Creating AK triggers TPM failure with hardware TPM on some Intel systems

I'm submitting this on behalf of my team, including @kkaarreell and @sergio-correia.

While running tests on the Rust Keylime agent using a hardware TPM, the team noticed that running the Create AK function triggers a failed TPM state on certain Intel systems (we don't have an exhaustive list). These same hardware TPMs work fine with our other Python agent, which does not use the rust-tss-esapi, leading us to believe this error may be rust-tss-esapi related. There is a miminal reproducer here that fails on this line.

• OS: RHEL 9.1 *see note
• TPM info: 'Issuer: C = DE, O = Infineon Technologies AG, OU = OPTIGA(TM) TPM2.0, CN = Infineon OPTIGA(TM) RSA Manufacturing CA 007'

Note: We don't currently have a Fedora test machine with hardware TPM as of today, but can get one if necessary. @sergio-correia did test this on F35 in the past week and reports similar results.

All tpm2_ commands produce expected output before running the mini reproducer. The output related to the error includes the following below:

This shows how the mini reproducer process hangs on ak::create_ak():

Hello, TPM world!
Preparing to kill the TPM of those Intel Whitley, Wilson City 2S, Ice Lake machines...
Preparing for ek::create_ek_object()
Preparing for ak::create_ak()

Output from dmesg after running the reproducer:

[ 1269.275809] tpm tpm0: tpm2_save_context: failed with a TPM error 0x0101
[ 1269.284334] tpm tpm0: A TPM error (257) occurred flushing context
[ 1269.292270] tpm tpm0: A TPM error (257) occurred flushing context
[ 1269.300268] tpm tpm0: A TPM error (257) occurred flushing context
[ 1269.306366] tpm tpm0: tpm2_commit_space: error -14

with that 257 error indicating...

# tpm2_rc_decode 257
tpm:error(2.0): commands not being accepted because of a TPM failure

Here is the output of tpm2_pcrread after running the reproducer:

# tpm2_pcrread
ERROR:tcti:src/tss2-tcti/tcti-device.c:501:Tss2_Tcti_Device_Init() timeout waiting for response from fd 3 
WARNING:esys:src/tss2-esys/api/Esys_GetCapability.c:303:Esys_GetCapability_Finish() Received TPM Error 
ERROR:esys:src/tss2-esys/api/Esys_GetCapability.c:107:Esys_GetCapability() Esys Finish ErrorCode (0x00000084) 
ERROR: Esys_GetCapability(0x84) - tpm:handle(unk):value is out of range or is not correct for the context
ERROR: Unable to run tpm2_pcrread

And tpm2_gettestresult (the value for data before the test was 0001f9db000000000000):

# tpm2_gettestresult
ERROR:tcti:src/tss2-tcti/tcti-device.c:501:Tss2_Tcti_Device_Init() timeout waiting for response from fd 3 
status:   success
data:   0001f9db0000000001d6

Any ideas on this?

Start with the linked minimal reproducer, especially src/main.rs:42 and the ak::create_ak() call, and run it on an affected Intel hardware TPM. Trace the rust-tss-esapi path used by Create AK while comparing the TPM and dmesg output. Done means the reproducer no longer hangs or leaves the TPM in a failed state on the affected systems.