@@ -2982,9 +2982,13 @@ static u64 cpu_avg_load_per_task(int cpu)
{
struct rq *rq = cpu_rq(cpu);
unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
+ u64 cfs_avg_load_per_task;
- if (nr_running)
- return rq->cfs.runnable_load_avg / nr_running;
+ if (nr_running) {
+ cfs_avg_load_per_task = rq->cfs.runnable_load_avg;
+ do_div(cfs_avg_load_per_task, nr_running);
+ return cfs_avg_load_per_task;
+ }
return 0;
}
@@ -3249,7 +3253,8 @@ find_idlest_group(struct sched_domain *sd, struct task_struct *p,
}
/* Adjust by relative CPU power of the group */
- avg_load = (avg_load * SCHED_POWER_SCALE) / group->sgp->power;
+ avg_load = (avg_load * SCHED_POWER_SCALE);
+ do_div(avg_load, group->sgp->power);
if (local_group) {
this_load = avg_load;
@@ -4756,7 +4761,8 @@ static inline void update_sg_lb_stats(struct lb_env *env,
}
/* Adjust by relative CPU power of the group */
- sgs->avg_load = (sgs->group_load*SCHED_POWER_SCALE) / group->sgp->power;
+ sgs->avg_load = (sgs->group_load*SCHED_POWER_SCALE);
+ do_div(sgs->avg_load, group->sgp->power);
/*
* Consider the group unbalanced when the imbalance is larger
@@ -4767,8 +4773,10 @@ static inline void update_sg_lb_stats(struct lb_env *env,
* normalized nr_running number somewhere that negates
* the hierarchy?
*/
- if (sgs->sum_nr_running)
- avg_load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
+ if (sgs->sum_nr_running) {
+ avg_load_per_task = sgs->sum_weighted_load;
+ do_div(avg_load_per_task, sgs->sum_nr_running);
+ }
if ((max_cpu_load - min_cpu_load) >= avg_load_per_task &&
(max_nr_running - min_nr_running) > 1)
@@ -4953,7 +4961,7 @@ void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
u64 scaled_busy_load_per_task;
if (sds->this_nr_running) {
- sds->this_load_per_task /= sds->this_nr_running;
+ do_div(sds->this_load_per_task, sds->this_nr_running);
if (sds->busiest_load_per_task >
sds->this_load_per_task)
imbn = 1;
@@ -4964,7 +4972,7 @@ void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
scaled_busy_load_per_task = sds->busiest_load_per_task
* SCHED_POWER_SCALE;
- scaled_busy_load_per_task /= sds->busiest->sgp->power;
+ do_div(scaled_busy_load_per_task, sds->busiest->sgp->power);
if (sds->max_load - sds->this_load + scaled_busy_load_per_task >=
(scaled_busy_load_per_task * imbn)) {
@@ -4985,20 +4993,21 @@ 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;
+ tmp = (sds->busiest_load_per_task * SCHED_POWER_SCALE);
+ do_div(tmp, sds->busiest->sgp->power);
if (sds->max_load > tmp)
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
- tmp = (sds->busiest_load_per_task * SCHED_POWER_SCALE) /
- sds->this->sgp->power;
+ sds->busiest_load_per_task * SCHED_POWER_SCALE) {
+ tmp = (sds->max_load * sds->busiest->sgp->power);
+ do_div(tmp, sds->this->sgp->power);
+ } else {
+ tmp = (sds->busiest_load_per_task * SCHED_POWER_SCALE);
+ do_div(tmp, sds->this->sgp->power);
+ }
pwr_move += sds->this->sgp->power *
min(sds->this_load_per_task, sds->this_load + tmp);
pwr_move /= SCHED_POWER_SCALE;
@@ -5018,7 +5027,7 @@ static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *s
{
u64 max_pull, load_above_capacity = ~0ULL;
- sds->busiest_load_per_task /= sds->busiest_nr_running;
+ do_div(sds->busiest_load_per_task, sds->busiest_nr_running);
if (sds->group_imb) {
sds->busiest_load_per_task =
min(sds->busiest_load_per_task, sds->avg_load);
@@ -5043,7 +5052,7 @@ static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *s
load_above_capacity *= (SCHED_LOAD_SCALE * SCHED_POWER_SCALE);
- load_above_capacity /= sds->busiest->sgp->power;
+ do_div(load_above_capacity, sds->busiest->sgp->power);
}
/*
@@ -5123,7 +5132,8 @@ find_busiest_group(struct lb_env *env, int *balance)
if (!sds.busiest || sds.busiest_nr_running == 0)
goto ret;
- sds.avg_load = (SCHED_POWER_SCALE * sds.total_load) / sds.total_pwr;
+ sds.avg_load = (SCHED_POWER_SCALE * sds.total_load);
+ do_div(sds.avg_load, sds.total_pwr);
/*
* If the busiest group is imbalanced the below checks don't
@@ -5223,7 +5233,8 @@ static struct rq *find_busiest_queue(struct lb_env *env,
* the load can be moved away from the cpu that is potentially
* running at a lower capacity.
*/
- wl = (wl * SCHED_POWER_SCALE) / power;
+ wl = (wl * SCHED_POWER_SCALE);
+ do_div(wl, power);
if (wl > max_load) {
max_load = wl;
Certain gcc tool chains convert the division on a 64-bit dividend into a __aeabi_uldivmod call which does unnecessary 64-bit by 64-bit divides although the divisor is 32-bit.This 64 by 64 bit division is not implemented in the kernel for reasons of efficiency,which results in undefined reference errors during link time.Hence perform the division on 64-bit dividends using do_div() function. The below use case is the integration of Per-entity-Load-Tracking metric with the load balancer,where cfs_rq->runnable_load_avg, a 64 bit unsigned integer is used to as the base metric for load balancing. Signed-off-by: Preeti U Murthy<preeti@linux.vnet.ibm.com> --- kernel/sched/fair.c | 51 +++++++++++++++++++++++++++++++-------------------- 1 file changed, 31 insertions(+), 20 deletions(-)