Claude review: drm/xe/xe_drm_ras: Add error-event support in XE DRM RAS

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

Patch Review

**Critical concern — `error_id` and `ras` variable scope in `gt_hw_error_handler`:**

```c
xe_drm_ras_notify(ras, error_id, severity, GFP_ATOMIC);
```

This is called at the end of `gt_hw_error_handler`. Looking at the patch context, `ras` and `error_id` are local variables in that function. The `error_id` appears to be derived from the error vector processing loop above. However, by the time we reach the notify call at the end of the function (after the loop), `error_id` holds the *last* error ID that was processed. If multiple error IDs triggered simultaneously, only the last one gets notified. This seems like a bug — the notification should likely be inside the loop body, so each error ID triggers its own notification.

**Same concern in `soc_hw_error_handler`:**

```c
xe_drm_ras_notify(ras, error_id, severity, GFP_ATOMIC);
```

Again called after the loop, so only the last processed `error_id` gets notified. The same fix would apply — move the notification inside the error processing loop.

**`GFP_ATOMIC` usage is correct** since both `gt_hw_error_handler` and `soc_hw_error_handler` are called from interrupt context (under `spin_lock_irqsave` in `hw_error_source_handler`).

**`xe_drm_ras_notify` wrapper is thin but useful:**

```c
void xe_drm_ras_notify(struct xe_drm_ras *ras, u32 error_id,
                       const enum drm_xe_ras_error_severity severity, gfp_t flags)
{
    struct drm_ras_node *node = &ras->node[severity];
    drm_ras_error_notify(node, error_id, flags);
}
```

This provides a clean mapping from the XE-specific severity-based node array to the generic DRM RAS node. The `const` on the `severity` parameter in the declaration is unnecessary (it's a value parameter) but harmless.

**Missing bounds check on `severity`:** If an invalid `severity` value is passed, `ras->node[severity]` would access out of bounds. The callers currently pass values derived from `hw_err_to_severity()` which should be safe, but a defensive check wouldn't hurt.

---
Generated by Claude Code Patch Reviewer