On 30 August 2014 19:07, Preeti U Murthy preeti@linux.vnet.ibm.com wrote:
Hi Vincent,
On 08/26/2014 04:36 PM, Vincent Guittot wrote:
capacity_orig is only changed for system with a SMT sched_domain level in order
I think I had asked this before, but why only capacity_orig? The capacity of a group is also being updated the same way. This patch fixes the capacity of a group to reflect the capacity of the heterogeneous CPUs in it, this capacity being both the full capacity of the group: capacity_orig and the capacity available for the fair tasks. So I feel in the subject as well as the changelog it would suffice to say 'capacity'.
IIRC, we discussed that point on a former version. The patch changes only the behavior of capacity_orig but the behavior of capacity stays unchanged as all archs can already set the capacity whereas the capacity_orig was configurable only if the SD_SHARE_CPUCAPACITY was set in the sched_domain. This is no more the case with this patch which creates arch_scale_cpu_capacity for setting capacity_orig.
to reflect the lower capacity of CPUs. Heterogenous system also have to reflect an original capacity that is different from the default value.
Create a more generic function arch_scale_cpu_capacity that can be also used by non SMT platform to set capacity_orig.
The weak behavior of arch_scale_cpu_capacity is the previous SMT one in order to keep backward compatibility in the use of capacity_orig.
arch_scale_smt_capacity and default_scale_smt_capacity have been removed as they were not use elsewhere than in arch_scale_cpu_capacity.
Signed-off-by: Vincent Guittot vincent.guittot@linaro.org
kernel/sched/fair.c | 25 ++++++------------------- 1 file changed, 6 insertions(+), 19 deletions(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index b85e9f7..8176bda 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -5705,19 +5705,12 @@ unsigned long __weak arch_scale_freq_capacity(struct sched_domain *sd, int cpu) return default_scale_capacity(sd, cpu); }
-static unsigned long default_scale_smt_capacity(struct sched_domain *sd, int cpu) +unsigned long __weak arch_scale_cpu_capacity(struct sched_domain *sd, int cpu) {
unsigned long weight = sd->span_weight;unsigned long smt_gain = sd->smt_gain;
if ((sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))return sd->smt_gain / sd->span_weight;
smt_gain /= weight;return smt_gain;-}
-unsigned long __weak arch_scale_smt_capacity(struct sched_domain *sd, int cpu) -{
return default_scale_smt_capacity(sd, cpu);
return SCHED_CAPACITY_SCALE;}
static unsigned long scale_rt_capacity(int cpu) @@ -5756,18 +5749,12 @@ static unsigned long scale_rt_capacity(int cpu)
static void update_cpu_capacity(struct sched_domain *sd, int cpu) {
unsigned long weight = sd->span_weight; unsigned long capacity = SCHED_CAPACITY_SCALE; struct sched_group *sdg = sd->groups;if ((sd->flags & SD_SHARE_CPUCAPACITY) && weight > 1) {if (sched_feat(ARCH_CAPACITY))Aren't you missing this check above? I understand that it is not crucial, but that would also mean removing ARCH_CAPACITY sched_feat altogether, wouldn't it?
Peter has proposed to remove all those checks and to keep only the default behavior because no arch uses arch_scale_smt_capacity. Then, ARCH_CAPACITY is still used in update_cpu_capacity
Regards, Vincent
Regards Preeti U Murthy
capacity *= arch_scale_smt_capacity(sd, cpu);elsecapacity *= default_scale_smt_capacity(sd, cpu);
capacity *= arch_scale_cpu_capacity(sd, cpu);
capacity >>= SCHED_CAPACITY_SHIFT;}
capacity >>= SCHED_CAPACITY_SHIFT; sdg->sgc->capacity_orig = capacity;