linux-kernel - Re: [PATCH 13/23] sched,ion: Convert to sched_set

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAKfTPtDa2JPtc2qPSe=d187u8pp7AdvV4wKJiUJsDiAvUTE9hg@mail.gmail.com>
Date:   Wed, 22 Apr 2020 17:09:15 +0200
From:   Vincent Guittot <vincent.guittot@...aro.org>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     Ingo Molnar <mingo@...nel.org>,
        linux-kernel <linux-kernel@...r.kernel.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        Steven Rostedt <rostedt@...dmis.org>,
        Qais Yousef <qais.yousef@....com>,
        Juri Lelli <juri.lelli@...hat.com>,
        Dietmar Eggemann <dietmar.eggemann@....com>,
        Ben Segall <bsegall@...gle.com>, Mel Gorman <mgorman@...e.de>,
        John Stultz <john.stultz@...aro.org>
Subject: Re: [PATCH 13/23] sched,ion: Convert to sched_set_normal()

On Wed, 22 Apr 2020 at 15:59, Peter Zijlstra <peterz@...radead.org> wrote:
>
> On Wed, Apr 22, 2020 at 03:36:22PM +0200, Vincent Guittot wrote:
> > On Wed, 22 Apr 2020 at 15:29, Peter Zijlstra <peterz@...radead.org> wrote:
> > >
> > > On Wed, Apr 22, 2020 at 03:21:45PM +0200, Vincent Guittot wrote:
> > > > On Wed, 22 Apr 2020 at 13:29, Peter Zijlstra <peterz@...radead.org> wrote:
> > > > >
> > > > > In an attempt to take away sched_setscheduler() from modules, change
> > > > > this into sched_set_normal(.nice = 19).
> > > > >
> > > > > Cc: john.stultz@...aro.org
> > > > > Signed-off-by: Peter Zijlstra (Intel) <peterz@...radead.org>
> > > > > Reviewed-by: Ingo Molnar <mingo@...nel.org>
> > > > > ---
> > > > >  drivers/staging/android/ion/ion_heap.c |    3 ---
> > > > >  1 file changed, 3 deletions(-)
> > > > >
> > > > > --- a/drivers/staging/android/ion/ion_heap.c
> > > > > +++ b/drivers/staging/android/ion/ion_heap.c
> > > > > @@ -244,8 +244,6 @@ static int ion_heap_deferred_free(void *
> > > > >
> > > > >  int ion_heap_init_deferred_free(struct ion_heap *heap)
> > > > >  {
> > > > > -       struct sched_param param = { .sched_priority = 0 };
> > > > > -
> > > > >         INIT_LIST_HEAD(&heap->free_list);
> > > > >         init_waitqueue_head(&heap->waitqueue);
> > > > >         heap->task = kthread_run(ion_heap_deferred_free, heap,
> > > > > @@ -255,7 +253,7 @@ int ion_heap_init_deferred_free(struct i
> > > > >                        __func__);
> > > > >                 return PTR_ERR_OR_ZERO(heap->task);
> > > > >         }
> > > > > -       sched_setscheduler(heap->task, SCHED_IDLE, &param);
> > > > > +       sched_set_normal(heap->task, 19);
> > > >
> > > > Would it make sense to have a sched_set_idle(task) to enable kernel
> > > > setting SCHED_IDLE task ?
> > > >
> > > > SCHED_NORMAL w/ nice 19 and SCHED_IDLE tasks are not treated in the
> > > > same way when checking for preemption at  wakeup
> > >
> > > Yeah, but does it really matter? I did indeed consider it, but got
> > > lazy. Is there a definite need for IDLE?
> >
> > John is the best to answer this for this driver but SCHED_IDLE will
> > let other tasks which might be involved in end user interaction like
> > on Android to run first
>
> So I don't much like SCHED_IDLE because it introduces some pretty
> horrible tail latencies. Consider the IDLE task holding a lock, then the
> lock waiter will have to wait until the task gets around to running.

Yes one must be careful when using it
>
> It's not unbounded, like a true idle-time scheduler would be, but it can
> still be pretty horrible. nice19 has some of that too of course, but
> idle has it worse, esp. also because it begs others to preempt it.

Yeah... you have to pay the benefit of letting other tasks to preempt
faster. But both sched_idle and nice19 have the same weight and as a
result they will have the same amount of running time.  So I'm not
sure that sched-idle will have bigger problems than nice 19 but may be
more often

>
> I should get back to proxy execution I suppose...