Re: [PATCH v5 2/5] tick: Add freeze timer events

[dbasehore .]

On Tue, Jul 18, 2017 at 3:22 PM, Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
> On Tue, 18 Jul 2017, dbasehore . wrote:
>> On Tue, Jul 18, 2017 at 2:53 PM, Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>> > On Tue, 18 Jul 2017, dbasehore . wrote:
>> >> On Mon, Jul 17, 2017 at 11:40 PM, Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>> >> > On Mon, 17 Jul 2017, dbasehore . wrote:
>> >> >> On Mon, Jul 17, 2017 at 6:33 PM, Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote:
>> >> >> I could make a patch to try it out. I would probably add a flag to rtc
>> >> >> timers to indicate whether it wakes the system (default true). We
>> >> >> would have to add a sync with the rtc irq and the rtc irqwork. I would
>> >> >> probably add a rtc_timer_sync function that would flush the rtc irq
>> >> >> and flush the irqwork. I would call this after the freeze_ops sync
>> >> >> function since the sci irq needs to finish before syncing with the rtc
>> >> >> irq. Also, pm_wakeup_irq seems racy with the current implementation of
>> >> >> s2idle_loop since the RTC irq could be mistakenly set as pm_wakeup_irq
>> >> >> when something else actually triggered the full wakeup. Fortunately, I
>> >> >> don't think pm_wakeup_irq is used for anything except debugging, but
>> >> >> we might change that.
>> >> >
>> >> > There is another option which you might consider. We can reserve one of the
>> >> > HPET comparators for that purpose and have a special interrupt handler for
>> >> > it. Checking the HPET for expiry from the low level code should be trivial.
>> >> >
>> >> Does that handle setting up new timers properly or does the timer sync
>> >> code still need to be written?
>> >
>> > Sorry, I don't understand the question. What is timer sync code?
>> >
>>
>> Does the comparator allow you to have a completely separate alarm set
>> in the hardware? If there's another timer setup (say some user
>> specified wake alarm), we need to program that alarm after the current
>> one goes off if we aren't able to program multiple alarms at the same
>> time.
>
> The HPET consists of several independent comparator units. The kernel uses
> only a limited set of them. We can reserve one or more for that purpose, so
> it does not require any multiplexing or whatever. How many of them do you
> need?
>

I just need one. Is it a different irq than the RTC? It would be nice
if we could avoid going through the s2idle_loop code.

> Thanks,
>
>         tglx
>