[v2,00/13] rust: miscellaneous cleanups + QOM integration tests

Message

Paolo Bonzini Oct. 21, 2024, 4:35 p.m. UTC

This series integrates some of the observations from the MSRV patches at
https://lore.kernel.org/qemu-devel/[e-mail]/.

The main changes here are two: first, build an integration test that
actually tries to create a QOM object that is defined by Rust code;
second, make the properties array immutable so that declare_properties!
is enforced to use only const-friendly constructs.  These are patches
6-11; the others consist of small cleanups.

Hidden in here is actually a C patch (#10) which makes the
bindgen-generated prototypes use "*const" instead of "*mut".

Tested with Rust nightly and (together with more patches from the
RFC), with Rust 1.63.0.

Unlike the MSRV patches, this should be ready for inclusion; the
changes should be mostly uncontroversial.

Paolo

Based-on: <[e-mail]>


v1->v2:
- add comment justifying -D unsafe_op_in_unsafe_fn
- use module_init! macro from #[derive(Object)]
- use test() instead of rust.test() for integration test
- pass rustfmt and clippy


Paolo Bonzini (13):
  meson: import rust module into a global variable
  meson: remove repeated search for rust_root_crate.sh
  meson: pass rustc_args when building all crates
  rust: do not use --no-size_t-is-usize
  rust: remove uses of #[no_mangle]
  rust: modernize link_section usage for ELF platforms
  rust: build integration test for the qemu_api crate
  rust: cleanup module_init!, use it from #[derive(Object)]
  rust: clean up define_property macro
  qdev: make properties array "const"
  rust: make properties array immutable
  rust: provide safe wrapper for MaybeUninit::zeroed()
  rust: do not use TYPE_CHARDEV unnecessarily

 meson.build                            | 33 +++++++---
 include/hw/qdev-core.h                 |  4 +-
 include/hw/qdev-properties.h           |  4 +-
 hw/core/qdev-properties.c              | 26 ++++----
 system/qdev-monitor.c                  |  2 +-
 rust/hw/char/pl011/src/device.rs       |  7 +--
 rust/hw/char/pl011/src/device_class.rs | 10 ++--
 rust/hw/char/pl011/src/memory_ops.rs   | 13 ++--
 rust/qemu-api-macros/meson.build       |  2 +-
 rust/qemu-api-macros/src/lib.rs        | 30 ++--------
 rust/qemu-api/meson.build              | 29 +++++++--
 rust/qemu-api/src/definitions.rs       | 67 ++++++++++-----------
 rust/qemu-api/src/device_class.rs      | 83 +++++++-------------------
 rust/qemu-api/src/lib.rs               |  4 +-
 rust/qemu-api/src/tests.rs             | 49 ---------------
 rust/qemu-api/src/zeroable.rs          | 23 +++++++
 rust/qemu-api/tests/tests.rs           | 78 ++++++++++++++++++++++++
 17 files changed, 240 insertions(+), 224 deletions(-)
 delete mode 100644 rust/qemu-api/src/tests.rs
 create mode 100644 rust/qemu-api/src/zeroable.rs
 create mode 100644 rust/qemu-api/tests/tests.rs

Comments

Kevin Wolf Oct. 22, 2024, 8:46 p.m. UTC | #1

Am 21.10.2024 um 18:35 hat Paolo Bonzini geschrieben:
> This series integrates some of the observations from the MSRV patches at
> https://lore.kernel.org/qemu-devel/[e-mail]/.
> 
> The main changes here are two: first, build an integration test that
> actually tries to create a QOM object that is defined by Rust code;
> second, make the properties array immutable so that declare_properties!
> is enforced to use only const-friendly constructs.  These are patches
> 6-11; the others consist of small cleanups.
> 
> Hidden in here is actually a C patch (#10) which makes the
> bindgen-generated prototypes use "*const" instead of "*mut".
> 
> Tested with Rust nightly and (together with more patches from the
> RFC), with Rust 1.63.0.
> 
> Unlike the MSRV patches, this should be ready for inclusion; the
> changes should be mostly uncontroversial.

I'm not convinced that Zeroable has sufficient justification when all it
does is saving us a few lines of code at the expense of making things
more implicit. But it's used correctly as far as I can tell, so:

Reviewed-by: Kevin Wolf <[e-mail]>

Paolo Bonzini Oct. 23, 2024, 7:14 a.m. UTC | #2

Il mar 22 ott 2024, 22:46 Kevin Wolf <[e-mail]> ha scritto:

> Am 21.10.2024 um 18:35 hat Paolo Bonzini geschrieben:
> > This series integrates some of the observations from the MSRV patches at
> >
> https://lore.kernel.org/qemu-devel/[e-mail]/
> .
> >
> > The main changes here are two: first, build an integration test that
> > actually tries to create a QOM object that is defined by Rust code;
> > second, make the properties array immutable so that declare_properties!
> > is enforced to use only const-friendly constructs.  These are patches
> > 6-11; the others consist of small cleanups.
> >
> > Hidden in here is actually a C patch (#10) which makes the
> > bindgen-generated prototypes use "*const" instead of "*mut".
> >
> > Tested with Rust nightly and (together with more patches from the
> > RFC), with Rust 1.63.0.
> >
> > Unlike the MSRV patches, this should be ready for inclusion; the
> > changes should be mostly uncontroversial.
>
> I'm not convinced that Zeroable has sufficient justification when all it
> does is saving us a few lines of code at the expense of making things
> more implicit


Implicit zero fields are used a lot in C code, including in
VMStateDescription and various const vtable structs, so I think we need
something of equivalent brevity.

There could be other solutions, for example a more generic
ConstDefault::DEFAULT trait. For now I went for something that resembles
existing code, as well as the C code we're converting.

. But it's used correctly as far as I can tell, so:
>
> Reviewed-by: Kevin Wolf <[e-mail]>
>

Thanks!

Paolo


>