Better Circular DMA support

[rudihorn]

This is my initial draft on PR on issue #242.

The example serial-dma-circ-len.rs shows how to use it in more detail, but the most important changes are demonstrate in the following snippet:

    let mut circ_buffer = rx.circ_read(buf);

    // wait until we have 5 bytes
    while circ_buffer.len() < 5 {}

    let mut dat = [0 as u8; 4];
    assert!(circ_buffer.read(&mut dat[..]) == 4);

One thing I have noticed is that currently a lot of the responsibility in setting up DMA is coded in the using module (e.g. in the serial module. I've introduced a setup function, so that it happens within the DMA module, and the setup within the serial module looks like:

                fn circ_read(self, buffer: &'static mut B) -> CircBuffer<u8, Self>{
                    let buffer = buffer.as_mut_slice();
                    let paddr = unsafe { &(*$USARTX::ptr()).dr as *const _ as u32 };

                    let mut buf = CircBuffer::new(buffer, self);
                    unsafe { buf.setup(paddr, Priority::MEDIUM) };
                    buf
                }

I have renamed the old CircBuffer behaviour renamed into e.g. CircDoubleBuffer.

I have only implemented this for the serial module so far.

[TheZoq2]

Sorry about the delay.

As an overview, I think this looks good (though I may be a bit tired to notice every detail at the moment :) )

Would you mind adding a changelog entry for the circular DMA stuff (since it looks like the only entry is from #239 which I guess this is based on)

[TheZoq2]

This is probably out of scope for this PR, but maybe we should consider returning a result here

[thalesfragoso]

Hmm, I think if this fails it's a configuration bug, so I'm not sure what a Result would give us here.

[TheZoq2]

Len doesn't sound like the right name for this, perhaps elements_available is more descriptive.

[thalesfragoso]

Thanks for your work, I just left some suggestions and questions. Could you also fix CI ?

This one is a bit tricky because it's hard to test for overruns without producing false positives, but since we are just copying data quickly here I think we're mostly okay.

[thalesfragoso]

You need a compiler_fence before reading the buffer, an Acquire fence should do here.

[thalesfragoso]

You need a compiler_fence before reading the buffer, an Acquire fence should do here.

[thalesfragoso]

Deref is already implied by StableDeref, so I don't think we need it here. Also, why do you think Unpin is needed here ?

[thalesfragoso]

Same comment as the one in dma.rs.

[thalesfragoso]

Same comment as the one in dma.rs.