xprtrdma: Use xprt_pin_rqst in rpcrdma_reply_handler
authorChuck Lever <chuck.lever@oracle.com>
Wed, 23 Aug 2017 21:05:58 +0000 (17:05 -0400)
committerTrond Myklebust <trond.myklebust@primarydata.com>
Tue, 5 Sep 2017 22:27:07 +0000 (18:27 -0400)
Adopt the use of xprt_pin_rqst to eliminate contention between
Call-side users of rb_lock and the use of rb_lock in
rpcrdma_reply_handler.

This replaces the mechanism introduced in 431af645cf66 ("xprtrdma:
Fix client lock-up after application signal fires").

Use recv_lock to quickly find the completing rqst, pin it, then
drop the lock. At that point invalidation and pull-up of the Reply
XDR can be done. Both are often expensive operations.

Finally, take recv_lock again to signal completion to the RPC
layer. It also protects adjustment of "cwnd".

This greatly reduces the amount of time a lock is held by the
reply handler. Comparing lock_stat results shows a marked decrease
in contention on rb_lock and recv_lock.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[trond.myklebust@primarydata.com: Remove call to rpcrdma_buffer_put() from
   the "out_norqst:" path in rpcrdma_reply_handler.]
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
net/sunrpc/xprt.c
net/sunrpc/xprtrdma/rpc_rdma.c
net/sunrpc/xprtrdma/transport.c
net/sunrpc/xprtrdma/verbs.c
net/sunrpc/xprtrdma/xprt_rdma.h

index 2af189c5ac3e513347e1d47a789dfc76920568f5..e741ec2b4d8e6ea5d08a8e942e3021e5a20e4c6a 100644 (file)
@@ -855,6 +855,7 @@ void xprt_pin_rqst(struct rpc_rqst *req)
 {
        set_bit(RPC_TASK_MSG_RECV, &req->rq_task->tk_runstate);
 }
+EXPORT_SYMBOL_GPL(xprt_pin_rqst);
 
 /**
  * xprt_unpin_rqst - Unpin a request on the transport receive list
@@ -870,6 +871,7 @@ void xprt_unpin_rqst(struct rpc_rqst *req)
        if (test_bit(RPC_TASK_MSG_RECV_WAIT, &task->tk_runstate))
                wake_up_bit(&task->tk_runstate, RPC_TASK_MSG_RECV);
 }
+EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
 
 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
 __must_hold(&req->rq_xprt->recv_lock)
index 7fec4039cd15ac6b5b92be13f6d61c97ed34cdaf..f1889f4d48030f3a0a07eef6fa524bbc12580216 100644 (file)
@@ -781,9 +781,6 @@ rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst)
                rtype = rpcrdma_areadch;
        }
 
-       req->rl_xid = rqst->rq_xid;
-       rpcrdma_insert_req(&r_xprt->rx_buf, req);
-
        /* This implementation supports the following combinations
         * of chunk lists in one RPC-over-RDMA Call message:
         *
@@ -1226,14 +1223,12 @@ rpcrdma_reply_handler(struct work_struct *work)
        struct rpcrdma_rep *rep =
                        container_of(work, struct rpcrdma_rep, rr_work);
        struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
-       struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
        struct rpc_xprt *xprt = &r_xprt->rx_xprt;
        struct xdr_stream *xdr = &rep->rr_stream;
        struct rpcrdma_req *req;
        struct rpc_rqst *rqst;
        __be32 *p, xid, vers, proc;
        unsigned long cwnd;
-       struct list_head mws;
        int status;
 
        dprintk("RPC:       %s: incoming rep %p\n", __func__, rep);
@@ -1259,21 +1254,14 @@ rpcrdma_reply_handler(struct work_struct *work)
        /* Match incoming rpcrdma_rep to an rpcrdma_req to
         * get context for handling any incoming chunks.
         */
-       spin_lock(&buf->rb_lock);
-       req = rpcrdma_lookup_req_locked(&r_xprt->rx_buf, xid);
-       if (!req)
-               goto out_nomatch;
-       if (req->rl_reply)
-               goto out_duplicate;
-
-       list_replace_init(&req->rl_registered, &mws);
-       rpcrdma_mark_remote_invalidation(&mws, rep);
-
-       /* Avoid races with signals and duplicate replies
-        * by marking this req as matched.
-        */
+       spin_lock(&xprt->recv_lock);
+       rqst = xprt_lookup_rqst(xprt, xid);
+       if (!rqst)
+               goto out_norqst;
+       xprt_pin_rqst(rqst);
+       spin_unlock(&xprt->recv_lock);
+       req = rpcr_to_rdmar(rqst);
        req->rl_reply = rep;
-       spin_unlock(&buf->rb_lock);
 
        dprintk("RPC:       %s: reply %p completes request %p (xid 0x%08x)\n",
                __func__, rep, req, be32_to_cpu(xid));
@@ -1285,17 +1273,12 @@ rpcrdma_reply_handler(struct work_struct *work)
         * waking the next RPC waits until this RPC has relinquished
         * all its Send Queue entries.
         */
-       if (!list_empty(&mws))
-               r_xprt->rx_ia.ri_ops->ro_unmap_sync(r_xprt, &mws);
+       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);
+       }
 
-       /* Perform XID lookup, reconstruction of the RPC reply, and
-        * RPC completion while holding the transport lock to ensure
-        * the rep, rqst, and rq_task pointers remain stable.
-        */
-       spin_lock(&xprt->recv_lock);
-       rqst = xprt_lookup_rqst(xprt, xid);
-       if (!rqst)
-               goto out_norqst;
        xprt->reestablish_timeout = 0;
        if (vers != rpcrdma_version)
                goto out_badversion;
@@ -1317,12 +1300,14 @@ rpcrdma_reply_handler(struct work_struct *work)
                goto out_badheader;
 
 out:
+       spin_lock(&xprt->recv_lock);
        cwnd = xprt->cwnd;
        xprt->cwnd = atomic_read(&r_xprt->rx_buf.rb_credits) << RPC_CWNDSHIFT;
        if (xprt->cwnd > cwnd)
                xprt_release_rqst_cong(rqst->rq_task);
 
        xprt_complete_rqst(rqst->rq_task, status);
+       xprt_unpin_rqst(rqst);
        spin_unlock(&xprt->recv_lock);
        dprintk("RPC:       %s: xprt_complete_rqst(0x%p, 0x%p, %d)\n",
                __func__, xprt, rqst, status);
@@ -1360,26 +1345,13 @@ out_badheader:
  */
 out_norqst:
        spin_unlock(&xprt->recv_lock);
-       rpcrdma_buffer_put(req);
-       dprintk("RPC:       %s: race, no rqst left for req %p\n",
-               __func__, req);
-       return;
-
-out_shortreply:
-       dprintk("RPC:       %s: short/invalid reply\n", __func__);
-       goto repost;
-
-out_nomatch:
-       spin_unlock(&buf->rb_lock);
        dprintk("RPC:       %s: no match for incoming xid 0x%08x\n",
                __func__, be32_to_cpu(xid));
        goto repost;
 
-out_duplicate:
-       spin_unlock(&buf->rb_lock);
-       dprintk("RPC:       %s: "
-               "duplicate reply %p to RPC request %p: xid 0x%08x\n",
-               __func__, rep, req, be32_to_cpu(xid));
+out_shortreply:
+       dprintk("RPC:       %s: short/invalid reply\n", __func__);
+       goto repost;
 
 /* If no pending RPC transaction was matched, post a replacement
  * receive buffer before returning.
index b680591f6763b078cef03678646846fe071edc41..c84e2b644e133ee13c94082b171acbe063bf3f97 100644 (file)
@@ -685,7 +685,6 @@ xprt_rdma_free(struct rpc_task *task)
 
        dprintk("RPC:       %s: called on 0x%p\n", __func__, req->rl_reply);
 
-       rpcrdma_remove_req(&r_xprt->rx_buf, req);
        if (!list_empty(&req->rl_registered))
                ia->ri_ops->ro_unmap_safe(r_xprt, req, !RPC_IS_ASYNC(task));
        rpcrdma_unmap_sges(ia, req);
index c78fb27c20ed2ff6e3f55fe486e0746601b95ce8..11a1fbf7e59e08e9172f4562652dc55d53240075 100644 (file)
@@ -1001,7 +1001,6 @@ rpcrdma_buffer_create(struct rpcrdma_xprt *r_xprt)
        spin_lock_init(&buf->rb_recovery_lock);
        INIT_LIST_HEAD(&buf->rb_mws);
        INIT_LIST_HEAD(&buf->rb_all);
-       INIT_LIST_HEAD(&buf->rb_pending);
        INIT_LIST_HEAD(&buf->rb_stale_mrs);
        INIT_DELAYED_WORK(&buf->rb_refresh_worker,
                          rpcrdma_mr_refresh_worker);
index 45dab2475c99ecc1ff83869cfae67582d6bc4a4a..e26a97d2f922fad2cc14aa11ea0cd4287d1c26b5 100644 (file)
@@ -340,7 +340,6 @@ enum {
 struct rpcrdma_buffer;
 struct rpcrdma_req {
        struct list_head        rl_list;
-       __be32                  rl_xid;
        unsigned int            rl_mapped_sges;
        unsigned int            rl_connect_cookie;
        struct rpcrdma_buffer   *rl_buffer;
@@ -404,7 +403,6 @@ struct rpcrdma_buffer {
        int                     rb_send_count, rb_recv_count;
        struct list_head        rb_send_bufs;
        struct list_head        rb_recv_bufs;
-       struct list_head        rb_pending;
        u32                     rb_max_requests;
        atomic_t                rb_credits;     /* most recent credit grant */
 
@@ -557,34 +555,6 @@ void rpcrdma_destroy_req(struct rpcrdma_req *);
 int rpcrdma_buffer_create(struct rpcrdma_xprt *);
 void rpcrdma_buffer_destroy(struct rpcrdma_buffer *);
 
-static inline void
-rpcrdma_insert_req(struct rpcrdma_buffer *buffers, struct rpcrdma_req *req)
-{
-       spin_lock(&buffers->rb_lock);
-       if (list_empty(&req->rl_list))
-               list_add_tail(&req->rl_list, &buffers->rb_pending);
-       spin_unlock(&buffers->rb_lock);
-}
-
-static inline struct rpcrdma_req *
-rpcrdma_lookup_req_locked(struct rpcrdma_buffer *buffers, __be32 xid)
-{
-       struct rpcrdma_req *pos;
-
-       list_for_each_entry(pos, &buffers->rb_pending, rl_list)
-               if (pos->rl_xid == xid)
-                       return pos;
-       return NULL;
-}
-
-static inline void
-rpcrdma_remove_req(struct rpcrdma_buffer *buffers, struct rpcrdma_req *req)
-{
-       spin_lock(&buffers->rb_lock);
-       list_del(&req->rl_list);
-       spin_unlock(&buffers->rb_lock);
-}
-
 struct rpcrdma_mw *rpcrdma_get_mw(struct rpcrdma_xprt *);
 void rpcrdma_put_mw(struct rpcrdma_xprt *, struct rpcrdma_mw *);
 struct rpcrdma_req *rpcrdma_buffer_get(struct rpcrdma_buffer *);