xprtrdma: Move Receive posting to Receive handler

Receive completion and Reply handling are done by a BOUND
workqueue, meaning they run on only one CPU.

Posting receives is currently done in the send_request path, which
on large systems is typically done on a different CPU than the one
handling Receive completions. This results in movement of
Receive-related cachelines between the sending and receiving CPUs.

More importantly, it means that currently Receives are posted while
the transport's write lock is held, which is unnecessary and costly.

Finally, allocation of Receive buffers is performed on-demand in
the Receive completion handler. This helps guarantee that they are
allocated on the same NUMA node as the CPU that handles Receive
completions.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

1 files changed, 7 insertions, 15 deletions

diff --git a/net/sunrpc/xprtrdma/rpc_rdma.c b/net/sunrpc/xprtrdma/rpc_rdma.c
index 8f89e3faae8e..d676106295ff 100644
--- a/net/sunrpc/xprtrdma/rpc_rdma.c
+++ b/net/sunrpc/xprtrdma/rpc_rdma.c
@@ -1027,8 +1027,6 @@ rpcrdma_is_bcall(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep)
 
 out_short:
 	pr_warn("RPC/RDMA short backward direction call\n");
-	if (rpcrdma_ep_post_recv(&r_xprt->rx_ia, rep))
-		xprt_disconnect_done(&r_xprt->rx_xprt);
 	return true;
 }
 #else	/* CONFIG_SUNRPC_BACKCHANNEL */
@@ -1334,13 +1332,14 @@ void rpcrdma_reply_handler(struct rpcrdma_rep *rep)
 	u32 credits;
 	__be32 *p;
 
+	--buf->rb_posted_receives;
+
 	if (rep->rr_hdrbuf.head[0].iov_len == 0)
 		goto out_badstatus;
 
+	/* Fixed transport header fields */
 	xdr_init_decode(&rep->rr_stream, &rep->rr_hdrbuf,
 			rep->rr_hdrbuf.head[0].iov_base);
-
-	/* Fixed transport header fields */
 	p = xdr_inline_decode(&rep->rr_stream, 4 * sizeof(*p));
 	if (unlikely(!p))
 		goto out_shortreply;
@@ -1379,17 +1378,10 @@ void rpcrdma_reply_handler(struct rpcrdma_rep *rep)
 
 	trace_xprtrdma_reply(rqst->rq_task, rep, req, credits);
 
+	rpcrdma_post_recvs(r_xprt, false);
 	queue_work(rpcrdma_receive_wq, &rep->rr_work);
 	return;
 
-out_badstatus:
-	rpcrdma_recv_buffer_put(rep);
-	if (r_xprt->rx_ep.rep_connected == 1) {
-		r_xprt->rx_ep.rep_connected = -EIO;
-		rpcrdma_conn_func(&r_xprt->rx_ep);
-	}
-	return;
-
 out_badversion:
 	trace_xprtrdma_reply_vers(rep);
 	goto repost;
@@ -1409,7 +1401,7 @@ out_shortreply:
  * receive buffer before returning.
  */
 repost:
-	r_xprt->rx_stats.bad_reply_count++;
-	if (rpcrdma_ep_post_recv(&r_xprt->rx_ia, rep))
-		rpcrdma_recv_buffer_put(rep);
+	rpcrdma_post_recvs(r_xprt, false);
+out_badstatus:
+	rpcrdma_recv_buffer_put(rep);
 }