diff --git a/lib/libthr/thread/thr_exit.c b/lib/libthr/thread/thr_exit.c
--- a/lib/libthr/thread/thr_exit.c
+++ b/lib/libthr/thread/thr_exit.c
@@ -55,7 +55,6 @@
 
 static void thread_unwind(void) __dead2;
 #ifdef PIC
-static void thread_uw_init(void);
 static _Unwind_Reason_Code thread_unwind_stop(int version,
 	_Unwind_Action actions,
 	uint64_t exc_class,
@@ -66,8 +65,8 @@
 	_Unwind_Stop_Fn, void *);
 static uintptr_t (*uwl_getcfa)(struct _Unwind_Context *);
 
-static void
-thread_uw_init(void)
+void
+_thread_uw_init(struct pthread *curthread)
 {
 	static int inited = 0;
 	Dl_info dli;
@@ -83,13 +82,14 @@
 		 * Make sure the address is always valid by holding the library,
 		 * also assume functions are in same library.
 		 */
-		if ((handle = dlopen(dli.dli_fname, RTLD_LAZY)) != NULL) {
+		if ((handle = dlopen(dli.dli_fname, RTLD_NOW)) != NULL) {
 		    forcedunwind = dlsym(handle, "_Unwind_ForcedUnwind");
 		    getcfa = dlsym(handle, "_Unwind_GetCFA");
 		    if (forcedunwind != NULL && getcfa != NULL) {
 			uwl_getcfa = getcfa;
 			atomic_store_rel_ptr((volatile void *)&uwl_forcedunwind,
 				(uintptr_t)forcedunwind);
+			curthread->unwind_dlhandle = handle;
 		    } else {
 			dlclose(handle);
 		    }
@@ -103,7 +103,11 @@
 _Unwind_ForcedUnwind(struct _Unwind_Exception *ex, _Unwind_Stop_Fn stop_func,
 	void *stop_arg)
 {
-	return (*uwl_forcedunwind)(ex, stop_func, stop_arg);
+	struct pthread *curthread = _get_curthread();
+	_Unwind_Reason_Code r = (*uwl_forcedunwind)(ex, stop_func, stop_arg);
+	if (curthread->unwind_dlhandle != NULL)
+		dlclose(curthread->unwind_dlhandle);
+	return r;
 }
 
 uintptr_t
@@ -238,7 +242,6 @@
 #ifdef _PTHREAD_FORCED_UNWIND
 
 #ifdef PIC
-	thread_uw_init();
 	if (uwl_forcedunwind != NULL) {
 #else
 	if (_Unwind_ForcedUnwind != NULL) {
@@ -249,17 +252,17 @@
 				_thread_printf(2, "Warning: old _pthread_cleanup_push was called, "
 				  	"stack unwinding is disabled.\n");
 			}
+			if (curthread->unwind_dlhandle != NULL)
+				dlclose(curthread->unwind_dlhandle);
 			goto cleanup;
 		}
 		thread_unwind();
-
 	} else {
 cleanup:
 		while (curthread->cleanup != NULL) {
 			__pthread_cleanup_pop_imp(1);
 		}
 		__cxa_thread_call_dtors();
-
 		exit_thread();
 	}
 
diff --git a/lib/libthr/thread/thr_fork.c b/lib/libthr/thread/thr_fork.c
--- a/lib/libthr/thread/thr_fork.c
+++ b/lib/libthr/thread/thr_fork.c
@@ -97,7 +97,7 @@
 	_thr_rwl_wrlock(&_thr_atfork_lock);
 	TAILQ_INSERT_TAIL(&_thr_atfork_list, af, qe);
 	_thr_rwl_unlock(&_thr_atfork_lock);
-	THR_CRITICAL_LEAVE(curthread);
+	THR_CRITICAL_LEAVE(curthread, 1);
 	return (0);
 }
 
@@ -122,7 +122,7 @@
 		}
 	}
 	_thr_rwl_unlock(&_thr_atfork_lock);
-	THR_CRITICAL_LEAVE(curthread);
+	THR_CRITICAL_LEAVE(curthread, 1);
 	while ((af = TAILQ_FIRST(&temp_list)) != NULL) {
 		TAILQ_REMOVE(&temp_list, af, qe);
 		free(af);
diff --git a/lib/libthr/thread/thr_init.c b/lib/libthr/thread/thr_init.c
--- a/lib/libthr/thread/thr_init.c
+++ b/lib/libthr/thread/thr_init.c
@@ -427,6 +427,9 @@
 
 #ifdef _PTHREAD_FORCED_UNWIND
 	thread->unwind_stackend = _usrstack;
+#ifdef PIC
+        _thread_uw_init(thread);
+#endif
 #endif
 
 	/* Others cleared to zero by thr_alloc() */
diff --git a/lib/libthr/thread/thr_list.c b/lib/libthr/thread/thr_list.c
--- a/lib/libthr/thread/thr_list.c
+++ b/lib/libthr/thread/thr_list.c
@@ -311,7 +311,7 @@
 	THR_THREAD_LOCK(curthread, thread);
 	thread->refcount--;
 	_thr_try_gc(curthread, thread);
-	THR_CRITICAL_LEAVE(curthread);
+	THR_CRITICAL_LEAVE(curthread, 1);
 }
 
 /* entered with thread lock held, exit with thread lock released */
diff --git a/lib/libthr/thread/thr_mutex.c b/lib/libthr/thread/thr_mutex.c
--- a/lib/libthr/thread/thr_mutex.c
+++ b/lib/libthr/thread/thr_mutex.c
@@ -639,7 +639,7 @@
 		_mutex_leave_robust(curthread, m);
 	if (ret != 0 && ret != EOWNERDEAD &&
 	    (m->m_flags & PMUTEX_FLAG_PRIVATE) != 0)
-		THR_CRITICAL_LEAVE(curthread);
+		THR_CRITICAL_LEAVE(curthread, 1);
 	return (ret);
 }
 
@@ -736,7 +736,7 @@
 		_mutex_leave_robust(curthread, m);
 	if (ret != 0 && ret != EOWNERDEAD &&
 	    (m->m_flags & PMUTEX_FLAG_PRIVATE) != 0 && !cvattach)
-		THR_CRITICAL_LEAVE(curthread);
+		THR_CRITICAL_LEAVE(curthread, 1);
 	return (ret);
 }
 
@@ -997,7 +997,7 @@
 			_mutex_leave_robust(curthread, m);
 	}
 	if (!cv && private)
-		THR_CRITICAL_LEAVE(curthread);
+		THR_CRITICAL_LEAVE(curthread, 1);
 	return (error);
 }
 
diff --git a/lib/libthr/thread/thr_private.h b/lib/libthr/thread/thr_private.h
--- a/lib/libthr/thread/thr_private.h
+++ b/lib/libthr/thread/thr_private.h
@@ -525,6 +525,7 @@
 	struct _Unwind_Exception	ex;
 	void			*unwind_stackend;
 	int			unwind_disabled;
+	void			*unwind_dlhandle;
 #endif
 
 	/*
@@ -587,12 +588,15 @@
 	((thrd)->critical_count > 0))
 
 #define	THR_CRITICAL_ENTER(thrd)			\
-	(thrd)->critical_count++
+	do {						\
+		(thrd)->critical_count++;		\
+	} while (0)
 
-#define	THR_CRITICAL_LEAVE(thrd)			\
+#define	THR_CRITICAL_LEAVE(thrd, do_ast)		\
 	do {						\
 		(thrd)->critical_count--;		\
-		_thr_ast(thrd);				\
+		if (do_ast)				\
+			_thr_ast(thrd);			\
 	} while (0)
 
 #define THR_UMUTEX_TRYLOCK(thrd, lck)			\
@@ -701,7 +705,7 @@
 
 #define THR_REF_DEL(curthread, pthread) {			\
 	pthread->refcount--;					\
-	THR_CRITICAL_LEAVE(curthread);				\
+	THR_CRITICAL_LEAVE(curthread, 1);			\
 } while (0)
 
 #define GC_NEEDED()	(_gc_count >= 5)
@@ -1103,6 +1107,9 @@
 bool __thr_get_main_stack_base(char **base);
 bool __thr_get_main_stack_lim(size_t *lim);
 int _Tthr_sigqueue(pthread_t pthread, int sig, const union sigval value);
+#ifdef PIC
+void _thread_uw_init(struct pthread *);
+#endif
 
 __END_DECLS
 __NULLABILITY_PRAGMA_POP
diff --git a/lib/libthr/thread/thr_rtld.c b/lib/libthr/thread/thr_rtld.c
--- a/lib/libthr/thread/thr_rtld.c
+++ b/lib/libthr/thread/thr_rtld.c
@@ -168,7 +168,14 @@
 	if (_thr_rwlock_unlock(&l->lock) == 0) {
 		if ((state & URWLOCK_WRITE_OWNER) == 0)
 			curthread->rdlock_count--;
-		THR_CRITICAL_LEAVE(curthread);
+		THR_CRITICAL_LEAVE(curthread, 1);
+	} else if (curthread->cancelling) {
+		/*
+		 * If the main thread is being destroyed as well and
+		 * tries to cancel us, some of the rtld locks might
+		 * already be gone and unlock() will err.
+		 */
+		THR_CRITICAL_LEAVE(curthread, 0);
 	}
 	RESTORE_ERRNO();
 }
diff --git a/lib/libthr/thread/thr_sig.c b/lib/libthr/thread/thr_sig.c
--- a/lib/libthr/thread/thr_sig.c
+++ b/lib/libthr/thread/thr_sig.c
@@ -426,7 +426,7 @@
 	 * because we are leaf code, we don't want to recursively call
 	 * ourself.
 	 */
-	curthread->critical_count++;
+	THR_CRITICAL_ENTER(curthread);
 	THR_UMUTEX_LOCK(curthread, &(curthread)->lock);
 	while ((curthread->flags & THR_FLAGS_NEED_SUSPEND) != 0) {
 		curthread->cycle++;
@@ -447,7 +447,7 @@
 		THR_UMUTEX_LOCK(curthread, &(curthread)->lock);
 	}
 	THR_UMUTEX_UNLOCK(curthread, &(curthread)->lock);
-	curthread->critical_count--;
+	THR_CRITICAL_LEAVE(curthread, 0);
 
 	_thr_signal_unblock(curthread);
 }
diff --git a/libexec/rtld-elf/rtld.c b/libexec/rtld-elf/rtld.c
--- a/libexec/rtld-elf/rtld.c
+++ b/libexec/rtld-elf/rtld.c
@@ -1044,8 +1044,10 @@
     where = (Elf_Addr *)(obj->relocbase + rel->r_offset);
     def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj, SYMLOOK_IN_PLT,
 	NULL, &lockstate);
-    if (def == NULL)
+    if (def == NULL) {
+	lock_release(rtld_bind_lock, &lockstate);
 	rtld_die();
+    }
     if (ELF_ST_TYPE(def->st_info) == STT_GNU_IFUNC)
 	target = (Elf_Addr)rtld_resolve_ifunc(defobj, def);
     else