Re: Bisected softirq accounting issue in v4.11-rc1~170^2~28

[Jesper Dangaard Brouer]

On Tue, 28 Mar 2017 15:06:04 +0200
Frederic Weisbecker <fweisbec@xxxxxxxxx> wrote:

> On Tue, Mar 28, 2017 at 02:26:42PM +0200, Peter Zijlstra wrote:
> > On Tue, Mar 28, 2017 at 06:34:52PM +0800, Wanpeng Li wrote:  
> > > 
> > > sched_clock_cpu(cpu) should be converted from cputime to ns.  
> > 
> > Uhm, no. sched_clock_cpu() returns u64 in ns.  
> 
> Yes, and most of the cputime_t have been converted to u64 so there
> should be no such conversion issue between u64 and cputime_t anymore.
> 
> Perhaps my commit has another side effect on softirq time accounting,
> I'll see if I can reproduce.

(Disclaimer without knowing anything about the scheduler code)
my theory is that irqtime_account_irq() does not get invoked often
enough, as in my pktgen "overload" use-case keeps softirq always
running. And your change moved updating cpustat[CPUTIME_SOFTIRQ] here.
Before it got updated by account_other_time() which gets invoked from
irqtime_account_process_tick().

-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer