[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 3 Mar 2022 14:43:22 +0200
From: Matti Vaittinen <mazziesaccount@...il.com>
To: Linus Walleij <linus.walleij@...aro.org>,
linux-power <linux-power@...rohmeurope.com>,
Thomas Gleixner <tglx@...utronix.de>,
Anna-Maria Gleixner <anna-maria@...utronix.de>
Cc: Sebastian Reichel <sre@...nel.org>,
Code Kipper <codekipper@...il.com>,
linux-kernel <linux-kernel@...r.kernel.org>,
Lee Jones <lee.jones@...aro.org>
Subject: Re: Question on expiring HRtimer in-kernel
Hi Linus,
On 3/1/22 14:19, Linus Walleij wrote:
> I have a problem with a premature expiring HRtimer.
>
> The HRtimer hrtimer_set_expires_range() is used in two places in
> the upstream kernel:
> kernel/futex/core.c
> drivers/power/supply/ab8500_chargalg.c
>
> Now I am testing the code in the latter, and it has seen some
> bitrot since merged in 2012. Maybe it was correct at one point.
> The timer is started like this:
>
> hrtimer_init(&di->safety_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
> (...)
> hrtimer_set_expires_range(&di->safety_timer,
> ktime_set(timer_expiration * ONE_HOUR_IN_SECONDS, 0),
> ktime_set(FIVE_MINUTES_IN_SECONDS, 0));
> hrtimer_start_expires(&di->safety_timer, HRTIMER_MODE_REL);
>
> What the author wanted to achieve is a very definitive callback in one
> hour relative to now +/- 5 min, and that is one hour later in the
> physical world,
> as this deals with battery charging.
>
> However sometimes this fires almost immediately rather than in an hour.
>
> My first thought is to pass HRTIMER_MODE_REL also to init as
> hrtimer_set_expires_range() could make things happen immediately
> if we have ABS set, but this is all just intuitive.
>
> Any hints? Better ways to create a definitive event in one hour?
>
_a lot_ of water has been flowing in the Oulu river since I last touched
on any code like this. Unfortunately I can't go back to my old code as
it was left in proprietary world. So no promises I am at all helpful here ;)
In any case, I remember few cases where I hit nasty issues because I
used CLOCK_REALTIME - which (AFAIR) is subject to the time adjustments.
NTP, GPS-time and so on can make the time tick in a strange way :) I
guess you would have noticed if time was set when timer did expire.
Anyways, I guess the battery charging should rather be tied CLOCK_MONOTONIC.
No guarantees though - please ignore me if I make no sense. Oh,
actually, please don't - please correct me instead :)
-- Matti
Powered by blists - more mailing lists