On Wed, Jul 23, 2025 at 10:55:20AM -0300, Daniel Almeida wrote:
> Hi Boqun,
>
> [...]
>
> >> + IrqRequest { dev, irq }
> >> + }
> >> +
> >> + /// Returns the IRQ number of an [`IrqRequest`].
> >> + pub fn irq(&self) -> u32 {
> >> + self.irq
> >> + }
> >> +}
> >> +
> >> +/// A registration of an IRQ handler for a given IRQ line.
> >> +///
> >> +/// # Examples
> >> +///
> >> +/// The following is an example of using `Registration`. It uses a
> >> +/// [`AtomicU32`](core::sync::AtomicU32) to provide the interior mutability.
> >
> > We are going to remove all usage of core::sync::Atomic* when the LKMM
> > atomics [1] land. You can probably use `Completion` here (handler does
> > complete_all(), and registration uses wait_for_completion()) because
> > `Completion` is irq-safe. And this brings my next comment..
>
> How are completions equivalent to atomics? I am trying to highlight interior
> mutability in this example.
>
Well, `Completion` also has interior mutability.
> Is the LKMM atomic series getting merged during the upcoming merge window? Because my
> understanding was that the IRQ series was ready to go in 6.17, pending a reply
Nope, it's likely to be in 6.18.
> from Thomas and some minor comments that have been mentioned in v7.
>
> If the LKMM series is not ready yet, my proposal is to leave the
> Atomics->Completion change for a future patch (or really, to just use the new
> Atomic types introduced by your series, because again, I don't think Completion
> is the right thing to have there).
>
Why? I can find a few examples that an irq handler does a
complete_all(), e.g. gpi_process_ch_ctrl_irq() in
drivers/dma/qcom/gpi.c. I think it's very normal for a driver thread to
use completions to wait for an irq to happen.
But sure, this and the handler pinned initializer thing is not a blocker
issue. However, I would like to see them resolved as soon as possible
once merged.
Regards,
Boqun
>
> - Daniel
