diff --git a/include/pybind11/cast.h b/include/pybind11/cast.h
index 6434d38a9..679ab7ab7 100644
--- a/include/pybind11/cast.h
+++ b/include/pybind11/cast.h
@@ -117,6 +117,18 @@ PYBIND11_NOINLINE inline handle get_object_handle(const void *ptr) {
     return handle((PyObject *) it->second);
 }
 
+inline PyThreadState *get_thread_state_unchecked() {
+#if   PY_VERSION_HEX < 0x03000000
+    return _PyThreadState_Current;
+#elif PY_VERSION_HEX < 0x03050000
+    return (PyThreadState*) _Py_atomic_load_relaxed(&_PyThreadState_Current);
+#elif PY_VERSION_HEX < 0x03050200
+    return (PyThreadState*) _PyThreadState_Current.value;
+#else
+    return _PyThreadState_UncheckedGet();
+#endif
+}
+
 class type_caster_generic {
 public:
     PYBIND11_NOINLINE type_caster_generic(const std::type_info &type_info)
diff --git a/include/pybind11/common.h b/include/pybind11/common.h
index 1cf0cd9b6..8aa0075cb 100644
--- a/include/pybind11/common.h
+++ b/include/pybind11/common.h
@@ -110,6 +110,13 @@
     extern "C" PYBIND11_EXPORT PyObject *init##name()
 #endif
 
+#if PY_VERSION_HEX >= 0x03050000 && PY_VERSION_HEX < 0x03050200
+extern "C" {
+    struct _Py_atomic_address { void *value; };
+    PyAPI_DATA(_Py_atomic_address) _PyThreadState_Current;
+};
+#endif
+
 #define PYBIND11_TRY_NEXT_OVERLOAD ((PyObject *) 1) // special failure return code
 #define PYBIND11_STRINGIFY(x) #x
 #define PYBIND11_TOSTRING(x) PYBIND11_STRINGIFY(x)
diff --git a/include/pybind11/pybind11.h b/include/pybind11/pybind11.h
index cf7f4b7b2..7c3642115 100644
--- a/include/pybind11/pybind11.h
+++ b/include/pybind11/pybind11.h
@@ -1053,11 +1053,13 @@ template <typename InputType, typename OutputType> void implicitly_convertible()
  *
  * 3. The reference count of an acquired thread state can be controlled. This
  *    can be handy to prevent cases where callbacks issued from an external
- *    thread constantly construct and destroy thread state data structures. */
+ *    thread would otherwise constantly construct and destroy thread state data
+ *    structures.
+ */
 
 class gil_scoped_acquire {
 public:
-    gil_scoped_acquire() {
+    PYBIND11_NOINLINE gil_scoped_acquire() {
         auto const &internals = detail::get_internals();
         tstate = (PyThreadState *) PyThread_get_key_value(internals.tstate);
 
@@ -1068,17 +1070,24 @@ public:
                     pybind11_fail("scoped_acquire: could not create thread state!");
             #endif
             tstate->gilstate_counter = 0;
+            #if PY_MAJOR_VERSION < 3
+                PyThread_delete_key_value(internals.tstate);
+            #endif
             PyThread_set_key_value(internals.tstate, tstate);
         } else {
-            release = PyThreadState_GET() != tstate;
+            release = detail::get_thread_state_unchecked() != tstate;
         }
 
         if (release) {
-            PyInterpreterState *interp = tstate->interp;
             /* Work around an annoying assertion in PyThreadState_Swap */
-            tstate->interp = nullptr;
+            #if defined(Py_DEBUG)
+                PyInterpreterState *interp = tstate->interp;
+                tstate->interp = nullptr;
+            #endif
             PyEval_AcquireThread(tstate);
-            tstate->interp = interp;
+            #if defined(Py_DEBUG)
+                tstate->interp = interp;
+            #endif
         }
 
         inc_ref();
@@ -1088,10 +1097,10 @@ public:
         ++tstate->gilstate_counter;
     }
 
-    void dec_ref() {
+    PYBIND11_NOINLINE void dec_ref() {
         --tstate->gilstate_counter;
         #if !defined(NDEBUG)
-            if (PyThreadState_GET() != tstate)
+            if (detail::get_thread_state_unchecked() != tstate)
                 pybind11_fail("scoped_acquire::dec_ref(): thread state must be current!");
             if (tstate->gilstate_counter < 0)
                 pybind11_fail("scoped_acquire::dec_ref(): reference count underflow!");
@@ -1103,12 +1112,12 @@ public:
             #endif
             PyThreadState_Clear(tstate);
             PyThreadState_DeleteCurrent();
-            PyThread_set_key_value(detail::get_internals().tstate, nullptr);
+            PyThread_delete_key_value(detail::get_internals().tstate);
             release = false;
         }
     }
 
-    ~gil_scoped_acquire() {
+    PYBIND11_NOINLINE ~gil_scoped_acquire() {
         dec_ref();
         if (release)
            PyEval_SaveThread();