On Fri, Aug 8, 2025 at 8:29 AM Saket Kumar Bhaskar [off-list ref] wrote:

On Thu, Aug 07, 2025 at 03:21:42PM -0700, Alexei Starovoitov wrote:

quoted

On Mon, Aug 4, 2025 at 11:29 PM Saket Kumar Bhaskar [off-list ref] wrote:

quoted

@@ -60,9 +65,16 @@ static void test_arena_spin_lock_size(int size)
                return;
        }

-       skel = arena_spin_lock__open_and_load();
-       if (!ASSERT_OK_PTR(skel, "arena_spin_lock__open_and_load"))
+       skel = arena_spin_lock__open();
+       if (!ASSERT_OK_PTR(skel, "arena_spin_lock__open"))
                return;
+
+       skel->rodata->nr_cpus = get_nprocs();

...

quoted

--- a/tools/testing/selftests/bpf/progs/bpf_arena_spin_lock.h
+++ b/tools/testing/selftests/bpf/progs/bpf_arena_spin_lock.h

@@ -20,8 +20,6 @@
 #define __arena __attribute__((address_space(1)))
 #endif

-extern unsigned long CONFIG_NR_CPUS __kconfig;
-
 /*
  * Typically, we'd just rely on the definition in vmlinux.h for qspinlock, but
  * PowerPC overrides the definition to define lock->val as u32 instead of

@@ -494,7 +492,7 @@ static __always_inline int arena_spin_lock(arena_spinlock_t __arena *lock)
 {
        int val = 0;

-       if (CONFIG_NR_CPUS > 1024)
+       if (nr_cpus > 1024)
                return -EOPNOTSUPP;

We cannot do this. It will make arena_spin_lock much harder to use.
BPF CI doesn't run on powerpc anyway, but you can document that this
test is disable by creating selftests/bpf/DENYLIST.powerpc.

Hi Alexie,
Sorry, I did not get it. Can you please help me to understand why it
makes arena_spin_lock harder to use.

because requiring user space to do
skel->rodata->nr_cpus = get_nprocs()
is a headache.