Re: [PATCH v3 3/4] sched/fair: Add sched_smt_active check for fastpaths

[Shrikanth Hegde <sshegde@linux.ibm.com>]

Hi Valentin. Thanks for going through the patches.

On 5/15/26 7:05 PM, Valentin Schneider wrote:
> On 13/05/26 19:09, Shrikanth Hegde wrote:
>> For fastpaths such as wakeup and load balance even minimal code additions
>> can add up. is_core_idle is accessed during load balance.
>>
>> Other callsites of is_core_idle make sched_smt_active() check first.
>> Make the same check in should_we_balance.
>>
>> Rest of access to cpu_smt_mask isn't in fastpath.
>>
>> Note: Remove the stale comment above is_core_idle. Enqueue methods
>> of fair aren't close to it anymore.
>>
>> Suggested-by: K Prateek Nayak <kprateek.nayak@amd.com>
>> Signed-off-by: Shrikanth Hegde <sshegde@linux.ibm.com>
>> ---
>>   kernel/sched/fair.c | 8 +++-----
>>   1 file changed, 3 insertions(+), 5 deletions(-)
>>
>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>> index 353e31ecaadc..964014a74cf9 100644
>> --- a/kernel/sched/fair.c
>> +++ b/kernel/sched/fair.c
>> @@ -1578,10 +1578,7 @@ update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
>>        se->exec_start = rq_clock_task(rq_of(cfs_rq));
>>   }
>>
>> -/**************************************************
>> - * Scheduling class queueing methods:
>> - */
>> -
>> +/* Check sched_smt_active before calling this to avoid overheads in fastpaths */
>>   static inline bool is_core_idle(int cpu)
>>   {
>>        int sibling;
>> @@ -11995,7 +11992,8 @@ static int should_we_balance(struct lb_env *env)
>>                 * balancing cores, but remember the first idle SMT CPU for
>>                 * later consideration.  Find CPU on an idle core first.
>>                 */
>> -		if (!(env->sd->flags & SD_SHARE_CPUCAPACITY) && !is_core_idle(cpu)) {
>> +		if (!(env->sd->flags & SD_SHARE_CPUCAPACITY) &&
>> +		    sched_smt_active() && !is_core_idle(cpu)) {
> 
> Nit: Since that whole if statement is about SMT, you could have the static
> key be the first thing checked in the condition so it can short circuit
> earlier.

Ok. We can do that too.

Only reason i had kept it that way was to avoid
calling it when load balancing at SMT domain assuming sched_smt_active() is
a more common in the field.

I can send out v4 out quickly with below change.

                if (sched_smt_active() &&
                    !(env->sd->flags & SD_SHARE_CPUCAPACITY) &&
                    !is_core_idle(cpu)) {

That's exactly what prateek had suggested.