arch/arm64/imx9: add key store, signing and persistence to the ELE

[royzah]

Summary

The i.MX9 ELE driver reaches the mailbox but not the key store the enclave offers. This adds sessions, key stores, key management, key generation, signing by handle, and the storage exchange that lets a key store outlive a boot.

A key generated in there is permitted one algorithm and one usage, and export can be withheld, so the private half has no command that returns it.

Storage runs the opposite way to every other command: the enclave asks the host to write its key store down and to give it back, and those requests arrive while a command of this side's is still outstanding. The reply tag is what tells a question from an answer.

Two things neither the reference implementation nor the headers say, each of which looks from outside like "this firmware has no key store":

• Key store commands carry a trailing crc, the exclusive or of every word including the header. Without it the enclave answers rating 0xb9.
• A persistent key lifetime is intent only. The strict flag on key generation is what writes the key to the store; without it the next signature answers rating 0x03.

Mailbox waits are now bounded in both directions, and the kilobyte reply buffer no longer sits on the caller's stack.

Impact

i.MX9 only, additive. No Kconfig, no init-time behaviour, nothing enabled by default. Boards that do not call the new functions are unaffected. The command payload layouts move to file scope as named types.

Testing

On an i.MX93 Cortex-A55 board, NuttX 12.11.0, kernel build. A P-256 key generated in the enclave, signing sha256(""):

$ openssl pkeyutl -verify -pubin -inkey pub.der -keyform DER -sigfile sig.der -in digest -pkeyopt digest:sha256
Signature Verified Successfully

Power cycled, then repeated. The key store is restored from the pieces the enclave exported, the public half is byte-identical, and a fresh signature verifies against it. ECDSA is randomised, so verification is the test, not comparison.

tools/checkpatch.sh -f passes on all three files.

[github-actions]

MemBrowse Memory Report

No memory changes detected for:

• arduino-mega2560
• esp32-devkitc
• hifive1-revb
• mirtoo
• qemu-armv8a
• qemu-intel64
• rx65n-rsk2mb
• s698pm-dkit
• stm32-nucleo-f103rb

[royzah]

@acassis @jerpelea could one of you take a look? It builds on #20191 and is already hardware-proven on the i.MX93 (key made, used, survives power loss).

[royzah]

Pushed two fixes found on hardware, squashed in:

• the ELE mailbox and the shared msg had no lock, so /dev/random could interleave with a key store command from another thread. There's now a recursive imx9_ele_lock(); imx9_ele_get_random() takes it itself.
• SIGNATURE_GENERATE was one word short of the PSA layout (salt_len + reserved were missing before the CRC).

nxstyle clean. Sorry for the re-review @xiaoxiang781216

[royzah]

@acassis @pussuw @tkaratapanis if any of you has ten minutes, extra eyes very welcome

[jlaitine]

I added just a few comments, nice work!

I'd re-check the cache operations. I made some quick notes about those, but didn't have time to really think twice. Please just re-check those points!

[jlaitine]

Thanks @royzah, looks nice

[royzah]

@xiaoxiang781216 and @acassis can you take another look ?