On Wed, Mar 23, 2022 at 10:57:23PM -0400, Steven Rostedt wrote:
> From: "Steven Rostedt (Google)" <rostedt@xxxxxxxxxxx>
>
> If the sched_waking (or sched_wakeup) event is found, then use it to time
> the wake up latency for each task.
>
> Signed-off-by: Steven Rostedt (Google) <rostedt@xxxxxxxxxxx>
> ---
> +static void process_wakeup(struct analysis_data *data,
> + struct tep_handle *tep,
> + struct tep_record *record)
> +{
> + struct cpu_data *cpu_data = &data->cpu_data[record->cpu];
> + struct task_cpu_item *cpu_task;
> + struct task_item *task;
> + unsigned long long val;
> + int pid;
> +
> + tep_read_number_field(data->wakeup_pid, record->data, &val);
> + pid = val;
> + cpu_task = get_cpu_task(cpu_data, pid);
Should this use get_task() instead of get_cpu_task()?
I was trying this out and it provides a great overview of a trace but I
found it strange that tasks affine to CPU N were also appearing in the
list for CPU M.
> + task = cpu_task->task;
> + task->wakeup.last = record->ts;
> + task->woken = true;
> +}
|