Measuring memory fragmentation

We have a lot of places which use the page allocator directly. We can use tools/mm/page-types to show what kind of allocations are preventing high order allocations.

It's probably slab pages and page tables getting intermingled with page cache / anon allocations. But that's just my opinion.

Worth noting, perhaps, that the Oppo measurements were being done with large anon folios but small FS folios because they were using one of the FS that haven't yet been converted to large folios.