DPDK patches and discussions
 help / color / mirror / Atom feed
* [dpdk-dev] [PATCH v2] test: new sched WRR unit-test
@ 2017-11-30  9:05 alangordondewar
  2017-11-30 10:48 ` Luca Boccassi
  2018-01-02 16:15 ` Dumitrescu, Cristian
  0 siblings, 2 replies; 4+ messages in thread
From: alangordondewar @ 2017-11-30  9:05 UTC (permalink / raw)
  To: cristian.dumitrescu; +Cc: dev, Alan Dewar

From: Alan Dewar <alan.dewar@att.com>

New unit-test for the librte_sched WRR weighting code.

With the standard 17.11 code, the first three sub-tests pass, but
the last three fail due to bugs in the WRR weighting code.

With v1 of the "sched: fix overflow errors in WRR weighting code"
patch the first five sub-tests pass, and the last sub-test fails badly.

With v2 of the "sched: fix overflow errors in WRR weighting code"
patch the first five sub-tests pass, and the last sub-test is a very
near miss (i.e. measured packets counts are one away from the expected
counts).

Signed-off-by: Alan Dewar <alan.dewar@att.com>
---
v2 - add new 255-254-253-1 weightings sub-test

 test/test/Makefile         |   1 +
 test/test/test_sched_wrr.c | 491 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 492 insertions(+)
 create mode 100644 test/test/test_sched_wrr.c

diff --git a/test/test/Makefile b/test/test/Makefile
index bb54c98..0ab0ed3 100644
--- a/test/test/Makefile
+++ b/test/test/Makefile
@@ -173,6 +173,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_NET) += test_crc.c
 ifeq ($(CONFIG_RTE_LIBRTE_SCHED),y)
 SRCS-y += test_red.c
 SRCS-y += test_sched.c
+SRCS-y += test_sched_wrr.c
 endif
 
 SRCS-$(CONFIG_RTE_LIBRTE_METER) += test_meter.c
diff --git a/test/test/test_sched_wrr.c b/test/test/test_sched_wrr.c
new file mode 100644
index 0000000..df5a231
--- /dev/null
+++ b/test/test/test_sched_wrr.c
@@ -0,0 +1,491 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   Copyright(c) 2017 ATT Intellectual Property. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ */
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <stdint.h>
+#include <unistd.h>
+
+#include "test.h"
+
+#include <rte_cycles.h>
+#include <rte_ether.h>
+#include <rte_ip.h>
+#include <rte_byteorder.h>
+#include <rte_sched.h>
+
+
+#define SUBPORT         0
+#define PIPE            1
+#define TC              0
+#define QUEUE           3
+
+static struct rte_sched_subport_params subport_param[] = {
+	{
+		.tb_rate = 1250000000,
+		.tb_size = 1000000,
+
+		.tc_rate = {1250000000, 1250000000, 1250000000, 1250000000},
+		.tc_period = 10,
+	},
+};
+
+static struct rte_sched_pipe_params pipe_profile[] = {
+	{ /* Profile #0 */
+		.tb_rate = 3051750,
+		.tb_size = 1000000,
+
+		.tc_rate = {3051750, 3051750, 3051750, 3051750},
+		.tc_period = 160,
+
+		.wrr_weights = {1, 1, 1, 1,
+				1, 1, 1, 1,
+				1, 1, 1, 1,
+				1, 1, 1, 1},
+	},
+};
+
+static struct rte_sched_port_params port_param = {
+	.socket = 0, /* computed */
+	.rate = 0, /* computed */
+	.mtu = 1522,
+	.frame_overhead = RTE_SCHED_FRAME_OVERHEAD_DEFAULT,
+	.n_subports_per_port = 1,
+	.n_pipes_per_subport = 1024,
+	.qsize = {32, 32, 32, 32},
+	.pipe_profiles = pipe_profile,
+	.n_pipe_profiles = 1,
+};
+
+#define NB_MBUF          (4 * 32)
+#define MBUF_DATA_SZ     (2048 + RTE_PKTMBUF_HEADROOM)
+#define MEMPOOL_CACHE_SZ 0
+#define SOCKET           0
+
+
+static struct rte_mempool *
+create_mempool(uint32_t total_packets)
+{
+	struct rte_mempool *mp;
+
+	mp = rte_mempool_lookup("test_sched wrr");
+	if (!mp)
+		mp = rte_pktmbuf_pool_create("test_sched wrr", total_packets,
+			MEMPOOL_CACHE_SZ, 0, MBUF_DATA_SZ, SOCKET);
+
+	return mp;
+}
+
+static void
+delete_mempool(struct rte_mempool *mp)
+{
+	rte_mempool_free(mp);
+}
+
+static void
+prepare_pkt(struct rte_mbuf *mbuf, uint32_t tc, uint32_t queue)
+{
+	struct ether_hdr *eth_hdr;
+	struct vlan_hdr *vlan1, *vlan2;
+	struct ipv4_hdr *ip_hdr;
+
+	/* Simulate a classifier */
+	eth_hdr = rte_pktmbuf_mtod(mbuf, struct ether_hdr *);
+	vlan1 = (struct vlan_hdr *)(&eth_hdr->ether_type);
+	vlan2 = (struct vlan_hdr *)((uintptr_t)&eth_hdr->ether_type +
+				    sizeof(struct vlan_hdr));
+	eth_hdr = (struct ether_hdr *)((uintptr_t)&eth_hdr->ether_type +
+				       2 * sizeof(struct vlan_hdr));
+	ip_hdr = (struct ipv4_hdr *)((uintptr_t)eth_hdr +
+				     sizeof(eth_hdr->ether_type));
+
+	vlan1->vlan_tci = rte_cpu_to_be_16(SUBPORT);
+	vlan2->vlan_tci = rte_cpu_to_be_16(PIPE);
+	eth_hdr->ether_type =  rte_cpu_to_be_16(ETHER_TYPE_IPv4);
+	ip_hdr->dst_addr = IPv4(0, 0, TC, QUEUE);
+
+
+	rte_sched_port_pkt_write(mbuf, SUBPORT, PIPE, tc, queue,
+				 e_RTE_METER_YELLOW);
+
+	/* 64 byte packet */
+	mbuf->pkt_len  = 60;
+	mbuf->data_len = 60;
+}
+
+/*
+ * This function carries out the core of the enqueue/dequeue testing.
+ * This is where we should detect failures if the WRR code is broken.
+ */
+static int
+test_sched_wrr_enqueue_dequeue(const char *subtest_name,
+			       struct rte_mempool *mp,
+			       struct rte_sched_port *port,
+			       struct rte_mbuf **in_mbufs,
+			       struct rte_mbuf **out_mbufs,
+			       int32_t enqueue_packets,
+			       int32_t dequeue_packets,
+			       uint32_t *expected_counts)
+{
+	uint32_t wrr_counts[RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS] = { 0 };
+	uint32_t pipe;
+	int32_t i;
+	int err;
+
+	/*
+	 * Create the packets to be enqueued, spread them evenly across
+	 * each of the four WRR queues of the test-TC
+	 */
+	for (i = 0; i < enqueue_packets; i++) {
+		in_mbufs[i] = rte_pktmbuf_alloc(mp);
+		TEST_ASSERT_NOT_NULL(in_mbufs[i],
+				     "%s: Packet allocation failed on packet "
+				     "%d\n", subtest_name, i);
+		prepare_pkt(in_mbufs[i], TC, (i & 0x3));
+	}
+
+	/*
+	 * Queue all the enqueue packets, none should be dropped as the
+	 * four queues should be long enough.
+	 */
+	err = rte_sched_port_enqueue(port, in_mbufs, enqueue_packets);
+	TEST_ASSERT_EQUAL(err, enqueue_packets,
+			  "%s: Enqueue failed, err: %d != %d\n",
+			  subtest_name, err, enqueue_packets);
+
+	/*
+	 * Dequeue the required number of packets.
+	 */
+	err = rte_sched_port_dequeue(port, out_mbufs, dequeue_packets);
+	TEST_ASSERT_EQUAL(err, dequeue_packets,
+			  "%s: Dequeue failed, err: %d != %d\n",
+			  subtest_name, err, dequeue_packets);
+
+	/*
+	 * Check each packet and count which WRR queue it came from.
+	 */
+	for (i = 0; i < dequeue_packets; i++) {
+		enum rte_meter_color color;
+		uint32_t subport, traffic_class, queue;
+
+		color = rte_sched_port_pkt_read_color(out_mbufs[i]);
+		TEST_ASSERT_EQUAL(color, e_RTE_METER_YELLOW,
+				  "%s: Wrong color\n", subtest_name);
+
+		rte_sched_port_pkt_read_tree_path(out_mbufs[i],
+				&subport, &pipe, &traffic_class, &queue);
+
+		TEST_ASSERT_EQUAL(subport, SUBPORT, "%s: Wrong subport\n",
+				  subtest_name);
+		TEST_ASSERT_EQUAL(pipe, PIPE, "%s: Wrong pipe\n", subtest_name);
+		TEST_ASSERT_EQUAL(traffic_class, TC,
+				  "%s: Wrong traffic-class\n", subtest_name);
+		wrr_counts[queue]++;
+		rte_pktmbuf_free(out_mbufs[i]);
+	}
+
+	/*
+	 * Check the number of packets dequeued from each WRR queue
+	 * against the expected counts.
+	 */
+	err = 0;
+	for (i = 0; i < RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS; i++) {
+		printf("%s - WRR queue %d, dequeued: %u, expected: %u\n",
+		       subtest_name, i, wrr_counts[i], expected_counts[i]);
+		if (wrr_counts[i] != expected_counts[i])
+			err = -1;
+	}
+	return err;
+}
+
+/*
+ * This function does the test config set-up and tear-down.
+ * If we see failures in here it is probably due to the test configuration.
+ */
+static int
+test_sched_wrr_test(const char *subtest_name, uint16_t *tc_qlengths,
+		    uint8_t *wrr_weights, int32_t enqueue_packets,
+		    int32_t dequeue_packets, uint32_t *expected_counts)
+{
+	struct rte_mbuf **in_mbufs;
+	struct rte_mbuf **out_mbufs;
+	struct rte_mempool *mp = NULL;
+	struct rte_sched_port *port = NULL;
+	int32_t total_expected = 0;
+	uint32_t pipe;
+	int32_t i;
+	int err;
+
+	/*
+	 * Some inbound argument checking
+	 */
+	TEST_ASSERT_EQUAL((tc_qlengths[TC] * 4), enqueue_packets,
+			  "%s: Queue length/Enqueue packet mismatch - "
+			  "%u vs %d\n",
+			  subtest_name, (tc_qlengths[TC] * 4), enqueue_packets);
+	TEST_ASSERT((dequeue_packets <= enqueue_packets),
+		    "%s: Dequeue packets %d > Enqueue packets %d\n",
+		    subtest_name, dequeue_packets, enqueue_packets);
+
+	for (i = 0; i < RTE_SCHED_QUEUES_PER_TRAFFIC_CLASS; i++)
+		total_expected += expected_counts[i];
+
+	TEST_ASSERT((dequeue_packets == total_expected),
+		    "%s: Dequeue packets %d != Total expected %u\n",
+		    subtest_name, dequeue_packets, total_expected);
+
+	/*
+	 * Create the mempool and allocate arrays to hold the rte_mbuf pointers
+	 */
+	mp = create_mempool(enqueue_packets);
+	TEST_ASSERT_NOT_NULL(mp, "%s: Error creating mempool\n", subtest_name);
+
+	in_mbufs = malloc(sizeof(struct rte_mbuf *) * enqueue_packets);
+	TEST_ASSERT_NOT_NULL(in_mbufs, "%s: Error creating in_mbuf array\n",
+			     subtest_name);
+
+	out_mbufs = malloc(sizeof(struct rte_mbuf *) * dequeue_packets);
+	TEST_ASSERT_NOT_NULL(out_mbufs, "%s: Error creating out_mbuf array\n",
+			     subtest_name);
+
+	/*
+	 * Set up the port and pipe profiles with the TC's queue lengths
+	 * and the WRR queue weightings
+	 */
+	port_param.socket = 0;
+	port_param.rate = (uint64_t) 10000 * 1000 * 1000 / 8;
+	for (i = 0; i < RTE_SCHED_TRAFFIC_CLASSES_PER_PIPE; i++)
+		port_param.qsize[i] = tc_qlengths[i];
+
+	for (i = 0; i < RTE_SCHED_QUEUES_PER_PIPE; i++)
+		pipe_profile[0].wrr_weights[i] = wrr_weights[i];
+
+	port = rte_sched_port_config(&port_param);
+	TEST_ASSERT_NOT_NULL(port, "%s: Error config sched port\n",
+			     subtest_name);
+
+	err = rte_sched_subport_config(port, SUBPORT, subport_param);
+	TEST_ASSERT_SUCCESS(err, "%s: Error config sched, err=%d\n",
+			    subtest_name, err);
+
+	for (pipe = 0; pipe < port_param.n_pipes_per_subport; pipe++) {
+		err = rte_sched_pipe_config(port, SUBPORT, pipe, 0);
+		TEST_ASSERT_SUCCESS(err,
+				    "%s: Error config sched pipe %u, err=%d\n",
+				    subtest_name, pipe, err);
+	}
+
+	/*
+	 * Enqueue and dequeue packets checking that each WRR queue dequeued
+	 * the correct number of packets.
+	 */
+	err = test_sched_wrr_enqueue_dequeue(subtest_name, mp, port, in_mbufs,
+					     out_mbufs, enqueue_packets,
+					     dequeue_packets,
+					     expected_counts);
+
+	/*
+	 * Free up allocated resources
+	 */
+	free(in_mbufs);
+	free(out_mbufs);
+	rte_sched_port_free(port);
+	delete_mempool(mp);
+
+	return err;
+}
+
+static int
+test_sched_wrr_even_weights(void)
+{
+	/*
+	 * Even weighting - each WRR queue should dequeue the same number of
+	 * packets
+	 */
+	uint16_t tc_qlengths[] = { 64, 64, 64, 64 };
+	uint8_t wrr_weights[] = { 1, 1, 1, 1,     // TC-0
+				  1, 1, 1, 1,     // TC-1
+				  1, 1, 1, 1,     // TC-2
+				  1, 1, 1, 1 };   // TC-3
+	int32_t enqueue_packets = 64 * 4;
+	int32_t dequeue_packets = 32;
+	uint32_t expected_counts[] = { 8, 8, 8, 8 };
+
+	return test_sched_wrr_test("wrr-even-weights", tc_qlengths, wrr_weights,
+				   enqueue_packets, dequeue_packets,
+				   expected_counts);
+}
+
+static int
+test_sched_wrr_8_4_2_1_weights(void)
+{
+	/*
+	 * Uneven weightings but with a low LCM of 8 * 4 * 2 * 1 = 64
+	 */
+	uint16_t tc_qlengths[] = { 64, 64, 64, 64 };
+	uint8_t wrr_weights[] = { 8, 4, 2, 1,     // TC-0
+				  1, 1, 1, 1,
+				  1, 1, 1, 1,
+				  1, 1, 1, 1 };
+	int32_t enqueue_packets = 64 * 4;
+	int32_t dequeue_packets = 15;
+	uint32_t expected_counts[] = { 8, 4, 2, 1 };
+
+	return test_sched_wrr_test("wrr-8-4-2-1-weights", tc_qlengths,
+				   wrr_weights, enqueue_packets,
+				   dequeue_packets, expected_counts);
+}
+
+static int
+test_sched_wrr_1_2_3_4_weights(void)
+{
+	/*
+	 * Uneven weightings with a low LCM, but weightings in reverse order
+	 * from previous test
+	 */
+	uint16_t tc_qlengths[] = { 64, 64, 64, 64 };
+	uint8_t wrr_weights[] = { 1, 2, 3, 4,
+				  1, 1, 1, 1,
+				  1, 1, 1, 1,
+				  1, 1, 1, 1 };
+	int32_t enqueue_packets = 64 * 4;
+	int32_t dequeue_packets = 40;
+	uint32_t expected_counts[] = { 4, 8, 12, 16 };
+
+	return test_sched_wrr_test("wrr-1-2-3-4-weights", tc_qlengths,
+				   wrr_weights, enqueue_packets,
+				   dequeue_packets, expected_counts);
+}
+
+static int
+test_sched_wrr_11_7_5_3_weights(void)
+{
+	/*
+	 * This test generates a LCM of 11 * 7 * 5 * 3 = 1155.
+	 * 1155 / 3 = 385 which is more than can be fitted in a uint8_t
+	 */
+	uint16_t tc_qlengths[] = { 64, 64, 64, 64 };
+	uint8_t wrr_weights[] = { 11, 7, 5, 3,
+				  1, 1, 1, 1,
+				  1, 1, 1, 1,
+				  1, 1, 1, 1 };
+	int32_t enqueue_packets = 64 * 4;
+	int32_t dequeue_packets = 26;
+	uint32_t expected_counts[] = { 11, 7, 5, 3 };
+
+	return test_sched_wrr_test("wrr-11-7-5-3-weights", tc_qlengths,
+				   wrr_weights, enqueue_packets,
+				   dequeue_packets, expected_counts);
+}
+
+static int
+test_sched_wrr_100_to_97_weights(void)
+{
+	/*
+	 * The ratios between the weightings is small how well will the
+	 * pseudo floating point wrr_cost perform.
+	 */
+	uint16_t tc_qlengths[] = { 128, 1, 1, 1 };
+	uint8_t wrr_weights[] = { 100, 99, 98, 97,
+				  1, 1, 1, 1,
+				  1, 1, 1, 1,
+				  1, 1, 1, 1 };
+	int32_t enqueue_packets = 128 * 4;
+	int32_t dequeue_packets = 394;
+	uint32_t expected_counts[] = { 100, 99, 98, 97 };
+
+	return test_sched_wrr_test("wrr-100-to-97-weights", tc_qlengths,
+				   wrr_weights, enqueue_packets,
+				   dequeue_packets, expected_counts);
+}
+
+static int
+test_sched_wrr_255_254_253_1_weights(void)
+{
+	/*
+	 * This test generates a LCM of 255 * 254 * 253 * 1 = 16386810
+	 * 16386810 / 255 =    64262 =   0xFB06
+	 * 16386810 / 1   = 16386810 = 0xFA0AFF
+	 * 0xFA0AFF needs to be shifted 16 bits left to fit into a uint8_t
+	 * but shifting 0xFB06 16 bits left results in a zero wrr_cost.
+	 *
+	 * This test can fail in two different ways.
+	 *
+	 * Very badly when the dequeued packet counts aren't even close to the
+	 * expected counts due a bug in a recent WRR patch.
+	 *
+	 * A very near miss, when the dequeued packet counts are just one or
+	 * two packets away from the expected counts.  This happens because
+	 * the weights of 255, 254, 254 and 1 get converted into wrr_costs
+	 * of 1, 1, 1, 250 respectively.  I think that this happens due to
+	 * rounding errors in the pseudo floating point code used by WRR.
+	 */
+	uint16_t tc_qlengths[] = { 1024, 1, 1, 1 };
+	uint8_t wrr_weights[] = { 254, 1, 253, 255,
+				  1, 1, 1, 1,
+				  1, 1, 1, 1,
+				  1, 1, 1, 1 };
+	int32_t enqueue_packets = 1024 * 4;
+	int32_t dequeue_packets = 255 + 254 + 253 + 1;
+	uint32_t expected_counts[] = { 254, 1, 253, 255 };
+
+	return test_sched_wrr_test("wrr-255-254-253-1-weights", tc_qlengths,
+				   wrr_weights, enqueue_packets,
+				   dequeue_packets, expected_counts);
+}
+
+/**
+ * WRR test main entrance for library sched
+ */
+static int
+test_sched_wrr(void)
+{
+	TEST_ASSERT_SUCCESS(test_sched_wrr_even_weights(),
+			    "even-weight test failed\n");
+	TEST_ASSERT_SUCCESS(test_sched_wrr_8_4_2_1_weights(),
+			    "8-4-2-1-weight test failed\n");
+	TEST_ASSERT_SUCCESS(test_sched_wrr_1_2_3_4_weights(),
+			    "1-2-3-4-weight test failed\n");
+	TEST_ASSERT_SUCCESS(test_sched_wrr_11_7_5_3_weights(),
+			    "11-7-5-3-weight test failed\n");
+	TEST_ASSERT_SUCCESS(test_sched_wrr_100_to_97_weights(),
+			    "100-to-97-weight test failed\n");
+	TEST_ASSERT_SUCCESS(test_sched_wrr_255_254_253_1_weights(),
+			    "255-254-253-1-weight test failed\n");
+	return 0;
+}
+
+REGISTER_TEST_COMMAND(sched_wrr_test, test_sched_wrr);
-- 
2.1.4

^ permalink raw reply	[flat|nested] 4+ messages in thread
end of thread, other threads:[~2018-01-03 14:03 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-11-30  9:05 [dpdk-dev] [PATCH v2] test: new sched WRR unit-test alangordondewar
2017-11-30 10:48 ` Luca Boccassi
2018-01-02 16:15 ` Dumitrescu, Cristian
2018-01-03 14:02   ` Dewar, Alan

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).