Re: [PATCH v2 7/8] lib: add test for for_each_numa_{cpu,hop_mask}()

[Yury Norov]

Hi Valentin,

Thanks for review!

On Mon, Apr 24, 2023 at 06:09:52PM +0100, Valentin Schneider wrote:
> On 19/04/23 22:19, Yury Norov wrote:
> > +	for (node = 0; node < sched_domains_numa_levels; node++) {
> > +		unsigned int hop, c = 0;
> > +
> > +		rcu_read_lock();
> > +		for_each_numa_cpu(cpu, hop, node, cpu_online_mask)
> > +			expect_eq_uint(cpumask_local_spread(c++, node), cpu);
> > +		rcu_read_unlock();
> > +	}
> 
> I'm not fond of the export of sched_domains_numa_levels, especially
> considering it's just there for tests.
> 
> Furthermore, is there any value is testing parity with
> cpumask_local_spread()?

I wanted to emphasize that new NUMA-aware functions are coherent with
each other, just like find_nth_bit() is coherent with find_next_bit().

But all that coherence looks important only in non-NUMA case, because
client code may depend on fact that next CPU is never less than current.
This doesn't hold for NUMA iterators anyways...

> Rather, shouldn't we check that using this API does
> yield CPUs of increasing NUMA distance?
> 
> Something like
> 
>         for_each_node(node) {
>                 unsigned int prev_cpu, hop = 0;
> 
>                 cpu = cpumask_first(cpumask_of_node(node));
>                 prev_cpu = cpu;
> 
>                 rcu_read_lock();
> 
>                 /* Assert distance is monotonically increasing */
>                 for_each_numa_cpu(cpu, hop, node, cpu_online_mask) {
>                         expect_ge_uint(cpu_to_node(cpu), cpu_to_node(prev_cpu));
>                         prev_cpu = cpu;
>                 }
> 
>                 rcu_read_unlock();
>         }

Your version of the test looks more straightforward. I need to think
for more, but it looks like I can take it in v3.

Thanks,
Yury