APPRO 8x GMSL camera issues

modules/v4l2/v4l2_device.cc

@@ -443,8 +443,20 @@ int V4L2DevicePoller::Poll(int timeout_ms, int events, std::vector<V4L2Device*>*
     }
     int ret = ::poll(poll_fds_.data(), poll_fds_.size(), timeout_ms);
     if (ret <= 0) {
-        for (size_t i = 0; i < devices_.size(); i++) {
-            LOGE("%s: Device node fd %d poll timeout.", __func__, devices_[i]->fd_);
+        if (ret < 0) {
+            LOGE("%s: poll error: %s", __func__, strerror(errno));
+        } else {
+            /* ret == 0: normal 1-second timeout — not an error, just no frames yet.
+             * Log at debug level to avoid flooding the error log every second. */
+            for (size_t i = 0; i < devices_.size(); i++) {
+                LOG2("%s: Device node fd %d poll timeout.", __func__, devices_[i]->fd_);
+            }
+            /* Defensive: if flush_fd fired in the same tick as timeout (race),
+             * treat it as a flush so the caller exits promptly. */
+            if (flush_fd_ != -1 && (poll_fds_.back().revents & (POLLIN | POLLPRI))) {
+                LOG1("%s: flush detected on timeout path, fd %d.", __func__, poll_fds_.back().fd);
+                return 1;
+            }
         }
         return ret;
     }

src/core/CameraDevice.cpp

@@ -850,7 +850,12 @@ int CameraDevice::stop() {
     LOG1("<id%d>@%s, mState:%d", mCameraId, __func__, mState);
     AutoMutex m(mDeviceLock);
 
-    mRequestThread->clearRequests();
+    /* Stop the RequestThread first (sets mState=EXIT + notifies mFrameAvailableSignal).
+     * This unblocks any icamerasrc streaming thread blocked inside camera_stream_dqbuf →
+     * waitFrame() immediately, instead of waiting for the 30-second dead-camera timeout.
+     * requestStop() includes clearRequests() so no separate call needed.
+     * The second requestStop() call in deinit() is safe — it is a no-op when already stopped. */
+    mRequestThread->requestStop();
 
     m3AControl->stop();
     mLensCtrl->stop();

src/core/CaptureUnit.cpp

@@ -419,12 +419,27 @@ int CaptureUnit::poll() {
         // Exiting, no error
         return -1;
     }
-    CheckAndLogError(ret < 0, UNKNOWN_ERROR, "%s: Poll error, ret:%d", __func__, ret);
+    if (ret < 0) {
+        LOGE("<id%d>%s: Unexpected poll error (POLLERR) while streaming — "
+             "camera may have been disrupted by another camera's STREAMOFF. "
+             "Poll thread will exit; other cameras should continue normally.",
+             mCameraId, __func__);
+        return ret;
+    }
     if (ret == 0) {
-        LOG1("<id%d>%s, timeout happens, wait recovery", mCameraId, __func__);
+        LOG2("<id%d>%s, timeout happens, wait recovery", mCameraId, __func__);
         return OK;
     }
 
+    /* ret > 0: one or more fds are ready.  If the flush pipe triggered this
+     * wake-up (stop() wrote to mFlushFd[1]) there will be no camera device in
+     * readyDevices, so the dequeue loop below is a no-op.  Check mExitPending
+     * first to avoid a spurious dequeue attempt after stop(). */
+    if (mExitPending) {
+        LOG2("%s: mExitPending true after poll wakeup, exit", __func__);
+        return -1;
+    }
+
     for (const auto& readyDevice : readyDevices) {
         for (auto device : mDevices) {
             if (device->getV4l2Device() == readyDevice) {

src/core/RequestThread.cpp

@@ -221,6 +221,8 @@ int RequestThread::waitFrame(int streamId, camera_buffer_t **ubuffer) {
     if (mState == EXIT) {
         return NO_INIT;
     }
+    static const int kMaxTimeouts = 2;  // ~10s before declaring camera dead and setting EXIT
+    int timeoutCount = 0;
     while (frameQueue.mFrameQueue.empty()) {
         std::cv_status ret = frameQueue.mFrameAvailableSignal.wait_for(
                       lock,
@@ -229,8 +231,28 @@ int RequestThread::waitFrame(int streamId, camera_buffer_t **ubuffer) {
             return NO_INIT;
         }
 
-        CheckWarning(ret == std::cv_status::timeout, TIMED_OUT,
-                     "<id%d>@%s, time out happens, wait recovery", mCameraId, __func__);
+        if (ret == std::cv_status::timeout) {
+            ++timeoutCount;
+            if (timeoutCount < kMaxTimeouts) {
+                LOGW("<id%d>@%s, no frame for ~%ds — waiting (attempt %d/%d)",
+                     mCameraId, __func__,
+                     (int)(kWaitFrameDuration * SLOWLY_MULTIPLIER / 1000000000LL) * timeoutCount,
+                     timeoutCount, kMaxTimeouts);
+            } else {
+                LOGE("<id%d>@%s, camera appears dead (no frame in ~%ds). "
+                     "Poll thread may have exited due to POLLERR from another camera's STREAMOFF. "
+                     "Marking EXIT so dqbuf() loop breaks and pipeline can shut down.",
+                     mCameraId, __func__,
+                     (int)(kWaitFrameDuration * SLOWLY_MULTIPLIER / 1000000000LL) * timeoutCount);
+                /* Set EXIT before releasing the lock so the next waitFrame() call (or a
+                 * concurrent one on another stream) returns NO_INIT immediately instead of
+                 * blocking for another kMaxTimeouts × kWaitFrameDuration seconds.
+                 * CameraDevice::dqbuf() loops while ret==TIMED_OUT — returning NO_INIT here
+                 * breaks that loop and lets icamerasrc post GST_FLOW_ERROR to start teardown. */
+                mState = EXIT;
+                return NO_INIT;
+            }
+        }
     }
 
     shared_ptr<CameraBuffer> camBuffer = frameQueue.mFrameQueue.front();

src/hal/CameraHal.cpp

@@ -280,6 +280,7 @@ int CameraHal::deviceStart(int cameraId) {
                              cameraId, mConfigTimes);
         }
     }
+    lock.unlock(); // allow concurent STREAMON for all VC Cameras
     // VIRTUAL_CHANNEL_E
 
     return device->start();

src/iutils/Thread.cpp

@@ -86,9 +86,16 @@ void Thread::wait() {
         return;
     }
 
+    const int64_t kWaitTimeout = 5000000000LL;  // 5 seconds
     while (mState != EXITED) {
         mState = EXITING;
-        mExitedCondition.wait(lock);
+        int waitRet = mExitedCondition.waitRelative(lock, kWaitTimeout);
+        if (waitRet == TIMED_OUT) {
+            LOGE("Thread::wait() timed out after 5s waiting for thread '%s' to exit — "
+                 "poll thread may be stuck in kernel; forcing exit",
+                 mName.empty() ? "NO_NAME" : mName.c_str());
+            break;
+        }
     }
 
     return;

src/platformdata/CameraSensorsParser.cpp

@@ -777,8 +777,8 @@ void CameraSensorsParser::parseSensorSection(const Json::Value& node) {
     if (node.isMember("vcId")) {
         mCurCam->mVCId = node["vcId"].asInt();
     }
-    if (node.isMember("vcGoupId")) {
-        mCurCam->mVCGroupId = node["vcGoupId"].asInt();
+    if (node.isMember("vcGroupId")) {
+        mCurCam->mVCGroupId = node["vcGroupId"].asInt();
     }
     // VIRTUAL_CHANNEL_E
     resolveCsiPortAndI2CBus();

src/v4l2/MediaControl.cpp

@@ -1022,10 +1022,12 @@ void MediaControl::mediaCtlClear(int cameraId, MediaCtlConf* mc) {
      * clearing routes here would disrupt other processes or camera instances that share the
      * same subdev and are still streaming.  Routes are left active on the hardware; the next
      * open() will skip SetRouting if the routes already match (see setRouting).
-     * Remove the receiver from the in-process tracking set so that a subsequent open() in
-     * this process re-evaluates the full setup (formats, links), even though SetRouting itself
-     * will be skipped when routes are still correctly configured.
+     * Reset mIsMediaCtlSetup so that a subsequent open() in this process re-evaluates the
+     * full setup (formats, links), even though SetRouting itself will be skipped when routes
+     * are still correctly configured.
      */
+    AutoMutex lock(sLock);
+    mIsMediaCtlSetup = false;
     // VIRTUAL_CHANNEL_E
 }
 
@@ -1125,8 +1127,9 @@ int MediaControl::getI2CBusAddress(const string& sensorEntityName, const string&
         char* entityName = nullptr;
         size_t sensorEntityNameLen = sensorEntityName.length();
         for (int i = 0; i < linksCount; i++) {
-            if (strcmp(links[i].sink->entity->info.name, sinkEntityName.c_str()) == 0) {
-                entityName = entity.info.name;
+            if (strcmp(links[i].sink->entity->info.name, sinkEntityName.c_str()) == 0 &&
+               strncmp(entity.info.name, sensorEntityName.c_str(), sensorEntityNameLen) == 0) {
+		entityName = entity.info.name;
                 break;
             }
         }