Re: [PATCH 2/4] rust: io_uring: introduce rust abstraction for io-uring cmd

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

On Tue, Jul 22, 2025 at 08:52:05PM +0200, Benno Lossin wrote:
> On Tue Jul 22, 2025 at 4:30 PM CEST, Sidong Yang wrote:
> > On Mon, Jul 21, 2025 at 06:28:09PM +0200, Benno Lossin wrote:
> >> On Mon Jul 21, 2025 at 5:47 PM CEST, Sidong Yang wrote:
> >> > It's safest to get NonNull from from_raw and it returns
> >> > Pin<&mut IoUringCmd>.
> >> 
> >> I don't think you need `NonNull<T>`.
> >
> > NonNull<T> gurantees that it's not null. It could be also dangling but it's
> > safer than *mut T. Could you tell me why I don't need it?
> 
> Raw pointers have better ergonomics and if you're just passing it back
> into ffi, I don't see the point of using `NonNull`...

Agreed, from_raw() would be called in rust kernel unsafe condition not in safe
driver. Thinking it over, using a raw pointer would be convenient.

> 
> >> > from_raw() name is weird. it should be from_nonnnull()? Also, done()
> >> > would get Pin<&mut Self>.
> >> 
> >> That sounds reasonable.
> >> 
> >> Are you certain that it's an exclusive reference?
> >
> > As far as I know, yes.
> 
> So the `IoUringCmd` is not refcounted and it is also not owned by the
> `done` callee?

Yes, io_uring_cmd would be allocated in io_submit_sqes() and it's not accessed
concurrently. io_uring_cmd_done() sets its result to io_uring_cmd and prepare
some field to completion.

Thanks,
Sidong

> 
> ---
> Cheers,
> Benno