From patchwork Fri Feb 12 21:07:00 2016 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Chuck Lever III X-Patchwork-Id: 8297981 Return-Path: X-Original-To: patchwork-linux-rdma@patchwork.kernel.org Delivered-To: patchwork-parsemail@patchwork1.web.kernel.org Received: from mail.kernel.org (mail.kernel.org [198.145.29.136]) by patchwork1.web.kernel.org (Postfix) with ESMTP id E69D79F372 for ; Fri, 12 Feb 2016 21:07:13 +0000 (UTC) Received: from mail.kernel.org (localhost [127.0.0.1]) by mail.kernel.org (Postfix) with ESMTP id 854282044B for ; Fri, 12 Feb 2016 21:07:12 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id EEDCA2042C for ; Fri, 12 Feb 2016 21:07:10 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751433AbcBLVHF (ORCPT ); Fri, 12 Feb 2016 16:07:05 -0500 Received: from mail-ig0-f194.google.com ([209.85.213.194]:33061 "EHLO mail-ig0-f194.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751293AbcBLVHB (ORCPT ); Fri, 12 Feb 2016 16:07:01 -0500 Received: by mail-ig0-f194.google.com with SMTP id hb3so2046676igb.0; Fri, 12 Feb 2016 13:07:01 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=sender:subject:from:to:date:message-id:in-reply-to:references :user-agent:mime-version:content-type:content-transfer-encoding; bh=XSvAmYUv8nI7KbBqdxD0amw6P9l27i/xR01G+HYzZ1o=; b=X/WQbJLasnEc/6UEqaiwaDM3kGRDmVwa776EtkCrQ+CJsTGBHCIj7knuByZaky+3Tj Qml20gLRyZbqkcK8ekiAF5FYBfsCWUnHUOuV2I7UIvkbNQtQ1KX1WygM5hqTHNJWoJha cdkqNmzR0OMsxlp/c8nqOfajcMUtriA18NdopXGZEwCbWmmhdSsw79FC0GxCMQE9BQIS gYKIblkhs4KBNI2Bt3wnMYTQQnEMfXfp7SfuS0YM2U0TJyBsD8/nfa6VYiMV0suWuBh0 Zd334kPpqFkv6jnJQPZX7KJOMldsJi5jvv5vTZ19mKm1Bpx8+916tMZbxiLVxIjnGEBq wbhg== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20130820; h=x-gm-message-state:sender:subject:from:to:date:message-id :in-reply-to:references:user-agent:mime-version:content-type :content-transfer-encoding; bh=XSvAmYUv8nI7KbBqdxD0amw6P9l27i/xR01G+HYzZ1o=; b=IFsEU0A/AUkxXEMWQ4DgKzNb0Z5DlXEaXY6W+yM65V8kC/auM1K7L9bsrwCGjhgRFJ G95oBxWhTHGPlDoovyXZJj3H9SxDucRRiSpKep416ql+/LhISohMNqQH/wRQgmWIAkEl TKSgq3Wr1Fo2x8t+U+8O4doAolzEJinkgWc54PzzfmMUse2ygzMv5zCaLXIfzQ9ElLXc A6vnMC+nxZ1/QyqEWazol0zxFDGEPNdTbgOX1u/XhH6YSHkYIF12dc5YtEjS02gpX8vF j9PyDTmvRrUpIfG3j+FaM3b7fY/y8lLx9Eurh1Awsw54atNrLvUffqHZi3kVv/I68D4r IgJg== X-Gm-Message-State: AG10YOSGNwOQUYLuQA4lflFV4zmn1cv4Tp5yLGEzuDO5RkrJMmH+rnBeE6kLdGKshBZW4w== X-Received: by 10.50.61.132 with SMTP id p4mr5389687igr.55.1455311221273; Fri, 12 Feb 2016 13:07:01 -0800 (PST) Received: from manet.1015granger.net ([2604:8800:100:81fc:ec4:7aff:fe6c:1dce]) by smtp.gmail.com with ESMTPSA id g67sm6853047ioe.34.2016.02.12.13.07.00 (version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Fri, 12 Feb 2016 13:07:00 -0800 (PST) Subject: [PATCH v1 8/8] xprtrdma: Use new CQ API for RPC-over-RDMA client send CQs From: Chuck Lever To: linux-rdma@vger.kernel.org, linux-nfs@vger.kernel.org Date: Fri, 12 Feb 2016 16:07:00 -0500 Message-ID: <20160212210700.5278.71010.stgit@manet.1015granger.net> In-Reply-To: <20160212205107.5278.55938.stgit@manet.1015granger.net> References: <20160212205107.5278.55938.stgit@manet.1015granger.net> User-Agent: StGit/0.17.1-dirty MIME-Version: 1.0 Sender: linux-rdma-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-rdma@vger.kernel.org X-Spam-Status: No, score=-6.9 required=5.0 tests=BAYES_00,DKIM_SIGNED, RCVD_IN_DNSWL_HI,RP_MATCHES_RCVD,T_DKIM_INVALID,UNPARSEABLE_RELAY autolearn=unavailable version=3.3.1 X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on mail.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP Calling ib_poll_cq() to sort through WCs during a completion is a common pattern amongst RDMA consumers. Since commit 14d3a3b2498e ("IB: add a proper completion queue abstraction"), WC sorting can be handled by the IB core. By converting to this new API, xprtrdma is made a better neighbor to other RDMA consumers, as it allows the core to schedule the delivery of completions more fairly amongst all active consumers. Because each ib_cqe carries a pointer to a completion method, the core can now post its own operations on a consumer's QP, and handle the completions itself, without changes to the consumer. Send completions were previously handled entirely in the completion upcall handler. Thus IB_POLL_SOFTIRQ is a direct replacement for the current xprtrdma send code path. Signed-off-by: Chuck Lever --- net/sunrpc/xprtrdma/frwr_ops.c | 99 +++++++++++++++++++++++++---------- net/sunrpc/xprtrdma/verbs.c | 109 ++++++++------------------------------- net/sunrpc/xprtrdma/xprt_rdma.h | 10 +--- 3 files changed, 94 insertions(+), 124 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html diff --git a/net/sunrpc/xprtrdma/frwr_ops.c b/net/sunrpc/xprtrdma/frwr_ops.c index 23382fa..3040fda 100644 --- a/net/sunrpc/xprtrdma/frwr_ops.c +++ b/net/sunrpc/xprtrdma/frwr_ops.c @@ -244,39 +244,80 @@ frwr_op_maxpages(struct rpcrdma_xprt *r_xprt) rpcrdma_max_segments(r_xprt) * ia->ri_max_frmr_depth); } -/* If FAST_REG or LOCAL_INV failed, indicate the frmr needs - * to be reset. +static void +__frwr_sendcompletion_flush(struct ib_wc *wc, struct rpcrdma_frmr *frmr, + const char *wr) +{ + frmr->fr_state = FRMR_IS_STALE; + if (wc->status != IB_WC_WR_FLUSH_ERR) + pr_err("RPC: %s: %s %s (%u, vendor %u)\n", + __func__, ib_wc_status_msg(wc->status), + wr, wc->status, wc->vendor_err); +} + +/** + * frwr_wc_fastreg - Invoked by RDMA provider for each polled FastReg WC + * @cq: completion queue (ignored) + * @wc: completed WR * - * WARNING: Only wr_id and status are reliable at this point + * FastReg WRs are typically unsignaled, but flushed WRs still need to + * be handled. */ static void -__frwr_sendcompletion_flush(struct ib_wc *wc, struct rpcrdma_mw *r) +frwr_wc_fastreg(struct ib_cq *cq, struct ib_wc *wc) { - if (likely(wc->status == IB_WC_SUCCESS)) - return; - - /* WARNING: Only wr_id and status are reliable at this point */ - r = (struct rpcrdma_mw *)(unsigned long)wc->wr_id; - if (wc->status == IB_WC_WR_FLUSH_ERR) - dprintk("RPC: %s: frmr %p flushed\n", __func__, r); - else - pr_warn("RPC: %s: frmr %p error, status %s (%d)\n", - __func__, r, ib_wc_status_msg(wc->status), wc->status); + struct rpcrdma_frmr *frmr; + struct ib_cqe *cqe; - r->frmr.fr_state = FRMR_IS_STALE; + /* WARNING: Only wr_cqe and status are reliable at this point */ + if (wc->status != IB_WC_SUCCESS) { + cqe = wc->wr_cqe; + frmr = container_of(cqe, struct rpcrdma_frmr, fr_cqe); + __frwr_sendcompletion_flush(wc, frmr, "fastreg"); + } } +/** + * frwr_wc_localinv - Invoked by RDMA provider for each polled LocalInv WC + * @cq: completion queue (ignored) + * @wc: completed WR + * + * LocalInv WRs are typically unsignaled, but flushed WRs still need to + * be handled. + */ static void -frwr_sendcompletion(struct ib_wc *wc) +frwr_wc_localinv(struct ib_cq *cq, struct ib_wc *wc) { - struct rpcrdma_mw *r = (struct rpcrdma_mw *)(unsigned long)wc->wr_id; - struct rpcrdma_frmr *f = &r->frmr; + struct rpcrdma_frmr *frmr; + struct ib_cqe *cqe; - if (unlikely(wc->status != IB_WC_SUCCESS)) - __frwr_sendcompletion_flush(wc, r); + /* WARNING: Only wr_cqe and status are reliable at this point */ + if (wc->status != IB_WC_SUCCESS) { + cqe = wc->wr_cqe; + frmr = container_of(cqe, struct rpcrdma_frmr, fr_cqe); + __frwr_sendcompletion_flush(wc, frmr, "localinv"); + } +} - if (f->fr_waiter) - complete(&f->fr_linv_done); +/** + * frwr_wc_localinv - Invoked by RDMA provider for each polled LocalInv WC + * @cq: completion queue (ignored) + * @wc: completed WR + * + * Awaken anyone waiting for an MR to finish being fenced. + */ +static void +frwr_wc_localinv_wake(struct ib_cq *cq, struct ib_wc *wc) +{ + struct rpcrdma_frmr *frmr; + struct ib_cqe *cqe; + + /* WARNING: Only wr_cqe and status are reliable at this point */ + cqe = wc->wr_cqe; + frmr = container_of(cqe, struct rpcrdma_frmr, fr_cqe); + if (wc->status != IB_WC_SUCCESS) + __frwr_sendcompletion_flush(wc, frmr, "localinv"); + complete(&frmr->fr_linv_done); } static int @@ -313,7 +354,6 @@ frwr_op_init(struct rpcrdma_xprt *r_xprt) list_add(&r->mw_list, &buf->rb_mws); list_add(&r->mw_all, &buf->rb_all); - r->mw_sendcompletion = frwr_sendcompletion; r->frmr.fr_xprt = r_xprt; } @@ -350,7 +390,6 @@ frwr_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg, } while (mw->frmr.fr_state != FRMR_IS_INVALID); frmr = &mw->frmr; frmr->fr_state = FRMR_IS_VALID; - frmr->fr_waiter = false; mr = frmr->fr_mr; reg_wr = &frmr->fr_regwr; @@ -400,7 +439,8 @@ frwr_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg, reg_wr->wr.next = NULL; reg_wr->wr.opcode = IB_WR_REG_MR; - reg_wr->wr.wr_id = (uintptr_t)mw; + frmr->fr_cqe.done = frwr_wc_fastreg; + reg_wr->wr.wr_cqe = &frmr->fr_cqe; reg_wr->wr.num_sge = 0; reg_wr->wr.send_flags = 0; reg_wr->mr = mr; @@ -437,12 +477,12 @@ __frwr_prepare_linv_wr(struct rpcrdma_mr_seg *seg) struct rpcrdma_frmr *f = &mw->frmr; struct ib_send_wr *invalidate_wr; - f->fr_waiter = false; f->fr_state = FRMR_IS_INVALID; invalidate_wr = &f->fr_invwr; memset(invalidate_wr, 0, sizeof(*invalidate_wr)); - invalidate_wr->wr_id = (unsigned long)(void *)mw; + f->fr_cqe.done = frwr_wc_localinv; + invalidate_wr->wr_cqe = &f->fr_cqe; invalidate_wr->opcode = IB_WR_LOCAL_INV; invalidate_wr->ex.invalidate_rkey = f->fr_mr->rkey; @@ -511,7 +551,7 @@ frwr_op_unmap_sync(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req) * are complete. */ f->fr_invwr.send_flags = IB_SEND_SIGNALED; - f->fr_waiter = true; + f->fr_cqe.done = frwr_wc_localinv_wake; init_completion(&f->fr_linv_done); INIT_CQCOUNT(&r_xprt->rx_ep); @@ -560,7 +600,8 @@ frwr_op_unmap(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg) invalidate_wr = &mw->frmr.fr_invwr; memset(invalidate_wr, 0, sizeof(*invalidate_wr)); - invalidate_wr->wr_id = (uintptr_t)mw; + frmr->fr_cqe.done = frwr_wc_localinv; + invalidate_wr->wr_cqe = &frmr->fr_cqe; invalidate_wr->opcode = IB_WR_LOCAL_INV; invalidate_wr->ex.invalidate_rkey = frmr->fr_mr->rkey; DECR_CQCOUNT(&r_xprt->rx_ep); diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c index 53c30e2..2f38388 100644 --- a/net/sunrpc/xprtrdma/verbs.c +++ b/net/sunrpc/xprtrdma/verbs.c @@ -112,73 +112,22 @@ rpcrdma_qp_async_error_upcall(struct ib_event *event, void *context) } } -static void -rpcrdma_cq_async_error_upcall(struct ib_event *event, void *context) -{ - struct rpcrdma_ep *ep = context; - - pr_err("RPC: %s: %s on device %s ep %p\n", - __func__, ib_event_msg(event->event), - event->device->name, context); - if (ep->rep_connected == 1) { - ep->rep_connected = -EIO; - rpcrdma_conn_func(ep); - wake_up_all(&ep->rep_connect_wait); - } -} - -static void -rpcrdma_sendcq_process_wc(struct ib_wc *wc) -{ - /* WARNING: Only wr_id and status are reliable at this point */ - if (wc->wr_id == RPCRDMA_IGNORE_COMPLETION) { - if (wc->status != IB_WC_SUCCESS && - wc->status != IB_WC_WR_FLUSH_ERR) - pr_err("RPC: %s: SEND: %s\n", - __func__, ib_wc_status_msg(wc->status)); - } else { - struct rpcrdma_mw *r; - - r = (struct rpcrdma_mw *)(unsigned long)wc->wr_id; - r->mw_sendcompletion(wc); - } -} - -/* The common case is a single send completion is waiting. By - * passing two WC entries to ib_poll_cq, a return code of 1 - * means there is exactly one WC waiting and no more. We don't - * have to invoke ib_poll_cq again to know that the CQ has been - * properly drained. - */ -static void -rpcrdma_sendcq_poll(struct ib_cq *cq) -{ - struct ib_wc *pos, wcs[2]; - int count, rc; - - do { - pos = wcs; - - rc = ib_poll_cq(cq, ARRAY_SIZE(wcs), pos); - if (rc < 0) - break; - - count = rc; - while (count-- > 0) - rpcrdma_sendcq_process_wc(pos++); - } while (rc == ARRAY_SIZE(wcs)); - return; -} - -/* Handle provider send completion upcalls. +/** + * rpcrdma_wc_send - Invoked by RDMA provider for each polled Send WC + * @cq: completion queue (ignored) + * @wc: completed WR + * + * Send WRs are typically unsignaled, but flushed WRs still need to + * be handled. */ static void -rpcrdma_sendcq_upcall(struct ib_cq *cq, void *cq_context) +rpcrdma_wc_send(struct ib_cq *cq, struct ib_wc *wc) { - do { - rpcrdma_sendcq_poll(cq); - } while (ib_req_notify_cq(cq, IB_CQ_NEXT_COMP | - IB_CQ_REPORT_MISSED_EVENTS) > 0); + /* WARNING: Only wr_cqe and status are reliable at this point */ + if (wc->status != IB_WC_SUCCESS && wc->status != IB_WC_WR_FLUSH_ERR) + pr_err("RPC: %s: Send: %s (%u, vendor %u)\n", + __func__, ib_wc_status_msg(wc->status), + wc->status, wc->vendor_err); } static void @@ -269,8 +218,6 @@ rpcrdma_flush_cqs(struct rpcrdma_ep *ep) while (ib_poll_cq(ep->rep_attr.recv_cq, 1, &wc) > 0) rpcrdma_recvcq_process_wc(&wc); - while (ib_poll_cq(ep->rep_attr.send_cq, 1, &wc) > 0) - rpcrdma_sendcq_process_wc(&wc); } static int @@ -562,9 +509,8 @@ rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia, struct rpcrdma_create_data_internal *cdata) { struct ib_cq *sendcq, *recvcq; - struct ib_cq_init_attr cq_attr = {}; unsigned int max_qp_wr; - int rc, err; + int rc; if (ia->ri_device->attrs.max_sge < RPCRDMA_MAX_IOVS) { dprintk("RPC: %s: insufficient sge's available\n", @@ -616,9 +562,9 @@ rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia, init_waitqueue_head(&ep->rep_connect_wait); INIT_DELAYED_WORK(&ep->rep_connect_worker, rpcrdma_connect_worker); - cq_attr.cqe = ep->rep_attr.cap.max_send_wr + 1; - sendcq = ib_create_cq(ia->ri_device, rpcrdma_sendcq_upcall, - rpcrdma_cq_async_error_upcall, NULL, &cq_attr); + sendcq = ib_alloc_cq(ia->ri_device, NULL, + ep->rep_attr.cap.max_send_wr + 1, + 0, IB_POLL_SOFTIRQ); if (IS_ERR(sendcq)) { rc = PTR_ERR(sendcq); dprintk("RPC: %s: failed to create send CQ: %i\n", @@ -626,13 +572,6 @@ rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia, goto out1; } - rc = ib_req_notify_cq(sendcq, IB_CQ_NEXT_COMP); - if (rc) { - dprintk("RPC: %s: ib_req_notify_cq failed: %i\n", - __func__, rc); - goto out2; - } - recvcq = ib_alloc_cq(ia->ri_device, NULL, ep->rep_attr.cap.max_recv_wr + 1, 0, IB_POLL_SOFTIRQ); @@ -667,10 +606,7 @@ rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia, return 0; out2: - err = ib_destroy_cq(sendcq); - if (err) - dprintk("RPC: %s: ib_destroy_cq returned %i\n", - __func__, err); + ib_free_cq(sendcq); out1: if (ia->ri_dma_mr) ib_dereg_mr(ia->ri_dma_mr); @@ -706,11 +642,7 @@ rpcrdma_ep_destroy(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia) } ib_free_cq(ep->rep_attr.recv_cq); - - rc = ib_destroy_cq(ep->rep_attr.send_cq); - if (rc) - dprintk("RPC: %s: ib_destroy_cq returned %i\n", - __func__, rc); + ib_free_cq(ep->rep_attr.send_cq); if (ia->ri_dma_mr) { rc = ib_dereg_mr(ia->ri_dma_mr); @@ -889,6 +821,7 @@ rpcrdma_create_req(struct rpcrdma_xprt *r_xprt) spin_lock(&buffer->rb_reqslock); list_add(&req->rl_all, &buffer->rb_allreqs); spin_unlock(&buffer->rb_reqslock); + req->rl_cqe.done = rpcrdma_wc_send; req->rl_buffer = &r_xprt->rx_buf; return req; } @@ -1252,7 +1185,7 @@ rpcrdma_ep_post(struct rpcrdma_ia *ia, } send_wr.next = NULL; - send_wr.wr_id = RPCRDMA_IGNORE_COMPLETION; + send_wr.wr_cqe = &req->rl_cqe; send_wr.sg_list = iov; send_wr.num_sge = req->rl_niovs; send_wr.opcode = IB_WR_SEND; diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h index 4be4c9b..fcfd4c0 100644 --- a/net/sunrpc/xprtrdma/xprt_rdma.h +++ b/net/sunrpc/xprtrdma/xprt_rdma.h @@ -95,10 +95,6 @@ struct rpcrdma_ep { #define INIT_CQCOUNT(ep) atomic_set(&(ep)->rep_cqcount, (ep)->rep_cqinit) #define DECR_CQCOUNT(ep) atomic_sub_return(1, &(ep)->rep_cqcount) -/* Force completion handler to ignore the signal - */ -#define RPCRDMA_IGNORE_COMPLETION (0ULL) - /* Pre-allocate extra Work Requests for handling backward receives * and sends. This is a fixed value because the Work Queues are * allocated when the forward channel is set up. @@ -205,11 +201,11 @@ struct rpcrdma_frmr { struct scatterlist *sg; int sg_nents; struct ib_mr *fr_mr; + struct ib_cqe fr_cqe; enum rpcrdma_frmr_state fr_state; + struct completion fr_linv_done; struct work_struct fr_work; struct rpcrdma_xprt *fr_xprt; - bool fr_waiter; - struct completion fr_linv_done;; union { struct ib_reg_wr fr_regwr; struct ib_send_wr fr_invwr; @@ -226,7 +222,6 @@ struct rpcrdma_mw { struct rpcrdma_fmr fmr; struct rpcrdma_frmr frmr; }; - void (*mw_sendcompletion)(struct ib_wc *); struct list_head mw_list; struct list_head mw_all; }; @@ -283,6 +278,7 @@ struct rpcrdma_req { struct rpcrdma_mr_seg rl_segments[RPCRDMA_MAX_SEGS]; struct rpc_task *rl_task; + struct ib_cqe rl_cqe; struct list_head rl_all; bool rl_backchannel; };