2018-05-04 14:59 GMT+02:00 Daniel Borkmann <dan...@iogearbox.net>:
[...]
>> +
>> +int xsk_rcv(struct xdp_sock *xs, struct xdp_buff *xdp)
>> +{
>> + int err;
>> +
>> + err = __xsk_rcv(xs, xdp);
>> + if (likely(!err))
>> + xdp_return_buff(xdp);
>> + else
>> + xs->rx_dropped++;
>
> This is triggered from __bpf_tx_xdp_map() -> __xsk_map_redirect().
> Should this be percpu counter instead?
>
No, it shouldn't be percpu, but the drop count shouldn't be increased
here. It should be increased as a result of full queue event. Thanks
for pointing this out, I'll fix it.
Björn
[...]
