[dpdk-dev] [PATCH v3 4/5] app/test-flow-perf: add memory dump to app

[Wisam Jaddo <wisamm@mellanox.com>]

Introduce new feature to dump memory statistics of each socket
and a total for all before and after the creation.

This will give two main advantage:
1- Check the memory consumption for large number of flows
"insertion rate scenario alone"

2- Check that no memory leackage after doing insertion then
deletion.

Signed-off-by: Suanming Mou <suanmingm@mellanox.com>
Signed-off-by: Wisam Jaddo <wisamm@mellanox.com>
---
 app/test-flow-perf/main.c      | 69 ++++++++++++++++++++++++++++++++++
 doc/guides/tools/flow-perf.rst |  6 ++-
 2 files changed, 74 insertions(+), 1 deletion(-)

diff --git a/app/test-flow-perf/main.c b/app/test-flow-perf/main.c
index 7c11c0b577..95435910de 100644
--- a/app/test-flow-perf/main.c
+++ b/app/test-flow-perf/main.c
@@ -62,6 +62,7 @@ static uint16_t flow_actions;
 static uint8_t flow_attrs;
 static volatile bool force_quit;
 static volatile bool dump_iterations;
+static volatile bool dump_socket_mem_flag;
 static volatile bool delete_flag;
 static struct rte_mempool *mbuf_mp;
 static uint32_t nb_lcores;
@@ -78,6 +79,7 @@ static void usage(char *progname)
 		" iteration\n");
 	printf("  --deletion-rate: Enable deletion rate"
 		" calculations\n");
+	printf("  --dump-socket-mem: to dump all socket memory\n");
 
 	printf("To set flow attributes:\n");
 	printf("  --ingress: set ingress attribute in flows\n");
@@ -127,6 +129,7 @@ args_parse(int argc, char **argv)
 		{ "flows-count",                1, 0, 0 },
 		{ "dump-iterations",            0, 0, 0 },
 		{ "deletion-rate",              0, 0, 0 },
+		{ "dump-socket-mem",            0, 0, 0 },
 		/* Attributes */
 		{ "ingress",                    0, 0, 0 },
 		{ "egress",                     0, 0, 0 },
@@ -310,6 +313,8 @@ args_parse(int argc, char **argv)
 				dump_iterations = true;
 			if (!strcmp(lgopts[opt_idx].name, "deletion-rate"))
 				delete_flag = true;
+			if (!strcmp(lgopts[opt_idx].name, "dump-socket-mem"))
+				dump_socket_mem_flag = true;
 			break;
 		default:
 			usage(argv[0]);
@@ -321,6 +326,62 @@ args_parse(int argc, char **argv)
 	printf("end_flow\n");
 }
 
+/* Dump the socket memory statistics on console */
+static size_t
+dump_socket_mem(FILE *f)
+{
+	struct rte_malloc_socket_stats socket_stats;
+	unsigned int i = 0;
+	size_t total = 0;
+	size_t alloc = 0;
+	size_t free = 0;
+	unsigned int n_alloc = 0;
+	unsigned int n_free = 0;
+	bool active_nodes = false;
+
+
+	for (i = 0; i < RTE_MAX_NUMA_NODES; i++) {
+		if (rte_malloc_get_socket_stats(i, &socket_stats) ||
+		    !socket_stats.heap_totalsz_bytes)
+			continue;
+		active_nodes = true;
+		total += socket_stats.heap_totalsz_bytes;
+		alloc += socket_stats.heap_allocsz_bytes;
+		free += socket_stats.heap_freesz_bytes;
+		n_alloc += socket_stats.alloc_count;
+		n_free += socket_stats.free_count;
+		if (dump_socket_mem_flag) {
+			fprintf(f, "::::::::::::::::::::::::::::::::::::::::");
+			fprintf(f,
+				"\nSocket %u:\nsize(M) total: %.6lf\nalloc:"
+				" %.6lf(%.3lf%%)\nfree: %.6lf"
+				"\nmax: %.6lf"
+				"\ncount alloc: %u\nfree: %u\n",
+				i,
+				socket_stats.heap_totalsz_bytes / 1.0e6,
+				socket_stats.heap_allocsz_bytes / 1.0e6,
+				(double)socket_stats.heap_allocsz_bytes * 100 /
+				(double)socket_stats.heap_totalsz_bytes,
+				socket_stats.heap_freesz_bytes / 1.0e6,
+				socket_stats.greatest_free_size / 1.0e6,
+				socket_stats.alloc_count,
+				socket_stats.free_count);
+				fprintf(f, "::::::::::::::::::::::::::::::::::::::::");
+		}
+	}
+	if (dump_socket_mem_flag && active_nodes) {
+		fprintf(f,
+			"\nTotal: size(M)\ntotal: %.6lf"
+			"\nalloc: %.6lf(%.3lf%%)\nfree: %.6lf"
+			"\ncount alloc: %u\nfree: %u\n",
+			total / 1.0e6, alloc / 1.0e6,
+			(double)alloc * 100 / (double)total, free / 1.0e6,
+			n_alloc, n_free);
+		fprintf(f, "::::::::::::::::::::::::::::::::::::::::\n");
+	}
+	return alloc;
+}
+
 static void
 print_flow_error(struct rte_flow_error error)
 {
@@ -657,6 +718,7 @@ main(int argc, char **argv)
 	uint16_t nr_ports;
 	int ret;
 	struct rte_flow_error error;
+	int64_t alloc, last_alloc;
 
 	nr_ports = rte_eth_dev_count_avail();
 	ret = rte_eal_init(argc, argv);
@@ -666,6 +728,7 @@ main(int argc, char **argv)
 	force_quit = false;
 	dump_iterations = false;
 	delete_flag = false;
+	dump_socket_mem_flag = false;
 	flows_count = 4000000;
 	iterations_number = 100000;
 	flow_group = 0;
@@ -686,7 +749,13 @@ main(int argc, char **argv)
 	if (nb_lcores <= 1)
 		rte_exit(EXIT_FAILURE, "This app needs at least two cores\n");
 
+	last_alloc = (int64_t)dump_socket_mem(stdout);
 	flows_handler();
+	alloc = (int64_t)dump_socket_mem(stdout);
+
+	if (last_alloc)
+		fprintf(stdout, ":: Memory allocation change(M): %.6lf\n",
+		(alloc - last_alloc) / 1.0e6);
 
 	RTE_LCORE_FOREACH_SLAVE(lcore_id)
 
diff --git a/doc/guides/tools/flow-perf.rst b/doc/guides/tools/flow-perf.rst
index e07e659df5..28d452fd06 100644
--- a/doc/guides/tools/flow-perf.rst
+++ b/doc/guides/tools/flow-perf.rst
@@ -18,7 +18,8 @@ give different flow each time, and all other items will have open masks.
 The current design have single core insertion rate. In the future we may
 have a multi core insertion rate measurement support in the app.
 
-The application also provide the ability to measure rte flow deletion rate.
+The application also provide the ability to measure rte flow deletion rate,
+in addition to memory consumption before and after the flows creation.
 
 
 Compiling the Application
@@ -94,6 +95,9 @@ The command line options are:
 *	``--deletion-rate``
 	Enable deletion rate calculations.
 
+*	``--dump-socket-mem``
+	Dump the memory stats for each socket before the insertion and after.
+
 Attributes:
 
 *	``--ingress``
-- 
2.17.1