[no subject]

What I am trying to say with the above example is:
- UFFD registeration + MADV_NOHUGEPAGE should be done by all applications before any data in the region is accessed. Using THPs and accessing data before UFFD registeration + MADV_NOHUGEPAGE can lead to unexpected behaviour and is wrong?
- even in the current kernel code in other places like khugepaged, its only checked if uffd is enabled currently. It is not tracked if it was ever enabled on any VMA.

Thanks for pointing to mm_forbids_zeropage. Incorporating that into the code, and if I am (hopefully :)) right about qemu and kernel above, then I believe the right code should be:

	if (userfaultfd_armed(pvmw->vma) && mm_forbids_zeropage(pvmw->vma->vm_mm))
		return false;

	if (!userfaultfd_armed(pvmw->vma)) {
		pte_clear_not_present_full(pvmw->vma->vm_mm, pvmw->address, pvmw->pte, false);
	} else {
		newpte = pte_mkspecial(pfn_pte(page_to_pfn(ZERO_PAGE(pvmw->address)),
					       pvmw->vma->vm_page_prot));
		ptep_clear_flush(pvmw->vma, pvmw->address, pvmw->pte);
		set_pte_at(pvmw->vma->vm_mm, pvmw->address, pvmw->pte, newpte);
	}
 


[1] https://github.com/qemu/qemu/blob/4e56e89d6c81589cc47cf5811f570c67889bd18a/migration/migration.c#L3817
[2] https://github.com/qemu/qemu/blob/4e56e89d6c81589cc47cf5811f570c67889bd18a/migration/migration.c#L3455
[3] https://github.com/qemu/qemu/blob/4e56e89d6c81589cc47cf5811f570c67889bd18a/migration/migration.c#L3591
[4] https://github.com/qemu/qemu/blob/4e56e89d6c81589cc47cf5811f570c67889bd18a/migration/migration.c#L3675
[5] https://github.com/torvalds/linux/blob/master/mm/khugepaged.c#L1307