Re: [PATCH 00/18] ARM/ARM64: Support hierarchical CPU arrangement for PSCI
From: Ulf Hansson
Date: Tue Jun 18 2019 - 08:01:52 EST
On Mon, 10 Jun 2019 at 20:57, Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
>
> On Mon, 10 Jun 2019 at 19:16, Lorenzo Pieralisi
> <lorenzo.pieralisi@xxxxxxx> wrote:
> >
> > On Mon, Jun 10, 2019 at 05:54:39PM +0200, Ulf Hansson wrote:
> >
> > [...]
> >
> > > My summary from the earlier ones, is that because the PSCI spec
> > > includes support for OSI, we should also support it in the kernel (and
> > > ATF). In a discussion offlist, Lorenzo agreed that it's okay to add,
> > > without an apple to apple comparison. Maybe Lorenzo can fill in and
> > > state this publicly, to save us all some time?
> >
> > The comparison should have been made before even requesting PSCI OSI
> > mode changes to the specifications, so we have a chip on our shoulders
> > anyway.
> >
> > We will enable PSCI OSI but that's not where the problem lies, enabling
> > PSCI OSI from a firmware perspective should take 10 lines of code,
> > not:
>
> Thanks for confirming!
>
> >
> > drivers/firmware/psci/Makefile | 2 +-
> > drivers/firmware/psci/psci.c | 219 ++++++++--
> > drivers/firmware/psci/psci.h | 29 ++
> > drivers/firmware/psci/psci_pm_domain.c | 403 ++++++++++++++++++
> >
> > I have some concerns about these changes that I will state in the
> > relevant patches.
>
> Most of the above changes isn't for solely for OSI, but to support a
> hierarchical topology described in the PSCI DT layout. This is for
> example needed when other resources shares the same power rail as the
> CPU cluster.
>
> In other words, the series is orthogonal to whether OSI or PC mode is
> used for PSCI, just to make that clear. BTW, this is what you
> requested me to change into, a while ago.
>
> >
> > > My final point in regards to the OSI mode support, it's a minor part
> > > of the series. I don't see how that should hurt from a maintenance
> > > point of view, or perhaps I am wrong? In any case, I offer my help
> > > with review/maintenance in any form as you may see need/fit.
> >
> > I will go through the series but most of this code should move
> > to core PM code, it has nothing to do with PSCI.
>
> I am looking forward to your review - and for sure, I am open to suggestions!
>
> >
> > BTW, apologies for the delay, I was away.
Lorenzo, a gentle ping.
Kind regards
Uffe