From patchwork Fri May 25 16:01:21 2018 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Devesh Sharma X-Patchwork-Id: 10427891 X-Patchwork-Delegate: jgg@ziepe.ca Return-Path: Received: from mail.wl.linuxfoundation.org (pdx-wl-mail.web.codeaurora.org [172.30.200.125]) by pdx-korg-patchwork.web.codeaurora.org (Postfix) with ESMTP id 9D833601D5 for ; Fri, 25 May 2018 16:01:47 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 84D3528A25 for ; Fri, 25 May 2018 16:01:47 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id 79B372974D; Fri, 25 May 2018 16:01:47 +0000 (UTC) X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on pdx-wl-mail.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-7.8 required=2.0 tests=BAYES_00,DKIM_SIGNED, MAILING_LIST_MULTI, RCVD_IN_DNSWL_HI, T_DKIM_INVALID autolearn=ham version=3.3.1 Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 8EF2B28A25 for ; Fri, 25 May 2018 16:01:46 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S965666AbeEYQBp (ORCPT ); Fri, 25 May 2018 12:01:45 -0400 Received: from mail-wm0-f67.google.com ([74.125.82.67]:53369 "EHLO mail-wm0-f67.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S966599AbeEYQBo (ORCPT ); Fri, 25 May 2018 12:01:44 -0400 Received: by mail-wm0-f67.google.com with SMTP id a67-v6so15646171wmf.3 for ; Fri, 25 May 2018 09:01:43 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=broadcom.com; s=google; h=from:to:cc:subject:date:message-id:in-reply-to:references; bh=fFj9ZMrhMTc8V+iCSsL4Li+UzARqAYXFyoE3b6ekUFI=; b=h2KycujHJAza1EiMuLR3HaHaPzCfVEX56X4pGYGjYVF/EwGzBa/xUk+FSczQBBQrcd nIgAO13DrhdhtVPrBCdCZh+9ULx94+XlthaX7VIUyrXUYUeKTnvZt2/HdlNClP8f8YU3 v1pR20kZjqM4KswAVRKUICRd9J+KsEbOjfyuI= X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references; bh=fFj9ZMrhMTc8V+iCSsL4Li+UzARqAYXFyoE3b6ekUFI=; b=BnXDmCuo7QeufBZLFb9INWVArEGFUpnBlO3fZYXld02/D8bUJG4xsYXmMKCd3r4XBG OrrETeYYMPKohqW2BJicuvnPh3nOpVer/Aik9LvSj/qiCvRGMsTEMs8d1i6hkcBuWX2B c3BhFfmO9d2BhFH/CSTe/RBB7yMgeP/rr/H5dtSIaw2Y4Xhf2rkjMOkhyM6KjXXhvqov 9KzpcRcxCOvV3+MsQyBAQhIqnfvsNL/RTu4hfBl7YYRocEvqUt6GkVk4HdZL/LUR2ZMe lpsuvBuLRJ6X8A7hcCmI0dAsUq2syRh5ZdCEnCAa1mK86UhyWzyAO+sTJ/4046gQsHLf 6nyQ== X-Gm-Message-State: ALKqPwdn4GyQHsQUXDdEXhJqXIcLyUpwPOUxgCMWw8NMNtNLSufzwN9K VfUf0KssCy/Y1qLEwukp4CNMiQ== X-Google-Smtp-Source: AB8JxZqjXBHEO2oEs0X5iDhsvHsAK9lqunTVrMAlKxTLmiUCQgBcPjB8KJ5oqQjY4zBAqT75k/YFHQ== X-Received: by 2002:a50:b585:: with SMTP id a5-v6mr3557904ede.136.1527264103068; Fri, 25 May 2018 09:01:43 -0700 (PDT) Received: from neo00-el73.dhcp.broadcom.net ([192.19.234.250]) by smtp.gmail.com with ESMTPSA id j5-v6sm13319603edd.37.2018.05.25.09.01.38 (version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Fri, 25 May 2018 09:01:42 -0700 (PDT) From: Devesh Sharma To: dledford@redhat.com, jgg@mellanox.com Cc: linux-rdma@vger.kernel.org, jtoppins@redhat.com, ddutile@redhat.com, Devesh Sharma Subject: [for-rc v3] RDMA/bnxt_re: Fix broken RoCE driver due to recent L2 driver changes Date: Fri, 25 May 2018 12:01:21 -0400 Message-Id: <1527264081-4451-2-git-send-email-devesh.sharma@broadcom.com> X-Mailer: git-send-email 1.8.3.1 In-Reply-To: <1527264081-4451-1-git-send-email-devesh.sharma@broadcom.com> References: <1527264081-4451-1-git-send-email-devesh.sharma@broadcom.com> Sender: linux-rdma-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-rdma@vger.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP The recent changes in Broadcom's ethernet driver(L2 driver) broke RoCE functionality in terms of MSIx vector allocation and de-allocation. There is a possibility that L2 driver would initiate MSIx vector reallocation depending upon the requests coming from administrator. In such cases L2 driver needs to free up all the MSIx vectors allocated previously and reallocate/initialize those. If RoCE driver is loaded and reshuffling is attempted, there will be kernel crashes because RoCE driver would still be holding the MSIx vectors but L2 driver would attempt to free in-use vectors. Thus leading to a kernel crash. Making changes in roce driver to fix crashes described above. As part of solution L2 driver tells RoCE driver to release the MSIx vector whenever there is a need. When RoCE driver get message it sync up with all the running tasklets and IRQ handlers and releases the vectors. L2 driver send one more message to RoCE driver to resume the MSIx vectors. L2 driver guarantees that RoCE vector do not change during reshuffling. Fixes: ec86f14ea506 ("bnxt_en: Add ULP calls to stop and restart IRQs.") Fixes: 08654eb213a8 ("bnxt_en: Change IRQ assignment for RDMA driver.") Signed-off-by: Devesh Sharma --- drivers/infiniband/hw/bnxt_re/main.c | 55 ++++++++++++++++- drivers/infiniband/hw/bnxt_re/qplib_fp.c | 94 +++++++++++++++++++----------- drivers/infiniband/hw/bnxt_re/qplib_fp.h | 3 + drivers/infiniband/hw/bnxt_re/qplib_rcfw.c | 61 +++++++++++++------ drivers/infiniband/hw/bnxt_re/qplib_rcfw.h | 3 + 5 files changed, 163 insertions(+), 53 deletions(-) diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c index f6c739e..20b9f31 100644 --- a/drivers/infiniband/hw/bnxt_re/main.c +++ b/drivers/infiniband/hw/bnxt_re/main.c @@ -185,12 +185,65 @@ static void bnxt_re_shutdown(void *p) bnxt_re_ib_unreg(rdev, false); } +static void bnxt_re_stop_irq(void *handle) +{ + struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle; + struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw; + struct bnxt_qplib_nq *nq; + int indx; + + for (indx = BNXT_RE_NQ_IDX; indx < rdev->num_msix; indx++) { + nq = &rdev->nq[indx - 1]; + bnxt_qplib_nq_stop_irq(nq, false); + } + + bnxt_qplib_rcfw_stop_irq(rcfw, false); +} + +static void bnxt_re_start_irq(void *handle, struct bnxt_msix_entry *ent) +{ + struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle; + struct bnxt_msix_entry *msix_ent = rdev->msix_entries; + struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw; + struct bnxt_qplib_nq *nq; + int indx, rc; + + if (!ent) { + /* Not setting the f/w timeout bit in rcfw. + * During the driver unload the first command + * to f/w will timeout and that will set the + * timeout bit. + */ + dev_err(rdev_to_dev(rdev), "Failed to re-start IRQs\n"); + return; + } + + /* Vectors may change after restart, so update with new vectors + * in device sctructure. + */ + for (indx = 0; indx < rdev->num_msix; indx++) + rdev->msix_entries[indx].vector = ent[indx].vector; + + bnxt_qplib_rcfw_start_irq(rcfw, msix_ent[BNXT_RE_AEQ_IDX].vector, + false); + for (indx = BNXT_RE_NQ_IDX ; indx < rdev->num_msix; indx++) { + nq = &rdev->nq[indx - 1]; + rc = bnxt_qplib_nq_start_irq(nq, indx - 1, + msix_ent[indx].vector, false); + if (rc) + dev_warn(rdev_to_dev(rdev), + "Failed to reinit NQ index %d\n", indx - 1); + } +} + static struct bnxt_ulp_ops bnxt_re_ulp_ops = { .ulp_async_notifier = NULL, .ulp_stop = bnxt_re_stop, .ulp_start = bnxt_re_start, .ulp_sriov_config = bnxt_re_sriov_config, - .ulp_shutdown = bnxt_re_shutdown + .ulp_shutdown = bnxt_re_shutdown, + .ulp_irq_stop = bnxt_re_stop_irq, + .ulp_irq_restart = bnxt_re_start_irq }; /* RoCE -> Net driver */ diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.c b/drivers/infiniband/hw/bnxt_re/qplib_fp.c index 3a78fab..50d8f1f 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_fp.c +++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.c @@ -336,22 +336,32 @@ static irqreturn_t bnxt_qplib_nq_irq(int irq, void *dev_instance) return IRQ_HANDLED; } +void bnxt_qplib_nq_stop_irq(struct bnxt_qplib_nq *nq, bool kill) +{ + tasklet_disable(&nq->worker); + /* Mask h/w interrupt */ + NQ_DB(nq->bar_reg_iomem, nq->hwq.cons, nq->hwq.max_elements); + /* Sync with last running IRQ handler */ + synchronize_irq(nq->vector); + if (kill) + tasklet_kill(&nq->worker); + if (nq->requested) { + irq_set_affinity_hint(nq->vector, NULL); + free_irq(nq->vector, nq); + nq->requested = false; + } +} + void bnxt_qplib_disable_nq(struct bnxt_qplib_nq *nq) { if (nq->cqn_wq) { destroy_workqueue(nq->cqn_wq); nq->cqn_wq = NULL; } + /* Make sure the HW is stopped! */ - synchronize_irq(nq->vector); - tasklet_disable(&nq->worker); - tasklet_kill(&nq->worker); + bnxt_qplib_nq_stop_irq(nq, true); - if (nq->requested) { - irq_set_affinity_hint(nq->vector, NULL); - free_irq(nq->vector, nq); - nq->requested = false; - } if (nq->bar_reg_iomem) iounmap(nq->bar_reg_iomem); nq->bar_reg_iomem = NULL; @@ -361,6 +371,40 @@ void bnxt_qplib_disable_nq(struct bnxt_qplib_nq *nq) nq->vector = 0; } +int bnxt_qplib_nq_start_irq(struct bnxt_qplib_nq *nq, int nq_indx, + int msix_vector, bool need_init) +{ + int rc; + + if (nq->requested) + return -EFAULT; + + nq->vector = msix_vector; + if (need_init) + tasklet_init(&nq->worker, bnxt_qplib_service_nq, + (unsigned long)nq); + else + tasklet_enable(&nq->worker); + + snprintf(nq->name, sizeof(nq->name), "bnxt_qplib_nq-%d", nq_indx); + rc = request_irq(nq->vector, bnxt_qplib_nq_irq, 0, nq->name, nq); + if (rc) + return rc; + + cpumask_clear(&nq->mask); + cpumask_set_cpu(nq_indx, &nq->mask); + rc = irq_set_affinity_hint(nq->vector, &nq->mask); + if (rc) { + dev_warn(&nq->pdev->dev, + "QPLIB: set affinity failed; vector: %d nq_idx: %d\n", + nq->vector, nq_indx); + } + nq->requested = true; + NQ_DB_REARM(nq->bar_reg_iomem, nq->hwq.cons, nq->hwq.max_elements); + + return rc; +} + int bnxt_qplib_enable_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq, int nq_idx, int msix_vector, int bar_reg_offset, int (*cqn_handler)(struct bnxt_qplib_nq *nq, @@ -372,41 +416,17 @@ int bnxt_qplib_enable_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq, resource_size_t nq_base; int rc = -1; - nq->pdev = pdev; - nq->vector = msix_vector; if (cqn_handler) nq->cqn_handler = cqn_handler; if (srqn_handler) nq->srqn_handler = srqn_handler; - tasklet_init(&nq->worker, bnxt_qplib_service_nq, (unsigned long)nq); - /* Have a task to schedule CQ notifiers in post send case */ nq->cqn_wq = create_singlethread_workqueue("bnxt_qplib_nq"); if (!nq->cqn_wq) - goto fail; - - nq->requested = false; - memset(nq->name, 0, 32); - sprintf(nq->name, "bnxt_qplib_nq-%d", nq_idx); - rc = request_irq(nq->vector, bnxt_qplib_nq_irq, 0, nq->name, nq); - if (rc) { - dev_err(&nq->pdev->dev, - "Failed to request IRQ for NQ: %#x", rc); - goto fail; - } - - cpumask_clear(&nq->mask); - cpumask_set_cpu(nq_idx, &nq->mask); - rc = irq_set_affinity_hint(nq->vector, &nq->mask); - if (rc) { - dev_warn(&nq->pdev->dev, - "QPLIB: set affinity failed; vector: %d nq_idx: %d\n", - nq->vector, nq_idx); - } + return -ENOMEM; - nq->requested = true; nq->bar_reg = NQ_CONS_PCI_BAR_REGION; nq->bar_reg_off = bar_reg_offset; nq_base = pci_resource_start(pdev, nq->bar_reg); @@ -419,7 +439,13 @@ int bnxt_qplib_enable_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq, rc = -ENOMEM; goto fail; } - NQ_DB_REARM(nq->bar_reg_iomem, nq->hwq.cons, nq->hwq.max_elements); + + rc = bnxt_qplib_nq_start_irq(nq, nq_idx, msix_vector, true); + if (rc) { + dev_err(&nq->pdev->dev, + "QPLIB: Failed to request irq for nq-idx %d", nq_idx); + goto fail; + } return 0; fail: diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.h b/drivers/infiniband/hw/bnxt_re/qplib_fp.h index ade9f13..72352ca 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_fp.h +++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.h @@ -467,7 +467,10 @@ struct bnxt_qplib_nq_work { struct bnxt_qplib_cq *cq; }; +void bnxt_qplib_nq_stop_irq(struct bnxt_qplib_nq *nq, bool kill); void bnxt_qplib_disable_nq(struct bnxt_qplib_nq *nq); +int bnxt_qplib_nq_start_irq(struct bnxt_qplib_nq *nq, int nq_indx, + int msix_vector, bool need_init); int bnxt_qplib_enable_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq, int nq_idx, int msix_vector, int bar_reg_offset, int (*cqn_handler)(struct bnxt_qplib_nq *nq, diff --git a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c index 80027a4..2852d35 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c +++ b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c @@ -582,19 +582,29 @@ int bnxt_qplib_alloc_rcfw_channel(struct pci_dev *pdev, return -ENOMEM; } -void bnxt_qplib_disable_rcfw_channel(struct bnxt_qplib_rcfw *rcfw) +void bnxt_qplib_rcfw_stop_irq(struct bnxt_qplib_rcfw *rcfw, bool kill) { - unsigned long indx; - - /* Make sure the HW channel is stopped! */ - synchronize_irq(rcfw->vector); tasklet_disable(&rcfw->worker); - tasklet_kill(&rcfw->worker); + /* Mask h/w interrupts */ + CREQ_DB(rcfw->creq_bar_reg_iomem, rcfw->creq.cons, + rcfw->creq.max_elements); + /* Sync with last running IRQ-handler */ + synchronize_irq(rcfw->vector); + if (kill) + tasklet_kill(&rcfw->worker); if (rcfw->requested) { free_irq(rcfw->vector, rcfw); rcfw->requested = false; } +} + +void bnxt_qplib_disable_rcfw_channel(struct bnxt_qplib_rcfw *rcfw) +{ + unsigned long indx; + + bnxt_qplib_rcfw_stop_irq(rcfw, true); + if (rcfw->cmdq_bar_reg_iomem) iounmap(rcfw->cmdq_bar_reg_iomem); rcfw->cmdq_bar_reg_iomem = NULL; @@ -614,6 +624,31 @@ void bnxt_qplib_disable_rcfw_channel(struct bnxt_qplib_rcfw *rcfw) rcfw->vector = 0; } +int bnxt_qplib_rcfw_start_irq(struct bnxt_qplib_rcfw *rcfw, int msix_vector, + bool need_init) +{ + int rc; + + if (rcfw->requested) + return -EFAULT; + + rcfw->vector = msix_vector; + if (need_init) + tasklet_init(&rcfw->worker, + bnxt_qplib_service_creq, (unsigned long)rcfw); + else + tasklet_enable(&rcfw->worker); + rc = request_irq(rcfw->vector, bnxt_qplib_creq_irq, 0, + "bnxt_qplib_creq", rcfw); + if (rc) + return rc; + rcfw->requested = true; + CREQ_DB_REARM(rcfw->creq_bar_reg_iomem, rcfw->creq.cons, + rcfw->creq.max_elements); + + return 0; +} + int bnxt_qplib_enable_rcfw_channel(struct pci_dev *pdev, struct bnxt_qplib_rcfw *rcfw, int msix_vector, @@ -675,27 +710,17 @@ int bnxt_qplib_enable_rcfw_channel(struct pci_dev *pdev, rcfw->creq_qp_event_processed = 0; rcfw->creq_func_event_processed = 0; - rcfw->vector = msix_vector; if (aeq_handler) rcfw->aeq_handler = aeq_handler; + init_waitqueue_head(&rcfw->waitq); - tasklet_init(&rcfw->worker, bnxt_qplib_service_creq, - (unsigned long)rcfw); - - rcfw->requested = false; - rc = request_irq(rcfw->vector, bnxt_qplib_creq_irq, 0, - "bnxt_qplib_creq", rcfw); + rc = bnxt_qplib_rcfw_start_irq(rcfw, msix_vector, true); if (rc) { dev_err(&rcfw->pdev->dev, "QPLIB: Failed to request IRQ for CREQ rc = 0x%x", rc); bnxt_qplib_disable_rcfw_channel(rcfw); return rc; } - rcfw->requested = true; - - init_waitqueue_head(&rcfw->waitq); - - CREQ_DB_REARM(rcfw->creq_bar_reg_iomem, 0, rcfw->creq.max_elements); init.cmdq_pbl = cpu_to_le64(rcfw->cmdq.pbl[PBL_LVL_0].pg_map_arr[0]); init.cmdq_size_cmdq_lvl = cpu_to_le16( diff --git a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h index c7cce2e..46416df 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h +++ b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h @@ -195,7 +195,10 @@ struct bnxt_qplib_rcfw { void bnxt_qplib_free_rcfw_channel(struct bnxt_qplib_rcfw *rcfw); int bnxt_qplib_alloc_rcfw_channel(struct pci_dev *pdev, struct bnxt_qplib_rcfw *rcfw, int qp_tbl_sz); +void bnxt_qplib_rcfw_stop_irq(struct bnxt_qplib_rcfw *rcfw, bool kill); void bnxt_qplib_disable_rcfw_channel(struct bnxt_qplib_rcfw *rcfw); +int bnxt_qplib_rcfw_start_irq(struct bnxt_qplib_rcfw *rcfw, int msix_vector, + bool need_init); int bnxt_qplib_enable_rcfw_channel(struct pci_dev *pdev, struct bnxt_qplib_rcfw *rcfw, int msix_vector,