From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754385AbaDWMYy (ORCPT ); Wed, 23 Apr 2014 08:24:54 -0400 Received: from mail-ob0-f182.google.com ([209.85.214.182]:40348 "EHLO mail-ob0-f182.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752103AbaDWMYw (ORCPT ); Wed, 23 Apr 2014 08:24:52 -0400 MIME-Version: 1.0 In-Reply-To: References: Date: Wed, 23 Apr 2014 17:54:51 +0530 Message-ID: Subject: Re: [Query]: hrtimers: why don't we consider hrtimers in get_next_timer_interrupt() From: Preeti Murthy To: Viresh Kumar Cc: Thomas Gleixner , =?UTF-8?B?RnLDqWTDqXJpYyBXZWlzYmVja2Vy?= , Lists linaro-kernel , Linux Kernel Mailing List , Linaro Networking , Preeti U Murthy Content-Type: text/plain; charset=UTF-8 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Viresh, On Wed, Apr 23, 2014 at 3:08 PM, Viresh Kumar wrote: > Hi, > > File: hrtimer.c > Function: hrtimer_get_next_event() > > I was looking closely at what's waking up my system and > found this piece of code: > > ktime_t hrtimer_get_next_event(void) > { > if (!hrtimer_hres_active()) { > find next hrtimer event .... > } > } > > Which probably means that: "don't consider hrtimers for getting > next timer event if we are running ticks in high resolution mode". > > And I couldn't understand why is it so? In case of high resolution mode, the clock device would have already been programmed for the upcoming hrtimer, the last time the list of hrtimers was evaluated. So we have already taken care of hrtimers to expire next. Note that in high resolution mode as long as the cpu is busy we queue the tick_sched_timer at the next periodic tick. Having said the above: The above mentioned function is called from cmp_next_hrtimer_event() when the scheduler tick needs to be stopped in dynamic tick mode. We now need to know the next appropriate time to setup the tick_sched_timer in high resolution mode and this has to co-incide at a periodic tick granularity. As long as the cpu was busy we were queuing it at the very next tick. But now that the cpu is going idle, we need to see the next opportune multiple of periodic tick to queue this timer at. That is nothing but the time at which the next *timer wheel event* is set to expire. In high resolution mode, the tick_sched_timer is queued at this time while in low resolution mode the clock device is programmed for this time. (Also note that in high resolution mode,if any hrtimer was to expire before the next timer wheel event, the cpu would exit idle and restart tick_sched_timer at the next period again.) So the purpose of hrtimer_get_next_event() is merely to tell you the next timer wheel event. If the system is in high resolution mode, then we do not need to probe the hrtimers for the reason mentioned above: that would have been taken care of in the previous hrtimer_interrupt() call or hrtimer_start() function. We only need to know the time at which tick_sched_timer needs to be queued and for that the timer wheel is sufficient to probe. In low resolution mode however we need to explicitly go through the hrtimer list since its a part and parcel of the timer wheel events. Regards Preeti U Murthy > > When hres_active isn't set, we run hrtimer handlers from timer > handlers, which means that timers would be sufficient in finding > the next event and we don't need to check for hrtimers. > > But when hres_active is set, hrtimers could be set to fire before > the next timer event and so we must take them into account. > > I am new to this whole world of core frameworks and might be > missing something very very obvious. But its always better to > ask than to ignore and think that things are probably fine and > just shut up. > > Thanks for your patience during my learning curve :) > > -- > viresh > -- > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > Please read the FAQ at http://www.tux.org/lkml/