Improvements and CI

[elpiel]

• I haven't tested the CI
• I fixed a few things to be able to build docs
• fixed a few places that add deasync attribute

[elpiel]

@qiuchengxuan I fixed a few missed places for the non-async API and exported a few private members. I've also included a docsrs attributes to showcase which members/modules are enabled by a feature.

I would like to see if the CI works but I suppose you need to enable to build first or we should merge it first in main?!

[qiuchengxuan]

You should meld 4 commits into one

[qiuchengxuan]

@qiuchengxuan I fixed a few missed places for the non-async API and exported a few private members. I've also included a docsrs attributes to showcase which members/modules are enabled by a feature.

I would like to see if the CI works but I suppose you need to enable to build first or we should merge it first in main?!

Up to you, both ok to me

[elpiel]

@qiuchengxuan I'm trying to test a littlefs storage impl with async-sdmmc, however, I keep getting timeouts on read_block or panics.

Another thing to note: tx & rx can have difference sizes and currently the Transfer::transfer impl for SPI is panicking.

https://github.com/qiuchengxuan/async-sdmmc/blob/464c1f44b89eb5c0b537520fb8a7b6ac26c6c582/src/bus/spi/bus.rs#L42

This is especially true since Bus::tx & Bus::rx already pass empty buffers to transfer: https://github.com/qiuchengxuan/async-sdmmc/blob/464c1f44b89eb5c0b537520fb8a7b6ac26c6c582/src/bus/spi/bus.rs#L96-L102

embedded_hal::spi Transfer trait seems to return the word read in the transaction, not sure what you meant that rx & tx are the same for embedded_hal. So far I have made this solution to the issue:

#[cfg(not(any(feature = "async", feature = "async-trait")))]
impl<E, T: spi::Transfer<u8, Error = E>> Transfer for T {
    type Error = E;

    fn transfer(&mut self, tx: &[u8], rx: &mut [u8]) -> Result<(), Self::Error> {
        // when rx > tx
        if rx.len() > tx.len() {
            // only replace the tx bytes length in rx
            rx[..tx.len()].copy_from_slice(tx);

            assert_eq!(&rx[..tx.len()], tx);

        } else if rx.len() < tx.len() {
            rx.copy_from_slice(&tx[..rx.len()])
        } else {
            // when rx == tx
            rx.copy_from_slice(tx);
        }

        self.transfer(rx).map(|_| ())
    }
}

[elpiel]

On the Timeout issue, I've found a wrongly connected wire and managed to get to the CRC checks now.

One thing to note - I had to alter the loop for the token to continue looping if unknown token has been met.

Now, for some reason, I get a CRC error, which happens before the read_block that was failing before :thinking:

[qiuchengxuan]

Oops, bus.rs:97 passed an empty slice as rx while bus.rs:42 copies tx to rx

Need to replace it with embedded_hal::blocking::spi::Write when tx only

[qiuchengxuan]

On the Timeout issue, I've found a wrongly connected wire and managed to get to the CRC checks now.

One thing to note - I had to alter the loop for the token to continue looping if unknown token has been met.

Now, for some reason, I get a CRC error, which happens before the read_block that was failing before 🤔

In theory 74 cycles should be sufficient, and several sdmmc library use identical amount of loop, so I've no idea why it's unstable either. Actually I'm not familiar with sdmmc too

[elpiel]

I haven't had time to work on a fix, however, you can check a fix for such problems with initialization of the card in this branch: https://github.com/rust-embedded-community/embedded-sdmmc-rs/pull/76

It has worked very reliably so far, even though embedded-sdmmc crate has other issues related to the fat filesystem implementation