- High-performance-network - mailweb.openeuler.org

[PATCH rdma-core 0/2] libhns: Two bugfixes in post_send flow
by Junxian Huang 08 Nov '24

08 Nov '24

Two hns bugfixes in post_send flow Chengchang Tang (1): libhns: Fix reference to uninitialized cq pointer Junxian Huang (1): libhns: Fix out-of-order issue of requester when setting FENCE providers/hns/hns_roce_u_hw_v2.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) -- 2.33.0

1 2

高性能网络sig例会
by openEuler conference 06 Nov '24

06 Nov '24

您好！ sig-high-performance-network 邀请您参加 2024-11-08 11:00 召开的WeLink会议(自动录制) 会议主题：高性能网络sig例会会议链接：https://meeting.huaweicloud.com:36443/#/j/960933891 会议纪要：https://etherpad.openeuler.org/p/sig-high-performance-network-meetings 更多资讯尽在：https://www.openeuler.org/zh/ Hello! sig-high-performance-network invites you to attend the WeLink conference(auto recording) will be held at 2024-11-08 11:00, The subject of the conference is 高性能网络sig例会, You can join the meeting at https://meeting.huaweicloud.com:36443/#/j/960933891. Add topics at https://etherpad.openeuler.org/p/sig-high-performance-network-meetings. More information: https://www.openeuler.org/en/

1 0

高性能网络sig会议
by openEuler conference 23 Oct '24

23 Oct '24

您好！ sig-high-performance-network 邀请您参加 2024-10-25 11:00 召开的WeLink会议会议主题：高性能网络sig会议会议链接：https://meeting.huaweicloud.com:36443/#/j/982220702 会议纪要：https://etherpad.openeuler.org/p/sig-high-performance-network-meetings 更多资讯尽在：https://www.openeuler.org/zh/ Hello! sig-high-performance-network invites you to attend the WeLink conference will be held at 2024-10-25 11:00, The subject of the conference is 高性能网络sig会议, You can join the meeting at https://meeting.huaweicloud.com:36443/#/j/982220702. Add topics at https://etherpad.openeuler.org/p/sig-high-performance-network-meetings. More information: https://www.openeuler.org/en/

1 0

Update kernel headers
by Junxian Huang 14 Oct '24

14 Oct '24

To commit: ?? ("RDMA/hns: Support mmapping reset state to userspace"). Signed-off-by: Junxian Huang <huangjunxian6(a)hisilicon.com> --- kernel-headers/rdma/hns-abi.h | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/kernel-headers/rdma/hns-abi.h b/kernel-headers/rdma/hns-abi.h index 94e861870..065eb2e0a 100644 --- a/kernel-headers/rdma/hns-abi.h +++ b/kernel-headers/rdma/hns-abi.h @@ -136,6 +136,7 @@ struct hns_roce_ib_alloc_ucontext_resp { __u32 max_inline_data; __u8 congest_type; __u8 reserved0[7]; + __aligned_u64 reset_mmap_key; }; struct hns_roce_ib_alloc_ucontext { @@ -153,4 +154,9 @@ struct hns_roce_ib_create_ah_resp { __u8 tc_mode; }; +struct hns_roce_reset_state { + __u32 hw_ready; + __u32 reserved; +}; + #endif /* HNS_ABI_USER_H */ -- 2.33.0

1 3

Re: rdma-core PR流程rpm差异分析已完成，待下一步处理
by li.fuyan＠zte.com.cn 27 Sep '24

27 Sep '24

你好，帮忙看下新的rpm差异分析是否还有问题。 https://gitee.com/src-openeuler/rdma-core/pulls/165 感谢！李富艳软件开发工程师 ICF六部/智算云底座研发中心/无线及算力研究院 ICF Dept. VI/Intelligent Computing and Cloud Foundation R&D Center/Wireless and Computing Product R&D Institute 中兴通讯股份有限公司西安市长安区西沣路五星段9号中兴通讯1E-401 邮编: 710114 T: +86 029 xxxxxxxx M: +86 15332482266 E: li.fuyan(a)zte.com.cn www.zte.com.cn Original From: 李富艳00122684 To: tangchengchang <tangchengchang(a)huawei.com>; Cc: high-performance-network(a)openeuler.org <high-performance-network(a)openeuler.org>;dev(a)openeuler.org <dev(a)openeuler.org>; Date: 2024年09月19日 14:49 Subject: rdma-core PR流程rpm差异分析已完成，待下一步处理您好，下面PR链接已经在9月11日针对RPM构建的差异完成了分析，下一步处理是不是到您那边了，请帮忙看一下。 https://gitee.com/src-openeuler/rdma-core/pulls/165 感谢！李富艳软件开发工程师 ICF六部/智算云底座研发中心/无线及算力研究院 ICF Dept. VI/Intelligent Computing and Cloud Foundation R&D Center/Wireless and Computing Product R&D Institute 中兴通讯股份有限公司西安市长安区西沣路五星段9号中兴通讯1E-401 邮编: 710114 T: +86 029 xxxxxxxx M: +86 15332482266 E: li.fuyan(a)zte.com.cn www.zte.com.cn

1 0

rdma-core PR流程rpm差异分析已完成，待下一步处理
by li.fuyan＠zte.com.cn 19 Sep '24

19 Sep '24

您好，下面PR链接已经在9月11日针对RPM构建的差异完成了分析，下一步处理是不是到您那边了，请帮忙看一下。 https://gitee.com/src-openeuler/rdma-core/pulls/165 感谢！李富艳软件开发工程师 ICF六部/智算云底座研发中心/无线及算力研究院 ICF Dept. VI/Intelligent Computing and Cloud Foundation R&D Center/Wireless and Computing Product R&D Institute 中兴通讯股份有限公司西安市长安区西沣路五星段9号中兴通讯1E-401 邮编: 710114 T: +86 029 xxxxxxxx M: +86 15332482266 E: li.fuyan(a)zte.com.cn www.zte.com.cn

1 0

关于rdma-core PR状态
by li.fuyan＠zte.com.cn 05 Sep '24

05 Sep '24

您好，请帮忙看一下我在高性能网络SIG下提交的关于“ZTE Dinghai rdma-core驱动合入”的PR是什么状态，我在页面上看不到构建信息了。 https://gitee.com/src-openeuler/rdma-core/pulls/165 李富艳中兴通讯股份有限公司 T: +86 029 xxxxxxxx M: +86 15332482266 E: li.fuyan(a)zte.com.cn www.zte.com.cn

1 0

Re: 关于欧拉社区合入rdma-core PR时门禁系统构建问题
by li.fuyan＠zte.com.cn 03 Sep '24

03 Sep '24

你好，帮忙看一下如下构建信息，我从日志里看不出来是哪里有问题，需要你的帮助。 PR链接: https://gitee.com/src-openeuler/rdma-core/pulls/165 构建链接： https://openeulerjenkins.osinfra.cn/job/multiarch/job/src-openeuler/job/x86… 构建错误日志： 3771 2024-09-03 17:22:56 patching file redhat/rdma-core.spec 3772 2024-09-03 17:22:56 Hunk #1 succeeded at 176 (offset 2 lines). 3773 2024-09-03 17:22:56 Hunk #2 succeeded at 204 (offset 2 lines). 3774 2024-09-03 17:22:56 Hunk #3 succeeded at 586 (offset 3 lines). 3775 2024-09-03 17:22:56 error: Bad exit status from /var/tmp/rpm-tmp.43Oe9Z (%prep) 3776 2024-09-03 17:22:56 Bad exit status from /var/tmp/rpm-tmp.43Oe9Z (%prep) 3777 2024-09-03 17:22:56 3778 2024-09-03 17:22:56 RPM build errors: 3779 2024-09-03 17:22:56 failed to build rpms with rdma-core.spec 看起来像是redhat/rdma-core.spec文件有异常，但是从如下修改内容来看，没有看到明显的错误，请帮忙看一下。李富艳软件开发工程师 ICF六部/智算云底座研发中心/无线及算力研究院 ICF Dept. VI/Intelligent Computing and Cloud Foundation R&D Center/Wireless and Computing Product R&D Institute 中兴通讯股份有限公司西安市长安区西沣路五星段9号中兴通讯1E-401 邮编: 710114 T: +86 029 xxxxxxxx M: +86 15332482266 E: li.fuyan(a)zte.com.cn www.zte.com.cn Original From: 李富艳00122684 To: tangchengchang <tangchengchang(a)huawei.com>;chenshaowei3(a)h-partners.com <chenshaowei3(a)h-partners.com>;wanghuan158(a)h-partners.com <wanghuan158(a)h-partners.com>; Cc: wanghuan158(a)h-partners.com <wanghuan158(a)h-partners.com>; Date: 2024年08月31日 14:07 Subject: 关于欧拉社区合入PR时门禁系统构建问题 tangchegnchang 你好，已经根据你之前提交的评审意见修改了ZTE Dinghai rdma-core的代码，但是由于操作问题，修改后的代码提交了新的PR，链接如下： https://gitee.com/src-openeuler/rdma-core/pulls/165 门禁系统当前状态为check_build失败，查看build日志，我发现提示信息如下红色标记，changelog部分错误看起来是基础内容，不是本次合入的，请你帮忙看一下如何解决这种问题。 https://eulermaker.compass-ci.openeuler.openatom.cn/package/build-record?os… 2024-08-31 12:59:19 Hunk #1 succeeded at 245 (offset 5 lines). 3772 2024-08-31 12:59:19 3773 2024-08-31 12:59:19 RPM build errors: 3774 2024-08-31 12:59:19 error: Bad exit status from /var/tmp/rpm-tmp.ikn0I8 (%prep) 3775 2024-08-31 12:59:19 bad date in %changelog: Wed May 29 zhangyaqi <zhangyaqi(a)kylinos.cn> - 50.0-11 3776 2024-08-31 12:59:19 Bad exit status from /var/tmp/rpm-tmp.ikn0I8 (%prep) 3777 2024-08-31 12:59:19 failed to build rpms with rdma-core.spec 另外，能否麻烦你加我微信或者你提供下你的电话，方便后续沟通。我的电话：15332482266 我的微信二维码如下，方便的话请你加一下我李富艳软件开发工程师 ICF六部/智算云底座研发中心/无线及算力研究院 ICF Dept. VI/Intelligent Computing and Cloud Foundation R&D Center/Wireless and Computing Product R&D Institute 中兴通讯股份有限公司西安市长安区西沣路五星段9号中兴通讯1E-401 邮编: 710114 T: +86 029 xxxxxxxx M: +86 15332482266 E: li.fuyan(a)zte.com.cn www.zte.com.cn

1 0

Perftest: Fix failure in creating cq when create_cq_ex is not supported by providers
by Junxian Huang 30 Aug '24

30 Aug '24

The previous TD lock-free patch introduced ibv_create_cq_ex() to perftest, but it breaks the provider devices which do not support this API yet. For these devices calling this API leads to an errno of EOPNOTSUPP. So add a check of errno, and if it is EOPNOTSUPP, use ibv_create_cq() as a fallback. Fixes: b6f957f6bc6c ("Perftest: Add support for TD lock-free mode") Signed-off-by: Junxian Huang <huangjunxian6(a)hisilicon.com> --- src/perftest_resources.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/src/perftest_resources.c b/src/perftest_resources.c index 2653207..814ffdb 100755 --- a/src/perftest_resources.c +++ b/src/perftest_resources.c @@ -1547,6 +1547,8 @@ int create_reg_cqs(struct pingpong_context *ctx, #endif ctx->send_cq = ibv_cq_ex_to_cq(ibv_create_cq_ex(ctx->context, &send_cq_attr)); if (!ctx->send_cq) { + if (!user_param->no_lock && errno == EOPNOTSUPP) + goto cq_ex_not_supported; fprintf(stderr, "Couldn't create CQ\n"); return FAILURE; } @@ -1571,7 +1573,9 @@ int create_reg_cqs(struct pingpong_context *ctx, } } return SUCCESS; -#else + +cq_ex_not_supported: +#endif ctx->send_cq = ibv_create_cq(ctx->context,tx_buffer_depth * user_param->num_of_qps, NULL, ctx->send_channel, user_param->eq_num); if (!ctx->send_cq) { @@ -1589,7 +1593,6 @@ int create_reg_cqs(struct pingpong_context *ctx, } return SUCCESS; -#endif } /****************************************************************************** -- 2.33.0

1 0

Perftest: Fix TD lock-free mode not working for QP
by Junxian Huang 19 Aug '24

19 Aug '24

From: Guofeng Yue <yueguofeng(a)h-partners.com> When creating QP in TD lock-free mode, set attr_ex.pd with ctx->pad instead of ctx->pd, otherwise the lock-free won't work. Fixes: 38d9fcfc3580 ("Perftest: Add support for TD lock-free mode") Signed-off-by: Guofeng Yue <yueguofeng(a)h-partners.com> Signed-off-by: Junxian Huang <huangjunxian6(a)hisilicon.com> --- src/perftest_resources.c | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/src/perftest_resources.c b/src/perftest_resources.c index 40b7ddf..2ea07fb 100755 --- a/src/perftest_resources.c +++ b/src/perftest_resources.c @@ -2384,7 +2384,13 @@ struct ibv_qp* ctx_qp_create(struct pingpong_context *ctx, else if (opcode == IBV_WR_RDMA_READ) attr_ex.send_ops_flags |= IBV_QP_EX_WITH_RDMA_READ; } + + #ifdef HAVE_TD_API + attr_ex.pd = user_param->no_lock ? ctx->pad : ctx->pd; + #else attr_ex.pd = ctx->pd; + #endif + attr_ex.comp_mask |= IBV_QP_INIT_ATTR_SEND_OPS_FLAGS | IBV_QP_INIT_ATTR_PD; attr_ex.send_cq = attr.send_cq; attr_ex.recv_cq = attr.recv_cq; -- 2.33.0

1 0

2025

2024

2023

2022

2021

2020

High-performance-network