@@ -793,6 +793,39 @@ TRACE_EVENT(xprtrdma_post_send,
)
);
+TRACE_EVENT(xprtrdma_post_send_err,
+ TP_PROTO(
+ const struct rpcrdma_xprt *r_xprt,
+ const struct rpcrdma_req *req,
+ int rc
+ ),
+
+ TP_ARGS(r_xprt, req, rc),
+
+ TP_STRUCT__entry(
+ __field(u32, cq_id)
+ __field(unsigned int, task_id)
+ __field(unsigned int, client_id)
+ __field(int, rc)
+ ),
+
+ TP_fast_assign(
+ const struct rpc_rqst *rqst = &req->rl_slot;
+ const struct rpcrdma_ep *ep = r_xprt->rx_ep;
+
+ __entry->cq_id = ep ? ep->re_attr.recv_cq->res.id : 0;
+ __entry->task_id = rqst->rq_task->tk_pid;
+ __entry->client_id = rqst->rq_task->tk_client ?
+ rqst->rq_task->tk_client->cl_clid : -1;
+ __entry->rc = rc;
+ ),
+
+ TP_printk("task:%u@%u cq.id=%u rc=%d",
+ __entry->task_id, __entry->client_id,
+ __entry->cq_id, __entry->rc
+ )
+);
+
TRACE_EVENT(xprtrdma_post_recv,
TP_PROTO(
const struct rpcrdma_rep *rep
@@ -394,6 +394,7 @@ int frwr_send(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req)
struct rpcrdma_ep *ep = r_xprt->rx_ep;
struct rpcrdma_mr *mr;
unsigned int num_wrs;
+ int ret;
num_wrs = 1;
post_wr = send_wr;
@@ -420,7 +421,10 @@ int frwr_send(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req)
}
trace_xprtrdma_post_send(req);
- return ib_post_send(ep->re_id->qp, post_wr, NULL);
+ ret = ib_post_send(ep->re_id->qp, post_wr, NULL);
+ if (ret)
+ trace_xprtrdma_post_send_err(r_xprt, req, ret);
+ return ret;
}
/**
Unlike xprtrdma_post_send(), this one can be left enabled all the time, and should almost never fire. But we do want to know about immediate errors when they happen. Note that there is already a similar post_linv_err tracepoint. Signed-off-by: Chuck Lever <chuck.lever@oracle.com> --- include/trace/events/rpcrdma.h | 33 +++++++++++++++++++++++++++++++++ net/sunrpc/xprtrdma/frwr_ops.c | 6 +++++- 2 files changed, 38 insertions(+), 1 deletion(-)