Claude review: Documentation: drm: Add entry for removing spsc_queue to TODO list

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

Patch Review

**Factual error — `ACCESS_ONCE()` reference is wrong:**

The patch states:

```
+This queue utilizes:
+
+- preempt_disable()
+- atomic instructions
+- memory barriers
+- ACCESS_ONCE()
```

`ACCESS_ONCE()` was removed from the kernel years ago and replaced with `READ_ONCE()`/`WRITE_ONCE()`. The current `spsc_queue.h` uses `READ_ONCE()` and `WRITE_ONCE()` exclusively. This should be corrected to `READ_ONCE() / WRITE_ONCE()`.

**Minor: "has to take a lock in some situations" claim is misleading:**

The patch says:

```
+Moreover, drm_sched_entity_push_job(), the only user of spsc_queue_push(), has
+to take a lock in some situations anyways and calls to it are often serialized
+with a driver lock.
```

Looking at the current code, `drm_sched_entity_push_job()` itself does not take any lock internally. Rather, its documentation states that callers should hold a "common lock" externally to guarantee ordering with `drm_sched_job_arm()`. This is an important distinction — the function relies on external serialization, but doesn't itself take any lock "in some situations." The wording could be made more precise to say something like "calls to it are typically externally serialized by a driver lock" to better reflect reality.

**Level classification:**

The task is marked as "Beginner" level. This seems reasonable for the replacement itself, though verifying correctness with benchmarks (especially on amdgpu) may push it toward Intermediate. The existing TODO entry immediately above it (about locking in `drm_sched_rq`) is marked Intermediate for similar work, so there's a minor inconsistency, but this is a judgment call.

**Structural nit — contact formatting:**

The existing TODO entries in the file use a single-line `Contact:` format like:

```
Contact: Philipp Stanner <phasta@kernel.org>
```

This patch uses a multi-line list format:

```
+Contact:
+
+- Philipp Stanner <phasta@kernel.org>
+- Christian König <christian.koenig@amd.com>
```

The list format is actually fine since there are two contacts, but worth noting the slight style difference. No change needed.

**Summary:** The patch is a reasonable TODO addition. The main actionable item is correcting the `ACCESS_ONCE()` reference to `READ_ONCE() / WRITE_ONCE()`, and optionally clarifying the locking claim about `drm_sched_entity_push_job()`.

---
Generated by Claude Code Patch Reviewer