Roadmaps / BeagleBone / Interview topics
Reference
Interview topics
Use these for embedded Linux, platform and firmware roles. Each follow-up is the question a good interviewer asks next; the Staff line is what lifts the answer.
Phase 1
| Question | Follow-up | A Staff answer adds |
|---|---|---|
What happens from power-on to init on an AM335x board? | Where does it fail if the eMMC is corrupt? | A factory and recovery flow, not only the boot order |
| Why is worst-case latency the number that matters? | How long must you test, and under what load? | A test whose load matches production, and a stated confidence |
| When does PREEMPT_RT help, and when not? | What still breaks determinism on an RT kernel? | The cost: throughput, carried patches, and the 32-bit Arm history |
| Why is 3.3 V I/O not 5 V tolerant a review item? | How would you interface a 5 V sensor? | Failure modes: hot-plug, ESD, one pin damage that looks like a software bug |
| How do you time an event with libgpiod? | Which clock is the timestamp from? | The IRQ-to-timestamp path and its own jitter |
System design prompt (20 minutes): boot and recovery for 10,000 field devices that boot from eMMC. Outline: boot chain and slots (3 min), recovery paths and who can trigger them (5), update and rollback (5), factory provisioning and identity (4), monitoring and failure rates (3).
Phase 2
| Question | Follow-up | A Staff answer adds |
|---|---|---|
| What is a device tree, and why does Arm need one? | How does an overlay change it at boot? | Ownership: who maintains the overlay when the base tree changes |
| How does the kernel bind a driver to a node? | What if two drivers match? | Probe deferral and the order of dependencies |
Kernel driver or /dev/i2c? | What if the sensor needs interrupts? | A rule for the team, with the cost of each |
| Why can an interrupt handler not sleep? | What changes on an RT kernel? | Threaded handlers and where hard-IRQ context still exists |
| How do you find a pin conflict? | How do you stop it recurring? | A checked pin table in CI |
System design prompt (20 minutes): add a new sensor bus to a shipped product without breaking updates. Outline: requirements and pins (4), driver against user space (5), overlay and compatibility with older images (5), test and rollout (4), rollback (2).
Phase 3
| Question | Follow-up | A Staff answer adds |
|---|---|---|
| PRU, RT thread or microcontroller? | At what step rate does the answer change? | A decision memo with numbers, cost and exit plan |
| How do PRU and Linux share data safely? | How do you detect torn data? | One writer, sequence numbers, and a test that injects the race |
| What is a resource table? | What happens if the firmware has none? | How it ties firmware to kernel versions |
| Why does external memory access break exact timing? | How do you budget for it? | A cycle budget measured on the board |
| What should a controller do when its supervisor dies? | How do you test it? | A safe state documented, exercised, and reviewed by safety owners |
System design prompt (20 minutes): a motion controller that must not lose a step when the Linux side crashes. Outline: requirements and budgets (4), split between Linux and PRU (5), protocol and shared state (5), failure and safe state (4), test plan (2).
Phase 4
| Question | Follow-up | A Staff answer adds |
|---|---|---|
| How do A/B updates survive a power cut? | Where can they still fail? | The bootloader variable write, the health-check window, flash wear |
| How do you prove an update bundle was not tampered with? | What does the signature cover, and what does it not? | Signing, versioning and who holds the keys; the 2026 RAUC plain-bundle CVE as a worked example |
| Secure boot on this board? | What do you do instead? | An honest threat model and a hardware upgrade path |
| Yocto or Buildroot? | When would you switch? | Cost of the switch, team skills, support window |
| How do you plan for a chip going end-of-life? | When do you start? | Time and cost of the port, and a test that tells you it worked |
Behavioural prompt (Staff and Principal): describe a time you moved a deadline-critical path from software timing to hardware, what you measured, who signed the decision and what you checked afterwards.
System design prompt (20 minutes): ship firmware updates to 1,000 devices with no remote hands. Outline: update format and signing (4), A/B and rollback (5), staged rollout and health signals (5), recovery for devices that do not come back (3), cost and supply (3).