Re: Re: [PATCH v2] bpf: Simplify code by using for_each_cpu_wrap()

[Punit Agrawal]

Hi Song,

Thanks for taking a look.

Song Liu <song@xxxxxxxxxx> writes:

> On Wed, Sep 7, 2022 at 8:58 AM Punit Agrawal
> <punit.agrawal@xxxxxxxxxxxxx> wrote:
>>
>> In the percpu freelist code, it is a common pattern to iterate over
>> the possible CPUs mask starting with the current CPU. The pattern is
>> implemented using a hand rolled while loop with the loop variable
>> increment being open-coded.
>>
>> Simplify the code by using for_each_cpu_wrap() helper to iterate over
>> the possible cpus starting with the current CPU. As a result, some of
>> the special-casing in the loop also gets simplified.
>>
>> No functional change intended.
>>
>> Signed-off-by: Punit Agrawal <punit.agrawal@xxxxxxxxxxxxx>
>> ---
>> v1 -> v2:
>> * Fixed the incorrect transformation changing semantics of __pcpu_freelist_push_nmi()
>>
>> Previous version -
>> v1: https://lore.kernel.org/all/20220817130807.68279-1-punit.agrawal@xxxxxxxxxxxxx/
>>
>>  kernel/bpf/percpu_freelist.c | 48 ++++++++++++------------------------
>>  1 file changed, 16 insertions(+), 32 deletions(-)
>>
>> diff --git a/kernel/bpf/percpu_freelist.c b/kernel/bpf/percpu_freelist.c
>> index 00b874c8e889..b6e7f5c5b9ab 100644
>> --- a/kernel/bpf/percpu_freelist.c
>> +++ b/kernel/bpf/percpu_freelist.c
>> @@ -58,23 +58,21 @@ static inline void ___pcpu_freelist_push_nmi(struct pcpu_freelist *s,
>>  {
>>         int cpu, orig_cpu;
>>
>> -       orig_cpu = cpu = raw_smp_processor_id();
>> +       orig_cpu = raw_smp_processor_id();
>>         while (1) {
>> -               struct pcpu_freelist_head *head;
>> +               for_each_cpu_wrap(cpu, cpu_possible_mask, orig_cpu) {
>> +                       struct pcpu_freelist_head *head;
>>
>> -               head = per_cpu_ptr(s->freelist, cpu);
>> -               if (raw_spin_trylock(&head->lock)) {
>> -                       pcpu_freelist_push_node(head, node);
>> -                       raw_spin_unlock(&head->lock);
>> -                       return;
>> +                       head = per_cpu_ptr(s->freelist, cpu);
>> +                       if (raw_spin_trylock(&head->lock)) {
>> +                               pcpu_freelist_push_node(head, node);
>> +                               raw_spin_unlock(&head->lock);
>> +                               return;
>> +                       }
>>                 }
>> -               cpu = cpumask_next(cpu, cpu_possible_mask);
>> -               if (cpu >= nr_cpu_ids)
>> -                       cpu = 0;
>
> I personally don't like nested loops here. Maybe we can keep
> the original while loop and use cpumask_next_wrap()?

Out of curiosity, is there a reason to avoid nesting here? The nested
loop avoids the "cpu == orig_cpu" unnecessary check every iteration.

As suggested, it's possible to use cpumask_next_wrap() like below -

diff --git a/kernel/bpf/percpu_freelist.c b/kernel/bpf/percpu_freelist.c
index 00b874c8e889..19e8eab70c40 100644
--- a/kernel/bpf/percpu_freelist.c
+++ b/kernel/bpf/percpu_freelist.c
@@ -68,9 +68,7 @@ static inline void ___pcpu_freelist_push_nmi(struct pcpu_freelist *s,
                        raw_spin_unlock(&head->lock);
                        return;
                }
-               cpu = cpumask_next(cpu, cpu_possible_mask);
-               if (cpu >= nr_cpu_ids)
-                       cpu = 0;
+               cpu = cpumask_next_wrap(cpu, cpu_possible_mask, orig_cpu, false);

                /* cannot lock any per cpu lock, try extralist */
                if (cpu == orig_cpu &&


I can send an updated patch if this is preferred.

Thanks,
Punit