Re: [PATCH 6/7] KVM: x86: Implement kvm_arch_vcpu_map_memory()

[Sean Christopherson <seanjc@xxxxxxxxxx>]

On Wed, Apr 17, 2024, Paolo Bonzini wrote:
> From: Isaku Yamahata <isaku.yamahata@xxxxxxxxx>
> 
> Wire KVM_MAP_MEMORY ioctl to kvm_mmu_map_tdp_page() to populate guest
> memory.  When KVM_CREATE_VCPU creates vCPU, it initializes the x86
> KVM MMU part by kvm_mmu_create() and kvm_init_mmu().  vCPU is ready to
> invoke the KVM page fault handler.
> 
> Signed-off-by: Isaku Yamahata <isaku.yamahata@xxxxxxxxx>
> Message-ID: <7138a3bc00ea8d3cbe0e59df15f8c22027005b59.1712785629.git.isaku.yamahata@xxxxxxxxx>
> Signed-off-by: Paolo Bonzini <pbonzini@xxxxxxxxxx>
> ---
>  arch/x86/kvm/Kconfig |  1 +
>  arch/x86/kvm/x86.c   | 43 +++++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 44 insertions(+)
> 
> diff --git a/arch/x86/kvm/Kconfig b/arch/x86/kvm/Kconfig
> index 7632fe6e4db9..e58360d368ec 100644
> --- a/arch/x86/kvm/Kconfig
> +++ b/arch/x86/kvm/Kconfig
> @@ -44,6 +44,7 @@ config KVM
>  	select KVM_VFIO
>  	select HAVE_KVM_PM_NOTIFIER if PM
>  	select KVM_GENERIC_HARDWARE_ENABLING
> +	select KVM_GENERIC_MAP_MEMORY
>  	help
>  	  Support hosting fully virtualized guest machines using hardware
>  	  virtualization extensions.  You will need a fairly recent
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 83b8260443a3..f84c75c2a47f 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -4715,6 +4715,9 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
>  	case KVM_CAP_MEMORY_FAULT_INFO:
>  		r = 1;
>  		break;
> +	case KVM_CAP_MAP_MEMORY:
> +		r = tdp_enabled;
> +		break;
>  	case KVM_CAP_EXIT_HYPERCALL:
>  		r = KVM_EXIT_HYPERCALL_VALID_MASK;
>  		break;
> @@ -5867,6 +5870,46 @@ static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
>  	}
>  }
>  
> +int kvm_arch_vcpu_map_memory(struct kvm_vcpu *vcpu,
> +			     struct kvm_map_memory *mapping)
> +{
> +	u64 mapped, end, error_code = 0;

Maybe add PFERR_GUEST_FINAL_MASK to the error code?  KVM doesn't currently consume
that except in svm_check_emulate_instruction(), which isn't reachable, but it
seems logical?

> +	u8 level = PG_LEVEL_4K;
> +	int r;
> +
> +	/*
> +	 * Shadow paging uses GVA for kvm page fault.  The first implementation
> +	 * supports GPA only to avoid confusion.
> +	 */
> +	if (!tdp_enabled)

Eh, I'd omit this explicit check since kvm_tdp_map_page() has a more complete
check.

Actually, why is this a separate function and a separate patch?  Just implement
kvm_arch_vcpu_map_memory() in mmu.c, in a single patch, e.g.

int kvm_arch_vcpu_map_memory(struct kvm_vcpu *vcpu,
			     struct kvm_map_memory *mapping)
{
	u64 mapped, end, error_code = 0;
	u8 level = PG_LEVEL_4K;
	int r;

	if (vcpu->arch.mmu->page_fault != kvm_tdp_page_fault)
		return -EOPNOTSUPP;

	kvm_mmu_reload(vcpu);

	if (kvm_arch_has_private_mem(vcpu->kvm) &&
	    kvm_mem_is_private(vcpu->kvm, gpa_to_gfn(mapping->base_address)))
		error_code |= PFERR_PRIVATE_ACCESS;

	r = __kvm_mmu_do_page_fault(vcpu, mapping->gpa, error_code, true, NULL, &level);
	if (r < 0)
		return r;

	switch (r) {
	case RET_PF_RETRY:
		return -EAGAIN;
	case RET_PF_FIXED:
	case RET_PF_SPURIOUS:
		break;
	case RET_PF_EMULATE:
		return -EBUSY;
	case RET_PF_CONTINUE:
	case RET_PF_INVALID:
	default:
		WARN_ON_ONCE(r);
		return -EIO;
	}

	/*
	 * Adjust the GPA down when accounting for the page size, as KVM could
	 * have created a hugepage that covers @gpa, but doesn't start at @gpa.
	 */
	end = (mapping->gpa & KVM_HPAGE_MASK(level)) + KVM_HPAGE_SIZE(level);
	return min(mapping->size, end - mapping->gpa);
}