Re: [PATCH v3 0/7] KVM: Guest Memory Pre-Population API

[Paolo Bonzini]

Il 18 aprile 2024 02:01:03 CEST, "Edgecombe, Rick P" <rick.p.edgecombe@xxxxxxxxx> ha scritto:
>On Wed, 2024-04-17 at 11:34 -0400, Paolo Bonzini wrote:
>> 
>> Compared to Isaku's v2, I have reduced the scope as much as possible:
>> 
>> - no vendor-specific hooks
>
>The TDX patches build on this, with the vendor callback looking like:
>
>"
>int tdx_pre_mmu_map_page(struct kvm_vcpu *vcpu,
>			 struct kvm_map_memory *mapping,
>			 u64 *error_code)
>{
>	struct kvm_tdx *kvm_tdx = to_kvm_tdx(vcpu->kvm);
>	struct kvm *kvm = vcpu->kvm;
>
>	if (!to_tdx(vcpu)->initialized)
>		return -EINVAL;
>
>	/* Shared-EPT case */
>	if (!(kvm_is_private_gpa(kvm, mapping->base_address)))
>		return 0;
>
>	/* Once TD is finalized, the initial guest memory is fixed. */
>	if (is_td_finalized(kvm_tdx))
>		return -EINVAL;

This is wrong, KVM_MAP_MEMORY should be idempotent. But anyway, you can post what you have on to of kvm-coco-queue (i.e., adding the hook in your patches) and we will sort it out a piece at a time.

Paolo

>
>	*error_code = TDX_SEPT_PFERR;
>	return 0;
>}
>"
>
>kvm_is_private_gpa() check is already handled in this series.
>
>The initialized and finalized checks are nice guard rails for userspace, but
>shouldn't be strictly required.
>
>The TDX_SEPT_PFERR is (PFERR_WRITE_MASK | PFERR_PRIVATE_ACCESS). The
>PFERR_WRITE_MASK doesn't seem to be required. Isaku, what was the intention?
>
>But, I think maybe we should add a hook back in the TDX series for the guard
>rails.

Paolo