fabtests: New fabtest fi_flood to test over subscription of resources

	  1.  MR cache based registrations
	  tests regsiter and send  in batch and sequential modes while
	  flooding the cache beyond the maximum size.
	  2. Test receipt of unexpected messages by overwhelming the receiver

Signed-off-by: nikhil nanal <[email protected]>

fabtests/Makefile.am

@@ -40,7 +40,7 @@ bin_PROGRAMS = \
 	functional/fi_rdm_atomic \
 	functional/fi_rdm_stress \
 	functional/fi_multi_recv \
-	functional/fi_bw \
+	functional/fi_flood \
 	functional/fi_rdm_multi_client \
 	functional/fi_loopback \
 	benchmarks/fi_msg_pingpong \
@@ -360,9 +360,9 @@ functional_fi_multi_recv_SOURCES = \
 	functional/multi_recv.c
 functional_fi_multi_recv_LDADD = libfabtests.la
 
-functional_fi_bw_SOURCES = \
-	functional/bw.c
-functional_fi_bw_LDADD = libfabtests.la
+functional_fi_flood_SOURCES = \
+	 functional/flood.c
+functional_fi_flood_LDADD = libfabtests.la
 
 functional_fi_rdm_multi_client_SOURCES = \
 	functional/rdm_multi_client.c
@@ -651,7 +651,7 @@ dummy_man_pages = \
 	man/man1/fi_eq_test.1 \
 	man/man1/fi_getinfo_test.1 \
 	man/man1/fi_mr_test.1 \
-	man/man1/fi_bw.1 \
+	man/man1/fi_flood.1 \
 	man/man1/fi_rdm_multi_client.1 \
 	man/man1/fi_ubertest.1 \
 	man/man1/fi_efa_ep_rnr_retry.1

fabtests/Makefile.win

@@ -77,7 +77,7 @@ benchmarks: $(outdir)\dgram_pingpong.exe $(outdir)\msg_bw.exe \
 	$(outdir)\rdm_pingpong.exe $(outdir)\rma_pingpong.exe $(outdir)\rdm_tagged_bw.exe \
 	$(outdir)\rdm_bw.exe $(outdir)\rdm_tagged_pingpong.exe $(outdir)\rma_bw.exe
 
-functional: $(outdir)\av_xfer.exe $(outdir)\bw.exe $(outdir)\cm_data.exe $(outdir)\cq_data.exe \
+functional: $(outdir)\av_xfer.exe $(outdir)\flood.exe $(outdir)\cm_data.exe $(outdir)\cq_data.exe \
 	$(outdir)\dgram.exe $(outdir)\msg.exe $(outdir)\msg_epoll.exe \
 	$(outdir)\inject_test.exe $(outdir)\msg_sockets.exe $(outdir)\multi_mr.exe \
 	$(outdir)\multi_ep.exe $(outdir)\multi_recv.exe $(outdir)\rdm.exe \
@@ -120,7 +120,7 @@ $(outdir)\rma_bw.exe: {benchmarks}rma_bw.c $(basedeps) {benchmarks}benchmark_sha
 
 $(outdir)\av_xfer.exe: {functional}av_xfer.c $(basedeps)
 
-$(outdir)\bw.exe: {functional}bw.c $(basedeps)
+$(outdir)\flood.exe: {functional}flood.c $(basedeps)
 
 $(outdir)\cm_data.exe: {functional}cm_data.c $(basedeps)
 

fabtests/fabtests.vcxproj

@@ -239,7 +239,7 @@
     <ClCompile Include="functional\rdm_netdir.c" />
     <ClCompile Include="functional\scalable_ep.c" />
     <ClCompile Include="functional\inject_test.c" />
-    <ClCompile Include="functional\bw.c" />
+    <ClCompile Include="functional\flood.c" />
     <ClCompile Include="unit\av_test.c" />
     <ClCompile Include="unit\cntr_test.c" />
     <ClCompile Include="unit\common.c" />

fabtests/fabtests.vcxproj.filters

@@ -225,7 +225,7 @@
     <ClCompile Include="functional\unexpected_msg.c">
       <Filter>Source Files\functional</Filter>
     </ClCompile>
-    <ClCompile Include="functional\bw.c">
+    <ClCompile Include="functional\flood.c">
       <Filter>Source Files\functional</Filter>
     </ClCompile>
     <ClCompile Include="functional\inject_test.c">

fabtests/functional/bw.c → fabtests/functional/flood.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019 Intel Corporation.  All rights reserved.
+ * Copyright (c) Intel Corporation.  All rights reserved.
  *
  * This software is available to you under the BSD license
  * below:
@@ -34,7 +34,7 @@
 
 #include <shared.h>
 
-int sleep_time = 0;
+static int sleep_time = 0;
 
 static ssize_t post_one_tx(struct ft_context *msg)
 {
@@ -99,94 +99,136 @@ static int post_rx_sync(void)
 	return ret;
 }
 
-static int run_loop(void)
+static void mr_close_all(struct ft_context *ctx_arr, int window_size)
+{
+	int i;
+
+	for (i = 0; i < window_size; i++)
+		FT_CLOSE_FID(ctx_arr[i].mr);
+}
+
+static int run_seq_mr_send(void) {
+
+	int ret;
+	int i;
+
+	mr_close_all(tx_ctx_arr, opts.window_size);
+	mr_close_all(rx_ctx_arr, opts.window_size);
+
+	printf("Sequential memory registration:");
+	if (opts.dst_addr) {
+		for (i = 0; i < opts.window_size; i++) {
+			ret = ft_reg_mr(fi, tx_ctx_arr[i].buf, tx_mr_size,
+					ft_info_to_mr_access(fi),
+					FT_TX_MR_KEY + i, opts.iface, opts.device,
+					&(tx_ctx_arr[i].mr), &(tx_ctx_arr[i].desc));
+			if (ret)
+				goto out;
+
+			ret = post_one_tx(&tx_ctx_arr[i]);
+			if (ret)
+				goto out;
+
+			ret = ft_get_tx_comp(tx_seq);
+			if (ret)
+				goto out;
+
+			FT_CLOSE_FID(tx_ctx_arr[i].mr);
+		}
+	} else {
+		for (i = 0; i < opts.window_size; i++) {
+			ret = ft_reg_mr(fi, rx_ctx_arr[i].buf, rx_mr_size,
+					ft_info_to_mr_access(fi), FT_RX_MR_KEY + i, opts.iface, opts.device,
+					&(rx_ctx_arr[i].mr),
+					&(rx_ctx_arr[i].desc));
+			if (ret)
+				goto out;
+
+			ret = ft_post_rx_buf(ep, opts.transfer_size,
+				             &(rx_ctx_arr[i].context),
+					     rx_ctx_arr[i].buf,
+					     rx_ctx_arr[i].desc, ft_tag);
+			if (ret)
+				goto out;
+
+			ret = wait_check_rx_bufs();
+			if (ret)
+				goto out;
+
+			FT_CLOSE_FID(rx_ctx_arr[i].mr);
+		}
+	}
+	if (opts.options & FT_OPT_OOB_SYNC)
+		ret = ft_sync();
+	else
+		ret = post_rx_sync();
+out:
+	printf("%s\n", ret ? "Fail" : "Pass");
+	return ret;
+}
+
+static int run_batch_mr_send(void)
 {
 	int ret, i;
 
 	/* Receive side delay is used in order to let the sender
- 	   get ahead of the receiver and post multiple sends
-	   before the receiver begins processing them. */
+	 * get ahead of the receiver and post multiple sends
+	 * before the receiver begins processing them.
+	 */
 	if (!opts.dst_addr)
 		sleep(sleep_time);
 
-	ft_start();
+	printf("Batch memory registration:");
 	if (opts.dst_addr) {
 		for (i = 0; i < opts.window_size; i++) {
 			ret = post_one_tx(&tx_ctx_arr[i]);
 			if (ret)
-				return ret;
+				goto out;
 		}
 
 		ret = ft_get_tx_comp(tx_seq);
 		if (ret)
-			return ret;
+			goto out;
 	} else {
 		for (i = 0; i < opts.window_size; i++) {
 			ret = ft_post_rx_buf(ep, opts.transfer_size,
 					     &rx_ctx_arr[i].context,
 					     rx_ctx_arr[i].buf,
 					     rx_ctx_arr[i].desc, 0);
 			if (ret)
-				return ret;
+				goto out;
 		}
 
 		ret = wait_check_rx_bufs();
 		if (ret)
-			return ret;
+			goto out;
 	}
-	ft_stop();
 
 	if (opts.options & FT_OPT_OOB_SYNC)
 		ret = ft_sync();
 	else
 		ret = post_rx_sync();
-	if (ret)
-		return ret;
-
-	if (opts.machr)
-		show_perf_mr(opts.transfer_size, opts.window_size, &start, &end, 1,
-				opts.argc, opts.argv);
-	else
-		show_perf(NULL, opts.transfer_size, opts.window_size, &start, &end, 1);
-
+out:
+	printf("%s\n", ret ? "Fail" : "Pass");
 	return ret;
 }
 
 static int run(void)
 {
-	int ret, i;
+	int ret;
 
 	ret = hints->ep_attr->type == FI_EP_MSG ?
 		ft_init_fabric_cm() : ft_init_fabric();
 	if (ret)
 		return ret;
-
-	ret = ft_tx(ep, remote_fi_addr, 1, &tx_ctx);
-	if (ret)
-		return ret;
 
-	ret = ft_get_tx_comp(tx_seq);
+	ret = run_batch_mr_send();
 	if (ret)
-		return ret;
+		goto out;
 
-	ret = ft_get_rx_comp(rx_seq);
+	ret = run_seq_mr_send();
 	if (ret)
-		return ret;
-
-	if (!(opts.options & FT_OPT_SIZE)) {
-		for (i = 0; i < TEST_CNT; i++) {
-			if (!ft_use_size(i, opts.sizes_enabled))
-				continue;
-			opts.transfer_size = test_size[i].size;
-			ret = run_loop();
-			if (ret)
-				goto out;
-		}
-	} else {
-		ret = run_loop();
-		if (ret)
-			goto out;
-	}
+		goto out;
 
 out:
 	return ret;
@@ -197,6 +239,8 @@ int main(int argc, char **argv)
 	int op, ret;
 
 	opts = INIT_OPTS;
+	opts.options |= FT_OPT_ALLOC_MULT_MR;
+	opts.options |= FT_OPT_NO_PRE_POSTED_RX;
 
 	hints = fi_allocinfo();
 	if (!hints)
@@ -225,7 +269,7 @@ int main(int argc, char **argv)
 			break;
 		case '?':
 		case 'h':
-			ft_usage(argv[0], "A bandwidth test with data verification.");
+			ft_usage(argv[0], "test to oversubscribe mr cache and receiver with unexpected msgs.");
 			FT_PRINT_OPTS_USAGE("-T sleep_time",
 				"Receive side delay before starting");
 			FT_PRINT_OPTS_USAGE("-v", "Enable data verification");
@@ -243,8 +287,6 @@ int main(int argc, char **argv)
 	hints->domain_attr->mr_mode = opts.mr_mode;
 	hints->addr_format = opts.address_format;
 
-	opts.options |= FT_OPT_ALLOC_MULT_MR;
-
 	if (hints->ep_attr->type == FI_EP_DGRAM) {
 		fprintf(stderr, "This test does not support DGRAM endpoints\n");
 		return -FI_EINVAL;
@@ -260,4 +302,4 @@ int main(int argc, char **argv)
 	ft_free_res();
 
 	return ft_exit_code(ret);
-}
+}

fabtests/man/fabtests.7.md

@@ -140,10 +140,13 @@ features of libfabric.
   buffer tries to remain the same.  This test is used to validate the
   correct behavior of memory registration caches.
 
-*fi_bw*
-: Performs a one-sided bandwidth test with an option for data verification.
-  A sleep time on the receiving side can be enabled in order to allow
-  the sender to get ahead of the receiver.
+*fi_flood*
+: The test performs a one-sided transfer by utilizing Bulk Memory Region (MR)
+  registration and flooding the receiver with unexpected messages. This is
+  followed by sequential MR registration transfers, which force the MR cache
+  to evict the least recently used MRs before making new transfers. An optional
+  sleep time can be enabled on the receiving side to allow the sender to get
+  ahead of the receiver.
 
 *fi_rdm_multi_client*
 : Tests a persistent server communicating with multiple clients, one at a

fabtests/pytest/default/test_rdm.py

@@ -39,7 +39,7 @@ def test_rdm_shared_av(cmdline_args):
 @pytest.mark.functional
 def test_rdm_bw_functional(cmdline_args, completion_semantic):
     from common import ClientServerTest
-    test = ClientServerTest(cmdline_args, "fi_bw -e rdm -v -T 1", completion_semantic=completion_semantic)
+    test = ClientServerTest(cmdline_args, "fi_flood -e rdm -v -T 1", completion_semantic=completion_semantic)
     test.run()
 
 @pytest.mark.parametrize("iteration_type",

fabtests/pytest/efa/test_flood_peer.py

@@ -3,6 +3,6 @@
 @pytest.mark.functional
 def test_flood_peer(cmdline_args):
     from common import ClientServerTest
-    test = ClientServerTest(cmdline_args, "fi_bw -e rdm -W 6400 -S 512 -T 5",
+    test = ClientServerTest(cmdline_args, "fi_flood -e rdm -W 6400 -S 512 -T 5",
                             timeout=300)
     test.run()

fabtests/scripts/runfabtests.cmd

@@ -74,9 +74,9 @@ set functional_tests=^
 	"inject_test -N -A inject -v"^
 	"inject_test -A inj_complete -v"^
 	"inject_test -N -A inj_complete -v"^
-	"bw -e rdm -v -T 1"^
-	"bw -e rdm -v -T 1 -U"^
-	"bw -e msg -v -T 1"^
+	"flood -e rdm -v -T 1"^
+	"flood -e rdm -v -T 1 -U"^
+	"flood -e msg -v -T 1"^
 	"rdm_multi_client -C 10 -I 5"^
 	"rdm_multi_client -C 10 -I 5 -U"
 

fabtests/scripts/runfabtests.sh

@@ -150,9 +150,9 @@ functional_tests=(
 	"fi_inject_test -N -A inject -v"
 	"fi_inject_test -A inj_complete -v"
 	"fi_inject_test -N -A inj_complete -v"
-	"fi_bw -e rdm -v -T 1"
-	"fi_bw -e rdm -v -T 1 -U"
-	"fi_bw -e msg -v -T 1"
+	"fi_flood -e rdm -v -T 1"
+	"fi_flood -e rdm -v -T 1 -U"
+	"fi_flood -e msg -v -T 1"
 	"fi_rdm_multi_client -C 10 -I 5"
 	"fi_rdm_multi_client -C 10 -I 5 -U"
 )

fabtests/test_configs/tcp/io_uring.exclude

@@ -70,13 +70,14 @@ fi_msg_sockets
 # fi_unexpected_msg -e rdm fails with no message
 fi_unexpected_msg -e rdm
 
-# fi_bw -e msg fails with
+# fi_flood -e msg fails with
 # fi_eq_sread(): common/shared.c:1165, ret=-4 (Interrupted system call)
-fi_bw -e msg
+fi_flood -e msg
 
-# fi_bw fails by hanging
+# fi_flood fails by hanging
+# fi_flood fails by runfabtest timeout only on the CI.
 # This is a suspected race condition
-fi_bw
+fi_flood
 
 # fi_msg_pingpong fails with
 # fi_eq_sread(): common/shared.c:1127, ret=-4 (Interrupted system call)