svcrdma: Add an async version of svc_rdma_send_ctxt_put()

DMA unmapping can take quite some time, so it should not be handled
in a single-threaded completion handler. Defer releasing send_ctxts
to the recently-added workqueue.

With this patch, DMA unmapping can be handled in parallel, and it
does not cause head-of-queue blocking of Send completions.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>

include/linux/sunrpc/svc_rdma.h

@@ -152,7 +152,9 @@ struct svc_rdma_recv_ctxt {
 struct svc_rdma_send_ctxt {
 	struct llist_node	sc_node;
 	struct rpc_rdma_cid	sc_cid;
+	struct work_struct	sc_work;
 
+	struct svcxprt_rdma	*sc_rdma;
 	struct ib_send_wr	sc_send_wr;
 	struct ib_cqe		sc_cqe;
 	struct xdr_buf		sc_hdrbuf;

net/sunrpc/xprtrdma/svc_rdma_sendto.c

@@ -143,6 +143,7 @@ svc_rdma_send_ctxt_alloc(struct svcxprt_rdma *rdma)
 
 	svc_rdma_send_cid_init(rdma, &ctxt->sc_cid);
 
+	ctxt->sc_rdma = rdma;
 	ctxt->sc_send_wr.next = NULL;
 	ctxt->sc_send_wr.wr_cqe = &ctxt->sc_cqe;
 	ctxt->sc_send_wr.sg_list = ctxt->sc_sges;
@@ -223,15 +224,8 @@ struct svc_rdma_send_ctxt *svc_rdma_send_ctxt_get(struct svcxprt_rdma *rdma)
 	goto out;
 }
 
-/**
- * svc_rdma_send_ctxt_put - Return send_ctxt to free list
- * @rdma: controlling svcxprt_rdma
- * @ctxt: object to return to the free list
- *
- * Pages left in sc_pages are DMA unmapped and released.
- */
-void svc_rdma_send_ctxt_put(struct svcxprt_rdma *rdma,
-			    struct svc_rdma_send_ctxt *ctxt)
+static void svc_rdma_send_ctxt_release(struct svcxprt_rdma *rdma,
+				       struct svc_rdma_send_ctxt *ctxt)
 {
 	struct ib_device *device = rdma->sc_cm_id->device;
 	unsigned int i;
@@ -255,6 +249,28 @@ void svc_rdma_send_ctxt_put(struct svcxprt_rdma *rdma,
 	llist_add(&ctxt->sc_node, &rdma->sc_send_ctxts);
 }
 
+static void svc_rdma_send_ctxt_put_async(struct work_struct *work)
+{
+	struct svc_rdma_send_ctxt *ctxt;
+
+	ctxt = container_of(work, struct svc_rdma_send_ctxt, sc_work);
+	svc_rdma_send_ctxt_release(ctxt->sc_rdma, ctxt);
+}
+
+/**
+ * svc_rdma_send_ctxt_put - Return send_ctxt to free list
+ * @rdma: controlling svcxprt_rdma
+ * @ctxt: object to return to the free list
+ *
+ * Pages left in sc_pages are DMA unmapped and released.
+ */
+void svc_rdma_send_ctxt_put(struct svcxprt_rdma *rdma,
+			    struct svc_rdma_send_ctxt *ctxt)
+{
+	INIT_WORK(&ctxt->sc_work, svc_rdma_send_ctxt_put_async);
+	queue_work(svcrdma_wq, &ctxt->sc_work);
+}
+
 /**
  * svc_rdma_wake_send_waiters - manage Send Queue accounting
  * @rdma: controlling transport