b200/dtor-stall: fixed bug that stalled b200 on shutdown.

1 files changed, 99 insertions, 0 deletions

diff --git a/host/lib/utils/tasks.cpp b/host/lib/utils/tasks.cpp
index 1f735de06..08c32a5fb 100644
--- a/host/lib/utils/tasks.cpp
+++ b/host/lib/utils/tasks.cpp
@@ -16,11 +16,13 @@
 //
 
 #include <uhd/utils/tasks.hpp>
+#include <uhd/utils/msg_task.hpp>
 #include <uhd/utils/msg.hpp>
 #include <boost/thread/thread.hpp>
 #include <boost/thread/barrier.hpp>
 #include <exception>
 #include <iostream>
+#include <vector>
 
 using namespace uhd;
 
@@ -80,3 +82,100 @@ private:
 task::sptr task::make(const task_fcn_type &task_fcn){
     return task::sptr(new task_impl(task_fcn));
 }
+
+/*
+ * During shutdown pointers to queues for radio_ctrl_core might not be available anymore.
+ * msg_task_impl provides a dump_queue for such messages.
+ * ctrl_cores can check this queue for stranded messages.
+ */
+
+class msg_task_impl : public msg_task{
+public:
+
+    msg_task_impl(const task_fcn_type &task_fcn):
+        _spawn_barrier(2)
+    {
+        _thread_group.create_thread(boost::bind(&msg_task_impl::task_loop, this, task_fcn));
+        _spawn_barrier.wait();
+    }
+
+    ~msg_task_impl(void){
+        _running = false;
+        _thread_group.interrupt_all();
+        _thread_group.join_all();
+    }
+
+    /*
+     * Returns the first message for the given SID.
+     * This way a radio_ctrl_core doesn't have to die in timeout but can check for stranded messages here.
+     * This might happen during shutdown when dtors are called.
+     * See also: comments in b200_io_impl->handle_async_task
+     */
+    msg_payload_t get_msg_from_dump_queue(boost::uint32_t sid)
+    {
+        boost::mutex::scoped_lock lock(_mutex);
+        msg_payload_t b;
+        for (size_t i = 0; i < _dump_queue.size(); i++) {
+            if (sid == _dump_queue[i].first) {
+                b = _dump_queue[i].second;
+                _dump_queue.erase(_dump_queue.begin() + i);
+                break;
+            }
+        }
+        return b;
+    }
+
+private:
+
+    void task_loop(const task_fcn_type &task_fcn){
+        _running = true;
+        _spawn_barrier.wait();
+
+        try{
+            while (_running){
+            	boost::optional<msg_type_t> buff = task_fcn();
+            	if(buff != boost::none){
+            	    /*
+            	     * If a message gets stranded it is returned by task_fcn and then pushed to the dump_queue.
+            	     * This way ctrl_cores can check dump_queue for missing messages.
+            	     */
+            	    boost::mutex::scoped_lock lock(_mutex);
+            	    _dump_queue.push_back(buff.get() );
+            	}
+            }
+        }
+        catch(const boost::thread_interrupted &){
+            //this is an ok way to exit the task loop
+        }
+        catch(const std::exception &e){
+            do_error_msg(e.what());
+        }
+        catch(...){
+            //FIXME
+            //Unfortunately, this is also an ok way to end a task,
+            //because on some systems boost throws uncatchables.
+        }
+    }
+
+    void do_error_msg(const std::string &msg){
+        UHD_MSG(error)
+            << "An unexpected exception was caught in a task loop." << std::endl
+            << "The task loop will now exit, things may not work." << std::endl
+            << msg << std::endl
+        ;
+    }
+
+    boost::mutex _mutex;
+    boost::thread_group _thread_group;
+    boost::barrier _spawn_barrier;
+    bool _running;
+
+    /*
+     * This queue holds stranded messages until a radio_ctrl_core grabs them via 'get_msg_from_dump_queue'.
+     */
+    std::vector <msg_type_t> _dump_queue;
+};
+
+msg_task::sptr msg_task::make(const task_fcn_type &task_fcn){
+    return msg_task::sptr(new msg_task_impl(task_fcn));
+}