Claude review: drm/gem/shmem: Introduce __drm_gem_shmem_free_sgt_locked()

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

Patch Review

**Status: Two issues to flag**

**Issue 1 — kdoc name mismatch**: The kernel-doc comment names the function `__drm_gem_shmem_release_sgt_locked` but the actual function is `__drm_gem_shmem_free_sgt_locked`:

```c
+/**
+ * __drm_gem_shmem_release_sgt_locked - Unpin and DMA unmap pages, and release the
```
vs.
```c
+void __drm_gem_shmem_free_sgt_locked(struct drm_gem_shmem_object *shmem)
```

The V10 changelog says "Fix incorrect function name in documentation" but the mismatch is still present.

**Issue 2 — No NULL check on `shmem->sgt`**: The new exported function does not check for `shmem->sgt == NULL` before dereferencing:

```c
+void __drm_gem_shmem_free_sgt_locked(struct drm_gem_shmem_object *shmem)
+{
+	dma_resv_assert_held(shmem->base.resv);
+
+	dma_unmap_sgtable(shmem->base.dev->dev, shmem->sgt, DMA_BIDIRECTIONAL, 0);
```

Both callers currently guard against this (the C side with `if (shmem->sgt)`, the Rust side via SGTableMap type invariants). However, since this is `EXPORT_SYMBOL_GPL` and documented as callable by drivers (specifically the Rust bindings), a defensive NULL check with early return or `WARN_ON` would be prudent. If a future caller misuses this, it will be a NULL pointer dereference in kernel context.

**Note on the kdoc**: The description says "Unpin and DMA unmap pages" but the function doesn't actually unpin pages (it doesn't call `drm_gem_shmem_put_pages_locked()`). It only unmaps the SGTable and frees it. The "unpin" claim is inaccurate.

---
Generated by Claude Code Patch Reviewer