Re: [PATCH 03/12] trace-cmd analyze: Show how much tasks run on each CPU

[John Keeping <john@metanate.com>]

On Fri, Mar 25, 2022 at 04:18:19PM -0400, Steven Rostedt wrote:
> On Fri, 25 Mar 2022 19:36:24 +0000
> John Keeping <john@metanate.com> wrote:
> 
> > On Wed, Mar 23, 2022 at 10:57:17PM -0400, Steven Rostedt wrote:
> > > From: "Steven Rostedt (Google)" <rostedt@goodmis.org>
> > > 
> > > Display for each CPU that was traced, the amount of time tasks ran on
> > > them. Listing the tasks from the longest runner to the least.
> > > 
> > > Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
> > > ---
> > > +	for (i = 0; i < nr_tasks; i++) {
> > > +		task = cpu_tasks[i]->task;
> > > +
> > > +		if (!i) {
> > > +			printf("    Task name        PID \t     Run time\n");
> > > +			printf("    ---------        --- \t     --------\n");
> > > +		}
> > > +		printf("%16s %8d\t",
> > > +		       tep_data_comm_from_pid(tep, task->pid),
> > > +		       task->pid);
> > > +		print_time(cpu_tasks[i]->runtime, '_');
> > > +		printf(" (%%%lld)\n", (task->runtime * 100) / total_time);  
> > 
> > Is there a reason for using the CPU-specific runtime for the value and
> > the total runtime for the percentage?
> > 
> > I expected the percentage to be the percentage of this CPU's time spend
> > running the task.
> 
> We modify it so that each CPU has the same run time, unless there's missed
> events at the start (later patches), and then we change total_time to be
> the total time of the events on the CPU and not the entire trace.

I think we're talking about different things here, I probably wasn't
entirely clear about this.  It's the numerator of this division that I'm
concerned about and I wonder if this should be:

	(cpu_tasks[i]->runtime * 100) / total_time

If total_time is 10 seconds and there's a task A which runs on CPU0 for
2 seconds and CPU1 for 1 second then I expect to see:

	CPU 0
	A	2.000  (20%)

	CPU 1
	A	1.000  (10%)

But at the moment both of those lines will give 30% (although the actual
run times are correct).