PSA: PSoC 6 Correct TRNG behaviour

[orenc17]

Description

On PSA targets the TRNG should be accessible from the secure-side only.
By removing NVSEED and restricting TRNG to secure-core we achieve that requirement.

Relevant tests passed

Pull request type

[ ] Fix
[X] Refactor
[ ] Target update
[ ] Functionality change
[ ] Docs update
[ ] Test update
[ ] Breaking change

Reviewers

Release Notes

@ARMmbed/team-cypress @ARMmbed/mbed-os-psa

[NirSonnenschein]

starting CI pending reviews

[alzix]

Removing an API from compilation is not sufficient from security perspective.
Is the peripheral accessible from cm4?

[orenc17]

I'm not removing the API, I'm letting the HRNG API take over

In PSoC 6 boards the crypto block is assigned to the m0+ core, access from the m4 core will result in Hardfault

[mbed-ci]

Test run: SUCCESS

Summary: 13 of 13 test jobs passed
Build number : 1
Build artifacts

[alzix]

TRNG must be restored
it is device_has_remove

[alzix]

LGTM

[0xc0170]

@ARMmbed/team-cypress Review please, we would like to integrate this PR soon

[NirSonnenschein]

Hi @0xc0170 : as this doesn't directly modify any cypress code Oren says he added cypress so they are aware of the change. will provide more details here shortly

[OlegKapshii]

For me, removing TRNG device from CM4_PSA target (as it was before) looks more logical, than a simple NV_SEED disabling. If there are no TRNG device - than nobody can work with it. If only one feature that needs TRNG is disabled, somebody can try to work with TRNG via another feature. In case anybody tries to work with TRNG (HW PSoC6 block) from CM4, it causes a CM4 HardFault.
But I am not enough familiar with the feature/system internals. If it is ok for ARM, than it is ok for me.

[orenc17]

@0xc0170 this PR doesn't actually change any of the Cypress code.
@ARMmbed/team-cypress The change is needed due to the nature of the PSA target with memory protection scheme.

if a user will try to access the TRNG from the M4 core on a PSA target, the user will get into HardFault.
that is why we enable the HRNG API which calls psa_crypto over SPM, access the TRNG on the M0+ core and return the random data.

the HRNG API relies on the platform having TRNG macro, that is why we add #if !(defined(TARGET_PSA) && defined(COMPONENT_NSPE)) to disable the access to the real TRNG from the M4 core.

[orenc17]

@OlegKapshii defining NVSEED basically sets it as the only entropy source, which is causing the issue
it masks the TRNG functionality

[OlegKapshii]

Now I understand. LGTM