diff --git a/changelog.d/5609.bugfix b/changelog.d/5609.bugfix
new file mode 100644
index 0000000000..534ee22a1b
--- /dev/null
+++ b/changelog.d/5609.bugfix
@@ -0,0 +1 @@
+Fix 'utime went backwards' errors on daemonization.
diff --git a/synapse/app/_base.py b/synapse/app/_base.py
index d50a9840d4..9b5c30f506 100644
--- a/synapse/app/_base.py
+++ b/synapse/app/_base.py
@@ -93,33 +93,36 @@ def start_reactor(
     install_dns_limiter(reactor)
 
     def run():
-        # make sure that we run the reactor with the sentinel log context,
-        # otherwise other PreserveLoggingContext instances will get confused
-        # and complain when they see the logcontext arbitrarily swapping
-        # between the sentinel and `run` logcontexts.
-        with PreserveLoggingContext():
-            logger.info("Running")
+        logger.info("Running")
+        change_resource_limit(soft_file_limit)
+        if gc_thresholds:
+            gc.set_threshold(*gc_thresholds)
+        reactor.run()
 
-            change_resource_limit(soft_file_limit)
-            if gc_thresholds:
-                gc.set_threshold(*gc_thresholds)
-            reactor.run()
+    # make sure that we run the reactor with the sentinel log context,
+    # otherwise other PreserveLoggingContext instances will get confused
+    # and complain when they see the logcontext arbitrarily swapping
+    # between the sentinel and `run` logcontexts.
+    #
+    # We also need to drop the logcontext before forking if we're daemonizing,
+    # otherwise the cputime metrics get confused about the per-thread resource usage
+    # appearing to go backwards.
+    with PreserveLoggingContext():
+        if daemonize:
+            if print_pidfile:
+                print(pid_file)
 
-    if daemonize:
-        if print_pidfile:
-            print(pid_file)
-
-        daemon = Daemonize(
-            app=appname,
-            pid=pid_file,
-            action=run,
-            auto_close_fds=False,
-            verbose=True,
-            logger=logger,
-        )
-        daemon.start()
-    else:
-        run()
+            daemon = Daemonize(
+                app=appname,
+                pid=pid_file,
+                action=run,
+                auto_close_fds=False,
+                verbose=True,
+                logger=logger,
+            )
+            daemon.start()
+        else:
+            run()
 
 
 def quit_with_error(error_string):
diff --git a/synapse/util/logcontext.py b/synapse/util/logcontext.py
index 9e1b537804..30dfa1d6b2 100644
--- a/synapse/util/logcontext.py
+++ b/synapse/util/logcontext.py
@@ -52,6 +52,15 @@ except Exception:
         return None
 
 
+# get an id for the current thread.
+#
+# threading.get_ident doesn't actually return an OS-level tid, and annoyingly,
+# on Linux it actually returns the same value either side of a fork() call. However
+# we only fork in one place, so it's not worth the hoop-jumping to get a real tid.
+#
+get_thread_id = threading.get_ident
+
+
 class ContextResourceUsage(object):
     """Object for tracking the resources used by a log context
 
@@ -225,7 +234,7 @@ class LoggingContext(object):
         # became active.
         self.usage_start = None
 
-        self.main_thread = threading.current_thread()
+        self.main_thread = get_thread_id()
         self.request = None
         self.tag = ""
         self.alive = True
@@ -318,7 +327,7 @@ class LoggingContext(object):
         record.request = self.request
 
     def start(self):
-        if threading.current_thread() is not self.main_thread:
+        if get_thread_id() != self.main_thread:
             logger.warning("Started logcontext %s on different thread", self)
             return
 
@@ -328,7 +337,7 @@ class LoggingContext(object):
             self.usage_start = get_thread_resource_usage()
 
     def stop(self):
-        if threading.current_thread() is not self.main_thread:
+        if get_thread_id() != self.main_thread:
             logger.warning("Stopped logcontext %s on different thread", self)
             return
 
@@ -355,7 +364,7 @@ class LoggingContext(object):
 
         # If we are on the correct thread and we're currently running then we
         # can include resource usage so far.
-        is_main_thread = threading.current_thread() is self.main_thread
+        is_main_thread = get_thread_id() == self.main_thread
         if self.alive and self.usage_start and is_main_thread:
             utime_delta, stime_delta = self._get_cputime()
             res.ru_utime += utime_delta