Hi Peng,

+TF-A folk.

My 0.02$.

What is the problem you are trying to solve? Why do you need to run OP-TEE and EHF together? EHF was originally written to support a S-EL0 SP that is managed directly by TF-A in EL3 (TF-A folk can chime in). 

The SP could perform RAS error handling for which it needs the EHF. The EHF triages asynchronous exceptions and hands RAS errors to the SP for further handling.

This is just one use case but there is no Trusted OS in these configurations.

So, it would help to understand the requirement. 

cheers,
Achin


From: OP-TEE <op-tee-bounces@lists.trustedfirmware.org> on behalf of Jens Wiklander via OP-TEE <op-tee@lists.trustedfirmware.org>
Sent: 17 March 2021 09:23
To: Peng Fan <peng.fan@nxp.com>
Cc: op-tee@lists.trustedfirmware.org <op-tee@lists.trustedfirmware.org>
Subject: Re: EHF + OPTEE on ARM64
 
On Wed, Mar 17, 2021 at 9:43 AM Peng Fan <peng.fan@nxp.com> wrote:
>
> > Subject: Re: EHF + OPTEE on ARM64
> >
> > On Wed, Mar 17, 2021 at 9:02 AM Peng Fan <peng.fan@nxp.com> wrote:
> > >
> > > > Subject: Re: EHF + OPTEE on ARM64
> > > >
> > > > On Wed, Mar 17, 2021 at 8:41 AM Peng Fan <peng.fan@nxp.com> wrote:
> > > > >
> > > > > > Subject: Re: EHF + OPTEE on ARM64
> > > > > >
> > > > > > On Tue, Mar 16, 2021 at 11:08 AM Peng Fan <peng.fan@nxp.com>
> > wrote:
> > > > > > >
> > > > > > > Hi,
> > > > > > >
> > > > > > > In bl31/ehf.c, there are following two lines, per my
> > > > > > > understanding, when cpu is in secure world, the non-secure
> > > > > > > interrupt as FIQ(GICv3) will be directly catched by EL3, not S-EL1
> > > > > > >         /* Route EL3 interrupts when in Secure and Non-secure.
> > */
> > > > > > >         set_interrupt_rm_flag(flags, NON_SECURE);
> > > > > > >         set_interrupt_rm_flag(flags, SECURE);
> > > > > > >
> > > > > > > So this will conflict with OP-TEE, because OP-TEE needs catch
> > > > > > > NS-interrupt as FIQ in S-EL1 world.
> > > > > >
> > > > > > In the case of GICv3, OP-TEE is configured to receive the
> > > > > > non-secure interrupts as FIQ and secure interrupts as IRQ. See
> > CFG_ARM_GICV3.
> > > > >
> > > > > But EHF needs NS-interrupt FIQ be catched by EL3 if I understand
> > > > > correct, per " set_interrupt_rm_flag(flags, SECURE);"
> > > > >
> > > > > So currently EHF could not work together with OP-TEE, right?
> > > >
> > > > To be honest, I'm not completely sure what EHF does. From OP-TEE
> > > > point of view we expect to receive the non-secure interrupts as a
> > > > way of doing a controlled exit. This allows OP-TEE to resume
> > > > execution with a different core on re-entry. If EL3 takes the
> > > > non-secure interrupts directly it will have to make sure to only re-enter
> > OP-TEE on this core as a return from exception.
> > >
> > > Is this easy to be achieved?
> >
> > I don't know, it depends on what you intend to do with this non-secure
> > interrupt. If it's handled at EL3 and then there's a return from exception back
> > to S-EL1 there's likely no harm done. But if there's a world switch involved
> > there might be trouble, OP-TEE might not be in a suitable state for a world
> > switch.
> >
> > >
> > > Or by using opteed_sel1_interrupt_handler, could we have similar
> > > behavior to allow the other core resume execution?
> >
> > Only OP-TEE itself can make a controlled exit as there's an internal state to
> > maintain. Currently that's signalled with a non-secure interrupt.
>
>
> Per EHF, https://trustedfirmware-a.readthedocs.io/en/latest/components/exception-handling.html?highlight=Exception%20Handling%20Framework
> On GICv3 systems, when executing in S-EL1, pending Non-secure interrupts of
> sufficient priority are signalled as FIQs, and therefore will be routed to EL3.
> As a result, S-EL1 software cannot expect to handle Non-secure interrupts at S-EL1.
> Essentially, this deprecates the routing mode described as CSS=0, TEL3=0.
>
> In order for S-EL1 software to handle Non-secure interrupts while having EHF enabled,
> the dispatcher must adopt a model where Non-secure interrupts are received at EL3,
> but are then synchronously handled over to S-EL1.
>
> The issue to me here how to synchronously handled over to S-EL1 and not break optee.

I understand. OP-TEE is masking interrupts in some critical sections,
while in such a state OP-TEE cannot handle any asynchronous interrupt.
Temporarily masking interrupts is normally a quick operation so we do
it in quite a few places.
So the crux of the problem is to make sure that OP-TEE is in a state
where it can make a controlled exit. I don't have any good ideas for
this right now.

Cheers,
Jens