From: robherring2@gmail.com (Rob Herring)
Date: Tue, 21 Jun 2011 08:18:22 -0500
Subject: [RFC PATCH 00/16] Consolidation: move SMP local timers to
	driver/clocksource
In-Reply-To: <4E009679.4010206@arm.com>
References: <1308251204-16719-1-git-send-email-marc.zyngier@arm.com>
	<201106171707.23989.arnd@arndb.de> <4DFB8E20.60802@gmail.com>
	<4E007CEB.6040104@arm.com> <4E0093CF.6030804@gmail.com>
	<4E009679.4010206@arm.com>
Message-ID: <4E009A1E.1080903@gmail.com>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

On 06/21/2011 08:02 AM, Marc Zyngier wrote:
> On 21/06/11 13:51, Rob Herring wrote:
>> Marc,
>>
>> On 06/21/2011 06:13 AM, Marc Zyngier wrote:
>>> On 17/06/11 18:25, Rob Herring wrote:
>>>> Also, device tree support for early platform devices would be useful for
>>>> this.
>>>
>>> What about the attached patch? I've used it to boot a VExpress board
>>> with the following DT snippet (and the corresponding support in the
>>> amp_smp_twd driver):
>>>
>>> 	localtimer at 2c000600 {
>>> 		device_type = "localtimer";
>>
>> My understanding is device_type is only supposed to be used for devices
>> with real OpenFirmware. You could just add localtimer or timer as a 2nd
>> compatible string. I think localtimer is a bit of a Linux name, so just
>> timer may be better.
> 
> Using "compatible" would not work. You'd have to add the same compatible
> string to all drivers, and get drivers probing the wrong device. Any
> other property we could use instead?

No, I'm saying you have a compatible property like this:

compatible = "arm,smp-twd", "timer";

You create the early platform device(s) based on matches with "timer",
but match the driver based on "arm,smp-twd".

> 
> As for the "localtimer" string, this is to differentiate the per-cpu
> timers versus the global timers, and they may need to be probed in a
> specific order (global first, and then local).
> 
> The reason for this is that TWD needs to be calibrated, hence the need
> for a global timer running. Colin's patch could help getting rid of this
> restriction though.

Yes, my original patch adding clock api allowed just that.

Rob

> 
> 	M.