diff --git a/include/grpc/grpc.h b/include/grpc/grpc.h
index e95a5f2e92c..2a7ddbd4f37 100644
--- a/include/grpc/grpc.h
+++ b/include/grpc/grpc.h
@@ -361,7 +361,8 @@ grpc_completion_queue *grpc_completion_queue_create(void);
 /** Blocks until an event is available, the completion queue is being shut down,
     or deadline is reached. 
 
-    Returns NULL on timeout, otherwise the event that occurred.
+    Returns a grpc_event with type GRPC_QUEUE_TIMEOUT on timeout,
+    otherwise a grpc_event describing the event that occurred.
 
     Callers must not call grpc_completion_queue_next and
     grpc_completion_queue_pluck simultaneously on the same completion queue. */
@@ -371,7 +372,8 @@ grpc_event grpc_completion_queue_next(grpc_completion_queue *cq,
 /** Blocks until an event with tag 'tag' is available, the completion queue is
     being shutdown or deadline is reached. 
 
-    Returns NULL on timeout, or a pointer to the event that occurred.
+    Returns a grpc_event with type GRPC_QUEUE_TIMEOUT on timeout,
+    otherwise a grpc_event describing the event that occurred.
 
     Callers must not call grpc_completion_queue_next and
     grpc_completion_queue_pluck simultaneously on the same completion queue. */
diff --git a/src/core/transport/chttp2_transport.c b/src/core/transport/chttp2_transport.c
index 630504565b9..bd259f7ae3d 100644
--- a/src/core/transport/chttp2_transport.c
+++ b/src/core/transport/chttp2_transport.c
@@ -230,7 +230,10 @@ struct transport {
   /* basic state management - what are we doing at the moment? */
   gpr_uint8 reading;
   gpr_uint8 writing;
-  gpr_uint8 calling_back;
+  /** are we calling back (via cb) with a channel-level event */
+  gpr_uint8 calling_back_channel;
+  /** are we calling back any grpc_transport_op completion events */
+  gpr_uint8 calling_back_ops;
   gpr_uint8 destroying;
   gpr_uint8 closed;
   error_state error_state;
@@ -357,7 +360,7 @@ static void push_setting(transport *t, grpc_chttp2_setting_id id,
                          gpr_uint32 value);
 
 static int prepare_callbacks(transport *t);
-static void run_callbacks(transport *t, const grpc_transport_callbacks *cb);
+static void run_callbacks(transport *t);
 static void call_cb_closed(transport *t, const grpc_transport_callbacks *cb);
 
 static int prepare_write(transport *t);
@@ -565,7 +568,7 @@ static void init_transport(transport *t, grpc_transport_setup_callback setup,
   }
 
   gpr_mu_lock(&t->mu);
-  t->calling_back = 1;
+  t->calling_back_channel = 1;
   ref_transport(t); /* matches unref at end of this function */
   gpr_mu_unlock(&t->mu);
 
@@ -574,7 +577,7 @@ static void init_transport(transport *t, grpc_transport_setup_callback setup,
   lock(t);
   t->cb = sr.callbacks;
   t->cb_user_data = sr.user_data;
-  t->calling_back = 0;
+  t->calling_back_channel = 0;
   if (t->destroying) gpr_cv_signal(&t->cv);
   unlock(t);
 
@@ -595,7 +598,7 @@ static void destroy_transport(grpc_transport *gt) {
      We need to be not writing as cancellation finalization may produce some
      callbacks that NEED to be made to close out some streams when t->writing
      becomes 0. */
-  while (t->calling_back || t->writing) {
+  while (t->calling_back_channel || t->writing) {
     gpr_cv_wait(&t->cv, &t->mu, gpr_inf_future);
   }
   drop_connection(t);
@@ -830,28 +833,29 @@ static void unlock(transport *t) {
   finish_reads(t);
 
   /* gather any callbacks that need to be made */
-  if (!t->calling_back) {
-    t->calling_back = perform_callbacks = prepare_callbacks(t);
-    if (cb) {
-      if (t->error_state == ERROR_STATE_SEEN && !t->writing) {
-        call_closed = 1;
-        t->calling_back = 1;
-        t->cb = NULL; /* no more callbacks */
-        t->error_state = ERROR_STATE_NOTIFIED;
-      }
-      if (t->num_pending_goaways) {
-        goaways = t->pending_goaways;
-        num_goaways = t->num_pending_goaways;
-        t->pending_goaways = NULL;
-        t->num_pending_goaways = 0;
-        t->cap_pending_goaways = 0;
-        t->calling_back = 1;
-      }
-    }
+  if (!t->calling_back_ops) {
+    t->calling_back_ops = perform_callbacks = prepare_callbacks(t);
+    if (perform_callbacks) ref_transport(t);
   }
 
-  if (perform_callbacks || call_closed || num_goaways) {
-    ref_transport(t);
+  if (!t->calling_back_channel && cb) {
+    if (t->error_state == ERROR_STATE_SEEN && !t->writing) {
+      call_closed = 1;
+      t->calling_back_channel = 1;
+      t->cb = NULL; /* no more callbacks */
+      t->error_state = ERROR_STATE_NOTIFIED;
+    }
+    if (t->num_pending_goaways) {
+      goaways = t->pending_goaways;
+      num_goaways = t->num_pending_goaways;
+      t->pending_goaways = NULL;
+      t->num_pending_goaways = 0;
+      t->cap_pending_goaways = 0;
+      t->calling_back_channel = 1;
+    }
+    if (call_closed || num_goaways) {
+      ref_transport(t);
+    }
   }
 
   /* finally unlock */
@@ -865,7 +869,11 @@ static void unlock(transport *t) {
   }
 
   if (perform_callbacks) {
-    run_callbacks(t, cb);
+    run_callbacks(t);
+    lock(t);
+    t->calling_back_ops = 0;
+    unlock(t);
+    unref_transport(t);
   }
 
   if (call_closed) {
@@ -878,9 +886,9 @@ static void unlock(transport *t) {
     perform_write(t, ep);
   }
 
-  if (perform_callbacks || call_closed || num_goaways) {
+  if (call_closed || num_goaways) {
     lock(t);
-    t->calling_back = 0;
+    t->calling_back_channel = 0;
     if (t->destroying) gpr_cv_signal(&t->cv);
     unlock(t);
     unref_transport(t);
@@ -2101,7 +2109,7 @@ static int prepare_callbacks(transport *t) {
   return t->executing_callbacks.count > 0;
 }
 
-static void run_callbacks(transport *t, const grpc_transport_callbacks *cb) {
+static void run_callbacks(transport *t) {
   size_t i;
   for (i = 0; i < t->executing_callbacks.count; i++) {
     op_closure c = t->executing_callbacks.callbacks[i];
diff --git a/test/core/network_benchmarks/low_level_ping_pong.c b/test/core/network_benchmarks/low_level_ping_pong.c
index 7d74d0e0781..78a0eef1a2e 100644
--- a/test/core/network_benchmarks/low_level_ping_pong.c
+++ b/test/core/network_benchmarks/low_level_ping_pong.c
@@ -238,6 +238,7 @@ static int set_socket_nonblocking(thread_args *args) {
 
 static int do_nothing(thread_args *args) { return 0; }
 
+#ifdef __linux__
 /* Special case for epoll, where we need to create the fd ahead of time. */
 static int epoll_setup(thread_args *args) {
   int epoll_fd;
@@ -258,6 +259,7 @@ static int epoll_setup(thread_args *args) {
   }
   return 0;
 }
+#endif
 
 static void server_thread(thread_args *args) {
   char *buf = malloc(args->msg_size);
diff --git a/test/cpp/qps/qps_driver.cc b/test/cpp/qps/qps_driver.cc
index 281e2e81198..4ae720722fe 100644
--- a/test/cpp/qps/qps_driver.cc
+++ b/test/cpp/qps/qps_driver.cc
@@ -112,7 +112,7 @@ static void QpsDriver() {
       FLAGS_warmup_seconds, FLAGS_benchmark_seconds, FLAGS_local_workers);
 
   GetReporter()->ReportQPS(*result);
-  GetReporter()->ReportQPSPerCore(*result, server_config);
+  GetReporter()->ReportQPSPerCore(*result);
   GetReporter()->ReportLatency(*result);
   GetReporter()->ReportTimes(*result);
 }
diff --git a/test/cpp/qps/qps_test.cc b/test/cpp/qps/qps_test.cc
index 63a37ae07ef..07b4834cc00 100644
--- a/test/cpp/qps/qps_test.cc
+++ b/test/cpp/qps/qps_test.cc
@@ -67,7 +67,7 @@ static void RunQPS() {
   const auto result =
       RunScenario(client_config, 1, server_config, 1, WARMUP, BENCHMARK, -2);
 
-  GetReporter()->ReportQPSPerCore(*result, server_config);
+  GetReporter()->ReportQPSPerCore(*result);
   GetReporter()->ReportLatency(*result);
 }
 
diff --git a/test/cpp/qps/report.cc b/test/cpp/qps/report.cc
index b4229889add..080f2af2713 100644
--- a/test/cpp/qps/report.cc
+++ b/test/cpp/qps/report.cc
@@ -50,10 +50,9 @@ void CompositeReporter::ReportQPS(const ScenarioResult& result) const {
   }
 }
 
-void CompositeReporter::ReportQPSPerCore(const ScenarioResult& result,
-                                         const ServerConfig& config) const {
+void CompositeReporter::ReportQPSPerCore(const ScenarioResult& result) const {
   for (size_t i = 0; i < reporters_.size(); ++i) {
-    reporters_[i]->ReportQPSPerCore(result, config);
+    reporters_[i]->ReportQPSPerCore(result);
   }
 }
 
@@ -77,15 +76,14 @@ void GprLogReporter::ReportQPS(const ScenarioResult& result) const {
                       [](ResourceUsage u) { return u.wall_time; }));
 }
 
-void GprLogReporter::ReportQPSPerCore(const ScenarioResult& result,
-                                      const ServerConfig& server_config) const {
+void GprLogReporter::ReportQPSPerCore(const ScenarioResult& result)  const {
   auto qps =
       result.latencies.Count() /
       average(result.client_resources,
           [](ResourceUsage u) { return u.wall_time; });
 
   gpr_log(GPR_INFO, "QPS: %.1f (%.1f/server core)", qps,
-          qps / server_config.threads());
+          qps / result.server_config.threads());
 }
 
 void GprLogReporter::ReportLatency(const ScenarioResult& result) const {
diff --git a/test/cpp/qps/report.h b/test/cpp/qps/report.h
index 2e5b627e581..ca29fc9937a 100644
--- a/test/cpp/qps/report.h
+++ b/test/cpp/qps/report.h
@@ -62,8 +62,7 @@ class Reporter {
   virtual void ReportQPS(const ScenarioResult& result) const = 0;
 
   /** Reports QPS per core as (YYY/server core). */
-  virtual void ReportQPSPerCore(const ScenarioResult& result,
-                                const ServerConfig& config) const = 0;
+  virtual void ReportQPSPerCore(const ScenarioResult& result) const = 0;
 
   /** Reports latencies for the 50, 90, 95, 99 and 99.9 percentiles, in ms. */
   virtual void ReportLatency(const ScenarioResult& result) const = 0;
@@ -84,8 +83,7 @@ class CompositeReporter : public Reporter {
   void add(std::unique_ptr<Reporter> reporter);
 
   void ReportQPS(const ScenarioResult& result) const GRPC_OVERRIDE;
-  void ReportQPSPerCore(const ScenarioResult& result,
-                        const ServerConfig& config) const GRPC_OVERRIDE;
+  void ReportQPSPerCore(const ScenarioResult& result) const GRPC_OVERRIDE;
   void ReportLatency(const ScenarioResult& result) const GRPC_OVERRIDE;
   void ReportTimes(const ScenarioResult& result) const GRPC_OVERRIDE;
 
@@ -100,8 +98,7 @@ class GprLogReporter : public Reporter {
 
  private:
   void ReportQPS(const ScenarioResult& result) const GRPC_OVERRIDE;
-  void ReportQPSPerCore(const ScenarioResult& result,
-                        const ServerConfig& config) const GRPC_OVERRIDE;
+  void ReportQPSPerCore(const ScenarioResult& result) const GRPC_OVERRIDE;
   void ReportLatency(const ScenarioResult& result) const GRPC_OVERRIDE;
   void ReportTimes(const ScenarioResult& result) const GRPC_OVERRIDE;
 };
diff --git a/test/cpp/qps/server_async.cc b/test/cpp/qps/server_async.cc
index 977dfc2372d..4b0678bb2c0 100644
--- a/test/cpp/qps/server_async.cc
+++ b/test/cpp/qps/server_async.cc
@@ -101,10 +101,11 @@ class AsyncQpsServerTest : public Server {
           ServerRpcContext *ctx = detag(got_tag);
           // The tag is a pointer to an RPC context to invoke
           bool still_going = ctx->RunNextState(ok);
-          std::lock_guard<std::mutex> g(shutdown_mutex_);
+          std::unique_lock<std::mutex> g(shutdown_mutex_);
           if (!shutdown_) {
             // this RPC context is done, so refresh it
             if (!still_going) {
+              g.unlock();
               ctx->Reset();
             }
           } else {