[Bug 273559] [iflib] numa allocation

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=273559

Zhenlei Huang <zlei@FreeBSD.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |zlei@FreeBSD.org

--- Comment #1 from Zhenlei Huang <zlei@FreeBSD.org> ---
(In reply to Konrad from comment #0)

The lagg(4) has no changes related to NUMA since Nov 19 2021.

From the logic in `sys/net/if_lagg.c`: 
```
struct lagg_port *
lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m, int *err)
{
        struct lacp_softc *lsc = LACP_SOFTC(sc);
        uint32_t hash;
        uint8_t numa_domain;

        if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) &&
            M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
                hash = m->m_pkthdr.flowid >> sc->flowid_shift;
        else
                hash = m_ether_tcpip_hash(sc->sc_flags, m, lsc->lsc_hashkey);

        numa_domain = m->m_pkthdr.numa_domain;
        return (lacp_select_tx_port_by_hash(sc, hash, numa_domain, err));
}
```

I guess `numa_domain` in mbuf packet header is not set correctly. That should
be done in driver IIUC.

I do not have NUMA devices, can you please share the debug info for
`m->m_pkthdr.numa_domain` ?

-- 
You are receiving this mail because:
You are the assignee for the bug.