Skip to content

testing.md

Latest commit

 

History

History
103 lines (84 loc) · 5.24 KB

testing.md

File metadata and controls

103 lines (84 loc) · 5.24 KB
Raw

Build and validation

Supported build profiles are x86_64, Cortex-M4 and Cortex-M7. The existing DMOD SDK, host tools and runtime are used without format changes. DMFs retain the original 240-byte header. The backend forks use typed DMOD_GLOBAL_POINTER slots for callbacks/coordinates and inline immutable descriptor text.

Final ELF metadata is audited at build time; it is never used to rewrite a DMF or generate a runtime relocation table. Absolute pointers outside existing header/API/GOT relocation sections fail the build. The audit currently reads ARM ELF32 and x86-64 ELF64; it rejects unsupported ELF architectures explicitly.

Native

cmake -S . -B build/native -DDMOD_DIR=/path/to/dmod
cmake --build build/native --parallel
ctest --test-dir build/native --output-on-failure

Three CTest entries cover different boundaries:

  • test_dmtls: three checks in the actual DMOD module loaded by dmod_loader, including crypto initialization, the entropy callback, CA parsing, ownership and shutdown. The deterministic entropy in this fixture is for tests only; that test never opens a connection or protects application data.

  • tls_interoperability: 18 checks compile the production engine and backend into a host test library with controllable SAL/UTC/timer providers. An independent OpenSSL peer exercises TLS 1.2/1.3, RSA/ECDSA certificates, hostname/CA/date failures, missing UTC, entropy failure, short I/O and WANT retries, write retry identity, truncation versus close_notify, deadlines, callback reentry, connection limits, malformed CA bundles and allocation quotas. Each test checks final TLS allocation cleanup.

  • dmtls_pointer_audit: three regression cases prove that GOT pointers are accepted, a pointer embedded in a mixed static structure is rejected, and missing relocation metadata cannot silently disable the check.

The public test CA contains no private key. Interoperability tests generate private keys in temporary directories; production trust stores are always supplied explicitly by the caller.

STM32F746G-DISCO

Build against the existing SDK with Cortex-M7 tools and the board CPU family:

cmake -S . -B build/cortex-m7 -DDMOD_DIR=/path/to/dmod \
  -DDMOD_TOOLS_NAME=arch/armv7/cortex-m7 -DDMOD_CPU_FAMILY=stm32f7 \
  -DDMTLS_BUILD_HARDWARE_PROBE=ON
cmake --build build/cortex-m7 --parallel
python3 tests/probe/make_fixtures.py build/board-fixtures

Install local dmtls, test_dmtls and tlsprobe packages into a dmod-boot test image with hardware RNG, RTC, the rule-driven UTC service, and rtc, test_dmutc, test_dmrtc, test_dmrng diagnostics. Use the local packages manifest to select these artifacts. Flash the board, wait for its console, then upload only ca.pem and other.pem to / using the boot repository's scripts/ymodem_send.py. Keep every private key on the host.

python3 tests/probe/run_board.py --boot-dir /path/to/dmod-boot \
  --fixtures build/board-fixtures --port /dev/ttyACM0

The opt-in probe uses real dmtls, hardware RNG and rule-selected dmutc. Its nonblocking callbacks exchange bytes through memory queues. UART waits happen in the diagnostic application, outside TLS. It claims terminal foreground while running; its input handle is released at process exit. Python OpenSSL MemoryBIO provides the independent peer. This validates TLS on hardware; it does not exercise dmtcp, HTTPS or external Internet connectivity.

Hardware validation completed with seven consecutive probes: TLS 1.2 and TLS 1.3 encrypted echo/close_notify, rejected hostname, untrusted CA, future and expired certificates, and missing UTC. The harness then passed UTC 6/6, RTC 12/12 and RNG 6/6 checks, restored host UTC and confirmed dmutc@rtc running. The real module's three lifecycle checks also passed on the board. Renode was omitted as requested for this change.

Cortex-M7 artifacts measured 140388 bytes DMF and 112574 bytes DMFC (about 137.1 and 109.9 KiB). The compressed image is storage size; loading and TLS heap need additional RAM. The diagnostic uses a 512 KiB heap quota and one connection; 32 KiB is only the minimum accepted quota, not a handshake budget. Raw board logs and generated fixture keys remain local under ignored build paths, outside release resources.

Cortex-M4 was built from fresh downloads of both fork archives, verified by SHA-256. Native tests ran with the existing installed dmod_loader/todmd tools. Hardware firmware used the original DMOD loader revision, with no relocation extension. The independent process/FreeRTOS lifetime fixes remain in the boot integration. Fork changes: Mbed TLS and TF-PSA-Crypto.

Arithmetic backend configuration

The backend enables upstream MBEDTLS_HAVE_ASM, as in the upstream default configuration. It selects supported architecture-specific multiplication kernels and otherwise retains portable arithmetic. This does not change verification, algorithms, public APIs, entropy requirements or the DMOD format. ARM and x86 builds, pointer audits and native interoperability tests are repeated for it. Public-server HTTPS on STM32 completed with the optimized configuration; the longer certificate chain exposed handshake latency in the previous configuration.