From mboxrd@z Thu Jan  1 00:00:00 1970
From: santosh.shilimkar@ti.com (Santosh Shilimkar)
Date: Thu, 02 Jun 2011 18:46:32 +0530
Subject: [PATCHv2 0/3] clocksource: add db8500 PRCMU timer
In-Reply-To: <20110602130406.GW3660@n2100.arm.linux.org.uk>
References: <1307007271-1004-1-git-send-email-mattias.wallin@stericsson.com>
	<20110602094622.GS3660@n2100.arm.linux.org.uk>
	<4DE7637B.3060901@stericsson.com>
	<20110602110137.GU3660@n2100.arm.linux.org.uk>
	<4DE77D9A.6090301@stericsson.com> <4DE788B2.5060308@ti.com>
	<20110602130406.GW3660@n2100.arm.linux.org.uk>
Message-ID: <4DE78D30.6020900@ti.com>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

On 6/2/2011 6:34 PM, Russell King - ARM Linux wrote:
> On Thu, Jun 02, 2011 at 06:27:22PM +0530, Santosh Shilimkar wrote:
>> Earlier this year at ELC SFO, I had a discussion with
>> John and Thomas on how to have a high-resolution clock-source
>> and a low-resolution clock-source working together to cover
>> the low power scenario and still manage to get the highest
>> timer resolution.
>> The idea was to do dynamic switching of clock-source
>> which initially looked simple. Here the idea was to
>> have this working for suspend and as well as cupidle.
>
> I don't think you can do this, because you'll lose precision whenever you
> switch from the high resolution clocksource to the low resolution
> clocksource.
>
> While you can quickly update the current time of day from the highres one,
> the lowres may or may not be a long way from its next tick rollover.  So
> you lose the precision whenever you switch.
>
> However, over a suspend/resume cycle, the precision loss is normally very
> small compared to the time which you have been suspended, so the %age
> error also becomes very small.
>
> With cpuidle, it becomes a completely different matter.  Here, the %age
> error is much larger because of the smaller sleep periods, and chances
> are we can't wait for the low-res timer to change.
>
> So if you're using cpuidle, you really need a clocksource which runs
> continuously, even in whatever states cpuidle drops you into.

According to John, that's what the meta-clock source will for.
It will be continuous and will make use of underneath high res,
low res clock-sources based on the availability.

I let John comment on this on details but I think any other
method would have shortcoming.

Regards
Santosh