Claude review: gpu: nova-core: convert registers to use the kernel register macro

[Claude Code Review Bot <claude-review@example.com>]

Overall Series Review

Subject: gpu: nova-core: convert registers to use the kernel register macro
Author: Alexandre Courbot <acourbot@nvidia.com>
Patches: 17
Reviewed: 2026-03-25T07:57:33.437301

---

This is a well-structured 10-patch series that migrates nova-core from its local `register!` macro to the kernel-wide `io::register!` macro. The conversion is methodical — one register group per patch — followed by cleanup (removing old macro, dropping `io::` qualifier, removing docs todo). The series is at v3 and already carries Reviewed-by tags from Eliot Courtney, Gary Guo, and Joel Fernandes on most patches.

The mechanical conversion looks correct overall. The new API style (`bar.read(REG)` / `bar.write(WithBase::of::<E>(), REG::zeroed().with_field(val))`) is more idiomatic and the `bounded_enum` macro is a nice improvement over the ad-hoc `impl_from_enum_to_u8!` + manual `TryFrom` implementations.

**Key observations:**

1. Patch 1 uses `self.inner` directly to access the raw register value in a `Display` impl, while the rest of the code uses `into_raw()`. This is inconsistent.
2. The `bounded_enum` macro's `From` variant uses `MAX.. => unreachable!()` per Gary Guo's suggestion from v3 — this is a good pattern for compile-time exhaustiveness checking.
3. The series cleanly removes 739 lines of macro infrastructure and 76 lines of TODO documentation, while the new code is more concise.

No major correctness issues found. A few minor observations below.

---
Generated by Claude Code Patch Reviewer