What runs here
This cell contains the guest operating system inside each virtual machine (VM) on the HPC’s application processor. Different VMs can use different operating systems; the label does not mean that every HPC needs all three.
| Component | Role |
|---|---|
| Kernel and scheduler | Run threads, handle interrupts and manage CPU time within the guest. |
| Memory management | Give processes separate address spaces and manage the guest’s allocated memory. |
| System libraries and IPC | Provide application interfaces and communication between processes. |
| Filesystems and networking | Support persistent data, sockets and network protocols. |
| Essential system processes | Start and supervise the guest’s basic operating environment. |
Linux supplies a general-purpose kernel with a selected userspace. QNX uses a microkernel architecture with operating-system services implemented in separate processes. Android is a Linux-based platform that also includes a runtime, framework and system services; its complete stack spans several rows of this grid. Android applications generally use Android APIs, so Android should not be assumed interchangeable with a POSIX environment required by another runtime. Android architecture
Why it looks like this
Cortex-A class processors provide memory management and hardware virtualisation suited to multiple protected application environments. HMI, feature services and ADAS software need substantial libraries, networking and access to GPU/NPU resources. A rich OS supports these workloads, but brings memory, storage, startup and maintenance costs.
Selection depends on workload and evidence. Linux offers a broad software ecosystem; Android supplies an application and HMI framework; a suitable QNX safety product can provide evidence for a safety-related integration. Assess licensing, BSP support, driver availability, qualification work and long-term patching together.
POSIX compatibility, real-time behaviour and functional safety are separate properties. Linux PREEMPT_RT reduces sources of scheduling latency, but meeting a deadline still requires analysis of the configured guest, hypervisor, drivers and shared hardware. Linux real-time documentation
The grid’s QM–ASIL B range describes intended deployment context, not an automatic OS rating. Safety claims depend on the exact product, configuration, assumptions and system safety argument. Safety products must be integrated according to their applicable safety manuals. QNX safety integration guidance
Separate VMs can support different update cadences, but shared devices and interfaces still create integration dependencies.
How it connects
Above, applications and middleware use system calls, libraries, threads, clocks and IPC. AUTOSAR Adaptive defines an Operating System Interface based on the POSIX PSE51 profile; its functional clusters sit above the OS. Android instead exposes its own framework and native interfaces. AUTOSAR Adaptive architecture
Below, the guest uses its BSP and drivers to access virtual devices or assigned physical devices. VirtIO can provide virtual-device interfaces where supported. The hypervisor controls VM resource allocation; the guest scheduler controls threads within its available virtual CPUs. Guest process protection and inter-VM isolation are distinct responsibilities.
Other VMs communicate through configured virtual networks or explicit shared-memory channels. Other vehicle nodes connect through network interfaces and middleware such as SOME/IP or DDS. Define ownership, validation, timeouts and restart behaviour at every boundary. Safety-relevant communication needs appropriate end-to-end protection; transport delivery alone is insufficient.
Compared with other node classes
- HPC safety island: an AUTOSAR OS or safety RTOS supports focused monitoring and control duties, with a smaller execution environment.
- Zone controller: a multicore AUTOSAR OS typically schedules configured tasks and interrupts close to local I/O and vehicle buses.
- MCU-based ECU: AUTOSAR OS or another RTOS usually serves a narrower function with tighter memory and timing budgets.
- Smart node / LIN slave: bare metal or a tiny scheduler handles limited behaviour without the rich process environment used here.
Best practices
- Select each guest OS from workload, timing and assurance requirements; document the supported configuration and lifecycle.
- Budget virtual CPU time, memory, storage and I/O together; assess contention on shared caches, memory and accelerators.
- Verify deadlines under representative peak load, including competing VMs, interrupts, thermal constraints and recovery activity.
- Minimise privileged services and device access; apply least privilege and maintain a tested security-patch process.
- Version the guest image, BSP, virtual-device interfaces and middleware together; test update compatibility and recovery.
- Define supervision across process, guest and hypervisor boundaries, including which failures require a VM restart or a degraded vehicle function.
Common pitfalls
- Treating a real-time scheduler, POSIX interface or certified component as sufficient evidence for the complete function.
- Assuming separate VMs eliminate interference through shared hardware or device backends.
- Porting Linux software to Android without checking libraries, permissions, lifecycle and interface compatibility.
- Restarting a guest without handling stale messages, unavailable services and actuator ownership.