Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <CAKZGPAOYPp3ANWfBWxcsT3TJdPt8jH-f2ZJzpin=UZ=-b_-QFg@mail.gmail.com>
 <CAKZGPAOY9uxvSTuta+4a41=GGqTxus8YDcv54T7iAL9P-fj_+w@mail.gmail.com> <20230720103746.GC3569127@hirez.programming.kicks-ass.net>
In-Reply-To: <20230720103746.GC3569127@hirez.programming.kicks-ass.net>
Sender: getarunks@gmail.com
From:   Arun KS <arunks.linux@gmail.com>
Date:   Mon, 24 Jul 2023 10:07:28 +0530
Message-ID: <CAKZGPAPuPQkpaZr4kMEg56VXAV+Ug7auxB2brb1=sbzdRz72rw@mail.gmail.com>
Subject: Re: Question on sched_clock
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     linux-kernel@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        Arun KS <getarunks@gmail.com>, pmladek@suse.com,
        tglx@linutronix.de
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Thu, Jul 20, 2023 at 4:07=E2=80=AFPM Peter Zijlstra <peterz@infradead.or=
g> wrote:
>
> On Thu, Jul 20, 2023 at 03:54:56PM +0530, Arun KS wrote:
> > CCing maintainers
> >
> > On Wed, Jul 19, 2023 at 10:36=E2=80=AFAM Arun KS <arunks.linux@gmail.co=
m> wrote:
> > >
> > > Hi,
> > >
> > > Kernel=E2=80=99s printk uses local_clock() for timestamps and it is m=
apped to
> > > sched_clock(). Two problems/requirements I see,
> > >
> > > One, Kernel=E2=80=99s printk timestamps start from 0, I want to chang=
e this to
> > > match with actual time since boot.
>
> You can fundamentally only consistently tell time since the clock gets
> initialized. Starting at 0 is what you get.
>
> > > Two, sched_clock() doesn=E2=80=99t account for time spend in low powe=
r
> > > state(suspend to ram)
>
> Why would we do that? The next person will complain that they don't want
> this. Then another person complains they also want time spend in
> suspend-to-disk, and another person wants a pony.

Thanks Peter, Got your point. I was trying to understand if there are
any other side effects of doing this downstream.

>
> > >
> > > Could workout patches to modify these behaviours and found working in
> > > my system. But need to hear expert opinion on why this is not done in
> > > the upstream.
> > >
> > > diff --git a/kernel/time/sched_clock.c b/kernel/time/sched_clock.c
> > > index 68d6c1190ac7..b63b2ded5727 100644
> > > --- a/kernel/time/sched_clock.c
> > > +++ b/kernel/time/sched_clock.c
>
> This is only one of many sched_clock implementations...
>
> > > @@ -190,7 +190,10 @@ sched_clock_register(u64 (*read)(void), int bits=
,
> > > unsigned long rate)
> > >         /* Update epoch for new counter and update 'epoch_ns' from ol=
d counter*/
> > >         new_epoch =3D read();
> > >         cyc =3D cd.actual_read_sched_clock();
> > > -       ns =3D rd.epoch_ns + cyc_to_ns((cyc - rd.epoch_cyc) &
> > > rd.sched_clock_mask, rd.mult, rd.shift);
> > > +       if (!cyc)
> > > +               ns =3D cyc_to_ns(new_epoch, new_mult, new_shift)
> > > +       else
> > > +               ns =3D rd.epoch_ns + cyc_to_ns((cyc - rd.epoch_cyc) &
> > > rd.sched_clock_mask, rd.mult, rd.shift);
> > >         cd.actual_read_sched_clock =3D read;
> > >
> > >         rd.read_sched_clock     =3D read;
> > >
> > > @@ -287,7 +290,6 @@ void sched_clock_resume(void)
> > >  {
> > >         struct clock_read_data *rd =3D &cd.read_data[0];
> > >
> > > -       rd->epoch_cyc =3D cd.actual_read_sched_clock();
>
> And what if you've been suspended long enough to wrap the clock ?!?
Fair point. Will take a note.

>
> > >         hrtimer_start(&sched_clock_timer, cd.wrap_kt, HRTIMER_MODE_RE=
L_HARD);
> > >         rd->read_sched_clock =3D cd.actual_read_sched_clock;
> > >  }
> > >
> > > Regards,
> > > Arun