Re: [PATCH v11 099/113] KVM: TDX: Handle TDX PV map_gpa hypercall

[Isaku Yamahata]

On Tue, Jan 31, 2023 at 09:30:29AM +0800,
Yuan Yao <yuan.yao@xxxxxxxxxxxxxxx> wrote:

> On Thu, Jan 12, 2023 at 08:32:47AM -0800, isaku.yamahata@xxxxxxxxx wrote:
> > From: Isaku Yamahata <isaku.yamahata@xxxxxxxxx>
> >
> > Wire up TDX PV map_gpa hypercall to the kvm/mmu backend.
> >
> > Signed-off-by: Isaku Yamahata <isaku.yamahata@xxxxxxxxx>
> > ---
> >  arch/x86/kvm/vmx/tdx.c | 53 ++++++++++++++++++++++++++++++++++++++++++
> >  1 file changed, 53 insertions(+)
> >
> > diff --git a/arch/x86/kvm/vmx/tdx.c b/arch/x86/kvm/vmx/tdx.c
> > index 4bbde58510a4..486d0f0c6dd1 100644
> > --- a/arch/x86/kvm/vmx/tdx.c
> > +++ b/arch/x86/kvm/vmx/tdx.c
> > @@ -1181,6 +1181,57 @@ static int tdx_emulate_wrmsr(struct kvm_vcpu *vcpu)
> >  	return 1;
> >  }
> >
> > +static int tdx_map_gpa(struct kvm_vcpu *vcpu)
> > +{
> > +	struct kvm *kvm = vcpu->kvm;
> > +	gpa_t gpa = tdvmcall_a0_read(vcpu);
> > +	gpa_t size = tdvmcall_a1_read(vcpu);
> > +	gpa_t end = gpa + size;
> > +	gfn_t s = gpa_to_gfn(gpa) & ~kvm_gfn_shared_mask(kvm);
> > +	gfn_t e = gpa_to_gfn(end) & ~kvm_gfn_shared_mask(kvm);
> > +	int i;
> > +
> > +	if (!IS_ALIGNED(gpa, 4096) || !IS_ALIGNED(size, 4096) ||
> > +	    end < gpa ||
> > +	    end > kvm_gfn_shared_mask(kvm) << (PAGE_SHIFT + 1) ||
> > +	    kvm_is_private_gpa(kvm, gpa) != kvm_is_private_gpa(kvm, end)) {
> > +		tdvmcall_set_return_code(vcpu, TDG_VP_VMCALL_INVALID_OPERAND);
> > +		return 1;
> > +	}
> > +
> > +	/*
> > +	 * Check how the requested region overlaps with the KVM memory slots.
> > +	 * For simplicity, require that it must be contained within a memslot or
> > +	 * it must not overlap with any memslots (MMIO).
> > +	 */
> > +	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
> > +		struct kvm_memslots *slots = __kvm_memslots(kvm, i);
> > +		struct kvm_memslot_iter iter;
> > +
> > +		kvm_for_each_memslot_in_gfn_range(&iter, slots, s, e) {
> > +			struct kvm_memory_slot *slot = iter.slot;
> > +			gfn_t slot_s = slot->base_gfn;
> > +			gfn_t slot_e = slot->base_gfn + slot->npages;
> > +
> > +			/* no overlap */
> > +			if (e < slot_s || s >= slot_e)
> > +				continue;
> > +
> > +			/* contained in slot */
> > +			if (slot_s <= s && e <= slot_e) {
> > +				if (kvm_slot_can_be_private(slot))
> > +					return tdx_vp_vmcall_to_user(vcpu);
> > +				continue;
> > +			}
> > +
> > +			break;
> > +		}
> > +	}
> > +
> > +	tdvmcall_set_return_code(vcpu, TDG_VP_VMCALL_INVALID_OPERAND);
> 
> This returns TDG_VP_VMCALL_INVALID_OPERAND if the TD is running with
> non-private slots, which looks incorrect to the caller in TD guest(because
> the operands are correct). Can we just refuse to create TD if private memory
> slot is the only supported slot type for it ?

TD needs to support both private and non-private memory slot for device
assignment via shared pages.

So question should be, what's the expectation for non-private memory slot with
memory attribute change or TDX map_gpa.

- setting memory attributes: KVM_SET_MEMORY_ATTRIBUTES
  - option 1. return error
  - option 2. silently ignore as nop
              Following get-attributes returns non-private.
  - option 3. Allow setting attributes value. but no effect.
              Following get-attributes returns private. but kvm page fault
              doesn't have any effect.

- TDX MAP GPA hypercall
  Let's assume the region is contained in a single memory slot.
  - option 1. return error
  - option 2. silently ignore as nop

For map gpa hypercall, it's better to return error because guest should know
which region is for private memory or device assignment.
-- 
Isaku Yamahata <isaku.yamahata@xxxxxxxxx>