Claude review: drm/gpuvm: Ensure correctness of unmap/remaps of repeated regions

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

Patch Review

**`validate_repeated_unmap_request` has an issue with `do_div` modifying `multiple`:** The variable `multiple` is set to `req_addr` or `req_end`, then `do_div(multiple, ...)` is called. `do_div` replaces `multiple` with the quotient, and the remainder is the return value. The intent is to check alignment (remainder != 0 → error). This is correct.

**Missing alignment check for `req_addr` on the last VA:** The function checks `req_addr` alignment only against the `first` VA and `req_end` alignment against both `first` (if it extends past) and `last`. But if `first` is fully contained within the unmap range and `last` is a repeat VA, the start of `last` is necessarily aligned (since it was created that way), so only `req_end` matters. This seems correct.

**`DRM_GPUVM_HAS_REPEAT_MAPS` flag:** This is set but never cleared. If all repeat mappings are removed, subsequent map/unmap operations will still walk VAs twice (once in `validate_repeated_unmap_request` and once in the main loop). This is a performance concern but not a correctness bug. Worth a `FIXME` comment.

**Variable shadowing in `__drm_gpuvm_sm_unmap`:**

```c
 	drm_gpuvm_for_each_va_range_safe(va, next, gpuvm, req_addr, req_end) {
 		struct drm_gpuva_op_map prev = {}, next = {};
```

The `next` variable in the for-each loop is shadowed by the local `struct drm_gpuva_op_map next`. This is pre-existing code, not introduced by this patch, but worth noting that adding `validate_repeated_unmap_request` before this loop (which also uses `drm_gpuvm_for_each_va_range`) means the VA tree is iterated twice.

---
Generated by Claude Code Patch Reviewer