Re: [PATCH net-next v2 0/7] selftests/net: Mixed select()+polling mode for TCP-AO tests
From: Jakub Kicinski
Date: Wed Mar 26 2025 - 16:00:19 EST
On Wed, 26 Mar 2025 19:48:16 +0000 Dmitry Safonov wrote:
> On Tue, 25 Mar 2025 at 13:15, Jakub Kicinski <kuba@xxxxxxxxxx> wrote:
> >
> > On Wed, 19 Mar 2025 03:13:33 +0000 Dmitry Safonov via B4 Relay wrote:
> > > Should fix flaky tcp-ao/connect-deny-ipv6 test.
> > > Begging pardon for the delay since the report and for sending it this
> > > late in the release cycle.
> >
> > Better late than never, thanks a lot! :)
>
> Thank you, Jakub!
>
> I also noticed that recently, self-connect-ipv6 became slightly flaky:
> https://netdev.bots.linux.dev/flakes.html?br-cnt=75&tn-needle=tcp-ao
>
> Seems unrelated to select()+poll selftests changes, but rather to
> timings in the kernel:
> # # 1249[lib/proc.c:213] Snmp6 Ip6OutNoRoutes: 0 => 1
>
> It seems that the test relied on kernel adding a link-local route with
> a loopback interface, but probably adding the interface got faster.
> Seems like a trivial two-line fix by manually adding the link-local
> route in userspace. Though, can't reproduce that flake locally on
> thousands of runs.
>
> Should I send the potential fix now for -net or wait until the merge
> window closes and send for -net-next?
I reckon you can send it now, maybe other maintainers will disagree
but to me test stability fixes should be okay during the MW.
You can tag it as net-next for the benefit of the build bot,
tho we'll end up merging it to net once/if Linus pulls.