From 229ea3f23c01543f98f4446bd9092dbf60c983cc Mon Sep 17 00:00:00 2001
From: Corinna Vinschen <corinna@vinschen.de>
Date: Tue, 22 Jan 2019 15:23:05 +0100
Subject: [PATCH] Cygwin: posix timers: reimplement using OS timer

- Rename files timer.* to posix_timer.*.
- Reimplement using an OS timer rather than a handcrafted wait loop.
- Use a Slim R/W Lock for synchronization.
- Drop timer chaining.  It doesn't server a purpose since all timers
  are local only.
- Rename ttstart to itimer_tracker to better reflect its purpose.
  It's not the anchor for a timer chain anymore anyway.
- Drop fixup_timers_after_fork.  Everything is process-local, nothing
  gets inherited.
- Rename timer_tracker::disarm_event to disarm_overrun_event for
  better readability.

Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
---
 winsup/cygwin/Makefile.in                  |   2 +-
 winsup/cygwin/cygheap_malloc.h             |   3 +-
 winsup/cygwin/exceptions.cc                |   8 +-
 winsup/cygwin/fork.cc                      |   2 -
 winsup/cygwin/{timer.cc => posix_timer.cc} | 494 +++++++++++----------
 winsup/cygwin/{timer.h => posix_timer.h}   |  33 +-
 winsup/cygwin/signal.cc                    |   4 +-
 7 files changed, 298 insertions(+), 248 deletions(-)
 rename winsup/cygwin/{timer.cc => posix_timer.cc} (50%)
 rename winsup/cygwin/{timer.h => posix_timer.h} (58%)

diff --git a/winsup/cygwin/Makefile.in b/winsup/cygwin/Makefile.in
index 374272f4c..32c02b87b 100644
--- a/winsup/cygwin/Makefile.in
+++ b/winsup/cygwin/Makefile.in
@@ -354,6 +354,7 @@ DLL_OFILES:= \
 	pinfo.o \
 	poll.o \
 	posix_ipc.o \
+	posix_timer.o \
 	pseudo-reloc.o \
 	pthread.o \
 	quotactl.o \
@@ -395,7 +396,6 @@ DLL_OFILES:= \
 	syslog.o \
 	termios.o \
 	thread.o \
-	timer.o \
 	timerfd.o \
 	times.o \
 	tls_pbuf.o \
diff --git a/winsup/cygwin/cygheap_malloc.h b/winsup/cygwin/cygheap_malloc.h
index 74f0bb619..cd545c35d 100644
--- a/winsup/cygwin/cygheap_malloc.h
+++ b/winsup/cygwin/cygheap_malloc.h
@@ -34,7 +34,8 @@ enum cygheap_types
   HEAP_2_DLL,
   HEAP_MMAP,
   HEAP_2_MAX = 200,
-  HEAP_3_FHANDLER
+  HEAP_3_FHANDLER,
+  HEAP_3_TIMER
 };
 
 extern "C" {
diff --git a/winsup/cygwin/exceptions.cc b/winsup/cygwin/exceptions.cc
index 205ad850e..491eedbe4 100644
--- a/winsup/cygwin/exceptions.cc
+++ b/winsup/cygwin/exceptions.cc
@@ -27,7 +27,7 @@ details. */
 #include "child_info.h"
 #include "ntdll.h"
 #include "exception.h"
-#include "timer.h"
+#include "posix_timer.h"
 
 /* Definitions for code simplification */
 #ifdef __x86_64__
@@ -1483,7 +1483,7 @@ sigpacket::process ()
   if (handler == SIG_IGN)
     {
       if (si.si_code == SI_TIMER)
-	((timer_tracker *) si.si_tid)->disarm_event ();
+	((timer_tracker *) si.si_tid)->disarm_overrun_event ();
       sigproc_printf ("signal %d ignored", si.si_signo);
       goto done;
     }
@@ -1508,7 +1508,7 @@ sigpacket::process ()
 	  || si.si_signo == SIGURG)
 	{
 	  if (si.si_code == SI_TIMER)
-	    ((timer_tracker *) si.si_tid)->disarm_event ();
+	    ((timer_tracker *) si.si_tid)->disarm_overrun_event ();
 	  sigproc_printf ("signal %d default is currently ignore", si.si_signo);
 	  goto done;
 	}
@@ -1637,7 +1637,7 @@ _cygtls::call_signal_handler ()
 	{
 	  timer_tracker *tt = (timer_tracker *)
 			      infodata.si_tid;
-	  infodata.si_overrun = tt->disarm_event ();
+	  infodata.si_overrun = tt->disarm_overrun_event ();
 	}
 
       /* Save information locally on stack to pass to handler. */
diff --git a/winsup/cygwin/fork.cc b/winsup/cygwin/fork.cc
index 5cfdd9559..6813446cf 100644
--- a/winsup/cygwin/fork.cc
+++ b/winsup/cygwin/fork.cc
@@ -22,7 +22,6 @@ details. */
 #include "tls_pbuf.h"
 #include "dll_init.h"
 #include "cygmalloc.h"
-#include "timer.h"
 #include "ntdll.h"
 
 #define NPIDS_HELD 4
@@ -196,7 +195,6 @@ frok::child (volatile char * volatile here)
   ForceCloseHandle1 (fork_info->forker_finished, forker_finished);
 
   pthread::atforkchild ();
-  fixup_timers_after_fork ();
   cygbench ("fork-child");
   ld_preload ();
   fixup_hooks_after_fork ();
diff --git a/winsup/cygwin/timer.cc b/winsup/cygwin/posix_timer.cc
similarity index 50%
rename from winsup/cygwin/timer.cc
rename to winsup/cygwin/posix_timer.cc
index e24edd5e9..d4f2575f1 100644
--- a/winsup/cygwin/timer.cc
+++ b/winsup/cygwin/posix_timer.cc
@@ -14,65 +14,37 @@ details. */
 #include "fhandler.h"
 #include "dtable.h"
 #include "cygheap.h"
-#include "timer.h"
+#include "posix_timer.h"
 #include <sys/param.h>
 
-#define EVENT_DISARMED	 0
-#define EVENT_ARMED	-1
-#define EVENT_LOCK	 1
+#define OVR_EVENT_DISARMED	0
+#define OVR_EVENT_ARMED		1
 
-/* Must not be NO_COPY to avoid memory leak after fork. */
-timer_tracker ttstart (CLOCK_REALTIME, NULL);
-
-class lock_timer_tracker
-{
-  static muto protect;
-public:
-  lock_timer_tracker ();
-  ~lock_timer_tracker ();
-};
-
-muto NO_COPY lock_timer_tracker::protect;
-
-lock_timer_tracker::lock_timer_tracker ()
-{
-  protect.init ("timer_protect")->acquire ();
-}
-
-lock_timer_tracker::~lock_timer_tracker ()
-{
-  protect.release ();
-}
+timer_tracker NO_COPY itimer_tracker (CLOCK_REALTIME, NULL);
 
 bool
 timer_tracker::cancel ()
 {
-  if (!hcancel)
-    return false;
+  DWORD res;
 
-  SetEvent (hcancel);
-  DWORD res = WaitForSingleObject (syncthread, INFINITE);
-  if (res != WAIT_OBJECT_0)
-    system_printf ("WFSO returned unexpected value %u, %E", res);
+  if (!cancel_evt)
+    return false;
+  SetEvent (cancel_evt);
+  if (sync_thr)
+    {
+      res = WaitForSingleObject (sync_thr, INFINITE);
+      if (res != WAIT_OBJECT_0)
+	system_printf ("WFSO returned unexpected value %u, %E", res);
+    }
   return true;
 }
 
-timer_tracker::~timer_tracker ()
-{
-  if (cancel ())
-    {
-      CloseHandle (hcancel);
-#ifdef DEBUGGING
-      hcancel = NULL;
-#endif
-    }
-  if (syncthread)
-    CloseHandle (syncthread);
-  magic = 0;
-}
-
 timer_tracker::timer_tracker (clockid_t c, const sigevent *e)
+: magic (TT_MAGIC), clock_id (c), timer (NULL), cancel_evt (NULL),
+  sync_thr (NULL), interval (0), exp_ts (0), overrun_event_running (0),
+  overrun_count_curr (0), overrun_count (0)
 {
+  srwlock = SRWLOCK_INIT;
   if (e != NULL)
     evp = *e;
   else
@@ -81,18 +53,17 @@ timer_tracker::timer_tracker (clockid_t c, const sigevent *e)
       evp.sigev_signo = SIGALRM;
       evp.sigev_value.sival_ptr = this;
     }
-  clock_id = c;
-  magic = TT_MAGIC;
-  hcancel = NULL;
-  event_running = EVENT_DISARMED;
-  overrun_count_curr = 0;
-  overrun_count = 0;
-  if (this != &ttstart)
-    {
-      lock_timer_tracker here;
-      next = ttstart.next;
-      ttstart.next = this;
-    }
+}
+
+timer_tracker::~timer_tracker ()
+{
+  AcquireSRWLockExclusive (&srwlock);
+  cancel ();
+  NtClose (cancel_evt);
+  NtClose (sync_thr);
+  NtClose (timer);
+  magic = 0;
+  ReleaseSRWLockExclusive (&srwlock);
 }
 
 static inline int64_t
@@ -105,29 +76,26 @@ timespec_to_us (const timespec& ts)
 }
 
 /* Returns 0 if arming successful, -1 if a signal is already queued.
-   If so, it also increments overrun_count. */
+   If so, it also increments overrun_count.  Only call under lock! */
 LONG
-timer_tracker::arm_event ()
+timer_tracker::arm_overrun_event ()
 {
   LONG ret;
 
-  while ((ret = InterlockedCompareExchange (&event_running, EVENT_ARMED,
-					    EVENT_DISARMED)) == EVENT_LOCK)
-    yield ();
-  if (ret == EVENT_ARMED)
-    InterlockedIncrement64 (&overrun_count);
+  ret = InterlockedExchange (&overrun_event_running, OVR_EVENT_ARMED);
+  if (ret == OVR_EVENT_ARMED)
+    ret = InterlockedIncrement64 (&overrun_count);
   return ret;
 }
 
 LONG
-timer_tracker::disarm_event ()
+timer_tracker::disarm_overrun_event ()
 {
   LONG ret;
 
-  while ((ret = InterlockedCompareExchange (&event_running, EVENT_LOCK,
-					    EVENT_ARMED)) == EVENT_LOCK)
-    yield ();
-  if (ret == EVENT_ARMED)
+  AcquireSRWLockExclusive (&srwlock);
+  ret = InterlockedExchange (&overrun_event_running, OVR_EVENT_DISARMED);
+  if (ret == OVR_EVENT_ARMED)
     {
       LONG64 ov_cnt;
 
@@ -138,8 +106,8 @@ timer_tracker::disarm_event ()
 	overrun_count_curr = ov_cnt;
       ret = overrun_count_curr;
       InterlockedExchange64 (&overrun_count, 0);
-      InterlockedExchange (&event_running, EVENT_DISARMED);
     }
+  ReleaseSRWLockExclusive (&srwlock);
   return ret;
 }
 
@@ -151,57 +119,69 @@ notify_thread_wrapper (void *arg)
   void * (*notify_func) (void *) = (void * (*) (void *))
 				   evt->sigev_notify_function;
 
-  tt->disarm_event ();
+  tt->disarm_overrun_event ();
   return notify_func (evt->sigev_value.sival_ptr);
 }
 
 DWORD
 timer_tracker::thread_func ()
 {
-  int64_t now;
-  int64_t cur_sleepto_us = sleepto_us;
+  HANDLE w4[2] = { timer, cancel_evt };
+
+  debug_printf ("%p timer armed", this);
   while (1)
     {
-      int64_t sleep_us;
-      LONG sleep_ms;
-      /* Account for delays in starting thread
-	and sending the signal */
-      now = get_clock (clock_id)->usecs ();
-      sleep_us = cur_sleepto_us - now;
-      if (sleep_us > 0)
+      switch (WaitForMultipleObjects (2, w4, FALSE, INFINITE))
 	{
-	  sleepto_us = cur_sleepto_us;
-	  sleep_ms = (sleep_us + (USPERSEC / MSPERSEC) - 1)
-		     / (USPERSEC / MSPERSEC);
-	}
-      else
-	{
-	  int64_t num_intervals = (now - cur_sleepto_us) / interval_us;
-	  InterlockedAdd64 (&overrun_count, num_intervals);
-	  cur_sleepto_us += num_intervals * interval_us;
-	  sleepto_us = cur_sleepto_us;
-	  sleep_ms = 0;
-	}
-
-      debug_printf ("%p waiting for %u ms", this, sleep_ms);
-      switch (WaitForSingleObject (hcancel, sleep_ms))
-	{
-	case WAIT_TIMEOUT:
-	  debug_printf ("timed out");
-	  break;
 	case WAIT_OBJECT_0:
-	  debug_printf ("%p cancelled", this);
+	  debug_printf ("%p timer expired", this);
+	  break;
+	case WAIT_OBJECT_0 + 1:
+	  debug_printf ("%p timer disarmed, %E", this);
 	  goto out;
 	default:
 	  debug_printf ("%p wait failed, %E", this);
+	  continue;
+	}
+      AcquireSRWLockExclusive (&srwlock);
+      /* Make sure we haven't been abandoned and/or disarmed in the meantime */
+      if (exp_ts == 0 && interval == 0)
+	{
+	  ReleaseSRWLockExclusive (&srwlock);
 	  goto out;
 	}
+      if (interval)
+	{
+	  /* Compute expiration count. */
+	  LONG64 now = get_clock_now ();
+	  LONG64 ts = get_exp_ts ();
+	  LONG64 exp_cnt;
 
+	  /* Make concessions for unexact realtime clock */
+	  if (ts > now)
+	    ts = now - 1;
+	  exp_cnt = (now - ts) / interval;
+	  InterlockedAdd64 (&overrun_count, exp_cnt);
+	  ts += interval * exp_cnt;
+	  set_exp_ts (ts);
+	  /* NtSetTimer allows periods of up to 24 days only.  If the time
+	     is longer, we set the timer up as one-shot timer for each
+	     interval.  Restart timer here with new due time. */
+	  if (interval > INT_MAX * (NS100PERSEC / MSPERSEC))
+	    {
+	      /* TODO: CLOCK_REALTIME_ALARM / CLOCK_BOOTTIME_ALARM
+		 See comment in arm_timer */
+	      BOOL Resume = FALSE;
+	      LARGE_INTEGER DueTime = { QuadPart: -interval };
+
+	      NtSetTimer (timer, &DueTime, NULL, NULL, Resume, 0, NULL);
+	    }
+	}
       switch (evp.sigev_notify)
 	{
 	case SIGEV_SIGNAL:
 	  {
-	    if (arm_event ())
+	    if (arm_overrun_event ())
 	      {
 		debug_printf ("%p timer signal already queued", this);
 		break;
@@ -217,7 +197,7 @@ timer_tracker::thread_func ()
 	  }
 	case SIGEV_THREAD:
 	  {
-	    if (arm_event ())
+	    if (arm_overrun_event ())
 	      {
 		debug_printf ("%p timer thread already queued", this);
 		break;
@@ -243,10 +223,16 @@ timer_tracker::thread_func ()
 	    break;
 	  }
 	}
-      if (!interval_us)
-	break;
-
-      cur_sleepto_us = sleepto_us + interval_us;
+      /* one-shot timer? */
+      if (!interval)
+	{
+	  memset (&time_spec, 0, sizeof time_spec);
+	  exp_ts = 0;
+	  overrun_event_running = OVR_EVENT_DISARMED;
+	  ReleaseSRWLockExclusive (&srwlock);
+	  goto out;
+	}
+      ReleaseSRWLockExclusive (&srwlock);
       debug_printf ("looping");
     }
 
@@ -262,58 +248,166 @@ timer_thread (VOID *x)
   return tt->thread_func ();
 }
 
-static inline bool
-timespec_bad (const timespec& t)
+int
+timer_tracker::gettime (itimerspec *curr_value, bool lock)
 {
-  if (t.tv_nsec < 0 || t.tv_nsec >= NSPERSEC || t.tv_sec < 0)
+  if (lock)
     {
-      set_errno (EINVAL);
-      return true;
+      AcquireSRWLockExclusive (&srwlock);
+      if (!is_timer_tracker ())
+	{
+	  ReleaseSRWLockExclusive (&srwlock);
+	  return -EINVAL;
+	}
     }
-  return false;
+  if (!cancel_evt)
+    memset (curr_value, 0, sizeof (*curr_value));
+  else
+    {
+      LONG64 next_relative_exp = get_exp_ts () - get_clock_now ();
+      curr_value->it_value.tv_sec = next_relative_exp / NS100PERSEC;
+      next_relative_exp -= curr_value->it_value.tv_sec * NS100PERSEC;
+      curr_value->it_value.tv_nsec = next_relative_exp
+				     * (NSPERSEC / NS100PERSEC);
+      curr_value->it_interval = time_spec.it_interval;
+    }
+  if (lock)
+    ReleaseSRWLockExclusive (&srwlock);
+  return 0;
 }
 
 int
-timer_tracker::settime (int in_flags, const itimerspec *value, itimerspec *ovalue)
+timer_tracker::settime (int flags, const itimerspec *new_value,
+			itimerspec *old_value)
 {
   int ret = -1;
 
   __try
     {
-      if (!value)
+      if (!new_value || !valid_timespec (new_value->it_value)
+	  || !valid_timespec (new_value->it_interval))
 	{
-	  set_errno (EINVAL);
+	  ret = -EINVAL;
 	  __leave;
 	}
 
-      if (timespec_bad (value->it_value) || timespec_bad (value->it_interval))
-	__leave;
+      AcquireSRWLockExclusive (&srwlock);
+      if (!is_timer_tracker ())
+	{
+	  ReleaseSRWLockExclusive (&srwlock);
+	  ret = -EINVAL;
+	  __leave;
+	}
 
-      lock_timer_tracker here;
-      cancel ();
+      if (old_value)
+	gettime (old_value, false);
 
-      if (ovalue)
-	gettime (ovalue);
-
-      if (!value->it_value.tv_sec && !value->it_value.tv_nsec)
-	interval_us = sleepto_us = 0;
+      if (!new_value->it_value.tv_sec && !new_value->it_value.tv_nsec)
+	{
+	  cancel ();
+	  memset (&time_spec, 0, sizeof time_spec);
+	  interval = 0;
+	  exp_ts = 0;
+	}
       else
 	{
-	  interval_us = timespec_to_us (value->it_interval);
-	  sleepto_us = timespec_to_us (value->it_value);
-	  if (!(in_flags & TIMER_ABSTIME))
-	    sleepto_us += get_clock (clock_id)->usecs ();
-	  it_interval = value->it_interval;
-	  if (!hcancel)
-	    hcancel = CreateEvent (&sec_none_nih, TRUE, FALSE, NULL);
+	  LONG64 ts;
+	  LARGE_INTEGER DueTime;
+	  BOOLEAN Resume;
+	  LONG Period;
+	  NTSTATUS status;
+
+	  if (!timer)
+	    {
+	      OBJECT_ATTRIBUTES attr;
+
+	      InitializeObjectAttributes (&attr, NULL, 0, NULL,
+					  sec_none_nih.lpSecurityDescriptor);
+	      status = NtCreateEvent (&cancel_evt, EVENT_ALL_ACCESS, &attr,
+				      NotificationEvent, FALSE);
+	      if (!NT_SUCCESS (status))
+		{
+		  ret = -geterrno_from_nt_status (status);
+		  __leave;
+		}
+	      status = NtCreateEvent (&sync_thr, EVENT_ALL_ACCESS, &attr,
+				      NotificationEvent, FALSE);
+	      if (!NT_SUCCESS (status))
+		{
+		  NtClose (cancel_evt);
+		  cancel_evt = NULL;
+		  ret = -geterrno_from_nt_status (status);
+		  __leave;
+		}
+	      status = NtCreateTimer (&timer, TIMER_ALL_ACCESS, &attr,
+				      SynchronizationTimer);
+	      if (!NT_SUCCESS (status))
+		{
+		  NtClose (cancel_evt);
+		  NtClose (sync_thr);
+		  cancel_evt = sync_thr = NULL;
+		  ret = -geterrno_from_nt_status (status);
+		  __leave;
+		}
+	    }
+	  ResetEvent (cancel_evt);
+	  ResetEvent (sync_thr);
+	  NtCancelTimer (timer, NULL);
+	  /* Convert incoming itimerspec into 100ns interval and timestamp */
+	  interval = new_value->it_interval.tv_sec * NS100PERSEC
+		      + (new_value->it_interval.tv_nsec
+			 + (NSPERSEC / NS100PERSEC) - 1)
+			/ (NSPERSEC / NS100PERSEC);
+	  ts = new_value->it_value.tv_sec * NS100PERSEC
+	       + (new_value->it_value.tv_nsec + (NSPERSEC / NS100PERSEC) - 1)
+		 / (NSPERSEC / NS100PERSEC);
+	  if (flags & TIMER_ABSTIME)
+	    {
+	      if (clock_id == CLOCK_REALTIME)
+		DueTime.QuadPart = ts + FACTOR;
+	      else /* non-REALTIME clocks require relative DueTime. */
+		{
+		  DueTime.QuadPart = get_clock_now () - ts;
+		  /* If the timestamp was earlier than now, compute number
+		     of expirations and offset DueTime to expire immediately. */
+		  if (DueTime.QuadPart >= 0)
+		    DueTime.QuadPart = -1LL;
+		}
+	    }
 	  else
-	    ResetEvent (hcancel);
-	  if (!syncthread)
-	    syncthread = CreateEvent (&sec_none_nih, TRUE, FALSE, NULL);
+	    {
+	      /* Keep relative timestamps relative for the timer, but store the
+		 expiry timestamp absolute for the timer thread. */
+	      DueTime.QuadPart = -ts;
+	      ts += get_clock_now ();
+	    }
+	  set_exp_ts (ts);
+	  time_spec = *new_value;
+	  overrun_count_curr = 0;
+	  overrun_count = 0;
+	  overrun_event_running = OVR_EVENT_DISARMED;
+	  /* TODO: CLOCK_REALTIME_ALARM / CLOCK_BOOTTIME_ALARM
+	     Note: Advanced Power Settings -> Sleep -> Allow Wake Timers
+	     since W10 1709 */
+	  Resume = FALSE;
+	  if (interval > INT_MAX * (NS100PERSEC / MSPERSEC))
+	    Period = 0;
 	  else
-	    ResetEvent (syncthread);
-	  new cygthread (timer_thread, this, "itimer", syncthread);
+	    Period = (interval + (NS100PERSEC / MSPERSEC) - 1)
+		     / (NS100PERSEC / MSPERSEC);
+	  status = NtSetTimer (timer, &DueTime, NULL, NULL, Resume, Period,
+			       NULL);
+	  if (!NT_SUCCESS (status))
+	    {
+	      memset (&time_spec, 0, sizeof time_spec);
+	      interval = 0;
+	      exp_ts = 0;
+	      ret = -geterrno_from_nt_status (status);
+	      __leave;
+	    }
+	  new cygthread (timer_thread, this, "itimer", sync_thr);
 	}
+      ReleaseSRWLockExclusive (&srwlock);
       ret = 0;
     }
   __except (EFAULT) {}
@@ -321,78 +415,12 @@ timer_tracker::settime (int in_flags, const itimerspec *value, itimerspec *ovalu
   return ret;
 }
 
-void
-timer_tracker::gettime (itimerspec *ovalue)
-{
-  if (!hcancel)
-    memset (ovalue, 0, sizeof (*ovalue));
-  else
-    {
-      ovalue->it_interval = it_interval;
-      int64_t now = get_clock (clock_id)->usecs ();
-      int64_t left_us = sleepto_us - now;
-      if (left_us < 0)
-       left_us = 0;
-      ovalue->it_value.tv_sec = left_us / USPERSEC;
-      ovalue->it_value.tv_nsec = (left_us % USPERSEC) * (NSPERSEC/USPERSEC);
-    }
-}
-
-int
-timer_tracker::clean_and_unhook ()
-{
-  for (timer_tracker *tt = &ttstart; tt->next != NULL; tt = tt->next)
-    if (tt->next == this)
-      {
-	tt->next = this->next;
-	return 0;
-      }
-  return -1;
-}
-
-void
-timer_tracker::fixup_after_fork ()
-{
-  ttstart.hcancel = ttstart.syncthread = NULL;
-  ttstart.event_running = EVENT_DISARMED;
-  ttstart.overrun_count_curr = 0;
-  ttstart.overrun_count = 0;
-  for (timer_tracker *tt = &ttstart; tt->next != NULL; /* nothing */)
-    {
-      timer_tracker *deleteme = tt->next;
-      tt->next = deleteme->next;
-      deleteme->hcancel = deleteme->syncthread = NULL;
-      delete deleteme;
-    }
-}
-
-void
-fixup_timers_after_fork ()
-{
-  timer_tracker::fixup_after_fork ();
-}
-
-extern "C" int
-timer_gettime (timer_t timerid, struct itimerspec *ovalue)
-{
-  int ret = -1;
-
-  __try
-    {
-      timer_tracker *tt = (timer_tracker *) timerid;
-      if (!tt->is_timer_tracker ())
-	{
-	  set_errno (EINVAL);
-	  return -1;
-	}
-
-      tt->gettime (ovalue);
-      ret = 0;
-    }
-  __except (EFAULT) {}
-  __endtry
-  return ret;
-}
+/* The timers are stored on the cygheap. */
+#define cnew(name, ...) \
+  ({ \
+    void* ptr = (void*) ccalloc (HEAP_3_TIMER, 1, sizeof (name)); \
+    ptr ? new (ptr) name (__VA_ARGS__) : NULL; \
+  })
 
 extern "C" int
 timer_create (clockid_t clock_id, struct sigevent *__restrict evp,
@@ -414,7 +442,7 @@ timer_create (clockid_t clock_id, struct sigevent *__restrict evp,
 	  return -1;
 	}
 
-      *timerid = (timer_t) new timer_tracker (clock_id, evp);
+      *timerid = (timer_t) cnew (timer_tracker, clock_id, evp);
       ret = 0;
     }
   __except (EFAULT) {}
@@ -422,6 +450,29 @@ timer_create (clockid_t clock_id, struct sigevent *__restrict evp,
   return ret;
 }
 
+extern "C" int
+timer_gettime (timer_t timerid, struct itimerspec *ovalue)
+{
+  int ret = -1;
+
+  __try
+    {
+      timer_tracker *tt = (timer_tracker *) timerid;
+      if (!tt->is_timer_tracker ())
+	{
+	  set_errno (EINVAL);
+	  return -1;
+	}
+
+      ret = tt->gettime (ovalue, true);
+      if (ret < 0)
+	set_errno (-ret);
+    }
+  __except (EFAULT) {}
+  __endtry
+  return ret;
+}
+
 extern "C" int
 timer_settime (timer_t timerid, int flags,
 	       const struct itimerspec *__restrict value,
@@ -438,6 +489,8 @@ timer_settime (timer_t timerid, int flags,
 	  __leave;
 	}
       ret = tt->settime (flags, value, ovalue);
+      if (ret < 0)
+	set_errno (-ret);
     }
   __except (EFAULT) {}
   __endtry
@@ -476,20 +529,12 @@ timer_delete (timer_t timerid)
   __try
     {
       timer_tracker *in_tt = (timer_tracker *) timerid;
-      if (!in_tt->is_timer_tracker ())
+      if (!in_tt->is_timer_tracker () || in_tt == &itimer_tracker)
 	{
 	  set_errno (EINVAL);
 	  __leave;
 	}
-
-      lock_timer_tracker here;
-      if (in_tt->clean_and_unhook () == 0)
-	{
-	  delete in_tt;
-	  ret = 0;
-	}
-      else
-	set_errno (EINVAL);
+      delete in_tt;
     }
   __except (EFAULT) {}
   __endtry
@@ -513,7 +558,8 @@ setitimer (int which, const struct itimerval *__restrict value,
       spec_value.it_interval.tv_nsec = value->it_interval.tv_usec * 1000;
       spec_value.it_value.tv_sec = value->it_value.tv_sec;
       spec_value.it_value.tv_nsec = value->it_value.tv_usec * 1000;
-      ret = timer_settime ((timer_t) &ttstart, 0, &spec_value, &spec_ovalue);
+      ret = timer_settime ((timer_t) &itimer_tracker, 0,
+			   &spec_value, &spec_ovalue);
       if (ret)
 	ret = -1;
       else if (ovalue)
@@ -541,7 +587,7 @@ getitimer (int which, struct itimerval *ovalue)
       __try
 	{
 	  struct itimerspec spec_ovalue;
-	  ret = timer_gettime ((timer_t) &ttstart, &spec_ovalue);
+	  ret = timer_gettime ((timer_t) &itimer_tracker, &spec_ovalue);
 	  if (!ret)
 	    {
 	      ovalue->it_interval.tv_sec = spec_ovalue.it_interval.tv_sec;
@@ -567,7 +613,7 @@ alarm (unsigned int seconds)
  if (seconds > (CLOCK_DELAY_MAX / 1000 - 1))
    seconds = (CLOCK_DELAY_MAX / 1000 - 1);
  newt.it_value.tv_sec = seconds;
- timer_settime ((timer_t) &ttstart, 0, &newt, &oldt);
+ timer_settime ((timer_t) &itimer_tracker, 0, &newt, &oldt);
  int ret = oldt.it_value.tv_sec + (oldt.it_value.tv_nsec > 0);
  syscall_printf ("%d = alarm(%u)", ret, seconds);
  return ret;
@@ -589,7 +635,7 @@ ualarm (useconds_t value, useconds_t interval)
      timer.it_interval.tv_sec = interval / USPERSEC;
      timer.it_interval.tv_nsec = (interval % USPERSEC) * (NSPERSEC/USPERSEC);
    }
- timer_settime ((timer_t) &ttstart, 0, &timer, &otimer);
+ timer_settime ((timer_t) &itimer_tracker, 0, &timer, &otimer);
  useconds_t ret = otimer.it_value.tv_sec * USPERSEC
 		  + (otimer.it_value.tv_nsec + (NSPERSEC/USPERSEC) - 1)
 		    / (NSPERSEC/USPERSEC);
diff --git a/winsup/cygwin/timer.h b/winsup/cygwin/posix_timer.h
similarity index 58%
rename from winsup/cygwin/timer.h
rename to winsup/cygwin/posix_timer.h
index dd5b165c7..04a383fd6 100644
--- a/winsup/cygwin/timer.h
+++ b/winsup/cygwin/posix_timer.h
@@ -13,38 +13,43 @@ details. */
 class timer_tracker
 {
   unsigned magic;
-  timer_tracker *next;
-
+  SRWLOCK srwlock;
   clockid_t clock_id;
   sigevent evp;
-  timespec it_interval;
-  HANDLE hcancel;
-  HANDLE syncthread;
-  int64_t interval_us;
-  int64_t sleepto_us;
-  LONG event_running;
+  struct itimerspec time_spec;
+  HANDLE timer;
+  HANDLE cancel_evt;
+  HANDLE sync_thr;
+  LONG64 interval;
+  LONG64 exp_ts;
+  LONG overrun_event_running;
   LONG overrun_count_curr;
   LONG64 overrun_count;
 
   bool cancel ();
 
  public:
+  void *operator new (size_t, void *p) __attribute__ ((nothrow)) {return p;}
+  void operator delete (void *p) { cfree (p); }
   timer_tracker (clockid_t, const sigevent *);
   ~timer_tracker ();
   inline bool is_timer_tracker () const { return magic == TT_MAGIC; }
   inline sigevent_t *sigevt () { return &evp; }
   inline int getoverrun () const { return overrun_count_curr; }
 
-  void gettime (itimerspec *);
+  LONG64 get_clock_now () const { return get_clock (clock_id)->n100secs (); }
+  LONG64 get_exp_ts () const { return exp_ts; }
+  LONG64 get_interval () const { return interval; }
+  void set_exp_ts (LONG64 ts) { exp_ts = ts; }
+
+  LONG arm_overrun_event ();
+  LONG disarm_overrun_event ();
+
+  int gettime (itimerspec *, bool);
   int settime (int, const itimerspec *, itimerspec *);
-  int clean_and_unhook ();
-  LONG arm_event ();
-  LONG disarm_event ();
 
   DWORD thread_func ();
   static void fixup_after_fork ();
 };
 
-extern void fixup_timers_after_fork ();
-
 #endif /* __TIMER_H__ */
diff --git a/winsup/cygwin/signal.cc b/winsup/cygwin/signal.cc
index fdde26058..abefedd7b 100644
--- a/winsup/cygwin/signal.cc
+++ b/winsup/cygwin/signal.cc
@@ -21,7 +21,7 @@ details. */
 #include "dtable.h"
 #include "cygheap.h"
 #include "cygwait.h"
-#include "timer.h"
+#include "posix_timer.h"
 
 #define _SA_NORESTART	0x8000
 
@@ -619,7 +619,7 @@ sigwait_common (const sigset_t *set, siginfo_t *info, PLARGE_INTEGER waittime)
 		{
 		  timer_tracker *tt = (timer_tracker *)
 				      _my_tls.infodata.si_tid;
-		  _my_tls.infodata.si_overrun = tt->disarm_event ();
+		  _my_tls.infodata.si_overrun = tt->disarm_overrun_event ();
 		}
 	      if (info)
 		*info = _my_tls.infodata;