Re: [RFC PATCH v2 4/4] samples: rust: rust_misc_device: add uring_cmd example

[Sidong Yang <sidong.yang@furiosa.ai>]

On Fri, Aug 01, 2025 at 11:11:44AM -0300, Daniel Almeida wrote:
> Hi Sidong,
> 
> > On 27 Jul 2025, at 12:03, Sidong Yang <sidong.yang@furiosa.ai> wrote:
> > 
> > This patch makes rust_misc_device handle uring_cmd. Command ops are like
> > ioctl that set or get values in simple way.
> > 
> > Signed-off-by: Sidong Yang <sidong.yang@furiosa.ai>
> > ---
> > samples/rust/rust_misc_device.rs | 34 ++++++++++++++++++++++++++++++++
> > 1 file changed, 34 insertions(+)
> > 
> > diff --git a/samples/rust/rust_misc_device.rs b/samples/rust/rust_misc_device.rs
> > index c881fd6dbd08..1044bde86e8d 100644
> > --- a/samples/rust/rust_misc_device.rs
> > +++ b/samples/rust/rust_misc_device.rs
> > @@ -101,6 +101,7 @@
> >     c_str,
> >     device::Device,
> >     fs::File,
> > +    io_uring::IoUringCmd,
> >     ioctl::{_IO, _IOC_SIZE, _IOR, _IOW},
> >     miscdevice::{MiscDevice, MiscDeviceOptions, MiscDeviceRegistration},
> >     new_mutex,
> > @@ -114,6 +115,9 @@
> > const RUST_MISC_DEV_GET_VALUE: u32 = _IOR::<i32>('|' as u32, 0x81);
> > const RUST_MISC_DEV_SET_VALUE: u32 = _IOW::<i32>('|' as u32, 0x82);
> > 
> > +const RUST_MISC_DEV_URING_CMD_SET_VALUE: u32 = _IOR::<i32>('|' as u32, 0x83);
> > +const RUST_MISC_DEV_URING_CMD_GET_VALUE: u32 = _IOW::<i32>('|' as u32, 0x84);
> > +
> > module! {
> >     type: RustMiscDeviceModule,
> >     name: "rust_misc_device",
> > @@ -190,6 +194,36 @@ fn ioctl(me: Pin<&RustMiscDevice>, _file: &File, cmd: u32, arg: usize) -> Result
> > 
> >         Ok(0)
> >     }
> > +
> > +    fn uring_cmd(
> > +        me: Pin<&RustMiscDevice>,
> 
> "me" ?

ioctl() uses the args name with "me".

> 
> > +        io_uring_cmd: Pin<&mut IoUringCmd>,
> > +        _issue_flags: u32,
> > +    ) -> Result<i32> {
> > +        dev_info!(me.dev, "UringCmd Rust Misc Device Sample\n");
> > +
> > +        let cmd = io_uring_cmd.cmd_op();
> > +        let cmd_data = io_uring_cmd.sqe().cmd_data().as_ptr() as *const usize;
> > +
> > +        // SAFETY: `cmd_data` is guaranteed to be a valid pointer to the command data
> > +        // within the SQE structure.
> 
> This is what core::ptr::read_volatile says:
> 
> Safety
> Behavior is undefined if any of the following conditions are violated:
>     o src must be valid for reads.
>     o src must be properly aligned.
>     o src must point to a properly initialized value of type T.
> 
> You must prove that the pre-conditions above are fulfilled here.

Okay, I would describe pre-conditions.
> 
> > +        // FIXME: switch to read_once() when it's available.
> > +        let addr = unsafe { core::ptr::read_volatile(cmd_data) };
> 
> So drivers have to write "unsafe" directly? It isn´t forbidden, but
> we should try our best to avoid it.

I don't know it's the best way to use &[u8] `cmd_data` in driver. The driver should
cast the `cmd_data` to user structure i32 in this time. is there any safe way
to provide safe interface for user driver?

> 
> > +
> > +        match cmd {
> > +            RUST_MISC_DEV_URING_CMD_SET_VALUE => {
> > +                me.set_value(UserSlice::new(addr, 8).reader())?;
> > +            }
> > +            RUST_MISC_DEV_URING_CMD_GET_VALUE => {
> > +                me.get_value(UserSlice::new(addr, 8).writer())?;
> > +            }
> > +            _ => {
> > +                dev_err!(me.dev, "-> uring_cmd not recognised: {}\n", cmd);
> > +                return Err(ENOTTY);
> > +            }
> > +        }
> > +        Ok(0)
> > +    }
> > }
> > 
> 
> Who calls this function?

This function would be called in io_uring subsystem. In detail, there is caller
io_uring_cmd() in io_uring/uring_cmd.c. When io_uring subsystem noticed that
submission queue entry which has uring cmd op pushed to submission queue, it
would be submitted and actually calls this function.

> 
> > #[pinned_drop]
> > -- 
> > 2.43.0
> > 
> > 
> 
> - Daniel
>