Adjust declared dependencies in test_suite_ecdsa

[mpg]

Currently, dependencies in test_suite_ecdsa (both the data and the function file) are expressed in terms of MBEDTLS_SHAxxx_C macros. However, except for deterministic ECDSA, the ECDSA module doesn't compute hashes, and after #6097, neither does its test suite. So, this dependency is too strong, and should be diminished when possible (info about the hash, for example to check length, should be enough).

[mpg]

I was thinking we could create a set of macros just for this test suite, in the header section of test_suite_ecdsa.function, like:

#include "legacy_or_psa.h"
#if ( defined(MBEDTLS_ECDSA_DETERMINISTIC) && defined(MBEDTLS_SHA256_C) ) || \
    defined(MBEDTLS_HAS_ALG_SHA_256_VIA_LOWLEVEL_OR_PSA)
#define MBEDTLS_HAS_ALG_SHA_256_VIA_MD_IF_DETERMINISTIC
#endif

That is: when DETERMINISTIC is defined, we absolutely need the software implementation of SHA-256 (because DETERMINISTIC depends on HMAC-DRBG which depends on MD which can only use the software implementation); otherwise we only need info about the hash, which is controlled by LOWLEVEL_OR_PSA.

(Not tested, just sharing my thoughts as I'm unassigning myself, as I won't have time to do it before going on leave.)

Edited by Andrzej Kurek: I took the liberty to change MBEDTLS_DETERMINISTIC_ECDSA to MBEDTLS_ECDSA_DETERMINISTIC, as this is the define present in the library.

[AndrzejKurek]

I was thinking we could create a set of macros just for this test suite, in the header section of test_suite_ecdsa.function, like:

#include "legacy_or_psa.h"
#if ( defined(MBEDTLS_ECDSA_DETERMINISTIC) && defined(MBEDTLS_SHA256_C) ) || \
    defined(MBEDTLS_HAS_ALG_SHA_256_VIA_LOWLEVEL_OR_PSA)
#define MBEDTLS_HAS_ALG_SHA_256_VIA_MD_IF_DETERMINISTIC
#endif

That is: when DETERMINISTIC is defined, we absolutely need the software implementation of SHA-256 (because DETERMINISTIC depends on HMAC-DRBG which depends on MD which can only use the software implementation); otherwise we only need info about the hash, which is controlled by LOWLEVEL_OR_PSA.

(Not tested, just sharing my thoughts as I'm unassigning myself, as I won't have time to do it before going on leave.)

This set of defines will be functionally identical to MBEDTLS_HAS_ALG_SHA_256_VIA_LOWLEVEL_OR_PSA:

( defined(MBEDTLS_ECDSA_DETERMINISTIC) && defined(MBEDTLS_SHA256_C) ) || 
defined(MBEDTLS_SHA256_C) || 
( defined(MBEDTLS_PSA_CRYPTO_C) && defined(PSA_WANT_ALG_SHA_256) ) 

No matter if MBEDTLS_ECDSA_DETERMINISTIC is defined or not, MBEDTLS_SHA256_C will be enough. This can cause a problem where MBEDTLS_DETERMINISTIC_ECDSA is defined, but MBEDTLS_SHA256_C is not and the basic required info comes from ( defined(MBEDTLS_PSA_CRYPTO_C) && defined(PSA_WANT_ALG_SHA_256) : the deterministic tests will fail.

( defined(MBEDTLS_ECDSA_DETERMINISTIC) && defined(MBEDTLS_SHA256_C) ) || 
( !defined(MBEDTLS_ECDSA_DETERMINISTIC) && defined(MBEDTLS_HAS_ALG_SHA_256_VIA_LOWLEVEL_OR_PSA) )

Would be good in this case. Our current set of tests wouldn't pick up this case, but I'm not sure if it's worth adding a new all.sh component for that.