Claude review: dma-fence: Use dma_fence_set_error() internally

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

Patch Review

**Status: Looks good**

The change at `drivers/dma-buf/dma-fence.c:602`:
```c
-		fence->error = -EDEADLK;
+		dma_fence_set_error(fence, -EDEADLK);
```

**Positive aspects:**
- Uses the proper API helper instead of poking the struct field directly, which is exactly what `dma_fence_set_error()` was designed for.
- The additional `WARN_ON` checks in the helper add a safety net — if someone ever manages to hit this path with an already-signaled fence, it will now produce a warning rather than silently corrupting state.
- Commit message is clear and accurately describes the change.

**Minor observations:**
- The commit message says "Only compile-tested" which is reasonable for a one-line cleanup, but the `WARN_ON(test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags))` in the helper is worth mentioning in the commit message, since it adds a new runtime check in an error path. It's a beneficial check, but reviewers should be aware it's not a pure no-op substitution.
- The pre-existing race between the unsynchronized `!dma_fence_test_signaled_flag(fence)` check (line 581) and the locked region (line 601) is not introduced by this patch and is not a concern for this review — but if the `WARN_ON` ever fires spuriously in the field, this is the path to investigate.

**Verdict:** Reviewed-by worthy. The patch improves code consistency with negligible risk.

---
Generated by Claude Code Patch Reviewer