[dpdk-dev] Intel 82599 ixgbe VF flow director support ?

[dpdk-dev] Intel 82599 ixgbe VF flow director support ?

[Vincent Li]

Hi,

I am testing https://github.com/iqiyi/dpvs in KVM VM running Ubuntu with 
Intel 82599 SR-IOV VF.

got error like:

# ./dpvs
current thread affinity is set to F
EAL: Detected 4 lcore(s)
EAL: Probing VFIO support...
EAL: WARNING: cpu flags constant_tsc=no nonstop_tsc=no -> using unreliable 
clock cycles !
EAL: PCI device 0000:00:03.0 on NUMA socket -1
EAL:   Invalid NUMA socket, default to 0
EAL:   probe driver: 1af4:1000 net_virtio
EAL: PCI device 0000:00:09.0 on NUMA socket -1
EAL:   Invalid NUMA socket, default to 0
EAL:   probe driver: 8086:10ed net_ixgbe_vf
EAL: PCI device 0000:00:10.0 on NUMA socket -1
EAL:   Invalid NUMA socket, default to 0
EAL:   probe driver: 8086:10ed net_ixgbe_vf
DPVS: dpvs version: 1.8-4, build on 2020.06.04.23:21:22
CFG_FILE: Opening configuration file '/etc/dpvs.conf'.
CFG_FILE: log_level = WARNING
NETIF: dpdk0:rx_queue_number = 8
NETIF: worker cpu1:dpdk0 rx_queue_id += 0
NETIF: worker cpu1:dpdk0 tx_queue_id += 0
NETIF: worker cpu2:dpdk0 rx_queue_id += 1
NETIF: worker cpu2:dpdk0 tx_queue_id += 1
IPSET: ipset_init: lcore 3: nothing to do.
IPVS: dp_vs_conn_init: lcore 3: nothing to do.
NETIF: Ethdev port_id=0 invalid rss_hf: 0x3afbc, valid value: 0x0
NETIF: Ethdev port_id=0 invalid tx_offload: 0x1000e, valid value: 0x3f
NETIF: fail to flush FDIR filters for device dpdk0
DPVS: Start dpdk0 failed, skipping ...


It appears fail at:

static int fdir_filter_flush(const struct netif_port *port)
{
    if (!port || port->type != PORT_TYPE_GENERAL)
        return EDPVS_OK;
    if (rte_eth_dev_filter_ctrl(port->id, RTE_ETH_FILTER_FDIR,
                RTE_ETH_FILTER_FLUSH, NULL) < 0)
        return EDPVS_DPDKAPIFAIL;
    return EDPVS_OK;
}

no problem if I provision ixgbe PF as PCI passthrough to the VM.

there is a old thread "Signature filter for virtual function" 
https://marc.info/?t=146622468100052&r=1&w=2

"
According to 82599 datasheet it is possible to route packets to different
queues within VF. So in not VT mode rx_queue defines absolute queue 
number,
and in VT rx_queue defines relative queue number within VF.
I can make patch later when I will have free time.

Regards,
Vladimir
"

I found no support in DPDK ixgbe VF flow director in ixgbe commit history, 
could anyone confirm that is the case? if no support, any plan to support 
it?