perftest/0012-Perftest-Add-support-for-TD-lock-free-mode.patch

From 641ab294a6bc344dffd042019c609d4e9839a2f5 Mon Sep 17 00:00:00 2001
From: Guofeng Yue <yueguofeng@h-partners.com>
Date: Tue, 13 Aug 2024 21:48:44 +0800
Subject: [PATCH] Perftest: Add support for TD lock-free mode

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@h-partners.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
---
 configure.ac              | 14 +++++++
 src/perftest_parameters.c | 30 ++++++++++++++-
 src/perftest_parameters.h |  1 +
 src/perftest_resources.c  | 81 +++++++++++++++++++++++++++++++++++++++
 src/perftest_resources.h  |  4 ++
 5 files changed, 128 insertions(+), 2 deletions(-)

diff --git a/configure.ac b/configure.ac
index 54fc2cc..d976663 100755
--- a/configure.ac
+++ b/configure.ac
@@ -298,6 +298,20 @@ 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 $HAVE_CQ_EX = yes]; then
+       AC_DEFINE([HAVE_CQ_EX], [1], [Have CQ EX API support])
+fi
+
+AC_TRY_LINK([#include <infiniband/verbs.h>],
+       [ibv_alloc_td], [HAVE_TD_API=yes], [HAVE_TD_API=no])
+AM_CONDITIONAL([HAVE_TD_API], [test "x$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 $CFLAGS"
 LIBS=$LIBS" -lpthread"
 AC_SUBST([LIBUMAD])
diff --git a/src/perftest_parameters.c b/src/perftest_parameters.c
index 16af503..4c6ba4f 100755
--- a/src/perftest_parameters.c
+++ b/src/perftest_parameters.c
@@ -464,6 +464,11 @@ 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, including post send, post recv, post srq recv and poll cq \n");
+	#endif
+
 	if (connection_type != RawEth) {
 		printf("      --ipv6 ");
 		printf(" Use IPv6 GID. Default is IPv4\n");
@@ -2225,6 +2230,9 @@ 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;
@@ -2300,6 +2308,9 @@ 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 = "report-both",	.has_arg = 0, .flag = &report_both_flag, .val = 1},
 			{ .name = "reversed",		.has_arg = 0, .flag = &is_reversed_flag, .val = 1},
 			{ .name = "pkey_index",		.has_arg = 1, .flag = &pkey_flag, .val = 1},
@@ -2944,6 +2955,12 @@ 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 (report_fmt_flag) {
 		user_param->report_fmt = GBS;
 	}
@@ -3284,15 +3301,24 @@ 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 ffc19ab..f49fb0f 100755
--- a/src/perftest_parameters.h
+++ b/src/perftest_parameters.h
@@ -462,6 +462,7 @@ 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 451f11d..0e2b9a3 100755
--- a/src/perftest_resources.c
+++ b/src/perftest_resources.c
@@ -1360,6 +1360,20 @@ 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;
@@ -1502,6 +1516,47 @@ 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->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->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->channel, user_param->eq_num);
 	if (!ctx->send_cq) {
@@ -1519,6 +1574,7 @@ int create_reg_cqs(struct pingpong_context *ctx,
 	}

 	return SUCCESS;
+#endif
 }

 /******************************************************************************
@@ -1906,6 +1962,31 @@ int ctx_init(struct pingpong_context *ctx, struct perftest_parameters *user_para
 		fprintf(stderr, "Couldn't allocate PD\n");
 		return FAILURE;
 	}
+
+	       #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");
+                        return FAILURE;
+                }
+
+                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");
+                        return FAILURE;
+                }
+        }
+        #endif
+
 	#ifdef HAVE_AES_XTS
 	if(user_param->aes_xts){
 		struct mlx5dv_dek_init_attr dek_attr = {};
diff --git a/src/perftest_resources.h b/src/perftest_resources.h
index cf0502c..ba8630b 100755
--- a/src/perftest_resources.h
+++ b/src/perftest_resources.h
@@ -170,6 +170,10 @@ struct pingpong_context {
 	#endif
 	struct ibv_comp_channel			*channel;
 	struct ibv_pd				*pd;
+	#ifdef HAVE_TD_API
+	struct ibv_td                           *td;
+	struct ibv_pd                           *pad;
+	#endif
 	struct ibv_mr				**mr;
 	struct ibv_cq				*send_cq;
 	struct ibv_cq				*recv_cq;
--
2.25.1