From 10694ac92c063681246a360a45019f05855ab755 Mon Sep 17 00:00:00 2001 From: Chuck Lever Date: Wed, 19 Jun 2019 10:32:32 -0400 Subject: xprtrdma: Fix a BUG when tracing is enabled with NFSv4.1 on RDMA A backchannel reply does not set task->tk_client. Fixes: 0c77668ddb4e ("SUNRPC: Introduce trace points in ... ") Signed-off-by: Chuck Lever Signed-off-by: Anna Schumaker --- include/trace/events/rpcrdma.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'include/trace/events') diff --git a/include/trace/events/rpcrdma.h b/include/trace/events/rpcrdma.h index df9851cb82b2..f0678e3ac2d4 100644 --- a/include/trace/events/rpcrdma.h +++ b/include/trace/events/rpcrdma.h @@ -559,7 +559,8 @@ TRACE_EVENT(xprtrdma_post_send, const struct rpc_rqst *rqst = &req->rl_slot; __entry->task_id = rqst->rq_task->tk_pid; - __entry->client_id = rqst->rq_task->tk_client->cl_clid; + __entry->client_id = rqst->rq_task->tk_client ? + rqst->rq_task->tk_client->cl_clid : -1; __entry->req = req; __entry->num_sge = req->rl_sendctx->sc_wr.num_sge; __entry->signaled = req->rl_sendctx->sc_wr.send_flags & -- cgit v1.2.1 From 1310051c720a83c5717658bcbff710b260f2bff9 Mon Sep 17 00:00:00 2001 From: Chuck Lever Date: Wed, 19 Jun 2019 10:32:43 -0400 Subject: xprtrdma: Replace use of xdr_stream_pos in rpcrdma_marshal_req This is a latent bug. xdr_stream_pos works by subtracting xdr_stream::nwords from xdr_buf::len. But xdr_stream::nwords is not initialized by xdr_init_encode(). It works today only because all fields in rpcrdma_req::rl_stream are initialized to zero by rpcrdma_req_create, making the subtraction in xdr_stream_pos always a no-op. I found this issue via code inspection. It was introduced by commit 39f4cd9e9982 ("xprtrdma: Harden chunk list encoding against send buffer overflow"), but the code has changed enough since then that this fix can't be automatically applied to stable. Signed-off-by: Chuck Lever Signed-off-by: Anna Schumaker --- include/trace/events/rpcrdma.h | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) (limited to 'include/trace/events') diff --git a/include/trace/events/rpcrdma.h b/include/trace/events/rpcrdma.h index f0678e3ac2d4..59492a93fe1d 100644 --- a/include/trace/events/rpcrdma.h +++ b/include/trace/events/rpcrdma.h @@ -470,13 +470,12 @@ TRACE_DEFINE_ENUM(rpcrdma_replych); TRACE_EVENT(xprtrdma_marshal, TP_PROTO( - const struct rpc_rqst *rqst, - unsigned int hdrlen, + const struct rpcrdma_req *req, unsigned int rtype, unsigned int wtype ), - TP_ARGS(rqst, hdrlen, rtype, wtype), + TP_ARGS(req, rtype, wtype), TP_STRUCT__entry( __field(unsigned int, task_id) @@ -491,10 +490,12 @@ TRACE_EVENT(xprtrdma_marshal, ), TP_fast_assign( + const struct rpc_rqst *rqst = &req->rl_slot; + __entry->task_id = rqst->rq_task->tk_pid; __entry->client_id = rqst->rq_task->tk_client->cl_clid; __entry->xid = be32_to_cpu(rqst->rq_xid); - __entry->hdrlen = hdrlen; + __entry->hdrlen = req->rl_hdrbuf.len; __entry->headlen = rqst->rq_snd_buf.head[0].iov_len; __entry->pagelen = rqst->rq_snd_buf.page_len; __entry->taillen = rqst->rq_snd_buf.tail[0].iov_len; -- cgit v1.2.1 From 05eb06d86685e7d9dac60e6bbb46d7f4c30b056e Mon Sep 17 00:00:00 2001 From: Chuck Lever Date: Wed, 19 Jun 2019 10:32:48 -0400 Subject: xprtrdma: Fix occasional transport deadlock Under high I/O workloads, I've noticed that an RPC/RDMA transport occasionally deadlocks (IOPS goes to zero, and doesn't recover). Diagnosis shows that the sendctx queue is empty, but when sendctxs are returned to the queue, the xprt_write_space wake-up never occurs. The wake-up logic in rpcrdma_sendctx_put_locked is racy. I noticed that both EMPTY_SCQ and XPRT_WRITE_SPACE are implemented via an atomic bit. Just one of those is sufficient. Removing EMPTY_SCQ in favor of the generic bit mechanism makes the deadlock un-reproducible. Without EMPTY_SCQ, rpcrdma_buffer::rb_flags is no longer used and is therefore removed. Unfortunately this patch does not apply cleanly to stable. If needed, someone will have to port it and test it. Fixes: 2fad659209d5 ("xprtrdma: Wait on empty sendctx queue") Signed-off-by: Chuck Lever Signed-off-by: Anna Schumaker --- include/trace/events/rpcrdma.h | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+) (limited to 'include/trace/events') diff --git a/include/trace/events/rpcrdma.h b/include/trace/events/rpcrdma.h index 59492a93fe1d..2fb415136f7b 100644 --- a/include/trace/events/rpcrdma.h +++ b/include/trace/events/rpcrdma.h @@ -539,6 +539,33 @@ TRACE_EVENT(xprtrdma_marshal_failed, ) ); +TRACE_EVENT(xprtrdma_prepsend_failed, + TP_PROTO(const struct rpc_rqst *rqst, + int ret + ), + + TP_ARGS(rqst, ret), + + TP_STRUCT__entry( + __field(unsigned int, task_id) + __field(unsigned int, client_id) + __field(u32, xid) + __field(int, ret) + ), + + TP_fast_assign( + __entry->task_id = rqst->rq_task->tk_pid; + __entry->client_id = rqst->rq_task->tk_client->cl_clid; + __entry->xid = be32_to_cpu(rqst->rq_xid); + __entry->ret = ret; + ), + + TP_printk("task:%u@%u xid=0x%08x: ret=%d", + __entry->task_id, __entry->client_id, __entry->xid, + __entry->ret + ) +); + TRACE_EVENT(xprtrdma_post_send, TP_PROTO( const struct rpcrdma_req *req, -- cgit v1.2.1 From 847568942f93e0af77e4bb8a098899f310cb3a88 Mon Sep 17 00:00:00 2001 From: Chuck Lever Date: Wed, 19 Jun 2019 10:32:59 -0400 Subject: xprtrdma: Remove fr_state Now that both the Send and Receive completions are handled in process context, it is safe to DMA unmap and return MRs to the free or recycle lists directly in the completion handlers. Doing this means rpcrdma_frwr no longer needs to track the state of each MR, meaning that a VALID or FLUSHED MR can no longer appear on an xprt's MR free list. Thus there is no longer a need to track the MR's registration state in rpcrdma_frwr. Signed-off-by: Chuck Lever Signed-off-by: Anna Schumaker --- include/trace/events/rpcrdma.h | 19 ++----------------- 1 file changed, 2 insertions(+), 17 deletions(-) (limited to 'include/trace/events') diff --git a/include/trace/events/rpcrdma.h b/include/trace/events/rpcrdma.h index 2fb415136f7b..1d2547084998 100644 --- a/include/trace/events/rpcrdma.h +++ b/include/trace/events/rpcrdma.h @@ -181,18 +181,6 @@ DECLARE_EVENT_CLASS(xprtrdma_wrch_event, ), \ TP_ARGS(task, mr, nsegs)) -TRACE_DEFINE_ENUM(FRWR_IS_INVALID); -TRACE_DEFINE_ENUM(FRWR_IS_VALID); -TRACE_DEFINE_ENUM(FRWR_FLUSHED_FR); -TRACE_DEFINE_ENUM(FRWR_FLUSHED_LI); - -#define xprtrdma_show_frwr_state(x) \ - __print_symbolic(x, \ - { FRWR_IS_INVALID, "INVALID" }, \ - { FRWR_IS_VALID, "VALID" }, \ - { FRWR_FLUSHED_FR, "FLUSHED_FR" }, \ - { FRWR_FLUSHED_LI, "FLUSHED_LI" }) - DECLARE_EVENT_CLASS(xprtrdma_frwr_done, TP_PROTO( const struct ib_wc *wc, @@ -203,22 +191,19 @@ DECLARE_EVENT_CLASS(xprtrdma_frwr_done, TP_STRUCT__entry( __field(const void *, mr) - __field(unsigned int, state) __field(unsigned int, status) __field(unsigned int, vendor_err) ), TP_fast_assign( __entry->mr = container_of(frwr, struct rpcrdma_mr, frwr); - __entry->state = frwr->fr_state; __entry->status = wc->status; __entry->vendor_err = __entry->status ? wc->vendor_err : 0; ), TP_printk( - "mr=%p state=%s: %s (%u/0x%x)", - __entry->mr, xprtrdma_show_frwr_state(__entry->state), - rdma_show_wc_status(__entry->status), + "mr=%p: %s (%u/0x%x)", + __entry->mr, rdma_show_wc_status(__entry->status), __entry->status, __entry->vendor_err ) ); -- cgit v1.2.1 From d8099feda4833bab96b1bf312e9e6aad6b771570 Mon Sep 17 00:00:00 2001 From: Chuck Lever Date: Wed, 19 Jun 2019 10:33:10 -0400 Subject: xprtrdma: Reduce context switching due to Local Invalidation Since commit ba69cd122ece ("xprtrdma: Remove support for FMR memory registration"), FRWR is the only supported memory registration mode. We can take advantage of the asynchronous nature of FRWR's LOCAL_INV Work Requests to get rid of the completion wait by having the LOCAL_INV completion handler take care of DMA unmapping MRs and waking the upper layer RPC waiter. This eliminates two context switches when local invalidation is necessary. As a side benefit, we will no longer need the per-xprt deferred completion work queue. Signed-off-by: Chuck Lever Signed-off-by: Anna Schumaker --- include/trace/events/rpcrdma.h | 1 + 1 file changed, 1 insertion(+) (limited to 'include/trace/events') diff --git a/include/trace/events/rpcrdma.h b/include/trace/events/rpcrdma.h index 1d2547084998..98023d91a72d 100644 --- a/include/trace/events/rpcrdma.h +++ b/include/trace/events/rpcrdma.h @@ -712,6 +712,7 @@ TRACE_EVENT(xprtrdma_wc_receive, DEFINE_FRWR_DONE_EVENT(xprtrdma_wc_fastreg); DEFINE_FRWR_DONE_EVENT(xprtrdma_wc_li); DEFINE_FRWR_DONE_EVENT(xprtrdma_wc_li_wake); +DEFINE_FRWR_DONE_EVENT(xprtrdma_wc_li_done); TRACE_EVENT(xprtrdma_frwr_alloc, TP_PROTO( -- cgit v1.2.1 From 675dd90ad0932f2c03912a5252458d792bd7033a Mon Sep 17 00:00:00 2001 From: Chuck Lever Date: Wed, 19 Jun 2019 10:33:42 -0400 Subject: xprtrdma: Modernize ops->connect Adapt and apply changes that were made to the TCP socket connect code. See the following commits for details on the purpose of these changes: Commit 7196dbb02ea0 ("SUNRPC: Allow changing of the TCP timeout parameters on the fly") Commit 3851f1cdb2b8 ("SUNRPC: Limit the reconnect backoff timer to the max RPC message timeout") Commit 02910177aede ("SUNRPC: Fix reconnection timeouts") Some common transport code is moved to xprt.c to satisfy the code duplication police. Signed-off-by: Chuck Lever Signed-off-by: Anna Schumaker --- include/trace/events/rpcrdma.h | 31 +++++++++++++++++++++++++++++++ 1 file changed, 31 insertions(+) (limited to 'include/trace/events') diff --git a/include/trace/events/rpcrdma.h b/include/trace/events/rpcrdma.h index 98023d91a72d..f6a4eaa85a3e 100644 --- a/include/trace/events/rpcrdma.h +++ b/include/trace/events/rpcrdma.h @@ -375,6 +375,37 @@ DEFINE_RXPRT_EVENT(xprtrdma_op_inject_dsc); DEFINE_RXPRT_EVENT(xprtrdma_op_close); DEFINE_RXPRT_EVENT(xprtrdma_op_connect); +TRACE_EVENT(xprtrdma_op_set_cto, + TP_PROTO( + const struct rpcrdma_xprt *r_xprt, + unsigned long connect, + unsigned long reconnect + ), + + TP_ARGS(r_xprt, connect, reconnect), + + TP_STRUCT__entry( + __field(const void *, r_xprt) + __field(unsigned long, connect) + __field(unsigned long, reconnect) + __string(addr, rpcrdma_addrstr(r_xprt)) + __string(port, rpcrdma_portstr(r_xprt)) + ), + + TP_fast_assign( + __entry->r_xprt = r_xprt; + __entry->connect = connect; + __entry->reconnect = reconnect; + __assign_str(addr, rpcrdma_addrstr(r_xprt)); + __assign_str(port, rpcrdma_portstr(r_xprt)); + ), + + TP_printk("peer=[%s]:%s r_xprt=%p: connect=%lu reconnect=%lu", + __get_str(addr), __get_str(port), __entry->r_xprt, + __entry->connect / HZ, __entry->reconnect / HZ + ) +); + TRACE_EVENT(xprtrdma_qp_event, TP_PROTO( const struct rpcrdma_xprt *r_xprt, -- cgit v1.2.1