Re: [PATCH bpf-next v2 5/5] selftests/bpf: Fix arena_spin_lock selftest failure

[Saket Kumar Bhaskar <skb99@linux.ibm.com>]

On Thu, Sep 04, 2025 at 01:55:40PM +0530, Saket Kumar Bhaskar wrote:
> On Thu, Sep 04, 2025 at 01:39:31PM +0530, Hari Bathini wrote:
> > 
> > 
> > On 29/08/25 10:21 pm, Saket Kumar Bhaskar wrote:
> > > For systems having CONFIG_NR_CPUS set to > 1024 in kernel config
> > > the selftest fails as arena_spin_lock_irqsave() returns EOPNOTSUPP.
> > > 
> > > The selftest is skipped incase bpf program returns EOPNOTSUPP,
> > > with a descriptive message logged.
> > > 
> > > Signed-off-by: Saket Kumar Bhaskar <skb99@linux.ibm.com>
> > > ---
> > >   .../selftests/bpf/prog_tests/arena_spin_lock.c      | 13 +++++++++++++
> > >   tools/testing/selftests/bpf/progs/arena_spin_lock.c |  5 ++++-
> > >   2 files changed, 17 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/tools/testing/selftests/bpf/prog_tests/arena_spin_lock.c b/tools/testing/selftests/bpf/prog_tests/arena_spin_lock.c
> > > index 0223fce4db2b..1ec1ca987893 100644
> > > --- a/tools/testing/selftests/bpf/prog_tests/arena_spin_lock.c
> > > +++ b/tools/testing/selftests/bpf/prog_tests/arena_spin_lock.c
> > > @@ -40,8 +40,13 @@ static void *spin_lock_thread(void *arg)
> > >   	err = bpf_prog_test_run_opts(prog_fd, &topts);
> > >   	ASSERT_OK(err, "test_run err");
> > > +
> > > +	if (topts.retval == -EOPNOTSUPP)
> > > +		goto end;
> > > +
> > >   	ASSERT_EQ((int)topts.retval, 0, "test_run retval");
> > > +end:
> > >   	pthread_exit(arg);
> > >   }
> > > @@ -63,6 +68,7 @@ static void test_arena_spin_lock_size(int size)
> > >   	skel = arena_spin_lock__open_and_load();
> > >   	if (!ASSERT_OK_PTR(skel, "arena_spin_lock__open_and_load"))
> > >   		return;
> > > +
> > >   	if (skel->data->test_skip == 2) {
> > >   		test__skip();
> > >   		goto end;
> > > @@ -86,6 +92,13 @@ static void test_arena_spin_lock_size(int size)
> > >   			goto end_barrier;
> > >   	}
> > > +	if (skel->data->test_skip == 2) {
> > > +		printf("%s:SKIP: %d CPUs exceed the maximum supported by arena spinlock\n",
> > > +		       __func__, get_nprocs());
> > > +		test__skip();
> > > +		goto end_barrier;
> > > +	}
> > > +
> > >   	ASSERT_EQ(skel->bss->counter, repeat * nthreads, "check counter value");
> > >   end_barrier:
> > > diff --git a/tools/testing/selftests/bpf/progs/arena_spin_lock.c b/tools/testing/selftests/bpf/progs/arena_spin_lock.c
> > > index c4500c37f85e..a475b974438e 100644
> > > --- a/tools/testing/selftests/bpf/progs/arena_spin_lock.c
> > > +++ b/tools/testing/selftests/bpf/progs/arena_spin_lock.c
> > > @@ -37,8 +37,11 @@ int prog(void *ctx)
> > >   #if defined(ENABLE_ATOMICS_TESTS) && defined(__BPF_FEATURE_ADDR_SPACE_CAST)
> > >   	unsigned long flags;
> > > -	if ((ret = arena_spin_lock_irqsave(&lock, flags)))
> > > +	if ((ret = arena_spin_lock_irqsave(&lock, flags))) {
> > > +		if (ret == -EOPNOTSUPP)
> > > +			test_skip = 2;
> > >   		return ret;
> > 
> > test_skip being set to `1` when the test runs seems counter intuitive.
> > How about setting test_skip to `0` when run conditions are met
> > and test_skip=1 if run conditions are not met and
> > test_skip=2 when operation is not supported?
> > 
> > - Hari
> That seems reasonable to me, but right now -EOPNOTSUPP is also
> returned when run condition is not met i.e.:
> 
>   if (CONFIG_NR_CPUS > 1024)
>                 return -EOPNOTSUPP;
> 
> So do we really need test_skip = 2 ?
> 
> Thanks,
> Saket
Also, when test_skip is initialized to 0 it is moved to bss segment 
from data segment:

        struct arena_spin_lock__arena {
                struct arena_qnode qnodes[1024][4];
                struct __qspinlock lock;
        } *arena;
        struct arena_spin_lock__bss {
                int test_skip;
                int counter;
                int limit;
                int cs_count;
        } *bss;

I dont have enough background here, as to if there is any specific 
reason to keep it in data segment:

        if (skel->data->test_skip == 2) {
                test__skip();
                goto end;
        }

Thanks,
Saket