[v3,07/10] sched/fair: Do not even the number of busy CPUs via asym_packing

Message ID 20230207045838.11243-8-ricardo.neri-calderon@linux.intel.com
State New
Headers
Series sched/fair: Avoid unnecessary migrations within SMT domains |

Commit Message

Ricardo Neri Feb. 7, 2023, 4:58 a.m. UTC
  Now that find_busiest_group() triggers load balancing between a fully_
busy SMT2 core and an idle non-SMT core, it is no longer needed to force
balancing via asym_packing. Use asym_packing only as intended: when there
is high-priority CPU that is idle.

After this change, the same logic apply to SMT and non-SMT local groups.
Simplify asym_smt_can_pull_tasks() accordingly.

Cc: Ben Segall <bsegall@google.com>
Cc: Daniel Bristot de Oliveira <bristot@redhat.com>
Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc: Len Brown <len.brown@intel.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tim C. Chen <tim.c.chen@intel.com>
Cc: Valentin Schneider <vschneid@redhat.com>
Cc: x86@kernel.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
---
Changes since v2:
 * Introduced this patch.

Changes since v1:
 * N/A
---
 kernel/sched/fair.c | 37 +++++--------------------------------
 1 file changed, 5 insertions(+), 32 deletions(-)
  

Comments

Dietmar Eggemann Feb. 13, 2023, 12:44 p.m. UTC | #1
On 07/02/2023 05:58, Ricardo Neri wrote:

[...]

> @@ -9269,33 +9264,11 @@ static bool asym_smt_can_pull_tasks(int dst_cpu, struct sd_lb_stats *sds,
>  				    struct sched_group *sg)
>  {
>  #ifdef CONFIG_SCHED_SMT
> -	bool local_is_smt;
>  	int sg_busy_cpus;
>  
> -	local_is_smt = sds->local->flags & SD_SHARE_CPUCAPACITY;
>  	sg_busy_cpus = sgs->group_weight - sgs->idle_cpus;
>  
> -	if (!local_is_smt) {
> -		/*
> -		 * If we are here, @dst_cpu is idle and does not have SMT
> -		 * siblings. Pull tasks if candidate group has two or more
> -		 * busy CPUs.
> -		 */
> -		if (sg_busy_cpus >= 2) /* implies sg_is_smt */
> -			return true;
> -
> -		/*
> -		 * @dst_cpu does not have SMT siblings. @sg may have SMT
> -		 * siblings and only one is busy. In such case, @dst_cpu
> -		 * can help if it has higher priority and is idle (i.e.,
> -		 * it has no running tasks).
> -		 */
> -		return sched_asym_prefer(dst_cpu, sg->asym_prefer_cpu);
> -	}
> -
>  	/*
> -	 * @dst_cpu has SMT siblings and are also idle.
> -	 *
>  	 * If the difference in the number of busy CPUs is two or more, let
>  	 * find_busiest_group() take care of it. We only care if @sg has
>  	 * exactly one busy CPU. This covers SMT and non-SMT sched groups.

Can't this be made lighter by removing asym_smt_can_pull_tasks() and
putting the logic to exclude the call to sched_asym_prefer() into
sched_asym() directly?
Not sure if we need the CONFIG_SCHED_SMT since it's all guarded by
`flags & SD_SHARE_CPUCAPACITY` already, which is only set under.
CONFIG_SCHED_SMT.

static inline bool
sched_asym(struct lb_env *env, struct sd_lb_stats *sds,
           struct sg_lb_stats *sgs, struct sched_group *group)
{
        bool local_is_smt = sds->local->flags & SD_SHARE_CPUCAPACITY;

        if (local_is_smt && !is_core_idle(env->dst_cpu))
                return false;

        if ((local_is_smt || group->flags & SD_SHARE_CPUCAPACITY)) {
                int sg_busy_cpus = sgs->group_weight - sgs->idle_cpus;

                if (sg_busy_cpus != 1)
                        return false;
        }

        return sched_asym_prefer(env->dst_cpu, group->asym_prefer_cpu);
}

[...]
  
Ricardo Neri Feb. 13, 2023, 7:47 p.m. UTC | #2
On Mon, Feb 13, 2023 at 01:44:20PM +0100, Dietmar Eggemann wrote:
> On 07/02/2023 05:58, Ricardo Neri wrote:
> 
> [...]
> 
> > @@ -9269,33 +9264,11 @@ static bool asym_smt_can_pull_tasks(int dst_cpu, struct sd_lb_stats *sds,
> >  				    struct sched_group *sg)
> >  {
> >  #ifdef CONFIG_SCHED_SMT
> > -	bool local_is_smt;
> >  	int sg_busy_cpus;
> >  
> > -	local_is_smt = sds->local->flags & SD_SHARE_CPUCAPACITY;
> >  	sg_busy_cpus = sgs->group_weight - sgs->idle_cpus;
> >  
> > -	if (!local_is_smt) {
> > -		/*
> > -		 * If we are here, @dst_cpu is idle and does not have SMT
> > -		 * siblings. Pull tasks if candidate group has two or more
> > -		 * busy CPUs.
> > -		 */
> > -		if (sg_busy_cpus >= 2) /* implies sg_is_smt */
> > -			return true;
> > -
> > -		/*
> > -		 * @dst_cpu does not have SMT siblings. @sg may have SMT
> > -		 * siblings and only one is busy. In such case, @dst_cpu
> > -		 * can help if it has higher priority and is idle (i.e.,
> > -		 * it has no running tasks).
> > -		 */
> > -		return sched_asym_prefer(dst_cpu, sg->asym_prefer_cpu);
> > -	}
> > -
> >  	/*
> > -	 * @dst_cpu has SMT siblings and are also idle.
> > -	 *
> >  	 * If the difference in the number of busy CPUs is two or more, let
> >  	 * find_busiest_group() take care of it. We only care if @sg has
> >  	 * exactly one busy CPU. This covers SMT and non-SMT sched groups.
> 
> Can't this be made lighter by removing asym_smt_can_pull_tasks() and
> putting the logic to exclude the call to sched_asym_prefer() into
> sched_asym() directly?

Yes, you are right. asym_smt_can_pull_tasks() was simplified significantly.
I'll take your suggestion.

> Not sure if we need the CONFIG_SCHED_SMT since it's all guarded by
> `flags & SD_SHARE_CPUCAPACITY` already, which is only set under.
> CONFIG_SCHED_SMT.

Yes, asym_smt_can_pull_tasks() now cares for a very specific case, which
only happens with CONFIG_SCHED_SMT. I'll remove the !CONFIG_SCHED_SMT part.
> 
> static inline bool
> sched_asym(struct lb_env *env, struct sd_lb_stats *sds,
>            struct sg_lb_stats *sgs, struct sched_group *group)
> {
>         bool local_is_smt = sds->local->flags & SD_SHARE_CPUCAPACITY;
> 
>         if (local_is_smt && !is_core_idle(env->dst_cpu))
>                 return false;
> 
>         if ((local_is_smt || group->flags & SD_SHARE_CPUCAPACITY)) {
>                 int sg_busy_cpus = sgs->group_weight - sgs->idle_cpus;
> 
>                 if (sg_busy_cpus != 1)
>                         return false;
>         }
> 
>         return sched_asym_prefer(env->dst_cpu, group->asym_prefer_cpu);

I'll take your suggestion. Thanks!
  

Patch

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index a37ad59f20ea..0ada2d18b934 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -9247,20 +9247,15 @@  group_type group_classify(unsigned int imbalance_pct,
  * @sgs:	Load-balancing statistics of the candidate busiest group
  * @sg:		The candidate busiest group
  *
- * Check the state of the SMT siblings of both @sds::local and @sg and decide
- * if @dst_cpu can pull tasks.
+ * Check the state of the SMT siblings of @sg and decide if @dst_cpu can pull
+ * tasks.
  *
  * This function must be called only if all the SMT siblings of @dst_cpu are
  * idle, if any.
  *
- * If @dst_cpu does not have SMT siblings, it can pull tasks if two or more of
- * the SMT siblings of @sg are busy. If only one CPU in @sg is busy, pull tasks
- * only if @dst_cpu has higher priority.
- *
- * If @dst_cpu has SMT siblings, decide based on the priority of @sg. Do it only
- * if @sg has exactly one busy CPU (i.e., one more than @sds::local). Bigger
- * imbalances in the number of busy CPUs will be dealt with in
- * find_busiest_group().
+ * @dst_cpu can pull tasks if @sg has exactly one busy CPU (i.e., one more than
+ * @sds::local) and has lower group priority than @sds::local. Bigger imbalances
+ * in the number of busy CPUs will be dealt with in find_busiest_group().
  *
  * Return: true if @dst_cpu can pull tasks, false otherwise.
  */
@@ -9269,33 +9264,11 @@  static bool asym_smt_can_pull_tasks(int dst_cpu, struct sd_lb_stats *sds,
 				    struct sched_group *sg)
 {
 #ifdef CONFIG_SCHED_SMT
-	bool local_is_smt;
 	int sg_busy_cpus;
 
-	local_is_smt = sds->local->flags & SD_SHARE_CPUCAPACITY;
 	sg_busy_cpus = sgs->group_weight - sgs->idle_cpus;
 
-	if (!local_is_smt) {
-		/*
-		 * If we are here, @dst_cpu is idle and does not have SMT
-		 * siblings. Pull tasks if candidate group has two or more
-		 * busy CPUs.
-		 */
-		if (sg_busy_cpus >= 2) /* implies sg_is_smt */
-			return true;
-
-		/*
-		 * @dst_cpu does not have SMT siblings. @sg may have SMT
-		 * siblings and only one is busy. In such case, @dst_cpu
-		 * can help if it has higher priority and is idle (i.e.,
-		 * it has no running tasks).
-		 */
-		return sched_asym_prefer(dst_cpu, sg->asym_prefer_cpu);
-	}
-
 	/*
-	 * @dst_cpu has SMT siblings and are also idle.
-	 *
 	 * If the difference in the number of busy CPUs is two or more, let
 	 * find_busiest_group() take care of it. We only care if @sg has
 	 * exactly one busy CPU. This covers SMT and non-SMT sched groups.