Re: [RFC PATCH 0/4] softirq: Per vector threading v3

[Paolo Abeni <pabeni@redhat.com>]

On Wed, 2018-01-24 at 10:05 -0500, David Miller wrote:
> From: Paolo Abeni <pabeni@redhat.com>
> Date: Wed, 24 Jan 2018 15:54:05 +0100
> 
> > Niklas suggested a possible relation with CONFIG_IRQ_TIME_ACCOUNTING=y
> > and indeed he was right.
> > 
> > The patched kernel under test had CONFIG_IRQ_TIME_ACCOUNTING set, and
> > very little CPU time was accounted to the kworker:
> > 
> > [2125 is the relevant kworker's pid]
> > grep sum_exec_runtime /proc/2125/sched; sleep 10; grep sum_exec_runtime /proc/2125/sched
> > se.sum_exec_runtime                          :         13408.239286
> > se.sum_exec_runtime                          :         13456.907197
> > 
> > despite such process was processing a lot of packets and basically
> > burning a CPU.
> 
> So IRQ_TIME_ACCOUNTING makes the scheduler think that the worker
> threads are using nearly no task time at all.

Yes, this is the behavior I observe in the test. But a quick look at
the scheduler code - I'm not very familiar with it - let me think this
is not the intended/expected behaviour for the ksoftirqd (and after
this series, for the kworker serving the softirq).

> The existing ksoftirqd code should hit the same problem, right?

I just tried the vanilla kernel with CONFIG_IRQ_TIME_ACCOUNTING=y, and
in the same test scenario, I observe the 'good' behavior: the user
space process and ksoftirqd share almost fairly the relevant CPU, and
the CPU time spend in softirq processing is accounted to ksoftirqd.

Cheers,

Paolo