Always enable MBEDTLS_USE_PSA_CRYPTO in all.sh

[gilles-peskine-arm]

A step towards #9630: we no longer test with MBEDTLS_USE_PSA_CRYPTO disabled. It's still possible to disable it explicitly; that will be handled in a follow-up.

Note: this may conflict with the work to migrate depends.py to the PSA configuration (#9292, #9612, #9633, #9634, #9654).

PR checklist

• changelog provided
• development PR here
• framework PR not required
• 3.6 PR not required because: new stuff
• 2.28 PR not required because: new stuff
• tests provided

[mpg]

Looks good to me except it doesn't do exactly what it says on the tin: now USE_PSA is enabled by default, but could still be disabled - either explicitly, or just by failing to define it when using a custom config.

Reading the PR's title, I would have expected something like removing USE_PSA from mbedtls_config.h and then defining it unconditionally in config_adjust_legacy_crypto.h for example. Reading the issue's description only re-inforced that impression.

[gilles-peskine-arm]

Looks good to me except it doesn't do exactly what it says on the tin: now USE_PSA is enabled by default, but could still be disabled - either explicitly, or just by failing to define it when using a custom config.

That's a good point. This accomplishes the title of #9630: we're no longer building code that's guarded by !defined(MBEDTLS_USE_PSA_CRYPTO). But it doesn't reach the definition of done of #9630, which says “MBEDTLS_USE_PSA_CRYPTO is no longer a selectable option in mbedtls_config.h`”.

I think this is a useful milestone, which has passed the CI and which you've reviewed. After this, we no longer have to worry about the non-use-PSA code paths in test code and test scripts. So I propose to declare this as a step towards #9630, and make a follow-up that fully resolves #9630. After that follow-up, we'll no longer have to worry about accidentally not enabling MBEDTLS_USE_PSA_CRYPTO.

The follow-up (#9727) will need to take care of test scripts that query MBEDTLS_USE_PSA_CRYPTO, since that will no longer be a public option covered by query_compile_time_config. So there'll at least be some code to remove in ssl-opt.sh.

[mpg]

Approving now, considering #9614 (comment)

[gilles-peskine-arm]

Uh, I clearly made some mistake in a rebase. Originally this branch had no unset MBEDTLS_USE_PSA_CRYPTO in components-*.sh, but now there are some. I don't think that the current state is a useful milestone: my intent here was that we would never do a CI run with MBEDTLS_USE_PSA_CRYPTO disabled (except possibly in configurations with only low-level crypto). I'm going to fix that.

[gilles-peskine-arm]

So I just accidentally dropped one commit in a rebase. It still applies cleanly so here goes. (But with an amended commit message, because the message didn't work with the changed commit order.)

[mpg]

LGTM

[mpg]

Looking at the outcomes file:

% grep 'Config: !MBEDTLS_USE_PSA_CRYPTO;PASS' outcomes.csv 
Linux-x86_64;component_test_ccm_aes_sha256;test_suite_config.mbedtls_boolean;Config: !MBEDTLS_USE_PSA_CRYPTO;PASS;
Linux-x86_64;component_test_tfm_config_as_is;test_suite_config.mbedtls_boolean;Config: !MBEDTLS_USE_PSA_CRYPTO;PASS;

The first one has nothing that's affected by USE_PSA so it's OK. The 2nd one is out of our control, but FWIW it doesn't seem to have anything that would be affected by USE_PSA either.

[yanesca]

LGTM.

(API break check expected to fail as the PR turns USE_PSA_CRYPTO on in the default configuration.)

[gilles-peskine-arm]

The successor #9727 is ready for review.