In preparation for the implementation of the accelerator APIs. This is
ramping up to the goal of only storing the export representation in the
key slot, and not keeping the crypto implementation-specific representations
around.

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

Change to on-demand loading of the internal representation when required
in order to call an mbed TLS cryptography API.

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

Change to on-demand loading of the internal representation when required
in order to call an mbed TLS cryptography API.

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

Now that both ECP and RSA keys are represented in export representation,
they can be treated more uniformly.

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

* Updated wording
* Split out buffer allocation to a convenience function
* Moved variable declarations to beginning of their code block

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

* Allocate internal representation contexts on the heap (i.e. don't change
  where they're being allocated)
* Unify load_xxx_representation in terms of allocation and init behaviour

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

* No null-check before calling free
* Close memory leak
* No need for double check of privkey validity

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

Avoids stack-allocating a key slot during ECDH, and mock-attaching a
key to a key slot during key import.

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

PSA Crypto was checking the byte length of a to-be-imported public ECP key
against the expected length for Weierstrass keys, forgetting that
Curve25519/Curve448 exists.

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

* No need to check for NULL before free'ing
* No need to reset variables that weren't touched
* Set output buffer to zero if key output fails
* Document internal functions and rearrange order of input arguments to
  better match other functions.
* Clean up Montgomery fix to be less verbose code

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

* return is treated as a function call
* space between opening and closing parentheses
* remove whiteline between assignment and checking of same variable

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

And zeroize key buffer before freeing to avoid keys hanging around on the
heap.

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>

Since it is being dereferenced by free on exit it should be inited to NULL.
Also added a small test that would trigger the issue.

Signed-off-by: Steven Cooreman <steven.cooreman@silabs.com>