On Tue, Jun 24, 2025 at 02:50:23PM +0100, Alice Ryhl wrote:
> On Tue, Jun 24, 2025 at 1:46 PM Benno Lossin <lossin@xxxxxxxxxx> wrote:
> >
> > On Tue Jun 24, 2025 at 2:31 PM CEST, Daniel Almeida wrote:
> > > On 23 Jun 2025, at 16:28, Benno Lossin <lossin@xxxxxxxxxx> wrote:
> > >> On Mon Jun 23, 2025 at 9:18 PM CEST, Boqun Feng wrote:
> > >>> try_pin_init!(&this in Self {
> > >>> handler,
> > >>> inner: Devres::new(
> > >>> dev,
> > >>> RegistrationInner {
> > >>> // Needs to use `handler` address as cookie, same for
> > >>> // request_irq().
> > >>> cookie: &raw (*(this.as_ptr().cast()).handler),
> > >>> irq: {
> > >>> to_result(unsafe { bindings::request_irq(...) })?;
> > >>> irq
> > >>> }
> > >>> },
> > >>> GFP_KERNEL,
> > >>> )?,
> > >>> _pin: PhantomPinned
> > >>> })
> > >>
> > >> Well yes and no, with the Devres changes, the `cookie` can just be the
> > >> address of the `RegistrationInner` & we can do it this way :)
> > >>
> > >> ---
> > >> Cheers,
> > >> Benno
> > >
> > >
> > > No, we need this to be the address of the the whole thing (i.e.
> > > Registration<T>), otherwise you can’t access the handler in the irq
> > > callback.
You only need the access of `handler` in the irq callback, right? I.e.
passing the address of `handler` would suffice (of course you need
to change the irq callback as well).
> >
> > Gotcha, so you keep the cookie field, but you should still be able to
> > use `try_pin_init` & the devres improvements to avoid the use of
> > `pin_init_from_closure`.
>
> It sounds like this is getting too complicated and that
> `pin_init_from_closure` is the simpler way to go.
Even if we use `pin_init_from_closure`, we still need the other
`try_pin_init` anyway for `Devres::new()` (or alternatively we can
implement a `RegistrationInner::new()`).
Below is what would look like with the Devres changes in mind:
try_pin_init!(&this in Self {
handler,
inner: <- Devres::new(
dev,
try_pin_init!( RegistrationInner {
// Needs to use `handler` address as cookie, same for
// request_irq().
cookie: &raw (*(this.as_ptr().cast()).handler),
// @Benno, would this "this" work here?
irq: {
to_result(unsafe { bindings::request_irq(...) })?;
irq
}
}),
)?,
_pin: PhantomPinned
})
Besides, working on this made me realize that we have to request_irq()
before `Devres::new()`, otherwise we may leak the irq resource,
considering the follow code from the current `pin_init_from_closure`
approach:
let closure = move |slot: *mut Self| {
// SAFETY: The slot passed to pin initializer is valid for writing.
unsafe {
slot.write(Self {
inner: Devres::new(
dev,
RegistrationInner {
irq,
cookie: slot.cast(),
},
GFP_KERNEL,
)?,
handler,
_pin: PhantomPinned,
})
};
`dev` can be unbound at here, right? If so, the devm callback will
revoke the `RegistrationInner`, `RegistrationInner::drop()` will then
call `free_irq()` before `request_irq()`, the best case is that we would
request_irq() with no one going to free it.
// SAFETY:
// - The callbacks are valid for use with request_irq.
// - If this succeeds, the slot is guaranteed to be valid until the
// destructor of Self runs, which will deregister the callbacks
// before the memory location becomes invalid.
let res = to_result(unsafe {
bindings::request_irq(
irq,
Some(handle_irq_callback::<T>),
flags.into_inner() as usize,
name.as_char_ptr(),
slot.cast(),
)
});
...
}
So seems to me the order of initialization has to be:
1. Initialize the `handler`.
2. `request_irq()`, i.e initialize the `RegistrationInner`.
3. `Devres::new()`, i.e initialize the `Devres`.
Regards,
Boqun
