Add Infineon PSOC Control C3 on-board TPM demo

[dgarske]

Description

Runs the whole demonstration on the microcontroller: post-quantum signing, a sealed secret bound to a measurement, PCR and endorsement certificate reads, an SPDM session, and the signature authenticating a post-quantum TLS 1.3 handshake. No host takes part in the cryptography; it supplies a screen.

What it adds

• Infineon/PSoC_Control_C3_Onboard_Demo/ - one firmware image whose acts are selected by a command byte on the console, emitting newline-delimited JSON
• tpm_io_i2c_c3.c, rng_seed.c - TPM IO over wolfBoot's I2C driver, and a DRBG seeded from the TPM
• act_tls_sign.c, host/tls/board_tls.c - a Hash-ML-DSA key derived in the TPM, and a TLS 1.3 peer whose signing is delegated to the board through a wolfCrypt crypto callback
• host/ - browser front end, a command-line driver for the same acts, and test_host.py
• wolfboot-demo.config, Makefile, patches/, BUILD.md - partitions for this image, a build that works against either wolfTPM SPDM layout, the wolfSSL patch the TLS tab needs, and the build commands

wolfBoot signs and verifies the application; make factory produces one flashable image. wolfTPM is unmodified. Needs wolfSSL/wolfBoot#915

Testing

Validated end to end on a PSOC Control C3 (C3M6) EVK with an Infineon SLB9678 on mikroBUS, firmware 45.91, and rebuilt from nothing on a clean Windows 11 machine.

The TLS tab needs the patch in patches/, which adds an RFC 8446 private-use signature algorithm for the pre-hash ML-DSA this firmware signs, so it is not interoperable with an unpatched peer. BUILD.md pins the wolfSSL revision: current master enforces FIPS 204 section 5.4 and rejects the SHA-256 pre-hash, which is the only one this part will sign with. patches/ carries the opt-in that lifts it, proposed as wolfSSL/wolfssl#11736. The endorsement tab verifies the key match for the P-256 certificate only.

[Copilot]

Copilot review overview

🟡 Changes recommended

Serial timeout and baud defects, unsafe certificate rendering, and incorrect SPDM success reporting can break or misrepresent core demonstrations.

Review effort: Balanced
Findings: 3 · 5 · 6

Open (14)

• Factory target does not track signing key generation · New
• Dispatcher invokes TPM operations after initialization failure · New
• Certificate SAN values are vulnerable to HTML injection · New
• Discovery-only results are falsely reported as secured sessions · New
• SPDM timeout is too short for the full handshake · New
• OnboardReader defaults to the wrong baud rate · New
• CLI default baud rate mismatches the firmware · New
• Benchmark reports PASS after incomplete iterations · New
• SPDM build hash is not reproducible across responder keys · New
• Command table omits the SPDM d command · New
• Documentation incorrectly describes two off-silicon acts · New
• Main claim incorrectly says every tab uses fitted silicon · New
• Sign act is mislabeled as host verification · New
• Transport is incorrectly documented as SPI · New

What changed in this PR

Adds a complete PSoC Control C3 firmware and host interface for driving an attached TPM over I2C, including ML-DSA, sealing, PCR, endorsement, and SPDM demonstrations.

Changes:

• Adds MCU firmware, TPM transport, cryptographic acts, diagnostics, and wolfBoot packaging.
• Adds a browser UI and serial command-line tooling.
• Documents building, flashing, operation, and measured performance.

File	Description
README.md	Links and summarizes the demo.
.../​.gitignore	Excludes generated artifacts.
.../​BUILD.md	Documents build and flashing.
.../​Makefile	Builds, signs, and assembles firmware.
.../​README.md	Documents firmware acts and benchmarks.
.../​act_sign.c	Implements ML-DSA signing and verification.
.../​act_sign.h	Declares the signing act.
.../​act_spdm.c	Implements the SPDM act.
.../​act_spdm.h	Declares the SPDM act.
.../​console_write.c	Routes diagnostic output to UART.
.../​demo_acts.c	Dispatches console-selected TPM acts.
.../​demo_util.c	Provides board, timing, and output utilities.
.../​demo_util.h	Declares shared utilities.
.../​fault_report.c	Reports MCU faults over UART.
.../​firmware/​reset-board.jlink	Resets the MCU and TPM.
.../​heap.c	Provides a fixed embedded heap.
.../​i2c_trace.c	Records TPM I2C transfers.
.../​i2c_trace.h	Declares optional I2C tracing.
.../​mldsa_bench.c	Benchmarks ML-DSA verification.
.../​mldsa_onboard.c	Runs the standalone signing demonstration.
.../​rng_seed.c	Seeds wolfCrypt from the TPM RNG.
.../​tpm_identity.c	Reads TPM identity on-device.
.../​tpm_io_i2c_c3.c	Implements wolfTPM I2C transport.
.../​user_settings.h	Configures ML-DSA verification.
.../​user_settings_tpm.h	Configures wolfTPM and SPDM support.
.../​wolfboot-demo.config	Defines the enlarged wolfBoot layout.
.../​host/​README.md	Documents host operation.
.../​host/​capture-tabs.sh	Automates UI screenshots.
.../​host/​onboard.py	Drives firmware over serial.
.../​host/​requirements.txt	Lists host dependencies.
.../​host/​shot.py	Captures headless browser screenshots.
.../​host/​sse_probe.py	Probes streamed demo events.
.../​host/​templates/​index.html	Implements the browser interface.
.../​host/​x509_lite.py	Parses displayed certificate fields.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🟡 Changes recommended

Unresolved input-framing, timeout, recovery-ordering, heap-boundary, build-reproducibility, and network-exposure issues affect reliability and security.

8 open findings

• Validate lower bound in _sbrk negative increments · New
• Bind unauthenticated control server to loopback · New
• Restore device before emitting terminal result · New
• Drain UART input after malformed hex requests · New
• Honor WOLFSSL_REV for existing wolfSSL trees · New
• Enforce process timeout while streaming stdout · New
• Restrict process termination to this demo · New
• Correct outdated wolfBoot PR support description · New

14 resolved since last review

• Certificate SAN values are vulnerable to HTML injection
• Dispatcher invokes TPM operations after initialization failure
• Factory target does not track signing key generation
• Benchmark reports PASS after incomplete iterations
• CLI default baud rate mismatches the firmware
• OnboardReader defaults to the wrong baud rate
• SPDM timeout is too short for the full handshake
• Discovery-only results are falsely reported as secured sessions
• Transport is incorrectly documented as SPI
• Sign act is mislabeled as host verification
• Main claim incorrectly says every tab uses fitted silicon
• Documentation incorrectly describes two off-silicon acts
• Command table omits the SPDM d command
• SPDM build hash is not reproducible across responder keys

🧠 Review effort: Balanced