Re: [RFC PATCH 3/9] rust: vmstate: add varray support to vmstate_of!

[Paolo Bonzini <pbonzini@redhat.com>]

On 1/8/25 04:28, Zhao Liu wrote:
>> +    #[must_use]
>> +    pub const fn with_varray_flag<T: VMState>(mut self, flag: VMStateFlags) -> VMStateField {
>> +        assert!((self.flags.0 & VMStateFlags::VMS_ARRAY.0) != 0);
> 
> I understand you checked VMS_ARRAY here since [T; N] has this array
> flag.
> 
> What if a Rust device just store a pointer to the array? If we allow
> this use, then it seems also possible to set varray flags...Then what
> about dropping this limitation?

Box can be added for that purpose:

impl_vmstate_pointer!(Box<T> where T: VMState);

Also I need to drop Option<NonNull<>> because of

             if (field->flags & VMS_POINTER) {
                 first_elem = *(void **)first_elem;
                 assert(first_elem || !n_elems || !size);
             }

in migration/vmstate.c.

> However, I also doube that pointer usage is bad; we should always use
> Vec.

Vec support is a bit tricky because the number of elements is not 
accessible from C.  But unbounded arrays are rare in devices.  We can 
think about it later.

Paolo