Improve DMA pop()

[Dominaezzz]

esp-hal/esp-hal/src/dma/mod.rs

Lines 740 to 756 in 822bef1

	fn pop(&mut self, data: &mut [u8]) -> Result<usize, DmaError> {
	let avail = self.available;
	if avail < data.len() {
	return Err(DmaError::Exhausted);
	}
	unsafe {
	let dst = data.as_mut_ptr();
	let src = self.read_buffer_start;
	let count = self.available;
	core::ptr::copy_nonoverlapping(src, dst, count);
	}
	self.available = 0;
	Ok(data.len())
	}

Found this while I was writing the camera driver, if pop is called with data of size less than available:

• You end up with a buffer overrun, right?
• All the remaining available data seems to be discarded too.

My proposed fix for this is.

fn pop(&mut self, data: &mut [u8]) -> Result<usize, DmaError> {
    let avail = self.available;

    if avail < data.len() {
        return Err(DmaError::Exhausted);
    }
    
    let dma_buf = unsafe {
        core::slice::from_raw_parts(self.read_buffer_start, avail)
    };
    data.copy_from_slice(&dma_buf[..data.len()]);

    self.available -= data.len();
    Ok(data.len())
}

What I don't understand is why pop returns a usize if it's always the same as data? Was pop meant to also work when the data buf is more than the available data?

[bjoernQ]

The current naive approach is that pop will always pop the buffer of one descriptor exactly.

I think for the fix we also would need to change available: Currently available will overwrite read_buffer_start and available.
Also pop would need to adjust read_buffer_start . We would need to also take wrapping into account

I think push is already working like that and probably it would be good to have them working similar

If the API works like this, we won't need the DmaError::Exhausted