Hi Stan,
Thank you very much for testing. I was wondering/worried that there
will be some churn in getting the test working on both our setups.
It's not unheard of I think because your ncdevmem changes had to go
through a couple of iterations to work for our slightly different
setups, but do bear with me. Thanks!
On Thu, Feb 20, 2025 at 12:01 PM Stanislav Fomichev
<stfomichev@xxxxxxxxx> wrote:
> > @@ -25,18 +25,36 @@ def check_rx(cfg) -> None:
> > require_devmem(cfg)
> >
> > port = rand_port()
> > - listen_cmd = f"./ncdevmem -l -f {cfg.ifname} -s {cfg.v6} -p {port}"
> > + listen_cmd = f"{cfg.bin_local} -l -f {cfg.ifname} -s {cfg.v6} -p {port}"
> >
> > with bkg(listen_cmd) as socat:
> > wait_port_listen(port)
> > - cmd(f"echo -e \"hello\\nworld\"| socat -u - TCP6:[{cfg.v6}]:{port}", host=cfg.remote, shell=True)
> > + cmd(f"echo -e \"hello\\nworld\"| socat -u - TCP6:{cfg.v6}:{port},bind={cfg.remote_v6}:{port}", host=cfg.remote, shell=True)
>
> IPv6 address need to be wrapped into [], so has to be at least:
> socat -u - TCP6:[{cfg.v6}]:{port},bind=[{cfg.remote_v6}]:{port}
>
Yeah, I will need to propagate the ncdevmem ipv4 support to devmem.py
in the future, but unnecessary for this series. Will do.
> But not sure why we care here about bind address here, let the kernel
> figure out the routing.
>
I will need to add this in the future to support my 5-tuple flow
steering setup in the future, but it is indeed unnecessary for this
series. Additionally the bind in the check_tx test is unnecessary,
removed there as well. Lets see if it works for you.
> Also, seems like "bkg(listen_cmd)" needs to be "bkg(listen_cmd,
> exit_wait=True)", otherwise sometimes I see racy empty result.
It passes for me with/without, but also fine, will do!
Thanks again for testing!
--
Thanks,
Mina
