Claude review: accel/amdxdna: Disable device buffer exporting

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

Patch Review

**Status: Looks good**

The patch adds a minimal `.export` callback that returns `-EOPNOTSUPP` and wires it into `amdxdna_gem_dev_obj_funcs`:

```c
static struct dma_buf *amdxdna_gem_dev_obj_export(struct drm_gem_object *gobj, int flags)
{
	return ERR_PTR(-EOPNOTSUPP);
}
```

**Observations:**

1. **Correct error code**: `-EOPNOTSUPP` is appropriate here — it signals that the operation is not supported, which is the right semantic for a buffer type that should never be exported.

2. **Correct override mechanism**: Without this callback, `drm_gem_prime_handle_to_fd()` in `drm_prime.c:387-390` would fall through to `drm_gem_prime_export()`, which would attempt the export. The `dev_obj_funcs` struct lacks `.pin`, `.get_sg_table`, and other callbacks that the default export path relies on, so the export would likely fail in a confusing way or produce a non-functional dma-buf. Explicitly returning an error is much cleaner.

3. **Consistent with driver design**: The shmem object funcs (`amdxdna_gem_shmem_funcs` at line 739) have a custom `.export = amdxdna_gem_prime_export`, showing the driver is intentional about which object types can be exported. This patch completes that story for device objects.

4. **Unused parameter**: The `gobj` and `flags` parameters are unused, but this is standard for stub/rejection callbacks in the kernel — no `__maybe_unused` annotation is needed since the kernel build suppresses these warnings for function pointer implementations.

No concerns with this patch. It's a clean, minimal hardening fix.

---
Generated by Claude Code Patch Reviewer