From: will.deacon@arm.com (Will Deacon) Date: Thu, 28 Feb 2013 16:14:26 +0000 Subject: [PATCH 0/4] Convert highbank to use PSCI calls In-Reply-To: <512E41A7.9040700@gmail.com> References: <1361936528-25998-1-git-send-email-robherring2@gmail.com> <20130227171830.GD22286@mudshark.cambridge.arm.com> <512E41A7.9040700@gmail.com> Message-ID: <20130228161426.GG32130@mudshark.cambridge.arm.com> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org On Wed, Feb 27, 2013 at 05:25:59PM +0000, Rob Herring wrote: > On 02/27/2013 11:18 AM, Will Deacon wrote: > > On Wed, Feb 27, 2013 at 03:42:04AM +0000, Rob Herring wrote: > >> Will, I'm using the id field to specify system level events which are > >> highbank specific. I worry that the id field will be abused for things > >> that could be standardized. Having standard PSCI calls for reset and > >> poweroff would avoid this for highbank. > > > > I don't like using the ID field for this because it places some of the > > number space in the device tree binding and other parts in the platform > > code. It might be better to use the bottom 16 bits of the power_state > > argument to CPU_OFF (which are designated as platform specific anyway) to > > indicate that you actually want a system-wide operation to take place. > > The platform specific ID in the power state field of CPU_OFF is what I'm > using. But there is no reason for this to be platform specific when > possibly all platforms need this. Yes, sorry, I misread your description above and assumed you were referring to the *function* ID. > > Since this document is currently undergoing a fairly major update (which > > will hopefully be released soon), I'll go and check with the authors to see > > if they have already considered this. > > It would be nice to have some visibility to this as this is an ABI and > will break if things change. As long as things are versioned sensibly, we just don't pretend to be compatible with versions of the ABI that we don't support; otherwise PSCI would be unable to evolve as a specification. The main thing is to avoid implementing features as platform-specific parts when an upcoming version of the specification actually describes them in a different way. We should talk more about this at Connect, where hopefully some other platforms can be converted too. Will