On Wed, Sep 26, 2018 at 02:30:02PM +1000, David Gibson wrote:
> On Wed, Sep 26, 2018 at 02:08:37PM +1000, David Gibson wrote:
> > > static void kvmppc_unmap_pte(struct kvm *kvm, pte_t *pte,
> > > - unsigned long gpa, unsigned int shift)
> > > + unsigned long gpa, unsigned int shift,
> > > + struct kvm_memory_slot *memslot)
> > >
> > > {
> > > - unsigned long page_size = 1ul << shift;
> > > unsigned long old;
> > >
> > > old = kvmppc_radix_update_pte(kvm, pte, ~0UL, 0, gpa, shift);
> > > kvmppc_radix_tlbie_page(kvm, gpa, shift);
> > > if (old & _PAGE_DIRTY) {
> > > unsigned long gfn = gpa >> PAGE_SHIFT;
> > > - struct kvm_memory_slot *memslot;
> > > + unsigned long page_size = PAGE_SIZE;
> > >
> > > - memslot = gfn_to_memslot(kvm, gfn);
> > > - if (memslot && memslot->dirty_bitmap)
> > > + if (shift)
> > > + page_size = 1ul << shift;
> > > + if (!memslot)
> > > + memslot = gfn_to_memslot(kvm, gfn);
>
> ..it might be nicer to avoid the explicit test on memslot by maping a
> __kvmppc_unmap_pte() which must have memslot passed in, and a wrapper
> which computes it from the gfn.
Then we would be looking up the memslot even when we aren't going to
use it. In a subsequent patch this function gets used on shadow page
tables and in that case we never need the memslot.
Paul.
