Re: [PATCH v12 54/84] KVM: arm64: Mark "struct page" pfns accessed/dirty before dropping mmu_lock

[Sean Christopherson <seanjc@google.com>]

On Tue, Aug 06, 2024, Marc Zyngier wrote:
> On Tue, 06 Aug 2024 00:26:54 +0100,
> Oliver Upton <oliver.upton@linux.dev> wrote:
> > 
> > On Mon, Aug 05, 2024 at 11:26:03PM +0000, Oliver Upton wrote:
> > > [+cc Fuad]
> > 
> > Take 2!
> > 
> > > Fuad, you mentioned in commit 9c30fc615daa ("KVM: arm64: Move setting
> > > the page as dirty out of the critical section") that restructuring
> > > around the MMU lock was helpful for reuse (presumably for pKVM), but I
> > > lack the context there.
> > > 
> > > On Fri, Jul 26, 2024 at 04:52:03PM -0700, Sean Christopherson wrote:
> > > > Mark pages/folios accessed+dirty prior to dropping mmu_lock, as marking a
> > > > page/folio dirty after it has been written back can make some filesystems
> > > > unhappy (backing KVM guests will such filesystem files is uncommon, and
> > > 
> > > typo: s/will/with/
> > > 
> > > > the race is minuscule, hence the lack of complaints).  See the link below
> > > > for details.
> 
> Should we consider reverting 9c30fc615daa then?

Aha!  After thinking through things more, I don't think a revert is necessary.
I _think_ the worst case scenario is that KVM would trigger this WARN in
filemap_unaccount_folio():

	/*
	 * At this point folio must be either written or cleaned by
	 * truncate.  Dirty folio here signals a bug and loss of
	 * unwritten data - on ordinary filesystems.
	 *
	 * But it's harmless on in-memory filesystems like tmpfs; and can
	 * occur when a driver which did get_user_pages() sets page dirty
	 * before putting it, while the inode is being finally evicted.
	 *
	 * Below fixes dirty accounting after removing the folio entirely
	 * but leaves the dirty flag set: it has no effect for truncated
	 * folio and anyway will be cleared before returning folio to
	 * buddy allocator.
	 */
	if (WARN_ON_ONCE(folio_test_dirty(folio) &&
			 mapping_can_writeback(mapping)))
		folio_account_cleaned(folio, inode_to_wb(mapping->host));

KVM won't actually write memory because the stage-2 mappings are protected by the
mmu_notifier, i.e. there is no risk of loss of data, even if the VM were backed
by memory that needs writeback.

And FWIW, given that multiple other KVM architectures mark folios dirty outside
of mmu_notifier protection and have never tripped over this, I think it's highly
unlikely the WARN will ever be triggered by a sane virtualization setup.

I can add something to that effect to the changelog, e.g. to document that this
isn't super urgent.