Re: [PATCH edk2-platforms v2 1/1] Silicon/SynQuacer: add optional OP-TEE DT node

[Ard Biesheuvel <ard.biesheuvel@linaro.org>]

On 27 July 2018 at 13:37, Sumit Garg <sumit.garg@linaro.org> wrote:
> On Fri, 27 Jul 2018 at 16:40, Daniel Thompson
> <daniel.thompson@linaro.org> wrote:
>>
>> On 26/07/18 09:42, Sumit Garg wrote:
>> > On Thu, 26 Jul 2018 at 13:20, Daniel Thompson
>> > <daniel.thompson@linaro.org> wrote:
>> >>
>> >> On Thu, Jul 26, 2018 at 09:39:37AM +0200, Ard Biesheuvel wrote:
>> >>> On 26 July 2018 at 09:36, Daniel Thompson <daniel.thompson@linaro.org> wrote:
>> >>>> On Wed, Jul 25, 2018 at 12:04:58PM +0200, Ard Biesheuvel wrote:
>> >>>>> On 23 July 2018 at 15:19, Sumit Garg <sumit.garg@linaro.org> wrote:
>> >>>>>> OP-TEE is optional on Developerbox controlled via SCP firmware. To check
>> >>>>>> if we need to delete OP-TEE DT node, we use DRAM1 region info as SCP
>> >>>>>> firmware conditionally carves out Secure memory from DRAM1 region.
>> >>>>>>
>> >>>>>> Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
>> >>>>>> Cc: Leif Lindholm <leif.lindholm@linaro.org>
>> >>>>>> Contributed-under: TianoCore Contribution Agreement 1.1
>> >>>>>> Signed-off-by: Sumit Garg <sumit.garg@linaro.org>
>> >>>>>> ---
>> >>>>>>
>> >>>>>
>> >>>>> As discussed on IRC, i am not a fan of inferring the presence of
>> >>>>> OP-TEE from the base/size values of the first DRAM region.
>> >>>>>
>> >>>>> Please refer to the existing PCIe code how to read a GPIO in PEI and
>> >>>>> set a dynamic PCD accordingly, so you can use its value in
>> >>>>> PlatformDxe.
>> >>>>
>> >>>> For Trusted Firmware I asked Sumit to look for the OP-TEE memory carve
>> >>>> out rather than looking at the switches. This was based on concerns
>> >>>> about version skew (new C-A53 firmware, old SCP firmware[1]), in particular
>> >>>> if TF-A jumps to an OP-TEE that isn't actually loaded the system will
>> >>>> fail in a not very transparent way (especially if the user hasn't found
>> >>>> the debug UART behind the back panel yet).
>> >>>>
>> >>>> What is the consequence of passing a DT with OP-TEE present if one is
>> >>>> not actually present? Do we at least get as far as bringing up the
>> >>>> framebuffer before things explode?
>> >>>>
>> >
>> > If we pass a DT with OP-TEE and OP-TEE not present, Linux TEE generic
>> > driver exits gracefully giving following message:
>> >
>> > [    1.976021] optee: probing for conduit method from DT.
>> > [    1.976033] optee: api uid mismatch
>>
>> That certainly means we can be pretty relaxed about version skew of
>> normal world components (since nothing bad happens if thinks get skewed).
>>
>>
>> >>> Is there any way we can let OP-TEE supply a DT overlay?
>> >>
>> >> I guess it could implement a secure monitor call to provide it. In
>> >> fact I find it a rather pleasing approach. However I think it still loops
>> >> us round to pretty much the same question as before. Does TF-A "protec
>> >> " a normal world that makes an SMC to an OP-TEE that isn't there by
>> >> failing the call in a nice way?
>> >>
>> >
>> > TF-A returns SMC call for OP-TEE as unknown (error code: -1 in "x0"
>> > register) if OP-TEE is not present.
>>
>> It is possible to experiment with getting EDK2 to detect OP-TEE using
>> SMC? This would be fully generic and presumably be the first step in
>> having an EFI OP-TEE driver.
>>
>
> Agree. I will try to detect OP-TEE version via SMC call. If SMC
> unknown is returned, then we say OP-TEE is not present and remove
> corresponding DT node.
>
> So I think this EFI OP-TEE driver makes more sense in edk2 rather than
> edk2-platforms.
>

Indeed.