From: Chuck Lever <chuck.lever@oracle.com>
Date: Mon, 16 Oct 2017 15:01:30 -0400
Subject: xprtrdma: Invoke rpcrdma_reply_handler directly from RECV completion
Patch-mainline: v4.15-rc1
Git-commit: d8f532d20ee43a0117284798d486bc4f98e3b196
References: bsc#1103992 FATE#326009
I noticed that the soft IRQ thread looked pretty busy under heavy
I/O workloads. perf suggested one area that was expensive was the
queue_work() call in rpcrdma_wc_receive. That gave me some ideas.
Instead of scheduling a separate worker to process RPC Replies,
promote the Receive completion handler to IB_POLL_WORKQUEUE, and
invoke rpcrdma_reply_handler directly.
Note that the poll workqueue is single-threaded. In order to keep
memory invalidation from serializing all RPC Replies, handle any
necessary invalidation tasks in a separate multi-threaded workqueue.
This provides a two-tier scheme, similar to OS I/O interrupt
handlers: A fast interrupt handler that schedules the slow handler
and re-enables the interrupt, and a slower handler that is invoked
for any needed heavy lifting.
Benefits include:
- One less context switch for RPCs that don't register memory
- Receive completion handling is moved out of soft IRQ context to
make room for other users of soft IRQ
- The same CPU core now DMA syncs and XDR decodes the Receive buffer
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Acked-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
---
net/sunrpc/xprtrdma/rpc_rdma.c | 46 ++++++++++++++++++++++++----------------
net/sunrpc/xprtrdma/verbs.c | 8 +++---
net/sunrpc/xprtrdma/xprt_rdma.h | 5 +++-
3 files changed, 36 insertions(+), 23 deletions(-)
--- a/net/sunrpc/xprtrdma/rpc_rdma.c
+++ b/net/sunrpc/xprtrdma/rpc_rdma.c
@@ -1265,16 +1265,36 @@ out_badheader:
goto out;
}
+/* Reply handling runs in the poll worker thread. Anything that
+ * might wait is deferred to a separate workqueue.
+ */
+void rpcrdma_deferred_completion(struct work_struct *work)
+{
+ struct rpcrdma_rep *rep =
+ container_of(work, struct rpcrdma_rep, rr_work);
+ struct rpcrdma_req *req = rpcr_to_rdmar(rep->rr_rqst);
+ struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
+
+ /* Invalidate and unmap the data payloads before waking
+ * the waiting application. This guarantees the memory
+ * regions are properly fenced from the server before the
+ * application accesses the data. It also ensures proper
+ * send flow control: waking the next RPC waits until this
+ * RPC has relinquished all its Send Queue entries.
+ */
+ rpcrdma_mark_remote_invalidation(&req->rl_registered, rep);
+ r_xprt->rx_ia.ri_ops->ro_unmap_sync(r_xprt, &req->rl_registered);
+
+ rpcrdma_complete_rqst(rep);
+}
+
/* Process received RPC/RDMA messages.
*
* Errors must result in the RPC task either being awakened, or
* allowed to timeout, to discover the errors at that time.
*/
-void
-rpcrdma_reply_handler(struct work_struct *work)
+void rpcrdma_reply_handler(struct rpcrdma_rep *rep)
{
- struct rpcrdma_rep *rep =
- container_of(work, struct rpcrdma_rep, rr_work);
struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
struct rpc_xprt *xprt = &r_xprt->rx_xprt;
struct rpcrdma_req *req;
@@ -1320,20 +1340,10 @@ rpcrdma_reply_handler(struct work_struct
dprintk("RPC: %s: reply %p completes request %p (xid 0x%08x)\n",
__func__, rep, req, be32_to_cpu(rep->rr_xid));
- /* Invalidate and unmap the data payloads before waking the
- * waiting application. This guarantees the memory regions
- * are properly fenced from the server before the application
- * accesses the data. It also ensures proper send flow control:
- * waking the next RPC waits until this RPC has relinquished
- * all its Send Queue entries.
- */
- if (!list_empty(&req->rl_registered)) {
- rpcrdma_mark_remote_invalidation(&req->rl_registered, rep);
- r_xprt->rx_ia.ri_ops->ro_unmap_sync(r_xprt,
- &req->rl_registered);
- }
-
- rpcrdma_complete_rqst(rep);
+ if (list_empty(&req->rl_registered))
+ rpcrdma_complete_rqst(rep);
+ else
+ queue_work(rpcrdma_receive_wq, &rep->rr_work);
return;
out_badstatus:
--- a/net/sunrpc/xprtrdma/verbs.c
+++ b/net/sunrpc/xprtrdma/verbs.c
@@ -73,7 +73,7 @@ static void rpcrdma_create_mrs(struct rp
static void rpcrdma_destroy_mrs(struct rpcrdma_buffer *buf);
static void rpcrdma_dma_unmap_regbuf(struct rpcrdma_regbuf *rb);
-static struct workqueue_struct *rpcrdma_receive_wq __read_mostly;
+struct workqueue_struct *rpcrdma_receive_wq __read_mostly;
int
rpcrdma_alloc_wq(void)
@@ -185,7 +185,7 @@ rpcrdma_wc_receive(struct ib_cq *cq, str
rpcrdma_update_granted_credits(rep);
out_schedule:
- queue_work(rpcrdma_receive_wq, &rep->rr_work);
+ rpcrdma_reply_handler(rep);
return;
out_fail:
@@ -583,7 +583,7 @@ rpcrdma_ep_create(struct rpcrdma_ep *ep,
recvcq = ib_alloc_cq(ia->ri_device, NULL,
ep->rep_attr.cap.max_recv_wr + 1,
- 0, IB_POLL_SOFTIRQ);
+ 0, IB_POLL_WORKQUEUE);
if (IS_ERR(recvcq)) {
rc = PTR_ERR(recvcq);
dprintk("RPC: %s: failed to create recv CQ: %i\n",
@@ -974,7 +974,7 @@ rpcrdma_create_rep(struct rpcrdma_xprt *
rep->rr_cqe.done = rpcrdma_wc_receive;
rep->rr_rxprt = r_xprt;
- INIT_WORK(&rep->rr_work, rpcrdma_reply_handler);
+ INIT_WORK(&rep->rr_work, rpcrdma_deferred_completion);
rep->rr_recv_wr.next = NULL;
rep->rr_recv_wr.wr_cqe = &rep->rr_cqe;
rep->rr_recv_wr.sg_list = &rep->rr_rdmabuf->rg_iov;
--- a/net/sunrpc/xprtrdma/xprt_rdma.h
+++ b/net/sunrpc/xprtrdma/xprt_rdma.h
@@ -533,6 +533,8 @@ void rpcrdma_ia_close(struct rpcrdma_ia
bool frwr_is_supported(struct rpcrdma_ia *);
bool fmr_is_supported(struct rpcrdma_ia *);
+extern struct workqueue_struct *rpcrdma_receive_wq;
+
/*
* Endpoint calls - xprtrdma/verbs.c
*/
@@ -617,7 +619,8 @@ void rpcrdma_unmap_sges(struct rpcrdma_i
int rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst);
void rpcrdma_set_max_header_sizes(struct rpcrdma_xprt *);
void rpcrdma_complete_rqst(struct rpcrdma_rep *rep);
-void rpcrdma_reply_handler(struct work_struct *work);
+void rpcrdma_reply_handler(struct rpcrdma_rep *rep);
+void rpcrdma_deferred_completion(struct work_struct *work);
static inline void rpcrdma_set_xdrlen(struct xdr_buf *xdr, size_t len)
{