Re: [PATCH net-next v1 4/9] net: devmem: ksft: remove ksft_disruptive

[Jakub Kicinski <kuba@kernel.org>]

On Mon, 19 May 2025 13:18:24 -0700 Stanislav Fomichev wrote:
> On 05/19, Mina Almasry wrote:
> > On Mon, May 19, 2025 at 8:25 AM Stanislav Fomichev <stfomichev@gmail.com> wrote:  
> > >
> > > On 05/19, Mina Almasry wrote:  
> > > > As far as I can tell the ksft_disruptive here is unnecessary. These
> > > > tests are largerly independent, and when one test fails, it's nice to
> > > > know the results from all the other test cases.  
> > >
> > > We currently don't do anything special for disruptive tests. I'm assuming
> > > anything that changes nic configuration is disruptive and was thinking of
> > > an option to run all disruptive tests at the end of the run. But so far we
> > > haven't had any problem with mixing disruptive and non-disruptive tests,
> > > so it's all moot. I'd prefer to keep everything as is for now (or remove
> > > this whole disruptive category).  
> > 
> > I've noticed that if all the tests are marked disruptive, and one test
> > fails, the others don't run at all, which seems unnecessary. I'd like
> > to see if the rx test passed if the tx one failed and vice versa for
> > example. Removing the disruptive tag seems to resolve that.  
> 
> I don't think that's the expected behavior. Disruptive should not
> have any effect on other tests if any one fails. Any idea why it happens?

Right, this sounds odd and needs investigating.

FWIW, in my mind disruptive tests were supposed to be the tests which
can cut a single NIC machine off the network. IOW the use case is that
the NIC under test is the same NIC over which we're SSH'ing into the
machine to run the test. So we shouldn't do things like take the link
down or flush the IP addresses because it may kill our SSH connection.

Obviously every test may be somewhat disruptive, but most shouldn't
break ssh?