Claude review: drm/xe/xe_vm: Add per VM fault info

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

Patch Review

**Comma operator instead of semicolons**: Lines 629-631 in `xe_vm_add_fault_entry_pf()`:
```c
e->fault_type = FIELD_GET(XE_PAGEFAULT_TYPE_MASK,
			  pf->consumer.fault_type_level),
e->fault_level = FIELD_GET(XE_PAGEFAULT_LEVEL_MASK,
			   pf->consumer.fault_type_level),
```
Both assignments end with commas (`,`) instead of semicolons (`;`). While the comma operator makes this technically work (the assignment still happens), it's a style error that could confuse readers and static analysis tools. Should be semicolons.

**kzalloc before spinlock then kfree under spinlock**: The pattern of allocating `e` before the lock, then potentially doing `kfree(e)` inside the `guard(spinlock)` scope if `len >= MAX_FAULTS_SAVED_PER_VM` is fine since `kfree` is safe to call with a spinlock held (it's GFP_ATOMIC internally). However, it means an allocation + free happens on every fault after the 50th, which could be frequent. Consider checking the length before allocating:

```c
/* Quick unlocked check to avoid unnecessary allocation */
if (READ_ONCE(vm->faults.len) >= MAX_FAULTS_SAVED_PER_VM)
    return;
```

**Typo in comment**: Line 482: "asociated" should be "associated".

**No mechanism to clear faults**: Faults accumulate up to 50 and are never cleared until VM destruction (`xe_vm_close_and_put`). The ioctl lets userspace read them but not consume/clear them. This means once 50 faults are reached, no new faults are ever recorded. This seems like a significant design limitation — userspace has no way to "acknowledge" faults and make room for new ones. This should at least be documented as a known limitation or addressed with a clear/reset mechanism.

---

---
Generated by Claude Code Patch Reviewer