include/hw/virtio/virtio-net.h | 1 + hw/net/virtio-net.c | 49 +++++++++++++++++++++++++++++----- 2 files changed, 44 insertions(+), 6 deletions(-)
v2: use validate_features() to disable the guest driver rather than setting vring.num to 0. With failover, when the guest virtio-net driver doesn't support the STANDBY feature, the primary device is not plugged and only the virtio-net device is kept. Doing like that we can migrate the machine and keep the network connection. But in some cases, when performance is more important than availability we would prefer to keep the VFIO device rather than the virtio-net one, even if it means to lose the network connection during the migration of the machine. To do that we can't simply unplug the virtio-net device and plug the VFIO one because for the migration the initial state must be kept (virtio-net plugged, VFIO unplugged) but we can try to disable the virtio-net driver and plug the VFIO card, so the initial state is correct (the virtio-net card is plugged, but disabled in guest, and the VFIO card is unplugged before migration). This change doesn't impact the case when guest and host support the STANDBY feature. I've introduced the "failover-default" property to virtio-net device to set which device to keep (failover-default=true keeps the virtio-net device, =off the other one). For example, with a guest driver that doesn't support STANDBY: ... -device virtio-net-pci,id=virtio0,failover=on,failover-default=on \ -device vfio-pci,host=$PCI,id=hostdev0,failover_pair_id=virtio0 \ ... [root@localhost ~]# ip a 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN qlen 1 link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 inet 127.0.0.1/8 scope host lo valid_lft forever preferred_lft forever inet6 ::1/128 scope host valid_lft forever preferred_lft forever 2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP q0 link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff inet 192.168.20.2/24 brd 192.168.20.255 scope global eth0 valid_lft forever preferred_lft forever inet6 fe80::2428:c5ff:fe7f:1424/64 scope link valid_lft forever preferred_lft forever # ethtool -i eth0 driver: virtio_net version: 1.0.0 firmware-version: expansion-rom-version: bus-info: 0000:04:00.0 supports-statistics: no supports-test: no supports-eeprom-access: no supports-register-dump: no supports-priv-flags: no ... -device virtio-net-pci,id=virtio0,failover=on,failover-default=off \ -device vfio-pci,host=$PCI,id=hostdev0,failover_pair_id=virtio0 \ ... [root@localhost ~]# ip a 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN qlen 1 link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 inet 127.0.0.1/8 scope host lo valid_lft forever preferred_lft forever inet6 ::1/128 scope host valid_lft forever preferred_lft forever 2: enp2s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP qlen 100 link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff inet 192.168.20.2/24 brd 192.168.20.255 scope global enp2s0 valid_lft forever preferred_lft forever inet6 fe80::2428:c5ff:fe7f:1424/64 scope link valid_lft forever preferred_lft forever [root@localhost ~]# ethtool -i enp2s0 driver: i40evf version: 1.6.27-k firmware-version: N/A expansion-rom-version: bus-info: 0000:02:00.0 supports-statistics: yes supports-test: no supports-eeprom-access: no supports-register-dump: no supports-priv-flags: no With guest driver that supports STANDBY, we would always have: [root@localhost ~]# ip a 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group defau0 link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 inet 127.0.0.1/8 scope host lo valid_lft forever preferred_lft forever inet6 ::1/128 scope host valid_lft forever preferred_lft forever 2: enp4s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP gr0 link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff inet 192.168.20.2/24 brd 192.168.20.255 scope global noprefixroute enp4s0 valid_lft forever preferred_lft forever inet6 fe80::2428:c5ff:fe7f:1424/64 scope link valid_lft forever preferred_lft forever 3: enp4s0nsby: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel master0 link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff 4: enp2s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq master enp4s0 st0 link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff [root@localhost ~]# ethtool -i enp4s0 driver: net_failover version: 0.1 firmware-version: expansion-rom-version: bus-info: supports-statistics: no supports-test: no supports-eeprom-access: no supports-register-dump: no supports-priv-flags: no [root@localhost ~]# ethtool -i enp4s0nsby driver: virtio_net version: 1.0.0 firmware-version: expansion-rom-version: bus-info: 0000:04:00.0 supports-statistics: yes supports-test: no supports-eeprom-access: no supports-register-dump: no supports-priv-flags: no [root@localhost ~]# ethtool -i enp2s0 driver: iavf version: 4.18.0-310.el8.x86_64 firmware-version: N/A expansion-rom-version: bus-info: 0000:02:00.0 supports-statistics: yes supports-test: no supports-eeprom-access: no supports-register-dump: no supports-priv-flags: yes Laurent Vivier (1): virtio: failover: define the default device to use in case of error include/hw/virtio/virtio-net.h | 1 + hw/net/virtio-net.c | 49 +++++++++++++++++++++++++++++----- 2 files changed, 44 insertions(+), 6 deletions(-) -- 2.31.1
On Tue, Aug 10, 2021 at 1:13 AM Laurent Vivier <lvivier@redhat.com> wrote: > > v2: use validate_features() to disable the guest driver rather > than setting vring.num to 0. > > With failover, when the guest virtio-net driver doesn't support the > STANDBY feature, the primary device is not plugged and only the virtio-net > device is kept. Doing like that we can migrate the machine and > keep the network connection. > > But in some cases, when performance is more important than availability > we would prefer to keep the VFIO device rather than the virtio-net one, > even if it means to lose the network connection during the migration of > the machine. I think we still need to seek a way to recover the network after migration. Thanks > > To do that we can't simply unplug the virtio-net device and plug the > VFIO one because for the migration the initial state must be kept > (virtio-net plugged, VFIO unplugged) but we can try to disable the > virtio-net driver and plug the VFIO card, so the initial state is > correct (the virtio-net card is plugged, but disabled in guest, and > the VFIO card is unplugged before migration). > > This change doesn't impact the case when guest and host support > the STANDBY feature. > > I've introduced the "failover-default" property to virtio-net device > to set which device to keep (failover-default=true keeps the virtio-net > device, =off the other one). > > For example, with a guest driver that doesn't support STANDBY: > > ... > -device virtio-net-pci,id=virtio0,failover=on,failover-default=on \ > -device vfio-pci,host=$PCI,id=hostdev0,failover_pair_id=virtio0 \ > ... > > [root@localhost ~]# ip a > 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN qlen 1 > link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 > inet 127.0.0.1/8 scope host lo > valid_lft forever preferred_lft forever > inet6 ::1/128 scope host > valid_lft forever preferred_lft forever > 2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP q0 > link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff > inet 192.168.20.2/24 brd 192.168.20.255 scope global eth0 > valid_lft forever preferred_lft forever > inet6 fe80::2428:c5ff:fe7f:1424/64 scope link > valid_lft forever preferred_lft forever > # ethtool -i eth0 > driver: virtio_net > version: 1.0.0 > firmware-version: > expansion-rom-version: > bus-info: 0000:04:00.0 > supports-statistics: no > supports-test: no > supports-eeprom-access: no > supports-register-dump: no > supports-priv-flags: no > > ... > -device virtio-net-pci,id=virtio0,failover=on,failover-default=off \ > -device vfio-pci,host=$PCI,id=hostdev0,failover_pair_id=virtio0 \ > ... > > [root@localhost ~]# ip a > 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN qlen 1 > link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 > inet 127.0.0.1/8 scope host lo > valid_lft forever preferred_lft forever > inet6 ::1/128 scope host > valid_lft forever preferred_lft forever > 2: enp2s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP qlen 100 > link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff > inet 192.168.20.2/24 brd 192.168.20.255 scope global enp2s0 > valid_lft forever preferred_lft forever > inet6 fe80::2428:c5ff:fe7f:1424/64 scope link > valid_lft forever preferred_lft forever > [root@localhost ~]# ethtool -i enp2s0 > driver: i40evf > version: 1.6.27-k > firmware-version: N/A > expansion-rom-version: > bus-info: 0000:02:00.0 > supports-statistics: yes > supports-test: no > supports-eeprom-access: no > supports-register-dump: no > supports-priv-flags: no > > With guest driver that supports STANDBY, we would always have: > > [root@localhost ~]# ip a > 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group defau0 > link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 > inet 127.0.0.1/8 scope host lo > valid_lft forever preferred_lft forever > inet6 ::1/128 scope host > valid_lft forever preferred_lft forever > 2: enp4s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP gr0 > link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff > inet 192.168.20.2/24 brd 192.168.20.255 scope global noprefixroute enp4s0 > valid_lft forever preferred_lft forever > inet6 fe80::2428:c5ff:fe7f:1424/64 scope link > valid_lft forever preferred_lft forever > 3: enp4s0nsby: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel master0 > link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff > 4: enp2s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq master enp4s0 st0 > link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff > [root@localhost ~]# ethtool -i enp4s0 > driver: net_failover > version: 0.1 > firmware-version: > expansion-rom-version: > bus-info: > supports-statistics: no > supports-test: no > supports-eeprom-access: no > supports-register-dump: no > supports-priv-flags: no > [root@localhost ~]# ethtool -i enp4s0nsby > driver: virtio_net > version: 1.0.0 > firmware-version: > expansion-rom-version: > bus-info: 0000:04:00.0 > supports-statistics: yes > supports-test: no > supports-eeprom-access: no > supports-register-dump: no > supports-priv-flags: no > [root@localhost ~]# ethtool -i enp2s0 > driver: iavf > version: 4.18.0-310.el8.x86_64 > firmware-version: N/A > expansion-rom-version: > bus-info: 0000:02:00.0 > supports-statistics: yes > supports-test: no > supports-eeprom-access: no > supports-register-dump: no > supports-priv-flags: yes > > Laurent Vivier (1): > virtio: failover: define the default device to use in case of error > > include/hw/virtio/virtio-net.h | 1 + > hw/net/virtio-net.c | 49 +++++++++++++++++++++++++++++----- > 2 files changed, 44 insertions(+), 6 deletions(-) > > -- > 2.31.1 > >
On 11/08/2021 06:17, Jason Wang wrote: > On Tue, Aug 10, 2021 at 1:13 AM Laurent Vivier <lvivier@redhat.com> wrote: >> >> v2: use validate_features() to disable the guest driver rather >> than setting vring.num to 0. >> >> With failover, when the guest virtio-net driver doesn't support the >> STANDBY feature, the primary device is not plugged and only the virtio-net >> device is kept. Doing like that we can migrate the machine and >> keep the network connection. >> >> But in some cases, when performance is more important than availability >> we would prefer to keep the VFIO device rather than the virtio-net one, >> even if it means to lose the network connection during the migration of >> the machine. > > I think we still need to seek a way to recover the network after migration. If the network device is configured to have an IP address at boot, the IP will be restored when the card is hotplugged. The only difference with virtio-net is _during_ the migration. Witht virtio-net we keep the networking while the VM is migrated, without it we lose the networking when the VFIO card is unplugged and restored when it is plugged back. Thanks, Laurent > > Thanks > >> >> To do that we can't simply unplug the virtio-net device and plug the >> VFIO one because for the migration the initial state must be kept >> (virtio-net plugged, VFIO unplugged) but we can try to disable the >> virtio-net driver and plug the VFIO card, so the initial state is >> correct (the virtio-net card is plugged, but disabled in guest, and >> the VFIO card is unplugged before migration). >> >> This change doesn't impact the case when guest and host support >> the STANDBY feature. >> >> I've introduced the "failover-default" property to virtio-net device >> to set which device to keep (failover-default=true keeps the virtio-net >> device, =off the other one). >> >> For example, with a guest driver that doesn't support STANDBY: >> >> ... >> -device virtio-net-pci,id=virtio0,failover=on,failover-default=on \ >> -device vfio-pci,host=$PCI,id=hostdev0,failover_pair_id=virtio0 \ >> ... >> >> [root@localhost ~]# ip a >> 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN qlen 1 >> link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 >> inet 127.0.0.1/8 scope host lo >> valid_lft forever preferred_lft forever >> inet6 ::1/128 scope host >> valid_lft forever preferred_lft forever >> 2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP q0 >> link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff >> inet 192.168.20.2/24 brd 192.168.20.255 scope global eth0 >> valid_lft forever preferred_lft forever >> inet6 fe80::2428:c5ff:fe7f:1424/64 scope link >> valid_lft forever preferred_lft forever >> # ethtool -i eth0 >> driver: virtio_net >> version: 1.0.0 >> firmware-version: >> expansion-rom-version: >> bus-info: 0000:04:00.0 >> supports-statistics: no >> supports-test: no >> supports-eeprom-access: no >> supports-register-dump: no >> supports-priv-flags: no >> >> ... >> -device virtio-net-pci,id=virtio0,failover=on,failover-default=off \ >> -device vfio-pci,host=$PCI,id=hostdev0,failover_pair_id=virtio0 \ >> ... >> >> [root@localhost ~]# ip a >> 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN qlen 1 >> link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 >> inet 127.0.0.1/8 scope host lo >> valid_lft forever preferred_lft forever >> inet6 ::1/128 scope host >> valid_lft forever preferred_lft forever >> 2: enp2s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP qlen 100 >> link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff >> inet 192.168.20.2/24 brd 192.168.20.255 scope global enp2s0 >> valid_lft forever preferred_lft forever >> inet6 fe80::2428:c5ff:fe7f:1424/64 scope link >> valid_lft forever preferred_lft forever >> [root@localhost ~]# ethtool -i enp2s0 >> driver: i40evf >> version: 1.6.27-k >> firmware-version: N/A >> expansion-rom-version: >> bus-info: 0000:02:00.0 >> supports-statistics: yes >> supports-test: no >> supports-eeprom-access: no >> supports-register-dump: no >> supports-priv-flags: no >> >> With guest driver that supports STANDBY, we would always have: >> >> [root@localhost ~]# ip a >> 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group defau0 >> link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 >> inet 127.0.0.1/8 scope host lo >> valid_lft forever preferred_lft forever >> inet6 ::1/128 scope host >> valid_lft forever preferred_lft forever >> 2: enp4s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP gr0 >> link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff >> inet 192.168.20.2/24 brd 192.168.20.255 scope global noprefixroute enp4s0 >> valid_lft forever preferred_lft forever >> inet6 fe80::2428:c5ff:fe7f:1424/64 scope link >> valid_lft forever preferred_lft forever >> 3: enp4s0nsby: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel master0 >> link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff >> 4: enp2s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq master enp4s0 st0 >> link/ether 26:28:c5:7f:14:24 brd ff:ff:ff:ff:ff:ff >> [root@localhost ~]# ethtool -i enp4s0 >> driver: net_failover >> version: 0.1 >> firmware-version: >> expansion-rom-version: >> bus-info: >> supports-statistics: no >> supports-test: no >> supports-eeprom-access: no >> supports-register-dump: no >> supports-priv-flags: no >> [root@localhost ~]# ethtool -i enp4s0nsby >> driver: virtio_net >> version: 1.0.0 >> firmware-version: >> expansion-rom-version: >> bus-info: 0000:04:00.0 >> supports-statistics: yes >> supports-test: no >> supports-eeprom-access: no >> supports-register-dump: no >> supports-priv-flags: no >> [root@localhost ~]# ethtool -i enp2s0 >> driver: iavf >> version: 4.18.0-310.el8.x86_64 >> firmware-version: N/A >> expansion-rom-version: >> bus-info: 0000:02:00.0 >> supports-statistics: yes >> supports-test: no >> supports-eeprom-access: no >> supports-register-dump: no >> supports-priv-flags: yes >> >> Laurent Vivier (1): >> virtio: failover: define the default device to use in case of error >> >> include/hw/virtio/virtio-net.h | 1 + >> hw/net/virtio-net.c | 49 +++++++++++++++++++++++++++++----- >> 2 files changed, 44 insertions(+), 6 deletions(-) >> >> -- >> 2.31.1 >> >> >
© 2016 - 2024 Red Hat, Inc.