Claude review: drm/amdkfd: Wire up batch allocation in ioctl handler

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

Patch Review

**Overview:** Integrates batch allocation into ioctl handler.

**Issues:**

1. **Integer Overflow - Critical:**
```c
+	for (i = 0; i < num_ranges; i++) {
+		total_size += ranges[i].size;
+	}
```
`total_size` can overflow. Should use `check_add_overflow()`.

2. **Missing Range Overlap Check - Critical:**
No check for overlapping ranges. Overlapping ranges could cause:
- Double-mapping to same GPU address
- Corruption during page operations
- Confusion in interval tree

3. **Memory Allocation Without Limit:**
```c
+	ranges = kvmalloc_array(num_ranges, sizeof(*ranges), GFP_KERNEL);
```
No limit on `num_ranges`. A malicious user could cause kernel memory exhaustion (DoS).

4. **SVM Check Overflow:**
```c
+			if (interval_tree_iter_first(&p->svms.objects,
+						     ranges[i].start >> PAGE_SHIFT,
+						     (ranges[i].start + ranges[i].size - 1) >> PAGE_SHIFT)) {
```
The `ranges[i].start + ranges[i].size - 1` can overflow.

---

## SUMMARY OF CRITICAL ISSUES

**Must Fix Before Merge:**
1. Race conditions around `valid` flag and interval tree access (need proper locking)
2. Use-after-free in cleanup paths when MMU notifier callbacks race with resource cleanup
3. Integer overflows in size calculations throughout the series
4. Memory leaks in error paths (especially patch 6)
5. Missing range overlap validation (patch 8)
6. Missing maximum limit on num_ranges (DoS vulnerability)

**Should Fix:**
7. UAPI design (mmap_offset overloading)
8. Missing error handling for partial failures
9. Incomplete validation of HMM page returns
10. Debug logging in hot paths
11. Type inconsistencies (uint64_t vs u64)

The core approach is sound, but the implementation needs careful review of locking protocols, error handling, and input validation before this can be safely merged.

---
Generated by Claude Code Patch Reviewer