On Thu, Jan 15, 2026 at 02:03:29PM +0100, Neil Armstrong wrote: > On 1/15/26 13:25, Sumit Garg wrote: > > + Jens (OP-TEE driver author in U-Boot) > > > > On Thu, Jan 15, 2026 at 11:49:49AM +0100, [email protected] wrote: > > > On 1/15/26 07:10, Sumit Garg wrote: > > > > On Wed, Jan 14, 2026 at 03:56:02PM +0100, Casey Connolly wrote: > > > > > > > > > > > > > > > On 09/01/2026 12:02, Sumit Garg wrote: > > > > > > On Thu, Jan 08, 2026 at 05:41:42PM +0100, Casey Connolly wrote: > > > > > > > > > > > > > > > > > > > > > On 29/12/2025 12:43, Sumit Garg wrote: > > > > > > > > From: Sumit Garg <[email protected]> > > > > > > > > > > > > > > > > Recently upstream TF-A/OP-TEE has started gaining support for > > > > > > > > Qcom > > > > > > > > platforms. RB3Gen2 being the first one and more to come. U-Boot > > > > > > > > in > > > > > > > > corresponding boot flow is packaged as a position independent > > > > > > > > executable. > > > > > > > > > > > > > > > > So, lets add a generic U-Boot defconfig for Qcom platforms to > > > > > > > > support > > > > > > > > TF-A/OP-TEE based TrustZone stack. Build command: > > > > > > > > > > > > > > > > $ make qcom_tfa_optee_defconfig > > > > > > > > $ make -j`nproc` DEVICE_TREE=qcom/qcs6490-rb3gen2 > > > > > > > > > > > > > > This would be better suited as a config fragment rather than a new > > > > > > > defconfig imo. > > > > > > > > > > > > That's fine with me to add it as a config fragment. > > > > > > > > > > > > > > > > > > > > But more importantly, enabling OPTEE support in U-Boot doesn't > > > > > > > imply > > > > > > > that it will be used, just that it's supported. > > > > > > > > > > > > There are real use-cases of OP-TEE in U-Boot for Qcom platforms like > > > > > > secure EFI variables based on OP-TEE secure storage. Have a look > > > > > > here [1]. > > > > > > > > > > > > And sure there will be more such use-cases like fTPM, KASLR etc. > > > > > > can be > > > > > > supported based on OP-TEE. > > > > > > > > > > I was referring literally to the fact that CONFIG_OPTEE being enabled > > > > > doesn't imply that OP-TEE is running, it's faulty logic to assume > > > > > that's > > > > > the case and add nodes to the DT. > > > > > > > > I don't disagree here as having a runtime check is always a better > > > > choice then a compile time config option. However, there isn't a common > > > > info method from properietary firmware that says if QTEE is running > > > > instead of OP-TEE. > > > > > > > > > > > > > > I just checked and there is an SMC call that tells you the UUID for > > > > > the > > > > > trusted OS, referred to as OPTEE_SMC_CALL_GET_OS_UUID in U-Boot and > > > > > OPTEE_ABI_CALL_GET_OS_UUID in OP-TEE. Presumably this identifies > > > > > OP-TEE > > > > > specifically. > > > > > > > > Also, we don't know how the QTEE will react to this OP-TEE specific SMC > > > > call given it's different variants running on legacy and the newer SoCs. > > > > So I would suggest to better gate OP-TEE presence behind a compile time > > > > check only. > > > > > > So you say it's fine to add the optee node, and the driver will bail out > > > if > > > OPTEE is not present, but it's not good to call OPTEE_SMC_CALL_GET_OS_UUID > > > in the fixup code to enable OPTEE only if present ? > > > > > > It's literally the same, my point in > > > https://lore.kernel.org/all/[email protected]/ > > > was exactly that, just call OPTEE_SMC_CALL_GET_OS_UUID and add the OPTEE > > > node only if present _AND_ if CONFIG_OPTEE is enabled. > > > > > > Move the CONFIG_OPTEE enable in a fragment and we're done, you will only > > > select OPTEE explicitly on desired platforms, and won't run the naughty > > > OPTEE_SMC_CALL_GET_OS_UUID on old crappy platforms. > > > > I am still trying to understand what benefit does invoking > > OPTEE_SMC_CALL_GET_OS_UUID from platform code provides us. Surely it > > can't be used to detect OP-TEE not present when QTEE is running due to > > unknown behaviour with QTEE. > > Sorry but what exactly do you expect that will happen if you enable the OPTEE > driver when running with QTEE ?
The OP-TEE SMC calls are not at all supported with QTEE, so the expected behaviour is undefined. IOW, the OP-TEE SMC ABI is not compatible with QTEE. However, it's going to be hit and trial method to see what QTEE responds to OP-TEE SMC calls. So it's not a reliable source of information we can use to detect which TEE is present or not. -Sumit > > > > > Jens, > > > > Will it be fine with you to expose is_optee_api() from the OP-TEE driver > > for the platform code to invoke it independently? Just for the sake of this > > discussion in case people still insist on it being the right thing to do. > > > > -Sumit > > > > > > > > Neil > > > > > > > > > > > > > > > > > My suggestion would be to make this SMC call if CONFIG_OPTEE is > > > > > enabled > > > > > in qcom_psci_fixup(), compare the UUID and add the node if it matches. > > > > > > > > That's exactly the first SMC call that U-Boot and Linux OP-TEE driver > > > > does to compare the UUID here [1] and bail out of the driver. I don't > > > > see a value of a redundant invoke in the Qcom specific platform code. > > > > > > > > [1] drivers/tee/optee/core.c:823: if (!is_optee_api(pdata->invoke_fn)) > > > > > > > > -Sumit > > > > > > > > > > > > > > > > > > > > > [1] lib/efi_loader/efi_variable_tee.c > > > > > > > > > > > > > > > > > > > > So I think the more appropriate patch here would be to just enable > > > > > > > OP-TEE in qcom_defconfig (assuming the binary size isn't > > > > > > > significantly > > > > > > > affected). > > > > > > > > > > > > The OP-TEE driver in U-Boot itself is probed based on DT and it's > > > > > > not > > > > > > only specific to Qcom platforms but every other platform using > > > > > > OP-TEE. > > > > > > > > > > > > > > > > > > > > Considering the other patch is based on this assumption that if > > > > > > > OP-TEE > > > > > > > support is enabled then the board must be using it, a different > > > > > > > approach > > > > > > > is definitely needed. > > > > > > > > > > > > Yeah that's true even with TF-A boot flow, there is possibility to > > > > > > boot > > > > > > without OP-TEE as well. However, TF-A generally doesn't provide a > > > > > > generic option to detect whether OP-TEE is running or not. > > > > > > > > > > > > > > > > > > > > When I was looking into this last year I remember discussing this > > > > > > > same > > > > > > > issue from the Linux side, there is a good argument to be made > > > > > > > that > > > > > > > OP-TEE support in Linux shouldn't be based on the devicetree - > > > > > > > particularly in the Qualcomm case where whether or not OP-TEE is > > > > > > > used is > > > > > > > a simple software change, nothing to do with hardware. > > > > > > > > > > > > Sadly it's true for every other silicon vendor too. But OP-TEE > > > > > > support > > > > > > based on DT has become an ABI unless migration for OP-TEE support > > > > > > based > > > > > > on FF-A comes into picture. > > > > > > > > > > > > > > > > > > > > So in general I'm not particularly keen on this approach, I think > > > > > > > it > > > > > > > /might/ be acceptable for U-Boot to have some fixup code to add > > > > > > > the > > > > > > > OP-TEE node if OP-TEE is in use with the idea of phasing that out > > > > > > > in > > > > > > > favour of runtime detection in the OS itself. I'd also expect > > > > > > > that fixup > > > > > > > code to go in the generic U-Boot DT fixup code that runs before > > > > > > > we jump > > > > > > > to the OS (like the EFI DT fixup function). > > > > > > > > > > > > The EFI DT fixup code is already there based on U-Boot DT. Have a > > > > > > look > > > > > > here: > > > > > > > > > > > > boot/image-fdt.c:627: fdt_ret = optee_copy_fdt_nodes(blob); > > > > > > > > > > > > In general on Arm platforms there isn't any SMC bus to detect > > > > > > dynamically if there is support for OP-TEE or not. That's why > > > > > > platform bus was choosen for the U-Boot and Linux OP-TEE driver. > > > > > > It's > > > > > > similar to how we have the SCM DT node for Qcom platforms. > > > > > > > > > > > > FF-A bus tries to solve that problem to unify that approach for > > > > > > future > > > > > > platform but U-Boot hasn't yet gained support for FF-A based OP-TEE > > > > > > driver too. > > > > > > > > > > > > Anyhow, this is the sanest way I can come up with to enable OP-TEE > > > > > > support in a general way for all the Qcom platforms. This is aligned > > > > > > with how OP-TEE support is detected for other silicon vendors too. > > > > > > > > > > > > -Sumit > > > > > > > > > > > > > > > > > > > > Kind regards, > > > > > > > > > > > > > > > > > > > > > > > For more information refer here: > > > > > > > > https://trustedfirmware-a.readthedocs.io/en/latest/plat/qti/rb3gen2.html > > > > > > > > > > > > > > > > Signed-off-by: Sumit Garg <[email protected]> > > > > > > > > --- > > > > > > > > configs/qcom_tfa_optee_defconfig | 7 +++++++ > > > > > > > > 1 file changed, 7 insertions(+) > > > > > > > > create mode 100644 configs/qcom_tfa_optee_defconfig > > > > > > > > > > > > > > > > diff --git a/configs/qcom_tfa_optee_defconfig > > > > > > > > b/configs/qcom_tfa_optee_defconfig > > > > > > > > new file mode 100644 > > > > > > > > index 00000000000..c398521770f > > > > > > > > --- /dev/null > > > > > > > > +++ b/configs/qcom_tfa_optee_defconfig > > > > > > > > @@ -0,0 +1,7 @@ > > > > > > > > +# Configuration for building a generic U-Boot image > > > > > > > > +# with support for TF-A/OP-TEE based Arm TrustZone stack. > > > > > > > > + > > > > > > > > +#include "qcom_defconfig" > > > > > > > > + > > > > > > > > +CONFIG_TEE=y > > > > > > > > +CONFIG_OPTEE=y > > > > > > > > > > > > > > -- > > > > > > > // Casey (she/her) > > > > > > > > > > > > > > > > > -- > > > > > // Casey (she/her) > > > > > > > > >
