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

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

Patch Review

**Comma instead of semicolon (bug):** 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 lines end with `,` instead of `;`. While this compiles (the comma operator evaluates left-to-right and discards the result of the left operand), it's clearly a typo. The `fault_type` assignment becomes the left operand of a comma expression whose result is the `fault_level` assignment. In practice this works, but it's wrong and confusing. These should be semicolons.

**Typo in comment:** "asociated" should be "associated" in `xe_pagefault_save_to_vm`.

**Duplicate forward declaration:** `struct xe_pagefault;` is forward-declared in both `xe_vm_types.h` and `xe_vm.h`. The one in `xe_vm_types.h` is unnecessary since nothing in that file references `xe_pagefault`.

**No mechanism to clear/consume faults:** Faults accumulate up to 50 and are never cleared until the VM is destroyed. There's no way for userspace to "consume" or reset the fault list after reading it. Once 50 faults are saved, no new faults are recorded. This seems like a significant design limitation — userspace can only ever see the first 50 faults in the VM's lifetime.

**Memory allocation before checking MAX:** `kzalloc_obj(*e)` allocates memory before checking `vm->faults.len >= MAX_FAULTS_SAVED_PER_VM`. If the list is full, the allocation is immediately freed. A minor inefficiency — checking `vm->faults.len` first (even without the lock, as a hint) would avoid the allocation in the common "full" case.

---
Generated by Claude Code Patch Reviewer