From: Yaqi Tang <yaqi.tang@intel.com>
To: dts@dpdk.org
Cc: Yaqi Tang <yaqi.tang@intel.com>
Subject: [dts][PATCH V3 3/3] tests/ice_iavf_packet_pacing: ice enable packet pacing
Date: Thu, 11 Aug 2022 11:16:58 +0000 [thread overview]
Message-ID: <20220811111658.78579-4-yaqi.tang@intel.com> (raw)
In-Reply-To: <20220811111658.78579-1-yaqi.tang@intel.com>
ICE support queue rate limitation configure, and parse the quanta_size for IGP pacing.
Signed-off-by: Yaqi Tang <yaqi.tang@intel.com>
---
tests/TestSuite_ice_iavf_packet_pacing.py | 641 ++++++++++++++++++++++
1 file changed, 641 insertions(+)
create mode 100644 tests/TestSuite_ice_iavf_packet_pacing.py
diff --git a/tests/TestSuite_ice_iavf_packet_pacing.py b/tests/TestSuite_ice_iavf_packet_pacing.py
new file mode 100644
index 00000000..46b8e341
--- /dev/null
+++ b/tests/TestSuite_ice_iavf_packet_pacing.py
@@ -0,0 +1,641 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright(c) 2010-2022 Intel Corporation
+#
+
+import copy
+import os
+import re
+import time
+from pprint import pformat
+
+from framework.packet import Packet
+from framework.pktgen import PacketGeneratorHelper
+from framework.pmd_output import PmdOutput
+from framework.settings import HEADER_SIZE
+from framework.test_case import TestCase
+from framework.utils import GREEN, RED
+
+
+class IAVFPacketPacingTest(TestCase):
+ def set_up_all(self):
+ """
+ Run at the start of each test suite.
+ """
+ self.verify(
+ self.nic in ["ICE_25G-E810C_SFP", "ICE_100G-E810C_QSFP"],
+ "%s nic not support vf timestamp" % self.nic,
+ )
+ self.dut_ports = self.dut.get_ports(self.nic)
+ self.ports_socket = self.dut.get_numa_id(self.dut_ports[0])
+ # Verify that enough ports are available
+ self.verify(len(self.dut_ports) >= 1, "Insufficient ports")
+ self.tester_port0 = self.tester.get_local_port(self.dut_ports[0])
+ self.tester_iface0 = self.tester.get_interface(self.tester_port0)
+ self.pkt = Packet()
+ self.pmdout = PmdOutput(self.dut)
+
+ self.vf_driver = self.get_suite_cfg()["vf_driver"]
+ if self.vf_driver is None:
+ self.vf_driver = "vfio-pci"
+ self.pf0_intf = self.dut.ports_info[self.dut_ports[0]]["intf"]
+
+ # get dts output path
+ if self.logger.log_path.startswith(os.sep):
+ self.output_path = self.logger.log_path
+ else:
+ cur_path = os.path.dirname(os.path.dirname(os.path.realpath(__file__)))
+ self.output_path = os.sep.join([cur_path, self.logger.log_path])
+ # create an instance to set stream field setting
+ self.pktgen_helper = PacketGeneratorHelper()
+ self.path = self.dut.apps_name["test-pmd"]
+
+ def set_up(self):
+ """
+ Run before each test case.
+ """
+ pass
+
+ def launch_testpmd(self, allowlist, line_option=""):
+ """
+ start testpmd
+ """
+ # Prepare testpmd EAL and parameters
+ output = self.pmdout.start_testpmd(
+ socket=self.ports_socket,
+ prefix="tx",
+ eal_param=allowlist,
+ param=line_option,
+ )
+ # test link status
+ res = self.pmdout.wait_link_status_up("all", timeout=15)
+ self.verify(res is True, "there have port link is down")
+ return output
+
+ def close_testpmd(self):
+ self.dut.send_expect("quit", "# ", 30)
+ time.sleep(5)
+
+ def create_vf(self, vf_num):
+ self.dut.bind_interfaces_linux("ice")
+ self.dut.generate_sriov_vfs_by_port(self.dut_ports[0], vf_num)
+ self.sriov_vfs_port = self.dut.ports_info[self.dut_ports[0]]["vfs_port"]
+ self.dut.send_expect("ifconfig %s up" % self.pf0_intf, "# ")
+ if vf_num >= 2:
+ self.dut.send_expect(
+ "ip link set dev %s vf 0 trust on" % self.pf0_intf, "# "
+ )
+ self.dut.send_expect(
+ "ip link set %s vf 1 mac 00:11:22:33:44:55" % self.pf0_intf, "#"
+ )
+ self.vf0_pci = self.sriov_vfs_port[0].pci
+ self.vf1_pci = self.sriov_vfs_port[1].pci
+ else:
+ self.dut.send_expect(
+ "ip link set %s vf 0 mac 00:11:22:33:44:55" % self.pf0_intf, "#"
+ )
+ self.vf0_pci = self.sriov_vfs_port[0].pci
+ try:
+ for port in self.sriov_vfs_port:
+ port.bind_driver(self.vf_driver)
+ except Exception as e:
+ self.destroy_vf()
+ raise Exception(e)
+
+ def destroy_vf(self):
+ self.dut.destroy_sriov_vfs_by_port(self.dut_ports[0])
+ self.dut.send_expect("rmmod ice", "# ", 15)
+ self.dut.send_expect("modprobe ice", "# ", 15)
+
+ def testpmd_query_stats(self):
+ output = self.dut.send_expect("show port stats all", "testpmd> ", 20)
+ self.logger.info(output)
+ if not output:
+ return
+ port_pat = ".*NIC statistics for (port \d+) .*"
+ tx_pat = ".*Tx-pps:\s+(\d+)\s+Tx-bps:\s+(\d+).*"
+ port = re.findall(port_pat, output, re.M)
+ tx = re.findall(tx_pat, output, re.M)
+ if not port or not tx:
+ return
+ stat = {}
+ for port_id, (tx_pps, tx_bps) in zip(port, tx):
+ stat[port_id] = {
+ "tx_pps": float(tx_pps),
+ "tx_bps": float(tx_bps),
+ }
+ self.pmd_stat = stat
+
+ def get_queue_packets_stats(self, port):
+ output = self.dut.send_expect("stop", "testpmd> ")
+ self.logger.info(output)
+ p = re.compile("TX Port= %d/Queue=.*\n.*TX-packets: ([0-9]+)\s" % port)
+ queue_pkts = p.findall(output)
+ queue_pkts = list(map(int, queue_pkts))
+ if not queue_pkts:
+ return {}
+ return queue_pkts
+
+ def is_expected_throughput(self, expected, pmd_stat):
+ _expected, unit, port = expected
+ port = port
+ real_stat = pmd_stat.get(f"port {port}", {})
+ key = "tx_bps"
+ real_bps = real_stat.get(key) or 0
+ if real_bps == 0 and _expected == 0:
+ return True
+ if not _expected:
+ return False
+ bias = 10
+ if unit == "MBps":
+ _bias = 100 * abs((real_bps / 8 / 1e6 - _expected) / _expected)
+ return _bias < bias
+ return True
+
+ def set_fields(self):
+ fields_config = {
+ "ip": {
+ "src": {
+ "start": "198.18.0.0",
+ "end": "198.18.0.255",
+ "step": 1,
+ "action": "random",
+ },
+ },
+ }
+ return fields_config
+
+ def check_traffic(self, expected, frame_size=64):
+ # create pcap file
+ dmac = "00:11:22:33:44:55"
+ payload_size = frame_size - HEADER_SIZE["ip"] - HEADER_SIZE["eth"]
+ pcap = os.sep.join([self.output_path, "dts0.pcap"])
+ self.tester.scapy_append(
+ 'wrpcap("%s", [Ether(dst="%s")/IP(src="198.18.0.0",dst="198.28.0.0")/("X"*%d)])'
+ % (pcap, dmac, payload_size)
+ )
+ self.tester.scapy_execute()
+ tgen_input = []
+ tgen_input.append(
+ (
+ self.tester.get_local_port(self.dut_ports[0]),
+ self.tester.get_local_port(self.dut_ports[0]),
+ "%s" % pcap,
+ )
+ )
+
+ # clear streams before add new streams
+ vm_config = self.set_fields()
+ self.tester.pktgen.clear_streams()
+
+ streams = self.pktgen_helper.prepare_stream_from_tginput(
+ tgen_input, 100, vm_config, self.tester.pktgen
+ )
+ traffic_opt = {
+ "method": "throughput",
+ "duration": 20,
+ "interval": 15,
+ "callback": self.testpmd_query_stats,
+ }
+ results = []
+ result = self.tester.pktgen.measure_throughput(
+ stream_ids=[0], options=traffic_opt
+ )
+
+ pmd_stat = self.pmd_stat
+ status = self.is_expected_throughput(expected, pmd_stat)
+ msg = (
+ f"{pformat(expected)}"
+ " not get expected throughput value, real is: "
+ f"{pformat(pmd_stat)}"
+ )
+ self.verify(status, msg)
+
+ _, _, port = expected
+ queue_pkts = self.get_queue_packets_stats(port)
+ results.append([result, pmd_stat, queue_pkts])
+ return results
+
+ def check_queue_pkts_ratio(self, expected, result):
+ total_pkts = sum(result)
+ total_ratio = sum(expected)
+ ratio = []
+ for idx, result in enumerate(result):
+ percentage = 100 * result / total_pkts
+ ratio.append(percentage)
+ bias = 10
+ for idx, percentage in enumerate(expected):
+ percentage = 100 * percentage / total_ratio
+ _bias = 100 * abs(ratio[idx] - percentage) / percentage
+ self.logger.info((ratio[idx], percentage))
+ if _bias < bias:
+ continue
+ else:
+ msg = "can not get expected queue ratio"
+ self.verify(False, msg)
+
+ def check_output(self, expected, output):
+ if isinstance(expected, str):
+ expected = [expected]
+ for _expected in expected:
+ self.verify(
+ _expected in output, f"expected <{_expected}> message not display"
+ )
+
+ def verify_without_quanta_size_check_peak_tb_rate(self):
+ allowlist = f"-a {self.vf0_pci}"
+ line_option = ""
+ self.launch_testpmd(allowlist, line_option)
+ self.dut.send_expect("port stop all", "testpmd> ")
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 1 1000000 0 612980769 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 1000 -1 0 1 0 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 900 1000 0 1 1 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 0 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect("port tm hierarchy commit 0 no", "testpmd> ")
+ self.dut.send_expect("port start all", "testpmd> ")
+ self.dut.send_expect("set fwd mac", "testpmd> ")
+ self.dut.send_expect("start", "testpmd> ")
+
+ expected_throughput = (612, "MBps", 0)
+
+ results = self.check_traffic(expected_throughput)
+
+ def verify_single_queue_check_peak_tb_rate(self):
+ allowlist = f"-a {self.vf0_pci},quanta_size=1024"
+ line_option = ""
+ self.launch_testpmd(allowlist, line_option)
+ self.dut.send_expect("port stop all", "testpmd> ")
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 1 1000000 0 612980769 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 1000 -1 0 1 0 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 900 1000 0 1 1 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 0 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect("port tm hierarchy commit 0 no", "testpmd> ")
+ self.dut.send_expect("port start all", "testpmd> ")
+ self.dut.send_expect("set fwd mac", "testpmd> ")
+ self.dut.send_expect("start", "testpmd> ")
+
+ expected_throughput = (612, "MBps", 0)
+
+ results = self.check_traffic(expected_throughput)
+
+ def verify_multi_queues_check_peak_tb_rate(self):
+ allowlist = f"-a {self.vf0_pci},quanta_size=1024"
+ line_option = "--rxq=8 --txq=8"
+ self.launch_testpmd(allowlist, line_option)
+ self.dut.send_expect("port stop all", "testpmd> ")
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 1 1000000 0 100000000 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 1000 -1 0 1 0 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 900 1000 0 1 1 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 0 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 1 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 2 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 3 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 4 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 5 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 6 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 7 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect("port tm hierarchy commit 0 no", "testpmd> ")
+ self.dut.send_expect("port start all", "testpmd> ")
+ self.dut.send_expect("set fwd mac", "testpmd> ")
+ self.dut.send_expect("start", "testpmd> ")
+
+ expected_throughput = (800, "MBps", 0)
+ expected_ratio = [1, 1, 1, 1, 1, 1, 1, 1]
+
+ results = self.check_traffic(expected_throughput)
+ self.check_queue_pkts_ratio(expected_ratio, results[-1][-1])
+
+ def verify_modify_quanta_size_check_peak_tb_rate(self):
+ allowlist = f"-a {self.vf0_pci},quanta_size=4096"
+ line_option = "--rxq=8 --txq=8"
+ self.launch_testpmd(allowlist, line_option)
+ self.dut.send_expect("port stop all", "testpmd> ")
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 1 1000000 0 100000000 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 1000 -1 0 1 0 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 900 1000 0 1 1 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 0 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 1 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 2 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 3 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 4 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 5 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 6 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 7 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect("port tm hierarchy commit 0 no", "testpmd> ")
+ self.dut.send_expect("port start all", "testpmd> ")
+ self.dut.send_expect("set fwd mac", "testpmd> ")
+ self.dut.send_expect("start", "testpmd> ")
+
+ expected_throughput = (800, "MBps", 0)
+ expected_ratio = [1, 1, 1, 1, 1, 1, 1, 1]
+
+ results = self.check_traffic(expected_throughput)
+ self.check_queue_pkts_ratio(expected_ratio, results[-1][-1])
+
+ def verify_multi_queues_with_diff_rate_limit_check_peak_tb_rate(self):
+ allowlist = f"-a {self.vf0_pci},quanta_size=1024"
+ line_option = "--rxq=8 --txq=8"
+ self.launch_testpmd(allowlist, line_option)
+ self.dut.send_expect("port stop all", "testpmd> ")
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 1 1000000 0 10000000 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 2 1000000 0 20000000 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 3 1000000 0 30000000 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 4 1000000 0 40000000 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 5 1000000 0 50000000 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 6 1000000 0 60000000 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 7 1000000 0 70000000 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 8 1000000 0 80000000 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 1000 -1 0 1 0 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 900 1000 0 1 1 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 0 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 1 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 2 900 0 1 2 3 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 3 900 0 1 2 4 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 4 900 0 1 2 5 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 5 900 0 1 2 6 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 6 900 0 1 2 7 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 7 900 0 1 2 8 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect("port tm hierarchy commit 0 no", "testpmd> ")
+ self.dut.send_expect("port start all", "testpmd> ")
+ self.dut.send_expect("set fwd mac", "testpmd> ")
+ self.dut.send_expect("start", "testpmd> ")
+
+ expected_throughput = (360, "MBps", 0)
+ expected_ratio = [1, 2, 3, 4, 5, 6, 7, 8]
+
+ results = self.check_traffic(expected_throughput)
+ self.check_queue_pkts_ratio(expected_ratio, results[-1][-1])
+
+ def verify_port_rate_limit_less_than_queue_rate_limit(self):
+ allowlist = f"-a {self.vf0_pci},cap=dcf -a {self.vf1_pci},quanta_size=1024"
+ line_option = "--rxq=8 --txq=8 --port-topology=loop"
+ self.launch_testpmd(allowlist, line_option)
+ self.dut.send_expect("port stop all", "testpmd> ")
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 1 1000000 0 100000000 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm node shaper profile 1 2 1000000 0 612980769 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 1000 -1 0 1 0 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 900 1000 0 1 1 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 0 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 1 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 1 1000 -1 0 1 0 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 1 900 1000 0 1 1 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 0 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 1 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 2 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 3 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 4 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 5 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 6 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 7 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect("port tm hierarchy commit 0 no", "testpmd> ")
+ self.dut.send_expect("port tm hierarchy commit 1 no", "testpmd> ")
+ self.dut.send_expect("port start all", "testpmd> ")
+ self.dut.send_expect("set fwd mac", "testpmd> ")
+ self.dut.send_expect("start", "testpmd> ")
+
+ expected_throughput = (100, "MBps", 1)
+ expected_ratio = [1, 1, 1, 1, 1, 1, 1, 1]
+
+ results = self.check_traffic(expected_throughput)
+ self.check_queue_pkts_ratio(expected_ratio, results[-1][-1])
+
+ def verify_port_rate_limit_more_than_queue_rate_limit(self):
+ allowlist = f"-a {self.vf0_pci},cap=dcf -a {self.vf1_pci},quanta_size=1024"
+ line_option = "--rxq=8 --txq=8 --port-topology=loop"
+ self.launch_testpmd(allowlist, line_option)
+ self.dut.send_expect("port stop all", "testpmd> ")
+ self.dut.send_expect(
+ "add port tm node shaper profile 0 1 1000000 0 200000000 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm node shaper profile 1 2 1000000 0 10000000 0 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 1000 -1 0 1 0 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 0 900 1000 0 1 1 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 0 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 0 1 900 0 1 2 1 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 1 1000 -1 0 1 0 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm nonleaf node 1 900 1000 0 1 1 -1 1 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 0 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 1 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 2 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 3 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 4 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 5 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 6 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect(
+ "add port tm leaf node 1 7 900 0 1 2 2 0 0xffffffff 0 0", "testpmd> "
+ )
+ self.dut.send_expect("port tm hierarchy commit 0 no", "testpmd> ")
+ self.dut.send_expect("port tm hierarchy commit 1 no", "testpmd> ")
+ self.dut.send_expect("port start all", "testpmd> ")
+ self.dut.send_expect("set fwd mac", "testpmd> ")
+ self.dut.send_expect("start", "testpmd> ")
+
+ expected_throughput = (80, "MBps", 1)
+ expected_ratio = [1, 1, 1, 1, 1, 1, 1, 1]
+
+ results = self.check_traffic(expected_throughput)
+ self.check_queue_pkts_ratio(expected_ratio, results[-1][-1])
+
+ def test_perf_without_quanta_size_check_peak_tb_rate(self):
+ self.create_vf(vf_num=1)
+ self.verify_without_quanta_size_check_peak_tb_rate()
+
+ def test_perf_single_queue_check_peak_tb_rate(self):
+ self.create_vf(vf_num=1)
+ self.verify_single_queue_check_peak_tb_rate()
+
+ def test_perf_multi_queues_check_peak_tb_rate(self):
+ self.create_vf(vf_num=1)
+ self.verify_multi_queues_check_peak_tb_rate()
+
+ def test_perf_modify_quanta_size_check_peak_tb_rate(self):
+ self.create_vf(vf_num=1)
+ self.verify_modify_quanta_size_check_peak_tb_rate()
+
+ def test_perf_invalid_quanta_size_check_peak_tb_rate(self):
+ self.create_vf(vf_num=1)
+ output = self.launch_testpmd(
+ allowlist=f"-a {self.vf0_pci},quanta_size=1000", line_option=""
+ )
+ expected = "iavf_parse_devargs(): invalid quanta size"
+ self.check_output(expected, output)
+
+ def test_perf_multi_queues_with_diff_rate_limit_check_peak_tb_rate(self):
+ self.create_vf(vf_num=1)
+ self.verify_multi_queues_with_diff_rate_limit_check_peak_tb_rate()
+
+ def test_perf_port_rate_limit_less_than_queue_rate_limit(self):
+ self.create_vf(vf_num=2)
+ self.verify_port_rate_limit_less_than_queue_rate_limit()
+
+ def test_perf_port_rate_limit_more_than_queue_rate_limit(self):
+ self.create_vf(vf_num=2)
+ self.verify_port_rate_limit_more_than_queue_rate_limit()
+
+ def tear_down(self):
+ """
+ Run after each test case.
+ """
+ self.close_testpmd()
+ self.destroy_vf()
+
+ def tear_down_all(self):
+ """
+ Run after each test suite.
+ """
+ self.dut.kill_all()
--
2.25.1
next prev parent reply other threads:[~2022-08-11 3:18 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-08-11 11:16 [PATCH V3 0/3] ice iavf " Yaqi Tang
2022-08-11 11:16 ` [dts][PATCH V3 1/3] test_plans/index: add new test plan for ice_iavf_packet_pacing Yaqi Tang
2022-08-12 7:40 ` Peng, Yuan
2022-08-11 11:16 ` [dts][PATCH V3 2/3] test_plans/ice_iavf_packet_pacing: ice enable packet pacing Yaqi Tang
2022-08-24 7:56 ` Tu, Lijuan
2022-08-11 11:16 ` Yaqi Tang [this message]
2022-08-22 7:52 ` [dts][PATCH V3 3/3] tests/ice_iavf_packet_pacing: " Chen, LingliX
2022-09-14 7:41 ` lijuan.tu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20220811111658.78579-4-yaqi.tang@intel.com \
--to=yaqi.tang@intel.com \
--cc=dts@dpdk.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).