Order of operations in mbedtls_pkcs5_pbkdf2_hmac does not release pico's SHA256 hardware lock

[nilswiersma]

The current way that the SHA256 hardware engine and the mbedtls implementation of mbedtls_pkcs5_pbkdf2_hmac interact with each other results in the situation where that function never releases the SHA256 engine lock. Below I included a trace of how functions are called in pico_mbedtls.c and in mbedtls' pkcs5.c.

You can see that it ends with a call to mbedtls_md_hmac_reset which claims the lock, but then it is never released. I would submit a pull request to address this, but I am not sure if it this a problem of mbedtls, or the link to the pico's SHA256 engine?

Iterations: '3'
Password: 'password' (length 8)
Salt: 686d6163dea2f2bf251b2275
KDF START

mbedtls_pkcs5_pbkdf2_hmac
mbedtls_md_hmac_starts
mbedtls_sha256_starts_ret
pico_sha256_lock
mbedtls_sha256_update_ret
mbedtls_md_hmac_update1
mbedtls_sha256_update_ret
mbedtls_md_hmac_update2
mbedtls_sha256_update_ret
mbedtls_md_hmac_finish1
mbedtls_sha256_finish_ret
pico_sha256_unlock
mbedtls_sha256_starts_ret
pico_sha256_lock
mbedtls_sha256_update_ret
mbedtls_sha256_update_ret
mbedtls_sha256_finish_ret
pico_sha256_unlock
mbedtls_md_hmac_reset1
mbedtls_sha256_starts_ret
pico_sha256_lock
mbedtls_sha256_update_ret
mbedtls_md_hmac_updater3
mbedtls_sha256_update_ret
mbedtls_md_hmac_finish2
mbedtls_sha256_finish_ret
pico_sha256_unlock
mbedtls_sha256_starts_ret
pico_sha256_lock
mbedtls_sha256_update_ret
mbedtls_sha256_update_ret
mbedtls_sha256_finish_ret
pico_sha256_unlock
mbedtls_md_hmac_reset2
mbedtls_sha256_starts_ret
pico_sha256_lock
mbedtls_sha256_update_ret
mbedtls_md_hmac_updater3
mbedtls_sha256_update_ret
mbedtls_md_hmac_finish2
mbedtls_sha256_finish_ret
pico_sha256_unlock
mbedtls_sha256_starts_ret
pico_sha256_lock
mbedtls_sha256_update_ret
mbedtls_sha256_update_ret
mbedtls_sha256_finish_ret
pico_sha256_unlock
mbedtls_md_hmac_reset2
mbedtls_sha256_starts_ret
pico_sha256_lock
mbedtls_sha256_update_ret

KDF took 111ms

[peterharperuk]

It does look a bit like mbedtls_sha256_free should release the lock if needed. Is that your fix? PR would be welcome.

[nilswiersma]

At the moment my fix is something hacky in the application that uses this library. Your suggestion sounds great, thanks! I will work on a PR.