AUTOSAR shwa

shwa — Safe Hardware Accelerator

← back to AUTOSAR

What ARA defines

ara::shwa is the platform's abstraction for safe access to hardware accelerators (GPU, DLA, dedicated compute) — exposing their state and managing their safe use on a high-performance compute ECU.

Our adaptation

We repurposed shwa as the hardware-telemetry + power plane for the compute node (NVIDIA Orin) and the central host — the part of "hardware accelerator awareness" that's actually actionable for us.

• Polls a build-selected backend on a tick and broadcasts a rich AccelTelemetry sample — CPU / GPU / memory / frequency / temperature / power / fan — the way jetson-stats (jtop) reports.
• Owns the platform power mode: the Orin nvpmodel profile + jetson_clocks. (This deliberately moved out of osi — osi is the POSIX/cgroup portability plane, not a Jetson knob.)
• Backend selected by bazel select (host vs jetson), not #ifdef. Read-only telemetry + the power knob; never a data path.

This is the home for all Jetson/accelerator knowledge. We lean toward "observe + one safe knob" rather than the spec's broader accelerator-management surface, because that's what the rigs need today.