Re: [PATCH v6 01/10] mm: add zone device coherent type memory support

[Jason Gunthorpe <jgg@nvidia.com>]

On Tue, Feb 15, 2022 at 05:49:07PM -0500, Felix Kuehling wrote:

> > Userspace does
> >   1) mmap(MAP_PRIVATE) to allocate anon memory
> >   2) something to trigger migration to install a ZONE_DEVICE page
> >   3) munmap()
> > 
> > Who decrements the refcout on the munmap?
> > 
> > When a ZONE_DEVICE page is installed in the PTE is supposed to be
> > marked as pte_devmap and that disables all the normal page refcounting
> > during munmap().
> > 
> > fsdax makes this work by working the refcounts backwards, the page is
> > refcounted while it exists in the driver, when the driver decides to
> > remove it then unmap_mapping_range() is called to purge it from all
> > PTEs and then refcount is decrd. munmap/fork/etc don't change the
> > refcount.
>
> Hmm, that just means, whether or not there are PTEs doesn't really
> matter.

Yes, that is the FSDAX model

> It should work the same as it does for DEVICE_PRIVATE pages. I'm not sure
> where DEVICE_PRIVATE page's refcounts are decremented on unmap, TBH. But I
> can't find it in our driver, or in the test_hmm driver for that matter.

It is not the same as DEVICE_PRIVATE because DEVICE_PRIVATE uses swap
entries. The put_page for that case is here:

static unsigned long zap_pte_range(struct mmu_gather *tlb,
				struct vm_area_struct *vma, pmd_t *pmd,
				unsigned long addr, unsigned long end,
				struct zap_details *details)
{
[..]
		if (is_device_private_entry(entry) ||
		    is_device_exclusive_entry(entry)) {
			struct page *page = pfn_swap_entry_to_page(entry);

			if (unlikely(zap_skip_check_mapping(details, page)))
				continue;
			pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
			rss[mm_counter(page)]--;

			if (is_device_private_entry(entry))
				page_remove_rmap(page, false);

			put_page(page);

However the devmap case will return NULL from vm_normal_page() and won't
do the put_page() embedded inside the __tlb_remove_page() in the
pte_present() block in the same function.

After reflecting for awhile, I think Christoph's idea is quite
good. Just make it so you don't set pte_devmap() on your pages and
then lets avoid pte_devmap for all refcount correct ZONE_DEVICE pages.

Jason