Re: [PATCH 2/9] rust/vmstate: Support varray's num field wrapped in BqlCell

[Zhao Liu]

> > HPET currently wraps num_timers in BqlCell<>. Although BqlCell<> is not
> > necessary from strictly speaking, it makes sense for vmstate to respect
> > BqlCell.
> 
> Dropping BqlCell<> from num_timers is indeed possible.

Hi Paolo,

I would like to further discuss whether there's any safe issues.

num_timers is a property:

    qemu_api::define_property!(
        c"timers",
        HPETState,
        num_timers,
        unsafe { &qdev_prop_uint8 },
        u8,
        default = HPET_MIN_TIMERS
    ),

Then this means someone could set this property in C side or Rust side
by:

DeviceState *hpet = qdev_new(TYPE_HPET);
qdev_prop_set_uint8(hpet, "timers", 8);

(Though we haven't provide safe interface at Rust side to set property.)

Whatever this happens at C side or Rust side, this depends on QOM core
code (in C) to overwrite the HPETState::num_timers directly.

Then after the call to qdev_prop_set_uint8() starts, all subsequent
processes happen on the C side, so even though the rewriting of num_timers
is runtime, there are no additional safety considerations because it
doesn't cross FFI boundaries. Am I understanding this correctly?

Thanks,
Zhao