Enable Serial Wire Output (SWO) for debug tracing #41

Summary

SWO (Serial Wire Output) provides a low-cost, single-pin trace channel for debug output via the ITM/TPIU. The hardware infrastructure exists in the codebase but is incomplete — the core trace peripherals are defined and the cortex::swo::emit() API works, but no initialization code connects them to the physical pin.

This issue covers enabling SWO on both the STM32H7 and STM32F4 families used in this project.

Current State

The following already exists (family-independent):

• modules/cortex/include/cortex/swo.hpp — Public API (cortex::swo::emit() for ports 0–1)
• modules/cortex/source/swo.cpp — Implementation (blocks on FIFO ready, writes ITM stimulus port)
• modules/cortex/include/cortex/peripherals/InstructionTraceMacrocell.hpp — Full register maps for ITM, DWT, and TPIU, with extern instances linked in peripherals.cpp
• modules/cortex/include/cortex/peripherals/DebugSystem.hpp — DEMCR register definition with enable_trace bit (TRCENA)

What's Missing

1. Family-specific DBGMCU register headers

No STM32 DBGMCU register header exists for either family. The key registers and fields differ:

STM32H7 (DBGMCU_CR at 0x5C001004):

Currently reads 0x00000007 — only debug-sleep/stop/standby bits are set.

STM32F4 (DBGMCU_CR at 0xE0042004):

Both are defined in their respective SVD files under modules/stm32/scripts/.

2. SWO pin configuration

The SWO pin is the same on both families:

• Pin: PB3, Alternate Function AF0 (TRACESWO)
• Mode: Alternate Function
• Output Type: Push-Pull
• Speed: Very High
• Pull: None

On the Nucleo-H753ZI, PB3 is also used as SPI1_SCL on the Arduino header (solder-bridge SB63/SB64 isolates it). On some STM32F4 packages PB3 may also serve as JTDO — board-specific handling may be needed.

3. System-level trace initialization sequence

A new cortex::initialize::trace() function that performs:

1. Core: Set DEMCR.TRCENA via DebugSystem::ExceptionMonitorControl.enable_trace
2. Family-specific DBGMCU:
   • H7: DBGMCU_CR.TRACECLKEN = 1, D1DBGCKEN = 1
   • F4: DBGMCU_CR.TRACE_IOEN = 1, TRACE_MODE = 0b01 (async NRZ)
3. GPIO: Configure PB3 as AF0 push-pull very high speed
4. TPIU:
   • TPIU_ACPR.SWOSCALER = (trace_ref_clock / desired_baud) - 1
   • TPIU_SPPR.Protocol = AsyncNRZ (preferred over Manchester)
5. DWT:
   • DWT_CTRL.CYCCNTENA = 1 (cycle counter)
   • DWT_CTRL.PCSAMPLENA = 1 (periodic PC sampling)
6. ITM:
   • ITM_TCR.ITMENA = 1, SWOENA = 1, DWTENA = 1, TSENA = 1
   • ITM_TER = enable stimulus ports 0 and 1
   • Unlock via ITM_LAR = 0xC5ACCE55

4. Clock frequency plumbing

The TPIU SWO baud prescaler formula is the same for both:

SWOSCALER = (trace_ref_clock / desired_baud) - 1

The clock tree must be initialized before trace init, since the prescaler depends on the measured frequency.

Measured Live Register Values (from J-Link on H753)

The ITM and DWT are partially initialized (likely by debugger or reset defaults), but the system-level path to the pin is gated off.

Acceptance Criteria

• Common cortex::initialize::trace() function handles both H7 and F4
• Family-specific DBGMCU register headers under modules/stm32/include/stm32/
• SWO stimulus port 0 emits characters visible via J-Link SWO viewer on both platforms
• SWO baud rate is configurable (default 2 MHz)
• DWT cycle counter and PC sampling packets appear on SWO
• Works after cortex::system::main() in normal startup
• No interference with existing PB3 functionality when SWO is not enabled