Re: Mainline loongarch64 defconfig+Rust build error

[Gary Guo <gary@garyguo.net>]

On Tue, 09 Dec 2025 19:28:33 +0800
Xi Ruoyao <xry111@xry111.site> wrote:

> On Tue, 2025-12-09 at 11:56 +0100, Miguel Ojeda wrote:
> > Hi all,
> > 
> > I am seeing in mainline loongarch64 defconfig+Rust the following build
> > error [1].
> > 
> > Cheers,
> > Miguel
> > 
> > [1]
> > 
> >  error[E0425]: cannot find value `compat_ptr_ioctl` in crate `bindings`
> >      --> drivers/android/binder/rust_binder_main.rs:317:38
> >       |
> > 317   |         compat_ioctl: Some(bindings::compat_ptr_ioctl),
> >       |                                      ^^^^^^^^^^^^^^^^ help: a
> > function with a similar name exists: `compat_sys_ioctl`
> >       |
> >      ::: /__w/quick/quick/linux/rust/bindings/bindings_generated.rs:78264:5
> >       |
> > 78264 |     pub fn compat_sys_ioctl(fd: ffi::c_uint, cmd: ffi::c_uint,
> > arg: compat_ulong_t) -> ffi::c_long;
> >       |     ----------------------------------------------------------------------------------------------
> > similarly named function `compat_sys_ioctl` defined here  
> 
> I highly doubt if this is really a LoongArch-only issue.  We have:
> 
> #ifdef CONFIG_COMPAT
> extern long compat_ptr_ioctl(struct file *file, unsigned int cmd, 
>                     unsigned long arg);
> #else
> #define compat_ptr_ioctl NULL
> #endif
> 
> AFAIK bindgen does not translate C preprocessor macro definitions, so
> when !CONFIG_COMPAT we won't have the Rust binding for it.
> 
> And as LA32 is not upstreamed it's pointless to support COMPAT for
> LoongArch, yet.
> 

I think bindgen can't handle this on its own anyway, as it doesn't
really know that you want a `Option<fn(..) -> ...>` type here instead
of just a function item.

Probably the best way is for each user of this function handle
CFG_COMPAT on its own for now. Alternatively, we can just provide a
helper that just returns `-ENONIOCTLCMD` instead of using NULL?

Besat,
Gary