From patchwork Tue Oct 5 14:17:59 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Chuck Lever III X-Patchwork-Id: 12536897 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 0F4CAC433FE for ; Tue, 5 Oct 2021 14:18:03 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id DF93161154 for ; Tue, 5 Oct 2021 14:18:02 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S235325AbhJEOTw (ORCPT ); Tue, 5 Oct 2021 10:19:52 -0400 Received: from mail.kernel.org ([198.145.29.99]:50070 "EHLO mail.kernel.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S235294AbhJEOTv (ORCPT ); Tue, 5 Oct 2021 10:19:51 -0400 Received: by mail.kernel.org (Postfix) with ESMTPSA id 8DBE260F4B; Tue, 5 Oct 2021 14:18:00 +0000 (UTC) Subject: [PATCH v3 1/2] xprtrdma: Provide a buffer to pad Write chunks of unaligned length From: Chuck Lever To: anna.schumaker@netapp.com, trondmy@hammerspace.com Cc: linux-nfs@vger.kernel.org, linux-rdma@vger.kernel.org Date: Tue, 05 Oct 2021 10:17:59 -0400 Message-ID: <163344347986.1933.1092012850713935427.stgit@morisot.1015granger.net> In-Reply-To: <163344340514.1933.10783386394620857061.stgit@morisot.1015granger.net> References: <163344340514.1933.10783386394620857061.stgit@morisot.1015granger.net> User-Agent: StGit/1.3 MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-rdma@vger.kernel.org This is a buffer to be left persistently registered while a connection is up. Connection tear-down will automatically DMA-unmap, invalidate, and dereg the MR. A persistently registered buffer is lower in cost to provide, and it can never be coalesced into the RDMA segment that carries the data payload. An RPC that provisions a Write chunk with a non-aligned length now uses this MR rather than the tail buffer of the RPC's rq_rcv_buf. Reviewed-By: Tom Talpey Signed-off-by: Chuck Lever --- include/trace/events/rpcrdma.h | 13 ++++++++++--- net/sunrpc/xprtrdma/frwr_ops.c | 35 +++++++++++++++++++++++++++++++++++ net/sunrpc/xprtrdma/rpc_rdma.c | 23 ++++++++++++++--------- net/sunrpc/xprtrdma/verbs.c | 1 + net/sunrpc/xprtrdma/xprt_rdma.h | 5 +++++ 5 files changed, 65 insertions(+), 12 deletions(-) diff --git a/include/trace/events/rpcrdma.h b/include/trace/events/rpcrdma.h index de4195499592..afb2e394797c 100644 --- a/include/trace/events/rpcrdma.h +++ b/include/trace/events/rpcrdma.h @@ -375,10 +375,16 @@ DECLARE_EVENT_CLASS(xprtrdma_mr_class, TP_fast_assign( const struct rpcrdma_req *req = mr->mr_req; - const struct rpc_task *task = req->rl_slot.rq_task; - __entry->task_id = task->tk_pid; - __entry->client_id = task->tk_client->cl_clid; + if (req) { + const struct rpc_task *task = req->rl_slot.rq_task; + + __entry->task_id = task->tk_pid; + __entry->client_id = task->tk_client->cl_clid; + } else { + __entry->task_id = 0; + __entry->client_id = -1; + } __entry->mr_id = mr->mr_ibmr->res.id; __entry->nents = mr->mr_nents; __entry->handle = mr->mr_handle; @@ -639,6 +645,7 @@ TRACE_EVENT(xprtrdma_nomrs_err, DEFINE_RDCH_EVENT(read); DEFINE_WRCH_EVENT(write); DEFINE_WRCH_EVENT(reply); +DEFINE_WRCH_EVENT(wp); TRACE_DEFINE_ENUM(rpcrdma_noch); TRACE_DEFINE_ENUM(rpcrdma_noch_pullup); diff --git a/net/sunrpc/xprtrdma/frwr_ops.c b/net/sunrpc/xprtrdma/frwr_ops.c index f700b34a5bfd..3eccf365fcb8 100644 --- a/net/sunrpc/xprtrdma/frwr_ops.c +++ b/net/sunrpc/xprtrdma/frwr_ops.c @@ -666,3 +666,38 @@ void frwr_unmap_async(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req) */ rpcrdma_force_disconnect(ep); } + +/** + * frwr_wp_create - Create an MR for padding Write chunks + * @r_xprt: transport resources to use + * + * Return 0 on success, negative errno on failure. + */ +int frwr_wp_create(struct rpcrdma_xprt *r_xprt) +{ + struct rpcrdma_ep *ep = r_xprt->rx_ep; + struct rpcrdma_mr_seg seg; + struct rpcrdma_mr *mr; + + mr = rpcrdma_mr_get(r_xprt); + if (!mr) + return -EAGAIN; + mr->mr_req = NULL; + ep->re_write_pad_mr = mr; + + seg.mr_len = XDR_UNIT; + seg.mr_page = virt_to_page(ep->re_write_pad); + seg.mr_offset = offset_in_page(ep->re_write_pad); + if (IS_ERR(frwr_map(r_xprt, &seg, 1, true, xdr_zero, mr))) + return -EIO; + trace_xprtrdma_mr_fastreg(mr); + + mr->mr_cqe.done = frwr_wc_fastreg; + mr->mr_regwr.wr.next = NULL; + mr->mr_regwr.wr.wr_cqe = &mr->mr_cqe; + mr->mr_regwr.wr.num_sge = 0; + mr->mr_regwr.wr.opcode = IB_WR_REG_MR; + mr->mr_regwr.wr.send_flags = 0; + + return ib_post_send(ep->re_id->qp, &mr->mr_regwr.wr, NULL); +} diff --git a/net/sunrpc/xprtrdma/rpc_rdma.c b/net/sunrpc/xprtrdma/rpc_rdma.c index c335c1361564..8035a983c8ce 100644 --- a/net/sunrpc/xprtrdma/rpc_rdma.c +++ b/net/sunrpc/xprtrdma/rpc_rdma.c @@ -255,15 +255,7 @@ rpcrdma_convert_iovs(struct rpcrdma_xprt *r_xprt, struct xdr_buf *xdrbuf, page_base = 0; } - if (type == rpcrdma_readch) - goto out; - - /* When encoding a Write chunk, some servers need to see an - * extra segment for non-XDR-aligned Write chunks. The upper - * layer provides space in the tail iovec that may be used - * for this purpose. - */ - if (type == rpcrdma_writech && r_xprt->rx_ep->re_implicit_roundup) + if (type == rpcrdma_readch || type == rpcrdma_writech) goto out; if (xdrbuf->tail[0].iov_len) @@ -405,6 +397,7 @@ static int rpcrdma_encode_write_list(struct rpcrdma_xprt *r_xprt, enum rpcrdma_chunktype wtype) { struct xdr_stream *xdr = &req->rl_stream; + struct rpcrdma_ep *ep = r_xprt->rx_ep; struct rpcrdma_mr_seg *seg; struct rpcrdma_mr *mr; int nsegs, nchunks; @@ -443,6 +436,18 @@ static int rpcrdma_encode_write_list(struct rpcrdma_xprt *r_xprt, nsegs -= mr->mr_nents; } while (nsegs); + if (xdr_pad_size(rqst->rq_rcv_buf.page_len)) { + if (encode_rdma_segment(xdr, ep->re_write_pad_mr) < 0) + return -EMSGSIZE; + + trace_xprtrdma_chunk_wp(rqst->rq_task, ep->re_write_pad_mr, + nsegs); + r_xprt->rx_stats.write_chunk_count++; + r_xprt->rx_stats.total_rdma_request += mr->mr_length; + nchunks++; + nsegs -= mr->mr_nents; + } + /* Update count of segments in this Write chunk */ *segcount = cpu_to_be32(nchunks); diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c index aaec3c9be8db..c3784b7b6855 100644 --- a/net/sunrpc/xprtrdma/verbs.c +++ b/net/sunrpc/xprtrdma/verbs.c @@ -551,6 +551,7 @@ int rpcrdma_xprt_connect(struct rpcrdma_xprt *r_xprt) goto out; } rpcrdma_mrs_create(r_xprt); + frwr_wp_create(r_xprt); out: trace_xprtrdma_connect(r_xprt, rc); diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h index d91f54eae00b..b6d8b3e6356c 100644 --- a/net/sunrpc/xprtrdma/xprt_rdma.h +++ b/net/sunrpc/xprtrdma/xprt_rdma.h @@ -68,12 +68,14 @@ /* * RDMA Endpoint -- connection endpoint details */ +struct rpcrdma_mr; struct rpcrdma_ep { struct kref re_kref; struct rdma_cm_id *re_id; struct ib_pd *re_pd; unsigned int re_max_rdma_segs; unsigned int re_max_fr_depth; + struct rpcrdma_mr *re_write_pad_mr; bool re_implicit_roundup; enum ib_mr_type re_mrtype; struct completion re_done; @@ -97,6 +99,8 @@ struct rpcrdma_ep { unsigned int re_inline_recv; /* negotiated */ atomic_t re_completion_ids; + + char re_write_pad[XDR_UNIT]; }; /* Pre-allocate extra Work Requests for handling reverse-direction @@ -535,6 +539,7 @@ int frwr_send(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req); void frwr_reminv(struct rpcrdma_rep *rep, struct list_head *mrs); void frwr_unmap_sync(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req); void frwr_unmap_async(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req); +int frwr_wp_create(struct rpcrdma_xprt *r_xprt); /* * RPC/RDMA protocol calls - xprtrdma/rpc_rdma.c From patchwork Tue Oct 5 14:18:06 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Chuck Lever III X-Patchwork-Id: 12536899 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 9FFF9C433EF for ; Tue, 5 Oct 2021 14:18:10 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 8695C6115B for ; Tue, 5 Oct 2021 14:18:10 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S235480AbhJEOT6 (ORCPT ); Tue, 5 Oct 2021 10:19:58 -0400 Received: from mail.kernel.org ([198.145.29.99]:50114 "EHLO mail.kernel.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S235294AbhJEOT5 (ORCPT ); Tue, 5 Oct 2021 10:19:57 -0400 Received: by mail.kernel.org (Postfix) with ESMTPSA id B77176115B; Tue, 5 Oct 2021 14:18:06 +0000 (UTC) Subject: [PATCH v3 2/2] xprtrdma: Remove rpcrdma_ep::re_implicit_roundup From: Chuck Lever To: anna.schumaker@netapp.com, trondmy@hammerspace.com Cc: linux-nfs@vger.kernel.org, linux-rdma@vger.kernel.org Date: Tue, 05 Oct 2021 10:18:06 -0400 Message-ID: <163344348604.1933.17126543709686424511.stgit@morisot.1015granger.net> In-Reply-To: <163344340514.1933.10783386394620857061.stgit@morisot.1015granger.net> References: <163344340514.1933.10783386394620857061.stgit@morisot.1015granger.net> User-Agent: StGit/1.3 MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-rdma@vger.kernel.org Clean up: this field is no longer used. xprt_rdma_pad_optimize is also no longer used, but is left in place because it is part of the kernel/userspace API. Signed-off-by: Chuck Lever --- net/sunrpc/xprtrdma/verbs.c | 2 -- net/sunrpc/xprtrdma/xprt_rdma.h | 1 - 2 files changed, 3 deletions(-) diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c index c3784b7b6855..3d3673ba9e1e 100644 --- a/net/sunrpc/xprtrdma/verbs.c +++ b/net/sunrpc/xprtrdma/verbs.c @@ -205,14 +205,12 @@ static void rpcrdma_update_cm_private(struct rpcrdma_ep *ep, unsigned int rsize, wsize; /* Default settings for RPC-over-RDMA Version One */ - ep->re_implicit_roundup = xprt_rdma_pad_optimize; rsize = RPCRDMA_V1_DEF_INLINE_SIZE; wsize = RPCRDMA_V1_DEF_INLINE_SIZE; if (pmsg && pmsg->cp_magic == rpcrdma_cmp_magic && pmsg->cp_version == RPCRDMA_CMP_VERSION) { - ep->re_implicit_roundup = true; rsize = rpcrdma_decode_buffer_size(pmsg->cp_send_size); wsize = rpcrdma_decode_buffer_size(pmsg->cp_recv_size); } diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h index b6d8b3e6356c..c79f92eeda76 100644 --- a/net/sunrpc/xprtrdma/xprt_rdma.h +++ b/net/sunrpc/xprtrdma/xprt_rdma.h @@ -76,7 +76,6 @@ struct rpcrdma_ep { unsigned int re_max_rdma_segs; unsigned int re_max_fr_depth; struct rpcrdma_mr *re_write_pad_mr; - bool re_implicit_roundup; enum ib_mr_type re_mrtype; struct completion re_done; unsigned int re_send_count;