xprtrdma: Clean up sendctx functions

Minor clean-ups I've stumbled on since sendctx was merged last year.
In particular, making Send completion processing more efficient
appears to have a measurable impact on IOPS throughput.

Note: test_and_clear_bit() returns a value, thus an explicit memory
barrier is not necessary.

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

1 files changed, 12 insertions, 15 deletions

diff --git a/net/sunrpc/xprtrdma/rpc_rdma.c b/net/sunrpc/xprtrdma/rpc_rdma.c
index 45cba06655ea..5cb060c87543 100644
--- a/net/sunrpc/xprtrdma/rpc_rdma.c
+++ b/net/sunrpc/xprtrdma/rpc_rdma.c
@@ -508,30 +508,26 @@ rpcrdma_encode_reply_chunk(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req,
 }
 
 /**
- * rpcrdma_unmap_sendctx - DMA-unmap Send buffers
+ * rpcrdma_sendctx_unmap - DMA-unmap Send buffer
  * @sc: sendctx containing SGEs to unmap
  *
  */
-void
-rpcrdma_unmap_sendctx(struct rpcrdma_sendctx *sc)
+void rpcrdma_sendctx_unmap(struct rpcrdma_sendctx *sc)
 {
-	struct rpcrdma_ia *ia = &sc->sc_xprt->rx_ia;
 	struct ib_sge *sge;
-	unsigned int count;
 
 	/* The first two SGEs contain the transport header and
 	 * the inline buffer. These are always left mapped so
 	 * they can be cheaply re-used.
 	 */
-	sge = &sc->sc_sges[2];
-	for (count = sc->sc_unmap_count; count; ++sge, --count)
-		ib_dma_unmap_page(ia->ri_device,
-				  sge->addr, sge->length, DMA_TO_DEVICE);
+	for (sge = &sc->sc_sges[2]; sc->sc_unmap_count;
+	     ++sge, --sc->sc_unmap_count)
+		ib_dma_unmap_page(sc->sc_device, sge->addr, sge->length,
+				  DMA_TO_DEVICE);
 
-	if (test_and_clear_bit(RPCRDMA_REQ_F_TX_RESOURCES, &sc->sc_req->rl_flags)) {
-		smp_mb__after_atomic();
+	if (test_and_clear_bit(RPCRDMA_REQ_F_TX_RESOURCES,
+			       &sc->sc_req->rl_flags))
 		wake_up_bit(&sc->sc_req->rl_flags, RPCRDMA_REQ_F_TX_RESOURCES);
-	}
 }
 
 /* Prepare an SGE for the RPC-over-RDMA transport header.
@@ -578,6 +574,7 @@ static bool rpcrdma_prepare_msg_sges(struct rpcrdma_xprt *r_xprt,
 	 */
 	if (!rpcrdma_regbuf_dma_map(r_xprt, rb))
 		goto out_regbuf;
+	sc->sc_device = rdmab_device(rb);
 	sge_no = 1;
 	sge[sge_no].addr = rdmab_addr(rb);
 	sge[sge_no].length = xdr->head[0].iov_len;
@@ -673,12 +670,12 @@ out_regbuf:
 	return false;
 
 out_mapping_overflow:
-	rpcrdma_unmap_sendctx(sc);
+	rpcrdma_sendctx_unmap(sc);
 	pr_err("rpcrdma: too many Send SGEs (%u)\n", sge_no);
 	return false;
 
 out_mapping_err:
-	rpcrdma_unmap_sendctx(sc);
+	rpcrdma_sendctx_unmap(sc);
 	trace_xprtrdma_dma_maperr(sge[sge_no].addr);
 	return false;
 }
@@ -698,7 +695,7 @@ rpcrdma_prepare_send_sges(struct rpcrdma_xprt *r_xprt,
 			  struct rpcrdma_req *req, u32 hdrlen,
 			  struct xdr_buf *xdr, enum rpcrdma_chunktype rtype)
 {
-	req->rl_sendctx = rpcrdma_sendctx_get_locked(&r_xprt->rx_buf);
+	req->rl_sendctx = rpcrdma_sendctx_get_locked(r_xprt);
 	if (!req->rl_sendctx)
 		return -EAGAIN;
 	req->rl_sendctx->sc_wr.num_sge = 0;