Async support without blocking #90
Comments
|
Hi @tarfu, I am currently on vacation, and just checking in once in a while from the pool 😄 Long story short is that we are currently working on an async first rewrite of this library along with the ublox wifi crate, that is much more along the same implementation strategy as embassy-net, and fits well with the embedded-nal-async way of doing things. |
|
Hi, that sounds great and a nice vacation to you. |
|
Thank you! |
|
Hi @tarfu As I said, we are currently experimenting with different architectures and designs on how we want to implement async-first implementations for this crate as well as the Wi-Fi cousin crate at ublox-short-range-rs. So far we have reached this https://github.com/BlackbirdHQ/ublox-short-range-rs/tree/async/no-channel/src/asynch as our most promising approach. The two examples at https://github.com/BlackbirdHQ/ublox-short-range-rs/tree/async/no-channel/examples/rpi-pico/src/bin compiles and works, so they should be a nice place to see how it looks in action (embassy based) TLDR of the implementation. It splits the crate into 3 parts:
As for the Others
Feedback, reviews and implementation help is HIGHLY appreciated! ❤️ Nothing is set in stone, this is merely our most promising approach based on a few experiments and reading a ton of reference implementations. We have not yet begun the cellular implementation, as we are still playing around in the Wi-Fi realm (less state and usually easier to debug with wireshark etc.), but we think it should be fairly easy to apply the same approach to cellular, even though it will obviously be more complex. |
|
I had a quick look at the Wi-Fi crate and I thought about extensibility, as cellular chips have a lot of functionality and tend to be on the more feature rich side. It would be great to have something that gives you a basic init, and then you maybe want to split stuff like PDP/PDS management and stuff like that to be reusable by multiple managers. This is basically what your point 3 is about. A probable issue with this design is RAM usage, and it could be a trait bound nightmare. My main things I would need are sockets, FOTA, location and time and my main module right now is a sara-r5 (SARA-R510M8S to be exact). |
I completely agree on the part about it usually being a lot more feature rich, hence my point about why we started with the Wi-Fi crate internally. I think we have a bit of different terminology going on here. I would not implement service specific features on I think it would make sense to have each of these handle their own relevant URC's, such that the crate would need to be able to have Furthermore, I think the PDP/PSD is common for pretty much all the services, and should be configured either through the
I don't see why we would introduce traits into much of this?
Yeah, that's a completely different story, and one I think will be mainly managed by modifications to ATAT (BlackbirdHQ/atat#76) Example#[embassy_executor::task]
async fn wifi_task(runner: Runner<'static, AtClient, Output<'static, PIN_26>, 8>) -> ! {
runner.run().await
}
#[embassy_executor::task]
async fn net_task(stack: &'static UbloxStack<AtClient>) -> ! {
stack.run().await
}
#[embassy_executor::task]
async fn ingress_task(
mut ingress: atat::Ingress<'static, AtDigester, EdmEvent, RX_BUF_LEN, URC_CAPACITY, 2>,
mut rx: uart::BufferedUartRx<'static, UART1>,
) -> ! {
ingress.read_from(&mut rx).await
}
bind_interrupts!(struct Irqs {
UART1_IRQ => BufferedInterruptHandler<UART1>;
});
#[embassy_executor::main]
async fn main(spawner: Spawner) {
defmt::info!("Hello World!");
let p = embassy_rp::init(Default::default());
let rst = Output::new(p.PIN_26, Level::High);
let (tx_pin, rx_pin, rts_pin, cts_pin, uart) =
(p.PIN_24, p.PIN_25, p.PIN_23, p.PIN_22, p.UART1);
let tx_buf = &mut make_static!([0u8; 64])[..];
let rx_buf = &mut make_static!([0u8; 64])[..];
let uart = uart::BufferedUart::new_with_rtscts(
uart,
Irqs,
tx_pin,
rx_pin,
rts_pin,
cts_pin,
tx_buf,
rx_buf,
uart::Config::default(),
);
let (rx, tx) = uart.split();
let buffers = &*make_static!(atat::Buffers::new());
let (ingress, client) = buffers.split(tx, AtDigester::default(), atat::Config::new());
defmt::unwrap!(spawner.spawn(ingress_task(ingress, rx)));
let state = make_static!(State::new(client));
let (device, mut control, runner) = new(state, &buffers.urc_channel, rst).await;
defmt::unwrap!(spawner.spawn(wifi_task(runner)));
// Configure PDP/PDS stuff on either `Device` or `Control`, not sure which makes most sense?
// Init network stack with room for 4 sockets
let stack = &*make_static!(UbloxStack::new(
device,
make_static!(StackResources::<4>::new()),
));
defmt::unwrap!(spawner.spawn(net_task(stack)));
// Not sure how fota is normally used, does it need to be static and have a background runner, or is it a temp thing?
let fota = &*make_static!(FOTA::new(
device
));
// And now we can use it!
defmt::info!("Device initialized!");
let addr = SocketAddr::V4(SocketAddrV4::new(Ipv4Addr::new(10, 42, 0, 1), 8000));
// `client` here implements `embedded-nal-async`, and issues `TcpSocket` on calls to `connect()`
let state: TcpClientState<1, 1024, 1024> = TcpClientState::new();
let client = TcpClient::new(&stack, &state);
info!("connecting...");
// `socket` implements `embeded-io-async`
let socket = client.connect(addr).await;
// Example code. Might look different IRL
let tls_config = tls_config::new().cert(b"some device cert").root_ca(b"some root CA").private_key(b"Some private key").build();
let tls_state: TlsClientState<1, 1024, 1024> = TlsClientState::new();
// `tls_client` here implements `embedded-nal-async`, and issues `TlsSocket` on calls to `connect()`
let tls_client = TlsClient::new(&stack, &tls_state, tls_config);
info!("connecting...");
// `tls_socket` implements `embeded-io-async`
let socket = tls_client.connect(addr).await.unwrap();
// It is also possible to go without the `embedded-nal-async` abstractions, and create sockets directly like:
let mut rx = [0u8; 128];
let mut tx = [0u8; 128];
// `socket` implements `embeded-io-async`
let socket = TcpSocket::new(stack, &mut rx, &mut tx);
socket.connect(addr).await.unwrap();
let mut tls_rx = [0u8; 128];
let mut tls_tx = [0u8; 128];
// `tls_socket` implements `embeded-io-async`.
let tls_socket = TlsSocket::new(stack, &mut tls_rx, &mut tls_tx);
tls_socket.connect(addr, tls_config).await.unwrap();
loop {
}
} |
Yes, we had some different terminology going on there. On the development side, would you want to directly put it in master or dev on an async branch and I PR against this branch? |
Think I would prefer a branch. Would you like to establish the initial structure, then we can PR to it from there, maybe with a small comment somewhere on what we are working on, so we don't end up with duplicate work 😅 |
|
If we take a branch, it would be a good idea that you create the branch and maybe the structure, so we can PR against that branch here in the same repo and target that branch for the PRs and base the PRs from all within this repo. I at least haven't seen any branch we could use for that or was already used for something like that here. |
Right now, the Ingress supports async but the sending, spinning and data service related functions have blocking operations which makes it impossible to use in an environment like embassy (with a single executor).
Copying the client and getting everything to async is not that bad, but copies a lot of code in the easy to do implementation.
The hard part is the data service and socket related stuff.
embedded-nal-async has quite a different API. You only implement a connect function which returns a socket which you can write to and read from via the embeeded-io-async traits, and dropping the socket would need to cause a disconnect.
The text was updated successfully, but these errors were encountered: