Claude review: drm/ttm/tests: add a test to exercise ttm_bo_swapout

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

Patch Review

**Positive aspects:**
- Well-written commit message with clear motivation and historical context.
- The test follows the established patterns in `ttm_bo_validate_test.c` (same allocation style, same cleanup pattern with `ttm_bo_fini`/`ttm_mock_manager_fini`).
- Good validation: checks the return value, the resulting mem_type, and the swapped flag.

**Minor observations:**

1. **`KUNIT_EXPECT_EQ` vs `KUNIT_ASSERT_EQ` for `ttm_bo_init_reserved` failure** (line 791):
   ```c
   KUNIT_EXPECT_EQ(test, err, 0);
   ```
   If `ttm_bo_init_reserved` fails, the test continues and will likely crash on the subsequent `dma_resv_unlock` and `ttm_bo_swapout` calls with an uninitialized BO. Using `KUNIT_ASSERT_EQ` would abort the test on failure, preventing cascading issues. That said, the existing tests in this file all use `KUNIT_EXPECT_EQ` for this same check (e.g., lines 129, 178, 220, 252), so this is consistent with the existing code style. It could be argued the existing tests all have this same weakness, but that's a separate cleanup.

2. **Potential NULL dereference in assertions after swapout** (lines 796-797):
   ```c
   KUNIT_EXPECT_EQ(test, bo->resource->mem_type, TTM_PL_SYSTEM);
   KUNIT_EXPECT_TRUE(test, bo->ttm->page_flags & TTM_TT_FLAG_SWAPPED);
   ```
   If `ttm_bo_swapout` fails or doesn't behave as expected, `bo->resource` or `bo->ttm` could potentially be NULL. A defensive check or using `KUNIT_ASSERT_NOT_NULL` before dereferencing would be safer. However, in practice, a failed swapout would likely leave these pointers valid (just not in the expected state), so this is a minor robustness concern rather than a real bug.

3. **Checking swapped return value** (line 794-795):
   ```c
   swapped = ttm_bo_swapout(priv->ttm_dev, &ctx_init, man, GFP_KERNEL, 1);
   KUNIT_EXPECT_EQ(test, swapped, MANAGER_SIZE / PAGE_SIZE);
   ```
   The `MANAGER_SIZE / PAGE_SIZE` expectation is consistent with having initialized the GEM object with size `MANAGER_SIZE` at line 787. This looks correct.

4. **Cleanup ordering**: The test calls `ttm_bo_fini(bo)` before `ttm_mock_manager_fini(priv->ttm_dev, mem_type)`, which is the correct order (destroy BO before tearing down the manager). This matches the pattern in other tests like `ttm_bo_validate_move_fence_not_signaled` at lines 757-759.

**Overall:** The patch is straightforward, well-structured, and follows the existing conventions. The observations above are all minor. The test provides useful coverage for `ttm_bo_swapout` that was previously missing.

**Reviewed-by worthy** with no blocking issues.

---
Generated by Claude Code Patch Reviewer