arch/loongarch/Kconfig | 9 ++++++ arch/loongarch/include/asm/smp.h | 1 + arch/loongarch/include/asm/topology.h | 8 +++++ arch/loongarch/kernel/smp.c | 46 +++++++++++++++++++++++++++ 4 files changed, 64 insertions(+)
From: wanghongliang <wanghongliang@loongson.cn>
In order to achieve more reasonable load balancing behavior,
support for SCHED_MC has been added.
The LLC distribution of Loongarch now is consistent with numa-node,
the balancing domain of SCHED_MC can effectively reduce the situation
where processes are awakened to smt_sibling
Co-developed-by: wanghongliang <wanghongliang@loongson.cn>
Signed-off-by: wanghongliang <wanghongliang@loongson.cn>
Signed-off-by: Tianyang Zhang <zhangtianyang@loongson.cn>
---
arch/loongarch/Kconfig | 9 ++++++
arch/loongarch/include/asm/smp.h | 1 +
arch/loongarch/include/asm/topology.h | 8 +++++
arch/loongarch/kernel/smp.c | 46 +++++++++++++++++++++++++++
4 files changed, 64 insertions(+)
diff --git a/arch/loongarch/Kconfig b/arch/loongarch/Kconfig
index 1a2cf012b..3d6d129ee 100644
--- a/arch/loongarch/Kconfig
+++ b/arch/loongarch/Kconfig
@@ -456,6 +456,15 @@ config SCHED_SMT
Improves scheduler's performance when there are multiple
threads in one physical core.
+config SCHED_MC
+ prompt "Multi-core scheduler support"
+ depends on SMP
+ default y
+ help
+ Multi-core scheduler support improves the CPU scheduler's decision
+ making when dealing with multi-core CPU chips at a cost of slightly
+ increased overhead in some places.
+
config SMP
bool "Multi-Processing support"
help
diff --git a/arch/loongarch/include/asm/smp.h b/arch/loongarch/include/asm/smp.h
index b87d1d5e5..13955d726 100644
--- a/arch/loongarch/include/asm/smp.h
+++ b/arch/loongarch/include/asm/smp.h
@@ -26,6 +26,7 @@ extern int num_processors;
extern int disabled_cpus;
extern cpumask_t cpu_sibling_map[];
extern cpumask_t cpu_core_map[];
+extern cpumask_t cpu_llc_shared_map[];
extern cpumask_t cpu_foreign_map[];
void loongson_smp_setup(void);
diff --git a/arch/loongarch/include/asm/topology.h b/arch/loongarch/include/asm/topology.h
index 50273c918..dfaf45d57 100644
--- a/arch/loongarch/include/asm/topology.h
+++ b/arch/loongarch/include/asm/topology.h
@@ -36,6 +36,14 @@ void numa_set_distance(int from, int to, int distance);
#define topology_sibling_cpumask(cpu) (&cpu_sibling_map[cpu])
#endif
+/*
+ * return cpus that shares the last level cache.
+ */
+static inline const struct cpumask *cpu_coregroup_mask(int cpu)
+{
+ return &cpu_llc_shared_map[cpu];
+}
+
#include <asm-generic/topology.h>
static inline void arch_fix_phys_package_id(int num, u32 slot) { }
diff --git a/arch/loongarch/kernel/smp.c b/arch/loongarch/kernel/smp.c
index 4b24589c0..7b9e996a1 100644
--- a/arch/loongarch/kernel/smp.c
+++ b/arch/loongarch/kernel/smp.c
@@ -50,6 +50,9 @@ EXPORT_SYMBOL(cpu_sibling_map);
cpumask_t cpu_core_map[NR_CPUS] __read_mostly;
EXPORT_SYMBOL(cpu_core_map);
+cpumask_t cpu_llc_shared_map[NR_CPUS] __read_mostly;
+EXPORT_SYMBOL(cpu_llc_shared_map);
+
static DECLARE_COMPLETION(cpu_starting);
static DECLARE_COMPLETION(cpu_running);
@@ -66,6 +69,10 @@ static cpumask_t cpu_sibling_setup_map;
/* representing cpus for which core maps can be computed */
static cpumask_t cpu_core_setup_map;
+/* representing cpus for which llc sibling maps can be computed */
+static cpumask_t cpu_llc_shared_setup_map;
+
+
struct secondary_data cpuboot_data;
static DEFINE_PER_CPU(int, cpu_state);
@@ -102,6 +109,42 @@ static inline void set_cpu_core_map(int cpu)
}
}
+static inline bool cpus_are_shared_llc(int cpua, int cpub)
+{
+ if (cpu_to_node(cpua) != cpu_to_node(cpub))
+ return false;
+
+ return true;
+}
+
+static inline void set_cpu_llc_shared_map(int cpu)
+{
+ int i;
+
+ cpumask_set_cpu(cpu, &cpu_llc_shared_setup_map);
+
+ for_each_cpu(i, &cpu_llc_shared_setup_map) {
+ if (cpus_are_shared_llc(cpu, i)) {
+ cpumask_set_cpu(i, &cpu_llc_shared_map[cpu]);
+ cpumask_set_cpu(cpu, &cpu_llc_shared_map[i]);
+ }
+ }
+}
+
+static inline void clear_cpu_llc_shared_map(int cpu)
+{
+ int i;
+
+ for_each_cpu(i, &cpu_llc_shared_setup_map) {
+ if (cpus_are_shared_llc(cpu, i)) {
+ cpumask_clear_cpu(i, &cpu_llc_shared_map[cpu]);
+ cpumask_clear_cpu(cpu, &cpu_llc_shared_map[i]);
+ }
+ }
+
+ cpumask_clear_cpu(cpu, &cpu_llc_shared_setup_map);
+}
+
static inline void set_cpu_sibling_map(int cpu)
{
int i;
@@ -406,6 +449,7 @@ int loongson_cpu_disable(void)
#endif
set_cpu_online(cpu, false);
clear_cpu_sibling_map(cpu);
+ clear_cpu_llc_shared_map(cpu);
calculate_cpu_foreign_map();
local_irq_save(flags);
irq_migrate_all_off_this_cpu();
@@ -573,6 +617,7 @@ void __init smp_prepare_cpus(unsigned int max_cpus)
loongson_prepare_cpus(max_cpus);
set_cpu_sibling_map(0);
set_cpu_core_map(0);
+ set_cpu_llc_shared_map(0);
calculate_cpu_foreign_map();
#ifndef CONFIG_HOTPLUG_CPU
init_cpu_present(cpu_possible_mask);
@@ -614,6 +659,7 @@ asmlinkage void start_secondary(void)
set_cpu_sibling_map(cpu);
set_cpu_core_map(cpu);
+ set_cpu_llc_shared_map(cpu);
notify_cpu_starting(cpu);
--
2.43.0Hi, Tianyang,
On Wed, Apr 30, 2025 at 5:08 PM Tianyang Zhang
<zhangtianyang@loongson.cn> wrote:
>
> From: wanghongliang <wanghongliang@loongson.cn>
>
> In order to achieve more reasonable load balancing behavior,
> support for SCHED_MC has been added.
> The LLC distribution of Loongarch now is consistent with numa-node,
> the balancing domain of SCHED_MC can effectively reduce the situation
> where processes are awakened to smt_sibling
>
> Co-developed-by: wanghongliang <wanghongliang@loongson.cn>
> Signed-off-by: wanghongliang <wanghongliang@loongson.cn>
> Signed-off-by: Tianyang Zhang <zhangtianyang@loongson.cn>
> ---
> arch/loongarch/Kconfig | 9 ++++++
> arch/loongarch/include/asm/smp.h | 1 +
> arch/loongarch/include/asm/topology.h | 8 +++++
> arch/loongarch/kernel/smp.c | 46 +++++++++++++++++++++++++++
> 4 files changed, 64 insertions(+)
>
> diff --git a/arch/loongarch/Kconfig b/arch/loongarch/Kconfig
> index 1a2cf012b..3d6d129ee 100644
> --- a/arch/loongarch/Kconfig
> +++ b/arch/loongarch/Kconfig
> @@ -456,6 +456,15 @@ config SCHED_SMT
> Improves scheduler's performance when there are multiple
> threads in one physical core.
>
> +config SCHED_MC
> + prompt "Multi-core scheduler support"
This should be bool....
To save your time, I have fixed this typo, refresh the title and
commit message [1], you can download it and then submit V3.
And if needed, also update the version in the internal repo.
[1] https://github.com/chenhuacai/linux/commit/17c9a8142696c428e2da37109ac30deb5133eaf7
Huacai
> + depends on SMP
> + default y
> + help
> + Multi-core scheduler support improves the CPU scheduler's decision
> + making when dealing with multi-core CPU chips at a cost of slightly
> + increased overhead in some places.
> +
> config SMP
> bool "Multi-Processing support"
> help
> diff --git a/arch/loongarch/include/asm/smp.h b/arch/loongarch/include/asm/smp.h
> index b87d1d5e5..13955d726 100644
> --- a/arch/loongarch/include/asm/smp.h
> +++ b/arch/loongarch/include/asm/smp.h
> @@ -26,6 +26,7 @@ extern int num_processors;
> extern int disabled_cpus;
> extern cpumask_t cpu_sibling_map[];
> extern cpumask_t cpu_core_map[];
> +extern cpumask_t cpu_llc_shared_map[];
> extern cpumask_t cpu_foreign_map[];
>
> void loongson_smp_setup(void);
> diff --git a/arch/loongarch/include/asm/topology.h b/arch/loongarch/include/asm/topology.h
> index 50273c918..dfaf45d57 100644
> --- a/arch/loongarch/include/asm/topology.h
> +++ b/arch/loongarch/include/asm/topology.h
> @@ -36,6 +36,14 @@ void numa_set_distance(int from, int to, int distance);
> #define topology_sibling_cpumask(cpu) (&cpu_sibling_map[cpu])
> #endif
>
> +/*
> + * return cpus that shares the last level cache.
> + */
> +static inline const struct cpumask *cpu_coregroup_mask(int cpu)
> +{
> + return &cpu_llc_shared_map[cpu];
> +}
> +
> #include <asm-generic/topology.h>
>
> static inline void arch_fix_phys_package_id(int num, u32 slot) { }
> diff --git a/arch/loongarch/kernel/smp.c b/arch/loongarch/kernel/smp.c
> index 4b24589c0..7b9e996a1 100644
> --- a/arch/loongarch/kernel/smp.c
> +++ b/arch/loongarch/kernel/smp.c
> @@ -50,6 +50,9 @@ EXPORT_SYMBOL(cpu_sibling_map);
> cpumask_t cpu_core_map[NR_CPUS] __read_mostly;
> EXPORT_SYMBOL(cpu_core_map);
>
> +cpumask_t cpu_llc_shared_map[NR_CPUS] __read_mostly;
> +EXPORT_SYMBOL(cpu_llc_shared_map);
> +
> static DECLARE_COMPLETION(cpu_starting);
> static DECLARE_COMPLETION(cpu_running);
>
> @@ -66,6 +69,10 @@ static cpumask_t cpu_sibling_setup_map;
> /* representing cpus for which core maps can be computed */
> static cpumask_t cpu_core_setup_map;
>
> +/* representing cpus for which llc sibling maps can be computed */
> +static cpumask_t cpu_llc_shared_setup_map;
> +
> +
> struct secondary_data cpuboot_data;
> static DEFINE_PER_CPU(int, cpu_state);
>
> @@ -102,6 +109,42 @@ static inline void set_cpu_core_map(int cpu)
> }
> }
>
> +static inline bool cpus_are_shared_llc(int cpua, int cpub)
> +{
> + if (cpu_to_node(cpua) != cpu_to_node(cpub))
> + return false;
> +
> + return true;
> +}
> +
> +static inline void set_cpu_llc_shared_map(int cpu)
> +{
> + int i;
> +
> + cpumask_set_cpu(cpu, &cpu_llc_shared_setup_map);
> +
> + for_each_cpu(i, &cpu_llc_shared_setup_map) {
> + if (cpus_are_shared_llc(cpu, i)) {
> + cpumask_set_cpu(i, &cpu_llc_shared_map[cpu]);
> + cpumask_set_cpu(cpu, &cpu_llc_shared_map[i]);
> + }
> + }
> +}
> +
> +static inline void clear_cpu_llc_shared_map(int cpu)
> +{
> + int i;
> +
> + for_each_cpu(i, &cpu_llc_shared_setup_map) {
> + if (cpus_are_shared_llc(cpu, i)) {
> + cpumask_clear_cpu(i, &cpu_llc_shared_map[cpu]);
> + cpumask_clear_cpu(cpu, &cpu_llc_shared_map[i]);
> + }
> + }
> +
> + cpumask_clear_cpu(cpu, &cpu_llc_shared_setup_map);
> +}
> +
> static inline void set_cpu_sibling_map(int cpu)
> {
> int i;
> @@ -406,6 +449,7 @@ int loongson_cpu_disable(void)
> #endif
> set_cpu_online(cpu, false);
> clear_cpu_sibling_map(cpu);
> + clear_cpu_llc_shared_map(cpu);
> calculate_cpu_foreign_map();
> local_irq_save(flags);
> irq_migrate_all_off_this_cpu();
> @@ -573,6 +617,7 @@ void __init smp_prepare_cpus(unsigned int max_cpus)
> loongson_prepare_cpus(max_cpus);
> set_cpu_sibling_map(0);
> set_cpu_core_map(0);
> + set_cpu_llc_shared_map(0);
> calculate_cpu_foreign_map();
> #ifndef CONFIG_HOTPLUG_CPU
> init_cpu_present(cpu_possible_mask);
> @@ -614,6 +659,7 @@ asmlinkage void start_secondary(void)
>
> set_cpu_sibling_map(cpu);
> set_cpu_core_map(cpu);
> + set_cpu_llc_shared_map(cpu);
>
> notify_cpu_starting(cpu);
>
> --
> 2.43.0
>
>
© 2016 - 2026 Red Hat, Inc.