Date:   Tue, 6 Jun 2017 09:51:22 -0600
From:   Jeffrey Hugo <jhugo@...eaurora.org>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     Ingo Molnar <mingo@...hat.com>, linux-kernel@...r.kernel.org,
        Dietmar Eggemann <dietmar.eggemann@....com>,
        Austin Christ <austinwc@...eaurora.org>,
        Tyler Baicar <tbaicar@...eaurora.org>,
        Timur Tabi <timur@...eaurora.org>
Subject: Re: [PATCH V4 1/2] sched/fair: Fix load_balance() affinity redo path

On 6/6/2017 6:30 AM, Peter Zijlstra wrote:
> On Fri, Jun 02, 2017 at 04:27:11PM -0600, Jeffrey Hugo wrote:
>> If load_balance() fails to migrate any tasks because all tasks were
>> affined, load_balance() removes the source cpu from consideration and
>> attempts to redo and balance among the new subset of cpus.
>>
>> There is a bug in this code path where the algorithm considers all active
>> cpus in the system (minus the source that was just masked out).  This is
>> not valid for two reasons: some active cpus may not be in the current
>> scheduling domain and one of the active cpus is dst_cpu. These cpus should
>> not be considered, as we cannot pull load from them.
>>
>> Instead of failing out of load_balance(), we may end up redoing the search
>> with no valid cpus and incorrectly concluding the domain is balanced.
>> Additionally, if the group_imbalance flag was just set, it may also be
>> incorrectly unset, thus the flag will not be seen by other cpus in future
>> load_balance() runs as that algorithm intends.
>>
>> Fix the check by removing cpus not in the current domain and the dst_cpu
>> from considertation, thus limiting the evaluation to valid remaining cpus
>> from which load might be migrated.
>>
>> Co-authored-by: Austin Christ <austinwc@...eaurora.org>
>> Co-authored-by: Dietmar Eggemann <dietmar.eggemann@....com>
>> Signed-off-by: Jeffrey Hugo <jhugo@...eaurora.org>
>> Tested-by: Tyler Baicar <tbaicar@...eaurora.org>
>> ---
>>   kernel/sched/fair.c | 22 ++++++++++------------
>>   1 file changed, 10 insertions(+), 12 deletions(-)
>>
>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>> index d711093..84255ab 100644
>> --- a/kernel/sched/fair.c
>> +++ b/kernel/sched/fair.c
>> @@ -6737,10 +6737,10 @@ int can_migrate_task(struct task_struct *p, struct lb_env *env)
>>   		 * our sched_group. We may want to revisit it if we couldn't
>>   		 * meet load balance goals by pulling other tasks on src_cpu.
>>   		 *
>> -		 * Also avoid computing new_dst_cpu if we have already computed
>> -		 * one in current iteration.
>> +		 * Avoid computing new_dst_cpu for NEWLY_IDLE or if we have
>> +		 * already computed one in current iteration.
>>   		 */
>> -		if (!env->dst_grpmask || (env->flags & LBF_DST_PINNED))
>> +		if (env->idle == CPU_NEWLY_IDLE || (env->flags & LBF_DST_PINNED))
>>   			return 0;
>>   
>>   		/* Prevent to re-select dst_cpu via env's cpus */
>> @@ -8091,14 +8091,7 @@ static int load_balance(int this_cpu, struct rq *this_rq,
>>   		.tasks		= LIST_HEAD_INIT(env.tasks),
>>   	};
>>   
>> -	/*
>> -	 * For NEWLY_IDLE load_balancing, we don't need to consider
>> -	 * other cpus in our group
>> -	 */
>> -	if (idle == CPU_NEWLY_IDLE)
>> -		env.dst_grpmask = NULL;
>> -
>> -	cpumask_copy(cpus, cpu_active_mask);
>> +	cpumask_and(cpus, sched_domain_span(sd), cpu_active_mask);
>>   
>>   	schedstat_inc(sd->lb_count[idle]);
>>   
>> @@ -8220,7 +8213,12 @@ static int load_balance(int this_cpu, struct rq *this_rq,
>>   		/* All tasks on this runqueue were pinned by CPU affinity */
>>   		if (unlikely(env.flags & LBF_ALL_PINNED)) {
>>   			cpumask_clear_cpu(cpu_of(busiest), cpus);
>> -			if (!cpumask_empty(cpus)) {
>> +			/*
>> +			 * Go back to "redo" iff the load-balance cpumask
>> +			 * contains other potential busiest cpus for the
>> +			 * current sched domain.
>> +			 */
>> +			if (!cpumask_subset(cpus, env.dst_grpmask)) {
>>   				env.loop = 0;
>>   				env.loop_break = sched_nr_migrate_break;
>>   				goto redo;
> 
> So I was struggling with that subset condition. You want to ensure there
> are CPUs outside of dst_grpmask left, otherwise balancing at this SD
> level doesn't make sense anymore, right?

Correct.

> 
> I think you might want to spell that out a little in that comment.
> Currently the comment only explains what it does, which is something we
> can read from the code. Comments should explain _why_ we do things and
> its failing there.

Agreed.  We will attempt to improve the comment in the next version.

> 
> 
> So with that the problem is that active_load_balance_cpu_stop() calls
> into can_migrate_task() with ->idle = CPU_IDLE and !dst_grpmask, which
> then goes *bang*. Now active_load_balance_cpu_stop() doesn't need to
> re-evaluate anything, so ideally it would just skip this entirely,
> right?

Correct.  Your #3 option below would also work based on our 
understanding, and seems cleaner.  We will validate it as a solution, 
and incorporate into the next version.

Thanks

> 
> So why not do #3:
> 
> ---
>   kernel/sched/fair.c | 7 +++++++
>   1 file changed, 7 insertions(+)
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 47a0c552c77b..fd639d32fa4c 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -8523,6 +8523,13 @@ static int active_load_balance_cpu_stop(void *data)
>   			.src_cpu	= busiest_rq->cpu,
>   			.src_rq		= busiest_rq,
>   			.idle		= CPU_IDLE,
> +			/*
> +			 * can_migrate_task() doesn't need to compute new_dst_cpu
> +			 * for active balancing. Since we have CPU_IDLE, but no
> +			 * @dst_grpmask we need to make that test go away with lying
> +			 * about DST_PINNED.
> +			 */
> +			.flags		= LBF_DST_PINNED,
>   		};
>   
>   		schedstat_inc(sd->alb_count);
> 


-- 
Jeffrey Hugo
Qualcomm Datacenter Technologies as an affiliate of Qualcomm 
Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the
Code Aurora Forum, a Linux Foundation Collaborative Project.

Powered by blists - more mailing lists