Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge branch 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull scheduler fixes from Ingo Molnar:
"Debug info and other statistics fixes and related enhancements"
* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/numa: Fix numa balancing stats in /proc/pid/sched
sched/numa: Show numa_group ID in /proc/sched_debug task listings
sched/debug: Move print_cfs_rq() declaration to kernel/sched/sched.h
sched/stat: Expose /proc/pid/schedstat if CONFIG_SCHED_INFO=y
sched/stat: Simplify the sched_info accounting dependency
+71
-35
+7
-4
fs/proc/base.c
+7
-4
fs/proc/base.c
···
491
491
}
492
492
#endif
493
493
494
-
#ifdef CONFIG_SCHEDSTATS
494
+
#ifdef CONFIG_SCHED_INFO
495
495
/*
496
496
* Provides /proc/PID/schedstat
497
497
*/
498
498
static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
499
499
struct pid *pid, struct task_struct *task)
500
500
{
501
-
seq_printf(m, "%llu %llu %lu\n",
501
+
if (unlikely(!sched_info_on()))
502
+
seq_printf(m, "0 0 0\n");
503
+
else
504
+
seq_printf(m, "%llu %llu %lu\n",
502
505
(unsigned long long)task->se.sum_exec_runtime,
503
506
(unsigned long long)task->sched_info.run_delay,
504
507
task->sched_info.pcount);
···
2790
2787
#ifdef CONFIG_STACKTRACE
2791
2788
ONE("stack", S_IRUSR, proc_pid_stack),
2792
2789
#endif
2793
-
#ifdef CONFIG_SCHEDSTATS
2790
+
#ifdef CONFIG_SCHED_INFO
2794
2791
ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2795
2792
#endif
2796
2793
#ifdef CONFIG_LATENCYTOP
···
3138
3135
#ifdef CONFIG_STACKTRACE
3139
3136
ONE("stack", S_IRUSR, proc_pid_stack),
3140
3137
#endif
3141
-
#ifdef CONFIG_SCHEDSTATS
3138
+
#ifdef CONFIG_SCHED_INFO
3142
3139
ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3143
3140
#endif
3144
3141
#ifdef CONFIG_LATENCYTOP
+3
-5
include/linux/sched.h
+3
-5
include/linux/sched.h
···
192
192
#ifdef CONFIG_SCHED_DEBUG
193
193
extern void proc_sched_show_task(struct task_struct *p, struct seq_file *m);
194
194
extern void proc_sched_set_task(struct task_struct *p);
195
-
extern void
196
-
print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
197
195
#endif
198
196
199
197
/*
···
836
838
struct backing_dev_info;
837
839
struct reclaim_state;
838
840
839
-
#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
841
+
#ifdef CONFIG_SCHED_INFO
840
842
struct sched_info {
841
843
/* cumulative counters */
842
844
unsigned long pcount; /* # of times run on this cpu */
···
846
848
unsigned long long last_arrival,/* when we last ran on a cpu */
847
849
last_queued; /* when we were last queued to run */
848
850
};
849
-
#endif /* defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) */
851
+
#endif /* CONFIG_SCHED_INFO */
850
852
851
853
#ifdef CONFIG_TASK_DELAY_ACCT
852
854
struct task_delay_info {
···
1395
1397
int rcu_tasks_idle_cpu;
1396
1398
#endif /* #ifdef CONFIG_TASKS_RCU */
1397
1399
1398
-
#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
1400
+
#ifdef CONFIG_SCHED_INFO
1399
1401
struct sched_info sched_info;
1400
1402
#endif
1401
1403
+1
init/Kconfig
+1
init/Kconfig
+1
-1
kernel/sched/core.c
+1
-1
kernel/sched/core.c
···
2164
2164
set_task_cpu(p, cpu);
2165
2165
raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2166
2166
2167
-
#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
2167
+
#ifdef CONFIG_SCHED_INFO
2168
2168
if (likely(sched_info_on()))
2169
2169
memset(&p->sched_info, 0, sizeof(p->sched_info));
2170
2170
#endif
+18
-22
kernel/sched/debug.c
+18
-22
kernel/sched/debug.c
···
142
142
0LL, 0L);
143
143
#endif
144
144
#ifdef CONFIG_NUMA_BALANCING
145
-
SEQ_printf(m, " %d", task_node(p));
145
+
SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p));
146
146
#endif
147
147
#ifdef CONFIG_CGROUP_SCHED
148
148
SEQ_printf(m, " %s", task_group_path(task_group(p)));
···
517
517
SEQ_printf(m, "%-45s:%14Ld.%06ld\n", #F, SPLIT_NS((long long)p->F))
518
518
519
519
520
+
#ifdef CONFIG_NUMA_BALANCING
521
+
void print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
522
+
unsigned long tpf, unsigned long gsf, unsigned long gpf)
523
+
{
524
+
SEQ_printf(m, "numa_faults node=%d ", node);
525
+
SEQ_printf(m, "task_private=%lu task_shared=%lu ", tsf, tpf);
526
+
SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gsf, gpf);
527
+
}
528
+
#endif
529
+
530
+
520
531
static void sched_show_numa(struct task_struct *p, struct seq_file *m)
521
532
{
522
533
#ifdef CONFIG_NUMA_BALANCING
523
534
struct mempolicy *pol;
524
-
int node, i;
525
535
526
536
if (p->mm)
527
537
P(mm->numa_scan_seq);
···
543
533
mpol_get(pol);
544
534
task_unlock(p);
545
535
546
-
SEQ_printf(m, "numa_migrations, %ld\n", xchg(&p->numa_pages_migrated, 0));
547
-
548
-
for_each_online_node(node) {
549
-
for (i = 0; i < 2; i++) {
550
-
unsigned long nr_faults = -1;
551
-
int cpu_current, home_node;
552
-
553
-
if (p->numa_faults)
554
-
nr_faults = p->numa_faults[2*node + i];
555
-
556
-
cpu_current = !i ? (task_node(p) == node) :
557
-
(pol && node_isset(node, pol->v.nodes));
558
-
559
-
home_node = (p->numa_preferred_nid == node);
560
-
561
-
SEQ_printf(m, "numa_faults_memory, %d, %d, %d, %d, %ld\n",
562
-
i, node, cpu_current, home_node, nr_faults);
563
-
}
564
-
}
565
-
536
+
P(numa_pages_migrated);
537
+
P(numa_preferred_nid);
538
+
P(total_numa_faults);
539
+
SEQ_printf(m, "current_node=%d, numa_group_id=%d\n",
540
+
task_node(p), task_numa_group_id(p));
541
+
show_numa_stats(p, m);
566
542
mpol_put(pol);
567
543
#endif
568
544
}
+21
-1
kernel/sched/fair.c
+21
-1
kernel/sched/fair.c
···
8473
8473
print_cfs_rq(m, cpu, cfs_rq);
8474
8474
rcu_read_unlock();
8475
8475
}
8476
-
#endif
8476
+
8477
+
#ifdef CONFIG_NUMA_BALANCING
8478
+
void show_numa_stats(struct task_struct *p, struct seq_file *m)
8479
+
{
8480
+
int node;
8481
+
unsigned long tsf = 0, tpf = 0, gsf = 0, gpf = 0;
8482
+
8483
+
for_each_online_node(node) {
8484
+
if (p->numa_faults) {
8485
+
tsf = p->numa_faults[task_faults_idx(NUMA_MEM, node, 0)];
8486
+
tpf = p->numa_faults[task_faults_idx(NUMA_MEM, node, 1)];
8487
+
}
8488
+
if (p->numa_group) {
8489
+
gsf = p->numa_group->faults[task_faults_idx(NUMA_MEM, node, 0)],
8490
+
gpf = p->numa_group->faults[task_faults_idx(NUMA_MEM, node, 1)];
8491
+
}
8492
+
print_numa_stats(m, node, tsf, tpf, gsf, gpf);
8493
+
}
8494
+
}
8495
+
#endif /* CONFIG_NUMA_BALANCING */
8496
+
#endif /* CONFIG_SCHED_DEBUG */
8477
8497
8478
8498
__init void init_sched_fair_class(void)
8479
8499
{
+13
kernel/sched/sched.h
+13
kernel/sched/sched.h
···
1689
1689
1690
1690
extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1691
1691
extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1692
+
1693
+
#ifdef CONFIG_SCHED_DEBUG
1692
1694
extern void print_cfs_stats(struct seq_file *m, int cpu);
1693
1695
extern void print_rt_stats(struct seq_file *m, int cpu);
1694
1696
extern void print_dl_stats(struct seq_file *m, int cpu);
1697
+
extern void
1698
+
print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1699
+
1700
+
#ifdef CONFIG_NUMA_BALANCING
1701
+
extern void
1702
+
show_numa_stats(struct task_struct *p, struct seq_file *m);
1703
+
extern void
1704
+
print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1705
+
unsigned long tpf, unsigned long gsf, unsigned long gpf);
1706
+
#endif /* CONFIG_NUMA_BALANCING */
1707
+
#endif /* CONFIG_SCHED_DEBUG */
1695
1708
1696
1709
extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1697
1710
extern void init_rt_rq(struct rt_rq *rt_rq);
+2
-2
kernel/sched/stats.h
+2
-2
kernel/sched/stats.h
···
47
47
# define schedstat_set(var, val) do { } while (0)
48
48
#endif
49
49
50
-
#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
50
+
#ifdef CONFIG_SCHED_INFO
51
51
static inline void sched_info_reset_dequeued(struct task_struct *t)
52
52
{
53
53
t->sched_info.last_queued = 0;
···
156
156
#define sched_info_depart(rq, t) do { } while (0)
157
157
#define sched_info_arrive(rq, next) do { } while (0)
158
158
#define sched_info_switch(rq, t, next) do { } while (0)
159
-
#endif /* CONFIG_SCHEDSTATS || CONFIG_TASK_DELAY_ACCT */
159
+
#endif /* CONFIG_SCHED_INFO */
160
160
161
161
/*
162
162
* The following are functions that support scheduler-internal time accounting.
+5
lib/Kconfig.debug
+5
lib/Kconfig.debug
···
841
841
that can help debug the scheduler. The runtime overhead of this
842
842
option is minimal.
843
843
844
+
config SCHED_INFO
845
+
bool
846
+
default n
847
+
844
848
config SCHEDSTATS
845
849
bool "Collect scheduler statistics"
846
850
depends on DEBUG_KERNEL && PROC_FS
851
+
select SCHED_INFO
847
852
help
848
853
If you say Y here, additional code will be inserted into the
849
854
scheduler and related routines to collect statistics about