Message-ID: <CAHUa44Fvk4_cayiV62b7yaYm2am3eG+goMFpmkLLx0de1_e20w@mail.gmail.com>
Date:   Tue, 20 Jul 2021 11:59:29 +0200
From:   Jens Wiklander <jens.wiklander@...aro.org>
To:     Sudeep Holla <sudeep.holla@....com>
Cc:     Etienne CARRIERE <etienne.carriere@...com>,
        Sumit Garg <sumit.garg@...aro.org>,
        Marc Zyngier <maz@...nel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        linux-arm-kernel <linux-arm-kernel@...ts.infradead.org>,
        OP-TEE TrustedFirmware <op-tee@...ts.trustedfirmware.org>,
        "devicetree@...r.kernel.org" <devicetree@...r.kernel.org>,
        Linux Doc Mailing List <linux-doc@...r.kernel.org>,
        Jerome Forissier <jerome@...issier.org>,
        Vincent Guittot <vincent.guittot@...aro.org>,
        Rob Herring <robh+dt@...nel.org>,
        Jonathan Corbet <corbet@....net>,
        Ard Biesheuvel <ardb@...nel.org>,
        Etienne Carriere <etienne.carriere@...aro.org>
Subject: Re: [PATCH v2 0/7] Asynchronous notifications from secure world

On Tue, Jul 20, 2021 at 9:51 AM Sudeep Holla <sudeep.holla@....com> wrote:
>
> On Tue, Jul 20, 2021 at 08:45:59AM +0200, Jens Wiklander wrote:
> > On Tue, Jul 13, 2021 at 1:12 PM Sudeep Holla <sudeep.holla@....com> wrote:
> > >
> > > On Fri, Jul 09, 2021 at 08:05:57AM +0000, Etienne CARRIERE wrote:
> > > > Hello Sudeep and all,
> > > >
> > > > On Wed, 7 Jul 2021 at 19:52, Sudeep Holla <sudeep.holla@....com> wrote:
> > > > >
> > > > > Hi Sumit,
> > > > >
> > > > > I was holding off you reply as I didn't have all the background on this.
> > > > > Achin did mention that this is preparatory work for FFA notifications.
> > > > > I did mention to him that this is more than that, it is custom extension
> > > > > to address what FF-A notification is trying to in standard way.
> >
> > Are you suggesting that we should use a hybrid implementation with
> > FF-A for notifications and keep the rest as is for armv7-a?
> >
>
> No I was just mentioning that this patch series addresses notifications from
> secure world(optee in this case) which is very similar to what FF-A is trying
> to address too.

I was hoping that it would be easy to integrate with FF-A in the
OP-TEE driver if it was done in this way.

>
> Anyways, you brought up interesting idea of hybrid model, it would be good if
> that is possible and the specification allows for that. I don't think it does
> in the current form, may need some amendments to allow that I think.

Agree, the scheduling part will need something too.

>
> > > > >
> > > > > I share same opinion as Marc Z.
> >
> > From what I've read in this thread this has mainly been about using
> > SGI notification and not whether asynchronous notification from OP-TEE
> > on non-FF-A systems is good or bad. I assume Sumit was asking about
> > SGI to find out why that wasn't used. This patch set uses SPI.
> >
>
> I understand. I was trying(ineffectively) to tell why it is not so trivial
> to use SGI and how FF-A is enabling that.

OK, thanks for clarifying.

>
> On SPI, so it is expected that platform has SPI available for this ?

Yes, either that or some other source usable with platform_get_irq().

>
> > > > >
> > > > > On Wed, Jul 07, 2021 at 11:22:23AM +0530, Sumit Garg wrote:
> > > > > > On Tue, 6 Jul 2021 at 18:16, Marc Zyngier <maz@...nel.org> wrote:
> > > > >
> > > > > [...]
> > > > >
> > > > > > >
> > > > > > > I don't care about OP-TEE. If you are proposing a contract between S
> > > > > > > and NS, it has to be TEE and OS independent. That's how the
> > > > > > > architecture works.
> > > > > > >
> > > > > >
> > > > > > Agree, here we are not proposing a common contract among the S and NS
> > > > > > world that every TEE (based on Arm TrustZone) will use to communicate
> > > > > > with REE (Linux in our case) but rather an OP-TEE specific
> > > > > > notifications feature that is built on top of OP-TEE specific ABIs.
> > > > > >
> > > > > > And I can see your arguments coming from an FFA perspective but there
> > > > > > are platforms like the ones based on Armv7 which don't support FFA
> > > > > > ABI. Maybe Jens can elaborate how this feature will fit in when FFA
> > > > > > comes into picture?
> > > > > >
> > > > >
> > > > > I can understand that but won't those platforms add the support both in
> > > > > the kernel(current series) and secure world to address notifications.
> > > > > While you could argue that it is small extension to what is already present
> > > > > but I prefer they support FF-A is they need such a support instead of adding
> > > > > custom mechanisms. It is hard to maintain and each vendor will deviate
> > > > > from this custom mechanism and soon we will have bunch of them to handle.
> >
> > Regarding deviation, are we still talking about the OP-TEE driver? So
> > far I haven't seen any vendor extensions at all in that driver.
> >
>
> Yes, I was referring to addition of notification support in both worlds.
> I was trying to emphasize that both OPTEE and FF-A needs changes in the
> secure world. OPTEE changes could be small compared to starting with FF-A
> but it may result in deviation in notification hadling(in both worlds).

Yes they will be different, but they will also be either or. In a FF-A
system it wouldn't make sense to use these OP-TEE notifications. In
fact it wouldn't be possible since it can only be enabled if probed
via the raw SMCs based interface.

>
> > > >
> > > > There exist armv7-a platforms that expect OP-TEE notification support and
> > > > will not move the FF-A, like the stm32mp15. This platform won't move to FF-A
> > > > mainly due to the memory cost of the added SPM layer and the device physical
> > > > constraints.
> > >
> > > Fair enough on the use-case and the analysis for not being able to use FF-A.
> > > As you may already know it doesn't simply this problem. This has been
> > > discussed for years and FF-A was assumed to be the solution when FF-A
> > > spec work started.
> > >
> > > > We have a usecase for OP-TEE notification. We're working on the integration
> > > > of an SCMI server in OP-TEE. SCMI notification is a feature needed is this
> > > > scope and it requires OP-TEE async notification means as those proposed
> > > > here.
> > > >
> > >
> > > I am aware of this use-case, I understand. But I can only share rants
> > > which I know doesn't help much.
> > >
> > > > This OP-TEE async notif also brings a lot of value in OP-TEE as it allows a
> > > > OP-TEE secure thread (i.e. executing a trusted application service) to
> > > > gently wait on a secure interrupt (as a slow bus transaction completion or
> > > > many other usecase) with the CPU relaxed. This support is provided by the
> > > > proposed series. I believe existing device should be able to leverage this
> > > > OP-TEE feature without needing their OP-TEE to move to the new FF-A
> > > > interface.
> > > >
> > >
> > > While I agree these are nice to have in OPTEE, the timing is just odd.
> > >
> > > We are trying hard to push FF-A as standard solution to address all such
> > > issues that couldn't be solved with OPTEE + DT, now we are back to address
> > > the same in parallel to FF-A.
> >
> > It's not exactly the same since the primary target here is armv7-a
> > where introducing FF-A isn't an obvious choice in all cases. For
> > OP-TEE armv7-a is special in the way that all secure world processing
> > is handled by OP-TEE. The internal secure monitor already takes care
> > of what's implemented in TF-A at EL3 for armv8-a.
> >
>
> Fair enough.
>
> > This isn't meant to compete with FF-A, it's to make sure that the
> > OP-TEE armv7-a user base isn't left behind. This doesn't rule out FF-A
> > support for armv7-a for those prepared to take that step.
> >
>
> Sure, as long as that is conveyed to the adopters of this, it should be
> fine. Do you have plans to disable this feature for armv8-a ? I see that
> as safe approach to avoid any kind of conflicts.

I'm not so keen on artificial limitations that encourage strange
workarounds. I'm not sure what would be the best way of disabling this
for armv8-a, it would for instance still make sense for armv8-a
aarch32.

>
> I just don't want similar arguments used as excuse on armv8-a.

As long as there are no technical arguments it shouldn't happen.

Cheers,
Jens

Powered by blists - more mailing lists