Re: [PATCH net-next] selftest: epoll_busy_poll: epoll busy poll tests

[Joe Damato]

On Fri, May 03, 2024 at 03:49:39PM -0700, Jakub Kicinski wrote:
> On Thu,  2 May 2024 21:20:11 +0000 Joe Damato wrote:
> > --- a/tools/testing/selftests/net/Makefile
> > +++ b/tools/testing/selftests/net/Makefile
> > @@ -84,6 +84,7 @@ TEST_GEN_FILES += sctp_hello
> >  TEST_GEN_FILES += csum
> >  TEST_GEN_FILES += ip_local_port_range
> >  TEST_GEN_FILES += bind_wildcard
> > +TEST_GEN_FILES += epoll_busy_poll
> 
> "GEN" is for files which are built for other tests to use.
> IOW unless there's also a wrapper script under TEST_PROGS
> (or the C code is itself under TEST_PROGS) this test won't
> be executed by most CIs.

Ah, I see. OK.

If I decided to go with the kselftest_harness as mentioned below, I'd need
to include a wrapper script to run the binary with the right cmd line
arg(s) and put that in TEST_PROGS?

> FWIW here's how we run the tests in our CI upstream CI:
> https://github.com/linux-netdev/nipa/wiki/How-to-run-netdev-selftests-CI-style

Thanks for the link, I'll give this a close read.

> >  TEST_PROGS += test_vxlan_mdb.sh
> >  TEST_PROGS += test_bridge_neigh_suppress.sh
> >  TEST_PROGS += test_vxlan_nolocalbypass.sh
> 
> > +static void do_simple_test(void)
> > +{
> > +	int fd;
> > +
> > +	fd = epoll_create1(0);
> > +	if (fd == -1)
> > +		error(1, errno, "epoll_create");
> > +
> > +	do_simple_test_invalid_fd();
> > +	do_simple_test_invalid_ioctl(fd);
> > +	do_simple_test_get_params(fd);
> > +	do_simple_test_set_invalid(fd);
> > +	do_simple_test_set_and_get_valid(fd);
> 
> You don't want to use the kselftest_harness for this?
> No strong preference here, but seems like you could
> pop the epoll_create1 into a FIXTURE() and then the
> test cases into TEST_F() and we'd get the KTAP output
> formatting, ability to run the tests selectively etc.
> for free.

I have no preference. I looked at some random .c file test in the directory
and it wasn't using the kselftest_harness stuff so I just went with that.

The advantages of kselftest_harness make sense, so I can give it a rewrite
to use kselftest_harness in v2.

> tools/testing/selftests/net/tap.c is probably a good example 
> to take a look at

Thanks, I'll look at that one. I had previously just kinda scanned
reuseaddr_conflict.c and rxtimestamp.c and some other ones. Seemed like a
bunch were just regular C programs so I went that route, but the advantages
you list make a lot of sense.