[v4,3/4] sched/fair: Rework sched_use_asym_prio() and sched_asym_prefer()
Commit Message
From: Alex Shi <alexs@kernel.org>
sched_use_asym_prio() sched_asym_prefer() are used together in various
places. Consolidate them into a single function sched_asym().
The existing sched_group_asym() is only used when collecting statistics
of a scheduling group. Rename it as sched_group_asym().
This makes the code easier to read. No functional changes.
Signed-off-by: Alex Shi <alexs@kernel.org>
To: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
To: Valentin Schneider <vschneid@redhat.com>
To: Vincent Guittot <vincent.guittot@linaro.org>
To: Peter Zijlstra <peterz@infradead.org>
To: Ingo Molnar <mingo@redhat.com>
---
kernel/sched/fair.c | 37 ++++++++++++++++++++-----------------
1 file changed, 20 insertions(+), 17 deletions(-)
Comments
On Wed, Feb 07, 2024 at 11:47:03AM +0800, alexs@kernel.org wrote:
> From: Alex Shi <alexs@kernel.org>
>
> sched_use_asym_prio() sched_asym_prefer() are used together in various
s/prio() sched/prio() and sched/
> places. Consolidate them into a single function sched_asym().
>
> The existing sched_group_asym() is only used when collecting statistics
> of a scheduling group. Rename it as sched_group_asym().
> This makes the code easier to read. No functional changes.
>
> Signed-off-by: Alex Shi <alexs@kernel.org>
> To: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
> To: Valentin Schneider <vschneid@redhat.com>
> To: Vincent Guittot <vincent.guittot@linaro.org>
> To: Peter Zijlstra <peterz@infradead.org>
> To: Ingo Molnar <mingo@redhat.com>
> ---
> kernel/sched/fair.c | 37 ++++++++++++++++++++-----------------
> 1 file changed, 20 insertions(+), 17 deletions(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 607dc310b355..942b6358f683 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -9746,8 +9746,18 @@ static bool sched_use_asym_prio(struct sched_domain *sd, int cpu)
> return sd->flags & SD_SHARE_CPUCAPACITY || is_core_idle(cpu);
> }
>
> +static inline bool sched_asym(struct sched_domain *sd, int dst_cpu, int src_cpu)
> +{
> + /*
> + * First check if @dst_cpu can do asym_packing load balance. Only do it
> + * if it has higher priority than @src_cpu.
> + */
> + return sched_use_asym_prio(sd, dst_cpu) &&
> + sched_asym_prefer(dst_cpu, src_cpu);
> +}
> +
> /**
> - * sched_asym - Check if the destination CPU can do asym_packing load balance
> + * sched_group_asym - Check if the destination CPU can do asym_packing balance
> * @env: The load balancing environment
> * @sgs: Load-balancing statistics of the candidate busiest group
> * @group: The candidate busiest group
After renaming and changing this function now its documentation has become
obsolete. Can you update it?
On 2/9/24 10:47 AM, Ricardo Neri wrote:
> On Wed, Feb 07, 2024 at 11:47:03AM +0800, alexs@kernel.org wrote:
>> From: Alex Shi <alexs@kernel.org>
>>
>> sched_use_asym_prio() sched_asym_prefer() are used together in various
>
> s/prio() sched/prio() and sched/
Will take it, Thanks!
>
>> places. Consolidate them into a single function sched_asym().
>>
>> The existing sched_group_asym() is only used when collecting statistics
>> of a scheduling group. Rename it as sched_group_asym().
>> This makes the code easier to read. No functional changes.
>>
>> Signed-off-by: Alex Shi <alexs@kernel.org>
>> To: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
>> To: Valentin Schneider <vschneid@redhat.com>
>> To: Vincent Guittot <vincent.guittot@linaro.org>
>> To: Peter Zijlstra <peterz@infradead.org>
>> To: Ingo Molnar <mingo@redhat.com>
>> ---
>> kernel/sched/fair.c | 37 ++++++++++++++++++++-----------------
>> 1 file changed, 20 insertions(+), 17 deletions(-)
>>
>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>> index 607dc310b355..942b6358f683 100644
>> --- a/kernel/sched/fair.c
>> +++ b/kernel/sched/fair.c
>> @@ -9746,8 +9746,18 @@ static bool sched_use_asym_prio(struct sched_domain *sd, int cpu)
>> return sd->flags & SD_SHARE_CPUCAPACITY || is_core_idle(cpu);
>> }
>>
>> +static inline bool sched_asym(struct sched_domain *sd, int dst_cpu, int src_cpu)
>> +{
>> + /*
>> + * First check if @dst_cpu can do asym_packing load balance. Only do it
>> + * if it has higher priority than @src_cpu.
>> + */
>> + return sched_use_asym_prio(sd, dst_cpu) &&
>> + sched_asym_prefer(dst_cpu, src_cpu);
>> +}
>> +
>> /**
>> - * sched_asym - Check if the destination CPU can do asym_packing load balance
>> + * sched_group_asym - Check if the destination CPU can do asym_packing balance
>> * @env: The load balancing environment
>> * @sgs: Load-balancing statistics of the candidate busiest group
>> * @group: The candidate busiest group
>
> After renaming and changing this function now its documentation has become
> obsolete. Can you update it?
Since the function sched_use_asym_prio() and sched_asym_prefer() both give detailed comments for their actions, as long as this function does, could we just remove the bit obsolete comments here?
@@ -9765,14 +9765,6 @@ static inline bool sched_asym(struct sched_domain *sd, int dst_cpu, int src_cpu)
* @env::dst_cpu can do asym_packing if it has higher priority than the
* preferred CPU of @group.
*
- * SMT is a special case. If we are balancing load between cores, @env::dst_cpu
- * can do asym_packing balance only if all its SMT siblings are idle. Also, it
- * can only do it if @group is an SMT group and has exactly on busy CPU. Larger
- * imbalances in the number of CPUS are dealt with in find_busiest_group().
- *
- * If we are balancing load within an SMT core, or at PKG domain level, always
- * proceed.
- *
* Return: true if @env::dst_cpu can do with asym_packing load balance. False
* otherwise.
>
On Fri, Feb 09, 2024 at 07:08:52PM +0800, kuiliang Shi wrote:
>
>
> On 2/9/24 10:47 AM, Ricardo Neri wrote:
> > On Wed, Feb 07, 2024 at 11:47:03AM +0800, alexs@kernel.org wrote:
> >> From: Alex Shi <alexs@kernel.org>
> >>
> >> sched_use_asym_prio() sched_asym_prefer() are used together in various
> >
> > s/prio() sched/prio() and sched/
>
> Will take it, Thanks!
>
> >
> >> places. Consolidate them into a single function sched_asym().
> >>
> >> The existing sched_group_asym() is only used when collecting statistics
> >> of a scheduling group. Rename it as sched_group_asym().
> >> This makes the code easier to read. No functional changes.
> >>
> >> Signed-off-by: Alex Shi <alexs@kernel.org>
> >> To: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
> >> To: Valentin Schneider <vschneid@redhat.com>
> >> To: Vincent Guittot <vincent.guittot@linaro.org>
> >> To: Peter Zijlstra <peterz@infradead.org>
> >> To: Ingo Molnar <mingo@redhat.com>
> >> ---
> >> kernel/sched/fair.c | 37 ++++++++++++++++++++-----------------
> >> 1 file changed, 20 insertions(+), 17 deletions(-)
> >>
> >> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> >> index 607dc310b355..942b6358f683 100644
> >> --- a/kernel/sched/fair.c
> >> +++ b/kernel/sched/fair.c
> >> @@ -9746,8 +9746,18 @@ static bool sched_use_asym_prio(struct sched_domain *sd, int cpu)
> >> return sd->flags & SD_SHARE_CPUCAPACITY || is_core_idle(cpu);
> >> }
> >>
> >> +static inline bool sched_asym(struct sched_domain *sd, int dst_cpu, int src_cpu)
> >> +{
> >> + /*
> >> + * First check if @dst_cpu can do asym_packing load balance. Only do it
> >> + * if it has higher priority than @src_cpu.
> >> + */
> >> + return sched_use_asym_prio(sd, dst_cpu) &&
> >> + sched_asym_prefer(dst_cpu, src_cpu);
> >> +}
> >> +
> >> /**
> >> - * sched_asym - Check if the destination CPU can do asym_packing load balance
> >> + * sched_group_asym - Check if the destination CPU can do asym_packing balance
> >> * @env: The load balancing environment
> >> * @sgs: Load-balancing statistics of the candidate busiest group
> >> * @group: The candidate busiest group
> >
> > After renaming and changing this function now its documentation has become
> > obsolete. Can you update it?
>
> Since the function sched_use_asym_prio() and sched_asym_prefer() both give detailed comments for their actions, as long as this function does, could we just remove the bit obsolete comments here?
>
>
> @@ -9765,14 +9765,6 @@ static inline bool sched_asym(struct sched_domain *sd, int dst_cpu, int src_cpu)
> * @env::dst_cpu can do asym_packing if it has higher priority than the
> * preferred CPU of @group.
> *
> - * SMT is a special case. If we are balancing load between cores, @env::dst_cpu
> - * can do asym_packing balance only if all its SMT siblings are idle. Also, it
> - * can only do it if @group is an SMT group and has exactly on busy CPU. Larger
> - * imbalances in the number of CPUS are dealt with in find_busiest_group().
> - *
> - * If we are balancing load within an SMT core, or at PKG domain level, always
> - * proceed.
> - *
> * Return: true if @env::dst_cpu can do with asym_packing load balance. False
> * otherwise.
Fine with me.
With this change,
Reviewed-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
Tested on Alder Lake and Meteor Lake, which use asym_packing.
Tested-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
> >
@@ -9746,8 +9746,18 @@ static bool sched_use_asym_prio(struct sched_domain *sd, int cpu)
return sd->flags & SD_SHARE_CPUCAPACITY || is_core_idle(cpu);
}
+static inline bool sched_asym(struct sched_domain *sd, int dst_cpu, int src_cpu)
+{
+ /*
+ * First check if @dst_cpu can do asym_packing load balance. Only do it
+ * if it has higher priority than @src_cpu.
+ */
+ return sched_use_asym_prio(sd, dst_cpu) &&
+ sched_asym_prefer(dst_cpu, src_cpu);
+}
+
/**
- * sched_asym - Check if the destination CPU can do asym_packing load balance
+ * sched_group_asym - Check if the destination CPU can do asym_packing balance
* @env: The load balancing environment
* @sgs: Load-balancing statistics of the candidate busiest group
* @group: The candidate busiest group
@@ -9767,22 +9777,17 @@ static bool sched_use_asym_prio(struct sched_domain *sd, int cpu)
* otherwise.
*/
static inline bool
-sched_asym(struct lb_env *env, struct sg_lb_stats *sgs, struct sched_group *group)
+sched_group_asym(struct lb_env *env, struct sg_lb_stats *sgs, struct sched_group *group)
{
- /* Ensure that the whole local core is idle, if applicable. */
- if (!sched_use_asym_prio(env->sd, env->dst_cpu))
- return false;
-
/*
- * CPU priorities does not make sense for SMT cores with more than one
+ * CPU priorities do not make sense for SMT cores with more than one
* busy sibling.
*/
- if (group->flags & SD_SHARE_CPUCAPACITY) {
- if (sgs->group_weight - sgs->idle_cpus != 1)
- return false;
- }
+ if ((group->flags & SD_SHARE_CPUCAPACITY) &&
+ (sgs->group_weight - sgs->idle_cpus != 1))
+ return false;
- return sched_asym_prefer(env->dst_cpu, group->asym_prefer_cpu);
+ return sched_asym(env->sd, env->dst_cpu, group->asym_prefer_cpu);
}
/* One group has more than one SMT CPU while the other group does not */
@@ -9938,7 +9943,7 @@ static inline void update_sg_lb_stats(struct lb_env *env,
/* Check if dst CPU is idle and preferred to this group */
if (!local_group && env->sd->flags & SD_ASYM_PACKING &&
env->idle != CPU_NOT_IDLE && sgs->sum_h_nr_running &&
- sched_asym(env, sgs, group)) {
+ sched_group_asym(env, sgs, group)) {
sgs->group_asym_packing = 1;
}
@@ -11037,8 +11042,7 @@ static struct rq *find_busiest_queue(struct lb_env *env,
* SMT cores with more than one busy sibling.
*/
if ((env->sd->flags & SD_ASYM_PACKING) &&
- sched_use_asym_prio(env->sd, i) &&
- sched_asym_prefer(i, env->dst_cpu) &&
+ sched_asym(env->sd, i, env->dst_cpu) &&
nr_running == 1)
continue;
@@ -11908,8 +11912,7 @@ static void nohz_balancer_kick(struct rq *rq)
* preferred CPU must be idle.
*/
for_each_cpu_and(i, sched_domain_span(sd), nohz.idle_cpus_mask) {
- if (sched_use_asym_prio(sd, i) &&
- sched_asym_prefer(i, cpu)) {
+ if (sched_asym(sd, i, cpu)) {
flags = NOHZ_STATS_KICK | NOHZ_BALANCE_KICK;
goto unlock;
}