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

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

[PATCH Perftest v3] Perftest: Add support for TD lock-free mode
by Junxian Huang 13 Aug '24

13 Aug '24

From: Guofeng Yue <yueguofeng(a)h-partners.com> Add support for TD lock-free mode New option: --no_lock Usage example: ./ib_send_bw -d hns_0 --no_lock ./ib_send_bw -d hns_0 --no_lock 192.168.100.100 Signed-off-by: Guofeng Yue <yueguofeng(a)h-partners.com> --- configure.ac | 14 ++++++ src/perftest_parameters.c | 31 +++++++++++++- src/perftest_parameters.h | 1 + src/perftest_resources.c | 90 +++++++++++++++++++++++++++++++++++++++ src/perftest_resources.h | 4 ++ 5 files changed, 138 insertions(+), 2 deletions(-) diff --git a/configure.ac b/configure.ac index e59bf39..862eb08 100755 --- a/configure.ac +++ b/configure.ac @@ -389,6 +389,20 @@ if [test $HAVE_HNSDV = yes]; then AC_SUBST([LIBHNS]) fi +AC_TRY_LINK([#include <infiniband/verbs.h>], + [ibv_cq_ex_to_cq], [HAVE_CQ_EX=yes], [HAVE_CQ_EX=no]) +AM_CONDITIONAL([HAVE_CQ_EX], [test "x$HAVE_CQ_EX" = "xyes"]) +if [test $HAVE_CQ_EX = yes]; then + AC_DEFINE([HAVE_CQ_EX], [1], [Have CQ EX APT support]) +fi + +AC_TRY_LINK([#include <infiniband/verbs.h>], + [ibv_alloc_td], [HAVE_TD_API=yes], [HAVE_TD_API=no]) +AM_CONDITIONAL([HAVE_TD_API], [test "x$HAVE_TD_API" = "xyes"]) +if [test $HAVE_TD_API = yes]; then + AC_DEFINE([HAVE_TD_API], [1], [Have TD API support]) +fi + CFLAGS="-g -Wall -D_GNU_SOURCE -O3 $CFLAGS" LDFLAGS="$LDFLAGS" LIBS=$LIBS" -lpthread" diff --git a/src/perftest_parameters.c b/src/perftest_parameters.c index ff34a2c..c6608b7 100755 --- a/src/perftest_parameters.c +++ b/src/perftest_parameters.c @@ -466,6 +466,11 @@ static void usage(const char *argv0, VerbType verb, TestType tst, int connection printf(" Use a Shared Receive Queue. --rx-depth controls max-wr size of the SRQ \n"); } + #ifdef HAVE_TD_API + printf(" --no_lock "); + printf(" No lock in IO, include post send, post recv, post srq recv, poll cq \n"); + #endif + if (connection_type != RawEth) { printf(" --ipv6 "); printf(" Use IPv6 GID. Default is IPv4\n"); @@ -872,6 +877,7 @@ static void init_perftest_params(struct perftest_parameters *user_param) user_param->has_source_ip = 0; user_param->use_write_with_imm = 0; user_param->congest_type = OFF; + user_param->no_lock = OFF; } static int open_file_write(const char* file_path) @@ -2322,6 +2328,9 @@ int parser(struct perftest_parameters *user_param,char *argv[], int argc) #ifdef HAVE_HNSDV static int congest_type_flag = 0; #endif + #ifdef HAVE_TD_API + static int no_lock_flag = 0; + #endif char *server_ip = NULL; char *client_ip = NULL; @@ -2401,6 +2410,9 @@ int parser(struct perftest_parameters *user_param,char *argv[], int argc) { .name = "run_infinitely", .has_arg = 0, .flag = &run_inf_flag, .val = 1 }, { .name = "report_gbits", .has_arg = 0, .flag = &report_fmt_flag, .val = 1}, { .name = "use-srq", .has_arg = 0, .flag = &srq_flag, .val = 1}, + #ifdef HAVE_TD_API + { .name = "no_lock", .has_arg = 0, .flag = &no_lock_flag, .val = 1}, + #endif { .name = "use-null-mr", .has_arg = 0, .flag = &use_null_mr_flag, .val = 1}, { .name = "report-both", .has_arg = 0, .flag = &report_both_flag, .val = 1}, { .name = "reversed", .has_arg = 0, .flag = &is_reversed_flag, .val = 1}, @@ -3144,6 +3156,12 @@ int parser(struct perftest_parameters *user_param,char *argv[], int argc) user_param->use_srq = 1; } + #ifdef HAVE_TD_API + if (no_lock_flag) { + user_param->no_lock = 1; + } + #endif + if (use_null_mr_flag) { user_param->use_null_mr = 1; } @@ -3488,15 +3506,24 @@ void ctx_print_test_info(struct perftest_parameters *user_param) printf(" Number of qps : %d\t\tTransport type : %s\n", user_param->num_of_qps, transport_str(user_param->transport_type)); printf(" Connection type : %s\t\tUsing SRQ : %s\n", connStr[user_param->connection_type], user_param->use_srq ? "ON" : "OFF"); #ifdef HAVE_RO - printf(" PCIe relax order: %s\n", user_param->disable_pcir ? "OFF" : "ON"); + #ifdef HAVE_TD_API + printf(" PCIe relax order: %s\t\tLock-free : %s\n", user_param->disable_pcir ? "OFF" : "ON", user_param->no_lock ? "ON" : "OFF"); + #else + printf(" PCIe relax order: %s\t\tLock-free : %s\n", user_param->disable_pcir ? "OFF" : "ON", "Unsupported"); + #endif //HAVE_TD_API if ((check_pcie_relaxed_ordering_compliant() == false) && (user_param->disable_pcir == 0)) { printf(" WARNING: CPU is not PCIe relaxed ordering compliant.\n"); printf(" WARNING: You should disable PCIe RO with `--disable_pcie_relaxed` for both server and client.\n"); } #else - printf(" PCIe relax order: %s\n", "Unsupported"); + #ifdef HAVE_TD_API + printf(" PCIe relax order: %s\t\tLock-free : %s\n", "Unsupported", user_param->no_lock ? "ON" : "OFF"); + #else + printf(" PCIe relax order: %s\t\tLock-free : %s\n", "Unsupported", "Unsupported"); + #endif //HAVE_TD_API #endif + printf(" ibv_wr* API : %s\n", user_param->use_old_post_send ? "OFF" : "ON"); if (user_param->machine == CLIENT || user_param->duplex) { printf(" TX depth : %d\n",user_param->tx_depth); diff --git a/src/perftest_parameters.h b/src/perftest_parameters.h index d26253d..e809c81 100755 --- a/src/perftest_parameters.h +++ b/src/perftest_parameters.h @@ -523,6 +523,7 @@ struct perftest_parameters { int recv_post_list; int duration; int use_srq; + int no_lock; int congest_type; int use_xrc; int use_rss; diff --git a/src/perftest_resources.c b/src/perftest_resources.c index be3890c..40b7ddf 100755 --- a/src/perftest_resources.c +++ b/src/perftest_resources.c @@ -1317,6 +1317,20 @@ int destroy_ctx(struct pingpong_context *ctx, } #endif + #ifdef HAVE_TD_API + if (user_param->no_lock) { + if (ibv_dealloc_pd(ctx->pad)) { + fprintf(stderr, "Failed to deallocate PAD - %s\n", strerror(errno)); + test_result = 1; + } + + if (ibv_dealloc_td(ctx->td)) { + fprintf(stderr, "Failed to deallocate TD - %s\n", strerror(errno)); + test_result = 1; + } + } + #endif + if (ibv_dealloc_pd(ctx->pd)) { fprintf(stderr, "Failed to deallocate PD - %s\n", strerror(errno)); test_result = 1; @@ -1517,6 +1531,47 @@ int create_reg_cqs(struct pingpong_context *ctx, struct perftest_parameters *user_param, int tx_buffer_depth, int need_recv_cq) { +#ifdef HAVE_CQ_EX + struct ibv_cq_init_attr_ex send_cq_attr = { + .cqe = tx_buffer_depth * user_param->num_of_qps, + .cq_context = NULL, + .channel = ctx->send_channel, + .comp_vector = user_param->eq_num, + }; + + #ifdef HAVE_TD_API + if (user_param->no_lock) { + send_cq_attr.parent_domain = ctx->pad; + send_cq_attr.comp_mask = IBV_CQ_INIT_ATTR_MASK_PD; + } + #endif + ctx->send_cq = ibv_cq_ex_to_cq(ibv_create_cq_ex(ctx->context, &send_cq_attr)); + if (!ctx->send_cq) { + fprintf(stderr, "Couldn't create CQ\n"); + return FAILURE; + } + + if (need_recv_cq) { + struct ibv_cq_init_attr_ex recv_cq_attr = { + .cqe = user_param->rx_depth * user_param->num_of_qps, + .cq_context = NULL, + .channel = ctx->recv_channel, + .comp_vector = user_param->eq_num, + }; + #ifdef HAVE_TD_API + if (user_param->no_lock) { + recv_cq_attr.parent_domain = ctx->pad; + recv_cq_attr.comp_mask = IBV_CQ_INIT_ATTR_MASK_PD; + } + #endif + ctx->recv_cq = ibv_cq_ex_to_cq(ibv_create_cq_ex(ctx->context, &recv_cq_attr)); + if (!ctx->recv_cq) { + fprintf(stderr, "Couldn't create a receiver CQ\n"); + return FAILURE; + } + } + return SUCCESS; +#else 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) { @@ -1534,6 +1589,7 @@ int create_reg_cqs(struct pingpong_context *ctx, } return SUCCESS; +#endif } /****************************************************************************** @@ -1941,6 +1997,30 @@ int ctx_init(struct pingpong_context *ctx, struct perftest_parameters *user_para goto comp_channel; } + #ifdef HAVE_TD_API + /* Allocating the Thread domain, Parent domain. */ + if (user_param->no_lock) { + struct ibv_td_init_attr td_attr = {0}; + ctx->td = ibv_alloc_td(ctx->context, &td_attr); + if (!ctx->td) { + fprintf(stderr, "Couldn't allocate TD\n"); + goto pd; + } + + struct ibv_parent_domain_init_attr pad_attr = { + .pd = ctx->pd, + .td = ctx->td, + .comp_mask = 0, + }; + + ctx->pad = ibv_alloc_parent_domain(ctx->context, &pad_attr); + if (!ctx->pad) { + fprintf(stderr, "Couldn't allocate PAD\n"); + goto td; + } + } + #endif + #ifdef HAVE_AES_XTS if(user_param->aes_xts){ struct mlx5dv_dek_init_attr dek_attr = {}; @@ -2126,6 +2206,16 @@ dek: mlx5dv_dek_destroy(ctx->dek[i]); #endif +#ifdef HAVE_TD_API + if (user_param->no_lock) + ibv_dealloc_pd(ctx->pad); + +td: + if (user_param->no_lock) + ibv_dealloc_td(ctx->td); +pd: +#endif + ibv_dealloc_pd(ctx->pd); comp_channel: diff --git a/src/perftest_resources.h b/src/perftest_resources.h index 565deda..6d95d75 100755 --- a/src/perftest_resources.h +++ b/src/perftest_resources.h @@ -178,6 +178,10 @@ struct pingpong_context { struct ibv_comp_channel *recv_channel; struct ibv_comp_channel *send_channel; struct ibv_pd *pd; + #ifdef HAVE_TD_API + struct ibv_td *td; + struct ibv_pd *pad; + #endif struct ibv_mr **mr; struct ibv_mr *null_mr; struct ibv_cq *send_cq; -- 2.33.0

1 0

2024

2023

2022

2021

2020

High-performance-network