From c61037e905a5cb74c7d786c35ee2cdbab9ed63af Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 28 Aug 2013 12:40:38 +0200 Subject: [PATCH] sched/fair: Fix the group_capacity computation Do away with 'phantom' cores due to N*frac(smt_power) >= 1 by limiting the capacity to the actual number of cores. The assumption of 1 < smt_power < 2 is an actual requirement because of what SMT is so this should work regardless of the SMT implementation. It can still be defeated by creative use of cpu hotplug, but if you're one of those freaks, you get to live with it. Signed-off-by: Peter Zijlstra Acked-by: Vincent Guittot Link: http://lkml.kernel.org/n/tip-dczmbi8tfgixacg1ji2av1un@git.kernel.org Signed-off-by: Ingo Molnar --- kernel/sched/fair.c | 20 +++++++++++++------- 1 file changed, 13 insertions(+), 7 deletions(-) diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 218f9c5b08ce..51c5c3ee77e3 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -4556,18 +4556,24 @@ static inline int sg_imbalanced(struct sched_group *group) /* * Compute the group capacity. * - * For now the capacity is simply the number of power units in the group_power. - * A power unit represents a full core. - * - * This has an issue where N*frac(smt_power) >= 1, in that case we'll see extra - * 'cores' that aren't actually there. + * Avoid the issue where N*frac(smt_power) >= 1 creates 'phantom' cores by + * first dividing out the smt factor and computing the actual number of cores + * and limit power unit capacity with that. */ static inline int sg_capacity(struct lb_env *env, struct sched_group *group) { + unsigned int capacity, smt, cpus; + unsigned int power, power_orig; + + power = group->sgp->power; + power_orig = group->sgp->power_orig; + cpus = group->group_weight; - unsigned int power = group->sgp->power; - unsigned int capacity = DIV_ROUND_CLOSEST(power, SCHED_POWER_SCALE); + /* smt := ceil(cpus / power), assumes: 1 < smt_power < 2 */ + smt = DIV_ROUND_UP(SCHED_POWER_SCALE * cpus, power_orig); + capacity = cpus / smt; /* cores */ + capacity = min_t(unsigned, capacity, DIV_ROUND_CLOSEST(power, SCHED_POWER_SCALE)); if (!capacity) capacity = fix_small_capacity(env->sd, group); -- 2.20.1