[PATCH v12 64/84] KVM: LoongArch: Mark "struct page" pfns dirty only in "slow" page fault path

[Sean Christopherson]

On Fri, Aug 02, 2024, maobibo wrote:
> On 2024/7/27 上午7:52, Sean Christopherson wrote:
> > Mark pages/folios dirty only the slow page fault path, i.e. only when
> > mmu_lock is held and the operation is mmu_notifier-protected, 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
> > the race is minuscule, hence the lack of complaints).
> > 
> > See the link below for details.
> > 
> > Link: https://lore.kernel.org/all/cover.1683044162.git.lstoakes@gmail.com
> > Signed-off-by: Sean Christopherson <seanjc at google.com>
> > ---
> >   arch/loongarch/kvm/mmu.c | 18 ++++++++++--------
> >   1 file changed, 10 insertions(+), 8 deletions(-)
> > 
> > diff --git a/arch/loongarch/kvm/mmu.c b/arch/loongarch/kvm/mmu.c
> > index 2634a9e8d82c..364dd35e0557 100644
> > --- a/arch/loongarch/kvm/mmu.c
> > +++ b/arch/loongarch/kvm/mmu.c
> > @@ -608,13 +608,13 @@ static int kvm_map_page_fast(struct kvm_vcpu *vcpu, unsigned long gpa, bool writ
> >   		if (kvm_pte_young(changed))
> >   			kvm_set_pfn_accessed(pfn);
> > -		if (kvm_pte_dirty(changed)) {
> > -			mark_page_dirty(kvm, gfn);
> > -			kvm_set_pfn_dirty(pfn);
> > -		}
> >   		if (page)
> >   			put_page(page);
> >   	}
> > +
> > +	if (kvm_pte_dirty(changed))
> > +		mark_page_dirty(kvm, gfn);
> > +
> >   	return ret;
> >   out:
> >   	spin_unlock(&kvm->mmu_lock);
> > @@ -915,12 +915,14 @@ static int kvm_map_page(struct kvm_vcpu *vcpu, unsigned long gpa, bool write)
> >   	else
> >   		++kvm->stat.pages;
> >   	kvm_set_pte(ptep, new_pte);
> > -	spin_unlock(&kvm->mmu_lock);
> > -	if (prot_bits & _PAGE_DIRTY) {
> > -		mark_page_dirty_in_slot(kvm, memslot, gfn);
> > +	if (writeable)
> Is it better to use write or (prot_bits & _PAGE_DIRTY) here?  writable is
> pte permission from function hva_to_pfn_slow(), write is fault action.

Marking folios dirty in the slow/full path basically necessitates marking the
folio dirty if KVM creates a writable SPTE, as KVM won't mark the folio dirty
if/when _PAGE_DIRTY is set.

Practically speaking, I'm 99.9% certain it doesn't matter.  The folio is marked
dirty by core MM when the folio is made writable, and cleaning the folio triggers
an mmu_notifier invalidation.  I.e. if the page is mapped writable in KVM's
stage-2 PTEs, then its folio has already been marked dirty.