mofanv / darknetz

runs several layers of a deep learning model in TrustZone
MIT License
85 stars 29 forks source link

darknetp: TEEC_Opensession failed with code 0xffff0008 origin 0x3 #7

Closed ZLKong closed 4 years ago

ZLKong commented 5 years ago

Hi Mofanv,

I am running a simulation on QEMU-v8 I followed the instructions and put the files in the right folders. But after rebuilding, I still cannot open darknet in the TEE. I tried hello_world, it worked, but darknetp does not. Do you know how to fix it?

Thanks! Kong

mofanv commented 4 years ago

Hi @ZLKong , sorry for the very late response. I'm wondering have you fix it? One reason for TEEC_Opensession failed with code 0xffff0008 origin 0x3 could be that the QEMU does not have enough secure memory to open the TA. Therefore, it has a problem with opening darknetp but can deal with HelloWorld successfully.

You can first try to decrease the required memory size, which is in ta/include/user_ta_header_defines.h

#define TA_DATA_SIZE

to around 2~5MB, then put less number of layers in the TEE, to see if this solve the problem.

Another way is to increase the mapping memory of QEMU. Check this out.

ZLKong commented 4 years ago

Hi Mofanv:

Thank you for your rely. I will try it out.

Best, Kong

ZLKong commented 4 years ago

It worked. Thank you very much!

21212124 commented 4 years ago

Hi @mofanv

I am running optee_example_hello_world on my imx8m mini board and I have this error: TEEC_Opensession failed with code 0xffff0008 origin 0x3 I tried to change TA_DATA_SIZE but it remains the same error, some other idea of why it happens?

Thanks

mofanv commented 4 years ago

Hi @21212124 , I don't have an imx8m board. but seems there is an issue with the board. Does the OPTEE support your board? I don't see this board on their list https://optee.readthedocs.io/en/latest/building/devices/index.html, maybe ask the OPTEE team..

21212124 commented 4 years ago

Hi @mofanv I think it is supported: https://optee.readthedocs.io/en/latest/general/platforms.html?highlight=imx#platforms-supported

I think it is a problem with the .TA