 }
 
-impl<K, V> Default for PersistedSingleThreadHashMap<K, V> {
+impl<K, V> Default for PersistedPerInstanceHashMap<K, V> {
   fn default() -> Self {
-    PersistedSingleThreadHashMap(Mutex::new(Default::default()))
+    let map = Default::default();
+    Self(Box::into_raw(Box::new(map)))
   }
 }
 
 type ModuleRegisterCallback =
-  PersistedSingleThreadVec<(Option<&'static str>, (&'static str, ExportRegisterCallback))>;
+  PersistedPerInstanceVec<(Option<&'static str>, (&'static str, ExportRegisterCallback))>;
 
-type ModuleClassProperty = PersistedSingleThreadHashMap<
+type ModuleClassProperty = PersistedPerInstanceHashMap<
   &'static str,
   HashMap<Option<&'static str>, (&'static str, Vec<Property>)>,
 >;
 
-unsafe impl<K, V> Send for PersistedSingleThreadHashMap<K, V> {}
-unsafe impl<K, V> Sync for PersistedSingleThreadHashMap<K, V> {}
+unsafe impl<K, V> Send for PersistedPerInstanceHashMap<K, V> {}
+unsafe impl<K, V> Sync for PersistedPerInstanceHashMap<K, V> {}
 
 type FnRegisterMap =
-  PersistedSingleThreadHashMap<ExportRegisterCallback, (sys::napi_callback, &'static str)>;
+  PersistedPerInstanceHashMap<ExportRegisterCallback, (sys::napi_callback, &'static str)>;
+type RegisteredClassesMap = PersistedPerInstanceHashMap<ThreadId, RegisteredClasses>;
 
 static MODULE_REGISTER_CALLBACK: Lazy<ModuleRegisterCallback> = Lazy::new(Default::default);
 static MODULE_CLASS_PROPERTIES: Lazy<ModuleClassProperty> = Lazy::new(Default::default);
-static MODULE_REGISTER_LOCK: Lazy<Mutex<()>> = Lazy::new(|| Mutex::new(()));
-static REGISTERED: Lazy<AtomicBool> = Lazy::new(|| AtomicBool::new(false));
-static REGISTERED_CLASSES: Lazy<thread_local::ThreadLocal<AtomicPtr<RegisteredClasses>>> =
-  Lazy::new(thread_local::ThreadLocal::new);
+static IS_FIRST_MODULE: AtomicBool = AtomicBool::new(true);
+static FIRST_MODULE_REGISTERED: AtomicBool = AtomicBool::new(false);
+static REGISTERED_CLASSES: Lazy<RegisteredClassesMap> = Lazy::new(Default::default);
 static FN_REGISTER_MAP: Lazy<FnRegisterMap> = Lazy::new(Default::default);
 
+type RegisteredClasses =
+  PersistedPerInstanceHashMap</* export name */ String, /* constructor */ sys::napi_ref>;
+
+#[cfg(feature = "compat-mode")]
+// compatibility for #[module_exports]
+static MODULE_EXPORTS: Lazy<PersistedPerInstanceVec<ModuleExportsCallback>> =
+  Lazy::new(Default::default);
+
 #[inline]
 fn wait_first_thread_registered() {
-  while !REGISTERED.load(Ordering::SeqCst) {
+  while !FIRST_MODULE_REGISTERED.load(Ordering::SeqCst) {
     std::hint::spin_loop();
   }
 }
 
-type RegisteredClasses =
-  HashMap</* export name */ String, /* constructor */ sys::napi_ref>;
-
-#[cfg(feature = "compat-mode")]
-// compatibility for #[module_exports]
-
-static MODULE_EXPORTS: Lazy<PersistedSingleThreadVec<ModuleExportsCallback>> =
-  Lazy::new(Default::default);
-
 #[doc(hidden)]
 pub fn get_class_constructor(js_name: &'static str) -> Option<sys::napi_ref> {
-  wait_first_thread_registered();
-  let registered_classes = REGISTERED_CLASSES.get().unwrap();
-  let registered_classes =
-    Box::leak(unsafe { Box::from_raw(registered_classes.load(Ordering::Relaxed)) });
-  registered_classes.get(js_name).copied()
+  let current_id = std::thread::current().id();
+  REGISTERED_CLASSES.borrow_mut(|map| {
+    map
+      .get(&current_id)
+      .map(|m| m.borrow_mut(|map| map.get(js_name).copied()))
+  })?
 }
 
 #[doc(hidden)]
@@ -203,7 +204,6 @@ pub fn register_class(
 /// ```
 ///
 pub fn get_js_function(env: &Env, raw_fn: ExportRegisterCallback) -> Result<JsFunction> {
-  wait_first_thread_registered();
   FN_REGISTER_MAP.borrow_mut(|inner| {
     inner
       .get(&raw_fn)
@@ -257,7 +257,6 @@ pub fn get_js_function(env: &Env, raw_fn: ExportRegisterCallback) -> Result<JsFu
 /// ```
 ///
 pub fn get_c_callback(raw_fn: ExportRegisterCallback) -> Result<crate::Callback> {
-  wait_first_thread_registered();
   FN_REGISTER_MAP.borrow_mut(|inner| {
     inner
       .get(&raw_fn)
@@ -271,20 +270,25 @@ pub fn get_c_callback(raw_fn: ExportRegisterCallback) -> Result<crate::Callback>
   })
 }
 
+#[cfg(windows)]
+#[ctor::ctor]
+fn load_host() {
+  unsafe {
+    sys::setup();
+  }
+}
+
 #[no_mangle]
 unsafe extern "C" fn napi_register_module_v1(
   env: sys::napi_env,
   exports: sys::napi_value,
 ) -> sys::napi_value {
-  #[cfg(windows)]
-  unsafe {
-    sys::setup();
+  if IS_FIRST_MODULE.load(Ordering::SeqCst) {
+    IS_FIRST_MODULE.store(false, Ordering::SeqCst);
+  } else {
+    wait_first_thread_registered();
   }
   crate::__private::___CALL_FROM_FACTORY.get_or_default();
-  let registered_classes_ptr = REGISTERED_CLASSES.get_or_default();
-  let lock = MODULE_REGISTER_LOCK
-    .lock()
-    .expect("Failed to acquire module register lock");
   let mut exports_objects: HashSet<String> = HashSet::default();
   MODULE_REGISTER_CALLBACK.borrow_mut(|inner| {
     inner
@@ -360,8 +364,7 @@ unsafe extern "C" fn napi_register_module_v1(
       })
   });
 
-  let mut registered_classes: RegisteredClasses =
-    HashMap::with_capacity(MODULE_CLASS_PROPERTIES.borrow_mut(|inner| inner.len()));
+  let mut registered_classes = HashMap::new();
 
   MODULE_CLASS_PROPERTIES.borrow_mut(|inner| {
     inner.iter().for_each(|(rust_name, js_mods)| {
@@ -447,10 +450,13 @@ unsafe extern "C" fn napi_register_module_v1(
         }
       }
     });
-    registered_classes_ptr.store(
-      Box::into_raw(Box::new(registered_classes)),
-      Ordering::Relaxed,
-    );
+
+    REGISTERED_CLASSES.borrow_mut(|map| {
+      map.insert(
+        std::thread::current().id(),
+        PersistedPerInstanceHashMap::from_hashmap(registered_classes),
+      )
+    });
   });
 
   #[cfg(feature = "compat-mode")]
@@ -462,26 +468,18 @@ unsafe extern "C" fn napi_register_module_v1(
     })
   });
 
-  #[cfg(all(feature = "tokio_rt", feature = "napi4"))]
+  #[cfg(all(windows, feature = "napi4", feature = "tokio_rt"))]
   {
-    let _ = crate::tokio_runtime::RT.clone();
-    crate::tokio_runtime::TOKIO_RT_REF_COUNT.fetch_add(1, Ordering::SeqCst);
-    assert_eq!(
-      unsafe {
-        sys::napi_add_env_cleanup_hook(
-          env,
-          Some(crate::shutdown_tokio_rt),
-          env as *mut std::ffi::c_void,
-        )
-      },
-      sys::Status::napi_ok
-    );
+    crate::tokio_runtime::RT_REFERENCE_COUNT.fetch_add(1, Ordering::SeqCst);
+    unsafe {
+      sys::napi_add_env_cleanup_hook(
+        env,
+        Some(crate::tokio_runtime::drop_runtime),
+        ptr::null_mut(),
+      )
+    };
   }
-  mem::drop(lock);
-
-  #[cfg(feature = "napi4")]
-  create_custom_gc(env);
-  REGISTERED.store(true, Ordering::SeqCst);
+  FIRST_MODULE_REGISTERED.store(true, Ordering::SeqCst);
   exports
 }
 
@@ -502,99 +500,3 @@ pub(crate) unsafe extern "C" fn noop(
   }
   ptr::null_mut()
 }
-
-#[cfg(feature = "napi4")]
-fn create_custom_gc(env: sys::napi_env) {
-  let mut custom_gc_fn = ptr::null_mut();
-  check_status_or_throw!(
-    env,
-    unsafe {
-      sys::napi_create_function(
-        env,
-        "custom_gc".as_ptr() as *const c_char,
-        9,
-        Some(empty),
-        ptr::null_mut(),
-        &mut custom_gc_fn,
-      )
-    },
-    "Create Custom GC Function in napi_register_module_v1 failed"
-  );
-  let mut async_resource_name = ptr::null_mut();
-  check_status_or_throw!(
-    env,
-    unsafe {
-      sys::napi_create_string_utf8(
-        env,
-        "CustomGC".as_ptr() as *const c_char,
-        8,
-        &mut async_resource_name,
-      )
-    },
-    "Create async resource string in napi_register_module_v1 napi_register_module_v1"
-  );
-  let mut custom_gc_tsfn = ptr::null_mut();
-  check_status_or_throw!(
-    env,
-    unsafe {
-      sys::napi_create_threadsafe_function(
-        env,
-        custom_gc_fn,
-        ptr::null_mut(),
-        async_resource_name,
-        0,
-        1,
-        ptr::null_mut(),
-        Some(custom_gc_finalize),
-        ptr::null_mut(),
-        Some(custom_gc),
-        &mut custom_gc_tsfn,
-      )
-    },
-    "Create Custom GC ThreadsafeFunction in napi_register_module_v1 failed"
-  );
-  check_status_or_throw!(
-    env,
-    unsafe { sys::napi_unref_threadsafe_function(env, custom_gc_tsfn) },
-    "Unref Custom GC ThreadsafeFunction in napi_register_module_v1 failed"
-  );
-  CUSTOM_GC_TSFN.store(custom_gc_tsfn, Ordering::SeqCst);
-  MAIN_THREAD_ID.get_or_init(|| std::thread::current().id());
-}
-
-#[cfg(feature = "napi4")]
-#[allow(unused)]
-unsafe extern "C" fn empty(env: sys::napi_env, info: sys::napi_callback_info) -> sys::napi_value {
-  ptr::null_mut()
-}
-
-#[cfg(feature = "napi4")]
-#[allow(unused)]
-unsafe extern "C" fn custom_gc_finalize(
-  env: sys::napi_env,
-  finalize_data: *mut c_void,
-  finalize_hint: *mut c_void,
-) {
-  CUSTOM_GC_TSFN_CLOSED.store(true, Ordering::SeqCst);
-}
-
-#[cfg(feature = "napi4")]
-// recycle the ArrayBuffer/Buffer Reference if the ArrayBuffer/Buffer is not dropped on the main thread
-extern "C" fn custom_gc(
-  env: sys::napi_env,
-  _js_callback: sys::napi_value,
-  _context: *mut c_void,
-  data: *mut c_void,
-) {
-  let mut ref_count = 0;
-  check_status_or_throw!(
-    env,
-    unsafe { sys::napi_reference_unref(env, data as sys::napi_ref, &mut ref_count) },
-    "Failed to unref Buffer reference in Custom GC"
-  );
-  check_status_or_throw!(
-    env,
-    unsafe { sys::napi_delete_reference(env, data as sys::napi_ref) },
-    "Failed to delete Buffer reference in Custom GC"
-  );
-}
diff --git a/crates/napi/src/env.rs b/crates/napi/src/env.rs
index 8b8568fc..7e5e473b 100644
--- a/crates/napi/src/env.rs
+++ b/crates/napi/src/env.rs
@@ -115,7 +115,7 @@ impl Env {
   #[cfg(feature = "napi6")]
   pub fn create_bigint_from_i128(&self, value: i128) -> Result<JsBigInt> {
     let mut raw_value = ptr::null_mut();
-    let sign_bit = if value > 0 { 0 } else { 1 };
+    let sign_bit = i32::from(value <= 0);
     let words = &value as *const i128 as *const u64;
     check_status!(unsafe {
       sys::napi_create_bigint_words(self.0, sign_bit, 2, words, &mut raw_value)
diff --git a/crates/napi/src/js_values/bigint.rs b/crates/napi/src/js_values/bigint.rs
index 245cddb1..2725c7be 100644
--- a/crates/napi/src/js_values/bigint.rs
+++ b/crates/napi/src/js_values/bigint.rs
@@ -204,7 +204,7 @@ impl TryFrom<JsBigInt> for u64 {
 impl JsBigInt {
   /// <https://nodejs.org/api/n-api.html#n_api_napi_get_value_bigint_words>
   pub fn get_words(&mut self) -> Result<(bool, Vec<u64>)> {
-    let mut words: Vec<u64> = Vec::with_capacity(self.word_count as usize);
+    let mut words: Vec<u64> = Vec::with_capacity(self.word_count);
     let word_count = &mut self.word_count;
     let mut sign_bit = 0;
     check_status!(unsafe {
@@ -218,7 +218,7 @@ impl JsBigInt {
     })?;
 
     unsafe {
-      words.set_len(self.word_count as usize);
+      words.set_len(self.word_count);
     };
 
     Ok((sign_bit == 1, words))
diff --git a/crates/napi/src/js_values/string/mod.rs b/crates/napi/src/js_values/string/mod.rs
index 0e5eeff8..313e46b1 100644
--- a/crates/napi/src/js_values/string/mod.rs
+++ b/crates/napi/src/js_values/string/mod.rs
@@ -35,7 +35,7 @@ impl JsString {
     check_status!(unsafe {
       sys::napi_get_value_string_utf8(self.0.env, self.0.value, ptr::null_mut(), 0, &mut length)
     })?;
-    Ok(length as usize)
+    Ok(length)
   }
 
   pub fn utf16_len(&self) -> Result<usize> {
@@ -43,7 +43,7 @@ impl JsString {
     check_status!(unsafe {
       sys::napi_get_value_string_utf16(self.0.env, self.0.value, ptr::null_mut(), 0, &mut length)
     })?;
-    Ok(length as usize)
+    Ok(length)
   }
 
   pub fn latin1_len(&self) -> Result<usize> {
@@ -51,7 +51,7 @@ impl JsString {
     check_status!(unsafe {
       sys::napi_get_value_string_latin1(self.0.env, self.0.value, ptr::null_mut(), 0, &mut length)
     })?;
-    Ok(length as usize)
+    Ok(length)
   }
 
   pub fn into_utf8(self) -> Result<JsStringUtf8> {
diff --git a/crates/napi/src/lib.rs b/crates/napi/src/lib.rs
index 9d5de8a6..09323b75 100644
--- a/crates/napi/src/lib.rs
+++ b/crates/napi/src/lib.rs
@@ -109,8 +109,6 @@ pub use error::*;
 pub use js_values::*;
 pub use status::Status;
 pub use task::Task;
-#[cfg(all(feature = "tokio_rt", feature = "napi4"))]
-pub use tokio_runtime::shutdown_tokio_rt;
 pub use value_type::*;
 pub use version::NodeVersion;
 #[cfg(feature = "serde-json")]
diff --git a/crates/napi/src/tokio_runtime.rs b/crates/napi/src/tokio_runtime.rs
index 932d387d..ef923ac5 100644
--- a/crates/napi/src/tokio_runtime.rs
+++ b/crates/napi/src/tokio_runtime.rs
@@ -1,54 +1,33 @@
-use std::ffi::c_void;
 use std::future::Future;
 use std::ptr;
-use std::sync::atomic::{AtomicUsize, Ordering};
 
 use once_cell::sync::Lazy;
-use tokio::{
-  runtime::Handle,
-  sync::mpsc::{self, error::TrySendError},
-};
+use tokio::runtime::Runtime;
 
 use crate::{check_status, sys, JsDeferred, JsUnknown, NapiValue, Result};
 
-pub(crate) static RT: Lazy<(Handle, mpsc::Sender<()>)> = Lazy::new(|| {
-  let runtime = tokio::runtime::Runtime::new();
-  let (sender, mut receiver) = mpsc::channel::<()>(1);
-  runtime
-    .map(|rt| {
-      let h = rt.handle();
-      let handle = h.clone();
-      handle.spawn(async move {
-        if receiver.recv().await.is_some() {
-          rt.shutdown_background();
-        }
-      });
-
-      (handle, sender)
-    })
-    .expect("Create tokio runtime failed")
+pub(crate) static mut RT: Lazy<Option<Runtime>> = Lazy::new(|| {
+  let runtime = tokio::runtime::Runtime::new().expect("Create tokio runtime failed");
+  Some(runtime)
 });
 
-pub(crate) static TOKIO_RT_REF_COUNT: AtomicUsize = AtomicUsize::new(0);
+#[cfg(windows)]
+pub(crate) static RT_REFERENCE_COUNT: std::sync::atomic::AtomicUsize =
+  std::sync::atomic::AtomicUsize::new(0);
 
-#[doc(hidden)]
-#[inline(never)]
-pub unsafe extern "C" fn shutdown_tokio_rt(arg: *mut c_void) {
-  if TOKIO_RT_REF_COUNT.fetch_sub(1, Ordering::SeqCst) == 0 {
-    let sender = &RT.1;
-    if let Err(e) = sender.clone().try_send(()) {
-      match e {
-        TrySendError::Closed(_) => {}
-        TrySendError::Full(_) => {
-          panic!("Send shutdown signal to tokio runtime failed, queue is full");
-        }
-      }
+#[cfg(windows)]
+pub(crate) unsafe extern "C" fn drop_runtime(arg: *mut std::ffi::c_void) {
+  use std::sync::atomic::Ordering;
+
+  if RT_REFERENCE_COUNT.fetch_sub(1, Ordering::SeqCst) == 1 {
+    if let Some(rt) = Lazy::get_mut(unsafe { &mut RT }) {
+      rt.take();
     }
   }
 
   unsafe {
     let env: sys::napi_env = arg as *mut sys::napi_env__;
-    sys::napi_remove_env_cleanup_hook(env, Some(shutdown_tokio_rt), arg);
+    sys::napi_remove_env_cleanup_hook(env, Some(drop_runtime), arg);
   }
 }
 
@@ -60,7 +39,7 @@ pub fn spawn<F>(fut: F) -> tokio::task::JoinHandle<F::Output>
 where
   F: 'static + Send + Future<Output = ()>,
 {
-  RT.0.spawn(fut)
+  unsafe { RT.as_ref() }.unwrap().spawn(fut)
 }
 
 /// Runs a future to completion
@@ -70,7 +49,7 @@ pub fn block_on<F>(fut: F) -> F::Output
 where
   F: 'static + Send + Future<Output = ()>,
 {
-  RT.0.block_on(fut)
+  unsafe { RT.as_ref() }.unwrap().block_on(fut)
 }
 
 // This function's signature must be kept in sync with the one in lib.rs, otherwise napi
@@ -78,8 +57,9 @@ where
 
 /// If the feature `tokio_rt` has been enabled this will enter the runtime context and
 /// then call the provided closure. Otherwise it will just call the provided closure.
+#[inline]
 pub fn within_runtime_if_available<F: FnOnce() -> T, T>(f: F) -> T {
-  let _rt_guard = RT.0.enter();
+  let _rt_guard = unsafe { RT.as_ref() }.unwrap().enter();
   f()
 }
 
diff --git a/crates/sys/src/functions.rs b/crates/sys/src/functions.rs
index 6d24f501..3041d1e4 100644
--- a/crates/sys/src/functions.rs
+++ b/crates/sys/src/functions.rs
@@ -738,7 +738,7 @@ pub use napi7::*;
 pub use napi8::*;
 
 #[cfg(windows)]
-pub(super) unsafe fn load() -> Result<(), libloading::Error> {
+pub(super) unsafe fn load() -> Result<libloading::Library, libloading::Error> {
   let host = match libloading::os::windows::Library::this() {
     Ok(lib) => lib.into(),
     Err(err) => {
@@ -764,5 +764,5 @@ pub(super) unsafe fn load() -> Result<(), libloading::Error> {
   napi8::load(&host)?;
   #[cfg(feature = "experimental")]
   experimental::load(&host)?;
-  Ok(())
+  Ok(host)
 }
diff --git a/crates/sys/src/lib.rs b/crates/sys/src/lib.rs
index a0d12621..5cf4fa5b 100644
--- a/crates/sys/src/lib.rs
+++ b/crates/sys/src/lib.rs
@@ -77,28 +77,21 @@ macro_rules! generate {
   };
 }
 
-#[cfg(windows)]
-use std::sync::Once;
-
 mod functions;
 mod types;
 
 pub use functions::*;
 pub use types::*;
 
-#[cfg(windows)]
-static SETUP: Once = Once::new();
-
 /// Loads N-API symbols from host process.
 /// Must be called at least once before using any functions in bindings or
 /// they will panic.
 /// Safety: `env` must be a valid `napi_env` for the current thread
 #[cfg(windows)]
 #[allow(clippy::missing_safety_doc)]
-pub unsafe fn setup() {
-  SETUP.call_once(|| {
-    if let Err(err) = load() {
-      panic!("{}", err);
-    }
-  });
+pub unsafe fn setup() -> libloading::Library {
+  match load() {
+    Err(err) => panic!("{}", err),
+    Ok(l) => l,
+  }
 }
diff --git a/examples/napi-compat-mode/src/object.rs b/examples/napi-compat-mode/src/object.rs
index a8f62ca6..bf5649d9 100644
--- a/examples/napi-compat-mode/src/object.rs
+++ b/examples/napi-compat-mode/src/object.rs
@@ -87,7 +87,7 @@ fn test_delete_named_property(ctx: CallContext) -> Result<JsBoolean> {
 fn test_get_property(ctx: CallContext) -> Result<JsUnknown> {
   let obj = ctx.get::<JsObject>(0)?;
   let key = ctx.get::<JsUnknown>(1)?;
-  obj.get_property(&key)
+  obj.get_property(key)
 }
 
 #[js_function(1)]
diff --git a/examples/napi/__test__/unload.spec.js b/examples/napi/__test__/unload.spec.js
new file mode 100644
index 00000000..21a32ca2
--- /dev/null
+++ b/examples/napi/__test__/unload.spec.js
@@ -0,0 +1,11 @@
+// use the commonjs syntax to prevent compiler from transpiling the module syntax
+
+const test = require('ava').default
+
+test('unload module', (t) => {
+  const { add } = require('../index.node')
+  t.is(add(1, 2), 3)
+  delete require.cache[require.resolve('../index.node')]
+  const { add: add2 } = require('../index.node')
+  t.is(add2(1, 2), 3)
+})
diff --git a/examples/napi/src/either.rs b/examples/napi/src/either.rs
index ab00955b..a316a603 100644
--- a/examples/napi/src/either.rs
+++ b/examples/napi/src/either.rs
@@ -22,13 +22,7 @@ fn either3(input: Either3<String, u32, bool>) -> u32 {
   match input {
     Either3::A(s) => s.len() as u32,
     Either3::B(n) => n,
-    Either3::C(b) => {
-      if b {
-        1
-      } else {
-        0
-      }
-    }
+    Either3::C(b) => u32::from(b),
   }
 }
 
@@ -42,13 +36,7 @@ fn either4(input: Either4<String, u32, bool, Obj>) -> u32 {
   match input {
     Either4::A(s) => s.len() as u32,
     Either4::B(n) => n,
-    Either4::C(b) => {
-      if b {
-        1
-      } else {
-        0
-      }
-    }
+    Either4::C(b) => u32::from(b),
     Either4::D(f) => match f.v {
       Either::A(s) => s.len() as u32,
       Either::B(n) => n,
diff --git a/examples/napi/src/typed_array.rs b/examples/napi/src/typed_array.rs
index 9cc5ae14..2d7caaef 100644
--- a/examples/napi/src/typed_array.rs
+++ b/examples/napi/src/typed_array.rs
@@ -69,5 +69,5 @@ impl Task for AsyncBuffer {
 
 #[napi]
 fn async_reduce_buffer(buf: Buffer) -> Result<AsyncTask<AsyncBuffer>> {
-  Ok(AsyncTask::new(AsyncBuffer { buf: buf.clone() }))
+  Ok(AsyncTask::new(AsyncBuffer { buf }))
 }
diff --git a/memory-testing/src/lib.rs b/memory-testing/src/lib.rs
index b41c1362..1b0c432d 100644
--- a/memory-testing/src/lib.rs
+++ b/memory-testing/src/lib.rs
@@ -75,7 +75,7 @@ pub fn test_async(env: Env) -> napi::Result<napi::JsObject> {
 
 #[napi]
 pub fn from_js(env: Env, input_object: Object) -> napi::Result<String> {
-  let a: Welcome = env.from_js_value(&input_object)?;
+  let a: Welcome = env.from_js_value(input_object)?;
   Ok(serde_json::to_string(&a)?)
 }