sched: add tracepoints related to NUMA task migration
authorMel Gorman <mgorman@suse.de>
Tue, 21 Jan 2014 23:51:03 +0000 (15:51 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 22 Jan 2014 00:19:48 +0000 (16:19 -0800)
This patch adds three tracepoints
 o trace_sched_move_numa when a task is moved to a node
 o trace_sched_swap_numa when a task is swapped with another task
 o trace_sched_stick_numa when a numa-related migration fails

The tracepoints allow the NUMA scheduler activity to be monitored and the
following high-level metrics can be calculated

 o NUMA migrated stuck  nr trace_sched_stick_numa
 o NUMA migrated idle  nr trace_sched_move_numa
 o NUMA migrated swapped nr trace_sched_swap_numa
 o NUMA local swapped  trace_sched_swap_numa src_nid == dst_nid (should never happen)
 o NUMA remote swapped  trace_sched_swap_numa src_nid != dst_nid (should == NUMA migrated swapped)
 o NUMA group swapped  trace_sched_swap_numa src_ngid == dst_ngid
 Maybe a small number of these are acceptable
 but a high number would be a major surprise.
 It would be even worse if bounces are frequent.
 o NUMA avg task migs.  Average number of migrations for tasks
 o NUMA stddev task mig  Self-explanatory
 o NUMA max task migs.  Maximum number of migrations for a single task

In general the intent of the tracepoints is to help diagnose problems
where automatic NUMA balancing appears to be doing an excessive amount
of useless work.

[akpm@linux-foundation.org: remove semicolon-after-if, repair coding-style]
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Alex Thorlton <athorlton@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/trace/events/sched.h
kernel/sched/core.c
kernel/sched/fair.c

index 04c3084..67e1bbf 100644 (file)
@@ -443,6 +443,93 @@ TRACE_EVENT(sched_process_hang,
 );
 #endif /* CONFIG_DETECT_HUNG_TASK */
 
+DECLARE_EVENT_CLASS(sched_move_task_template,
+
+       TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
+
+       TP_ARGS(tsk, src_cpu, dst_cpu),
+
+       TP_STRUCT__entry(
+               __field( pid_t, pid                     )
+               __field( pid_t, tgid                    )
+               __field( pid_t, ngid                    )
+               __field( int,   src_cpu                 )
+               __field( int,   src_nid                 )
+               __field( int,   dst_cpu                 )
+               __field( int,   dst_nid                 )
+       ),
+
+       TP_fast_assign(
+               __entry->pid            = task_pid_nr(tsk);
+               __entry->tgid           = task_tgid_nr(tsk);
+               __entry->ngid           = task_numa_group_id(tsk);
+               __entry->src_cpu        = src_cpu;
+               __entry->src_nid        = cpu_to_node(src_cpu);
+               __entry->dst_cpu        = dst_cpu;
+               __entry->dst_nid        = cpu_to_node(dst_cpu);
+       ),
+
+       TP_printk("pid=%d tgid=%d ngid=%d src_cpu=%d src_nid=%d dst_cpu=%d dst_nid=%d",
+                       __entry->pid, __entry->tgid, __entry->ngid,
+                       __entry->src_cpu, __entry->src_nid,
+                       __entry->dst_cpu, __entry->dst_nid)
+);
+
+/*
+ * Tracks migration of tasks from one runqueue to another. Can be used to
+ * detect if automatic NUMA balancing is bouncing between nodes
+ */
+DEFINE_EVENT(sched_move_task_template, sched_move_numa,
+       TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
+
+       TP_ARGS(tsk, src_cpu, dst_cpu)
+);
+
+DEFINE_EVENT(sched_move_task_template, sched_stick_numa,
+       TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
+
+       TP_ARGS(tsk, src_cpu, dst_cpu)
+);
+
+TRACE_EVENT(sched_swap_numa,
+
+       TP_PROTO(struct task_struct *src_tsk, int src_cpu,
+                struct task_struct *dst_tsk, int dst_cpu),
+
+       TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu),
+
+       TP_STRUCT__entry(
+               __field( pid_t, src_pid                 )
+               __field( pid_t, src_tgid                )
+               __field( pid_t, src_ngid                )
+               __field( int,   src_cpu                 )
+               __field( int,   src_nid                 )
+               __field( pid_t, dst_pid                 )
+               __field( pid_t, dst_tgid                )
+               __field( pid_t, dst_ngid                )
+               __field( int,   dst_cpu                 )
+               __field( int,   dst_nid                 )
+       ),
+
+       TP_fast_assign(
+               __entry->src_pid        = task_pid_nr(src_tsk);
+               __entry->src_tgid       = task_tgid_nr(src_tsk);
+               __entry->src_ngid       = task_numa_group_id(src_tsk);
+               __entry->src_cpu        = src_cpu;
+               __entry->src_nid        = cpu_to_node(src_cpu);
+               __entry->dst_pid        = task_pid_nr(dst_tsk);
+               __entry->dst_tgid       = task_tgid_nr(dst_tsk);
+               __entry->dst_ngid       = task_numa_group_id(dst_tsk);
+               __entry->dst_cpu        = dst_cpu;
+               __entry->dst_nid        = cpu_to_node(dst_cpu);
+       ),
+
+       TP_printk("src_pid=%d src_tgid=%d src_ngid=%d src_cpu=%d src_nid=%d dst_pid=%d dst_tgid=%d dst_ngid=%d dst_cpu=%d dst_nid=%d",
+                       __entry->src_pid, __entry->src_tgid, __entry->src_ngid,
+                       __entry->src_cpu, __entry->src_nid,
+                       __entry->dst_pid, __entry->dst_tgid, __entry->dst_ngid,
+                       __entry->dst_cpu, __entry->dst_nid)
+);
 #endif /* _TRACE_SCHED_H */
 
 /* This part must be outside protection */
index 36c951b..5ae36cc 100644 (file)
@@ -1108,6 +1108,7 @@ int migrate_swap(struct task_struct *cur, struct task_struct *p)
        if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task)))
                goto out;
 
+       trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu);
        ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg);
 
 out:
@@ -4603,6 +4604,7 @@ int migrate_task_to(struct task_struct *p, int target_cpu)
 
        /* TODO: This is not properly updating schedstats */
 
+       trace_sched_move_numa(p, curr_cpu, target_cpu);
        return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg);
 }
 
index b24b6cf..867b0a4 100644 (file)
@@ -1250,11 +1250,15 @@ static int task_numa_migrate(struct task_struct *p)
        p->numa_scan_period = task_scan_min(p);
 
        if (env.best_task == NULL) {
-               int ret = migrate_task_to(p, env.best_cpu);
+               ret = migrate_task_to(p, env.best_cpu);
+               if (ret != 0)
+                       trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
                return ret;
        }
 
        ret = migrate_swap(p, env.best_task);
+       if (ret != 0)
+               trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
        put_task_struct(env.best_task);
        return ret;
 }