RDKEMW-22339: BLE key thread spins due to fds not cleaned up properly

src/ble/ctrlm_ble_rcu_interface.cpp

@@ -120,6 +120,14 @@ void ctrlm_ble_rcu_interface_t::shutdown()
         m_controller->shutdown();
     }
 
+    // Tell ctrlm_ble_key_m to drop every managed device before destructing the controller/adapter.
+    // Without this, the key monitor thread is left polling an fd that bluez destroys.
+    if (m_controller) {
+        for (const BleAddress &address : m_controller->managedDevices()) {
+            removeDeviceKeyMonitorThread(address);
+        }
+    }
+
     // delete the adapter and controller when going into deepsleep so that we don't
     // get notified of the remote disconnection that occurs when going to sleep.
     // Upon waking, these will be re-initialized.
@@ -173,10 +181,7 @@ void ctrlm_ble_rcu_interface_t::initialize()
             {
                 XLOGD_DEBUG("deviceRemovedSlot %s", address.toString().c_str());
 
-                ctrlm_ble_key_queue_device_changed_msg_t msg;
-                msg.header.type = CTRLM_BLE_KEY_QUEUE_MSG_TYPE_DEVICE_REMOVED;
-                msg.address = address;
-                ctrlm_utils_queue_msg_push(m_keyThreadMsgQ, (const char *)&msg, sizeof(msg));
+                removeDeviceKeyMonitorThread(address);
 
                 ctrlm_hal_ble_IndUnPaired_params_t params;
                 params.ieee_address = address.toUInt64();
@@ -283,9 +288,11 @@ bool ctrlm_ble_rcu_interface_t::handleAddedDevice(const BleAddress &address)
         {
             XLOGD_INFO("BLE RCU %s connected changed to <%s>", address.toString().c_str(), connected ? "TRUE" : "FALSE");
 
-            if (connected) {
-                if (m_keyMonitorThread.running) {
+            if (m_keyMonitorThread.running) {
+                if (connected) {
                     addNewDeviceKeyMonitorThread(address);
+                } else {
+                    removeDeviceKeyMonitorThread(address);
                 }
             }
 
@@ -1448,6 +1455,13 @@ void ctrlm_ble_rcu_interface_t::addNewDeviceKeyMonitorThread(BleAddress address)
     msg.address = address;
     ctrlm_utils_queue_msg_push(m_keyThreadMsgQ, (const char *)&msg, sizeof(msg));
 }
+
+void ctrlm_ble_rcu_interface_t::removeDeviceKeyMonitorThread(BleAddress address){
+    ctrlm_ble_key_queue_device_changed_msg_t msg;
+    msg.header.type = CTRLM_BLE_KEY_QUEUE_MSG_TYPE_DEVICE_REMOVED;
+    msg.address = address;
+    ctrlm_utils_queue_msg_push(m_keyThreadMsgQ, (const char *)&msg, sizeof(msg));
+}
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 // BEGIN - Key Monitor Thread
 // -------------------------------------------------------------------------------------------------------------
@@ -1681,15 +1695,29 @@ void *KeyMonitorThread(void *data)
         }
 
         // loop the rcu fds to see if any has data to read
-        for (auto const &rcu : rcuKeypressFds) {
+        for (auto &rcu : rcuKeypressFds) {
             if (rcu.second >= 0) {
                 if (FD_ISSET(rcu.second, &rfds)) {
                     safec_rc = memset_s ((void*) &event, sizeof(event), 0, sizeof(event));
                     ERR_CHK(safec_rc);
                     ret = read(rcu.second, (void*)&event, sizeof(event));
-                    if (ret < 0) {
-                        // int errsv = errno;
-                        // XLOGD_ERROR("Error reading event: error = <%d>, <%s>", errsv, strerror(errsv));
+                    if (ret == 0) {
+                        // EOF - the underlying input device node is gone.
+                        XLOGD_WARN("Input device for RCU <%s> hit EOF, closing fd <%d>",
+                                rcu.first.toString().c_str(), rcu.second);
+                        close(rcu.second);
+                        rcu.second = -1;
+                    } else if (ret < 0) {
+                        int errsv = errno;
+                        if (errsv == ENODEV || errsv == EBADF || errsv == EINVAL) {
+                            // Device node was destroyed out from under us leaving 100% CPU spin.
+                            // Close fd and FindRcuInputDevices() reopens the device node if it comes back.
+                            XLOGD_WARN("Input device for RCU <%s> gone: error = <%d>, <%s>, closing fd <%d>",
+                                    rcu.first.toString().c_str(), errsv, strerror(errsv), rcu.second);
+                            close(rcu.second);
+                            rcu.second = -1;
+                        }
+                        // else: transient error (e.g. EAGAIN) - retry on the next wakeup.
                     } else {
                         HandleKeypress(metadata, &event, rcu.first);
                     }

src/ble/ctrlm_ble_rcu_interface.h

@@ -177,6 +177,7 @@ class ctrlm_ble_rcu_interface_t
     std::shared_ptr<BleRcuController> m_controller;
 
     void addNewDeviceKeyMonitorThread(BleAddress address);
+    void removeDeviceKeyMonitorThread(BleAddress address);
 };
 
 #endif //__CTRLM_BLE_RCU_INTERFACE_H__

src/ctrlm_main.cpp

@@ -986,14 +986,13 @@ gboolean ctrlm_thread_monitor(gpointer user_data) {
    }
    #endif
 
-   if(ctrlm_was_cpu_halted()) {
+   if(ctrlm_was_cpu_halted() && !g_ctrlm.thread_monitor_active) {
       XLOGD_INFO("skipping response check due to power state <%s>",ctrlm_power_state_str(g_ctrlm.power_state));
-      g_ctrlm.thread_monitor_active = false; // Deactivate thread monitoring
    } else if(!g_ctrlm.thread_monitor_active) {
       XLOGD_INFO("activate due to power state <%s>",ctrlm_power_state_str(g_ctrlm.power_state));
       g_ctrlm.thread_monitor_active = true;  // Activate thread monitoring again
    } else {
-      // Check the response from each thread on the previous attempt
+      // Check the response from each thread on the previous attempt, including final iteration before DEEP_SLEEP.
       for(vector<ctrlm_thread_monitor_t>::iterator it = g_ctrlm.monitor_threads.begin(); it != g_ctrlm.monitor_threads.end(); it++) {
          XLOGD_DEBUG("Checking %s", it->name);
 
@@ -1024,6 +1023,12 @@ gboolean ctrlm_thread_monitor(gpointer user_data) {
             return (FALSE);
          }
       }
+
+      if(ctrlm_was_cpu_halted()) {
+         // The DEEP_SLEEP transition has now been checked, go quiet until we wake.
+         XLOGD_INFO("power state <%s>: transition checked, deactivating thread monitor",ctrlm_power_state_str(g_ctrlm.power_state));
+         g_ctrlm.thread_monitor_active = false; // Deactivate thread monitoring
+      }
    }
 
    if(g_ctrlm.thread_monitor_active) { // Thread monitoring is active