nrf51 uart write blocks in qemu

[JoshMcguigan]

Hello! I am following along the rust discovery uart example by running the microbit v1 (nrf51) in qemu.

$ cargo build --features v1 --target thumbv6m-none-eabi

$ qemu-system-arm -machine microbit -display none -serial stdio -s -kernel target/thumbv6m-none-eabi/debug/uart

At first I thought the qemu emulation might be at fault, but I found this forum post where they seem to see a similar problem on real hardware with code that matches the uart write impl here:

nrf-hal/nrf-hal-common/src/uart.rs

Lines 134 to 141 in 939c017

	fn write(&mut self, byte: u8) -> nb::Result<(), Self::Error> {
	// Are we ready for sending out next byte?
	if self.0.events_txdrdy.read().bits() == 1 {
	// Reset ready for transmit event.
	self.0.events_txdrdy.reset();
	// Send byte.
	self.0.txd.write(|w| unsafe { w.bits(u32::from(byte)) });

This check for txdrdy hangs when trying to send the first byte. The reference manual says the txdrdy event is triggered only when a byte is actually sent. That aligns with the findings in the forum post that show very similar code blocking indefinitely.

Unfortunately I don't have real hardware to test this on at the moment.

[kaidokert]

There's a workaround for this - after configuring the UART, simply write a dummy byte via PAC.

    let mut actual_uart = Uart::new(
        pac_uart,
        pins,
        uart::Parity::EXCLUDED,
        uart::Baudrate::BAUD115200,
    );
    // Write one byte to get txrdy to be set
    unsafe {
        let uart0 = &*hal::pac::UART0::ptr();
        uart0.txd.write(|w| w.txd().bits(b' '));
    }

The code in Uart::new actually already does that before configuring the peripheral, and that works on real hardware. But QEMU emulation is slightly different and it it ignores everything written to it before .enabled() bit gets set.

[BartMassey]

The micro::bit v1 version of this book is pretty much out of maintenance at this point. That said, we would take a fix if it works on hardware without an unnecessary byte sent out on the UART. Bonus points if it also works on QEMU. Someone with a micro::bit v1 should test a proposed fix on both UART and QEMU.

Sorry, wrong repo. Let me look at this a bit.

[BartMassey]

Yeah, thanks to @kaidokert for pointing out the issue. I think this should probably be fixed in QEMU, not here, but contrary opinions are welcome. Don't want to put a dummy byte in the UART after enabling it, as it will presumably go out on the wire.

[JoshMcguigan]

Thanks @kaidokert, I agree this sounds like a QEMU issue. I'm going to close this issue.