Re: [PATCH v8 12/18] KVM: SVM: Add support for static allocation of unified Page Encryption Bitmap.

From: Ashish Kalra
Date: Sat May 30 2020 - 01:49:36 EST


Hello Steve,

On Fri, May 29, 2020 at 07:07:33PM -0700, Steve Rutherford wrote:
> On Tue, May 5, 2020 at 2:18 PM Ashish Kalra <Ashish.Kalra@xxxxxxx> wrote:
> >
> > From: Ashish Kalra <ashish.kalra@xxxxxxx>
> >
> > Add support for static allocation of the unified Page encryption bitmap by
> > extending kvm_arch_commit_memory_region() callack to add svm specific x86_ops
> > which can read the userspace provided memory region/memslots and calculate
> > the amount of guest RAM managed by the KVM and grow the bitmap based
> > on that information, i.e. the highest guest PA that is mapped by a memslot.
> >
> > Signed-off-by: Ashish Kalra <ashish.kalra@xxxxxxx>
> > ---
> > arch/x86/include/asm/kvm_host.h | 1 +
> > arch/x86/kvm/svm/sev.c | 35 +++++++++++++++++++++++++++++++++
> > arch/x86/kvm/svm/svm.c | 1 +
> > arch/x86/kvm/svm/svm.h | 1 +
> > arch/x86/kvm/x86.c | 5 +++++
> > 5 files changed, 43 insertions(+)
> >
> > diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> > index fc74144d5ab0..b573ea85b57e 100644
> > --- a/arch/x86/include/asm/kvm_host.h
> > +++ b/arch/x86/include/asm/kvm_host.h
> > @@ -1254,6 +1254,7 @@ struct kvm_x86_ops {
> >
> > bool (*apic_init_signal_blocked)(struct kvm_vcpu *vcpu);
> > int (*enable_direct_tlbflush)(struct kvm_vcpu *vcpu);
> > + void (*commit_memory_region)(struct kvm *kvm, enum kvm_mr_change change);
> > int (*page_enc_status_hc)(struct kvm *kvm, unsigned long gpa,
> > unsigned long sz, unsigned long mode);
> > int (*get_page_enc_bitmap)(struct kvm *kvm,
> > diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
> > index 30efc1068707..c0d7043a0627 100644
> > --- a/arch/x86/kvm/svm/sev.c
> > +++ b/arch/x86/kvm/svm/sev.c
> > @@ -1377,6 +1377,41 @@ static int sev_resize_page_enc_bitmap(struct kvm *kvm, unsigned long new_size)
> > return 0;
> > }
> >
> > +void svm_commit_memory_region(struct kvm *kvm, enum kvm_mr_change change)
> > +{
> > + struct kvm_memslots *slots;
> > + struct kvm_memory_slot *memslot;
> > + gfn_t start, end = 0;
> > +
> > + spin_lock(&kvm->mmu_lock);
> > + if (change == KVM_MR_CREATE) {
> > + slots = kvm_memslots(kvm);
> > + kvm_for_each_memslot(memslot, slots) {
> > + start = memslot->base_gfn;
> > + end = memslot->base_gfn + memslot->npages;
> > + /*
> > + * KVM memslots is a sorted list, starting with
> > + * the highest mapped guest PA, so pick the topmost
> > + * valid guest PA.
> > + */
> > + if (memslot->npages)
> > + break;
> > + }
> > + }
> > + spin_unlock(&kvm->mmu_lock);
> > +
> > + if (end) {
> > + /*
> > + * NORE: This callback is invoked in vm ioctl
> > + * set_user_memory_region, hence we can use a
> > + * mutex here.
> > + */
> > + mutex_lock(&kvm->lock);
> > + sev_resize_page_enc_bitmap(kvm, end);
> > + mutex_unlock(&kvm->lock);
> > + }
> > +}
> > +
> > int svm_page_enc_status_hc(struct kvm *kvm, unsigned long gpa,
> > unsigned long npages, unsigned long enc)
> > {
> > diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> > index 501e82f5593c..442adbbb0641 100644
> > --- a/arch/x86/kvm/svm/svm.c
> > +++ b/arch/x86/kvm/svm/svm.c
> > @@ -4015,6 +4015,7 @@ static struct kvm_x86_ops svm_x86_ops __initdata = {
> >
> > .check_nested_events = svm_check_nested_events,
> >
> > + .commit_memory_region = svm_commit_memory_region,
> > .page_enc_status_hc = svm_page_enc_status_hc,
> > .get_page_enc_bitmap = svm_get_page_enc_bitmap,
> > .set_page_enc_bitmap = svm_set_page_enc_bitmap,
> > diff --git a/arch/x86/kvm/svm/svm.h b/arch/x86/kvm/svm/svm.h
> > index 2ebdcce50312..fd99e0a5417a 100644
> > --- a/arch/x86/kvm/svm/svm.h
> > +++ b/arch/x86/kvm/svm/svm.h
> > @@ -406,6 +406,7 @@ int svm_page_enc_status_hc(struct kvm *kvm, unsigned long gpa,
> > unsigned long npages, unsigned long enc);
> > int svm_get_page_enc_bitmap(struct kvm *kvm, struct kvm_page_enc_bitmap *bmap);
> > int svm_set_page_enc_bitmap(struct kvm *kvm, struct kvm_page_enc_bitmap *bmap);
> > +void svm_commit_memory_region(struct kvm *kvm, enum kvm_mr_change change);
> >
> > /* avic.c */
> >
> > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> > index c4166d7a0493..8938de868d42 100644
> > --- a/arch/x86/kvm/x86.c
> > +++ b/arch/x86/kvm/x86.c
> > @@ -10133,6 +10133,11 @@ void kvm_arch_commit_memory_region(struct kvm *kvm,
> > kvm_mmu_change_mmu_pages(kvm,
> > kvm_mmu_calculate_default_mmu_pages(kvm));
> >
> > + if (change == KVM_MR_CREATE || change == KVM_MR_DELETE) {
> > + if (kvm_x86_ops.commit_memory_region)
> > + kvm_x86_ops.commit_memory_region(kvm, change);
> Why not just call this every time (if it exists) and have the
> kvm_x86_op determine if it should do anything?
>
> It seems like it's a nop anyway unless you are doing a create.
>

Yes, this makes sense.

I will call it unconditionally it it exits and let the callback
determine what to do eventually with it.

Thanks,
Ashish

> > + }
> > +
> > /*
> > * Dirty logging tracks sptes in 4k granularity, meaning that large
> > * sptes have to be split. If live migration is successful, the guest
> > --
> > 2.17.1
> >