Contents
SpecificationArm

Arm ARM (A-profile)

Arm Architecture Reference Manual for A-profile architecture

The Arm Architecture Reference Manual for A-profile is the rulebook for how Cortex-A application processors execute instructions, take exceptions, and separate privileged and secure software.

What it is

The Arm Architecture Reference Manual for A-profile (the Arm ARM) is the specification of the instruction set and system architecture used by Arm Cortex-A application processors. In a vehicle those processors sit in high-performance computers and similar ECUs, where they run boot firmware, hypervisors, and rich operating systems.

The manual states the behavior software can rely on: which instructions exist, what the system registers do, how exceptions are taken, how addresses are translated, and how execution is split across privilege levels and security states. A silicon vendor implements one version of this architecture in a core. Pipeline structure, cache sizes, and the on-chip interconnect are not in this book. Those facts belong to the Technical Reference Manual for a specific core and to the SoC documentation.

A-profile is separate from Arm M-profile, used by many microcontrollers, and from R-profile, used by real-time cores. Behavior that exists only in those profiles is outside this manual.

Who it applies to

Read it if you write or review software that runs with privilege on the application processor: boot ROM and bootloader code, the secure monitor, a hypervisor, or the operating-system kernel and its board support. E/E architects who write the HPC boot, partitioning, and virtualization concept need the same vocabulary, so requirements name real architectural mechanisms.

Cybersecurity and functional-safety engineers use it to check that isolation claims match processor behavior. A secure world, a guest virtual machine, or a split between privileged and unprivileged software means what this manual says it means on that core. People writing ordinary application code above the operating system rarely need the manual itself.

What it covers

The parts used most often in automotive HPC work are:

Topic What the manual defines Where it shows up
Exception levels Privilege levels EL0 to EL3, and how exceptions are entered and returned from Applications, kernel, hypervisor, secure monitor
Security states Separate execution worlds and the controls used to cross between them Secure boot and hardware-isolated security services
Memory system Translation, memory types, caches, and ordering rules Page tables, device registers, DMA
Instruction set Instructions used to compute and to configure the processor Boot, context switch, barriers, cache maintenance
Optional extensions Features a core may include, such as extra virtualization or hardening support Only when that core’s ID registers report them

Each A-profile core implements a base architecture version. Optional extensions are present only when that core includes them. The interrupt controller, reset and power sequence, bus firewalls, and the peripheral memory map are specified in other documents.

How it fits with other documents

This manual is the architecture layer. The Technical Reference Manual for a named core records how that core implements the architecture, including implementation-defined registers and which optional features were built. The SoC manual records the rest of the chip: boot configuration, the memory map, firewalls, and accelerators.

Firmware and operating-system stacks are written to the architecture and then adapted to one SoC. They explain how a project uses the processor; they are not the specification. Automotive process standards for safety, cybersecurity, and software update set engineering obligations and evidence. They leave exception levels and security states to this manual. When a safety or security argument depends on hardware isolation inside the application processor, the mechanism should be traceable here and to the features the chosen core actually implements.

In practice

Identify the architecture version of the application processor on the SoC before reading. Use the current edition of the manual, then confirm implemented features in the core Technical Reference Manual. Firmware work usually stays in a few places: the state entered at reset, how each boot stage drops privilege, how the memory management unit is programmed, and how the security state is selected. Kernel and hypervisor work adds exception entry and return, separate address spaces, and the ordering rules drivers need for peripherals and DMA.

Record the issue of the manual you designed against. Later issues describe architecture features that older cores do not implement. Keep the manual beside you for boot, privilege changes, and isolation decisions. Day-to-day application logic stays on the operating-system interface.

What it means for your discipline

  • E/E architectureCore

    E/E architects use it when they specify how an HPC application processor boots, how software is split across privilege levels, and which isolation features the SoC must implement. Requirements should name exception levels, security states, and the memory system, not only vendor feature names.

  • CybersecurityCore

    Secure boot, hardware-backed isolation, and debug control on the application processor rest on security states, the highest exception level, and system registers defined here. A cybersecurity concept should cite those mechanisms and confirm the chosen core implements them.

  • HPC & application softwareCore

    Bootloaders, secure monitors, hypervisors, and kernel ports are written against the instruction set, exception model, and memory model in this manual. Any code that runs before the operating system is up, or that changes privilege, translation, or cache state, has to follow it.

  • Hardware & siliconPartial

    Hardware engineers select the SoC rather than implement the instruction set, and they need the A-profile version and optional features that core actually includes. That choice limits what firmware may rely on and how boot pins, firewalls, and the memory map must line up.

  • Base softwarePartial

    Board support, hypervisor configuration, and the lowest drivers on an HPC must use the exception levels and memory types the architecture defines for peripherals and DMA. Classic microcontroller stacks follow a different Arm profile, so A-profile rules do not carry over unchanged.

  • Functional safetyPartial

    A safety case that places functions of different integrity on one application processor has to show how interference is prevented. Use the manual to see whether exception levels, address spaces, or virtualization actually provide that separation, and where the guarantee stops.

  • Software update & OTAPartial

    The update agent depends on a boot path whose privilege and security state are defined by this architecture. The manual does not describe the campaign format; it specifies the processor state the bootloader is in when it chooses which image to start.

No direct stake: Networks & communication · Diagnostics & lifecycle