Fix breaking vf-lag functionality in os-net-config
Because of racing issue to activate vf-lag after moving the second sriov_pf interface to switchdev mode in Nvidia\Mellanox nics, we may bind sriov_vfs while the LAG is not active yet. Another reason for breaking vf-lag functionality is that we are doing ifdown/ifup for sriov_pfs after binding the vfs(in case of linux_bond is member of ovs_bridge). As a solution for this issue, we are doing the binding after assuring the LAG is active, and also moving the ifdown/ifup before start binding Closes-Bug: #2020085 Change-Id: If0cad8c856ee62064205b9a88f0148980653fcb2
This commit is contained in:
waleedm
parent
0350a82f19
commit
b1a7c9c5f0
|
|
@ -1004,6 +1004,9 @@ class LinuxBond(_BaseOpts):
|
|||
self.bonding_options = bonding_options
|
||||
self.ethtool_opts = ethtool_opts
|
||||
for member in self.members:
|
||||
if isinstance(member, SriovPF):
|
||||
utils.update_sriov_pf_map(member.name, member.numvfs, False,
|
||||
lag_candidate=True)
|
||||
if isinstance(member, SriovVF):
|
||||
LinuxBond.update_vf_config(member)
|
||||
member.linux_bond_name = name
|
||||
|
|
|
|||
|
|
@ -50,6 +50,7 @@ MLNX5_VDPA_KMODS = [
|
|||
]
|
||||
|
||||
MAX_RETRIES = 10
|
||||
MLNX_LAG_PATH = "/sys/kernel/debug/mlx5/{pf_pci}/lag/state"
|
||||
PF_FUNC_RE = re.compile(r"\.(\d+)$", 0)
|
||||
|
||||
VF_PCI_RE = re.compile(r'/[\d]{4}\:(\d+):(\d+)\.(\d+)/net/[^\/]+$')
|
||||
|
|
@ -303,6 +304,7 @@ def configure_sriov_pf(execution_from_cli=False, restart_openvswitch=False):
|
|||
|
||||
sriov_map = common.get_sriov_map()
|
||||
dpdk_vfs_pcis_list = []
|
||||
vf_lag_sriov_pfs_list = []
|
||||
trigger_udev_rule = False
|
||||
|
||||
# Cleanup the previous config by puppet-tripleo
|
||||
|
|
@ -372,6 +374,10 @@ def configure_sriov_pf(execution_from_cli=False, restart_openvswitch=False):
|
|||
|
||||
# Configure switchdev mode
|
||||
configure_switchdev(item['name'])
|
||||
# Add sriovpf to vf_lag_sriov_pfs_list if it's
|
||||
# a linux bond member (lag_candidate)
|
||||
if item.get('lag_candidate', False):
|
||||
vf_lag_sriov_pfs_list.append(item['name'])
|
||||
# Adding a udev rule to rename vf-representors
|
||||
else:
|
||||
trigger_udev_rule = add_udev_rule_for_vdpa_representors(
|
||||
|
|
@ -387,6 +393,12 @@ def configure_sriov_pf(execution_from_cli=False, restart_openvswitch=False):
|
|||
if execution_from_cli:
|
||||
if_up_interface(item['name'])
|
||||
|
||||
if restart_openvswitch:
|
||||
restart_ovs_and_pfs_netdevs()
|
||||
|
||||
if vf_lag_sriov_pfs_list and execution_from_cli:
|
||||
_wait_for_lag_creation(vf_lag_sriov_pfs_list)
|
||||
|
||||
if dpdk_vfs_pcis_list and not vdpa:
|
||||
sriov_bind_pcis_map = {_MLNX_DRIVER: dpdk_vfs_pcis_list}
|
||||
if not execution_from_cli:
|
||||
|
|
@ -400,8 +412,6 @@ def configure_sriov_pf(execution_from_cli=False, restart_openvswitch=False):
|
|||
trigger_udev_rules()
|
||||
|
||||
udev_monitor_stop(observer)
|
||||
if restart_openvswitch:
|
||||
restart_ovs_and_pfs_netdevs()
|
||||
|
||||
|
||||
def _wait_for_uplink_rep_creation(pf_name):
|
||||
|
|
@ -416,6 +426,26 @@ def _wait_for_uplink_rep_creation(pf_name):
|
|||
raise RuntimeError(f"{pf_name}: Timeout waiting uplink representor")
|
||||
|
||||
|
||||
def _wait_for_lag_creation(lag_sriov_pf_list):
|
||||
for sriov_pf in lag_sriov_pf_list:
|
||||
pf_pci = get_pf_pci(sriov_pf)
|
||||
lag_path = MLNX_LAG_PATH.format(pf_pci=pf_pci)
|
||||
if os.path.exists(lag_path):
|
||||
for i in range(MAX_RETRIES):
|
||||
lag_state = common.get_file_data(lag_path).strip()
|
||||
if lag_state == "active":
|
||||
logger.info(f"VF-LAG is enabled for interface {sriov_pf}"
|
||||
f" after {i} retries")
|
||||
break
|
||||
time.sleep(1)
|
||||
else:
|
||||
raise RuntimeError("VF-LAG is not created for interface"
|
||||
f" {sriov_pf} after {i} retries")
|
||||
else:
|
||||
logger.warning(f"Lag path {lag_path} does not exist for this "
|
||||
"kernel, skipping..")
|
||||
|
||||
|
||||
def create_rep_link_name_script():
|
||||
with open(_REP_LINK_NAME_FILE, "w") as f:
|
||||
f.write(_REP_LINK_NAME_DATA)
|
||||
|
|
|
|||
|
|
@ -243,6 +243,82 @@ class TestUtils(base.TestCase):
|
|||
self.assertEqual(1, len(pf_map))
|
||||
self.assertListEqual(pf_final, pf_map)
|
||||
|
||||
def test_update_sriov_pf_map_new_with_lag_candidate(self):
|
||||
def get_numvfs_stub(pf_name):
|
||||
return 0
|
||||
self.stub_out('os_net_config.sriov_config.get_numvfs',
|
||||
get_numvfs_stub)
|
||||
utils.update_sriov_pf_map('eth1', 10, False, lag_candidate=True)
|
||||
contents = common.get_file_data(common.SRIOV_CONFIG_FILE)
|
||||
sriov_pf_map = yaml.safe_load(contents) if contents else []
|
||||
self.assertEqual(1, len(sriov_pf_map))
|
||||
test_sriov_pf_map = [{'device_type': 'pf', 'link_mode': 'legacy',
|
||||
'name': 'eth1', 'numvfs': 10, 'vdpa': False,
|
||||
'lag_candidate': True}]
|
||||
self.assertListEqual(test_sriov_pf_map, sriov_pf_map)
|
||||
|
||||
def test_update_sriov_pf_map_exist_with_lag_candidate(self):
|
||||
def get_numvfs_stub(pf_name):
|
||||
return 10
|
||||
self.stub_out('os_net_config.sriov_config.get_numvfs',
|
||||
get_numvfs_stub)
|
||||
pf_initial = [{'device_type': 'pf', 'link_mode': 'legacy',
|
||||
'name': 'eth1', 'numvfs': 10, 'promisc': 'on',
|
||||
'vdpa': False, 'lag_candidate': False}]
|
||||
utils.write_yaml_config(common.SRIOV_CONFIG_FILE, pf_initial)
|
||||
|
||||
utils.update_sriov_pf_map('eth1', 10, False, lag_candidate=True)
|
||||
pf_final = [{'device_type': 'pf', 'link_mode': 'legacy',
|
||||
'name': 'eth1', 'numvfs': 10, 'promisc': 'on',
|
||||
'vdpa': False, 'lag_candidate': True}]
|
||||
contents = common.get_file_data(common.SRIOV_CONFIG_FILE)
|
||||
|
||||
pf_map = yaml.safe_load(contents) if contents else []
|
||||
self.assertEqual(1, len(pf_map))
|
||||
self.assertListEqual(pf_final, pf_map)
|
||||
|
||||
def test_update_sriov_pf_map_exist_with_lag_candidate_not_exist_true(
|
||||
self):
|
||||
def get_numvfs_stub(pf_name):
|
||||
return 10
|
||||
self.stub_out('os_net_config.sriov_config.get_numvfs',
|
||||
get_numvfs_stub)
|
||||
pf_initial = [{'device_type': 'pf', 'link_mode': 'legacy',
|
||||
'name': 'eth1', 'numvfs': 10, 'promisc': 'on',
|
||||
'vdpa': False}]
|
||||
utils.write_yaml_config(common.SRIOV_CONFIG_FILE, pf_initial)
|
||||
|
||||
utils.update_sriov_pf_map('eth1', 10, False, lag_candidate=True)
|
||||
pf_final = [{'device_type': 'pf', 'link_mode': 'legacy',
|
||||
'name': 'eth1', 'numvfs': 10, 'promisc': 'on',
|
||||
'vdpa': False, 'lag_candidate': True}]
|
||||
contents = common.get_file_data(common.SRIOV_CONFIG_FILE)
|
||||
|
||||
pf_map = yaml.safe_load(contents) if contents else []
|
||||
self.assertEqual(1, len(pf_map))
|
||||
self.assertListEqual(pf_final, pf_map)
|
||||
|
||||
def test_update_sriov_pf_map_exist_with_lag_candidate_not_exist_false(
|
||||
self):
|
||||
def get_numvfs_stub(pf_name):
|
||||
return 10
|
||||
self.stub_out('os_net_config.sriov_config.get_numvfs',
|
||||
get_numvfs_stub)
|
||||
pf_initial = [{'device_type': 'pf', 'link_mode': 'legacy',
|
||||
'name': 'eth1', 'numvfs': 10, 'promisc': 'on',
|
||||
'vdpa': False}]
|
||||
utils.write_yaml_config(common.SRIOV_CONFIG_FILE, pf_initial)
|
||||
|
||||
utils.update_sriov_pf_map('eth1', 10, False, lag_candidate=False)
|
||||
pf_final = [{'device_type': 'pf', 'link_mode': 'legacy',
|
||||
'name': 'eth1', 'numvfs': 10, 'promisc': 'on',
|
||||
'vdpa': False, 'lag_candidate': False}]
|
||||
contents = common.get_file_data(common.SRIOV_CONFIG_FILE)
|
||||
|
||||
pf_map = yaml.safe_load(contents) if contents else []
|
||||
self.assertEqual(1, len(pf_map))
|
||||
self.assertListEqual(pf_final, pf_map)
|
||||
|
||||
def test_update_sriov_vf_map_minimal_new(self):
|
||||
utils.update_sriov_vf_map('eth1', 2, 'eth1_2')
|
||||
contents = common.get_file_data(common.SRIOV_CONFIG_FILE)
|
||||
|
|
|
|||
|
|
@ -375,7 +375,8 @@ def _update_dpdk_map(ifname, pci_address, mac_address, driver):
|
|||
|
||||
|
||||
def update_sriov_pf_map(ifname, numvfs, noop, promisc=None,
|
||||
link_mode='legacy', vdpa=False, steering_mode=None):
|
||||
link_mode='legacy', vdpa=False, steering_mode=None,
|
||||
lag_candidate=None):
|
||||
if not noop:
|
||||
cur_numvfs = sriov_config.get_numvfs(ifname)
|
||||
if cur_numvfs > 0 and cur_numvfs != numvfs:
|
||||
|
|
@ -391,6 +392,8 @@ def update_sriov_pf_map(ifname, numvfs, noop, promisc=None,
|
|||
item['link_mode'] = link_mode
|
||||
if steering_mode is not None:
|
||||
item['steering_mode'] = steering_mode
|
||||
if lag_candidate is not None:
|
||||
item['lag_candidate'] = lag_candidate
|
||||
break
|
||||
else:
|
||||
new_item = {}
|
||||
|
|
@ -403,6 +406,8 @@ def update_sriov_pf_map(ifname, numvfs, noop, promisc=None,
|
|||
new_item['link_mode'] = link_mode
|
||||
if steering_mode is not None:
|
||||
new_item['steering_mode'] = steering_mode
|
||||
if lag_candidate is not None:
|
||||
new_item['lag_candidate'] = lag_candidate
|
||||
sriov_map.append(new_item)
|
||||
|
||||
write_yaml_config(common.SRIOV_CONFIG_FILE, sriov_map)
|
||||
|
|
@ -516,8 +521,9 @@ def _configure_sriov_config_service():
|
|||
|
||||
def configure_sriov_pfs(execution_from_cli=False, restart_openvswitch=False):
|
||||
logger.info("Configuring PFs now")
|
||||
sriov_config.configure_sriov_pf(execution_from_cli=execution_from_cli,
|
||||
restart_openvswitch=restart_openvswitch)
|
||||
sriov_config.configure_sriov_pf(
|
||||
execution_from_cli=execution_from_cli,
|
||||
restart_openvswitch=restart_openvswitch)
|
||||
_configure_sriov_config_service()
|
||||
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue