Claude review: mm: Add helper to convert HMM pfn to migrate pfn

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

Patch Review

Adds `HMM_PFN_MIGRATE`, `HMM_PFN_COMPOUND`, and `HMM_PFN_REQ_MIGRATE` flags, and the `migrate_hmm_range_setup()` conversion function.

**Issue 1 — `EXPORT_SYMBOL` vs `EXPORT_SYMBOL_GPL`:**
```c
+EXPORT_SYMBOL(migrate_hmm_range_setup);
```
All other `migrate_vma_*` functions in this file use `EXPORT_SYMBOL()`, so this is consistent. However, it's worth noting that device migration is inherently a privileged kernel operation — `EXPORT_SYMBOL_GPL` would be more appropriate and consistent with the broader mm subsystem trend.

**Issue 2 — HMM_PFN_ORDER_SHIFT compression:**
```c
-	HMM_PFN_ORDER_SHIFT = (BITS_PER_LONG - 11),
+	HMM_PFN_ORDER_SHIFT = (BITS_PER_LONG - 13),
```
Two new flag bits (`HMM_PFN_MIGRATE` at bit -7, `HMM_PFN_COMPOUND` at bit -8) consume space from the order field, shifting it from bits [53..57] to bits [51..55] on 64-bit. This reduces the maximum representable order from 31 to 31 (5 bits, unchanged width), so no functional impact. But the flags are getting tight in this namespace — 8 of the top 13 bits are now consumed.

**Issue 3 — `migrate_hmm_range_setup` calls static `migrate_vma_unmap`:**
```c
+	if (migrate->cpages)
+		migrate_vma_unmap(migrate);
```
This works because `migrate_hmm_range_setup` is defined in `migrate_device.c` alongside `migrate_vma_unmap` (which is static). Good.

**Minor — extra blank line:**
```c
+	if (migrate->cpages)
+		migrate_vma_unmap(migrate);
+
+}
```
Unnecessary blank line before the closing brace.

---
Generated by Claude Code Patch Reviewer