Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge tag 'sched-urgent-2023-09-17' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull scheduler fixes from Ingo Molnar:
"Fix a performance regression on large SMT systems, an Intel SMT4
balancing bug, and a topology setup bug on (Intel) hybrid processors"
* tag 'sched-urgent-2023-09-17' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/sched: Restore the SD_ASYM_PACKING flag in the DIE domain
sched/fair: Fix SMT4 group_smt_balance handling
sched/fair: Optimize should_we_balance() for large SMT systems
+34
-5
+9
-3
arch/x86/kernel/smpboot.c
+9
-3
arch/x86/kernel/smpboot.c
···
579
579
}
580
580
581
581
582
-
#if defined(CONFIG_SCHED_SMT) || defined(CONFIG_SCHED_CLUSTER) || defined(CONFIG_SCHED_MC)
583
582
static inline int x86_sched_itmt_flags(void)
584
583
{
585
584
return sysctl_sched_itmt_enabled ? SD_ASYM_PACKING : 0;
···
602
603
return cpu_cluster_flags() | x86_sched_itmt_flags();
603
604
}
604
605
#endif
605
-
#endif
606
+
607
+
static int x86_die_flags(void)
608
+
{
609
+
if (cpu_feature_enabled(X86_FEATURE_HYBRID_CPU))
610
+
return x86_sched_itmt_flags();
611
+
612
+
return 0;
613
+
}
606
614
607
615
/*
608
616
* Set if a package/die has multiple NUMA nodes inside.
···
646
640
*/
647
641
if (!x86_has_numa_in_package) {
648
642
x86_topology[i++] = (struct sched_domain_topology_level){
649
-
cpu_cpu_mask, SD_INIT_NAME(DIE)
643
+
cpu_cpu_mask, x86_die_flags, SD_INIT_NAME(DIE)
650
644
};
651
645
}
652
646
+25
-2
kernel/sched/fair.c
+25
-2
kernel/sched/fair.c
···
6619
6619
/* Working cpumask for: load_balance, load_balance_newidle. */
6620
6620
static DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
6621
6621
static DEFINE_PER_CPU(cpumask_var_t, select_rq_mask);
6622
+
static DEFINE_PER_CPU(cpumask_var_t, should_we_balance_tmpmask);
6622
6623
6623
6624
#ifdef CONFIG_NO_HZ_COMMON
6624
6625
···
9580
9579
imbalance /= ncores_local + ncores_busiest;
9581
9580
9582
9581
/* Take advantage of resource in an empty sched group */
9583
-
if (imbalance == 0 && local->sum_nr_running == 0 &&
9582
+
if (imbalance <= 1 && local->sum_nr_running == 0 &&
9584
9583
busiest->sum_nr_running > 1)
9585
9584
imbalance = 2;
9586
9585
···
9768
9767
break;
9769
9768
9770
9769
case group_smt_balance:
9770
+
/*
9771
+
* Check if we have spare CPUs on either SMT group to
9772
+
* choose has spare or fully busy handling.
9773
+
*/
9774
+
if (sgs->idle_cpus != 0 || busiest->idle_cpus != 0)
9775
+
goto has_spare;
9776
+
9777
+
fallthrough;
9778
+
9771
9779
case group_fully_busy:
9772
9780
/*
9773
9781
* Select the fully busy group with highest avg_load. In
···
9816
9806
else
9817
9807
return true;
9818
9808
}
9809
+
has_spare:
9819
9810
9820
9811
/*
9821
9812
* Select not overloaded group with lowest number of idle cpus
···
10928
10917
10929
10918
static int should_we_balance(struct lb_env *env)
10930
10919
{
10920
+
struct cpumask *swb_cpus = this_cpu_cpumask_var_ptr(should_we_balance_tmpmask);
10931
10921
struct sched_group *sg = env->sd->groups;
10932
10922
int cpu, idle_smt = -1;
10933
10923
···
10952
10940
return 1;
10953
10941
}
10954
10942
10943
+
cpumask_copy(swb_cpus, group_balance_mask(sg));
10955
10944
/* Try to find first idle CPU */
10956
-
for_each_cpu_and(cpu, group_balance_mask(sg), env->cpus) {
10945
+
for_each_cpu_and(cpu, swb_cpus, env->cpus) {
10957
10946
if (!idle_cpu(cpu))
10958
10947
continue;
10959
10948
···
10966
10953
if (!(env->sd->flags & SD_SHARE_CPUCAPACITY) && !is_core_idle(cpu)) {
10967
10954
if (idle_smt == -1)
10968
10955
idle_smt = cpu;
10956
+
/*
10957
+
* If the core is not idle, and first SMT sibling which is
10958
+
* idle has been found, then its not needed to check other
10959
+
* SMT siblings for idleness:
10960
+
*/
10961
+
#ifdef CONFIG_SCHED_SMT
10962
+
cpumask_andnot(swb_cpus, swb_cpus, cpu_smt_mask(cpu));
10963
+
#endif
10969
10964
continue;
10970
10965
}
10971
10966
···
12939
12918
for_each_possible_cpu(i) {
12940
12919
zalloc_cpumask_var_node(&per_cpu(load_balance_mask, i), GFP_KERNEL, cpu_to_node(i));
12941
12920
zalloc_cpumask_var_node(&per_cpu(select_rq_mask, i), GFP_KERNEL, cpu_to_node(i));
12921
+
zalloc_cpumask_var_node(&per_cpu(should_we_balance_tmpmask, i),
12922
+
GFP_KERNEL, cpu_to_node(i));
12942
12923
12943
12924
#ifdef CONFIG_CFS_BANDWIDTH
12944
12925
INIT_CSD(&cpu_rq(i)->cfsb_csd, __cfsb_csd_unthrottle, cpu_rq(i));