Skip to content

Renode STM32F7 testing

Renode (Antmicro, MIT-licensed) is the open-source emulator that models full STM32F7 silicon — IWDG, SAI, DMA, FMC/SDRAM, LTDC, NVIC, and friends — at a fidelity where the unmodified stm32f7xx_hal.h code runs. oveRTOS uses Renode 1.16.1 with the platforms/boards/stm32f7_discovery-bb.repl platform to gate every PR on full-board execution of the CMocka suite.

Why Renode

The QEMU MPS2-AN500 target exercises the CPU core, NVIC, SysTick, scheduler, and the kernel-level parts of oveRTOS — but not any STM32-specific peripheral code. backends/freertos/freertos_watchdog.c writes to IWDG_HandleTypeDef registers QEMU doesn't model; stm32f7xx_hal.c is compiled-in-CI but never executed. Renode closes that gap on the fastest feedback loop CI can provide — the same firmware.elf that goes on a real Discovery board boots under Renode. The complementary HW layer (manual-only, separate page) closes the gap to actual silicon.

Test matrix

Six targets total — one per RTOS backend, two heap modes each:

Target Tests Notes
test-renode-stm32f746-freertos 219 FreeRTOS heap-mode; full HAL + lwIP linked
test-renode-stm32f746-freertos-zeroheap 192 FreeRTOS static-only allocation
test-renode-stm32f746-zephyr 211 Zephyr stm32f746g_disco, heap pool 128 KiB
test-renode-stm32f746-zephyr-zeroheap 184 Zephyr HW_STACK_PROTECTION=n (workaround for Renode's MPU drift)
test-renode-stm32f746-nuttx 212 NuttX stm32f746g-disco:nsh
test-renode-stm32f746-nuttx-zeroheap 185 Same with CONFIG_OVE_ZERO_HEAP

Group invocation:

make test-renode                                  # all six
make test-renode-stm32f746-freertos               # one
make test-renode-stm32f746-zephyr-zeroheap        # one

ove test renode is the equivalent inside the Python CLI.

Architecture

flowchart LR
    subgraph Build["Build (per RTOS variant)"]
        SRC["tests/sim/renode-stm32f746-&lt;rtos&gt;{-zeroheap}/<br/>(CMakeLists.txt or nuttx_app/Makefile)"]
        ELF["firmware.elf"]
        SRC --> ELF
    end

    subgraph Run["Run"]
        RESC["test.resc"]
        RENODE["renode --console --plain"]
        RESC --> RENODE
    end

    subgraph Capture
        UART["uart.log<br/>(SemihostingUart on FreeRTOS;<br/>USART1 on Zephyr / NuttX)"]
        PARSE["_parse_cmocka()"]
        TABLE["pass/fail/skip table"]
        UART --> PARSE --> TABLE
    end

    ELF --> RENODE --> UART

    style ELF fill:#5c6bc0,stroke:#3949ab,color:#fff
    style RENODE fill:#1e88e5,stroke:#1565c0,color:#fff
    style UART fill:#43a047,stroke:#2e7d32,color:#fff
    style TABLE fill:#f57c00,stroke:#e65100,color:#fff

The runner (config/ove-cli/ove/test.py::_renode_run_elf) is a thin shim around renode -e ... -e include @<resc> -e quit. Each RTOS has its own build helper (_build_renode_stm32f746, _build_renode_stm32f746_zephyr, _build_renode_stm32f746_nuttx) that the HW runner reuses to avoid duplicating build logic.

Per-RTOS notes

FreeRTOS

Three sim trees: tests/sim/renode-stm32f746-freertos/ (heap) and tests/sim/renode-stm32f746-freertos-zeroheap/. Both link the real STM32F7 HAL, FreeRTOS ARM_CM7/r0p1 port, and lwIP. Renode 1.16's semihosting handler doesn't implement SYS_WRITE (0x05); the trees ship a semihosting_io.c that overrides newlib's _write to emit one byte at a time via SYS_WRITEC (0x03). Test output is captured through a UART.SemihostingUart attached by test.resc.

The FreeRTOS test target also exercises real HAL-backed drivers against Renode's modelled peripherals:

  • STM32F7_I2C answers FT5336 register reads at i2c3:0x38.
  • STM32F7_USART accepts USART2 transmits.
  • STM32SPI + SPI.SPILoopback @ spi1 echoes SPI transfers.
  • Network.SynopsysEthernetMAC + EthernetPhysicalLayer accepts the full lwIP TCP/IP path; the suite verifies static-IP bring-up and UDP loopback.

Zephyr

tests/sim/renode-stm32f746-{zephyr,zephyr-zeroheap}/. Built via west build -b stm32f746g_disco. printk goes through Zephyr's UART_CONSOLE on USART1; test.resc taps that into a file backend. Zero-heap variant disables CONFIG_HW_STACK_PROTECTION because Zephyr's MPU programming for static thread stacks trips a USAGE_FAULT ("Illegal use of EPSR") on the first context switch back from k_sleep under Renode 1.16's STM32F7 + ARM core combo.

main() ends in for(;;) rather than semihosting_exit() — Renode 1.16 doesn't catch SYS_EXIT_EXTENDED, the bkpt fall-through trips Zephyr's k_thread_abort cleanup into a fault loop, and the resulting wall:sim ratio collapse blows past the 300 s harness ceiling.

NuttX

tests/sim/renode-stm32f746-{nuttx,nuttx-zeroheap}/. Built via NuttX's configure.sh stm32f746g-disco:nsh followed by make olddefconfig && make. USART1 is the console. The defconfig overlay bumps DEFAULT_TASK_STACKSIZE, PTHREAD_STACK_DEFAULT, POSIX_SPAWN_DEFAULT_STACKSIZE, and IRQ_WORK_STACKSIZE to 4 KiB (the upstream stm32f746g-disco:nsh ships 2 KiB defaults that overflow under cmocka). It also sets CONFIG_SCHED_WORKQUEUE=y + CONFIG_SCHED_HPWORK=y so nuttx_timer.c's SIGEV_THREAD path compiles.

main() returns rather than calling semihosting_exit() — same rationale as Zephyr but additionally because the NuttX panic path on Renode 1.16's STM32F7 halts the CPU at PC=0, which freezes Renode's virtual clock so RunFor blocks forever.

What Renode catches that QEMU doesn't

Concrete bugs Renode caught (or would catch) that QEMU misses:

  • HAL struct-layout regressions — IWDG_HandleTypeDef, SD_HandleTypeDef, ETH_HandleTypeDef etc. depend on CubeF7 release. QEMU never executes them, so a struct-size mismatch lands silently; Renode invokes them and trips canaries.
  • Driver descriptor wiring — freertos_eth.c::DMARxDscrTab / DMATxDscrTab initialisation, ring-buffer pointer arithmetic.
  • Peripheral register-write bugs — wrong bit positions, missing enable bits, incorrect prescaler computation.
  • lwIP integration bugs — netif/socket lifecycles, ARP cache population (when paired with Renode's Network.NetworkServer).

What Renode doesn't catch — see the HW layer

  • Real silicon errata (cache coherency, DMA-vs-CPU memory ordering).
  • Real PHY autonegotiation timing (Renode's EthernetPhysicalLayer always reports link up).
  • Real wall-clock from SysTick @ HCLK/1 kHz (Renode's virtual time is fake — PLL miscompute is invisible).
  • DWT cycle counter advancing (Renode 1.16 leaves DWT->CYCCNT stuck at 0; oveRTOS's Renode-FreeRTOS firmware swaps in stub_time.c for this reason).
  • IWDG actually resetting the MCU.
  • SD card protocol — Renode's STM32FSDMMC rejects several registers BSP_SD_Init writes; FS/NVS coverage on FatFS is deferred (see the comment block at the bottom of tests/suites/test_renode_stm32_periph.c).

Renode is downloaded automatically

Renode is tracked in manifest.yaml under tools.renode:

tools:
  renode:
    version: "1.16.1"
    url:     "https://github.com/renode/renode/releases/download/v1.16.1/renode-1.16.1.linux-portable.tar.gz"

make download (or running any test-renode-… target the first time) pulls the portable build into output/tools/renode/. ove doctor reports its presence.