On Thu, Jul 29, 2021, David Edmondson wrote:
> When passing the failing address and size out to user space, SGX must
> ensure not to trample on the earlier fields of the emulation_failure
> sub-union of struct kvm_run.
>
> Signed-off-by: David Edmondson <david.edmondson@xxxxxxxxxx>
> ---
> arch/x86/kvm/vmx/sgx.c | 8 +++-----
> 1 file changed, 3 insertions(+), 5 deletions(-)
>
> diff --git a/arch/x86/kvm/vmx/sgx.c b/arch/x86/kvm/vmx/sgx.c
> index 6693ebdc0770..63fb93163383 100644
> --- a/arch/x86/kvm/vmx/sgx.c
> +++ b/arch/x86/kvm/vmx/sgx.c
> @@ -53,11 +53,9 @@ static int sgx_get_encls_gva(struct kvm_vcpu *vcpu, unsigned long offset,
> static void sgx_handle_emulation_failure(struct kvm_vcpu *vcpu, u64 addr,
> unsigned int size)
> {
> - vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
> - vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
> - vcpu->run->internal.ndata = 2;
> - vcpu->run->internal.data[0] = addr;
> - vcpu->run->internal.data[1] = size;
> + uint64_t data[2] = { addr, size };
> +
> + kvm_prepare_emulation_failure_exit(vcpu, false, data, sizeof(data));
Assuming we go with my suggestion to have kvm_prepare_emulation_failure_exit()
capture the exit reason/info, it's probably worth converting all the
KVM_EXIT_INTERNAL_ERROR paths in sgx.c, even though the others don't clobber flags.
