From: Wanpeng Li <[email protected]>
In real product setup, there will be houseeking cpus in each nodes, it
is prefer to do housekeeping from local node, fallback to global online
cpumask if failed to find houseeking cpu from local node.
Cc: Ingo Molnar <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Frederic Weisbecker <[email protected]>
Signed-off-by: Wanpeng Li <[email protected]>
---
v1 -> v2:
* introduce sched_numa_find_closest
kernel/sched/isolation.c | 12 ++++++++++--
kernel/sched/sched.h | 5 ++---
kernel/sched/topology.c | 14 ++++++++++++++
3 files changed, 26 insertions(+), 5 deletions(-)
diff --git a/kernel/sched/isolation.c b/kernel/sched/isolation.c
index 123ea07..589afba 100644
--- a/kernel/sched/isolation.c
+++ b/kernel/sched/isolation.c
@@ -16,9 +16,17 @@ static unsigned int housekeeping_flags;
int housekeeping_any_cpu(enum hk_flags flags)
{
- if (static_branch_unlikely(&housekeeping_overridden))
- if (housekeeping_flags & flags)
+ int cpu;
+
+ if (static_branch_unlikely(&housekeeping_overridden)) {
+ if (housekeeping_flags & flags) {
+ cpu = sched_numa_find_closest(housekeeping_mask, smp_processor_id());
+ if (cpu < nr_cpu_ids)
+ return cpu;
+
return cpumask_any_and(housekeeping_mask, cpu_online_mask);
+ }
+ }
return smp_processor_id();
}
EXPORT_SYMBOL_GPL(housekeeping_any_cpu);
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index b08dee2..0db7431 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1212,9 +1212,6 @@ enum numa_topology_type {
extern enum numa_topology_type sched_numa_topology_type;
extern int sched_max_numa_distance;
extern bool find_numa_distance(int distance);
-#endif
-
-#ifdef CONFIG_NUMA
extern void sched_init_numa(void);
extern void sched_domains_numa_masks_set(unsigned int cpu);
extern void sched_domains_numa_masks_clear(unsigned int cpu);
@@ -1224,6 +1221,8 @@ static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
#endif
+extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
+
#ifdef CONFIG_NUMA_BALANCING
/* The regions in numa_faults array from task_struct */
enum numa_faults_stats {
diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index 63184cf..3d3fb04 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -1726,6 +1726,20 @@ void sched_domains_numa_masks_clear(unsigned int cpu)
#endif /* CONFIG_NUMA */
+int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
+{
+#ifdef CONFIG_NUMA
+ int i, j = cpu_to_node(cpu);
+
+ for (i = 0; i < sched_domains_numa_levels; i++) {
+ cpu = cpumask_any_and(cpus, sched_domains_numa_masks[i][j]);
+ if (cpu < nr_cpu_ids)
+ return cpu;
+ }
+#endif
+ return nr_cpu_ids;
+}
+
static int __sdt_alloc(const struct cpumask *cpu_map)
{
struct sched_domain_topology_level *tl;
--
2.7.4
On Fri, Jun 21, 2019 at 09:35:11AM +0800, Wanpeng Li wrote:
> From: Wanpeng Li <[email protected]>
>
> In real product setup, there will be houseeking cpus in each nodes, it
> is prefer to do housekeeping from local node, fallback to global online
> cpumask if failed to find houseeking cpu from local node.
>
> Cc: Ingo Molnar <[email protected]>
> Cc: Peter Zijlstra <[email protected]>
> Cc: Frederic Weisbecker <[email protected]>
> Signed-off-by: Wanpeng Li <[email protected]>
Looks good; did it actually work? :-)
> diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
> index 63184cf..3d3fb04 100644
> --- a/kernel/sched/topology.c
> +++ b/kernel/sched/topology.c
> @@ -1726,6 +1726,20 @@ void sched_domains_numa_masks_clear(unsigned int cpu)
>
> #endif /* CONFIG_NUMA */
Please double check this function; I wrote it in a hurry :-) Also maybe
add a wee comment on top like:
/*
* sched_numa_file_closest() - given the NUMA topology, find the cpu
* closest to @cpu from @cpumask.
* cpumask: cpumask to find a cpu from
* cpu: cpu to be close to
*
* returns: cpu, or >= nr_cpu_ids when nothing found (or !NUMA).
*/
> +int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
> +{
> +#ifdef CONFIG_NUMA
> + int i, j = cpu_to_node(cpu);
> +
> + for (i = 0; i < sched_domains_numa_levels; i++) {
> + cpu = cpumask_any_and(cpus, sched_domains_numa_masks[i][j]);
> + if (cpu < nr_cpu_ids)
> + return cpu;
> + }
> +#endif
> + return nr_cpu_ids;
> +}
On Fri, 21 Jun 2019 at 14:58, Peter Zijlstra <[email protected]> wrote:
>
> On Fri, Jun 21, 2019 at 09:35:11AM +0800, Wanpeng Li wrote:
> > From: Wanpeng Li <[email protected]>
> >
> > In real product setup, there will be houseeking cpus in each nodes, it
> > is prefer to do housekeeping from local node, fallback to global online
> > cpumask if failed to find houseeking cpu from local node.
> >
> > Cc: Ingo Molnar <[email protected]>
> > Cc: Peter Zijlstra <[email protected]>
> > Cc: Frederic Weisbecker <[email protected]>
> > Signed-off-by: Wanpeng Li <[email protected]>
>
> Looks good; did it actually work? :-)
Yeah, it works! :)
>
> > diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
> > index 63184cf..3d3fb04 100644
> > --- a/kernel/sched/topology.c
> > +++ b/kernel/sched/topology.c
> > @@ -1726,6 +1726,20 @@ void sched_domains_numa_masks_clear(unsigned int cpu)
> >
> > #endif /* CONFIG_NUMA */
>
>
> Please double check this function; I wrote it in a hurry :-) Also maybe
So smart. :)
> add a wee comment on top like:
>
> /*
> * sched_numa_file_closest() - given the NUMA topology, find the cpu
> * closest to @cpu from @cpumask.
> * cpumask: cpumask to find a cpu from
> * cpu: cpu to be close to
> *
> * returns: cpu, or >= nr_cpu_ids when nothing found (or !NUMA).
> */
>
Will do in v3.
Regards,
Wanpeng Li
> > +int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
> > +{
> > +#ifdef CONFIG_NUMA
> > + int i, j = cpu_to_node(cpu);
> > +
> > + for (i = 0; i < sched_domains_numa_levels; i++) {
> > + cpu = cpumask_any_and(cpus, sched_domains_numa_masks[i][j]);
> > + if (cpu < nr_cpu_ids)
> > + return cpu;
> > + }
> > +#endif
> > + return nr_cpu_ids;
> > +}