High-performance-network January 2025

high-performance-network@openeuler.org

1 participants
2 discussions

[PATCH] libhns: Add error logs to help diagnosis
by Chengchang Tang 03 Jan '25

03 Jan '25

From: Xinghai Cen <cenxinghai(a)h-partners.com> Add error logs to help diagnosis. Signed-off-by: Xinghai Cen <cenxinghai(a)h-partners.com> --- ...hns-Add-error-logs-to-help-diagnosis.patch | 242 ++++++++++++++++++ rdma-core.spec | 9 +- 2 files changed, 250 insertions(+), 1 deletion(-) create mode 100644 0042-libhns-Add-error-logs-to-help-diagnosis.patch diff --git a/0042-libhns-Add-error-logs-to-help-diagnosis.patch b/0042-libhns-Add-error-logs-to-help-diagnosis.patch new file mode 100644 index 0000000..9d880ea --- /dev/null +++ b/0042-libhns-Add-error-logs-to-help-diagnosis.patch @@ -0,0 +1,242 @@ +From 60c45b5f7c2cd0c2e7139d472406f071f327bb91 Mon Sep 17 00:00:00 2001 +From: Junxian Huang <huangjunxian6(a)hisilicon.com> +Date: Fri, 27 Dec 2024 14:02:29 +0800 +Subject: [PATCH] libhns: Add error logs to help diagnosis + +mainline inclusion +from mainline-master +commit 7849f1b17f89b8baa0065adaf9cd04204698ea82 +category: feature +bugzilla: https://gitee.com/src-openeuler/rdma-core/issues/IBFGPH +CVE: NA + +Reference: https://github.com/linux-rdma/rdma-core/pull/1533/commits/7849f1b17f89b8baa… + +---------------------------------------------------------------------- + +Add error logs to help diagnosis. + +Signed-off-by: Junxian Huang <huangjunxian6(a)hisilicon.com> +--- + providers/hns/hns_roce_u.c | 4 +- + providers/hns/hns_roce_u_hw_v2.c | 3 ++ + providers/hns/hns_roce_u_verbs.c | 79 ++++++++++++++++++++++++++------ + 3 files changed, 70 insertions(+), 16 deletions(-) + +diff --git a/providers/hns/hns_roce_u.c b/providers/hns/hns_roce_u.c +index e219b9e..ec995e7 100644 +--- a/providers/hns/hns_roce_u.c ++++ b/providers/hns/hns_roce_u.c +@@ -424,8 +424,10 @@ static struct verbs_context *hns_roce_alloc_context(struct ibv_device *ibdev, + + context->uar = mmap(NULL, hr_dev->page_size, PROT_READ | PROT_WRITE, + MAP_SHARED, cmd_fd, 0); +- if (context->uar == MAP_FAILED) ++ if (context->uar == MAP_FAILED) { ++ verbs_err(&context->ibv_ctx, "failed to mmap uar page.\n"); + goto err_set_attr; ++ } + + if (init_dca_context(context, cmd_fd, + &resp, ctx_attr, hr_dev->page_size)) +diff --git a/providers/hns/hns_roce_u_hw_v2.c b/providers/hns/hns_roce_u_hw_v2.c +index c746e03..0628646 100644 +--- a/providers/hns/hns_roce_u_hw_v2.c ++++ b/providers/hns/hns_roce_u_hw_v2.c +@@ -3057,6 +3057,9 @@ static int fill_send_wr_ops(const struct ibv_qp_init_attr_ex *attr, + fill_send_wr_ops_ud(qp_ex); + break; + default: ++ verbs_err(verbs_get_ctx(qp_ex->qp_base.context), ++ "QP type %d not supported for qp_ex send ops.\n", ++ attr->qp_type); + return -EOPNOTSUPP; + } + +diff --git a/providers/hns/hns_roce_u_verbs.c b/providers/hns/hns_roce_u_verbs.c +index c733b21..e9acfab 100644 +--- a/providers/hns/hns_roce_u_verbs.c ++++ b/providers/hns/hns_roce_u_verbs.c +@@ -422,8 +422,11 @@ static int verify_cq_create_attr(struct ibv_cq_init_attr_ex *attr, + { + struct hns_roce_pad *pad = to_hr_pad(attr->parent_domain); + +- if (!attr->cqe || attr->cqe > context->max_cqe) ++ if (!attr->cqe || attr->cqe > context->max_cqe) { ++ verbs_err(&context->ibv_ctx, "unsupported cq depth %u.\n", ++ attr->cqe); + return EINVAL; ++ } + + if (!check_comp_mask(attr->comp_mask, CREATE_CQ_SUPPORTED_COMP_MASK)) { + verbs_err(&context->ibv_ctx, "unsupported cq comps 0x%x\n", +@@ -431,8 +434,11 @@ static int verify_cq_create_attr(struct ibv_cq_init_attr_ex *attr, + return EOPNOTSUPP; + } + +- if (!check_comp_mask(attr->wc_flags, CREATE_CQ_SUPPORTED_WC_FLAGS)) ++ if (!check_comp_mask(attr->wc_flags, CREATE_CQ_SUPPORTED_WC_FLAGS)) { ++ verbs_err(&context->ibv_ctx, "unsupported wc flags 0x%llx.\n", ++ attr->wc_flags); + return EOPNOTSUPP; ++ } + + if (attr->comp_mask & IBV_CQ_INIT_ATTR_MASK_PD) { + if (!pad) { +@@ -504,8 +510,11 @@ static int exec_cq_create_cmd(struct ibv_context *context, + ret = ibv_cmd_create_cq_ex(context, attr, &cq->verbs_cq, + &cmd_ex.ibv_cmd, sizeof(cmd_ex), + &resp_ex.ibv_resp, sizeof(resp_ex), 0); +- if (ret) ++ if (ret) { ++ verbs_err(verbs_get_ctx(context), ++ "failed to exec create cq cmd, ret = %d.\n", ret); + return ret; ++ } + + cq->cqn = resp_drv->cqn; + cq->flags = resp_drv->cap_flags; +@@ -724,13 +733,20 @@ static int verify_srq_create_attr(struct hns_roce_context *context, + struct ibv_srq_init_attr_ex *attr) + { + if (attr->srq_type != IBV_SRQT_BASIC && +- attr->srq_type != IBV_SRQT_XRC) ++ attr->srq_type != IBV_SRQT_XRC) { ++ verbs_err(&context->ibv_ctx, ++ "unsupported srq type, type = %d.\n", attr->srq_type); + return -EINVAL; ++ } + + if (!attr->attr.max_sge || + attr->attr.max_wr > context->max_srq_wr || +- attr->attr.max_sge > context->max_srq_sge) ++ attr->attr.max_sge > context->max_srq_sge) { ++ verbs_err(&context->ibv_ctx, ++ "invalid srq attr size, max_wr = %u, max_sge = %u.\n", ++ attr->attr.max_wr, attr->attr.max_sge); + return -EINVAL; ++ } + + attr->attr.max_wr = max_t(uint32_t, attr->attr.max_wr, + HNS_ROCE_MIN_SRQ_WQE_NUM); +@@ -862,8 +878,12 @@ static int exec_srq_create_cmd(struct ibv_context *context, + ret = ibv_cmd_create_srq_ex(context, &srq->verbs_srq, init_attr, + &cmd_ex.ibv_cmd, sizeof(cmd_ex), + &resp_ex.ibv_resp, sizeof(resp_ex)); +- if (ret) ++ if (ret) { ++ verbs_err(verbs_get_ctx(context), ++ "failed to exec create srq cmd, ret = %d.\n", ++ ret); + return ret; ++ } + + srq->srqn = resp_ex.srqn; + srq->cap_flags = resp_ex.cap_flags; +@@ -1086,9 +1106,12 @@ static int check_qp_create_mask(struct hns_roce_context *ctx, + struct ibv_qp_init_attr_ex *attr) + { + struct hns_roce_device *hr_dev = to_hr_dev(ctx->ibv_ctx.context.device); ++ int ret = 0; + +- if (!check_comp_mask(attr->comp_mask, CREATE_QP_SUP_COMP_MASK)) +- return EOPNOTSUPP; ++ if (!check_comp_mask(attr->comp_mask, CREATE_QP_SUP_COMP_MASK)) { ++ ret = EOPNOTSUPP; ++ goto out; ++ } + + if (attr->comp_mask & IBV_QP_INIT_ATTR_SEND_OPS_FLAGS && + !check_comp_mask(attr->send_ops_flags, SEND_OPS_FLAG_MASK)) +@@ -1102,17 +1125,21 @@ static int check_qp_create_mask(struct hns_roce_context *ctx, + case IBV_QPT_RC: + case IBV_QPT_XRC_SEND: + if (!(attr->comp_mask & IBV_QP_INIT_ATTR_PD)) +- return EINVAL; ++ ret = EINVAL; + break; + case IBV_QPT_XRC_RECV: + if (!(attr->comp_mask & IBV_QP_INIT_ATTR_XRCD)) +- return EINVAL; ++ ret = EINVAL; + break; + default: + return EOPNOTSUPP; + } + +- return 0; ++out: ++ if (ret) ++ verbs_err(&ctx->ibv_ctx, "invalid comp_mask 0x%x.\n", ++ attr->comp_mask); ++ return ret; + } + + static int hns_roce_qp_has_rq(struct ibv_qp_init_attr_ex *attr) +@@ -1137,8 +1164,13 @@ static int verify_qp_create_cap(struct hns_roce_context *ctx, + if (cap->max_send_wr > ctx->max_qp_wr || + cap->max_recv_wr > ctx->max_qp_wr || + cap->max_send_sge > ctx->max_sge || +- cap->max_recv_sge > ctx->max_sge) ++ cap->max_recv_sge > ctx->max_sge) { ++ verbs_err(&ctx->ibv_ctx, ++ "invalid qp cap size, max_send/recv_wr = {%u, %u}, max_send/recv_sge = {%u, %u}.\n", ++ cap->max_send_wr, cap->max_recv_wr, ++ cap->max_send_sge, cap->max_recv_sge); + return -EINVAL; ++ } + + has_rq = hns_roce_qp_has_rq(attr); + if (!has_rq) { +@@ -1147,12 +1179,20 @@ static int verify_qp_create_cap(struct hns_roce_context *ctx, + } + + min_wqe_num = HNS_ROCE_V2_MIN_WQE_NUM; +- if (cap->max_send_wr < min_wqe_num) ++ if (cap->max_send_wr < min_wqe_num) { ++ verbs_debug(&ctx->ibv_ctx, ++ "change sq depth from %u to minimum %u.\n", ++ cap->max_send_wr, min_wqe_num); + cap->max_send_wr = min_wqe_num; ++ } + + if (cap->max_recv_wr) { +- if (cap->max_recv_wr < min_wqe_num) ++ if (cap->max_recv_wr < min_wqe_num) { ++ verbs_debug(&ctx->ibv_ctx, ++ "change rq depth from %u to minimum %u.\n", ++ cap->max_recv_wr, min_wqe_num); + cap->max_recv_wr = min_wqe_num; ++ } + + if (!cap->max_recv_sge) + return -EINVAL; +@@ -1646,6 +1686,11 @@ static int qp_exec_create_cmd(struct ibv_qp_init_attr_ex *attr, + ret = ibv_cmd_create_qp_ex2(&ctx->ibv_ctx.context, &qp->verbs_qp, attr, + &cmd_ex.ibv_cmd, sizeof(cmd_ex), + &resp_ex.ibv_resp, sizeof(resp_ex)); ++ if (ret) { ++ verbs_err(&ctx->ibv_ctx, ++ "failed to exec create qp cmd, ret = %d.\n", ret); ++ return ret; ++ } + + qp->flags = resp_ex.drv_payload.cap_flags; + *dwqe_mmap_key = resp_ex.drv_payload.dwqe_mmap_key; +@@ -1707,8 +1752,12 @@ static int mmap_dwqe(struct ibv_context *ibv_ctx, struct hns_roce_qp *qp, + { + qp->dwqe_page = mmap(NULL, HNS_ROCE_DWQE_PAGE_SIZE, PROT_WRITE, + MAP_SHARED, ibv_ctx->cmd_fd, dwqe_mmap_key); +- if (qp->dwqe_page == MAP_FAILED) ++ if (qp->dwqe_page == MAP_FAILED) { ++ verbs_err(verbs_get_ctx(ibv_ctx), ++ "failed to mmap direct wqe page, QPN = %u.\n", ++ qp->verbs_qp.qp.qp_num); + return -EINVAL; ++ } + + return 0; + } +-- +2.33.0 + diff --git a/rdma-core.spec b/rdma-core.spec index 121f2f5..db8e113 100644 --- a/rdma-core.spec +++ b/rdma-core.spec @@ -1,6 +1,6 @@ Name: rdma-core Version: 50.0 -Release: 17 +Release: 18 Summary: RDMA core userspace libraries and daemons License: GPL-2.0-only OR BSD-2-Clause AND BSD-3-Clause Url: https://github.com/linux-rdma/rdma-core @@ -47,6 +47,7 @@ patch38: 0038-libhns-Fix-reference-to-uninitialized-cq-pointer.patch patch39: 0039-libhns-Fix-the-exception-branch-of-wr_start-is-not-l.patch patch40: 0040-libhns-Fix-memory-leakage-when-DCA-is-enabled.patch patch41: 0041-libhns-Fix-coredump-during-QP-destruction-when-send_.patch +patch42: 0042-libhns-Add-error-logs-to-help-diagnosis.patch BuildRequires: binutils cmake >= 2.8.11 gcc libudev-devel pkgconfig pkgconfig(libnl-3.0) BuildRequires: pkgconfig(libnl-route-3.0) systemd systemd-devel @@ -626,6 +627,12 @@ fi %doc %{_docdir}/%{name}-%{version}/70-persistent-ipoib.rules %changelog +* Fri Jan 3 2025 Xinghai Cen <cenxinghai(a)h-partners.com> - 50.0-18 +- Type: requirement +- ID: NA +- SUG: NA +- DESC: Add error logs to help diagnosis + * Thu Nov 28 2024 Xinghai Cen <cenxinghai(a)h-partners.com> - 50.0-17 - Type: bugfix - ID: NA -- 2.33.0

1 0

[PATCH] libhns: Fixed two bugs in libhns
by Chengchang Tang 03 Jan '25

03 Jan '25

From: Xinghai Cen <cenxinghai(a)h-partners.com> Fixed two bugs in libhns: libhns: Fix bypassed vendor check in hnsdv_query_device() libhns: Fix coredump during QP destruction when send_cq： == recv_cq Signed-off-by: Xinghai Cen <cenxinghai(a)h-partners.com> --- ...sed-vendor-check-in-hnsdv_query_devi.patch | 40 ++++++++++++++ ...ump-during-QP-destruction-when-send_.patch | 53 +++++++++++++++++++ rdma-core.spec | 10 +++- 3 files changed, 102 insertions(+), 1 deletion(-) create mode 100644 0089-libhns-Fix-bypassed-vendor-check-in-hnsdv_query_devi.patch create mode 100644 0090-libhns-Fix-coredump-during-QP-destruction-when-send_.patch diff --git a/0089-libhns-Fix-bypassed-vendor-check-in-hnsdv_query_devi.patch b/0089-libhns-Fix-bypassed-vendor-check-in-hnsdv_query_devi.patch new file mode 100644 index 0000000..14f2edc --- /dev/null +++ b/0089-libhns-Fix-bypassed-vendor-check-in-hnsdv_query_devi.patch @@ -0,0 +1,40 @@ +From 485cddd47c83d6f229450b28d55d8e07f60ddcc0 Mon Sep 17 00:00:00 2001 +From: Yuyu Li <liyuyu6(a)huawei.com> +Date: Thu, 21 Nov 2024 21:37:15 +0800 +Subject: [PATCH] libhns: Fix bypassed vendor check in hnsdv_query_device() + +driver inclusion +category: bugfix +bugzilla: https://gitee.com/src-openeuler/rdma-core/issues/IBF87T + +-------------------------------------------------------------------------- + +The device vendor check is actually bypassed currently due +to the wrong if-condition. It should be a '||' statement. + +Fixes: 19e1eabc154f ("libhns: Add input parameter check for hnsdv_query_device()") +Signed-off-by: Yuyu Li <liyuyu6(a)huawei.com> +--- + providers/hns/hns_roce_u_verbs.c | 4 ++-- + 1 file changed, 2 insertions(+), 2 deletions(-) + +diff --git a/providers/hns/hns_roce_u_verbs.c b/providers/hns/hns_roce_u_verbs.c +index 090efbf..a6afce2 100644 +--- a/providers/hns/hns_roce_u_verbs.c ++++ b/providers/hns/hns_roce_u_verbs.c +@@ -128,10 +128,10 @@ int hnsdv_query_device(struct ibv_context *context, + struct hns_roce_context *ctx = context ? to_hr_ctx(context) : NULL; + struct hns_roce_device *hr_dev; + +- if (!ctx || !attrs_out) ++ if (!ctx || !context->device || !attrs_out) + return EINVAL; + +- if (!context->device && !is_hns_dev(context->device)) { ++ if (!is_hns_dev(context->device)) { + verbs_err(verbs_get_ctx(context), "not a HNS RoCE device!\n"); + return EOPNOTSUPP; + } +-- +2.33.0 + diff --git a/0090-libhns-Fix-coredump-during-QP-destruction-when-send_.patch b/0090-libhns-Fix-coredump-during-QP-destruction-when-send_.patch new file mode 100644 index 0000000..4909bd3 --- /dev/null +++ b/0090-libhns-Fix-coredump-during-QP-destruction-when-send_.patch @@ -0,0 +1,53 @@ +From ad5055f9b32ab0915803575385985fb10a29434a Mon Sep 17 00:00:00 2001 +From: Yuyu Li <liyuyu6(a)huawei.com> +Date: Mon, 25 Nov 2024 15:42:16 +0800 +Subject: [PATCH] libhns: Fix coredump during QP destruction when send_cq + == recv_cq + +driver inclusion +category: bugfix +bugzilla: https://gitee.com/src-openeuler/rdma-core/issues/IBF87T + +-------------------------------------------------------------------------- + +If the specified send CQ and recv CQ are both +the same CQ, the QP node in SCQ is not deleted. +which causes a segfault to occur when recreating +the QP. Here fix it. + +coredump info: +0x0000ffff8fbc37d4 in list_add_before_ +0x0000ffff8fbc381c in list_add_tail_ +0x0000ffff8fbc9d9c in add_qp_to_cq_list +0x0000ffff8fbca008 in create_qp +0x0000ffff8fbca110 in hns_roce_u_create_qp +0x0000ffff8feae39c in __ibv_create_qp_1_1 +0x0000000000401420 in test_ctrl_path + +Fixes: 5bebdb5ba77b ("libhns: Support reporting wc as software mode") +Signed-off-by: Yuyu Li <liyuyu6(a)huawei.com> +--- + providers/hns/hns_roce_u_hw_v2.c | 7 ++++--- + 1 file changed, 4 insertions(+), 3 deletions(-) + +diff --git a/providers/hns/hns_roce_u_hw_v2.c b/providers/hns/hns_roce_u_hw_v2.c +index 8f071e1..48a7566 100644 +--- a/providers/hns/hns_roce_u_hw_v2.c ++++ b/providers/hns/hns_roce_u_hw_v2.c +@@ -2033,9 +2033,10 @@ static int hns_roce_u_v2_destroy_qp(struct ibv_qp *ibqp) + list_del(&qp->rcq_node); + } + +- if (ibqp->send_cq && ibqp->send_cq != ibqp->recv_cq) { +- __hns_roce_v2_cq_clean(to_hr_cq(ibqp->send_cq), ibqp->qp_num, +- NULL); ++ if (ibqp->send_cq) { ++ if (ibqp->send_cq != ibqp->recv_cq) ++ __hns_roce_v2_cq_clean(to_hr_cq(ibqp->send_cq), ibqp->qp_num, ++ NULL); + list_del(&qp->scq_node); + } + +-- +2.33.0 + diff --git a/rdma-core.spec b/rdma-core.spec index 229810a..5e7be99 100644 --- a/rdma-core.spec +++ b/rdma-core.spec @@ -1,6 +1,6 @@ Name: rdma-core Version: 41.0 -Release: 28 +Release: 29 Summary: RDMA core userspace libraries and daemons License: GPLv2 or BSD Url: https://github.com/linux-rdma/rdma-core @@ -94,6 +94,8 @@ patch85: 0085-libhns-Fix-memory-leakage-when-DCA-is-enabled.patch patch86: 0086-libhns-Fix-the-exception-branch-of-wr_start-is-not-l.patch patch87: 0087-libhns-Fix-out-of-order-issue-of-requester-when-sett.patch patch88: 0088-libhns-Fix-reference-to-uninitialized-cq-pointer.patch +patch89: 0089-libhns-Fix-bypassed-vendor-check-in-hnsdv_query_devi.patch +patch90: 0090-libhns-Fix-coredump-during-QP-destruction-when-send_.patch BuildRequires: binutils cmake >= 2.8.11 gcc libudev-devel pkgconfig pkgconfig(libnl-3.0) BuildRequires: pkgconfig(libnl-route-3.0) valgrind-devel systemd systemd-devel @@ -341,6 +343,12 @@ fi %{_mandir}/* %changelog +* Thu Jan 2 2025 Xinghai Cen <cenxinghai(a)h-partners.com> - 41.0-29 +- Type: bugfix +- ID: NA +- SUG: NA +- DESC: Fixed two bugs in libhns + * Thu Nov 21 2024 Wentao Hu <huwentao19(a)h-partners.com> - 41.0-28 - Type: bugfix - ID: NA -- 2.33.0

1 0

2025

2024

2023

2022

2021

2020

High-performance-network January 2025