summaryrefslogtreecommitdiffstats
path: root/net/sunrpc
diff options
context:
space:
mode:
authorChuck Lever <chuck.lever@oracle.com>2018-05-04 21:35:57 +0200
committerAnna Schumaker <Anna.Schumaker@Netapp.com>2018-06-01 19:56:30 +0200
commit2fad659209d5b1dbaa1f58606372571c61fc8cbd (patch)
treecc0cdc1adb78753269485d4754f5182d989ef5c6 /net/sunrpc
parentxprtrdma: Move common wait_for_buffer_space call to parent function (diff)
downloadlinux-2fad659209d5b1dbaa1f58606372571c61fc8cbd.tar.xz
linux-2fad659209d5b1dbaa1f58606372571c61fc8cbd.zip
xprtrdma: Wait on empty sendctx queue
Currently, when the sendctx queue is exhausted during marshaling, the RPC/RDMA transport places the RPC task on the delayq, which forces a wait for HZ >> 2 before the marshal and send is retried. With this change, the transport now places such an RPC task on the pending queue, and wakes it just as soon as more sendctxs become available. This typically takes less than a millisecond, and the write_space waking mechanism is less deadlock-prone. Moreover, the waiting RPC task is holding the transport's write lock, which blocks the transport from sending RPCs. Therefore faster recovery from sendctx queue exhaustion is desirable. Cf. commit 5804891455d5 ("xprtrdma: ->send_request returns -EAGAIN when there are no free MRs"). Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Diffstat (limited to 'net/sunrpc')
-rw-r--r--net/sunrpc/xprtrdma/rpc_rdma.c2
-rw-r--r--net/sunrpc/xprtrdma/verbs.c8
-rw-r--r--net/sunrpc/xprtrdma/xprt_rdma.h6
3 files changed, 14 insertions, 2 deletions
diff --git a/net/sunrpc/xprtrdma/rpc_rdma.c b/net/sunrpc/xprtrdma/rpc_rdma.c
index b12b0443d33a..a373d0322a80 100644
--- a/net/sunrpc/xprtrdma/rpc_rdma.c
+++ b/net/sunrpc/xprtrdma/rpc_rdma.c
@@ -695,7 +695,7 @@ rpcrdma_prepare_send_sges(struct rpcrdma_xprt *r_xprt,
{
req->rl_sendctx = rpcrdma_sendctx_get_locked(&r_xprt->rx_buf);
if (!req->rl_sendctx)
- return -ENOBUFS;
+ return -EAGAIN;
req->rl_sendctx->sc_wr.num_sge = 0;
req->rl_sendctx->sc_unmap_count = 0;
req->rl_sendctx->sc_req = req;
diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c
index 0e0b7d5cb74d..7276e82db3b4 100644
--- a/net/sunrpc/xprtrdma/verbs.c
+++ b/net/sunrpc/xprtrdma/verbs.c
@@ -878,6 +878,7 @@ static int rpcrdma_sendctxs_create(struct rpcrdma_xprt *r_xprt)
sc->sc_xprt = r_xprt;
buf->rb_sc_ctxs[i] = sc;
}
+ buf->rb_flags = 0;
return 0;
@@ -935,7 +936,7 @@ out_emptyq:
* completions recently. This is a sign the Send Queue is
* backing up. Cause the caller to pause and try again.
*/
- dprintk("RPC: %s: empty sendctx queue\n", __func__);
+ set_bit(RPCRDMA_BUF_F_EMPTY_SCQ, &buf->rb_flags);
r_xprt = container_of(buf, struct rpcrdma_xprt, rx_buf);
r_xprt->rx_stats.empty_sendctx_q++;
return NULL;
@@ -970,6 +971,11 @@ rpcrdma_sendctx_put_locked(struct rpcrdma_sendctx *sc)
/* Paired with READ_ONCE */
smp_store_release(&buf->rb_sc_tail, next_tail);
+
+ if (test_and_clear_bit(RPCRDMA_BUF_F_EMPTY_SCQ, &buf->rb_flags)) {
+ smp_mb__after_atomic();
+ xprt_write_space(&sc->sc_xprt->rx_xprt);
+ }
}
static void
diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h
index c60687934092..e4a408d7b195 100644
--- a/net/sunrpc/xprtrdma/xprt_rdma.h
+++ b/net/sunrpc/xprtrdma/xprt_rdma.h
@@ -400,6 +400,7 @@ struct rpcrdma_buffer {
spinlock_t rb_lock; /* protect buf lists */
struct list_head rb_send_bufs;
struct list_head rb_recv_bufs;
+ unsigned long rb_flags;
u32 rb_max_requests;
u32 rb_credits; /* most recent credit grant */
int rb_posted_receives;
@@ -417,6 +418,11 @@ struct rpcrdma_buffer {
};
#define rdmab_to_ia(b) (&container_of((b), struct rpcrdma_xprt, rx_buf)->rx_ia)
+/* rb_flags */
+enum {
+ RPCRDMA_BUF_F_EMPTY_SCQ = 0,
+};
+
/*
* Internal structure for transport instance creation. This
* exists primarily for modularity.