sched: simplify sched_init_numa()

[PATCH] sched: simplify sched_init_numa()

Posted by Yury Norov 2 months, 2 weeks ago

From: Yury Norov (NVIDIA) <yury.norov@gmail.com>

The function opencodes for_each_set_bit() macro.

Signed-off-by: Yury Norov (NVIDIA) <yury.norov@gmail.com>
---
 kernel/sched/topology.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index b958fe48e020..7dc3c79aa480 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -1932,10 +1932,9 @@ void sched_init_numa(int offline_node)
 		return;
 	}
 
-	for (i = 0, j = 0; i < nr_levels; i++, j++) {
-		j = find_next_bit(distance_map, NR_DISTANCE_VALUES, j);
-		distances[i] = j;
-	}
+	for_each_set_bit(j, distance_map, NR_DISTANCE_VALUES)
+		distances[i++] = j;
+
 	rcu_assign_pointer(sched_domains_numa_distance, distances);
 
 	bitmap_free(distance_map);
-- 
2.43.0

Re: [PATCH] sched: simplify sched_init_numa()

Posted by Srikar Dronamraju 2 months, 2 weeks ago

* Yury Norov <yury.norov@gmail.com> [2025-07-19 17:07:51]:

> From: Yury Norov (NVIDIA) <yury.norov@gmail.com>
> 
> The function opencodes for_each_set_bit() macro.
> 
> Signed-off-by: Yury Norov (NVIDIA) <yury.norov@gmail.com>
> ---
>  kernel/sched/topology.c | 7 +++----
>  1 file changed, 3 insertions(+), 4 deletions(-)
> 
> diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
> index b958fe48e020..7dc3c79aa480 100644
> --- a/kernel/sched/topology.c
> +++ b/kernel/sched/topology.c
> @@ -1932,10 +1932,9 @@ void sched_init_numa(int offline_node)
>  		return;
>  	}
>  
> -	for (i = 0, j = 0; i < nr_levels; i++, j++) {
> -		j = find_next_bit(distance_map, NR_DISTANCE_VALUES, j);
> -		distances[i] = j;
> -	}
> +	for_each_set_bit(j, distance_map, NR_DISTANCE_VALUES)
> +		distances[i++] = j;
> +

Dont we need to reset the value of i;

Also now we may be iterating for NR_DISTANCE_VALUES instead of nr_levels.
It should be okay, since NR_DISTANCE_VALUES is just 8.

>  	rcu_assign_pointer(sched_domains_numa_distance, distances);
>  
>  	bitmap_free(distance_map);
> -- 
> 2.43.0
> 

-- 
Thanks and Regards
Srikar Dronamraju

Re: [PATCH] sched: simplify sched_init_numa()

Posted by Valentin Schneider 1 month, 2 weeks ago

On 21/07/25 14:02, Srikar Dronamraju wrote:
> * Yury Norov <yury.norov@gmail.com> [2025-07-19 17:07:51]:
>
>> From: Yury Norov (NVIDIA) <yury.norov@gmail.com>
>>
>> The function opencodes for_each_set_bit() macro.
>>
>> Signed-off-by: Yury Norov (NVIDIA) <yury.norov@gmail.com>
>> ---
>>  kernel/sched/topology.c | 7 +++----
>>  1 file changed, 3 insertions(+), 4 deletions(-)
>>
>> diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
>> index b958fe48e020..7dc3c79aa480 100644
>> --- a/kernel/sched/topology.c
>> +++ b/kernel/sched/topology.c
>> @@ -1932,10 +1932,9 @@ void sched_init_numa(int offline_node)
>>              return;
>>      }
>>
>> -	for (i = 0, j = 0; i < nr_levels; i++, j++) {
>> -		j = find_next_bit(distance_map, NR_DISTANCE_VALUES, j);
>> -		distances[i] = j;
>> -	}
>> +	for_each_set_bit(j, distance_map, NR_DISTANCE_VALUES)
>> +		distances[i++] = j;
>> +
>
> Dont we need to reset the value of i;
>

That.

> Also now we may be iterating for NR_DISTANCE_VALUES instead of nr_levels.
> It should be okay, since NR_DISTANCE_VALUES is just 8.
>

DISTANCE_BITS is 8, NR_DISTANCE_VALUES is 1 << 8 aka 256.

But here the use of for_each_set_bit() means we'll get just one extra
iteration compared to using `nr_levels`. That said, since we *have* to
compute `nr_levels` to allocate sched_domains_numa_distance, IMO we're not
gaining much by using for_each_set_bit() here.

>>      rcu_assign_pointer(sched_domains_numa_distance, distances);
>>
>>      bitmap_free(distance_map);
>> --
>> 2.43.0
>>
>
> --
> Thanks and Regards
> Srikar Dronamraju

Re: [PATCH] sched: simplify sched_init_numa()

Posted by Yury Norov 1 month, 2 weeks ago

On Tue, Aug 19, 2025 at 10:52:19AM +0200, Valentin Schneider wrote:
> On 21/07/25 14:02, Srikar Dronamraju wrote:
> > * Yury Norov <yury.norov@gmail.com> [2025-07-19 17:07:51]:
> >
> >> From: Yury Norov (NVIDIA) <yury.norov@gmail.com>
> >>
> >> The function opencodes for_each_set_bit() macro.
> >>
> >> Signed-off-by: Yury Norov (NVIDIA) <yury.norov@gmail.com>
> >> ---
> >>  kernel/sched/topology.c | 7 +++----
> >>  1 file changed, 3 insertions(+), 4 deletions(-)
> >>
> >> diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
> >> index b958fe48e020..7dc3c79aa480 100644
> >> --- a/kernel/sched/topology.c
> >> +++ b/kernel/sched/topology.c
> >> @@ -1932,10 +1932,9 @@ void sched_init_numa(int offline_node)
> >>              return;
> >>      }
> >>
> >> -	for (i = 0, j = 0; i < nr_levels; i++, j++) {
> >> -		j = find_next_bit(distance_map, NR_DISTANCE_VALUES, j);
> >> -		distances[i] = j;
> >> -	}
> >> +	for_each_set_bit(j, distance_map, NR_DISTANCE_VALUES)
> >> +		distances[i++] = j;
> >> +
> >
> > Dont we need to reset the value of i;
> >
> 
> That.

Yes we need.
 
> > Also now we may be iterating for NR_DISTANCE_VALUES instead of nr_levels.
> > It should be okay, since NR_DISTANCE_VALUES is just 8.
> >
> 
> DISTANCE_BITS is 8, NR_DISTANCE_VALUES is 1 << 8 aka 256.

I mistaken this DISTANCE_BITS vs NR_DISTANCE_VALUES. The first one is
8, so we can benefit from small_const_nbits() optimization. But we
actually iterate over NR_DISTANCE_VALUES - and no benefit for using
it over nr_levels.

> But here the use of for_each_set_bit() means we'll get just one extra
> iteration compared to using `nr_levels`. That said, since we *have* to
> compute `nr_levels` to allocate sched_domains_numa_distance, IMO we're not
> gaining much by using for_each_set_bit() here.

I'm not sure I understand that. If we switch the loop terminator to
nr_levels, the for_each() will be the exact functional equivalent.

I'll get back to this function shortly and send v2.

Thanks for the discussion.