[00/10] rust: (mostly) type safe VMState

Message

Paolo Bonzini Jan. 17, 2025, 9 a.m. UTC

The existing translation of the C macros for vmstate does not make
any attempt to type-check vmstate declarations against the struct, so
introduce a new system that computes VMStateField based on the actual
struct declaration.

Macros do not have full access to the type system, therefore a full
implementation of this scheme requires a helper trait to analyze the type
and produce a VMStateField from it; the trait stores the VMStateField
as ana associated const.  Then, a macro "vmstate_of!" accepts arguments
similar to "offset_of!" and tricks the compiler into looking up the
trait for the right type.

Structs and cells have their own sub-VMStateDescription.  This poses
a problem in that references can only be included in associated consts
starting with Rust 1.83.0, when the const_refs_static was stabilized.
So for now structs and cells cannot be done in a type-safe manner,
but arrays, pointers and scalars can.

Technically, scalar types have the same issue in that they point to a
VMStateInfo.  For this case however there is onlyya limited list of
VMStateInfos, so I am placing them in an enum, and going from enum
to &VMStateInfo only when building the VMStateField (i.e. in a static's
value rather than in an associated const, and in a macro rather than in
a const fn).  This isn't great, but it's easily removed when QEMU starts
accepting a newer Rust version.

The series also introduces the pattern of applying changes to a struct
using methods with a fn(mut self) -> Self signature, for example

    pub const fn with_version_id(mut self, version_id: i32) -> Self {
        self.version_id = version_id;
        self
    }

This is not quite the builder pattern because there is no need for a
final ".build()" call, but it's similar/related.

The unfavorable diffstat comes half from the const_refs_static workaround,
half from new documentation (so in that case it's a plus rather than
a minus).  And also, there is more functionality and extensibility in
the new mechanism, whereas only a subset of the C macros was included
in the previous implementation.  The exception is "field_exists", which
may be added later if needed as a modifier, similar to with_version_id().

Paolo

v1->v2:
- add attribution for call_func_with_field macro
- add VMS_VARRAY_FLAGS, check it in with_array_flag
- remove Option<NonNull<T>> support, add Box<T>
- fix cut and paste typo in vmstate_varray_flag
- fix rustfmt
- move SCALAR_TYPE definitions to the right patch
- keep $num as an ident rather than tt
- remove vmstate_cell
- make order of parameters consistent in vmstate_clock (last patch, new)

Paolo Bonzini (10):
  rust: vmstate: add new type safe implementation
  rust: vmstate: implement VMState for non-leaf types
  rust: vmstate: add varray support to vmstate_of!
  rust: vmstate: implement Zeroable for VMStateField
  rust: vmstate: implement VMState for scalar types
  rust: vmstate: add public utility macros to implement VMState
  rust: qemu_api: add vmstate_struct
  rust: pl011: switch vmstate to new-style macros
  rust: vmstate: remove translation of C vmstate macros
  rust: vmstate: make order of parameters consistent in vmstate_clock

 rust/hw/char/pl011/src/device.rs       |   3 +-
 rust/hw/char/pl011/src/device_class.rs |  38 +-
 rust/hw/char/pl011/src/lib.rs          |   6 +
 rust/qemu-api/src/prelude.rs           |   2 +
 rust/qemu-api/src/vmstate.rs           | 700 ++++++++++++++++---------
 rust/qemu-api/src/zeroable.rs          |  31 ++
 6 files changed, 504 insertions(+), 276 deletions(-)