SUNRPC: RPC transport queue must be low latency

rpciod can easily get congested due to the long list of queued rpc_tasks.
Having the receive queue wait in turn for those tasks to complete can
therefore be a bottleneck.

Address the problem by separating the workqueues into:
- rpciod: manages rpc_tasks
- xprtiod: manages transport related work.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index 216a138..71df082 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -220,7 +220,7 @@
 		clear_bit(XPRT_LOCKED, &xprt->state);
 		smp_mb__after_atomic();
 	} else
-		queue_work(rpciod_workqueue, &xprt->task_cleanup);
+		queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 }
 
 /*
@@ -645,7 +645,7 @@
 	set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 	/* Try to schedule an autoclose RPC call */
 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
-		queue_work(rpciod_workqueue, &xprt->task_cleanup);
+		queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 	xprt_wake_pending_tasks(xprt, -EAGAIN);
 	spin_unlock_bh(&xprt->transport_lock);
 }
@@ -672,7 +672,7 @@
 	set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 	/* Try to schedule an autoclose RPC call */
 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
-		queue_work(rpciod_workqueue, &xprt->task_cleanup);
+		queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 	xprt_wake_pending_tasks(xprt, -EAGAIN);
 out:
 	spin_unlock_bh(&xprt->transport_lock);
@@ -689,7 +689,7 @@
 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 		goto out_abort;
 	spin_unlock(&xprt->transport_lock);
-	queue_work(rpciod_workqueue, &xprt->task_cleanup);
+	queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 	return;
 out_abort:
 	spin_unlock(&xprt->transport_lock);