On Fri, Aug 14, 2026 at 04:32:59PM +0800, Xiaoyao Li wrote:
> > } \
> > })
> > +#define __tdx_vcpu_ioctl(vcpu, cmd, _flags, arg) \
> > +({ \
> > + union { \
> > + struct kvm_tdx_cmd c; \
> > + unsigned long raw; \
> > + } tdx_cmd = { .c = { \
> > + .id = (cmd), \
> > + .flags = (u32)(_flags), \
> > + .data = (u64)(arg), \
> > + } }; \
> > + \
> > + __vcpu_ioctl(vcpu, KVM_MEMORY_ENCRYPT_OP, &tdx_cmd.raw); \
> > +})
> > +
> > +#define tdx_vcpu_ioctl(vcpu, cmd, flags, arg)
> > \
> > +({ \
> > + int ret = __tdx_vcpu_ioctl(vcpu, cmd, flags, arg); \
> > + TEST_ASSERT(!ret, "%s failed, errno: %d (%s)", \
> > + #cmd, errno, strerror(errno)); \
> > +})
>
> It doesn't handle the tdx_cmd.c.hw_error for vcpu ioctl, but handle it for
> vm ioctl, which looks inconsistent. This at least deserves a justification
> in the changelog.
Looks like this assymetry came from a v13 comment [1].
tdx_vcpu_ioctl() actually doesn't write back hw_error to userspace like
tdx_vm_ioctl() does, so hw_error is vCPU-scoped only right now. This is
a bit subtle so probably deserves to be called out.
[1]
https://lore.kernel.org/all/[email protected]/
>
