small fix for fix_small_imbalance
cover small fix for fix_small_imbalance
Here is two patches which correct the scale usage in the fix_small_balance, and give out comment over when fix_small_imbalance would cause load change.
Lei Wen (2):
sched: reduce calculation effort in fix_small_imbalance
sched: scale the busy and this queue's per-task load before compare
kernel/sched/fair.c | 37 ++++++++++++++++++++++--------------- 1 file changed, 22 insertions(+), 15 deletions(-)
Signature · 1 line
1.7.10.4
PATCH 1/2 sched: reduce calculation effort in fix_small_imbalance
Actually all below item could be repalced by scaled_busy_load_per_task
(sds->busiest_load_per_task * SCHED_POWER_SCALE) /sds->busiest->sgp->power;
--- kernel/sched/fair.c | 19 ++++++++----------- 1 file changed, 8 insertions(+), 11 deletions(-)
kernel/sched/fair.c +8 -11
@@ -4727,20 +4727,17 @@ void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds) pwr_now /=SCHED_POWER_SCALE ;/* Amount of load we'd subtract */ - tmp = (sds->busiest_load_per_task *SCHED_POWER_SCALE ) / - sds->busiest->sgp->power; -if (sds->max_load > tmp) +if (sds->max_load > scaled_busy_load_per_task) { pwr_move += sds->busiest->sgp->power * -min (sds->busiest_load_per_task, sds->max_load - tmp); - -/* Amount of load we'd add */ -if (sds->max_load * sds->busiest->sgp->power < - sds->busiest_load_per_task *SCHED_POWER_SCALE ) - tmp = (sds->max_load * sds->busiest->sgp->power) / - sds->this->sgp->power; -else +min (sds->busiest_load_per_task, + sds->max_load - scaled_busy_load_per_task); tmp = (sds->busiest_load_per_task *SCHED_POWER_SCALE ) / sds->this->sgp->power; + }else + tmp = (sds->max_load * sds->busiest->sgp->power) / + sds->this->sgp->power; + +/* Amount of load we'd add */ pwr_move += sds->this->sgp->power *min (sds->this_load_per_task, sds->this_load + tmp); pwr_move /=SCHED_POWER_SCALE ;
Signature · 1 line
1.7.10.4
PATCH 2/2 sched: scale the busy and this queue's per-task load before compare
In reply to Lei WenSince for max_load and this_load, they are the value that already be scaled. It is not reasonble to get a minimum value between the scaled and non-scaled value, like below example.
min(sds->busiest_load_per_task, sds->max_load);
Also add comment over in what condition, there would be cpu power gain in move the load.
--- kernel/sched/fair.c | 28 +++++++++++++++++++--------- 1 file changed, 19 insertions(+), 9 deletions(-)
kernel/sched/fair.c +19 -9
@@ -4692,7 +4692,7 @@ void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds) {unsigned long tmp, pwr_now =0 , pwr_move =0 ;unsigned int imbn =2 ; -unsigned long scaled_busy_load_per_task; +unsigned long scaled_busy_load_per_task, scaled_this_load_per_task;if (sds->this_nr_running) { sds->this_load_per_task /= sds->this_nr_running;
@@ -4714,6 +4714,9 @@ void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)return ; } + scaled_this_load_per_task = sds->this_load_per_task + *SCHED_POWER_SCALE ; + scaled_this_load_per_task /= sds->this->sgp->power;/* * OK, we don't have enough imbalance to justify moving tasks, * however we may be able to increase total CPU power used by
@@ -4721,28 +4724,35 @@ void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)*/ pwr_now += sds->busiest->sgp->power * -min (sds->busiest_load_per_task, sds->max_load); +min (scaled_busy_load_per_task, sds->max_load); pwr_now += sds->this->sgp->power * -min (sds->this_load_per_task, sds->this_load); +min (scaled_this_load_per_task, sds->this_load); pwr_now /=SCHED_POWER_SCALE ;/* Amount of load we'd subtract */ if (sds->max_load > scaled_busy_load_per_task) { pwr_move += sds->busiest->sgp->power * -min (sds->busiest_load_per_task, +min (scaled_busy_load_per_task, sds->max_load - scaled_busy_load_per_task); - tmp = (sds->busiest_load_per_task *SCHED_POWER_SCALE ) / - sds->this->sgp->power; + tmp = scaled_busy_load_per_task; }else - tmp = (sds->max_load * sds->busiest->sgp->power) / - sds->this->sgp->power; + tmp = sds->max_load; +/* Scale to this queue from busiest queue */ + tmp = (tmp * sds->busiest->sgp->power) / + sds->this->sgp->power;/* Amount of load we'd add */ pwr_move += sds->this->sgp->power * -min (sds->this_load_per_task, sds->this_load + tmp); +min (scaled_this_load_per_task, sds->this_load + tmp); pwr_move /=SCHED_POWER_SCALE ;/* Move if we gain throughput */ +/* +* The only possibilty for below statement be true, is: +* sds->max_load is larger than scaled_busy_load_per_task, while, +* scaled_this_load_per_task is larger than sds->this_load plus by +* the scaled scaled_busy_load_per_task moved into this queue +*/ if (pwr_move > pwr_now) env->imbalance = sds->busiest_load_per_task; }
Signature · 1 line
1.7.10.4