[openib-general] (no subject)

[Or Gerlitz]

Hi Sean,

Looking in the code i have realized that it is a must for the CMA 
consumer to call rdma_disconnect to have the QP state moved into ERROR.

Am i correct? with this understanding at hand, i have changed iSER code to 
call rdma_disconnect even if it got a DISCONNECTED event caused by the 
passive side initiating the disconnect flow (ie sending a DREQ), since 
otherwise in such case i never got the FLUSHES on the posted RX/TX WRs.

This makes sense to me, just wanted to have your confirmation.

Or. 

Index: iser_verbs.c
===================================================================
--- iser_verbs.c	(revision 6965)
+++ iser_verbs.c	(working copy)
@@ -318,18 +318,20 @@ void iser_conn_terminate(struct iser_con
 {
 	int err = 0;
 
-	if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP,
-				      ISER_CONN_TERMINATING)) {
-		err = rdma_disconnect(ib_conn->cma_id);
-		if (err)
-			iser_err("Failed to disconnect, conn: 0x%p err %d\n",
-				 ib_conn,err);
-
-	}
-
+	/* change the ib conn state only if the conn is UP, however always call 
+	 * rdma_disconnect since this is the only way to cause the CMA to change
+	 * the QP state to ERROR 
+	 */
+
+	iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP, ISER_CONN_TERMINATING);
+	err = rdma_disconnect(ib_conn->cma_id);
+	if (err)
+		iser_err("Failed to disconnect, conn: 0x%p err %d\n",
+			 ib_conn,err);       
+	
 	wait_event_interruptible(ib_conn->wait,
 				 ib_conn->state == ISER_CONN_DOWN);
-
+	
 	iser_conn_release(ib_conn);
 }