Skip to content

STM32H743 simulator partial peripherals ​

Simulated peripherals — Cortex-M7 core (NVIC, SysTick, MPU, SCB fault registers, DWT cycle counter) · USART3 console · TIM2 and TIM4 counters and their periodic interrupt · RCC/PWR clock-and-PLL handshake · FLASH ACR · GPIOD · memory: 2 MB flash (also aliased at 0x00000000), 128 KB DTCM, 512 KB AXI SRAM, 256 KB SRAM1/2, 64 KB SRAM4. Every other peripheral reads as zero and ignores writes.

The simulator below runs real, unmodified H743 .bin images compiled for this project. Everything you see is computed by the WASM core on this page — no board, no server.

What to expect?

What the demo benchmark firmware does ​

Each button loads the project's own benchmark image — the same binaries that run on the H743 boards, byte for byte. Every image contains the RVM hypervisor, one guest RTOS running inside a virtual machine (RT-Thread, FreeRTOS, uC/OS-II, uC/OS-III, RMP, RTEMS, Zephyr, NuttX, ChibiOS, DjyOS or KLite), and the benchmark suite itself. When it starts, the hypervisor prints its own table first; the guest then prints its table through a virtual UART, which is the USART3 console on the right.

The numbers are CPU cycles, shown as average / maximum / minimum over 10 000 rounds — so they capture both the steady-state cost and the worst case caused by interrupts, preemption and cache effects. Together the two tables cover:

  • Hypervisor cost — interrupt vector entry, signal/notification between threads and between processes, invocation round-trip, and a cross-process yield. This is what virtualization adds on top of a bare-metal OS.
  • Guest OS primitives — thread yield, semaphore, mailbox, FIFO, message queue, blocking message queue, software timer/alarm, and memory allocation/free.
  • Interrupt path — the same primitives triggered from an ISR, plus the ISR-to-RVM activation relay: the cost of delivering an interrupt into a VM.

A full run takes about a minute at the default speed. The hypervisor table appears after about 10 s, the guest table completes in 30–60 s depending on the guest (the speed slider trades scheduling smoothness for throughput). CoreMark is separate: it is the standard industry benchmark, shortened to about 1 s of guest time so it finishes in about 20 s; its score matches the full 10 s run. The 10000 overflows lines that follow are the timer calibration loop counting 16-bit counter wraps — not an error.

Aquarium — all operating systems at once ​

Aquarium (all operating systems) is the multi-VM showcase: a single image that boots eleven virtual machines side by side — FreeRTOS, uC/OS-III, RT-Thread, uC/OS-II, ChibiOS, RMP, KLite, DjyOS, NuttX, Zephyr and RTEMS — on one H743 under one hypervisor. Instead of benchmark tables, every guest periodically prints a heartbeat on the console:

[FRT] alive 1
[UO3] alive 1
[RTT] alive 1
...

The number after alive grows every round, so you can watch all eleven systems time-sharing one CPU in real time.

This is also the best image for the ⚠ Inject VM fault button: the hijack lands in whichever guest happens to be running, that guest crashes and is restarted by the safety daemon — and the other ten keep running without a pause. The heartbeats prove it: after the fault, one guest reboots from its boot banner and its counter starts over, while every other guest's alive number keeps climbing uninterrupted. That is multi-machine virtualization in one picture: independent systems, one chip, a crash contained to a single VM.

Why "Inject VM fault"? ​

The button hijacks the guest's program counter to 0x00000000. That address lies outside every MPU region granted to the guest, so the instruction fetch raises a MemManage fault (instruction access violation) inside the virtual machine. RVM's safety daemon catches it, prints the faulting thread's register set and stacked frame, and then recovers the VM: the guest restarts from its reset vector and prints its banner again, while the hypervisor, the monitor and the other processes keep running.

That is the point of the button: it demonstrates fault containment and VM restart. A crashed guest cannot take the system down and does not need a board reset — the isolated VM is simply restarted. The button only unlocks once the VM is up (the benchmark table printed, or the first alive heartbeat wave for the Aquarium image), because before that there is no virtual-machine context to fault.

You can also load your own image: drop any H743 .bin into the firmware panel and it is loaded at 0x08000000 and reset.

An official work of EDI, released into the public domain.

GitHub·Discord·QQ 群 1038895132·rtospro.com