On 5/17/2020 11:42 AM, Cong Wang wrote:
On Sat, May 16, 2020 at 9:53 AM Vladimir Oltean <olte...@gmail.com> wrote:
Thanks a lot for presenting the options. In general, xmit in DSA is
relatively simple and most of the time stateless. My stacked DSA setup
appears to work just fine with NETIF_F_LLTX, including the updating of
percpu counters. I'm not really sure if there's something in
particular to test?
Anyway, will you send a patch with NETIF_F_LLTX or should I do it? I
can do further testing if necessary.
If DSA is software based, there is a large chance it can be just
using NETIF_F_LLTX, like you said.
Please do send a patch for this NETIF_F_LLTX. Note my patch
simply reverts to the old code, this issue probably exists before it.
It transmits frames by calling dev_queue_xmit() after having assigned
skb->dev to be pointing to the master network device, very much like
what VLAN devices do for instance, so I believe setting NETIF_F_LLTX is
fine here.
--
Florian
