Re: [PATCH 7/7] rtc: cmos: Add suspend/resume endurance testing hook

[Zhang Rui <rui.zhang@intel.com>]

Hi, Alexandre,

May I know if this addresses your concern or not?

thanks,
rui

On Sat, 2022-05-07 at 15:41 +0800, Zhang Rui wrote:
> On Sat, 2022-05-07 at 09:31 +0200, Alexandre Belloni wrote:
> > On 07/05/2022 10:00:40+0800, Zhang Rui wrote:
> > > Hi, Alexandre,
> > > 
> > > Thanks for reviewing the patch.
> > > 
> > > On Fri, 2022-05-06 at 23:46 +0200, Alexandre Belloni wrote:
> > > > Hello,
> > > > 
> > > > I assume I can ignore this patch as this seems to be only for
> > > > testing
> > > 
> > > The main purpose of this patch is for automate testing.
> > > But this doesn't mean it cannot be part of upstream code, right?
> > > 
> > > > I'm not even sure why this is needed as this completely breaks
> > > > setting
> > > > the alarm time.
> > > 
> > > Or overrides the alarm time, :)
> > > 
> > > People rely on the rtc alarm in the automated suspend stress
> > > test,
> > > which suspend and resume the system for over 1000 iterations.
> > > As I mentioned in the cover letter of this patch series, if the
> > > system
> > > suspend time varies from under 1 second to over 60 seconds, how
> > > to
> > > alarm the RTC before suspend?
> > > This feature is critical in this scenario.
> > > 
> > > Plus, the current solution is transparent to people who don't
> > > known/use
> > > this "rtc_wake_override_sec" parameter. And for people who use
> > > this,
> > > they should know that the previous armed RTC alarm will be
> > > overrode
> > > whenever a system suspend is triggered. I can add a message when
> > > the
> > > parameter is set, if needed.
> > > 
> > 
> > It is not transparent, if I read your patch properly, this breaks
> > wakeup
> > for everyone...
> > 
> > > > On 05/05/2022 09:58:14+0800, Zhang Rui wrote:
> > > > > +static int cmos_pm_notify(struct notifier_block *nb,
> > > > > unsigned
> > > > > long
> > > > > mode, void *_unused)
> > > > > +{
> > > > > +	struct cmos_rtc *cmos = container_of(nb, struct
> > > > > cmos_rtc,
> > > > > pm_nb);
> > > > > +	struct rtc_device       *rtc = cmos->rtc;
> > > > > +	unsigned long           now;
> > > > > +	struct rtc_wkalrm       alm;
> > > > > +
> > > > > +	if (rtc_wake_override_sec == 0)
> > > > > +		return NOTIFY_OK;
> > > > > +
> > > > > +	switch (mode) {
> > > > > +	case PM_SUSPEND_PREPARE:
> > > > > +		/*
> > > > > +		 * Cancel the timer to make sure it won't fire
> > > > > +		 * before rtc is rearmed later.
> > > > > +		 */
> > > > > +		rtc_timer_cancel(rtc, &rtc->aie_timer);
> > > > > +		break;
> > > > > +	case PM_SUSPEND_LATE:
> > > > > +		if (rtc_read_time(rtc, &alm.time))
> > > > > +			return NOTIFY_BAD;
> > > > > +
> > > > > +		now = rtc_tm_to_time64(&alm.time);
> > > > > +		memset(&alm, 0, sizeof(alm));
> > > > > +		rtc_time64_to_tm(now + rtc_wake_override_sec,
> > > > > &alm.time);
> > > > > +		alm.enabled = true;
> > > > > +		if (rtc_set_alarm(rtc, &alm))
> > > > > +			return NOTIFY_BAD;
> > 
> > ... because if rtc_wake_override_sec is not set, this sets the
> > alarm
> > to
> > now which is the current RTC time, ensuring the alarm will never
> > trigger.
> 
> No.
> As the code below
> > > > > 
> > > > > 	if (rtc_wake_override_sec == 0)
> > > > > +		return NOTIFY_OK;
> 
> We check for rtc_wake_override_sec at the beginning of the notifier
> callback. So it takes effect only if a) rtc_wake_override_sec is set,
> AND b) a suspend is triggered.
> 
> thanks,
> rui
> 
>