From: "Steven Rostedt (Google)" <rostedt@xxxxxxxxxxx>
If the sched_switch event is available, use that to figure out how long a
task is running on the CPU. It gives better accuracy than just basing it
off of events.
Signed-off-by: Steven Rostedt (Google) <rostedt@xxxxxxxxxxx>
---
tracecmd/trace-analyze.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/tracecmd/trace-analyze.c b/tracecmd/trace-analyze.c
index dec705ce8e15..56123f9b4f86 100644
--- a/tracecmd/trace-analyze.c
+++ b/tracecmd/trace-analyze.c
@@ -270,6 +270,8 @@ static void process_switch(struct analysis_data *data,
cpu_task = get_cpu_task(cpu_data, pid);
task = cpu_task->task;
+ update_cpu_task_times(cpu_data, cpu_task, record->ts);
+
/* Fill in missing comms */
if (pid && data->prev_comm && !task->comm) {
comm = (char *)(record->data + data->prev_comm->offset);
@@ -283,6 +285,8 @@ static void process_switch(struct analysis_data *data,
pid = val;
cpu_task = get_cpu_task(cpu_data, pid);
task = cpu_task->task;
+ task->start_ts = record->ts;
+ cpu_data->current_pid = pid;
/* Fill in missing comms */
if (pid && data->next_comm && !task->comm) {
--
2.35.1
