transceivers: convert to counted_ringbufs

[Aaron-Hartwig]

I'm going to use a counted_ringbuf in the transceivers task because there is a lot of activity that happens, specifically in the UDP handler. I also bumped up the size from 16 -> 32 of that one because there's nearly constant activity and each UDP message involves at least three entries.

Fixes #1778

[hawkw]

I left a couple suggestions. At some point we may also want to add #[count(children)] to some of the errors, so that we can generate a separate counter for each error variant, but it's up to you whether we ought to do that now, and which errors are worth counting individually.

[Aaron-Hartwig]

At some point we may also want to add #[count(children)] to some of the errors, so that we can generate a separate counter for each error variant, but it's up to you whether we ought to do that now, and which errors are worth counting individually.

@hawkw I did add it in a few places, but some of the other places I'd like to add it are blocked by #[derive(Count)] being limited to enums right now. Most of the places I'd like to add it would be keyed off an event happening at a particular port (currently a usize in the Trace, but could easily be made a LogicalPort which is a struct that wraps a u8). Thanks for the feedback as always!

[hawkw]

I did add it in a few places, but some of the other places I'd like to add it are blocked by #[derive(Count)] being limited to enums right now. Most of the places I'd like to add it would be keyed off an event happening at a particular port (currently a usize in the Trace, but could easily be made a LogicalPort which is a struct that wraps a u8). Thanks for the feedback as always!

Yeah, I think that if you want to count events per LogicalPort, you would need to actually implement Count for that type manually. I would probably do it using an array of counters that's indexed by port, like this:

impl Count for LogicalPort {
    type Counters = [AtomicU32; NUM_PORTS],;

    #[allow(clippy::declare_interior_mutable_const)]
    const NEW_COUNTERS: Self::Counters = {
        const COUNTER_INIT: AtomicU32 = AtomicU32::new(0);
        [COUNTER_INIT; NUM_PORTS]
    };

    fn count(&self, counters: &Self::Counters) {
        let Some(ctr) = counters.get(self.0 as usize) else {
            // `LogicalPort`s should never index the counters array out of
            // bounds, but if that does ever happen, let's just bail instead of
            // panicking...
            return;
        }
        armv6m_atomic_hack::AtomicU32Ext::fetch_add(ctr, 1, Ordering::Relaxed);
    }
}

Although, OTTOMH, I'm not sure how well the humility code for reading counters would handle an array, so I might have to go back and fix that if you did do something like this...

[Aaron-Hartwig]

@hawkw should we do the Count impl for LogicalPort in this PR? I don't want to force Humility work upon you if you've something more pressing to work on. I can always write an issue for this and handle it in The Future.

[hawkw]

@hawkw should we do the Count impl for LogicalPort in this PR? I don't want to force Humility work upon you if you've something more pressing to work on. I can always write an issue for this and handle it in The Future.

My preference would be to go ahead and merge this as-is, and add something like that later --- I can think of a few other places where it would be nice to be able to use arrays for generating counters from a numeric index (e.g. SensorIds in task-sensors) if I get around to making it work nicely in humility!.

[Aaron-Hartwig]

That works for me. I've opened oxidecomputer/humility#487 so we remember where we talked about this. If you give this a +1 I'll get it merged!