Claude review: accel/amdxdna: Handle DETACH_DEBUG_BO through config_debug_bo path

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

Patch Review

**Correctness: Good.** The root cause is correctly identified. Looking at `aie2_ctx.c:858-884`, when `attach` is false, `cmd.opcode` is set to `DETACH_DEBUG_BO`, which then gets submitted to the scheduler. But in `aie2_sched_job_run()` at line 358, the switch only matched `ATTACH_DEBUG_BO`, so `DETACH_DEBUG_BO` fell through to `default: ret = -EINVAL`. The handler `aie2_config_debug_bo()` in `aie2_message.c:1179` already correctly checks the opcode to decide between `DEBUG_BO_REGISTER` and `DEBUG_BO_UNREGISTER`, so routing `DETACH_DEBUG_BO` there is the right fix.

**Style nit:** The explicit `fallthrough;` annotation is unnecessary here. In kernel style, `fallthrough` is used when there is code in the preceding case before falling into the next one. When two cases simply share the same body with no intervening code, stacking the `case` labels is the standard idiom:

```c
case ATTACH_DEBUG_BO:
case DETACH_DEBUG_BO:
    ret = aie2_config_debug_bo(hwctx, job, aie2_sched_drvcmd_resp_handler);
    break;
```

The `fallthrough` keyword is mandated by `-Wimplicit-fallthrough` only when there are statements between the case labels. With no statements, there is no implicit fallthrough warning, and the bare stacked-case form is both idiomatic and more concise. The existing `SYNC_DEBUG_BO` case just above demonstrates this — it doesn't use `fallthrough` before its own handler. This is a minor style point and shouldn't block the patch, but it would be cleaner without it.

**Overall:** The fix is correct and important — without it, debug BO detach would always fail silently (well, with `-EINVAL`). Recommend merging with the optional style cleanup.

---
Generated by Claude Code Patch Reviewer