diff --git a/winsup/cygwin/cygserver_ipc.h b/winsup/cygwin/cygserver_ipc.h
index 31eea3464..2f7c3531f 100644
--- a/winsup/cygwin/cygserver_ipc.h
+++ b/winsup/cygwin/cygserver_ipc.h
@@ -43,7 +43,7 @@ ipc_set_proc_info (proc &blk)
   blk.gidcnt = 0;
   blk.gidlist = NULL;
   blk.is_admin = false;
-  blk.signal_arrived = _my_tls.signal_arrived;
+  blk.signal_arrived = signal_arrived;
 }
 #endif /* __INSIDE_CYGWIN__ */
 
diff --git a/winsup/cygwin/cygthread.cc b/winsup/cygwin/cygthread.cc
index 604521c58..4cbd3b064 100644
--- a/winsup/cygwin/cygthread.cc
+++ b/winsup/cygwin/cygthread.cc
@@ -373,16 +373,20 @@ cygthread::detach (HANDLE sigwait)
 	  LONG prio = GetThreadPriority (hth);
 	  ::SetThreadPriority (hth, THREAD_PRIORITY_BELOW_NORMAL);
 
+	  HANDLE w4[2];
+	  unsigned n = 2;
 	  DWORD howlong = INFINITE;
+	  w4[0] = sigwait;
+	  w4[1] = signal_arrived;
 	  /* For a description of the below loop see the end of this file */
 	  for (int i = 0; i < 2; i++)
-	    switch (res = cygwait (sigwait, howlong))
+	    switch (res = WaitForMultipleObjects (n, w4, FALSE, howlong))
 	      {
 	      case WAIT_OBJECT_0:
 		if (n == 1)
 		  howlong = 50;
 		break;
-	      case WAIT_SIGNALED:
+	      case WAIT_OBJECT_0 + 1:
 		n = 1;
 		if (i--)
 		  howlong = 50;
@@ -391,7 +395,20 @@ cygthread::detach (HANDLE sigwait)
 		break;
 	      default:
 		if (!exiting)
-		  api_fatal ("WFMO failed waiting for cygthread '%s', %E", __name);
+		  {
+		    system_printf ("WFMO failed waiting for cygthread '%s', %E", __name);
+		    for (unsigned j = 0; j < n; j++)
+		      switch (WaitForSingleObject (w4[j], 0))
+			{
+			case WAIT_OBJECT_0:
+			case WAIT_TIMEOUT:
+			  break;
+			default:
+			  system_printf ("%s handle %p is bad", (j ? "signal_arrived" : "semaphore"), w4[j]);
+			  break;
+			}
+		    api_fatal ("exiting on fatal error");
+		  }
 		break;
 	      }
 	  /* WAIT_OBJECT_0 means that the thread successfully read something,
diff --git a/winsup/cygwin/cygtls.h b/winsup/cygwin/cygtls.h
index e4240984e..40c6151c2 100644
--- a/winsup/cygwin/cygtls.h
+++ b/winsup/cygwin/cygtls.h
@@ -174,8 +174,7 @@ public:
   int sa_flags;
   sigset_t oldmask;
   sigset_t deltamask;
-  HANDLE signal_arrived;
-  bool waiting;
+  HANDLE event;
   int *errno_addr;
   sigset_t sigmask;
   sigset_t sigwait_mask;
diff --git a/winsup/cygwin/cygwait.cc b/winsup/cygwin/cygwait.cc
index 377012a1a..9b22bd01f 100644
--- a/winsup/cygwin/cygwait.cc
+++ b/winsup/cygwin/cygwait.cc
@@ -21,8 +21,6 @@
 
 #define is_cw_sig_handle	(mask & (is_cw_sig | is_cw_sig_eintr))
 
-TIMER_BASIC_INFORMATION cw_nowait;
-
 DWORD
 cancelable_wait (HANDLE object, PLARGE_INTEGER timeout, unsigned mask)
 {
@@ -85,11 +83,8 @@ cancelable_wait (HANDLE object, PLARGE_INTEGER timeout, unsigned mask)
 	/* all set */;
       else if (is_cw_sig_eintr)
 	res = WAIT_SIGNALED;
-      else
-	{
-	  _my_tls.call_signal_handler ();
-	  continue;
-	}
+      else if (_my_tls.call_signal_handler () || &_my_tls != _main_tls)
+	continue;
       break;
     }
 
diff --git a/winsup/cygwin/cygwait.h b/winsup/cygwin/cygwait.h
index a7daf88eb..9b2be02ac 100644
--- a/winsup/cygwin/cygwait.h
+++ b/winsup/cygwin/cygwait.h
@@ -11,6 +11,9 @@
 
 #pragma once
 
+#define WAIT_CANCELED   (WAIT_OBJECT_0 + 2)
+#define WAIT_SIGNALED  (WAIT_OBJECT_0 + 1)
+
 enum cw_wait_mask
 {
   cw_cancel =		0x0001,
@@ -19,8 +22,6 @@ enum cw_wait_mask
   cw_sig_eintr =	0x0008
 };
 
-extern TIMER_BASIC_INFORMATION cw_nowait;
-
 const unsigned cw_std_mask = cw_cancel | cw_cancel_self | cw_sig;
 
 DWORD cancelable_wait (HANDLE, PLARGE_INTEGER timeout = NULL,
@@ -28,7 +29,7 @@ DWORD cancelable_wait (HANDLE, PLARGE_INTEGER timeout = NULL,
   __attribute__ ((regparm (3)));
 
 static inline DWORD __attribute__ ((always_inline))
-cancelable_wait (HANDLE h, DWORD howlong, unsigned mask)
+cygwait (HANDLE h, DWORD howlong = INFINITE)
 {
   PLARGE_INTEGER pli_howlong;
   LARGE_INTEGER li_howlong;
@@ -39,14 +40,7 @@ cancelable_wait (HANDLE h, DWORD howlong, unsigned mask)
       li_howlong.QuadPart = 10000ULL * howlong;
       pli_howlong = &li_howlong;
     }
-
-  return cancelable_wait (h, pi_howlong, mask);
-}
-
-static inline DWORD __attribute__ ((always_inline))
-cygwait (HANDLE h, DWORD howlong = INFINITE)
-{
-  return cancelable_wait (h, howlong, cw_cancel | cw_sig_eintr);
+  return cancelable_wait (h, pli_howlong, cw_cancel | cw_sig);
 }
 
 static inline DWORD __attribute__ ((always_inline))
@@ -54,29 +48,3 @@ cygwait (DWORD howlong)
 {
   return cygwait ((HANDLE) NULL, howlong);
 }
-
-class set_thread_waiting
-{
-  void doit (bool setit, DWORD& here)
-  {
-    if (setit)
-      {
-	if (_my_tls.signal_arrived == NULL)
-	  _my_tls.signal_arrived = CreateEvent (&sec_none_nih, false, false, NULL);
-	here = _my_tls.signal_arrived;
-	_my_tls.waiting = true;
-      }
-  }
-public:
-  set_thread_waiting (bool setit, DWORD& here) { doit (setit, here); }
-  set_thread_waiting (DWORD& here) { doit (true, here); }
-
-  ~set_thread_waiting ()
-  {
-    if (_my_tls.waiting)
-      {
-	_my_tls.waiting = false;
-	ResetEvent (_my_tls.signal_arrived);
-      }
-  }
-};
diff --git a/winsup/cygwin/dcrt0.cc b/winsup/cygwin/dcrt0.cc
index 71b3d2ac3..e28aecde9 100644
--- a/winsup/cygwin/dcrt0.cc
+++ b/winsup/cygwin/dcrt0.cc
@@ -842,6 +842,8 @@ dll_crt0_1 (void *)
   strace.microseconds ();
 #endif
 
+  create_signal_arrived (); /* FIXME: move into wait_sig? */
+
   /* Initialize debug muto, if DLL is built with --enable-debugging.
      Need to do this before any helper threads start. */
   debug_init ();
diff --git a/winsup/cygwin/exceptions.cc b/winsup/cygwin/exceptions.cc
index fe233cf49..3c22e35c3 100644
--- a/winsup/cygwin/exceptions.cc
+++ b/winsup/cygwin/exceptions.cc
@@ -715,7 +715,7 @@ handle_sigsuspend (sigset_t tempmask)
   sigproc_printf ("oldmask %p, newmask %p", oldmask, tempmask);
 
   pthread_testcancel ();
-  cancelable_wait (NULL, NULL, cw_cancel | cw_cancel_self | cw_sig_eintr);
+  cancelable_wait (signal_arrived, NULL, cw_cancel | cw_cancel_self);
 
   set_sig_errno (EINTR);	// Per POSIX
 
@@ -748,7 +748,8 @@ sig_handle_tty_stop (int sig)
   sigproc_printf ("process %d stopped by signal %d", myself->pid, sig);
   HANDLE w4[2];
   w4[0] = sigCONT;
-  switch (cancelable_wait (sigCONT, NULL, cw_sig_eintr))
+  w4[1] = signal_arrived;
+  switch (WaitForMultipleObjects (2, w4, TRUE, INFINITE))
     {
     case WAIT_OBJECT_0:
     case WAIT_OBJECT_0 + 1:
@@ -1247,7 +1248,7 @@ sigpacket::process ()
 	{
 	  sigproc_printf ("default signal %d ignored", si.si_signo);
 	  if (continue_now)
-	    SetEvent (use_tls->signal_arrived);
+	    SetEvent (signal_arrived);
 	  goto done;
 	}
 
diff --git a/winsup/cygwin/fhandler_socket.cc b/winsup/cygwin/fhandler_socket.cc
index 611e70619..f78002801 100644
--- a/winsup/cygwin/fhandler_socket.cc
+++ b/winsup/cygwin/fhandler_socket.cc
@@ -130,7 +130,7 @@ get_inet_addr (const struct sockaddr *in, int inlen,
 	  pthread_testcancel ();
 	  /* Using IsEventSignalled like this is racy since another thread could
 	     be waiting for signal_arrived. */
-	  if (cancelable_wait (NULL, cw_nowait, cw_sig_eintr) == WAIT_SIGNALED
+	  if (IsEventSignalled (signal_arrived)
 	      && !_my_tls.call_signal_handler ())
 	    {
 	      set_errno (EINTR);
@@ -662,8 +662,7 @@ fhandler_socket::wait_for_events (const long event_mask, const DWORD flags)
 	  return SOCKET_ERROR;
 	}
 
-      WSAEVENT ev[2] = { wsock_evt };
-      set_thread_waiting (ev[1]);
+      WSAEVENT ev[2] = { wsock_evt, signal_arrived };
       switch (WSAWaitForMultipleEvents (2, ev, FALSE, 50, FALSE))
 	{
 	  case WSA_WAIT_TIMEOUT:
@@ -1785,7 +1784,7 @@ fhandler_socket::close ()
 	  res = -1;
 	  break;
 	}
-      if (cygwait (10) == WAIT_SIGNALED)
+      if (WaitForSingleObject (signal_arrived, 10) == WAIT_OBJECT_0)
 	{
 	  set_errno (EINTR);
 	  res = -1;
diff --git a/winsup/cygwin/fhandler_tape.cc b/winsup/cygwin/fhandler_tape.cc
index 2e936c3b1..c394d862d 100644
--- a/winsup/cygwin/fhandler_tape.cc
+++ b/winsup/cygwin/fhandler_tape.cc
@@ -1142,14 +1142,26 @@ mtinfo::initialize ()
 inline bool
 fhandler_dev_tape::_lock (bool cancelable)
 {
+  HANDLE w4[3] = { mt_mtx, signal_arrived, NULL };
+  DWORD cnt = 2;
+  if (cancelable && (w4[2] = pthread::get_cancel_event ()) != NULL)
+    cnt = 3;
   /* O_NONBLOCK is only valid in a read or write call.  Only those are
      cancelable. */
   DWORD timeout = cancelable && is_nonblocking () ? 0 : INFINITE;
 restart:
-  switch (cancelable_wait (mt_mtx, timeout, cw_sig | cw_cancel | cw_cancel_self))
+  switch (WaitForMultipleObjects (cnt, w4, FALSE, timeout))
     {
     case WAIT_OBJECT_0:
       return true;
+    case WAIT_OBJECT_0 + 1:
+      if (_my_tls.call_signal_handler ())
+	goto restart;
+      set_errno (EINTR);
+      return false;
+    case WAIT_OBJECT_0 + 2:
+      pthread::static_cancel_self ();
+      /*NOTREACHED*/
     case WAIT_TIMEOUT:
       set_errno (EAGAIN);
       return false;
diff --git a/winsup/cygwin/fhandler_termios.cc b/winsup/cygwin/fhandler_termios.cc
index 3bfa38a6b..7fddba5b3 100644
--- a/winsup/cygwin/fhandler_termios.cc
+++ b/winsup/cygwin/fhandler_termios.cc
@@ -204,7 +204,7 @@ fhandler_termios::bg_check (int sig)
     {
       /* Don't raise a SIGTT* signal if we have already been
 	 interrupted by another signal. */
-      if (cygwait (0) != WAIT_SIGNALED)
+      if (WaitForSingleObject (signal_arrived, 0) != WAIT_OBJECT_0)
 	{
 	  siginfo_t si = {0};
 	  si.si_signo = sig;
diff --git a/winsup/cygwin/fhandler_tty.cc b/winsup/cygwin/fhandler_tty.cc
index c5f1d4618..fe7e283a9 100644
--- a/winsup/cygwin/fhandler_tty.cc
+++ b/winsup/cygwin/fhandler_tty.cc
@@ -281,7 +281,7 @@ fhandler_pty_master::process_slave_output (char *buf, size_t len, int pktmode_on
 	      goto out;
 	    }
 	  pthread_testcancel ();
-	  if (cancelable_wait (NULL, 10, cw_sig_eintr) == WAIT_SIGNALED
+	  if (WaitForSingleObject (signal_arrived, 10) == WAIT_OBJECT_0
 	      && !_my_tls.call_signal_handler ())
 	    {
 	      set_errno (EINTR);
diff --git a/winsup/cygwin/fhandler_windows.cc b/winsup/cygwin/fhandler_windows.cc
index b592e3217..c89d98e16 100644
--- a/winsup/cygwin/fhandler_windows.cc
+++ b/winsup/cygwin/fhandler_windows.cc
@@ -96,8 +96,7 @@ fhandler_windows::read (void *buf, size_t& len)
       return;
     }
 
-  HANDLE w4[3] = { get_handle (), };
-  set_thread_waiting (w4[1]);
+  HANDLE w4[3] = { get_handle (), signal_arrived, NULL };
   DWORD cnt = 2;
   if ((w4[cnt] = pthread::get_cancel_event ()) != NULL)
     ++cnt;
diff --git a/winsup/cygwin/flock.cc b/winsup/cygwin/flock.cc
index f40c1e5c6..176e4869e 100644
--- a/winsup/cygwin/flock.cc
+++ b/winsup/cygwin/flock.cc
@@ -1247,7 +1247,7 @@ lf_setlock (lockf_t *lock, inode_t *node, lockf_t **clean, HANDLE fhdl)
 	timeout = 100L;
 
       DWORD WAIT_SIGNAL_ARRIVED = WAIT_OBJECT_0 + wait_count;
-      set_thread_waiting (w4[wait_count++]);
+      w4[wait_count++] = signal_arrived;
 
       DWORD WAIT_THREAD_CANCELED = WAIT_TIMEOUT + 1;
       HANDLE cancel_event = pthread::get_cancel_event ();
diff --git a/winsup/cygwin/posix_ipc.cc b/winsup/cygwin/posix_ipc.cc
index b427d11d6..b9d224e43 100644
--- a/winsup/cygwin/posix_ipc.cc
+++ b/winsup/cygwin/posix_ipc.cc
@@ -119,12 +119,14 @@ ipc_mutex_init (HANDLE *pmtx, const char *name)
 static int
 ipc_mutex_lock (HANDLE mtx)
 {
-  switch (cancelable_wait (mtx, NULL, cw_sig_eintr | cw_cancel | cw_cancel_self))
+  HANDLE h[2] = { mtx, signal_arrived };
+
+  switch (WaitForMultipleObjects (2, h, FALSE, INFINITE))
     {
     case WAIT_OBJECT_0:
     case WAIT_ABANDONED_0:
       return 0;
-    case WAIT_SIGNALED:
+    case WAIT_OBJECT_0 + 1:
       set_errno (EINTR);
       return 1;
     default:
@@ -172,12 +174,11 @@ ipc_cond_init (HANDLE *pevt, const char *name, char sr)
 static int
 ipc_cond_timedwait (HANDLE evt, HANDLE mtx, const struct timespec *abstime)
 {
-  HANDLE w4[4] = { evt, };
+  HANDLE w4[4] = { evt, signal_arrived, NULL, NULL };
   DWORD cnt = 2;
   DWORD timer_idx = 0;
   int ret = 0;
 
-  set_thread_waiting (w4[1]);
   if ((w4[cnt] = pthread::get_cancel_event ()) != NULL)
     ++cnt;
   if (abstime)
diff --git a/winsup/cygwin/select.cc b/winsup/cygwin/select.cc
index d98e13255..0186af733 100644
--- a/winsup/cygwin/select.cc
+++ b/winsup/cygwin/select.cc
@@ -312,7 +312,7 @@ select_stuff::wait (fd_set *readfds, fd_set *writefds, fd_set *exceptfds,
   select_record *s = &start;
   DWORD m = 0;
 
-  set_thread_waiting (w4[m++]);
+  w4[m++] = signal_arrived;  /* Always wait for the arrival of a signal. */
   if ((w4[m] = pthread::get_cancel_event ()) != NULL)
     m++;
 
diff --git a/winsup/cygwin/signal.cc b/winsup/cygwin/signal.cc
index 595bb01d7..a62c038cd 100644
--- a/winsup/cygwin/signal.cc
+++ b/winsup/cygwin/signal.cc
@@ -120,9 +120,12 @@ clock_nanosleep (clockid_t clk_id, int flags, const struct timespec *rqtp,
 
   syscall_printf ("clock_nanosleep (%ld.%09ld)", rqtp->tv_sec, rqtp->tv_nsec);
 
-  int rc = cancelable_wait (NULL, &timeout, cw_sig | cw_cancel | cw_cancel_self);
-  if (rc == WAIT_SIGNALED)
-    res = EINTR;
+  int rc = cancelable_wait (signal_arrived, &timeout);
+  if (rc == WAIT_OBJECT_0)
+    {
+      _my_tls.call_signal_handler ();
+      res = EINTR;
+    }
 
   /* according to POSIX, rmtp is used only if !abstime */
   if (rmtp && !abstime)
@@ -562,14 +565,21 @@ extern "C" int
 sigwaitinfo (const sigset_t *set, siginfo_t *info)
 {
   pthread_testcancel ();
+  HANDLE h;
+  h = _my_tls.event = CreateEvent (&sec_none_nih, FALSE, FALSE, NULL);
+  if (!h)
+    {
+      __seterrno ();
+      return -1;
+    }
 
   _my_tls.sigwait_mask = *set;
   sig_dispatch_pending (true);
 
   int res;
-  switch (cancelable_wait (NULL, NULL, cw_sig | cw_cancel | cw_cancel_self))
+  switch (WaitForSingleObject (h, INFINITE))
     {
-    case WAIT_SIGNALED:
+    case WAIT_OBJECT_0:
       if (!sigismember (set, _my_tls.infodata.si_signo))
 	{
 	  set_errno (EINTR);
@@ -588,6 +598,8 @@ sigwaitinfo (const sigset_t *set, siginfo_t *info)
       res = -1;
     }
 
+  _my_tls.event = NULL;
+  CloseHandle (h);
   sigproc_printf ("returning signal %d", res);
   return res;
 }
diff --git a/winsup/cygwin/sigproc.cc b/winsup/cygwin/sigproc.cc
index 56fce1663..05d98729f 100644
--- a/winsup/cygwin/sigproc.cc
+++ b/winsup/cygwin/sigproc.cc
@@ -43,6 +43,9 @@ int __sp_ln;
 
 char NO_COPY myself_nowait_dummy[1] = {'0'};// Flag to sig_send that signal goes to
 					//  current process but no wait is required
+HANDLE NO_COPY signal_arrived;		// Event signaled when a signal has
+					//  resulted in a user-specified
+					//  function call
 
 #define Static static NO_COPY
 
@@ -515,6 +518,17 @@ sig_dispatch_pending (bool fast)
     sig_send (myself, fast ? __SIGFLUSHFAST : __SIGFLUSH);
 }
 
+void __stdcall
+create_signal_arrived ()
+{
+  if (signal_arrived)
+    return;
+  /* local event signaled when main thread has been dispatched
+     to a signal handler function. */
+  signal_arrived = CreateEvent (&sec_none_nih, false, false, NULL);
+  ProtectHandle (signal_arrived);
+}
+
 /* Signal thread initialization.  Called from dll_crt0_1.
    This routine starts the signal handling thread.  */
 void __stdcall
diff --git a/winsup/cygwin/sigproc.h b/winsup/cygwin/sigproc.h
index ba8a231b9..d980f1a6e 100644
--- a/winsup/cygwin/sigproc.h
+++ b/winsup/cygwin/sigproc.h
@@ -58,6 +58,7 @@ struct sigpacket
   int __stdcall process () __attribute__ ((regparm (1)));
 };
 
+extern HANDLE signal_arrived;
 extern HANDLE sigCONT;
 
 void __stdcall sig_dispatch_pending (bool fast = false);
@@ -85,6 +86,7 @@ int __stdcall sig_send (_pinfo *, siginfo_t&, class _cygtls *tls = NULL) __attri
 int __stdcall sig_send (_pinfo *, int) __attribute__ ((regparm (2)));
 void __stdcall signal_fixup_after_exec ();
 void __stdcall sigalloc ();
+void __stdcall create_signal_arrived ();
 
 int kill_pgrp (pid_t, siginfo_t&);
 int killsys (pid_t, int);
diff --git a/winsup/cygwin/syscalls.cc b/winsup/cygwin/syscalls.cc
index c0cacac97..2077a0a3c 100644
--- a/winsup/cygwin/syscalls.cc
+++ b/winsup/cygwin/syscalls.cc
@@ -2295,7 +2295,7 @@ retry:
     {
       debug_printf ("status %p", status);
       if (status == STATUS_SHARING_VIOLATION
-	  && cygwait (10L) != WAIT_SIGNALED)
+	  && WaitForSingleObject (signal_arrived, 10L) != WAIT_OBJECT_0)
 	{
 	  /* Typical BLODA problem.  Some virus scanners check newly generated
 	     files and while doing that disallow DELETE access.  That's really
diff --git a/winsup/cygwin/thread.h b/winsup/cygwin/thread.h
index 0dd9b893b..62a06f60b 100644
--- a/winsup/cygwin/thread.h
+++ b/winsup/cygwin/thread.h
@@ -344,9 +344,6 @@ public:
   pthread_spinlock (int);
 };
 
-#define WAIT_CANCELED   (WAIT_OBJECT_0 + 1)
-#define WAIT_SIGNALED  (WAIT_OBJECT_0 + 2)
-
 class _cygtls;
 class pthread: public verifyable_object
 {