On Mon, Mar 25, 2024 at 2:58 PM Heng Qi <hengqi@xxxxxxxxxxxxxxxxx> wrote:
>
>
>
> 在 2024/3/25 下午2:29, Jason Wang 写道:
> > On Mon, Mar 25, 2024 at 10:11 AM Heng Qi <hengqi@xxxxxxxxxxxxxxxxx> wrote:
> >>
> >>
> >> 在 2024/3/22 下午1:17, Jason Wang 写道:
> >>> On Thu, Mar 21, 2024 at 7:46 PM Heng Qi <hengqi@xxxxxxxxxxxxxxxxx> wrote:
> >>>> When the dim worker is scheduled, if it fails to acquire the lock,
> >>>> dim may not be able to return to the working state later.
> >>>>
> >>>> For example, the following single queue scenario:
> >>>> 1. The dim worker of rxq0 is scheduled, and the dim status is
> >>>> changed to DIM_APPLY_NEW_PROFILE;
> >>>> 2. The ethtool command is holding rtnl lock;
> >>>> 3. Since the rtnl lock is already held, virtnet_rx_dim_work fails
> >>>> to acquire the lock and exits;
> >>>>
> >>>> Then, even if net_dim is invoked again, it cannot work because the
> >>>> state is not restored to DIM_START_MEASURE.
> >>>>
> >>>> Fixes: 6208799553a8 ("virtio-net: support rx netdim")
> >>>> Signed-off-by: Heng Qi <hengqi@xxxxxxxxxxxxxxxxx>
> >>>> ---
> >>>> drivers/net/virtio_net.c | 4 +++-
> >>>> 1 file changed, 3 insertions(+), 1 deletion(-)
> >>>>
> >>>> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> >>>> index c22d111..0ebe322 100644
> >>>> --- a/drivers/net/virtio_net.c
> >>>> +++ b/drivers/net/virtio_net.c
> >>>> @@ -3563,8 +3563,10 @@ static void virtnet_rx_dim_work(struct work_struct *work)
> >>>> struct dim_cq_moder update_moder;
> >>>> int i, qnum, err;
> >>>>
> >>>> - if (!rtnl_trylock())
> >>>> + if (!rtnl_trylock()) {
> >>>> + schedule_work(&dim->work);
> >>>> return;
> >>>> + }
> >>> Patch looks fine but I wonder if a delayed schedule is better.
> >> The work in net_dim() core layer uses non-delayed-work, and the two
> >> cannot be mixed.
> > Well, I think we need first to figure out if delayed work is better here.
>
> I tested a VM with 16 NICs, 128 queues per NIC (2kq total). With dim
> enabled on all queues,
> there are many opportunities for contention for rtnl lock, and this
> patch introduces no visible hotspots.
> The dim performance is also stable. So I think there doesn't seem to be
> a strong motivation right now.
That's fine, let's add them to the changelog.
Acked-by: Jason Wang <jasowang@xxxxxxxxxx>
Thanks
>
> Thanks,
> Heng
>
> >
> > Switching to use delayed work for dim seems not hard anyhow.
> >
> > Thanks
> >
> >> Thanks,
> >> Heng
> >>
> >>> Thanks
> >>>
> >>>> /* Each rxq's work is queued by "net_dim()->schedule_work()"
> >>>> * in response to NAPI traffic changes. Note that dim->profile_ix
> >>>> --
> >>>> 1.8.3.1
> >>>>
>
