Claude review: drm/exec, drm/xe: Avoid abusing the drm_exec retry pointer

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

Patch Review

**Functionally correct but the WARN_ON condition may be wrong for the "newly initialized" case described in the doc comment.**

The new `drm_exec_retry()` macro:
```c
#define drm_exec_retry(_exec)
	do {
		WARN_ON(!drm_exec_is_contended(_exec));
		goto *__drm_exec_retry_ptr;
	} while (0)
```

The doc comment says:
> Unconditionally retry the loop to lock all objects. For consistency,
> the exec object needs to be **newly initialized or contended**.

But `WARN_ON(!drm_exec_is_contended(_exec))` will fire when the exec is newly initialized (since `contended` will be NULL after init). The WARN_ON only checks for the contended case, not the "newly initialized" case. Either the comment or the assertion needs updating. If the intent is that only the contended case is valid, the comment should be fixed. If newly-initialized is also valid (as needed for the OOM retry path in xe), then the assertion is too strict.

Looking at the xe usage: `xe_validation_should_retry()` is called on `-ENOMEM`, which could happen before any contention. If `exec->contended` is NULL at that point, this will produce a spurious WARN.

The commit message also says "This is relying on documented behaviour" — perhaps this should say "undocumented" given the context of the series (cleaning up abuses)?

---

---
Generated by Claude Code Patch Reviewer