On Mon, 31 Aug 2026 18:40:48 +0300
Raslan Darawsheh <[email protected]> wrote:
> Hi,
>
>
> On 31/07/2026 12:23 PM, David Marchand wrote:
> > The kernel netdev promiscuous or all multicast mode could be enabled
> > before DPDK starts the port. This would cause an inconsistency between
> > the kernel state and DPDK's expected state
> > (dev->data->promiscuous / dev->data->all_multicast).
> >
> > Add mlx5_os_get_promisc() and mlx5_os_get_allmulti() to query the
> > current kernel state via netlink. On port start, compare the kernel
> > state with DPDK's state and sync if they differ.
> >
> > On Windows, these getters always return false as there seems to be no
> > API for getting the states.
> >
> > Fixes: dd4bb90bc3cb ("net/mlx5: use Netlink to enable promisc/allmulti
> > mode")
> > Cc: [email protected]
> >
> > Signed-off-by: David Marchand <[email protected]>
>
> Series applied to next-net-mlx,
>
> Kindest regards
> Raslan Darawsheh
>
AI review found that there is potential issue if tcpdump is used on kernel side.
963565124c reset promiscuous modes on close — behaviorally deliberate, but the
condition is kernel state only, not "did DPDK set this." If an operator had
promiscuous mode on independently (tcpdump, a bridge port), closing a DPDK
port
clears it and disrupts that unrelated user. Kernel promiscuity is refcounted
internally, but a netlink flag-set overrides that accounting rather than
decrementing it. Scoping to modes DPDK actually enabled would be safer.
