Re: [syzbot] [mm?] WARNING in __folio_rmap_sanity_checks

[David Hildenbrand]

> > vm_normal_page() works on these mappings, so we'd also have to skip rmap
> > code when unmapping these pages etc. Maybe that's the whole reason we
> > have the rmap handling here: to not special-case the unmap path.
>
> vm_insert_page() will set VM_MIXEDMAP and vm_normal_page() will skip
> the page if CONFIG_ARCH_HAS_PTE_SPECIAL is enabled (it's enabled for
> x86).  So the unmap path will skip these kind of folios?

I think we run into the
	if (likely(!pte_special(pte)))
		goto check_pfn;

first and return these folios. That also matches the comment of 
vm_normal_page: "VM_MIXEDMAP mappings can likewise contain memory with 
or without ... _all_ pages with a struct page (that is, those where 
pfn_valid is true) are refcounted and considered normal pages by the VM."

--
Cheers,

David / dhildenb