From cb586decbb88fcd068116af2d4e1e3e2e86978d6 Mon Sep 17 00:00:00 2001 From: Chuck Lever Date: Fri, 3 Jan 2020 11:56:32 -0500 Subject: [PATCH] xprtrdma: Make sendctx queue lifetime the same as connection lifetime The size of the sendctx queue depends on the value stored in ia->ri_max_send_sges. This value is determined by querying the underlying device. Eventually, rpcrdma_ia_open() and rpcrdma_ep_create() will be called in the connect worker rather than at transport set-up time. The underlying device will not have been chosen device set-up time. The sendctx queue will thus have to be created after the underlying device has been chosen via address and route resolution; in other words, in the connect worker. Signed-off-by: Chuck Lever Signed-off-by: Anna Schumaker --- include/trace/events/rpcrdma.h | 12 ++++++++---- net/sunrpc/xprtrdma/verbs.c | 22 +++++++++++++++------- 2 files changed, 23 insertions(+), 11 deletions(-) diff --git a/include/trace/events/rpcrdma.h b/include/trace/events/rpcrdma.h index 18790582d2a5..c0e4c93324f5 100644 --- a/include/trace/events/rpcrdma.h +++ b/include/trace/events/rpcrdma.h @@ -729,6 +729,7 @@ TRACE_EVENT(xprtrdma_post_send, TP_STRUCT__entry( __field(const void *, req) + __field(const void *, sc) __field(unsigned int, task_id) __field(unsigned int, client_id) __field(int, num_sge) @@ -743,14 +744,15 @@ TRACE_EVENT(xprtrdma_post_send, __entry->client_id = rqst->rq_task->tk_client ? rqst->rq_task->tk_client->cl_clid : -1; __entry->req = req; + __entry->sc = req->rl_sendctx; __entry->num_sge = req->rl_wr.num_sge; __entry->signaled = req->rl_wr.send_flags & IB_SEND_SIGNALED; __entry->status = status; ), - TP_printk("task:%u@%u req=%p (%d SGE%s) %sstatus=%d", + TP_printk("task:%u@%u req=%p sc=%p (%d SGE%s) %sstatus=%d", __entry->task_id, __entry->client_id, - __entry->req, __entry->num_sge, + __entry->req, __entry->sc, __entry->num_sge, (__entry->num_sge == 1 ? "" : "s"), (__entry->signaled ? "signaled " : ""), __entry->status @@ -849,6 +851,7 @@ TRACE_EVENT(xprtrdma_wc_send, TP_STRUCT__entry( __field(const void *, req) + __field(const void *, sc) __field(unsigned int, unmap_count) __field(unsigned int, status) __field(unsigned int, vendor_err) @@ -856,13 +859,14 @@ TRACE_EVENT(xprtrdma_wc_send, TP_fast_assign( __entry->req = sc->sc_req; + __entry->sc = sc; __entry->unmap_count = sc->sc_unmap_count; __entry->status = wc->status; __entry->vendor_err = __entry->status ? wc->vendor_err : 0; ), - TP_printk("req=%p, unmapped %u pages: %s (%u/0x%x)", - __entry->req, __entry->unmap_count, + TP_printk("req=%p sc=%p unmapped=%u: %s (%u/0x%x)", + __entry->req, __entry->sc, __entry->unmap_count, rdma_show_wc_status(__entry->status), __entry->status, __entry->vendor_err ) diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c index 945603662419..b6aba0c85998 100644 --- a/net/sunrpc/xprtrdma/verbs.c +++ b/net/sunrpc/xprtrdma/verbs.c @@ -74,6 +74,8 @@ /* * internal functions */ +static int rpcrdma_sendctxs_create(struct rpcrdma_xprt *r_xprt); +static void rpcrdma_sendctxs_destroy(struct rpcrdma_xprt *r_xprt); static void rpcrdma_sendctx_put_locked(struct rpcrdma_xprt *r_xprt, struct rpcrdma_sendctx *sc); static void rpcrdma_reqs_reset(struct rpcrdma_xprt *r_xprt); @@ -428,6 +430,7 @@ rpcrdma_ia_remove(struct rpcrdma_ia *ia) rpcrdma_regbuf_dma_unmap(req->rl_recvbuf); } rpcrdma_mrs_destroy(r_xprt); + rpcrdma_sendctxs_destroy(r_xprt); ib_dealloc_pd(ia->ri_pd); ia->ri_pd = NULL; @@ -705,6 +708,10 @@ retry: rpcrdma_reset_cwnd(r_xprt); rpcrdma_post_recvs(r_xprt, true); + rc = rpcrdma_sendctxs_create(r_xprt); + if (rc) + goto out; + rc = rdma_connect(ia->ri_id, &ep->rep_remote_cma); if (rc) goto out; @@ -757,6 +764,7 @@ rpcrdma_ep_disconnect(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia) rpcrdma_xprt_drain(r_xprt); rpcrdma_reqs_reset(r_xprt); rpcrdma_mrs_destroy(r_xprt); + rpcrdma_sendctxs_destroy(r_xprt); } /* Fixed-size circular FIFO queue. This implementation is wait-free and @@ -776,13 +784,17 @@ rpcrdma_ep_disconnect(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia) * queue activity, and rpcrdma_xprt_drain has flushed all remaining * Send requests. */ -static void rpcrdma_sendctxs_destroy(struct rpcrdma_buffer *buf) +static void rpcrdma_sendctxs_destroy(struct rpcrdma_xprt *r_xprt) { + struct rpcrdma_buffer *buf = &r_xprt->rx_buf; unsigned long i; + if (!buf->rb_sc_ctxs) + return; for (i = 0; i <= buf->rb_sc_last; i++) kfree(buf->rb_sc_ctxs[i]); kfree(buf->rb_sc_ctxs); + buf->rb_sc_ctxs = NULL; } static struct rpcrdma_sendctx *rpcrdma_sendctx_create(struct rpcrdma_ep *ep) @@ -810,7 +822,6 @@ static int rpcrdma_sendctxs_create(struct rpcrdma_xprt *r_xprt) * Sends are posted. */ i = buf->rb_max_requests + RPCRDMA_MAX_BC_REQUESTS; - dprintk("RPC: %s: allocating %lu send_ctxs\n", __func__, i); buf->rb_sc_ctxs = kcalloc(i, sizeof(sc), GFP_KERNEL); if (!buf->rb_sc_ctxs) return -ENOMEM; @@ -824,6 +835,8 @@ static int rpcrdma_sendctxs_create(struct rpcrdma_xprt *r_xprt) buf->rb_sc_ctxs[i] = sc; } + buf->rb_sc_head = 0; + buf->rb_sc_tail = 0; return 0; } @@ -1166,10 +1179,6 @@ int rpcrdma_buffer_create(struct rpcrdma_xprt *r_xprt) init_llist_head(&buf->rb_free_reps); - rc = rpcrdma_sendctxs_create(r_xprt); - if (rc) - goto out; - return 0; out: rpcrdma_buffer_destroy(buf); @@ -1245,7 +1254,6 @@ static void rpcrdma_mrs_destroy(struct rpcrdma_xprt *r_xprt) void rpcrdma_buffer_destroy(struct rpcrdma_buffer *buf) { - rpcrdma_sendctxs_destroy(buf); rpcrdma_reps_destroy(buf); while (!list_empty(&buf->rb_send_bufs)) {