svcrdma: Fewer calls to wake_up() in Send completion handler

Because wake_up() takes an IRQ-safe lock, it can be expensive,
especially to call inside of a single-threaded completion handler.
What's more, the Send wait queue almost never has waiters, so
most of the time, this is an expensive no-op.

As always, the goal is to reduce the average overhead of each
completion, because a transport's completion handlers are single-
threaded on one CPU core. This change reduces CPU utilization of
the Send completion thread by 2-3% on my server.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-By: Tom Talpey <tom@talpey.com>

1 files changed, 15 insertions, 3 deletions

diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index d6bbafb773e1..fba2ee4eb607 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -259,6 +259,20 @@ void svc_rdma_send_ctxt_put(struct svcxprt_rdma *rdma,
 }
 
 /**
+ * svc_rdma_wake_send_waiters - manage Send Queue accounting
+ * @rdma: controlling transport
+ * @avail: Number of additional SQEs that are now available
+ *
+ */
+void svc_rdma_wake_send_waiters(struct svcxprt_rdma *rdma, int avail)
+{
+	atomic_add(avail, &rdma->sc_sq_avail);
+	smp_mb__after_atomic();
+	if (unlikely(waitqueue_active(&rdma->sc_send_wait)))
+		wake_up(&rdma->sc_send_wait);
+}
+
+/**
  * svc_rdma_wc_send - Invoked by RDMA provider for each polled Send WC
  * @cq: Completion Queue context
  * @wc: Work Completion object
@@ -275,11 +289,9 @@ static void svc_rdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
 
 	trace_svcrdma_wc_send(wc, &ctxt->sc_cid);
 
+	svc_rdma_wake_send_waiters(rdma, 1);
 	complete(&ctxt->sc_done);
 
-	atomic_inc(&rdma->sc_sq_avail);
-	wake_up(&rdma->sc_send_wait);
-
 	if (unlikely(wc->status != IB_WC_SUCCESS))
 		svc_xprt_deferred_close(&rdma->sc_xprt);
 }