torvalds_skill

Overview

This skill teaches writing systems and kernel-level code in the pragmatic, performance-first style associated with Linus Torvalds. It emphasizes getting data structures right, optimizing the common case, and enforcing crisp code review and commit hygiene. Use it when you need reliable, maintainable, high-performance system software.

How this skill works

The skill inspects design choices and code patterns, prioritizing data layout, cache behavior, and clear control flow. It recommends concrete patterns: array-first data structures, branch-prediction hints, single-exit cleanup with goto, and explicit reference-counting. It also evaluates commit messages and patch size to ensure easy review and traceability.

When to use it

• Implementing kernel modules, drivers, or low-level OS services
• Designing performance-sensitive data structures and memory layouts
• Refactoring subsystems for clarity and auditability
• Preparing small, reviewable patches and clean commit history
• Writing code that must be safe on real hardware and under contention

Best practices

• Design data structures before writing algorithms; let the layout drive code
• Optimize the common case; use likely()/unlikely() and measure first
• Prefer arrays and contiguous storage for cache locality over linked lists
• Keep patches small, write clear commit messages, and stage hunks with git add -p
• Use single-exit cleanup patterns (goto) for easy reasoning and audit
• Avoid abstractions that hide performance costs; prefer explicit, simple code

Example use cases

• Implement a high-throughput network driver with cache-friendly packet queues
• Refactor a subsystem to expose a minimal public API and well-namespaced internals
• Add reference counting and safe release paths to a shared kernel object
• Optimize a hot path by rearranging data layout and adding branch hints
• Prepare a sequence of small, testable commits for upstream review