Re: [PATCH] sched/fair: Add SMT4 group_smt_balance handling

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

On 8/22/23 12:49 AM, Tim Chen wrote:
> On Mon, 2023-08-07 at 15:06 +0530, Shrikanth Hegde wrote:
>>>
>>> From: Tim Chen <tim.c.chen@linux.intel.com>
>>> Date: Fri, 14 Jul 2023 16:09:30 -0700
>>> Subject: [PATCH] sched/fair: Add SMT4 group_smt_balance handling
>>>
>>> For SMT4, any group with more than 2 tasks will be marked as
>>> group_smt_balance. Retain the behaviour of group_has_spare by marking
>>> the busiest group as the group which has the least number of idle_cpus.
>>>
>>> Also, handle rounding effect of adding (ncores_local + ncores_busy)
>>> when the local is fully idle and busy group has more than 2 tasks.
>>> Local group should try to pull at least 1 task in this case.
>>>
>>> Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
>>> ---
>>>  kernel/sched/fair.c | 18 ++++++++++++++++--
>>>  1 file changed, 16 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>>> index a87988327f88..566686c5f2bd 100644
>>> --- a/kernel/sched/fair.c
>>> +++ b/kernel/sched/fair.c
>>> @@ -9563,7 +9563,7 @@ static inline long sibling_imbalance(struct lb_env *env,
>>>  	imbalance /= ncores_local + ncores_busiest;
>>>  
>>>  	/* Take advantage of resource in an empty sched group */
>>> -	if (imbalance == 0 && local->sum_nr_running == 0 &&
>>> +	if (imbalance <= 1 && local->sum_nr_running == 0 &&
>>>  	    busiest->sum_nr_running > 1)
>>>  		imbalance = 2;
>>>  
>>> @@ -9751,6 +9751,20 @@ static bool update_sd_pick_busiest(struct lb_env *env,
>>>  		break;
>>>  
>>>  	case group_smt_balance:
>>> +		/* no idle cpus on both groups handled by group_fully_busy below */
>>> +		if (sgs->idle_cpus != 0 || busiest->idle_cpus != 0) {
>>> +			if (sgs->idle_cpus > busiest->idle_cpus)
>>> +				return false;
>>> +			if (sgs->idle_cpus < busiest->idle_cpus)
>>> +				return true;
>>> +			if (sgs->sum_nr_running <= busiest->sum_nr_running)
>>> +				return false;
>>> +			else
>>> +				return true;
>>> +		}
>>> +		goto fully_busy;
>>> +		break;
>>> +
>>>  	case group_fully_busy:
>>>  		/*
>>>  		 * Select the fully busy group with highest avg_load. In
>>> @@ -9763,7 +9777,7 @@ static bool update_sd_pick_busiest(struct lb_env *env,
>>>  		 * select the 1st one, except if @sg is composed of SMT
>>>  		 * siblings.
>>>  		 */
>>> -
>>> +fully_busy:
>>>  		if (sgs->avg_load < busiest->avg_load)
>>>  			return false;
>>>  
>>
>> Hi Tim, Peter. 
>>
>> group_smt_balance(cluster scheduling), patches are in tip/sched/core. I dont 
>> see this above patch there yet. Currently as is, this can cause function difference 
>> in SMT4 systems( such as Power10). 
>>
>> Can we please have the above patch as well in tip/sched/core?
>>
>> Acked-by: Shrikanth Hegde <sshegde@linux.vnet.ibm.com>
> 
> Hi Peter,
> 
> Just back from my long vacation.  Wonder if you have any comments on the above patch
> for fixing the SMT4 case?
> 
> Tim

Hi Tim, Peter. 

are there any concerns with the above patch for fixing the SMT4 case. 
Currently the behavior is group_smt_balance is set for having even 2 tasks in 
SMT4, ideally it should be same as the group_has_spare. 

The above patch copies the same behavior to group_smt_balance. 
>