The first part of this description is taken from the bug report: https://gitlab.com/libvirt/libvirt/-/work_items/875 By default, a macvtap device will have the qdisc for the tap set to `noqueue` to prevent lock contention limiting the available bandwidth for the VM (see [BZ 1329644](https://bugzilla.redhat.com/show_bug.cgi?id=1329644)): ```xml <interface type='direct'> <mac address='52:54:00:f5:7e:97'/> <source dev='ens2f0np0' mode='passthrough'/> <target dev='macvtap0'/> <model type='virtio'/> <alias name='net1'/> <address type='pci' domain='0x0000' bus='0x02' slot='0x00' function='0x0'/> </interface> ``` ``` 12: macvtap0@ens2f0np0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 500 ``` However, defining an interface with only an `inbound` bandwidth limit sets the iface's root qdisc to the system default: ```xml <interface type='direct'> <mac address='52:54:00:05:8f:18'/> <source dev='ens2f0np0' mode='passthrough'/> <bandwidth> <inbound average='3125000' peak='3125000'/> </bandwidth> <target dev='macvtap5'/> <model type='virtio'/> <alias name='net0'/> <address type='pci' domain='0x0000' bus='0x01' slot='0x00' function='0x0'/> </interface> ``` ``` 16: macvtap5@ens2f0np0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel state UP group default qlen 500 ``` If I understand correctly, the `root` and `ingress` directions can be configured independently. So the above bandwidth definition should correspond to this (configured manually): ``` $ sudo tc qdisc show dev macvtap5 qdisc noqueue 8007: root refcnt 2 qdisc ingress ffff: parent ffff:fff1 ---------------- ``` Rather than: ``` $ sudo tc qdisc show dev macvtap6 qdisc fq_codel 0: root refcnt 2 limit 10240p flows 1024 quantum 1514 target 5ms interval 100ms memory_limit 32Mb ecn drop_batch 64 qdisc ingress ffff: parent ffff:fff1 ---------------- ``` When libvirt sets up a macvtap device, qemuDomainInterfaceSetDefaultQDisc configures qdisc noqueue for the interface. However, if bandwidth limits are defined, it also calls virNetDevBandwidthSet, which will call virNetDevBandwidthClear before reconfiguring the qdisc for the tx/rx. If only rx is defined, the root qdisc will be the system default. d53b0923539 prevents the noqueue configuration from being lost when reconfiguring bandwidth limits. All usages of virNetDevBandwidthSet are either covered by a call to qemuDomainInterfaceSetDefaultQDisc or qdisc noqueue does not apply there. This change doesn't change the semantics of virNetDevBandwidthClear. Alternatively, virNetDevBandwidthSet/virNetDevBandwidthClear could take a default qdisc which would be used to call virNetDevBandwidthSetRootQDisc in virNetDevBandwidthClear, removing the need for the two changes in d53b0923539. I've confirmed that the fix works when creating a domain with an ingress-only bandwidth limit. Please let me know what you think. Thanks! Signed-off-by: Wesley Hershberger <wesley.hershberger@canonical.com> --- Wesley Hershberger (2): virnetdevbandwidth: Split virNetDevBandwidthClear into two helpers virnetdevbandwidth: Only clear qdisc for defined directions src/util/virnetdevbandwidth.c | 70 +++++++++++++++++++++++++++++-------------- src/util/virnetdevbandwidth.h | 2 ++ 2 files changed, 49 insertions(+), 23 deletions(-) --- base-commit: 543280d8c8fe06526ce9871e7d25f46f6ee89017 change-id: 20260506-qdisc-clear-11a1f86407a2 Best regards, -- Wesley Hershberger <wesley.hershberger@canonical.com>