On Tue, Jan 25, 2011 at 02:57:34PM +0800, Wen Congyang wrote:
When we kill the qemu, the function qemuMonitorSetCapabilities() failed and then we close monitor.
In another thread, mon->fd is broken and the function qemuHandleMonitorEOF() is called. The function qemuHandleMonitorEOF() calls qemudShutdownVMDaemon() to shutdown vm. The monitor will be closed in the function qemudShutdownVMDaemon().
The monitor close twice and the reference is decreased to 0 unexpectedly. The memory will be freed when reference is decreased to 0.
We will remove the watch of mon->fd when the monitor is closed. This request will be done in the function qemuMonitorUnwatch() in the qemuloop thread. In the function qemuMonitorUnwatch(), we will lock monitor, but the lock is destroyed and we will block here,
In the main thread, we may add some watch or timeout, and will be blocked because the lock of eventLoop is hold by qemuLoop thread.
We should close monitor only once.
I think the problem actually lies in the qemuConnectMonitor() call. This method calls qemuMonitorSetCapabilities() and if that fails it calls qemuMonitorClose(). The caller of qemuConnectMonitor() will see the error code and also try to kill the QEMU process, by calling qemuShutdownVMDaemon(), which calls qemuMonitorClose() again. So I think we need to remove the call to qemuMonitorClose() from qemuConnectMonitor() and just let the calls cleanup up via normal VM shutdown procedure. Regards, Daniel