Nithin, The driver is never blocked and it always executes in asynchronous fashion. A driver can immediately complete an I/O request by returning a status success code or an error. If the I/O is not completed the driver returns a pending status and the I/O will be completed later from a different thread context (this enables asynchronous/overlapped I/O for user mode process). Regardless of the driver specific I/O implementation, User mode thread can opt to wait on a I/O device request or not. If an overlapped structure is sent in the DeviceIoControl, the DeviceIOControl returns immediately regardless if the I/O has been completed or not. If a NULL sent rather than an overlapped structure, the thread which calls DeviceIOControl is blocked until the I/O is completed by the driver.
With respect to WFP related I/O I don't think we need to return pending status from the driver. It should be fine to block the user mode thread during the WFP transaction processing. Thanks, Eitan -----Original Message----- From: Nithin Raju Sent: Wednesday, May 20, 2015 11:04 AM To: Eitan Eliahu; Sorin Vinturis; dev@openvswitch.org Subject: Re: [ovs-dev] [PATCH v2 3/4] netlink: support for asynchronous NL transactions Sorin/Eitan, Do we need this patch or not? Eitan tells me that if return status is STATUS_PENDING, then the context is blocked in the kernel. If that is true, we probably don’t need this patch since control won’t return to userspace at all, till the IRP is complete. This is especially true since we are passing NULL for the ‘lpOverlapped’ parameter to DeviceIoControl(). Pls. see the following documentation on MSDN: ====== https://msdn.microsoft.com/en-us/library/windows/desktop/aa363216%28v=vs.85%29.aspx lpOverlapped [in, out, optional]: For overlapped operations, DeviceIoControl returns immediately, and the event object is signaled when the operation has been completed. Otherwise, the function does not return until the operation has been completed or an error occurs. ====== Based on this, my belief is that the call to DeviceIoControl() would block indefinitely until the IRP is completed. In your testing, did you think this patch was necessary? thanks, -- Nithin _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
